Explainable Artificial Intelligence in Mechanical Engineering: A Synthetic Dataset for Comprehensive Failure Mode Analysis

Transparency and comprehensibility of artificial intelligence (AI) algorithms is of central importance across vertical industries. However, comprehensive datasets for evaluating explainable AI (XAI) methods are lacking in mechanical engineering. This paper introduces a new synthetic dataset for failure mode analysis in drilling that is generated with real-world applicability in mind. It is optimized for XAI purposes and can be used to evaluate explanations of such methods. We show that local explanations and overall feature importance scores derived from SHapley Additive exPlanations (SHAP) values match expert-defined failure modes on our data, and thus lay the groundwork for future XAI research in the field.