Fault Diagnosis for Industrial Robots Based on Informer

Abstract

Timely and accurate fault diagnosis plays an important role in the reliability and safety of modern industrial robot systems, such as directed energy deposition additive manufacturing (DED-AM) systems. Aiming at the fact that few fault samples can be collected under actual working conditions, an end-to-end deep learning fault diagnosis approach for industrial robot systems based on Informer is proposed. This model solves the problem of poor fault diagnosis and generalization ability caused by limited fault samples under actual working conditions. Based on self-attention distillation and probabilistic sparse self-attention mechanism, the model focuses on hidden knowledge related to faults, thereby improving fault feature utilization. Furthermore, a deep one-dimensional convolutional network is deployed to extract features for classification. An experimental study based on the real-world industrial robot operational dataset was implemented. The results indicate that Informer shows merits in the fault diagnosis for industrial robot compared with the existing intelligent fault diagnosis algorithms.