Essais & Simulations n° 138

DOSSIER AUTO MOTIVE

DOSSIER AUTO MOTIVE TECHNOLOGY Acceleration measurementsand Usage Classification for avehidein different driving conditions ln this paper, two experts from the French company Metravib highlight the concrete use of artificial intelligence and machine learning in the implementation of automotive component tests. Daniel Vaucher de la Croix and Jean-Yves Disson, from the French company Metravib uite generally more and more attention is paid to Q the effect of severe environment conditions to critical parts of industrial structures, in various fields such as Energy, Oil&Gas, Automotive, Aerospace to mention the largest ones. Extracting from field measurement data most relevant information for anticipating potential failures and ultimately predicting remaining life time is what is expected from a performing Health and Usage Monitoring System (HUMS). In this context Artificial Intelligence (AI) and Machine Learning (ML) techniques become essential tools which prove their efficiency every day. This is why Metravib decided to invest in such technologies, already providing quite interesting results in several projects such as the one reported in this paper. o Wnkdamai• O Mode rate damai• 1.5 O S.vua da _ ma1• The focus is given here on the dynamic behavior of the vehicle suspension and adequate processing of data collected from the vehicle driven in tough conditions. Determining relevant information to get the clearest view about possible failures way before they can appear is no easy task as direct measurements are often qui te difficult to achieve if not impossible. Metravib proposed approach takes operational constraints into consideration and first concentrates about an effective way to rank Usage as part of a dedicated Health and Usage Monitoring System (HUMS). It will also be exposed in this paper how a dedicated Usage Classifier has been developed to provide 3 classes of damage for the suspension: weak, moderate or strong. A test campaign has successfully been carried out demonstrating how Machine Learning methods can help training and validating that classifier fed with indirect and simplified measurements. TEST CAMPAIGN AND INSTRUMENTATION 0.5 .5 -1 -2 ·2.5 ·2 -1.5 -1 -0.5 0 0.5 1.5 1 ·Definition of 3 classes for each elementary usage 2-diagram indicators related to direct measurements In order to feed an initial database several tests were performed driving a 4X4 rugged vehicle in different conditions (road, gravels, potholes, African washboard, ail terrain, concrete bumps, sinus, turns ... ) on tracks available on site at close distance from Metravib. The deployed instrumentation included: GPS, direct instrumentation used for usages identification (wire displacement sensors to measure the displacement of the suspension arms) and indirect instrumentation (3 axis accelerometer in a safe location inside the vehicle). 52 IESSAIS & SIMULATIONS• N ° 138 • Septembre - Octobre 2019

DOSSIER AUTO MOTIVE USAGE IDE N TIFICATION FROM DIR ECT MEASUREMENTS Measured signals are eut into blocks of dedicated time length called elementary usages. Tuen two so-called "damage indicators" have been defined in order to identify the class of each elementary usage of each track. In fact these indicators are not the true damage but allow performing a ranking of the damage when the vehicle is rolling. The first indicator (referring to the torsion bar and calculated from the measured displacement of the suspension) is called virtual damage as the calculated stress in the bar is a macroscopic one, not taking into account the local elfects. The second indicator relates to virtual damage of the