AI failure analysis of technical elastomers

Developing an automated detection process for elastomer failure

KI-basierte Schadensanalyse von technischen Elastomeren
© pixabay, HansBraxmeier

Elastomer failure needs to be detected quickly and easily — but how? Due the variety of failure patterns that occur on elastomer components, it is not always easy to establish the exact cause of a failure. In the internal Fraunhofer project, AI Failure Analysis of Technical Elastomers, researchers are exploring the possibility of objectively analyzing elasto-mer failures through artificial intelligence. The resulting process will be transferred to customer-specific applications at a later stage.

Objectivity and efficiency

Elastomers are exposed to extremely high and complex requirements, such as mechanical and thermal loads and different media influences. Damage to these components can result from these loads as well as from aging, deviations in the manufacturing process and other factors. An analysis of damage to the components can be carried out with the help of VDI 3822. However, experience and expertise as well as similar damage patterns with different causes of damage lead to a subjective damage assessment in this process. The Fraunhofer-internal development project of "AI-based damage analysis of technical elastomers" is intended to automate and objectify this time-consuming and cost-intensive process and to sound out the necessary framework conditions (e.g. input parameters, training data basis, extrapolation potential, etc.) for a practice-relevant implementation.

Automated detection of specified causes of damage

The objective of the first project phase is to establish a process chain from image capture to reliable damage assessment of different component damage. In addition to the actual damage images, additional material or process parameters are included in the damage assessment and thus support the assignment of damage causes to the damage images according to the cause-effect relationship. In addition to the correct detection of damage including its interpretation and cause assignment by the implemented process chain, a major project focus will be on the required training database. While many components are manufactured in considerable lot sizes and are used in systems, damage to components should remain the exception rather than the rule and may be due to improper conditions of use. For various applications, this results in significant limitations in the training database as well as the derivation of the corresponding process requirements. 

Validity range and extrapolation potential

After successful detection, analysis and assignment of damage causes and damage patterns, the validity range of a model approach is analyzed in detail in the second project phase under variation of the component and the damage. For example, the influence of scaling effects or the location and extent of the damage will be investigated in order to draw conclusions about generalization errors of the model. Based on these findings on the boundary conditions of the model, the transferability of the process to further application cases is discussed.