Research on Data-Driven Intelligent Fault Diagnosis Method for Industrial Robots

Abstract

Industrial robots play an indispensable role in realizing intelligent production and industrial upgrading. In order to ensure the healthy and smooth operation of industrial robots, a reliable fault diagnosis system for industrial robots needs to be established. However, many shallow learning fault diagnosis methods rely on manual extraction of signal features and selection of appropriate classifier combinations, which rely heavily on expert experience. The fault diagnosis model optimization process is time-consuming and has poor generalization ability, making it difficult to meet the actual fault diagnosis needs of industrial production. In robot fault diagnosis tasks, problems such as lack of fault data and category imbalance are also faced. Aiming at the difficulty of complex fault diagnosis of industrial robots, we propose an improved one-dimensional (1D) convolutional neural network fault diagnosis model (SRIPCNN-1D). The SRIPCNN-1D model has fewer model layers, fewer training parameters and strong model expression ability, and is suitable for online fault diagnosis of robots. This model achieved a diagnosis accuracy of more than 98% on the multi-axis industrial robot compound fault data set. It was compared with WDCNN, CNN-1D and other models as well as single fault diagnosis models to verify the effectiveness of the proposed model. At the same time, we also studied the impact of different sampling frequency data on the fault diagnosis effect of the model established by this algorithm. Experimental results show that the model is still effective when we adjust the sampling interval to 1 s.