Compound Fault Diagnosis of Industrial Robot Based on Improved Multi-label One-Dimensional Convolutional Neural Network

Abstract

Industrial robot is one kind of complex infrastructure in industrial production and applications. Fault diagnosis is an important part of the intelligent application and monitoring of industrial robots. For multi-axis industrial robot compound fault prediction and diagnosis problem, this paper proposes a fault diagnosis model based on improved multi-label one-dimensional convolutional neural network (ML-SRIPCNN-1D). Firstly, the compound fault data set is enhanced by random sampling and Mixup. Then, the single fault data and compound fault data were trained end-to-end by the improved multi-label one-dimensional convolutional neural network. Finally, accurate diagnosis and prediction of compound faults of industrial robots are implemented. The compound fault data set was derived from a company's multi-axis industrial robot. The characteristic variables of fault diagnosis are torque, current, velocity, position, etc. Compared with SRIPCNN-1D, MLCNN, WT-MLCNN, T-FSM-MLCNN, ELM + AE + SVM, LMD + TDSF + ML-KNN models, the average diagnosis accuracy of ML-SRIPCNN-1D reached 98.67%. The model has good diagnosis effect and high accuracy for the prediction and diagnosis of industrial robot compound fault.