Fault Diagnosis Method for Harmonic Reducers under Different Working Conditions Based on Digital Twin

The harmonic reducer, a crucial component of industrial robots, works in complex and variable environments, leading to significant losses when failures occur. Due to the challenges in acquiring actual vibration data of harmonic reducers, the limited number of fault sample, missing data labels, and differences in data distribution under varying working conditions, a fault diagnosis method for harmonic reducer under different working conditions based on digital twin is proposed. Firstly, a digital twin model of the faulty harmonic reducer is constructed using dynamic modeling to generate twin data. Secondly, a virtual-real mapping method based on a cyclic generative adversarial network is proposed to achieve the mapping between twin data and real measured data. To enhance feature extraction and suppress noise interference, an improved semi-soft threshold function is integrated into a deep residual shrinkage network. Meanwhile, the extracted features are subjected to domain adaptation in unsupervised scenarios, using the maximum mean discrepancy to reduce distribution differences between domains, thereby achieving fault diagnosis under different working conditions. Finally, a fault simulation test bench for the harmonic reducer is established, and experimental verification shows that the proposed method achieves an average accuracy of 99.2% in all transfer tasks. It effectively addresses the fault diagnosis challenges of harmonic reducers in unsupervised scenarios under different working conditions.