Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Investigations on the execution and evaluation of the Pummel test for polyvinyl butyral based interlayers

Investigations on the execution and evaluation of the Pummel test for polyvinyl butyral based... Glass Struct Eng (2020) 5:371–396 https://doi.org/10.1007/s40940-020-00120-y(0123456789().,-volV)(0123456789().,-volV) SI: C HALLENGING GLASS Investigations on the execution and evaluation of the Pummel test for polyvinyl butyral based interlayers . . Miriam Schuster Jens Schneider Tuong An Nguyen Received: 1 February 2020 / Accepted: 7 April 2020 / Published online: 4 June 2020 The Author(s) 2020 Abstract Laminated safety glass (LSG) is increas- Three image evaluation methods in Matlab are ingly used as structural element in buildings. Of presented, discussed and compared: (1) analysis of central importance for safety are the adhesion and the binary images, (2) statistical evaluation of the grey- residual load-bearing capacity in the post fractured scale images and (3) texture analysis using co- state. In literature a large number of tests to assess occurrence matrices. adhesion is mentioned. These include, e.g. peel tests, through-cracked-tensile/-bending tests, VW-pull tests Keywords Laminated safety glass  Adhesion and compressive shear tests. However, especially in Pummeltest  Automatization  Digital image industry, the Pummel test is widespread for determin- evaluation ing the quality of adhesion in LSG with polyvinyl butyral based interlayers. This test method proves to be simple and quick to carry out: The laminate is stored at - 18 C and then completely destroyed at 1 Introduction room temperature with hammer blows. The adhesion level (0–10) is determined by visually comparing the 1.1 Laminated safety glass adhering glass fragments with reference pictures or with the help of diagrams and tables which indicate the The use of glass in the building industry has risen Pummel value as a function of the free film surface. sharply in recent years with the ongoing architectural Pummel value 0 is to be interpreted as no adhesion and trend towards transparent structures. In this context, Pummel value 10 as very high adhesion. Due to the the glass must increasingly meet safety-relevant lack of standardization, the execution and evaluation requirements with regard to breakage and post-break- is very much dependent on the test institution and age behaviour. The unannounced failure of the brittle executive person. This paper shows different Pummel material glass is countered by the use of laminated classifications that can currently be found on the safety glass (LSG). According to EN ISO 12543-2 market. Subsequently, approaches to the automatiza- (2011), laminated safety glass, consisting of two or tion and standardization of the execution and espe- more glass panes that are laminated together to form a cially the evaluation of the Pummel test are shown. package by means of polymer interlayers, must bind the resulting glass splinters after glass breakage and enable a residual load-bearing behaviour of the M. Schuster (&)  J. Schneider  T. A. Nguyen glazing. Accordingly, the adhesion between glass TU Darmstadt, Darmstadt, Germany e-mail: schuster@ismd.tu-darmstadt.de 123 372 M. Schuster et al. and film plays a decisive role. In laminated safety glass The bond between PVB and glass is characterized with low to no adhesion, the glass pane will delam- by dense reversible hydrogen bonds between the inate and after glass breakage the glass splinters do not polyvinyl alcohol in PVB and the silanol group in the stick on the interlayer. Contrary, a very high adhesion glass, see Fig. 1. Accordingly, the degree of adhesion results in a quasi-monolithic brittle failure of the can be controlled via the film composition, e.g. the laminate. The film adheres so strongly to the glass content of polyvinyl alcohol, the presence of adhesion- surface that elastic deformations are only possible to a regulating additives or the interlayer moisture content. limited extent and the impact energy cannot be In (Ensslen 2005) the optimum moisture content of dissipated. With an optimal degree of adhesion, the 0.45% ± 0.07% in the PVB-based film is given. This interlayer has a splinter-binding effect and can range remains largely constant at a relative humidity partially detach from the glass plate so that the impact of 30%. At a relative humidity between 50 and 70%, energy can be absorbed and the glass fragments will the water content in the film increases linearly. From a continue to adhere to the film. water content of 1%, the adhesive bond deteriorates Application situations in which laminated safety enormously and from 2% the film becomes milky glass is required are, for example, when the glass is cloudy. In addition to the film properties, other factors used for fall protection, can be walked on, is used as also have a considerable influence on adhesion. If the horizontal glazing with the traffic area underneath glass panes are not cleaned properly, grease and (overhead glazing) or as a structural component (e.g. limescale may be present on the surface and thus as a glass beam). The most common interlayer prevent hydrogen bonding. Kuraray (2012) recom- products in building industry are polyvinyl butyral mends distilled or fully desalinated washing water, ?? ?? (PVB)-based with a market share of over 90%. Other since alkaline earth ions (Ca and Mg ) reduce the interlayers include ethylene vinyl acetate (EVA) and glass adhesion even at low quantities enormously ionoplastics (SentryGlas ). (Keller et al. 2002; Kuraray 2012). The glass surface This article deals exclusively with laminated glass roughness also has an influence, since a rough where PVB interlayers are used. The adhesive microstructure favors mechanical interlocking. Since behaviour between glass and film will be discussed the tin side of glass has a smoother surface structure ?? in more detail in the following sections. and can present Sn ions, usually a slightly lower adhesion than the air side is observed. Finally, errors in 1.2 Adhesion the lamination process may result in air bubbles between the interlayer and glass, which prevent flat First the production process of laminated safety glass contact between glass and film and therefore lead to is briefly explained. The first step in the production of insufficient adhesion levels. laminated safety glass consists of cutting of the glass sheets to the desired size, edge processing and cleaning. The previously air-conditioned interlayers are then cut and the laminate is layered. This inevitably leads to air inclusions between the film and glass, which are mostly removed in the pre- lamination process using rollers in heating tunnels, where PVB and glass are heated to 35 C and 60 C and using a negative pressure of - 0.6 to - 0.9 bar. In addition, the edges are sealed in this process step to prevent air from penetrating the sandwich. After the pre-laminate production, the adhesive effect between film and glass is already given, but small air inclusions can remain. These are dissolved by the final laminate Fig. 1 Adhesion between glass and PVB, based on Kuntsche production in the autoclave at a temperature of (2015) approximately 135 C and a pressure of maximum 14 bar. 123 Investigations on the execution and evaluation of the Pummel test 373 1.3 Adhesion and post-fractured tests maximum shear force until breakage is a measure of adhesion (Kuraray 2012; Schneider et al. 2016). There are various test methods in the construction The adhesive tensile test, also known as the VW industry for quantifying the adhesion between glass Pull Test (Fig. 2c), was developed as an alternative to and polymer. Besides the ball drop test and the the compression shear test. The LSG sample is joined pendulum impact test, the tests are not standardized, so to a steel plate and then clamped into a universal that differences are possible in their performance testing machine. During the test procedure the plates (Schneider et al. 2016). In this section, the adhesion are separated at a constant speed at room temperature. tests are briefly explained, whereby the Pummel test is The force is applied until the glass plate is released dealt with in more detail with regard to the procedure from the foil. The adhesive tensile strength is calcu- and the advantages and disadvantages. lated via the technical/engineering breaking stress The ball drop test (Fig. 2a) is regulated in the (Franz 2015). In the adhesive shear test (Fig. 2d), German Standard DIN 52338 (2016) and the European laminated safety glass samples are subjected to a shear Standards EN 14449 (2005) and EN 356 (1999). In this test until the glass plate has detached itself from the test a defined steel ball falls from a certain height onto foil (Franz 2015). Compared to the adhesive shear test, the laminated safety glass. It is evaluated whether the the adhesive tensile test is much more complex. The ball penetrates the glass. Additionally the mass of the sample preparation with bonding is time-consuming detached glass splinters is evaluated. The pendulum and requires extreme accuracy (parallelism of both impact test (Fig. 2b) according to the European metal anvil plates). In addition, the glass panes must be Standard EN 12600 (2003) is an impact test similar separated from the steel plate after the test. For these to the ball drop test. Here, the double-tyre pendulum reasons the adhesive shear test is preferred. with different drop heights simulates the impact of a For the peel test (Fig. 2f), the test specimens cannot person at different speeds. The test is considered to be be taken from conventional production. A release film passed if none of the specimens breaks or breaks in a must be added during the production process so that non-hazardous manner. For the latter, the result is the interlayer only adheres to one glass pane. Alu- defined by the weight and size of the glass fragments. minum foil is laminated to the PVB film to prevent In the compression shear test (Fig. 2e), the adhe- high elongation when peeling the PVB film. The PVB sion resistance between glass and foil is determined at film should also protrude so that it can be clamped in room temperature in a universal testing machine, the tensile testing machine. The protruding film is where the sample, which is clamped at an angle of 45, pulled off the glass pane at a defined angle with a is loaded both in compression and in shear. The constant force level. The applied force is a measure for adhesion (Kuraray 2012; Schneider et al. 2016). Fig. 2 a Ball drop test (Schneider et al. 2016), b Pendulum (based on EN 12600 2003), c VW-pull test (Franz 2015), d shear test (Schneider et al. 2016), e CSS (Fahlbusch 2007), f Peel test (based on Kinloch et al. 1994), g, h TCT and TCB (Scholz Campos 2019) 123 374 M. Schuster et al. Disadvantages of the test method are the costly the subjective evaluation, which is judged by the production of the test specimens and the influence of naked eye of the examiner. Standardization of the the elongation of the PVB film and the aluminum film. execution procedure and evaluation would finally This test is used especially in the photovoltaic industry allow an objective statement. For this purpose, Sect. 2 for EVA films (Kuraray 2012). first presents the state of the art and research for For the Through-Cracked-Tensile Test (TCT, realization and evaluation. Subsequently, conducted Fig. 2g) and the Through-Cracked-Bending Test experiments are presented in Sect. 3 and automated (TCB, Fig. 2h), the glass panes of the laminated glass evaluation methods are proposed and discussed in sample are broken in a defined manner. The sample is Sect. 4. then subjected to a tensile test or a four point bending test respectively. The TCT test determines not only the delamination behaviour between glass and interlayer 2 Pummel test but also the material behaviour of the interlayer itself. Additionally there is the possibility to determine Although the Pummel test is widely used, there is no parameters for cohesive zone models, such as the standardization for its implementation and evaluation. energy release rate (Franz 2015; Schneider et al. In literature, however, there are different patent 2016). specifications, manufacturer guidelines and isolated research papers, which are briefly presented here. The Pummel test (further description in Sect. 2)is the most common test method for estimating the adhesion between glass and interlayer in production 2.1 Patent specifications facilities. After the sample has been brought to the desired temperature of - 18 C for several hours, it is The Pummel test was mentioned for the first time in placed on an inclined hard metal base. Then a certain the United States Patent US 3,434,915, which was glass surface is pulverized by means of hammer blows applied by Garrison (1965) in 1965 for laminated (manual, semi-automatic e.g. by a hydraulic or safety glass in car windscreens. According to this electrical controlled hammer that always strikes the patent, the test specimens (no dimensions specified) same spot so that the laminate has to be moved are first stored for 16 ± 4 h at a temperature of manually, or fully-automatic). One side of the sample approx. - 18 C. The cooled specimen is held against is Pummeled on the tin side (Sn) and one side of the an angled metal surface. The sample is inclined so that sample on the air/fire side (A/F). The evaluation is one side of the glass only touches the metal with the done visually with the help of reference pictures or edge. Then a flat-headed hammer is used to pulverize tables indicating percentages of free film surface, see the surface. The destroyed surface is at least 7.62 cm Sects. 2.1 and 2.2. The adhesion scale, also known as in diameter and the remaining glass particles should Pummel values, ranges from 0 (low adhesion) to 10 not be larger than 0.64 cm. Afterwards, the loose glass (high adhesion). Compared to the other test methods fragments are carefully removed by shaking gently presented above, the Pummel test has the advantage and the degree of adhesion is determined using a that it can be carried out without much effort. table containing Pummel values for different free film Furthermore, no complex and expensive testing surfaces, see Table 1. machines or sample preparations are required. For a In 1977 the United States Patent US 4,144,376 manual execution only a hammer and an angled metal (Beckmann and Knackstedt 1977) was registered. support are necessary. However, the lack of standards Here, a procedure to regulate the adhesion of a leads to enormous differences in implementation and plasticized, partially acetalized polyvinyl alcohol film evaluation. The test method depends strongly on the for required Pummel adhesion values is described. experience of the test person. In particular, the number PVB resin was processed with conventional methods and intensity of hammer blows can lead to different in newly determined quantities for laminated safety results, see Sect. 3. Furthermore, changes in the angle glass. The samples for the Pummel test had the of inclination between specimen and metal surface dimensions 150 9 300 mm and were cooled for 2–8 h may result in different fracture patterns (Hark 2012). at a temperature of - 18 C(± 0.5 C). The samples The different Pummel values result additionally from were then placed on a metal block angled at 45. The 123 Investigations on the execution and evaluation of the Pummel test 375 Table 1 Pummel Values in dependency of free film surface Free film surface (%) Free film surface (%) Free film surface (%) Pummel value US 3,434,915 and US 4,144,376 US 6,984,679 B2 EP 1 470 182 B1 – – Large areas of total detachment - 1 100 n.e. 100 0 95 n.e. 99 1 90 97 97 2 85 n.e. 92 3 60 83 83 4 40 n.e. 67 5 20 43 43 6 10 n.e. 20 7 5 n.e. 8 8 2 n.e. 1 9 0 n.e. 0 10 n.e. not evaluated Estimated values out of graph for selected Pummel values glass surface was treated with a flat-headed hammer based. If one compares the graph with Table 1, one until the glass was completely pulverized. The test sees differences in the Pummel value assignment. For area was 100 9 150 mm. The Pummel test was example, in US 3,434,915, a Pummel value of 2 performed and evaluated according to the US Patent corresponds to a free film surface of 90%, a Pummel 3,434,915. value of 4 corresponds to a free film surface of 60% The Pummel test is also described in the United and Pummel 6 corresponds to a free film surface of States Patent US 6,984,679 B2 (Papenfuhs and Steuer 20%, while in US 6,984,679 B2 Pummel 2 corre- 2006) and is used to check the level of adhesion. The sponds to a free foil area of * 97% Pummel 4 test specimens are each stored with 2 9 2mm float corresponds to a free foil area of * 83% and Pummel glass (100 9 300 mm) for 4 h at a temperature of 6 corresponds to a free foil area of * 43%. - 18 C. The laminate is placed on a 45 inclined The European Patent EP 1 470 182 B1 (Keller et al. support and destroyed with an automatic hammer. 2002) treats a plasticized film of partially acetalized Figure 3a shows the impact pattern and Fig. 3b the polyvinyl alcohols as an interlayer for laminated graph, on which the evaluation of the Pummel test is safety glass. Among other things, these laminates are Fig. 3 a Sketch of the impact pattern according to US 6,984,679 B2, b modified evaluation graph, based on US 6,984,679 B2 (Papenfuhs and Steuer 2006) 123 376 M. Schuster et al. subjected to the Pummel test described in the patent. temperature equilibrium. After the samples are con- The test specimens (80 9 300 mm) consist of two ditioned, the Pummeling procedure should start within 2.1 mm thick float glass panes and are stored for at seconds of removal from the freezer. It is recom- least 24 h in a cold chamber at a temperature of mended, that the sample is held at about a 5 angle to - 18 C(± 2 C). The specimens are then placed on the plane of the Pummel plate, so that only the edge of a steel base at an inclination of 5 and Pummeled the unbroken glass contacts the plate. The laminate evenly with a hammer head (500 g, round head). The should be struck progressively in about 10 mm impact intensity is 1 Nm per impact. To ensure that the increments along the bottom 15 mm of the laminate. glass plate is completely destroyed at every point all When the bottom edge has been completely pulver- the blows overlap by 75%. At least 6 cm (from the ized, the next 15 mm has to be pulverized in the same lower edge measured) is destroyed. After the sample is manner. This has to be repeated until at least 7–10 cm lightly tapped, a visual inspection is carried out at of the laminate has been Pummeled. Care is exercised room temperature. The Pummel values (0–10) are to insure that all smooth glass is pulverized. Loose assigned via the exposed foil areas. If a test specimen glass dust has to be removed by slightly brushing the shows large areas of total detachment, Pummel value Pummeled laminate. For the evaluation and Pummel - 1 can be assigned to it. The classification of the classification it is advised to place the laminate on a Pummel values as a function of the free film surface black background and to compare the sample to the corresponds to the graph from US 6,984,679 B2. Pummel adhesion references. Everlam has also provided the authors with a 2.2 Manufacturer guidelines brochure containing a description of the Pummel TM procedure as well as the evaluation for EVERLAM - Eastman provides a description of the Pummel test and PVB laminates. Here the test specimens (no dimen- its evaluation. Eastman has sent reference pictures and sions) have to be cooled down to a temperature of a description of the experiment to the authors. - 18 C for 3 h. The sample is then Pummeled Reference pictures of the Pummel classes 1–9 are manually with a flat-head hammer with a weight of given in Table 2. The description of the experiment 500 g on an anvil at 5. The blows are made with a can also be found in (Eastman 2013). It is recom- distance of 1.25 cm in a row and with 2 cm distance mended to put the laminated samples (no dimension between the rows. The destroyed laminate is visually given) in an - 18 C freezer for enough time (1–6 h inspected with the reference pictures (Table 2). Fur- depending on freezer) in order that they may reach thermore, an evaluation diagram based on the values Table 2 Reference pictures from different PVB manufacturers Pummel Pummel Pummel Pummel Pummel Pummel Pummel Pummel Pummel Pummel 1 2 3 4 5 6 7 8 9 10 First row: Eastman (sent via Email); second row: Kuraray (Pictures from poster, sent via E-mail); third row: Everlam (Pummel Standards sent via Email) 123 Investigations on the execution and evaluation of the Pummel test 377 of the US 4,144,376 and correlation diagrams of adhesion, the Pummel test and the compression shear Pummel values with CSS and ball drop performances test. In the Pummel test, the laminated glass sample are included. (no dimensions given) was cooled to - 18 C and In Kuraray’s manual dealing with the processing of Pummelled with a hammer weighing approx. 453 g on Trosifol PVB, the experimental execution is briefly a steel plate until both sides of the sample were described (Kuraray 2012). The sample dimensions are pulverized. The destroyed sample was then visually 80 9 200 mm (resp. 80 9 300 mm for windscreen evaluated according to the Pummel scale 0–10. By samples) and the float glass thickness should not comparing the results of the two test methods, it can be exceed 4 mm. The sample is stored for at least 2 h at a seen that these usually correlate almost linearly, if the temperature of - 18 C. The Pummeling process is condition is fulfilled that the test specimens are semi-automatic. The glass forms a small angle to the derived from the same batch. inclined metal base. According to the manual, a rough Franz (2015) evaluated in his dissertation the state distinction is made between high and medium glass of knowledge on fractured laminated safety glass and adhesion according to the classification in a diagram. the further development of the residual strength tests. In addition to the manual, Kuraray also sent us a poster Two test methods were used to assess the adhesive with reference pictures of Pummel values from 1 to 10 strength and quality. On the one hand, the Pummel test for Trosifol interlayers. Those are represented in to check the adhesive quality, on the other hand the Table 2. Additionally, a Pummel manual can be found adhesive shear test to verify the adhesive strength. The on the manufacturers website (Kuraray 2014). Regard- description of the execution of the adhesive shear test ing laminates with PVB, the Pummel standards is omitted here. The Pummel samples consisted of (Pummeling procedure, reference pictures, graph with 2 9 3 mm float glass and 1.52 mm PVB foil. The foil free film surface, correlations to CSS and ball drop had three different degrees of adhesion (BG R10, BG tests) correspond to those of Everlam. It should be R15 and BG R20, Trosifol (R) interlayers). The noted, that the reference pictures differ from the poster dimensions were 80 9 300 mm. The laminates were and the Pummel standard manual. stored at - 18 C for 2 h. Then the tin and fire sides There is also a working instruction for the Pummel were Pummelled with a hammer to determine the test on laminated safety glass with ionoplast (Sen- adhesion of both glass sides. For evaluation the tryGlas ). However, since this interlayer material is destroyed samples were visually compared with not part of the investigations presented, it will not be reference pictures and classified on a Pummel scale discussed in detail here. Further information can be from 0 to 10. Franz (2015) compared the Pummeling found in (Stu ¨ we 2007). results with the shear strengths resulting from the Looking at the classification based on the reference shear test. He showed that the adhesive stresses images of different manufacturers (Table 2), one can increase with increasing Pummel value. see that already the adhesion scale differs among the Hof and Oechsner (2017) investigated the adhesion film manufacturers. Kuraray compares the samples of laminated glass as a function of moisture content with Pummel values from 1 to 10, Eastman with the within the film with three different test methods. With values from 1 to 9 and Everlam only with the Pummel the help of a spectrophotometer they were able to values from 2 to 9. Reasons for further differences are determine the light transmittance at a specific wave- different lighting conditions and the different record- length (which is dependent on the interlayer moisture level) of several glass panes. Furthermore, the VW ing angle of the samples. Pull Test and the Pummel test were applied. The 2.3 Research samples had the dimensions 100 9 200 mm and a total thickness of about 8.5 mm. Both sides of the Pummel tests were carried out in several research sample were destroyed with a hammer. The results projects on the subject of adhesion and correlations showed that samples with a moisture content below with other adhesion tests were investigated. 0.67% led to a high Pummel value (Pummel 7–8) and Keller and Mortelmans (1999) investigated the samples with a moisture content above 0.67% led to adhesion mechanism between glass and interlayer. low Pummel values (Pummel \ 3). Two test methods were used to determine the 123 378 M. Schuster et al. In his bachelor’s thesis, Hark (2012) was mainly were then cut to 100 9 300 mm large samples by the concerned with the evaluation of the Pummel test. For manufacturer. According to the manufacturer, the this purpose he examined different procedures, such as samples should have a Pummel value of 7. weight change, change in the light transmission of the All samples were provided with an impact pattern laminated safety glass, profile analysis of the test consisting of 56 impact fields (7 rows of 8 impact specimen, maximum stress and maximum elongation fields, see Fig. 4b) before storage. In addition, the tin of Pummelled specimens. The test specimens had a and air side was detected with a UV lamp and marked. dimension of 25 9 7.5 cm. They consisted of two After noting the glass thickness, the laminate thick- float glass panes of 2.95 mm each and a 0.76 mm PVB ness and the mass, the samples were stored in a film. Three different adhesion levels were investi- climate chamber at a constant temperature of - 18 C gated. The samples were stored at temperatures of and a relative humidity of 25–30%rH for * 24 h. The - 24 C, - 18 C and above - 18 C (no details tests were carried out at room temperature. First the given) for about 2 h. The Pummel test was performed tin side was tested. The glass sample was then lightly on a sloping metal surface. This prevented the glass tapped. The visual evaluation was performed each fragments from remaining on the sample. The sample time directly on site using Kuraray’s reference was also slightly tilted away from the metal surface to images. In addition, the final weight was weighed achieve a better impact effect. The evaluation showed and a picture was taken for later evaluation. A camera that at temperatures higher than - 18 C, the inter- system is required to ensure identical recording layer became too elastic, so that the fragments were conditions for the sample pictures after the test has pressed into the foil by the hammer blows and did not been carried out. This consists of a box (Fig. 4a) made detach from the foil. of cardboard with black interior walls, and two opposite open sides. On one open side a light source is positioned at a defined distance. The selected 3 Experimental investigation setting is retained for all pictures. Like the box, the light construction is made of cardboard, to which three In this section, the human influence during the LED strips have been attached. The sample can be pummeling process is examined. Three different test placed in the box through the other open side. The persons were used and the results were compared to Pummelled sample is placed in the same position after those of the experiments with a Pummel apparatus. each test. A hole has been cut in the upper side of the box to allow the camera to be placed on the box and 3.1 Samples and performed tests take pictures from the same position. Before the air side was tested, the sample was placed back in the The composition of the LSG samples was 3 mm Float climate chamber for another 15 min. Again, this side Glass / 0.76 mm PVB / 3 mm Float Glass. was also visually evaluated, weighed and pho- 300 9 300 mm large laminates were created, which tographed in the camera device. An infrared Fig. 4 a Sketch of the camera device, b sketch of the impact pattern, c modified hammer including a ring load cell (Nguyen 2019) 123 Investigations on the execution and evaluation of the Pummel test 379 thermometer was used to record the temperature of the 45, on which the glass plate can be fixed, so that it is at glass surface before and after the experiment. a flat angle of about 5 to the inclined metal plate. The The human influence was investigated by having upper part of the sample is clamped, while the lower three different people carry out the experiment. part leans on the inclined plate. A threaded rod on the Further test series were carried out with the Pummel clamp makes it possible to move the sample along the apparatus shown in Fig. 5, so that comparisons could inclined plate. The hammer strikes the sample from be made between the manual and semi-automatic above. It is adjusted so that it hits the glass plate at a Pummel test. The same hammer was used for all tests. right angle. The maximum drop height corresponds to In preliminary tests, a fitter’s hammer with a hammer the vertical alignment of the hammer handle (75 cm head weight of 800 g was selected. In all manual tests between centre axis of the hammer head and impact and part of the tests with the Pummel apparatus the surface on the glass plate). Furthermore, the position applied force was measured. For this purpose the of the hammer can be moved horizontally along the hammer was equipped with a ring load cell (Fig. 4c). axis. The hammer mounting is designed so that The Pummel apparatus consists of a steel frame. A different hammers can be used. metal plate is welded to the steel frame at an angle of Fig. 5 a Sketch of the Pummel apparatus, b Pummel apparatus (Nguyen 2019) Fig. 6 Applied forces while Pummeling with the Pummel apparatus (1 row) and during manual Pummeling (row 2–4), (Nguyen 2019) 123 380 M. Schuster et al. Table 3 Comparison of different tests Testperson 1 Testperson 2 Testperson 3 Number of rows Test 1 Test 2 Test 3 Test 1 Test 2 Test 3 Test 1 Test 2 Test 3 Sn 36 7 Sn 5 44 Sn 75 5 A 5 6 7 A4 5 5 A7 6 6 Total number of hammer blows Sn 26 50 60 Sn 46 40 60 Sn 54 50 44 A 55 60 57 A 70 79 85 A 79 65 62 The bold elements are those represented in Fig. 6 3.2 Test results sample from the same height of fall. This is probably unavoidable due to the overlapping blows and the Figure 6 (rows 2–4) shows the applied forces during associated changing nature of the impact surface. the manual Pummel procedure of three different test To finally evaluate the results, assign pummel persons using the example of one sample at a time. A values and investigate whether the manually induced peak grouping corresponds to the Pummeling of one variations during the pummeling effect the pummel row. The sections in between show the time it took to result, a suitable automated evaluation method and a move the sample down. uniform Pummel scaling rate must first be established. As can be seen from the force–time curves and from For this purpose, Sect. 4 will first present general Table 3, the three test persons performed the Pummel methods of digital image evaluation for pummel rating test differently. The required rows vary between 3 and and apply them to the Eastman reference pictures. 7, the required strokes between 26 and 57. Also the Furthermore, approaches to reduce external influences amount of the forces differs from person to person, such as different exposure situations and image especially when looking at the variations between the resolutions are presented in Sect. 5. rows. This leads to different visual impressions of the texture, see Fig. 7. Figure 6 (row 1) shows the applied forces during 4 Evaluation methods the semi-automatic Pummel procedure with the Pum- mel apparatus. Here, the number of rows was fixed to 7 4.1 General and the number of hammer blows to 56 (8 per row). Compared to Fig. 6 (rows 2–4) we can see that the In order to eliminate the human influence on the Pummeling process, as expected, was performed more evaluation of the test samples, automated image evenly regarding the variances between the individual analysis methods have been created in MATLAB. rows. It can also be seen that the forces at the Three different evaluation types were examined: beginning and end of a row are lower than in the middle of the row, although the hammer hits the Fig. 7 Results for different Testpersons 123 Investigations on the execution and evaluation of the Pummel test 381 • Method A Analysis of the free film surface using RGB images are converted to greyscale images by Binary Images, eliminating the colour tone and saturation information • Method B First order statistical evaluation of the while retaining the luminance. The luminance is greyscale images and calculated with a weighted sum of the three basic • Method C Texture analysis using co-occurrence colours: 0.299 9 Red ? 0.587 9 Green ? 0.114 9 matrices and second order statistics Blue. A threshold value is required to convert a greyscale image into a binary image. This defines up to First a short insight into image parameters and which grey value a pixel is still considered black or digital image processing is given. from which grey value a pixel is still considered white. A digital image consists of different picture The threshold value can be set either manually or elements, called pixels, and is represented via a two- automatically. dimensional matrix. The dimensions of this matrix The easiest and most intuitive image features for correspond to the image size. Each element of the image analysis are computed from histograms of the matrix is occupied with a finite, discrete quantile of greyscales in the image, which describes the statistical numeric representation for its intensity in a certain occurrence of the different greyscales. For this colour map. purpose, the number of pixels with the same grey Binary Images only consist of two quantization value is counted and stored in a vector. Often this levels 0 and 1, which usually represent black and vector is displayed as a bar chart whose X-axis white. Greyscale Images are single-channel images indicates the intensity and whose Y-axis represents the and are displayed with positive integer numbers. The number of pixels with exactly this intensity. From the value range is determined by the bit-number (colour intensity diagram of an image, first order statistics can depth) of the image. A typical greyscale image uses be computed. This includes, for example, the calcu- k = 8 bits, which means that a pixel in the image can lation of the mean grey value, the median grey value, k 8 take 2 =2 = 256 different intensity values (Table 4). value of average contrast expressed with the standard The lowest intensity level corresponds to black (0) and deviation, variance of grey values, skewness of the the highest level to white (255). Colour images are histogram, etc. These features are based on the values multi-channel images. Usually they are created in the of individual pixels, but neglect the relative position of primary colours red, green, blue (RGB model). Each pixels with respect to each other and hence neglect of these colour components consists of usually 8 bits. what is called the image texture in this paper (Lofstedt Similar to the 8bit greyscale images, the value range of et al. 2019). Figure 8 shows two images A and B with the components lies between [0,…, 255]. Through the same number of quantization levels (1 = black, additive colour mixing of the primary colours, a 4 = white). Although our eye obviously perceives two variety of colours and tones can be displayed. Besides different images, both images result in the same the RGB model there are also other colour maps, such histogram. as CMYK colour model (Cyan-Magenta-Yellow- To analyze the image texture, higher-order statis- Black). tics are required. These consider the interaction between two or more pixels at a time. Methods that make this possible are, for example, the Local Binary Table 4 Number of quantization levels Pattern method, the use of Gabor filters or the creation k (bit) Number of quantization levels and evaluation of Grey-Level-Co-Occurrence-Matri- ces (GLCM). GLCM have successfully been applied 12 =2 in medical analysis, e.g. X-ray mammography, brain 22 =4 cancer, in fabric defect detection, in image classifica- 32 =8 tion of satellite images, in face recognition and many 42 =16 others, which is why this method is investigated 82 = 256 further in this paper (Chan et al. 1995; Eleyan and k2 Demirel 2011; Elshinawy et al. 2011; Gomez et al. 2012; Haralick et al. 1973; Lofstedt et al. 2019; Raheja 123 382 M. Schuster et al. Fig. 8 Two different greyscale images with the same histogram and identical first order statistic values et al. 2013; Ulaby et al. 1986; Zayed and Elnemr matrix can be described with the Haralick parameters 2015). (Lofstedt et al. 2019; Talibi Alaoui and Sbihi 2013). The GLCM is a square matrix which size N 9 N is The main Haralick features are (Haralick et al. specified by the number of grey-levels N in the 1973; Zayed and Elnemr 2015): quantized image. The matrix elements GLCM[i,j] indi- cate how often a pixel with the grey value i [ [0:N] 4.1.1 Contrast and a pixel with the grey value j [ [0:N] occur in a spatial relationship in the image. The spatial relation- The contrast, also known as variance or inertia, is a ship, also called offset, defines the direction and measure of intensity variations between a pixel and its distance between the pixel of interest and its neigh- neighbour over the whole image. The higher the contrast, the more the entries of the normalized GLCM bour. For the right immediate neighbour, the direction corresponds to 0 and the distance to 1 which corre- move away from the matrix diagonal. The minimum sponds to a offset vector of (0,1). It is also possible to value is 0, which is obtained for a constant image. investigate more distant neighbours. The considera- piðÞ ; j jj i  j ð1Þ tion of a single direction leads to a GLCM sensitive to i;j rotation. For eight immediate neighbours (dis- tance = 1) of a pixel, eight offset vectors and eight corresponding GLCMs may be created, and fused see 4.1.2 Correlation Fig. 9. Considering offset vectors with opposite directions equals to an evaluation of pixel relations The correlation is a measure of how correlated a pixel in e.g. horizontal or vertical direction instead of right is to its neighbour over the whole image. The range is or left resp. up or down and lead to symmetrical [- 1:1], where 1 equals a perfectly positively corre- GLCMs. In order to combine different offsets, the lation and - 1 equals a perfectly negative correlation. GLCMs are normalized by dividing each matrix For a constant image no correlation can be computed. element with the sum of all elements. The normalized ðÞ i  l j  l x y elements can be considered as the joint probability piðÞ ; j  ð2Þ r r x y occurrence p(i,j) of the considered pixel pair with the i;j greyscales i and j. The final, fused, GLCM is obtained l = mean value of the partial probability function in by generating the mean values of the joint occurrence x-direction (column); r = standard deviation of the probabilities of the considered offsets. This final partial probability function in x-direction (column); 123 Investigations on the execution and evaluation of the Pummel test 383 Fig. 9 Creation of GLCM for the eight immediate neighbours rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P P 2 4.1.3 Energy l ¼ i  piðÞ ; j and r ¼ ðÞ i  l piðÞ ; j ; l x x y i;j i;j The energy, also known as uniformity or angular and r are calculated analogously by substituting j for second moment, equals the sum of the squared i. For symmetric GLCMs: l = l and r = r . i j i j elements in the normalized GLCM. It is a measure 123 384 M. Schuster et al. Fig. 10 Determination of GLCM and Haralick features, example 1 of texture roughness. When pixels have similar combined, so that only 4 single GLCM are shown. intensity, the energy is high. The range is [0:1]. The GLCMs can also be displayed as a 2-dimensional energy of a constant image equals 1. histogram by showing the matrix elements in the grey value corresponding to the matrix entry. Light grey piðÞ ; j ð3Þ values represent a high probability of occurrence i;j (upper limit: white, 1), dark ones a lower probability (lower limit: black, 0). If one compares the three simple examples with 4.1.4 Homogeneity pictures consisting exclusively of 2 grey values, the Haralick features can be explained very clearly. The homogeneity is a measure of how close the elements of the normalized GLCM are to its diagonal. In the case of the constant image (Fig. 10), both the fused GLCM and the four individual normalized Typically, homogeneity increases with decreasing contrast. Homogeneity has a range of [0:1]. For a GLCMs are identical. This is because the image is independent of rotation. Only the element GLCM(1,1) diagonal GLCM, the homogeneity equals 1. is occupied. The value is 1, which means that the piðÞ ; j probability that the neighbour has the same grey value ð3Þ 1 þjj i  j i;j as the pixel under consideration is 100%. The second image (Fig. 11) consists of two halves, In order to clarify the significance of the individual with the upper half consisting of black pixels and the features, simple artificially created greyscale images lower half of white pixels. If only the horizontal will be examined as examples (Figs. 10, 11, 12). In neighbours are considered, a GLCM is created where these pictures the opposite directions are already only the diagonal is occupied. The probability that 2 123 Investigations on the execution and evaluation of the Pummel test 385 Fig. 11 Determination of GLCM and Haralick features, example 2 black pixels are adjacent is just as high as the value 0.5. The contrast is fully pronounced. Two probability that 2 white neighbours are horizontally horizontally or vertically adjacent pixels force the adjacent (0.5 and 0.5). The probability that a black correlation - 1. The energy and the homogeneity are pixel is horizontally adjacent to a white pixel (and vice in the middle range. In the diagonal direction (left or versa) is 0. The 2D histogram therefore appears on the right) there are always pixels with the same colour, diagonal in a medium grey tone, while the remaining which means that diagonal neighbouring pixels have a elements are displayed in black. The contrast in the perfect correlation. Only the diagonals of the GLCMs horizontal direction is therefore 0 (non-existent), the are occupied here. The contrast is 0 in this case, while homogeneity 1 (fully developed). Horizontally adja- the homogeneity is fully developed with the value 1. cent pixels corrode to 100% and the energy is in the The combined GLCM only has matrix entries of the middle range. Looking at the remaining three direc- same size, so the 2D histogram is constant. This means tions, the GLCMs are identical. This means, for that it is equally likely that one of the eight pixel example, that it is equally likely that the pixel pair neighbours has the same or a different pixel colour. (1,1) will occur in diagonal or vertical direction, If one now compares the fused GLCMs of the three namely 33.333%. examples, the contrast increases, homogeneity, corre- The repetitive pattern in Fig. 12 best illustrates the lation and energy decrease. Haralick features. In both the horizontal and vertical The GLCMs and Haralick parameters of the images directions, the probability that the neighbouring pixel shown in Fig. 13 show clear differences. At this point will differ in colour from the pixel under consideration it should be remembered that these images were not is 100%. Thus the diagonals of the GLSM are empty distinguishable when looking at the 1st order statistics and the remaining elements are each assigned the (see Fig. 8). In the upper image, the contrast between 123 386 M. Schuster et al. 12 Contrast 0.5 1 2 1 2 Correlation 0 10.25 0.25 2 1 2 1 Energy 0.25 20.25 0.25 1 2 1 2 Homogeneity 0.75 2 1 2 1 Pixel pair Normalized GLCM - 2D histogramm Haralick Features Distance 1 Contrast 1 10 0.5 Direction Correlation -1 20.5 0 0&180° Energy 0.5 Homogeneity 0.5 Contrast 0 Distance 1 10.4444 0 Direction Correlation 1 200.5556 45°&225° Energy 0.5062 Homogeneity 1 Distance 1 Contrast 1 10 0.5 Direction Correlation -1 90°&270° 20.5 0 Energy 0.5 Homogeneity 0.5 Distance 1 Contrast 0 10.5556 0 Direction Correlation 1 20 0.4444 135°&315° Energy 0.5062 Homogeneity 1 Fig. 12 Determination of GLCM and Haralick features, example 3e Fig. 13 fused and normalized GLCM and Haralick features for two images having the same histogram directly adjacent pixels is significantly higher, while high potential of this texture recognition tool. In the homogeneity and energy are lower than in the relation to the evaluation of Pummel images, the lower image. In the upper image the correlation even texture of the pulverized glass is also evaluated (coarse takes on a negative value. This example shows the for low Pummel values, fine for high Pummel values). 123 Investigations on the execution and evaluation of the Pummel test 387 This makes the evaluation more independent of the pixels should be determined. By comparing the sample lighting conditions. with the reference images (e.g. by calculating a In the following sections the three evaluation correlation factor), a Pummel value is finally assigned. methods A (Analysis of Binary Images), B (First However, the determination of a threshold value for Order Statistical Evaluation of the Greyscale Images) conversion into a binary image has proven to be and C (Texture Analysis using Co-Occurrence Matri- problematic. In order to keep the human influence as ces and Second Order Statistics) will be tested on the low as possible, it was decided to use the Otsu Eastman reference Pummel pictures. The procedure is algorithm (Otsu 1979; Gonzalez et al. 2009)to first briefly explained using two different Pummel determine the threshold value. The pixels of the pictures (Pummel 1 and Pummel 7). In addition, the greyscale image are divided into two groups in such a results of the three methods are shown for all 9 way that the variance within the group is as small as reference pictures. The aim is to find an evaluation possible and at the same time the variance between the methodology that finds clear differences in the nine groups is as high as possible. Otsu’s method exhibits reference pictures and thus allows a later automated the relatively good performance if the histogram can assignment into a Pummel class. be assumed to have bimodal distribution and assumed to possess a deep and sharp valley between two peaks. 4.2 Method A: analysis of binary images If one looks at the greyscale histogram of the two reference Pummel images (Pummel 1 and 7, Fig. 14), The first idea for automated evaluation is based on the it becomes clear that there is no bimodality, which can evaluation method described in the patents (see lead to errors in the automated threshold determina- Sect. 2.1). As already described above, the assignment tion. If one looks at the automatically determined of a Pummel value is usually done via the percentage thresholds of all nine reference pictures, three different of free film area. areas can be identified: the first area comprises Accordingly, both the image of the Pummelled Pummel 1–3 and has comparatively low thresholds. sample and the reference images should first be For Pummel pictures 4–6 (area 2), identical, medium– converted into a binary black and white image and high threshold values were determined. The third area then the number of black (free foil) and white (glass) Fig. 14 Transformation of greyscale image into binary image for reference picture Pummel 1 and 7 123 388 M. Schuster et al. Table 5 Black/White pixels and threshold value according to Otsu Algorithm for the 9 reference pictures Pummel 123456789 Black part (%) 74.98 64.59 60.38 53.77 51.79 45.94 34.08 36.48 39.07 White part (%) 25.02 35.41 39.62 46.23 48.21 54.06 65.92 63.52 60.93 Normalized threshold value 0.26 0.28 0.31 0.34 0.34 0.34 0.41 0.44 0.49 Absolute threshold value 66 72 79 86 86 86 105 112 125 Fig. 15 Free film surface (%) and Otsu threshold value for the 9 reference pictures Table 6 Black/White pixels with a constant threshold value of 0.356 (mean of threshold values of Table 5) Pummel 1234567 89 Black part (%) 86.33 76.08 69.23 56.98 55.10 50.25 19.29 12.37 2.31 White part (%) 13.67 23.92 30.77 43.02 44.90 49.75 80.71 87.63 97.69 Normalized threshold value 0.3560 0.3560 0.3560 0.3560 0.3560 0.3560 0.3560 0.3560 0.3560 Absolute threshold value 91 91 91 91 91 91 91 91 91 comprises Pummel 7–9 with comparatively high obtained. For Pummel values greater than 7, the share threshold values. of black pixels increases again, which excludes a If the reference images are now converted into Pummel class assignment based on this criterion. binary images with the individual threshold values and Classification using the threshold value seems to make the proportions of black pixels are then determined more sense here. (Table 5, resp. Fig. 15), the desired result is not 123 Investigations on the execution and evaluation of the Pummel test 389 Fig. 16 Free film surface (%) for constant threshold value for the 9 reference pictures Fig. 17 Determination of cumulative grey value distribution for reference Pummel 1 and 7 If a constant threshold value is used for all nine compares the result with the values given in the reference images (in this case the mean value of the patents, Fig. 16 is obtained. individual threshold values determined according to Otsu was used, see Table 6a,) a different result is 4.3 Method B: first order statistical evaluation obtained. Here the percentage of black always of the greyscale images decreases as the Pummel value increases. However, it is also apparent here that the black components of In order to avoid the problem of finding a suit- the mean Pummel values are very close together, and able threshold value (see Method A), method B an allocation in this range can be difficult. If one evaluates the grey value images with 1st order 123 390 M. Schuster et al. Fig. 18 Empirical cumulative distribution functions of the reference greyscale images Table 7 Median and mean values, standard deviations and variances for all 9 Pummel references Pummel 1234567 8 9 Median value (–) 38.00 52.00 68.00 81.00 86.00 91.00 118.00 121.00 131.00 Mean value (–) 50.97 62.09 74.46 84.24 86.82 90.01 113.87 117.55 129.17 Standard deviation (–) 35.81 39.16 37.01 41.38 41.33 33.02 28.85 26.90 17.90 Variance (–) 1282.10 1533.80 1369.72 1712.14 1707.83 1090.07 832.59 723.63 320.36 statistics. For this purpose, the histograms of the grey- lower grey value. If the median and the mean value are value reference images are examined. This is exem- identical, the grey value distribution is symmetrical; if plarily shown in Fig. 17 for Pummel 1 and Pummel 7. they differ, the grey value distribution is The average value is a measure of the brightness of asymmetrical. the image, wherein a high mean value corresponds to The assignment criteria that are examined here with an overall brighter and a low mean value to an overall this method are: mean value, median and cumulative darker image impression. The standard deviation and distribution function. The cumulative distribution the square deviation (also called variance) are param- functions of all 9 Pummel reference pictures are eters that characterize the grey value distribution. The shown in Fig. 18. Here it is particularly noticeable that variance is a measure of the scattering of the grey the distribution functions of the middle Pummel values around the mean value and therefore provides values intersect (Pummel 4–Pummel 6). Mean values, information about the global contrast (high variance) median values, standard deviations and variances are and homogeneity (low variance). summarized in Table 7. Figure 19 shows the mean Another important characteristic of the grey value grey values and median values. The standard devia- distribution is the median. This is the grey value at tions are also shown in form of error bars around the which 50% of all pixels in the image have a higher mean values. grey value and 50% of all pixels in the image have a 123 Investigations on the execution and evaluation of the Pummel test 391 Fig. 19 Mean grey value and median grey value for the reference pictures Fig. 20 a Different illuminations of Pummel 7, b images after histogram equalization Overall it can be stated that the mean and median the same histograms and thus cannot be differentiated grey value increases with increasing Pummel value. from each other with 1st order statistics. In addition, methods A and B have the disadvantage that they are 4.4 Method C: texture analysis using co- very dependent on the lighting conditions under which occurrence matrices and second order the images were taken. statistics To minimize the influence of different illumina- tions, the grey tone histograms of the images can be The previously presented methods A and B are not to equalized before texture analysis. Histogram equal- be rejected in principle. However, in Sect. 4.1 a ization is a method that stretches the intensity range of decisive disadvantage of these two evaluation meth- the histogram to the entire intensity range and hence ods could be shown: different image patterns can have improves the contrast. The equalized histograms have 123 392 M. Schuster et al. Fig. 21 a Co-occurrence matrix Pummel 1, b co- occurrence matrix Pummel Fig. 22 Haralick features for all 9 reference pictures, pixel distance: 4, number of grey levels: 64 64 bins and are flat compared to the original picture Pummel 7, the greyscale image with subse- histograms. This is illustrated in Fig. 20. Figure 20a quently reduced brightness and the greyscale image shows the original greyscale image of the reference with subsequently increased brightness. In addition, 123 Investigations on the execution and evaluation of the Pummel test 393 the respective grey value histograms are shown. It is Figure 22 shows the four main Haralick features for obvious that the evaluation methods according to 1st all 9 equalized reference images. Except for the order statistics (derived from grey value histogram) energy, all features have a more or less monotonous are directly influenced by this. The original image, for course, so that the Pummel scale can be characterized example, has a mean grey value of 113.87 (see by a vector containing the three features Correlation, Table 7), the darkened image of 71.20 and the Contrast and Homogeneity. Furthermore, Fig. 22 illuminated image of 160.76. Figure 20b shows the illustrates, that identical results are obtained for those corresponding images after histogram equalization. three parameters, when using the equalized images of Figure 21 shows the combined GLCM (all eight different illuminations. Therefore, this method seems neighbours were considered) exemplary for Pummel 1 to be effective in terms of eliminating different and 7. Since 64 grey values are represented in an illumination conditions. equalized pummel image, the GLCMs are not dis- played in matrix form here, but only the 2D histograms are shown. 5 Evaluation of the test samples The distance factor between neighbours was set to 4. The value 4 resulted from a preliminary study, Especially when analyzing structures e.g. with Method where the correlation feature at variant distances was C, the image scale (how large is the imaged object in evaluated. When observing the direct neighbour cm and with how many pixels is it displayed) plays a (distance = 1) there were hardly any differences in decisive role. If one looks at the individual Pummel the correlation factor of the nine reference images. The reference images (Table 2), it is important to note that reason why the directly adjacent pixels always corre- the texture becomes finer as the pummel value late well is that the texture differences only occur with increases. As a result, an image of a lower pummel a larger pixel pitch. Only at a distance of 4 or more, value can be considered a zoom of the image of a differences became noticeable. Accordingly, the dis- higher pummel, which can lead to misinterpretations tance is dependent on the image resolution, which has when the image scale isn’t specified. According to the to be further investigated. manufacturer, the samples shown in the reference Fig. 23 First row: grey image sections, second row: binary pictures created with a constant threshold 0.356, third row: equalized histogram images 123 394 M. Schuster et al. images have the dimensions 15 9 15 cm, which Pummel Apparatus would be assigned to Pummel 8–9; corresponds to an image scale of 56 pixel/cm. For those pummeled by Testperson 1 and 3 to Pummel 8 and this reason the scale of the own pictures was adjusted. the sample pummeled by Testperson 2 to a pummel 7–8. 336 9 224 pixel image sections were then analyzed If the threshold value is constantly set to 0.356, all with Methods A, B and C. This corresponds to binary images consist of 23–28% black pixels. The evaluation areas of 6 9 4 cm. The corresponding lowest black ratio, which corresponds to the free film evaluation images are shown in Fig. 23. To minimize surface, occurs in the sample pummeled by Testperson differences in illumination, the equalized images were 3 and the highest black ratio occurs in the sample used for Method C, see Fig. 23, third row. pummeled by Testperson 2. According to Table 6 this Table 8 summarizes the results of all three evalu- corresponds to a Pummel value of 6–7 for all samples. ation methods for the samples shown in Fig. 7. It has already been shown that the evaluation 5.2 Method B: first order statistical evaluation methods according to 1st order statistics (Method A of greyscale images and B) depend very much on the illumination condi- tions of the image. Since it cannot be guaranteed that The median and mean grey values of all four samples these are the same for the pummelled sample and the are close together (between 107 and 113). The lowest reference images, the comparison of the results of the median and mean grey values occur in the sample self-pummelled samples with the pummel scales pummeled by Testperson 2, the highest in the sample according to Tables 5, 6 and 7 must be viewed very pummeled by Testperson 1. According to Table 7, all critically. Nevertheless, 1st order statistics can be used samples would be assigned to Pummel 6–7. to qualitatively compare the images in Fig. 7, as they were taken under the same lighting conditions. 5.3 Method C: texture analysis using co- occurrence matrices and second order 5.1 Method A: analysis of binary images statistics The threshold values determined by Otsu’s algorithm The results for the correlation, contrast and homo- are varying between 0.43 (Testperson 2) and 0.45 geneity factors of the test evaluation Method C, (Pummel Apparatus). If the evaluation scale according deliver Pummel 5–6 for the samples pummeled by the to Table 5 is used, the sample pummeled with the Pummel Apparatus and Testperson 1. For the sample Table 8 Results according to evaluation methods A, B and C for the samples shown in Fig. 7 Pummel apparatus Testperson 1 Testperson 2 Testperson 3 Method A: threshold: Otsu Threshold 0.45 0.44 0.43 0.44 Pummel 8–9 8 7–8 8 Method A: threshold: 0.356 Black part (%) 23.36 23.79 27.84 23.25 Pummel 6–7 6–7 6–7 6–7 Method B: Mean grey value 112.28 112.29 107.72 111.56 Median 111 113 107 111 Pummel 6–7 6–7 6–7 6–7 Method C: with equalized images Correlation 0.3097 0.3236 0.3540 0.2193 Contrast 470.22 462.14 440.87 534.06 Homogeneity 0.1384 0.1376 0.1448 0.1251 Pummel 5–6 5–6 4–5 6–7 123 Investigations on the execution and evaluation of the Pummel test 395 pummeled by Testperson 2, Pummel 4–5 and for the reduces the effect of different illuminations. The GLCM sample pummeled by Testperson 3, Pummel 6–7 is contains information about the spatial arrangement of achieved. neighbouring pixel intensities. Certain parameters (Har- It should be pointed out once again that until now, alick features) can be derived from the GLCM. In this only one set of reference pictures has been taken into paper the features contrast, correlation, energy and account. In order to get a uniform pummel scale, other homogeneity were investigated. Using a simple exam- reference images should be included in the study. ple, it could be shown that the images that could not be Nevertheless, it can be concluded that the self- distinguished according to statistical 1st order could now pummelled pictures hardly differ according to 1st be differentiated. Hence, in theory the texture analysis order statistics. However, if the texture is analysed, based on GLCM is the most promising for becoming one can see that the sample of Testperson 3 is the finest standardized. However, it is essential to define an image and that of Testperson 2 the coarsest. scale in order to get a uniformed reference pummel scale and a reliable assignment of the tests. Until now, the three evaluation methods were tested 6 Summary and outlook with only one set of reference images. To be able to make a general statement, further studies are still In this paper, the Pummel test was examined as an required, especially for texture analysis, which according to theory is very promising. Here, especially adhesion test. For this purpose, a research on the state of the art of performance and evaluation was presented the definition of a required image scale (number of first. Since neither the execution nor the evaluation are pixels/depicted sample section in cm) seems indis- standardized, the test results are very objective. pensable for the creation of a uniformed reference The first part of the paper deals with the execution pummel scale and a subsequent reliable assignment of of experiments. Since manual tests with a hammer as the test samples. For this purpose it is useful to use well as semi-automatic or even fully automatic reference images from different manufacturers and Pummel machines are used in industry to perform compare the results with each other. the Pummel test, this paper presented a test series in Acknowledgements Open Access funding provided by which the Pummel procedure of three different test Projekt DEAL. At this point we would like to thank the persons as well as of a Pummel apparatus was following institutions and persons for their valuable support in investigated. The hammer was equipped with a load this work: Marcel Ho ¨ rbert, laboratory employee of ISM ? D, cell. The force–time curves show that both the number who constructed the Pummel apparatus. Furthermore, we would like to thank the DIBt, which supports us in the project, and the of blows and the forces vary greatly from person to interlayer manufacturer Eastman, which provided us with the person. In contrast, more reproducible force–time test samples. curves could be achieved with the Pummeling appa- ratus. This also became apparent during the visual Compliance with ethical standards inspection of the sample structure. For the final Conflict of interest On behalf of all authors, the corre- evaluation of these experiments, however, an objec- sponding author states that there is no conflict of interest. tive evaluation method must first be found. Therefore typical image evaluation methods were Open Access This article is licensed under a Creative Com- mons Attribution 4.0 International License, which permits use, presented in the second part of this paper. Method A sharing, adaptation, distribution and reproduction in any med- and B are very easy to implement. They can be derived ium or format, as long as you give appropriate credit to the from the grey tone histogram. However, it could be original author(s) and the source, provide a link to the Creative shown that different images may have identical Commons licence, and indicate if changes were made. The images or other third party material in this article are included in histograms and therefore cannot be differentiated with the article’s Creative Commons licence, unless indicated these two methods. Furthermore, two images of the otherwise in a credit line to the material. If material is not same sample under different lighting conditions can included in the article’s Creative Commons licence and your lead to different results. intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly To avoid these problems, a texture analysis based on from the copyright holder. To view a copy of this licence, visit the grey value Co-occurrence matrix can be carried out. http://creativecommons.org/licenses/by/4.0/. It was shown that the use of equalized grey value images 123 396 M. Schuster et al. References Keller, U., Mortelmans, H.: Adhesion in laminated safety glass—what makes it work? gloss processing days. In: The Sixth International Conference on Architectural and Beckmann, R., Knackstedt, W.: Process for the production of Auromotive Glass, Today and in the 21st Century, Tam- modified partially acetaliziedpolyvinyl alcohol films— pere, Finland, 13.6.1999–16.6.1999 (1999) United States Patent US 4,144,376 (1977) Keller, U., Stenzel, H., Hoss, M.: PVB film for composite safety Chan, H.-P., Wei, D., Helvie, M., Sahiner, B., Adler, D., glass and composite safety glass. EP 1 470 182 B1 (2002) Goodsitt, M., Petrick, N.: Computer-aided classification of Kinloch, A., Lau, C.C., Williams, J.G.: The peeling of flexible mammographic masses and normal tissue: linear discrim- laminates. Int. J. Fract. 66(1), 45–70 (1994) inant analysis in texture feature space. Phys. Med. Biol. 40, Kuntsche, J.K.: Mechanisches Verhalten von Verbundglas unter 857–876 (1995). https://doi.org/10.1088/0031-9155/40/5/ zeitabha ¨ngiger Belastung und Explosionsbeanspruchung (Mechanical behaviour of laminated glass under time-de- DIN 52338: Test methods for flat glass in building—Ball drop pendent and explosion loading). Ph.D. Thesis, Technische test for laminated glass (2016) Universita ¨t Darmstadt (2015) Eastman: Architectural lamination guide (2013) Kuraray: Manual Verarbeitung von TROSIFOL PVB-Folie Eleyan, A., Demirel, H.: Co-occurrence matrix and its statistical (2012) features as a new approach for face recognition. Turk. Kuraray: Trosifol Pummel test standards (2014) J. Electr. Eng. Comput. Sci. 19, 97–107 (2011). https://doi. Lofstedt, T., Brynolfsson, P., Asklund, T., Nyholm, T., Garpe- org/10.3906/elk-0906-27 bring, A.: Grey-level invariant Haralick texture features. Elshinawy, M.Y., Badawy, A.-H.A., Abdelmageed, W.W., PLoS ONE 14(2), e0212110 (2019). https://doi.org/10. Chouikha, M.F.: Detection of normal mammograms based 1371/journal.pone.0212110 on breast tissue density using GLCM features. Paper pre- Nguyen, T.A.: Investigations on the Experimental Realisation sented at the biomedical engineering (2011) and Evaluation of the ‘‘Pummel test’’ as Adhesion Tests for EN ISO 12543-2 (2011) Glass in building—laminated glass and Laminated Safety Glass. TU Darmstadt, Darmstadt (2019) laminated safety glass—Part 2: laminated safety glass (ISO Otsu, N.: A threshold selection method from grey-level his- 12543-2:2011); German version (2011) tograms. IEEE Trans. Syst. Man Cybern. 9, 62–66 (1979). EN 356: Glass in building—Security glazing—Testing and https://doi.org/10.1109/tsmc.1979.4310076 classification of resistance against manual attack; German Papenfuhs, B., Steuer, M.: Plasticizer-containing Polyvinylbu- version (1999) tyrals, method for producing the same and the use thereof, EN 12600: Glass in building—Pendulum test—impact test especially for producings film for use in laminated safety method and classification for flat glass (2003) glasses—United States Patent US 6,984,679 B2 (2006) EN 14449: Laminated glass and laminated safety glass—eval- Raheja, J.L., Ajay, B., Chaudhary, A.: Real time fabric defect uation of conformity/Product standard (2005) detection system on an embedded DSP platform. Optik Ensslen, F.: Zum Tragverhalten von Verbund-Sicherheitsglas 124(21), 5280–5284 (2013). https://doi.org/10.1016/j.ijleo. unter Beru ¨ cksichtigung der Alterung der Polyvinylbutyral- 2013.03.038 Folie. Ph.D. thesis, Ruhr-Universita ¨t Bochum (2005) Schneider, J., Kuntsche, J., Schula, S., Schneider, F., Wo ¨ rner, J.- Everlam: Pummel standards D.: Glasbau Grundlagen, Berechnung, Konstruktion, vol. Fahlbusch, M.: Zur Ermittlung der Resttragfa ¨higkeit von Ver- 2. Springer, New York (2016) bundsicherheitsglas am Beispiel eines Glasbogens mit Scholz Campos, H.I.: Investigation on the Experimental Reali- Zugstab. TU Darmstadt, Darmstadt (2007) sation and Evaluation of Residual Load-Bearing Tests on Franz, J.: Investigation of the Residual Load-Bearing Behaviour Laminated Safety Glass. TU Darmstadt, Darmstadt (2019) of Fractured Glazing. TU Darmstadt, Darmstadt (2015) Stu ¨ we, M.: Arbeitsanweisung—SentryGlas Laminate Pummel Garrison, E.E.: Glass Laminate—United States Patent Office, Test. Du Pont De Nemours (Deutschland) GMBH—Glass US 3,434,915 (1965) Laminating Solutions (2007) Gomez, W., Pereira, W.C., Infantosi, A.F.: Analysis of co-oc- Talibi Alaoui, M., Sbihi, A.: Texture Classification Based on currence texture statistics as a function of grey-level Co-occurrence Matrix and Neuro-Morphological quantization for classifying breast ultrasound. IEEE Trans. Approach. Image Analysis and Processing—ICIAP 2013, Med. Imaging 31(10), 1889–1899 (2012). https://doi.org/ pp. 510–521. Springer, Berlin (2013) 10.1109/TMI.2012.2206398 Ulaby, F.T., Kouyate, F., Brisco, B., Williams, T.H.L.: Textural Gonzalez, R.C., Woods, R.E., Eddins, S.L.: Digital Image information in SAR images. IEEE Trans. Geosci. Remote Processing Using MATLAB , vol. 2. United States (2009) Sens. GE-24(2), 235–245 (1986). https://doi.org/10.1109/ Haralick, R.M., Shanmugam, K., Dinstein, I.: Textural features tgrs.1986.289643 for image classification. IEEE Trans. Syst. Man Cybern. Zayed, N., Elnemr, H.A.: Statistical analysis of haralick texture SMC-3(6), 610–621 (1973). https://doi.org/10.1109/tsmc. features to discriminate lung abnormalities. Int. J. Biomed. 1973.4309314 Imaging 2015, 267807 (2015). https://doi.org/10.1155/ Hark, M.: Approach to Quantify the Pummel Test. TU Darm- 2015/267807 stadt, Darmstadt (2012) Hof, P., Oechsner, M.: Testing of adhesion on laminated glass using photometric measurements. Paper presented at the Publisher’s Note Springer Nature remains neutral with Glass Performance Days, Tampere, Finland, regard to jurisdictional claims in published maps and 28.06.2017–30.06.2017 institutional affiliations. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Glass Structures & Engineering Springer Journals

Investigations on the execution and evaluation of the Pummel test for polyvinyl butyral based interlayers

Loading next page...
 
/lp/springer-journals/investigations-on-the-execution-and-evaluation-of-the-pummel-test-for-0KcrdvDNWa
Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2020
ISSN
2363-5142
eISSN
2363-5150
DOI
10.1007/s40940-020-00120-y
Publisher site
See Article on Publisher Site

Abstract

Glass Struct Eng (2020) 5:371–396 https://doi.org/10.1007/s40940-020-00120-y(0123456789().,-volV)(0123456789().,-volV) SI: C HALLENGING GLASS Investigations on the execution and evaluation of the Pummel test for polyvinyl butyral based interlayers . . Miriam Schuster Jens Schneider Tuong An Nguyen Received: 1 February 2020 / Accepted: 7 April 2020 / Published online: 4 June 2020 The Author(s) 2020 Abstract Laminated safety glass (LSG) is increas- Three image evaluation methods in Matlab are ingly used as structural element in buildings. Of presented, discussed and compared: (1) analysis of central importance for safety are the adhesion and the binary images, (2) statistical evaluation of the grey- residual load-bearing capacity in the post fractured scale images and (3) texture analysis using co- state. In literature a large number of tests to assess occurrence matrices. adhesion is mentioned. These include, e.g. peel tests, through-cracked-tensile/-bending tests, VW-pull tests Keywords Laminated safety glass  Adhesion and compressive shear tests. However, especially in Pummeltest  Automatization  Digital image industry, the Pummel test is widespread for determin- evaluation ing the quality of adhesion in LSG with polyvinyl butyral based interlayers. This test method proves to be simple and quick to carry out: The laminate is stored at - 18 C and then completely destroyed at 1 Introduction room temperature with hammer blows. The adhesion level (0–10) is determined by visually comparing the 1.1 Laminated safety glass adhering glass fragments with reference pictures or with the help of diagrams and tables which indicate the The use of glass in the building industry has risen Pummel value as a function of the free film surface. sharply in recent years with the ongoing architectural Pummel value 0 is to be interpreted as no adhesion and trend towards transparent structures. In this context, Pummel value 10 as very high adhesion. Due to the the glass must increasingly meet safety-relevant lack of standardization, the execution and evaluation requirements with regard to breakage and post-break- is very much dependent on the test institution and age behaviour. The unannounced failure of the brittle executive person. This paper shows different Pummel material glass is countered by the use of laminated classifications that can currently be found on the safety glass (LSG). According to EN ISO 12543-2 market. Subsequently, approaches to the automatiza- (2011), laminated safety glass, consisting of two or tion and standardization of the execution and espe- more glass panes that are laminated together to form a cially the evaluation of the Pummel test are shown. package by means of polymer interlayers, must bind the resulting glass splinters after glass breakage and enable a residual load-bearing behaviour of the M. Schuster (&)  J. Schneider  T. A. Nguyen glazing. Accordingly, the adhesion between glass TU Darmstadt, Darmstadt, Germany e-mail: schuster@ismd.tu-darmstadt.de 123 372 M. Schuster et al. and film plays a decisive role. In laminated safety glass The bond between PVB and glass is characterized with low to no adhesion, the glass pane will delam- by dense reversible hydrogen bonds between the inate and after glass breakage the glass splinters do not polyvinyl alcohol in PVB and the silanol group in the stick on the interlayer. Contrary, a very high adhesion glass, see Fig. 1. Accordingly, the degree of adhesion results in a quasi-monolithic brittle failure of the can be controlled via the film composition, e.g. the laminate. The film adheres so strongly to the glass content of polyvinyl alcohol, the presence of adhesion- surface that elastic deformations are only possible to a regulating additives or the interlayer moisture content. limited extent and the impact energy cannot be In (Ensslen 2005) the optimum moisture content of dissipated. With an optimal degree of adhesion, the 0.45% ± 0.07% in the PVB-based film is given. This interlayer has a splinter-binding effect and can range remains largely constant at a relative humidity partially detach from the glass plate so that the impact of 30%. At a relative humidity between 50 and 70%, energy can be absorbed and the glass fragments will the water content in the film increases linearly. From a continue to adhere to the film. water content of 1%, the adhesive bond deteriorates Application situations in which laminated safety enormously and from 2% the film becomes milky glass is required are, for example, when the glass is cloudy. In addition to the film properties, other factors used for fall protection, can be walked on, is used as also have a considerable influence on adhesion. If the horizontal glazing with the traffic area underneath glass panes are not cleaned properly, grease and (overhead glazing) or as a structural component (e.g. limescale may be present on the surface and thus as a glass beam). The most common interlayer prevent hydrogen bonding. Kuraray (2012) recom- products in building industry are polyvinyl butyral mends distilled or fully desalinated washing water, ?? ?? (PVB)-based with a market share of over 90%. Other since alkaline earth ions (Ca and Mg ) reduce the interlayers include ethylene vinyl acetate (EVA) and glass adhesion even at low quantities enormously ionoplastics (SentryGlas ). (Keller et al. 2002; Kuraray 2012). The glass surface This article deals exclusively with laminated glass roughness also has an influence, since a rough where PVB interlayers are used. The adhesive microstructure favors mechanical interlocking. Since behaviour between glass and film will be discussed the tin side of glass has a smoother surface structure ?? in more detail in the following sections. and can present Sn ions, usually a slightly lower adhesion than the air side is observed. Finally, errors in 1.2 Adhesion the lamination process may result in air bubbles between the interlayer and glass, which prevent flat First the production process of laminated safety glass contact between glass and film and therefore lead to is briefly explained. The first step in the production of insufficient adhesion levels. laminated safety glass consists of cutting of the glass sheets to the desired size, edge processing and cleaning. The previously air-conditioned interlayers are then cut and the laminate is layered. This inevitably leads to air inclusions between the film and glass, which are mostly removed in the pre- lamination process using rollers in heating tunnels, where PVB and glass are heated to 35 C and 60 C and using a negative pressure of - 0.6 to - 0.9 bar. In addition, the edges are sealed in this process step to prevent air from penetrating the sandwich. After the pre-laminate production, the adhesive effect between film and glass is already given, but small air inclusions can remain. These are dissolved by the final laminate Fig. 1 Adhesion between glass and PVB, based on Kuntsche production in the autoclave at a temperature of (2015) approximately 135 C and a pressure of maximum 14 bar. 123 Investigations on the execution and evaluation of the Pummel test 373 1.3 Adhesion and post-fractured tests maximum shear force until breakage is a measure of adhesion (Kuraray 2012; Schneider et al. 2016). There are various test methods in the construction The adhesive tensile test, also known as the VW industry for quantifying the adhesion between glass Pull Test (Fig. 2c), was developed as an alternative to and polymer. Besides the ball drop test and the the compression shear test. The LSG sample is joined pendulum impact test, the tests are not standardized, so to a steel plate and then clamped into a universal that differences are possible in their performance testing machine. During the test procedure the plates (Schneider et al. 2016). In this section, the adhesion are separated at a constant speed at room temperature. tests are briefly explained, whereby the Pummel test is The force is applied until the glass plate is released dealt with in more detail with regard to the procedure from the foil. The adhesive tensile strength is calcu- and the advantages and disadvantages. lated via the technical/engineering breaking stress The ball drop test (Fig. 2a) is regulated in the (Franz 2015). In the adhesive shear test (Fig. 2d), German Standard DIN 52338 (2016) and the European laminated safety glass samples are subjected to a shear Standards EN 14449 (2005) and EN 356 (1999). In this test until the glass plate has detached itself from the test a defined steel ball falls from a certain height onto foil (Franz 2015). Compared to the adhesive shear test, the laminated safety glass. It is evaluated whether the the adhesive tensile test is much more complex. The ball penetrates the glass. Additionally the mass of the sample preparation with bonding is time-consuming detached glass splinters is evaluated. The pendulum and requires extreme accuracy (parallelism of both impact test (Fig. 2b) according to the European metal anvil plates). In addition, the glass panes must be Standard EN 12600 (2003) is an impact test similar separated from the steel plate after the test. For these to the ball drop test. Here, the double-tyre pendulum reasons the adhesive shear test is preferred. with different drop heights simulates the impact of a For the peel test (Fig. 2f), the test specimens cannot person at different speeds. The test is considered to be be taken from conventional production. A release film passed if none of the specimens breaks or breaks in a must be added during the production process so that non-hazardous manner. For the latter, the result is the interlayer only adheres to one glass pane. Alu- defined by the weight and size of the glass fragments. minum foil is laminated to the PVB film to prevent In the compression shear test (Fig. 2e), the adhe- high elongation when peeling the PVB film. The PVB sion resistance between glass and foil is determined at film should also protrude so that it can be clamped in room temperature in a universal testing machine, the tensile testing machine. The protruding film is where the sample, which is clamped at an angle of 45, pulled off the glass pane at a defined angle with a is loaded both in compression and in shear. The constant force level. The applied force is a measure for adhesion (Kuraray 2012; Schneider et al. 2016). Fig. 2 a Ball drop test (Schneider et al. 2016), b Pendulum (based on EN 12600 2003), c VW-pull test (Franz 2015), d shear test (Schneider et al. 2016), e CSS (Fahlbusch 2007), f Peel test (based on Kinloch et al. 1994), g, h TCT and TCB (Scholz Campos 2019) 123 374 M. Schuster et al. Disadvantages of the test method are the costly the subjective evaluation, which is judged by the production of the test specimens and the influence of naked eye of the examiner. Standardization of the the elongation of the PVB film and the aluminum film. execution procedure and evaluation would finally This test is used especially in the photovoltaic industry allow an objective statement. For this purpose, Sect. 2 for EVA films (Kuraray 2012). first presents the state of the art and research for For the Through-Cracked-Tensile Test (TCT, realization and evaluation. Subsequently, conducted Fig. 2g) and the Through-Cracked-Bending Test experiments are presented in Sect. 3 and automated (TCB, Fig. 2h), the glass panes of the laminated glass evaluation methods are proposed and discussed in sample are broken in a defined manner. The sample is Sect. 4. then subjected to a tensile test or a four point bending test respectively. The TCT test determines not only the delamination behaviour between glass and interlayer 2 Pummel test but also the material behaviour of the interlayer itself. Additionally there is the possibility to determine Although the Pummel test is widely used, there is no parameters for cohesive zone models, such as the standardization for its implementation and evaluation. energy release rate (Franz 2015; Schneider et al. In literature, however, there are different patent 2016). specifications, manufacturer guidelines and isolated research papers, which are briefly presented here. The Pummel test (further description in Sect. 2)is the most common test method for estimating the adhesion between glass and interlayer in production 2.1 Patent specifications facilities. After the sample has been brought to the desired temperature of - 18 C for several hours, it is The Pummel test was mentioned for the first time in placed on an inclined hard metal base. Then a certain the United States Patent US 3,434,915, which was glass surface is pulverized by means of hammer blows applied by Garrison (1965) in 1965 for laminated (manual, semi-automatic e.g. by a hydraulic or safety glass in car windscreens. According to this electrical controlled hammer that always strikes the patent, the test specimens (no dimensions specified) same spot so that the laminate has to be moved are first stored for 16 ± 4 h at a temperature of manually, or fully-automatic). One side of the sample approx. - 18 C. The cooled specimen is held against is Pummeled on the tin side (Sn) and one side of the an angled metal surface. The sample is inclined so that sample on the air/fire side (A/F). The evaluation is one side of the glass only touches the metal with the done visually with the help of reference pictures or edge. Then a flat-headed hammer is used to pulverize tables indicating percentages of free film surface, see the surface. The destroyed surface is at least 7.62 cm Sects. 2.1 and 2.2. The adhesion scale, also known as in diameter and the remaining glass particles should Pummel values, ranges from 0 (low adhesion) to 10 not be larger than 0.64 cm. Afterwards, the loose glass (high adhesion). Compared to the other test methods fragments are carefully removed by shaking gently presented above, the Pummel test has the advantage and the degree of adhesion is determined using a that it can be carried out without much effort. table containing Pummel values for different free film Furthermore, no complex and expensive testing surfaces, see Table 1. machines or sample preparations are required. For a In 1977 the United States Patent US 4,144,376 manual execution only a hammer and an angled metal (Beckmann and Knackstedt 1977) was registered. support are necessary. However, the lack of standards Here, a procedure to regulate the adhesion of a leads to enormous differences in implementation and plasticized, partially acetalized polyvinyl alcohol film evaluation. The test method depends strongly on the for required Pummel adhesion values is described. experience of the test person. In particular, the number PVB resin was processed with conventional methods and intensity of hammer blows can lead to different in newly determined quantities for laminated safety results, see Sect. 3. Furthermore, changes in the angle glass. The samples for the Pummel test had the of inclination between specimen and metal surface dimensions 150 9 300 mm and were cooled for 2–8 h may result in different fracture patterns (Hark 2012). at a temperature of - 18 C(± 0.5 C). The samples The different Pummel values result additionally from were then placed on a metal block angled at 45. The 123 Investigations on the execution and evaluation of the Pummel test 375 Table 1 Pummel Values in dependency of free film surface Free film surface (%) Free film surface (%) Free film surface (%) Pummel value US 3,434,915 and US 4,144,376 US 6,984,679 B2 EP 1 470 182 B1 – – Large areas of total detachment - 1 100 n.e. 100 0 95 n.e. 99 1 90 97 97 2 85 n.e. 92 3 60 83 83 4 40 n.e. 67 5 20 43 43 6 10 n.e. 20 7 5 n.e. 8 8 2 n.e. 1 9 0 n.e. 0 10 n.e. not evaluated Estimated values out of graph for selected Pummel values glass surface was treated with a flat-headed hammer based. If one compares the graph with Table 1, one until the glass was completely pulverized. The test sees differences in the Pummel value assignment. For area was 100 9 150 mm. The Pummel test was example, in US 3,434,915, a Pummel value of 2 performed and evaluated according to the US Patent corresponds to a free film surface of 90%, a Pummel 3,434,915. value of 4 corresponds to a free film surface of 60% The Pummel test is also described in the United and Pummel 6 corresponds to a free film surface of States Patent US 6,984,679 B2 (Papenfuhs and Steuer 20%, while in US 6,984,679 B2 Pummel 2 corre- 2006) and is used to check the level of adhesion. The sponds to a free foil area of * 97% Pummel 4 test specimens are each stored with 2 9 2mm float corresponds to a free foil area of * 83% and Pummel glass (100 9 300 mm) for 4 h at a temperature of 6 corresponds to a free foil area of * 43%. - 18 C. The laminate is placed on a 45 inclined The European Patent EP 1 470 182 B1 (Keller et al. support and destroyed with an automatic hammer. 2002) treats a plasticized film of partially acetalized Figure 3a shows the impact pattern and Fig. 3b the polyvinyl alcohols as an interlayer for laminated graph, on which the evaluation of the Pummel test is safety glass. Among other things, these laminates are Fig. 3 a Sketch of the impact pattern according to US 6,984,679 B2, b modified evaluation graph, based on US 6,984,679 B2 (Papenfuhs and Steuer 2006) 123 376 M. Schuster et al. subjected to the Pummel test described in the patent. temperature equilibrium. After the samples are con- The test specimens (80 9 300 mm) consist of two ditioned, the Pummeling procedure should start within 2.1 mm thick float glass panes and are stored for at seconds of removal from the freezer. It is recom- least 24 h in a cold chamber at a temperature of mended, that the sample is held at about a 5 angle to - 18 C(± 2 C). The specimens are then placed on the plane of the Pummel plate, so that only the edge of a steel base at an inclination of 5 and Pummeled the unbroken glass contacts the plate. The laminate evenly with a hammer head (500 g, round head). The should be struck progressively in about 10 mm impact intensity is 1 Nm per impact. To ensure that the increments along the bottom 15 mm of the laminate. glass plate is completely destroyed at every point all When the bottom edge has been completely pulver- the blows overlap by 75%. At least 6 cm (from the ized, the next 15 mm has to be pulverized in the same lower edge measured) is destroyed. After the sample is manner. This has to be repeated until at least 7–10 cm lightly tapped, a visual inspection is carried out at of the laminate has been Pummeled. Care is exercised room temperature. The Pummel values (0–10) are to insure that all smooth glass is pulverized. Loose assigned via the exposed foil areas. If a test specimen glass dust has to be removed by slightly brushing the shows large areas of total detachment, Pummel value Pummeled laminate. For the evaluation and Pummel - 1 can be assigned to it. The classification of the classification it is advised to place the laminate on a Pummel values as a function of the free film surface black background and to compare the sample to the corresponds to the graph from US 6,984,679 B2. Pummel adhesion references. Everlam has also provided the authors with a 2.2 Manufacturer guidelines brochure containing a description of the Pummel TM procedure as well as the evaluation for EVERLAM - Eastman provides a description of the Pummel test and PVB laminates. Here the test specimens (no dimen- its evaluation. Eastman has sent reference pictures and sions) have to be cooled down to a temperature of a description of the experiment to the authors. - 18 C for 3 h. The sample is then Pummeled Reference pictures of the Pummel classes 1–9 are manually with a flat-head hammer with a weight of given in Table 2. The description of the experiment 500 g on an anvil at 5. The blows are made with a can also be found in (Eastman 2013). It is recom- distance of 1.25 cm in a row and with 2 cm distance mended to put the laminated samples (no dimension between the rows. The destroyed laminate is visually given) in an - 18 C freezer for enough time (1–6 h inspected with the reference pictures (Table 2). Fur- depending on freezer) in order that they may reach thermore, an evaluation diagram based on the values Table 2 Reference pictures from different PVB manufacturers Pummel Pummel Pummel Pummel Pummel Pummel Pummel Pummel Pummel Pummel 1 2 3 4 5 6 7 8 9 10 First row: Eastman (sent via Email); second row: Kuraray (Pictures from poster, sent via E-mail); third row: Everlam (Pummel Standards sent via Email) 123 Investigations on the execution and evaluation of the Pummel test 377 of the US 4,144,376 and correlation diagrams of adhesion, the Pummel test and the compression shear Pummel values with CSS and ball drop performances test. In the Pummel test, the laminated glass sample are included. (no dimensions given) was cooled to - 18 C and In Kuraray’s manual dealing with the processing of Pummelled with a hammer weighing approx. 453 g on Trosifol PVB, the experimental execution is briefly a steel plate until both sides of the sample were described (Kuraray 2012). The sample dimensions are pulverized. The destroyed sample was then visually 80 9 200 mm (resp. 80 9 300 mm for windscreen evaluated according to the Pummel scale 0–10. By samples) and the float glass thickness should not comparing the results of the two test methods, it can be exceed 4 mm. The sample is stored for at least 2 h at a seen that these usually correlate almost linearly, if the temperature of - 18 C. The Pummeling process is condition is fulfilled that the test specimens are semi-automatic. The glass forms a small angle to the derived from the same batch. inclined metal base. According to the manual, a rough Franz (2015) evaluated in his dissertation the state distinction is made between high and medium glass of knowledge on fractured laminated safety glass and adhesion according to the classification in a diagram. the further development of the residual strength tests. In addition to the manual, Kuraray also sent us a poster Two test methods were used to assess the adhesive with reference pictures of Pummel values from 1 to 10 strength and quality. On the one hand, the Pummel test for Trosifol interlayers. Those are represented in to check the adhesive quality, on the other hand the Table 2. Additionally, a Pummel manual can be found adhesive shear test to verify the adhesive strength. The on the manufacturers website (Kuraray 2014). Regard- description of the execution of the adhesive shear test ing laminates with PVB, the Pummel standards is omitted here. The Pummel samples consisted of (Pummeling procedure, reference pictures, graph with 2 9 3 mm float glass and 1.52 mm PVB foil. The foil free film surface, correlations to CSS and ball drop had three different degrees of adhesion (BG R10, BG tests) correspond to those of Everlam. It should be R15 and BG R20, Trosifol (R) interlayers). The noted, that the reference pictures differ from the poster dimensions were 80 9 300 mm. The laminates were and the Pummel standard manual. stored at - 18 C for 2 h. Then the tin and fire sides There is also a working instruction for the Pummel were Pummelled with a hammer to determine the test on laminated safety glass with ionoplast (Sen- adhesion of both glass sides. For evaluation the tryGlas ). However, since this interlayer material is destroyed samples were visually compared with not part of the investigations presented, it will not be reference pictures and classified on a Pummel scale discussed in detail here. Further information can be from 0 to 10. Franz (2015) compared the Pummeling found in (Stu ¨ we 2007). results with the shear strengths resulting from the Looking at the classification based on the reference shear test. He showed that the adhesive stresses images of different manufacturers (Table 2), one can increase with increasing Pummel value. see that already the adhesion scale differs among the Hof and Oechsner (2017) investigated the adhesion film manufacturers. Kuraray compares the samples of laminated glass as a function of moisture content with Pummel values from 1 to 10, Eastman with the within the film with three different test methods. With values from 1 to 9 and Everlam only with the Pummel the help of a spectrophotometer they were able to values from 2 to 9. Reasons for further differences are determine the light transmittance at a specific wave- different lighting conditions and the different record- length (which is dependent on the interlayer moisture level) of several glass panes. Furthermore, the VW ing angle of the samples. Pull Test and the Pummel test were applied. The 2.3 Research samples had the dimensions 100 9 200 mm and a total thickness of about 8.5 mm. Both sides of the Pummel tests were carried out in several research sample were destroyed with a hammer. The results projects on the subject of adhesion and correlations showed that samples with a moisture content below with other adhesion tests were investigated. 0.67% led to a high Pummel value (Pummel 7–8) and Keller and Mortelmans (1999) investigated the samples with a moisture content above 0.67% led to adhesion mechanism between glass and interlayer. low Pummel values (Pummel \ 3). Two test methods were used to determine the 123 378 M. Schuster et al. In his bachelor’s thesis, Hark (2012) was mainly were then cut to 100 9 300 mm large samples by the concerned with the evaluation of the Pummel test. For manufacturer. According to the manufacturer, the this purpose he examined different procedures, such as samples should have a Pummel value of 7. weight change, change in the light transmission of the All samples were provided with an impact pattern laminated safety glass, profile analysis of the test consisting of 56 impact fields (7 rows of 8 impact specimen, maximum stress and maximum elongation fields, see Fig. 4b) before storage. In addition, the tin of Pummelled specimens. The test specimens had a and air side was detected with a UV lamp and marked. dimension of 25 9 7.5 cm. They consisted of two After noting the glass thickness, the laminate thick- float glass panes of 2.95 mm each and a 0.76 mm PVB ness and the mass, the samples were stored in a film. Three different adhesion levels were investi- climate chamber at a constant temperature of - 18 C gated. The samples were stored at temperatures of and a relative humidity of 25–30%rH for * 24 h. The - 24 C, - 18 C and above - 18 C (no details tests were carried out at room temperature. First the given) for about 2 h. The Pummel test was performed tin side was tested. The glass sample was then lightly on a sloping metal surface. This prevented the glass tapped. The visual evaluation was performed each fragments from remaining on the sample. The sample time directly on site using Kuraray’s reference was also slightly tilted away from the metal surface to images. In addition, the final weight was weighed achieve a better impact effect. The evaluation showed and a picture was taken for later evaluation. A camera that at temperatures higher than - 18 C, the inter- system is required to ensure identical recording layer became too elastic, so that the fragments were conditions for the sample pictures after the test has pressed into the foil by the hammer blows and did not been carried out. This consists of a box (Fig. 4a) made detach from the foil. of cardboard with black interior walls, and two opposite open sides. On one open side a light source is positioned at a defined distance. The selected 3 Experimental investigation setting is retained for all pictures. Like the box, the light construction is made of cardboard, to which three In this section, the human influence during the LED strips have been attached. The sample can be pummeling process is examined. Three different test placed in the box through the other open side. The persons were used and the results were compared to Pummelled sample is placed in the same position after those of the experiments with a Pummel apparatus. each test. A hole has been cut in the upper side of the box to allow the camera to be placed on the box and 3.1 Samples and performed tests take pictures from the same position. Before the air side was tested, the sample was placed back in the The composition of the LSG samples was 3 mm Float climate chamber for another 15 min. Again, this side Glass / 0.76 mm PVB / 3 mm Float Glass. was also visually evaluated, weighed and pho- 300 9 300 mm large laminates were created, which tographed in the camera device. An infrared Fig. 4 a Sketch of the camera device, b sketch of the impact pattern, c modified hammer including a ring load cell (Nguyen 2019) 123 Investigations on the execution and evaluation of the Pummel test 379 thermometer was used to record the temperature of the 45, on which the glass plate can be fixed, so that it is at glass surface before and after the experiment. a flat angle of about 5 to the inclined metal plate. The The human influence was investigated by having upper part of the sample is clamped, while the lower three different people carry out the experiment. part leans on the inclined plate. A threaded rod on the Further test series were carried out with the Pummel clamp makes it possible to move the sample along the apparatus shown in Fig. 5, so that comparisons could inclined plate. The hammer strikes the sample from be made between the manual and semi-automatic above. It is adjusted so that it hits the glass plate at a Pummel test. The same hammer was used for all tests. right angle. The maximum drop height corresponds to In preliminary tests, a fitter’s hammer with a hammer the vertical alignment of the hammer handle (75 cm head weight of 800 g was selected. In all manual tests between centre axis of the hammer head and impact and part of the tests with the Pummel apparatus the surface on the glass plate). Furthermore, the position applied force was measured. For this purpose the of the hammer can be moved horizontally along the hammer was equipped with a ring load cell (Fig. 4c). axis. The hammer mounting is designed so that The Pummel apparatus consists of a steel frame. A different hammers can be used. metal plate is welded to the steel frame at an angle of Fig. 5 a Sketch of the Pummel apparatus, b Pummel apparatus (Nguyen 2019) Fig. 6 Applied forces while Pummeling with the Pummel apparatus (1 row) and during manual Pummeling (row 2–4), (Nguyen 2019) 123 380 M. Schuster et al. Table 3 Comparison of different tests Testperson 1 Testperson 2 Testperson 3 Number of rows Test 1 Test 2 Test 3 Test 1 Test 2 Test 3 Test 1 Test 2 Test 3 Sn 36 7 Sn 5 44 Sn 75 5 A 5 6 7 A4 5 5 A7 6 6 Total number of hammer blows Sn 26 50 60 Sn 46 40 60 Sn 54 50 44 A 55 60 57 A 70 79 85 A 79 65 62 The bold elements are those represented in Fig. 6 3.2 Test results sample from the same height of fall. This is probably unavoidable due to the overlapping blows and the Figure 6 (rows 2–4) shows the applied forces during associated changing nature of the impact surface. the manual Pummel procedure of three different test To finally evaluate the results, assign pummel persons using the example of one sample at a time. A values and investigate whether the manually induced peak grouping corresponds to the Pummeling of one variations during the pummeling effect the pummel row. The sections in between show the time it took to result, a suitable automated evaluation method and a move the sample down. uniform Pummel scaling rate must first be established. As can be seen from the force–time curves and from For this purpose, Sect. 4 will first present general Table 3, the three test persons performed the Pummel methods of digital image evaluation for pummel rating test differently. The required rows vary between 3 and and apply them to the Eastman reference pictures. 7, the required strokes between 26 and 57. Also the Furthermore, approaches to reduce external influences amount of the forces differs from person to person, such as different exposure situations and image especially when looking at the variations between the resolutions are presented in Sect. 5. rows. This leads to different visual impressions of the texture, see Fig. 7. Figure 6 (row 1) shows the applied forces during 4 Evaluation methods the semi-automatic Pummel procedure with the Pum- mel apparatus. Here, the number of rows was fixed to 7 4.1 General and the number of hammer blows to 56 (8 per row). Compared to Fig. 6 (rows 2–4) we can see that the In order to eliminate the human influence on the Pummeling process, as expected, was performed more evaluation of the test samples, automated image evenly regarding the variances between the individual analysis methods have been created in MATLAB. rows. It can also be seen that the forces at the Three different evaluation types were examined: beginning and end of a row are lower than in the middle of the row, although the hammer hits the Fig. 7 Results for different Testpersons 123 Investigations on the execution and evaluation of the Pummel test 381 • Method A Analysis of the free film surface using RGB images are converted to greyscale images by Binary Images, eliminating the colour tone and saturation information • Method B First order statistical evaluation of the while retaining the luminance. The luminance is greyscale images and calculated with a weighted sum of the three basic • Method C Texture analysis using co-occurrence colours: 0.299 9 Red ? 0.587 9 Green ? 0.114 9 matrices and second order statistics Blue. A threshold value is required to convert a greyscale image into a binary image. This defines up to First a short insight into image parameters and which grey value a pixel is still considered black or digital image processing is given. from which grey value a pixel is still considered white. A digital image consists of different picture The threshold value can be set either manually or elements, called pixels, and is represented via a two- automatically. dimensional matrix. The dimensions of this matrix The easiest and most intuitive image features for correspond to the image size. Each element of the image analysis are computed from histograms of the matrix is occupied with a finite, discrete quantile of greyscales in the image, which describes the statistical numeric representation for its intensity in a certain occurrence of the different greyscales. For this colour map. purpose, the number of pixels with the same grey Binary Images only consist of two quantization value is counted and stored in a vector. Often this levels 0 and 1, which usually represent black and vector is displayed as a bar chart whose X-axis white. Greyscale Images are single-channel images indicates the intensity and whose Y-axis represents the and are displayed with positive integer numbers. The number of pixels with exactly this intensity. From the value range is determined by the bit-number (colour intensity diagram of an image, first order statistics can depth) of the image. A typical greyscale image uses be computed. This includes, for example, the calcu- k = 8 bits, which means that a pixel in the image can lation of the mean grey value, the median grey value, k 8 take 2 =2 = 256 different intensity values (Table 4). value of average contrast expressed with the standard The lowest intensity level corresponds to black (0) and deviation, variance of grey values, skewness of the the highest level to white (255). Colour images are histogram, etc. These features are based on the values multi-channel images. Usually they are created in the of individual pixels, but neglect the relative position of primary colours red, green, blue (RGB model). Each pixels with respect to each other and hence neglect of these colour components consists of usually 8 bits. what is called the image texture in this paper (Lofstedt Similar to the 8bit greyscale images, the value range of et al. 2019). Figure 8 shows two images A and B with the components lies between [0,…, 255]. Through the same number of quantization levels (1 = black, additive colour mixing of the primary colours, a 4 = white). Although our eye obviously perceives two variety of colours and tones can be displayed. Besides different images, both images result in the same the RGB model there are also other colour maps, such histogram. as CMYK colour model (Cyan-Magenta-Yellow- To analyze the image texture, higher-order statis- Black). tics are required. These consider the interaction between two or more pixels at a time. Methods that make this possible are, for example, the Local Binary Table 4 Number of quantization levels Pattern method, the use of Gabor filters or the creation k (bit) Number of quantization levels and evaluation of Grey-Level-Co-Occurrence-Matri- ces (GLCM). GLCM have successfully been applied 12 =2 in medical analysis, e.g. X-ray mammography, brain 22 =4 cancer, in fabric defect detection, in image classifica- 32 =8 tion of satellite images, in face recognition and many 42 =16 others, which is why this method is investigated 82 = 256 further in this paper (Chan et al. 1995; Eleyan and k2 Demirel 2011; Elshinawy et al. 2011; Gomez et al. 2012; Haralick et al. 1973; Lofstedt et al. 2019; Raheja 123 382 M. Schuster et al. Fig. 8 Two different greyscale images with the same histogram and identical first order statistic values et al. 2013; Ulaby et al. 1986; Zayed and Elnemr matrix can be described with the Haralick parameters 2015). (Lofstedt et al. 2019; Talibi Alaoui and Sbihi 2013). The GLCM is a square matrix which size N 9 N is The main Haralick features are (Haralick et al. specified by the number of grey-levels N in the 1973; Zayed and Elnemr 2015): quantized image. The matrix elements GLCM[i,j] indi- cate how often a pixel with the grey value i [ [0:N] 4.1.1 Contrast and a pixel with the grey value j [ [0:N] occur in a spatial relationship in the image. The spatial relation- The contrast, also known as variance or inertia, is a ship, also called offset, defines the direction and measure of intensity variations between a pixel and its distance between the pixel of interest and its neigh- neighbour over the whole image. The higher the contrast, the more the entries of the normalized GLCM bour. For the right immediate neighbour, the direction corresponds to 0 and the distance to 1 which corre- move away from the matrix diagonal. The minimum sponds to a offset vector of (0,1). It is also possible to value is 0, which is obtained for a constant image. investigate more distant neighbours. The considera- piðÞ ; j jj i  j ð1Þ tion of a single direction leads to a GLCM sensitive to i;j rotation. For eight immediate neighbours (dis- tance = 1) of a pixel, eight offset vectors and eight corresponding GLCMs may be created, and fused see 4.1.2 Correlation Fig. 9. Considering offset vectors with opposite directions equals to an evaluation of pixel relations The correlation is a measure of how correlated a pixel in e.g. horizontal or vertical direction instead of right is to its neighbour over the whole image. The range is or left resp. up or down and lead to symmetrical [- 1:1], where 1 equals a perfectly positively corre- GLCMs. In order to combine different offsets, the lation and - 1 equals a perfectly negative correlation. GLCMs are normalized by dividing each matrix For a constant image no correlation can be computed. element with the sum of all elements. The normalized ðÞ i  l j  l x y elements can be considered as the joint probability piðÞ ; j  ð2Þ r r x y occurrence p(i,j) of the considered pixel pair with the i;j greyscales i and j. The final, fused, GLCM is obtained l = mean value of the partial probability function in by generating the mean values of the joint occurrence x-direction (column); r = standard deviation of the probabilities of the considered offsets. This final partial probability function in x-direction (column); 123 Investigations on the execution and evaluation of the Pummel test 383 Fig. 9 Creation of GLCM for the eight immediate neighbours rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P P 2 4.1.3 Energy l ¼ i  piðÞ ; j and r ¼ ðÞ i  l piðÞ ; j ; l x x y i;j i;j The energy, also known as uniformity or angular and r are calculated analogously by substituting j for second moment, equals the sum of the squared i. For symmetric GLCMs: l = l and r = r . i j i j elements in the normalized GLCM. It is a measure 123 384 M. Schuster et al. Fig. 10 Determination of GLCM and Haralick features, example 1 of texture roughness. When pixels have similar combined, so that only 4 single GLCM are shown. intensity, the energy is high. The range is [0:1]. The GLCMs can also be displayed as a 2-dimensional energy of a constant image equals 1. histogram by showing the matrix elements in the grey value corresponding to the matrix entry. Light grey piðÞ ; j ð3Þ values represent a high probability of occurrence i;j (upper limit: white, 1), dark ones a lower probability (lower limit: black, 0). If one compares the three simple examples with 4.1.4 Homogeneity pictures consisting exclusively of 2 grey values, the Haralick features can be explained very clearly. The homogeneity is a measure of how close the elements of the normalized GLCM are to its diagonal. In the case of the constant image (Fig. 10), both the fused GLCM and the four individual normalized Typically, homogeneity increases with decreasing contrast. Homogeneity has a range of [0:1]. For a GLCMs are identical. This is because the image is independent of rotation. Only the element GLCM(1,1) diagonal GLCM, the homogeneity equals 1. is occupied. The value is 1, which means that the piðÞ ; j probability that the neighbour has the same grey value ð3Þ 1 þjj i  j i;j as the pixel under consideration is 100%. The second image (Fig. 11) consists of two halves, In order to clarify the significance of the individual with the upper half consisting of black pixels and the features, simple artificially created greyscale images lower half of white pixels. If only the horizontal will be examined as examples (Figs. 10, 11, 12). In neighbours are considered, a GLCM is created where these pictures the opposite directions are already only the diagonal is occupied. The probability that 2 123 Investigations on the execution and evaluation of the Pummel test 385 Fig. 11 Determination of GLCM and Haralick features, example 2 black pixels are adjacent is just as high as the value 0.5. The contrast is fully pronounced. Two probability that 2 white neighbours are horizontally horizontally or vertically adjacent pixels force the adjacent (0.5 and 0.5). The probability that a black correlation - 1. The energy and the homogeneity are pixel is horizontally adjacent to a white pixel (and vice in the middle range. In the diagonal direction (left or versa) is 0. The 2D histogram therefore appears on the right) there are always pixels with the same colour, diagonal in a medium grey tone, while the remaining which means that diagonal neighbouring pixels have a elements are displayed in black. The contrast in the perfect correlation. Only the diagonals of the GLCMs horizontal direction is therefore 0 (non-existent), the are occupied here. The contrast is 0 in this case, while homogeneity 1 (fully developed). Horizontally adja- the homogeneity is fully developed with the value 1. cent pixels corrode to 100% and the energy is in the The combined GLCM only has matrix entries of the middle range. Looking at the remaining three direc- same size, so the 2D histogram is constant. This means tions, the GLCMs are identical. This means, for that it is equally likely that one of the eight pixel example, that it is equally likely that the pixel pair neighbours has the same or a different pixel colour. (1,1) will occur in diagonal or vertical direction, If one now compares the fused GLCMs of the three namely 33.333%. examples, the contrast increases, homogeneity, corre- The repetitive pattern in Fig. 12 best illustrates the lation and energy decrease. Haralick features. In both the horizontal and vertical The GLCMs and Haralick parameters of the images directions, the probability that the neighbouring pixel shown in Fig. 13 show clear differences. At this point will differ in colour from the pixel under consideration it should be remembered that these images were not is 100%. Thus the diagonals of the GLSM are empty distinguishable when looking at the 1st order statistics and the remaining elements are each assigned the (see Fig. 8). In the upper image, the contrast between 123 386 M. Schuster et al. 12 Contrast 0.5 1 2 1 2 Correlation 0 10.25 0.25 2 1 2 1 Energy 0.25 20.25 0.25 1 2 1 2 Homogeneity 0.75 2 1 2 1 Pixel pair Normalized GLCM - 2D histogramm Haralick Features Distance 1 Contrast 1 10 0.5 Direction Correlation -1 20.5 0 0&180° Energy 0.5 Homogeneity 0.5 Contrast 0 Distance 1 10.4444 0 Direction Correlation 1 200.5556 45°&225° Energy 0.5062 Homogeneity 1 Distance 1 Contrast 1 10 0.5 Direction Correlation -1 90°&270° 20.5 0 Energy 0.5 Homogeneity 0.5 Distance 1 Contrast 0 10.5556 0 Direction Correlation 1 20 0.4444 135°&315° Energy 0.5062 Homogeneity 1 Fig. 12 Determination of GLCM and Haralick features, example 3e Fig. 13 fused and normalized GLCM and Haralick features for two images having the same histogram directly adjacent pixels is significantly higher, while high potential of this texture recognition tool. In the homogeneity and energy are lower than in the relation to the evaluation of Pummel images, the lower image. In the upper image the correlation even texture of the pulverized glass is also evaluated (coarse takes on a negative value. This example shows the for low Pummel values, fine for high Pummel values). 123 Investigations on the execution and evaluation of the Pummel test 387 This makes the evaluation more independent of the pixels should be determined. By comparing the sample lighting conditions. with the reference images (e.g. by calculating a In the following sections the three evaluation correlation factor), a Pummel value is finally assigned. methods A (Analysis of Binary Images), B (First However, the determination of a threshold value for Order Statistical Evaluation of the Greyscale Images) conversion into a binary image has proven to be and C (Texture Analysis using Co-Occurrence Matri- problematic. In order to keep the human influence as ces and Second Order Statistics) will be tested on the low as possible, it was decided to use the Otsu Eastman reference Pummel pictures. The procedure is algorithm (Otsu 1979; Gonzalez et al. 2009)to first briefly explained using two different Pummel determine the threshold value. The pixels of the pictures (Pummel 1 and Pummel 7). In addition, the greyscale image are divided into two groups in such a results of the three methods are shown for all 9 way that the variance within the group is as small as reference pictures. The aim is to find an evaluation possible and at the same time the variance between the methodology that finds clear differences in the nine groups is as high as possible. Otsu’s method exhibits reference pictures and thus allows a later automated the relatively good performance if the histogram can assignment into a Pummel class. be assumed to have bimodal distribution and assumed to possess a deep and sharp valley between two peaks. 4.2 Method A: analysis of binary images If one looks at the greyscale histogram of the two reference Pummel images (Pummel 1 and 7, Fig. 14), The first idea for automated evaluation is based on the it becomes clear that there is no bimodality, which can evaluation method described in the patents (see lead to errors in the automated threshold determina- Sect. 2.1). As already described above, the assignment tion. If one looks at the automatically determined of a Pummel value is usually done via the percentage thresholds of all nine reference pictures, three different of free film area. areas can be identified: the first area comprises Accordingly, both the image of the Pummelled Pummel 1–3 and has comparatively low thresholds. sample and the reference images should first be For Pummel pictures 4–6 (area 2), identical, medium– converted into a binary black and white image and high threshold values were determined. The third area then the number of black (free foil) and white (glass) Fig. 14 Transformation of greyscale image into binary image for reference picture Pummel 1 and 7 123 388 M. Schuster et al. Table 5 Black/White pixels and threshold value according to Otsu Algorithm for the 9 reference pictures Pummel 123456789 Black part (%) 74.98 64.59 60.38 53.77 51.79 45.94 34.08 36.48 39.07 White part (%) 25.02 35.41 39.62 46.23 48.21 54.06 65.92 63.52 60.93 Normalized threshold value 0.26 0.28 0.31 0.34 0.34 0.34 0.41 0.44 0.49 Absolute threshold value 66 72 79 86 86 86 105 112 125 Fig. 15 Free film surface (%) and Otsu threshold value for the 9 reference pictures Table 6 Black/White pixels with a constant threshold value of 0.356 (mean of threshold values of Table 5) Pummel 1234567 89 Black part (%) 86.33 76.08 69.23 56.98 55.10 50.25 19.29 12.37 2.31 White part (%) 13.67 23.92 30.77 43.02 44.90 49.75 80.71 87.63 97.69 Normalized threshold value 0.3560 0.3560 0.3560 0.3560 0.3560 0.3560 0.3560 0.3560 0.3560 Absolute threshold value 91 91 91 91 91 91 91 91 91 comprises Pummel 7–9 with comparatively high obtained. For Pummel values greater than 7, the share threshold values. of black pixels increases again, which excludes a If the reference images are now converted into Pummel class assignment based on this criterion. binary images with the individual threshold values and Classification using the threshold value seems to make the proportions of black pixels are then determined more sense here. (Table 5, resp. Fig. 15), the desired result is not 123 Investigations on the execution and evaluation of the Pummel test 389 Fig. 16 Free film surface (%) for constant threshold value for the 9 reference pictures Fig. 17 Determination of cumulative grey value distribution for reference Pummel 1 and 7 If a constant threshold value is used for all nine compares the result with the values given in the reference images (in this case the mean value of the patents, Fig. 16 is obtained. individual threshold values determined according to Otsu was used, see Table 6a,) a different result is 4.3 Method B: first order statistical evaluation obtained. Here the percentage of black always of the greyscale images decreases as the Pummel value increases. However, it is also apparent here that the black components of In order to avoid the problem of finding a suit- the mean Pummel values are very close together, and able threshold value (see Method A), method B an allocation in this range can be difficult. If one evaluates the grey value images with 1st order 123 390 M. Schuster et al. Fig. 18 Empirical cumulative distribution functions of the reference greyscale images Table 7 Median and mean values, standard deviations and variances for all 9 Pummel references Pummel 1234567 8 9 Median value (–) 38.00 52.00 68.00 81.00 86.00 91.00 118.00 121.00 131.00 Mean value (–) 50.97 62.09 74.46 84.24 86.82 90.01 113.87 117.55 129.17 Standard deviation (–) 35.81 39.16 37.01 41.38 41.33 33.02 28.85 26.90 17.90 Variance (–) 1282.10 1533.80 1369.72 1712.14 1707.83 1090.07 832.59 723.63 320.36 statistics. For this purpose, the histograms of the grey- lower grey value. If the median and the mean value are value reference images are examined. This is exem- identical, the grey value distribution is symmetrical; if plarily shown in Fig. 17 for Pummel 1 and Pummel 7. they differ, the grey value distribution is The average value is a measure of the brightness of asymmetrical. the image, wherein a high mean value corresponds to The assignment criteria that are examined here with an overall brighter and a low mean value to an overall this method are: mean value, median and cumulative darker image impression. The standard deviation and distribution function. The cumulative distribution the square deviation (also called variance) are param- functions of all 9 Pummel reference pictures are eters that characterize the grey value distribution. The shown in Fig. 18. Here it is particularly noticeable that variance is a measure of the scattering of the grey the distribution functions of the middle Pummel values around the mean value and therefore provides values intersect (Pummel 4–Pummel 6). Mean values, information about the global contrast (high variance) median values, standard deviations and variances are and homogeneity (low variance). summarized in Table 7. Figure 19 shows the mean Another important characteristic of the grey value grey values and median values. The standard devia- distribution is the median. This is the grey value at tions are also shown in form of error bars around the which 50% of all pixels in the image have a higher mean values. grey value and 50% of all pixels in the image have a 123 Investigations on the execution and evaluation of the Pummel test 391 Fig. 19 Mean grey value and median grey value for the reference pictures Fig. 20 a Different illuminations of Pummel 7, b images after histogram equalization Overall it can be stated that the mean and median the same histograms and thus cannot be differentiated grey value increases with increasing Pummel value. from each other with 1st order statistics. In addition, methods A and B have the disadvantage that they are 4.4 Method C: texture analysis using co- very dependent on the lighting conditions under which occurrence matrices and second order the images were taken. statistics To minimize the influence of different illumina- tions, the grey tone histograms of the images can be The previously presented methods A and B are not to equalized before texture analysis. Histogram equal- be rejected in principle. However, in Sect. 4.1 a ization is a method that stretches the intensity range of decisive disadvantage of these two evaluation meth- the histogram to the entire intensity range and hence ods could be shown: different image patterns can have improves the contrast. The equalized histograms have 123 392 M. Schuster et al. Fig. 21 a Co-occurrence matrix Pummel 1, b co- occurrence matrix Pummel Fig. 22 Haralick features for all 9 reference pictures, pixel distance: 4, number of grey levels: 64 64 bins and are flat compared to the original picture Pummel 7, the greyscale image with subse- histograms. This is illustrated in Fig. 20. Figure 20a quently reduced brightness and the greyscale image shows the original greyscale image of the reference with subsequently increased brightness. In addition, 123 Investigations on the execution and evaluation of the Pummel test 393 the respective grey value histograms are shown. It is Figure 22 shows the four main Haralick features for obvious that the evaluation methods according to 1st all 9 equalized reference images. Except for the order statistics (derived from grey value histogram) energy, all features have a more or less monotonous are directly influenced by this. The original image, for course, so that the Pummel scale can be characterized example, has a mean grey value of 113.87 (see by a vector containing the three features Correlation, Table 7), the darkened image of 71.20 and the Contrast and Homogeneity. Furthermore, Fig. 22 illuminated image of 160.76. Figure 20b shows the illustrates, that identical results are obtained for those corresponding images after histogram equalization. three parameters, when using the equalized images of Figure 21 shows the combined GLCM (all eight different illuminations. Therefore, this method seems neighbours were considered) exemplary for Pummel 1 to be effective in terms of eliminating different and 7. Since 64 grey values are represented in an illumination conditions. equalized pummel image, the GLCMs are not dis- played in matrix form here, but only the 2D histograms are shown. 5 Evaluation of the test samples The distance factor between neighbours was set to 4. The value 4 resulted from a preliminary study, Especially when analyzing structures e.g. with Method where the correlation feature at variant distances was C, the image scale (how large is the imaged object in evaluated. When observing the direct neighbour cm and with how many pixels is it displayed) plays a (distance = 1) there were hardly any differences in decisive role. If one looks at the individual Pummel the correlation factor of the nine reference images. The reference images (Table 2), it is important to note that reason why the directly adjacent pixels always corre- the texture becomes finer as the pummel value late well is that the texture differences only occur with increases. As a result, an image of a lower pummel a larger pixel pitch. Only at a distance of 4 or more, value can be considered a zoom of the image of a differences became noticeable. Accordingly, the dis- higher pummel, which can lead to misinterpretations tance is dependent on the image resolution, which has when the image scale isn’t specified. According to the to be further investigated. manufacturer, the samples shown in the reference Fig. 23 First row: grey image sections, second row: binary pictures created with a constant threshold 0.356, third row: equalized histogram images 123 394 M. Schuster et al. images have the dimensions 15 9 15 cm, which Pummel Apparatus would be assigned to Pummel 8–9; corresponds to an image scale of 56 pixel/cm. For those pummeled by Testperson 1 and 3 to Pummel 8 and this reason the scale of the own pictures was adjusted. the sample pummeled by Testperson 2 to a pummel 7–8. 336 9 224 pixel image sections were then analyzed If the threshold value is constantly set to 0.356, all with Methods A, B and C. This corresponds to binary images consist of 23–28% black pixels. The evaluation areas of 6 9 4 cm. The corresponding lowest black ratio, which corresponds to the free film evaluation images are shown in Fig. 23. To minimize surface, occurs in the sample pummeled by Testperson differences in illumination, the equalized images were 3 and the highest black ratio occurs in the sample used for Method C, see Fig. 23, third row. pummeled by Testperson 2. According to Table 6 this Table 8 summarizes the results of all three evalu- corresponds to a Pummel value of 6–7 for all samples. ation methods for the samples shown in Fig. 7. It has already been shown that the evaluation 5.2 Method B: first order statistical evaluation methods according to 1st order statistics (Method A of greyscale images and B) depend very much on the illumination condi- tions of the image. Since it cannot be guaranteed that The median and mean grey values of all four samples these are the same for the pummelled sample and the are close together (between 107 and 113). The lowest reference images, the comparison of the results of the median and mean grey values occur in the sample self-pummelled samples with the pummel scales pummeled by Testperson 2, the highest in the sample according to Tables 5, 6 and 7 must be viewed very pummeled by Testperson 1. According to Table 7, all critically. Nevertheless, 1st order statistics can be used samples would be assigned to Pummel 6–7. to qualitatively compare the images in Fig. 7, as they were taken under the same lighting conditions. 5.3 Method C: texture analysis using co- occurrence matrices and second order 5.1 Method A: analysis of binary images statistics The threshold values determined by Otsu’s algorithm The results for the correlation, contrast and homo- are varying between 0.43 (Testperson 2) and 0.45 geneity factors of the test evaluation Method C, (Pummel Apparatus). If the evaluation scale according deliver Pummel 5–6 for the samples pummeled by the to Table 5 is used, the sample pummeled with the Pummel Apparatus and Testperson 1. For the sample Table 8 Results according to evaluation methods A, B and C for the samples shown in Fig. 7 Pummel apparatus Testperson 1 Testperson 2 Testperson 3 Method A: threshold: Otsu Threshold 0.45 0.44 0.43 0.44 Pummel 8–9 8 7–8 8 Method A: threshold: 0.356 Black part (%) 23.36 23.79 27.84 23.25 Pummel 6–7 6–7 6–7 6–7 Method B: Mean grey value 112.28 112.29 107.72 111.56 Median 111 113 107 111 Pummel 6–7 6–7 6–7 6–7 Method C: with equalized images Correlation 0.3097 0.3236 0.3540 0.2193 Contrast 470.22 462.14 440.87 534.06 Homogeneity 0.1384 0.1376 0.1448 0.1251 Pummel 5–6 5–6 4–5 6–7 123 Investigations on the execution and evaluation of the Pummel test 395 pummeled by Testperson 2, Pummel 4–5 and for the reduces the effect of different illuminations. The GLCM sample pummeled by Testperson 3, Pummel 6–7 is contains information about the spatial arrangement of achieved. neighbouring pixel intensities. Certain parameters (Har- It should be pointed out once again that until now, alick features) can be derived from the GLCM. In this only one set of reference pictures has been taken into paper the features contrast, correlation, energy and account. In order to get a uniform pummel scale, other homogeneity were investigated. Using a simple exam- reference images should be included in the study. ple, it could be shown that the images that could not be Nevertheless, it can be concluded that the self- distinguished according to statistical 1st order could now pummelled pictures hardly differ according to 1st be differentiated. Hence, in theory the texture analysis order statistics. However, if the texture is analysed, based on GLCM is the most promising for becoming one can see that the sample of Testperson 3 is the finest standardized. However, it is essential to define an image and that of Testperson 2 the coarsest. scale in order to get a uniformed reference pummel scale and a reliable assignment of the tests. Until now, the three evaluation methods were tested 6 Summary and outlook with only one set of reference images. To be able to make a general statement, further studies are still In this paper, the Pummel test was examined as an required, especially for texture analysis, which according to theory is very promising. Here, especially adhesion test. For this purpose, a research on the state of the art of performance and evaluation was presented the definition of a required image scale (number of first. Since neither the execution nor the evaluation are pixels/depicted sample section in cm) seems indis- standardized, the test results are very objective. pensable for the creation of a uniformed reference The first part of the paper deals with the execution pummel scale and a subsequent reliable assignment of of experiments. Since manual tests with a hammer as the test samples. For this purpose it is useful to use well as semi-automatic or even fully automatic reference images from different manufacturers and Pummel machines are used in industry to perform compare the results with each other. the Pummel test, this paper presented a test series in Acknowledgements Open Access funding provided by which the Pummel procedure of three different test Projekt DEAL. At this point we would like to thank the persons as well as of a Pummel apparatus was following institutions and persons for their valuable support in investigated. The hammer was equipped with a load this work: Marcel Ho ¨ rbert, laboratory employee of ISM ? D, cell. The force–time curves show that both the number who constructed the Pummel apparatus. Furthermore, we would like to thank the DIBt, which supports us in the project, and the of blows and the forces vary greatly from person to interlayer manufacturer Eastman, which provided us with the person. In contrast, more reproducible force–time test samples. curves could be achieved with the Pummeling appa- ratus. This also became apparent during the visual Compliance with ethical standards inspection of the sample structure. For the final Conflict of interest On behalf of all authors, the corre- evaluation of these experiments, however, an objec- sponding author states that there is no conflict of interest. tive evaluation method must first be found. Therefore typical image evaluation methods were Open Access This article is licensed under a Creative Com- mons Attribution 4.0 International License, which permits use, presented in the second part of this paper. Method A sharing, adaptation, distribution and reproduction in any med- and B are very easy to implement. They can be derived ium or format, as long as you give appropriate credit to the from the grey tone histogram. However, it could be original author(s) and the source, provide a link to the Creative shown that different images may have identical Commons licence, and indicate if changes were made. The images or other third party material in this article are included in histograms and therefore cannot be differentiated with the article’s Creative Commons licence, unless indicated these two methods. Furthermore, two images of the otherwise in a credit line to the material. If material is not same sample under different lighting conditions can included in the article’s Creative Commons licence and your lead to different results. intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly To avoid these problems, a texture analysis based on from the copyright holder. To view a copy of this licence, visit the grey value Co-occurrence matrix can be carried out. http://creativecommons.org/licenses/by/4.0/. It was shown that the use of equalized grey value images 123 396 M. Schuster et al. References Keller, U., Mortelmans, H.: Adhesion in laminated safety glass—what makes it work? gloss processing days. In: The Sixth International Conference on Architectural and Beckmann, R., Knackstedt, W.: Process for the production of Auromotive Glass, Today and in the 21st Century, Tam- modified partially acetaliziedpolyvinyl alcohol films— pere, Finland, 13.6.1999–16.6.1999 (1999) United States Patent US 4,144,376 (1977) Keller, U., Stenzel, H., Hoss, M.: PVB film for composite safety Chan, H.-P., Wei, D., Helvie, M., Sahiner, B., Adler, D., glass and composite safety glass. EP 1 470 182 B1 (2002) Goodsitt, M., Petrick, N.: Computer-aided classification of Kinloch, A., Lau, C.C., Williams, J.G.: The peeling of flexible mammographic masses and normal tissue: linear discrim- laminates. Int. J. Fract. 66(1), 45–70 (1994) inant analysis in texture feature space. Phys. Med. Biol. 40, Kuntsche, J.K.: Mechanisches Verhalten von Verbundglas unter 857–876 (1995). https://doi.org/10.1088/0031-9155/40/5/ zeitabha ¨ngiger Belastung und Explosionsbeanspruchung (Mechanical behaviour of laminated glass under time-de- DIN 52338: Test methods for flat glass in building—Ball drop pendent and explosion loading). Ph.D. Thesis, Technische test for laminated glass (2016) Universita ¨t Darmstadt (2015) Eastman: Architectural lamination guide (2013) Kuraray: Manual Verarbeitung von TROSIFOL PVB-Folie Eleyan, A., Demirel, H.: Co-occurrence matrix and its statistical (2012) features as a new approach for face recognition. Turk. Kuraray: Trosifol Pummel test standards (2014) J. Electr. Eng. Comput. Sci. 19, 97–107 (2011). https://doi. Lofstedt, T., Brynolfsson, P., Asklund, T., Nyholm, T., Garpe- org/10.3906/elk-0906-27 bring, A.: Grey-level invariant Haralick texture features. Elshinawy, M.Y., Badawy, A.-H.A., Abdelmageed, W.W., PLoS ONE 14(2), e0212110 (2019). https://doi.org/10. Chouikha, M.F.: Detection of normal mammograms based 1371/journal.pone.0212110 on breast tissue density using GLCM features. Paper pre- Nguyen, T.A.: Investigations on the Experimental Realisation sented at the biomedical engineering (2011) and Evaluation of the ‘‘Pummel test’’ as Adhesion Tests for EN ISO 12543-2 (2011) Glass in building—laminated glass and Laminated Safety Glass. TU Darmstadt, Darmstadt (2019) laminated safety glass—Part 2: laminated safety glass (ISO Otsu, N.: A threshold selection method from grey-level his- 12543-2:2011); German version (2011) tograms. IEEE Trans. Syst. Man Cybern. 9, 62–66 (1979). EN 356: Glass in building—Security glazing—Testing and https://doi.org/10.1109/tsmc.1979.4310076 classification of resistance against manual attack; German Papenfuhs, B., Steuer, M.: Plasticizer-containing Polyvinylbu- version (1999) tyrals, method for producing the same and the use thereof, EN 12600: Glass in building—Pendulum test—impact test especially for producings film for use in laminated safety method and classification for flat glass (2003) glasses—United States Patent US 6,984,679 B2 (2006) EN 14449: Laminated glass and laminated safety glass—eval- Raheja, J.L., Ajay, B., Chaudhary, A.: Real time fabric defect uation of conformity/Product standard (2005) detection system on an embedded DSP platform. Optik Ensslen, F.: Zum Tragverhalten von Verbund-Sicherheitsglas 124(21), 5280–5284 (2013). https://doi.org/10.1016/j.ijleo. unter Beru ¨ cksichtigung der Alterung der Polyvinylbutyral- 2013.03.038 Folie. Ph.D. thesis, Ruhr-Universita ¨t Bochum (2005) Schneider, J., Kuntsche, J., Schula, S., Schneider, F., Wo ¨ rner, J.- Everlam: Pummel standards D.: Glasbau Grundlagen, Berechnung, Konstruktion, vol. Fahlbusch, M.: Zur Ermittlung der Resttragfa ¨higkeit von Ver- 2. Springer, New York (2016) bundsicherheitsglas am Beispiel eines Glasbogens mit Scholz Campos, H.I.: Investigation on the Experimental Reali- Zugstab. TU Darmstadt, Darmstadt (2007) sation and Evaluation of Residual Load-Bearing Tests on Franz, J.: Investigation of the Residual Load-Bearing Behaviour Laminated Safety Glass. TU Darmstadt, Darmstadt (2019) of Fractured Glazing. TU Darmstadt, Darmstadt (2015) Stu ¨ we, M.: Arbeitsanweisung—SentryGlas Laminate Pummel Garrison, E.E.: Glass Laminate—United States Patent Office, Test. Du Pont De Nemours (Deutschland) GMBH—Glass US 3,434,915 (1965) Laminating Solutions (2007) Gomez, W., Pereira, W.C., Infantosi, A.F.: Analysis of co-oc- Talibi Alaoui, M., Sbihi, A.: Texture Classification Based on currence texture statistics as a function of grey-level Co-occurrence Matrix and Neuro-Morphological quantization for classifying breast ultrasound. IEEE Trans. Approach. Image Analysis and Processing—ICIAP 2013, Med. Imaging 31(10), 1889–1899 (2012). https://doi.org/ pp. 510–521. Springer, Berlin (2013) 10.1109/TMI.2012.2206398 Ulaby, F.T., Kouyate, F., Brisco, B., Williams, T.H.L.: Textural Gonzalez, R.C., Woods, R.E., Eddins, S.L.: Digital Image information in SAR images. IEEE Trans. Geosci. Remote Processing Using MATLAB , vol. 2. United States (2009) Sens. GE-24(2), 235–245 (1986). https://doi.org/10.1109/ Haralick, R.M., Shanmugam, K., Dinstein, I.: Textural features tgrs.1986.289643 for image classification. IEEE Trans. Syst. Man Cybern. Zayed, N., Elnemr, H.A.: Statistical analysis of haralick texture SMC-3(6), 610–621 (1973). https://doi.org/10.1109/tsmc. features to discriminate lung abnormalities. Int. J. Biomed. 1973.4309314 Imaging 2015, 267807 (2015). https://doi.org/10.1155/ Hark, M.: Approach to Quantify the Pummel Test. TU Darm- 2015/267807 stadt, Darmstadt (2012) Hof, P., Oechsner, M.: Testing of adhesion on laminated glass using photometric measurements. Paper presented at the Publisher’s Note Springer Nature remains neutral with Glass Performance Days, Tampere, Finland, regard to jurisdictional claims in published maps and 28.06.2017–30.06.2017 institutional affiliations.

Journal

Glass Structures & EngineeringSpringer Journals

Published: Nov 4, 2020

There are no references for this article.