Case study: Predictive Maintenance

The sudden failure of vehicle components often leads to unscheduled maintenance and service appointments which increase the service costs and forces companies to maintain vehicle reserves. Mathematical models can help in reliably predicting potential failures. By doing so, maintenance and repair can be optimized, which reduces maintenance costs.

• Predicting failures and malfunctions of individual vehicle components
• Identifying failure causes (cause and effect) in a safe manner
• Optimizing service processes
• Providing live monitoring of the vehicle condition

Maintenance is a key cost driver in the transport industry. Although vehicles are equipped with more sensors than ever before, the data is hardly used to plan and control maintenance and servicing. Therefore, equipment failures and damages always lead to taking unscheduled maintenance measures. This subsequently

• decreases the transportation quality
• incurs additional costs and
• forces the company to provide a vehicle reserve.

The aim of the project was to optimize the total costs by using a holistic mathematical model, with which maintenance and repair can be planned and controlled more sustainably in the future.

Vehicle manufacturers supply information about the vehicle via the FMS CAN bus data, but are rarely familiar with operational procedures. That’s why the analysis used the high-resolution CAN bus data (8 Hz) as well as the operational data along with the timetable and maintenance data. A complex data preparation process was used to identify the data patterns that lead to a failure and recognize a multitude of different influencing variables and their effects. In order to optimize further processes, several mathematical methods were combined and transformed into a holistic model (e.g. dimension reduction and feature engineering).

The forecast was based on a random forest machine learning model, which was programmed and then used. Afterwards, the relevance of the influencing variables could be determined and, through a backtest, the forecast quality could be evaluated. In the end, the optimization of maintenance and repair costs was achieved through integration with operations research methods.

With Visual Analytics and R, it was possible to integrate the model into the customer's infrastructure. Today, the company can accurately predict impending failures and reduce the number of service appointments. The detailed forecast of the defective components can also determine the errors that will cause the failure in advance. Thus, necessary parts for the upcoming maintenance can be ordered in good time. The optimized planning and control of maintenance and servicing is regarded as a strategic measure to sustainably increase economic success.