Neural Network Based Digital Twin Health Monitoring of BLDC Motor Drives for Robots

Abstract

Robots have shown promising prospect in numerous applications, such as space exploration and disaster rescue. Due to the harsh environmental conditions (e.g., high temperature) in many applications, motor drive systems in the robotic arms are vulnerable to hardware failures such as inverter switching aging or faults. To address this challenge and avoid significant downtime cost, a digital twin based online health monitoring model, is developed for diagnosing potential switching faults that could occur to the robotic brushless DC (BLDC) motor drives. Specifically, the online digital twin health monitoring model is based on a dynamic neural network (DNN). Various DNN architectures have been tested to determine the best trade-off between the model accuracy and computational efficiency, which is to ensure that the proposed model can be embedded into a microprocessor and used in real-time applications. Finally, the efficacy of the proposed DNN-based digital twin approach is validated with testing data in a BLDC motor-drive prototype.