Bearing Fault Diagnosis Based on Lightweight Model Combined with DA and TL
In order to achieve accurate and real-time diagnosis of bearing faults on platforms with limited computing resources,a method combining lightweight Mobilenet V3 model with data augmentation and transfer learning techniques was proposed.The 1D vibra-tion signals was transformed into 2D time-frequency images using continuous wavelet transform to better reveal the time-frequency characteristics of signal.Data augmentation techniques were applied to enhance the time-frequency image,it was used as input to the network model,thus improving the robustness and generalization performance of the model.Finally,the transfer learning was utilized to fine-tune the network model,reducing the number of training iterations and improving diagnostic accuracy.The effectiveness of the pro-posed method was validated on the Case Western Reserve University dataset.The experimental results show that the proposed method achieves a diagnostic accuracy of 100%in the source domain,with a diagnosis time of 41.3 ms and a model size of 16.3 MB,the accu-racy is improved by 0.437%compared with the optimal network model in the same type.In the noise with different signal-to-noise rati-os,the average diagnostic accuracy still reaches 97.406%.In cross-domain experiments,the average accuracy reaches 98.188%,which is improved by 1.563%compared with the optimal model in the same level.By considering diagnostic accuracy,diagnosis time,model size,noise resistance and generalization performance,the proposed method can achieve accurate diagnosis and real-time response for bearing faults.
万方数据
维普
