24th International Colloquium Tribology - January 2024 227 Parallel Wear Testing - an Update Can We Produce Enough Data to Enable AI in Tribology? Lais Lopes 1 , Dirk Drees 1 , Pedro Bai-o 1 , Emmanouil Georgiou 2 1 Falex Tribology NV, Rotselaar, Belgium 2 Hellenic Air-Force Academy, Faculty of Aerospace Studies, Dekelia Air Force Base, Athens, Greece * Corresponding author: ddrees@falex.eu 1. Introduction The durability of materials (wear resistance) or anti-wear efficiency of lubricants can be measured in many different ways, both non-standardized and standardized. Non-standard methods are used to be close to an application. However, such methods cannot be used to compare materials or lubricants systematically from one another. In the case of polymers, for instance, available standard methods are very limited, the major ones being ASTM D3702 and G137. Each method has their advantage, but both have one important drawback : there is a large variation in test results. This large variation is mostly the result of inhomogeneity in materials and wear mechanisms, and can not be reduced by better test methods. Dry wear of polymers is inherently a very stochastic property. Since it is not feasible, and even not useful to improve on the repeatability of the test methods - due to the lack of repeatability in the materials themselves, it is a better approach to consider multiple data points per materal, and to use trends and outlier statistics to evaluate wear resistance. Multiple data points can be easily produced by reversing the mind-set of the tribological experiment : instead of producing one very expensive test result with an equally expensive single station tribometer with many sensors, we aim to generate only wear data with a multi-station ‘wear generator’ with a single measurand, namely material wear loss. This enables the plotting of wear trends, rather than a single point with a very wide variation. The risk of misinterpreting outliers reduces, and an interpretation of wear evolution becomes more feasible. The ability to produce larger datasets economically, may also open possibilities for AI based evaluation of materials. 2. Experimental approach The wear generator used is a 10-station, constant load, constant speed system, comprised of a rotating shaft where 10 individual loading stations apply the same load to 10 test samples. The rotating shafts can be acquired in different materials, or with different coatings, but the base shafts are available at very moderate costs. This reduces the test cost considerably. The materials under study can be machined in the shapes of blocks or cylinders can be used, they are held in sample holders. 10-station wear generator setup 10-station wear generator sample holders (10) Once the machine has been set up with 10 sample holders running against the rotating shaft, the machine does not require supervision or any online monitoring devices, although it may be beneficial to add some sensors to this system. This Parallel Wear Testing - an Update 228 24th International Colloquium Tribology - January 2024 is the subject of further research and development. To date, we focus on running standard sets of conditions to compare many different materials. One such case is the wear resistance of ‘lubricant containing’ polymers. We tested the wear resistance of 16 different polymers against a standard steel shaft. Test conditions were optimized for time: test duration of no longer than 1 week, or 200.000 test cycles. The wear evaluation method is optimised, rejecting a weight loss measurement (effect of water absorption, and too light wear losses), comparing 2Dand 3D-wear scar measurements. It is found that the 3D measurement is over-complicated in this case, and gives no more information than the simpler 2D-method. The wear volume on the blocks (see picture below) can be easily estimated geometrically by measuring just the wear scar width. So a more time-efficient method has proven to be adequate for this characterisation. 3D image of a typical wear scar of plastic material test against rotating steel shaft. Typical wear scars, easily measureable with 2D (optical) technique. Illustrating typical scar width variation for a single material in the same test. 3. Results and conclusions The previous image shows the typical variation that is measured on nominally same materials. All 5 samples in this test have undergone the exact same, and parallel wear test (same test conditions), and samples were taken from a single production batch. Nevertheless, there is some notable variation in wear scars. Only by collecting enough repeats, will it be possible to conclude trends and general test results for different materials. Within a few days, a total of 16 materials, each tested 5 times (80 test results) can be produced easily. Comparison of 16 different polymer grades, in terms of wear (scar) resistance. Higher is poorer wear resistance. Note the variation per material, as indicated. Some materials are clearly more repeatable than others. Wear evolution: In addition to the total wear of a material after given duration, it is also efficient to plot a wear evolution : 10 stations are available, and 10 samples of the same material can be loaded at the same setup. Then, pairs of loading stations can be retracted after given intervals during a long test, and the wear of each can be measured after the final test duration. This quickly gives an indication of the wear evolution of a material. The following graph shows the evolution for two materials, showing in both cases a rapid run-in wear, followed by a more steady low wear rate. The question becomes : how to define a wear rate ? Total wear divided by total duration ? Or slope of the wear curve, ignoring run-in wear ? Wear evolution measured in one run : subsequent removal of wear stations. CONCLUSION: parallel wear testing opens the possibility to produce a lot of wear data efficiently, both in personnel time, materials, and characterisation methods. It allows for the first time to study the variability of materials durability, the wear evolution of materials, and the confidence level of any wear metric in an economically feasible way.