The maintenance of chassis components is currently carried out on the basis of defined thresholds (preventive maintenance) or in the event of acute defects (reactive maintenance). In both cases, unnecessary costs are incurred: if components are replaced too early, the remaining useful life is unused and replacement too late can result in expensive consequential damages.
Chassis components are also crucial for vehicle safety, so that any defect that occurs must be detected as quickly as possible. With the increasing emergence of car sharing services and autonomous vehicles, the driver is no longer available as a monitoring authority, as he or she is not aware of the normal driving status or the responsibility has even been completely transferred to the vehicle. This means that a defect remains undetected for longer if the vehicle cannot take over the monitoring task.
The vehicle should be able to detect the condition of individual chassis components such as tires, brakes or dampers, either independently or in combination with a back-end architecture. Additionally, the remaining useful life should be predicted. This method, called predictive maintenance, not only increases general road safety but also enables cost savings for fleet operators, workshops and customers, since procurement of spare parts can be planned accordingly.
In addition to the selection of suitable sensor signals, the first step is to examine the required signal quality. After onboard preprocessing, these measured values are transferred to a cloud infrastructure and evaluated by means of machine learning methods. Both supervised and unsupervised learning is conceivable here. In order to obtain a scalable solution, the use of big-data frameworks is aimed at.
A major challenge is to derive suitable models for the enormous variety of vehicle derivatives without having to train new machine learning models each time. Furthermore, the generation of training data is time-consuming and expensive, so that a continuous spectrum of component states must be derived from a limited number of known defects.