Predictive Maintenance in Automotive

Challenge

Planning the costs associated with warranty claims and investigating factors influencing potential field failures are central questions in quality analysis. Our client, an international automotive company, faced the challenge of improving their tools for predicting failures of various engine and drivetrain components to optimize the planning of warranty costs.

Approach

Modeling failure times can be done in various ways. Best-practice approaches like Weibull analyses are often too one-dimensional to calculate the complex interactions between different factors influencing failure. More complex models like survival models or non-parametric estimations of expected failure distribution offer more flexibility but are often not an option due to high-dimensional influences and very large datasets. To calculate failure times, we used a machine learning model capable of adequately processing the high-dimensional data structure while providing accurate failure time predictions.

Result

The developed machine learning model showed a significant improvement in forecast accuracy compared to the existing tool. The models and insights developed during the project help the client today to best plan and estimate expected warranty costs. For optimal use, the model was implemented into a dashboard environment.