In-situ fault diagnosis for the harmonic reducer of industrial robots via multi-scale mixed convolutional neural networks

The faults of harmonic reducers result in excessive vibration affecting the joint stabilization of industrial robots and manufacturing quality. In-situ fault diagnosis of harmonic reducers can avoid the disassembly of industrial robots to reduce the downtime of production lines. Compared with disassembly diagnosis in an experimental environment, in-situ signals for diagnosis are more complex due to the multi-scale characteristics of harmonic reducers and industrial noise interference. In this paper, an in-situ fault diagnosis method via the multi-scale mixed convolutional neural networks (MSMCNN) model is proposed. The MSMCNN model with the multi-scale feature extraction ability is specially designed for the harmonic reducer of multi-joint industrial robots with multi-scale characteristics, which can extract more comprehensive and complementary fault features from complex in-situ multi-channel signals with industrial noise. Integrated experiments are performed on real industrial robot datasets and public disassembling part datasets for assessing and analyzing the effectiveness of the proposed method. The experiment results show that the MSMCNN achieves 97.08% and is superior to classical and some state-of-the-art congener DL methods in terms of diagnosis accuracy.

Automated summary

In this paper, an in-situ fault diagnosis method via the multi-scale mixed convolutional neural networks (MSMCNN) model is proposed for harmonic reducers. The MSMCNN model extracts more comprehensive and complementary fault features from complex in-situ multi-channel signals with industrial noise. Experimental results show that the MSMCNN achieves superior diagnosis accuracy compared to classical and some state-of-the-art DL methods.