Abstract: Robust identification of the running state of transmission parts is of great significance to ensure the service performance of electric mountain wreckages.In the high-frequency variable load operation and high noise environment,the collected vibration signals are often affected by complex transmission paths and component coupling,resulting in the interference of irrelevant noise,which brings challenges to the accurate diagnosis of bearing faults of mountain wreckages. A fault diagnosis model of rolling bearing based on frequency domain Granian Angular Fidles and multi-scale Residual Networks (ResNet) is proposed. The frequency component of time series signal is reconstructed by Granian Angular Fidles,and the features are weighted and enhanced by multi-scale attention enhancement mechanism. This enables the model to adaptively focus on the features that are most important for fault diagnosis while suppressing the impact of noise. The residual connection is introduced to promote the training and information flow of the deep network,so as to promote the learning of complex features. The fault accelerated life test data set of rolling bearing in Xi’an Jiaotong University and the rolling bearing data set of mountain wrecking truck collected by using the rolling bearing test bed are used for verification. The fault recognition rate in the two data sets is more than 99 %,which verifies the effectiveness of the proposed fault diagnosis model. Compared with different variable load conditions,the fault recognition rate of the model is more than 98.5 %,and the recognition rate is still more than 90 % in the high-frequency variable load environment of -6 dB noise,which further verifies that the model can be used for bearing fault recognition of mountain wrecktrucks and has good robustness and generalization ability.

Key words: Granian Angular Fidles, multi-scale, Residual Networks(ResNet), high frequency variable load, bearing fault diagnosis, mountain wrecker