Research On Multi-state Assessment Method Of A Rolling Bearing Based On Sparse Auto-encoder

As a key component of rotating machinery,rolling bearing plays a very important role in the safety performance of the whole equipment.If the running condition of the rolling bearing can't be quantitatively assessed timely and accurately,the equipment will be out of service,or even life-threatening.Therefore,it has a great significance to achieve multi-state assessment of a rolling bearing.In order to fully exploit the running information of vibration signal of a rolling bearing,the shallow and deep features of vibration signals of the rolling bearing are extracted respectively.Shallow features include time domain,frequency domain and time-frequency domain features.Deep features are adaptively extracted based on sparse auto-encoder of deep learning theory.Deep features extraction eliminates can eliminate many extraction steps and the efficiency can be improved.In order to further compare the advantages and disadvantages of shallow and deep features,the t-SNE algorithm in data visualization is further studied.Based on the t-SNE algorithm,two types of features are visualized,and the two types of the features are compared visually.The results show that the deep features are more conducive to the multi-state assessment of the rolling bearing.In addition,compared with other dimensionality reduction algorithm,verify the superiority of t-SNE in visualization.Because of the unique advantages in dealing with heterogeneous data and uneven data distribution problems,hypersphere support vector machine(HSVM)is chosen as the classification model,and combining with the improved decision rules,the multi-state assessment of the bearing can be achieved.Simultaneously,genetic algorithm is used to optimize the parameters of the model.The effectiveness of the proposed method is verified by experiments.Finally,to grasp the multiple operating conditions of the bearing correctly,the deep features as the input of HSVM are used in the multi-state assessment of a rolling bearing,and combining the decay coefficient and relative compensation distance of each state relative to the normal state,then the multi-state unified assessment index is constructed and the multi-state of the rolling bearing can be effectively assessed.