Comparison of different ANFIS models for the condition monitoring of a rack and pinion contact using methods of explainable artificial intelligence

Abstract

This paper investigates the use of XAI and TAI methods for condition monitoring on a laser cutting machine. The focus is on the analysis of the rack and pinion contact with wear being predicted by four differently derived ANFIS models. Using both model-agnostic and model-specific parameters integrated in a weighted evaluation framework, the models are evaluated with respect to the effectiveness of explanations. This framework is based on the observation of the outputs of the individual layers of ANFIS, also focusing on aspects of two multi-valued logics, namely fuzzy logic and support logic. The results show that the introduced weighted evaluation framework makes it possible to quantify the explainability of the individual models in terms of XAI and TAI. Finally, a preselection of a model for predicting the wear of the rack and pinion contact can be made.

Conflict of Interest Statement

The authors declare no conflicts of interest