Interpretable artificial Intelligence for COndition Monitoring (IICOM)

Purpose and goal:

Traditional condition monitoring is based on physical and rule-based models. In recent years some modern manufacturing industries embraced the big data revolution and have gone beyond the traditional physics-based models and implemented data-driven machine health monitoring systems.

Nevertheless, most data-driven models lack interpretability, robustness and adaptability, which makes it difficult to use them in real world situations. We aim to develop causal machine learning for condition monitoring models which are interpretable. Expected results and effects:

In this project we aim to develop causal methods for ML models in the CM field of heavy vehicle industry. Through this we aim to demonstrate how the causal methods will help to solve the above mentioned

obstacles; adaptability, robustness and interpretability. Three different vehicle components, belonging to the after-treatment system of the vehicle will serve as test-bed of this research. The results of this project will have a reach beyond the vehicle industry, as it will show a path and strategy for applying causal methods to general CM problems using ML RUL models.

Approach and implementation:

The proposed project setting with focus on an industrial PhD student has the advantage that the student will be able to take the necessary time to dive into the active research topic of causality and interpretability, making the student into an expert in the field within the project time. At the same time Scania’s data and use-cases will lie on the focus during the research.

This combination will bring the cutting-edge research of causality and interpretability closer to actual industrial implementations, in this case for CM systems.