Fault Diagnosis Method Of Crane Reducer Based On Transfer Learning

The crane is a large industrial machinery and equipment indispensable in construction,ports,railways and various construction plants,and its safe operation is directly related to the economic benefits of enterprises and people’s personal safety.As the core equipment of the crane,the reducer provides a large torque for the crane to lift heavy objects smoothly.therefore,the reducer’health status is directly related to the safe operation of the crane.Due to the crane’s high intensity work and changing working conditions,the reducer as the core equipment of the crance becomes a major faulty component.So this paper was dedicated to the crane reducer fault diagnosis method research,to timely detection and diagnosis of the reducer fault characteristics,to do the corresponding maintenance measures to avoid enterprise economic losses and protect people’s lives.Specific research content is as follows:Firstly,in view of the strong non-stationary vibration signal of the crane reducer in the actual production environment and the difficulty of extracting its hidden fault characteristics,a reducer fault feature extraction method based on the parameter adapation variational modal decomposition was proposed.In the method,an optimisation objective function was constructed.The optimisation objective function incorporated the Pearson correlation coefficient,the similarity of the time-frequency spectrum and the maximum cliff value.In addition,the grey wolf optimisation algorithm was used to adaptively search for the optimal parameters of the variational modal decomposition to extract the fault features of the crane reducer in the actual production environment.Then,aiming at the problem that the fault diagnosis accuracy of the traditional deep learning model was reduced because of lacking effective training by the vibration signal data of the crane reducer at the time of fault in the actual production environment,a single-case migration learning-based reducer fault diagnosis method was proposed.In this method,the auxiliary feature samples with labels in the source domain were adopted to improve the least squares support vector machine migration learning strategy according to the principle of minimising structural risk.In addition,the method proposed above was applied to the input sample data for feature extraction to reduce the noise interference of the sample data and improve the performance of the target fault diagnosis model,so as to achieve the fault diagnosis of crane reducers under single working condition in a practical production environment.Finally,in order to solve the problem of unstable performance of the migration training diagnostic model assisted by using sample data from a single working condition source domain,a reducer fault diagnosis method based on migration learning of multiple working conditions was proposed.In the method,the probabilistic latent semantic analysis model was used to explore the implicit features between the crane reducer fault vibration signals under different working conditions.In addition,combined the fault feature dictionary and Fisher kernel function,the auxiliary feature samples with labels from multiple source domains under different working conditions were adopted to establish a multi-source domain migration LSSVM fault diagnosis model based on probabilistic latent semantic analysis.Therefore,the stability of the diagnosis effect was improve and the fault diagnosis of crane reducer under multiple working conditions in real production environment was achieved.Combined with examples,this paper provided the practical value of fault diagnosis of crane speed reducers in practical production environments.