Towards Predictive Safety Maintenance for IoT Equipped Bikes

We present a novel approach for predictive maintenance using acceleration data. Modern bikes can be equipped with additional smart features that enable early detection of deteriorating brake performance. This allows individual user feedback based on the condition of their bikes and therefore improve safety. We evaluate the suitability of various machine learning approaches for predictive maintenance using acceleration data of bike rides with good and bad brake performance. Here we compare two methods of measuring acceleration, namely we use hall sensors and inertial sensors. Overall, we achieve a F1-score of up to 0.76 using the time series specialized k-nearest neighbor in a preliminary evaluation. Furthermore, our results show that inertial sensors are better suited for measuring acceleration data than hall sensors.