Ball-bearings fault detection for an independent cart system: experimental campaign and preliminary results

The independent cart system is a recent innovation in electric drives, comprising a fleet of linear motors (carts) moving along a predetermined circuit. Monitoring the condition of this new system is challenging, especially for fault detection in the bearings that connect the moving carts to the fixed rail. Sensors on the frame can capture vibrations from a nearby cart but also pick up noise from other carts and impacts related to the dynamics of the specific application. In a real industrial environment, for example in a packaging machine, the system might include up to one hundred carts operating simultaneously. Additionally, wired sensors on moving carts are not feasible. The resulting environment is highly non-stationary with non-Gaussian heavy-tailed noise. This paper presents an autoencoder-based anomaly detection of such independent cart systems, starting from an extended experimental campaign.