Research On The Intelligent Diagnosis Of Electromechanical Equipment Based On Hybrid Physics-based And Data-driven Models

With the increasing automation level of modern electromechanical systems and the gradual increase in system scale,the reliability and safety issues of modern electromechanical systems have also attracted more and more attention.Therefore,it is of great significance to study the fault diagnosis and prediction methods of modern electromechanical equipment.Traditional modern electromechanical equipment fault diagnosis and prediction methods mostly rely on physics-based methods and data-driven methods.However,these two methods still face some challenges when used alone.Physics-based methods are complex and difficult to solve.Data-driven methods are less robust in the case of small samples,resulting in poor model confidence.Therefore,it is necessary to consider combining the two and propose a computationally efficient and more reliable fault diagnosis and prediction method for electromechanical equipment.In this thesis,by combining physics-based methods and data-driven methods for fault diagnosis and prediction of electromechanical equipment,a fusion reasoning method of physical mechanism and data-driven is proposed.Specific research contents include:(1)A physical mechanism and data-driven collaborative reasoning analysis method is established.Aiming at the complex solution process and insufficient reliability of the traditional intelligent diagnosis method for electromechanical equipment,by analyzing the complementarity between the physics-based method and the data-driven method,the physical mechanism and the data-driven system coordination mechanism(physical information machine learning method)are summarized.methods,machine learning-assisted simulation technology,and interpretable machine learning methods),proposed a physical mechanism and a data-driven collaborative reasoning analysis framework,constructed a closed loop of information interaction between the intelligent analysis system and users,analyzed and discussed the physical mechanism and the pplication status and development prospects of data-driven collaborative reasoning technology in the field of electromechanical equipment intelligent diagnosis have realized the transformation from traditional intelligent diagnosis methods to intelligent analysis and diagnosis methods with strong reliability and high computational efficiency.(2)A self-attention analysis method for fault diagnosis is established.Aiming at the problems of unclear feature contribution and weak model interpretation ability of traditional data fusion methods,the self-attention mechanism is used to fuse multi-resolution features from multi-sensors,and a multi-resolution fusion model based on self-attention mechanism is constructed.The model is used to initially fuse multi-sensor features of the same resolution,and the self-attention mechanism is used to establish the correlation between multi-resolution features.A self-attention analysis method for fault diagnosis is proposed,and t-SNE technology is used to perform post-processing on input features.The feature extraction ability of the model is explained and analyzed.Taking the fault data of induction motor as an example,the validity of the proposed method and model is verified,and the diagnostic accuracy rate can reach 98.959%.Compared with the traditional fault diagnosis modeling method,the proposed method can effectively establish the correlation between multi-resolution features and improve the diagnostic accuracy and post-explainability of the data-driven model.(3)A physics-guided meta-learning method for fault prediction is established.Aiming at the problems of poor parameter robustness of the physical model and insufficient interpretability of the data-driven model caused by the rate of equipment deterioration With the data fusion model,a physics-guided meta-learning model is constructed to learn the robust relationship between the degree of equipment degradation and other physical variables,and a physics-based loss function is proposed to ensure the interpretability of meta-learning.Taking the tool failure data as an example,the effectiveness of the proposed method and model is verified.Compared with the traditional fault prediction modeling method,the proposed method has stronger interpretability and can provide fault information of equipment in each deterioration stage.