IDEA: Identifying key variables in monitoring of production processes in automotive industry

Purpose and goal: The IDEA project aims to extract key features from original signals for anomaly detection in process monitoring.

Different learning methods have been investigated in this project to create a low number of features from the original data while still capturing the significant information. This work has led to substantial reduction of input dimensionality of anomaly detection models in process monitoring. Expected results and effects:

This pre-study project gained valuable experience of applying deep learning methods to reduce input dimensionality for anomaly detection models in process monitoring. Different detection models combined with low dimensional feature learning have been constructed and evaluated. The acquired lower dimensionality and model complexity brings the following benefits:

* More precise and reliable detection of anomaly * Faster detection in real-time monitoring * Lower energy consumption in deployment Approach and implementation: The following tasks have been performed in the implementation of the IDEA project * Scenario analysis and use case selection * Data understanding and interpretation

* Temporal approximation from original data * New feature creation by means of data analysis and learning * Construction of detection models using learned features * Evaluation of experiment results IDEA has been conducted with close collaboration between Mälardalen University and Volvo Trucks