基于时频优化盲解卷积的工业机器人故障诊断OA

To address the problem where periodic fault features of industrial robot joint motor bearings are sub-merged by strong noise and lack stability of traditional blind deconvolution methods,a fault diagnosis and prog-nosis method based on time-frequency optimized blind deconvolution(TFO-BD)is proposed.This method ex-plicitly embeds time-domain quadratic convolution and frequency-domain linear filtering into a deep learning framework to suppress non-periodic noise and enhance fault characteristic frequency(FCF)information.By in-troducing a physics-informed loss function that fuses kurtosis constraints,frequency-domain energy constraints,and cross-entropy,end-to-end joint optimization of the blind deconvolution parameters and classification net-work is achieved.Furthermore,multi-step time-series prediction model based on gated recurrent unit(GRU)network is constructed to predict fault trends and generate early warnings,forming a"prediction-diagnosis"closed loop.Experimental results demonstrate that under-4 dB noise,the proposed method significantly im-proves performance across mainstream backbone networks,increasing the average F1-score by over 10％.The GRU model also maintains high training efficiency while ensuring accuracy,validating the effectiveness and en-gineering potential of proposed method in strong-noise industrial scenarios.