Research On Fault Diagnosis Method Of Key Components Of Hoist Frequency Conversion Control System

The hoist variable frequency control system is an important part of the hoist.Because of its complex and harsh operating environment,failures are prone to occur.Therefore,it is of great significance to diagnose the fault of the hoist’s frequency conversion control system.This thesis takes the hoist variable frequency control system as the research object,and studies the fault diagnosis method of the inverter and the motor two key components in the variable frequency control system.The inverter and the motor form a complex nonlinear system,which leads to its signal characteristics having nonlinear and instability characteristics.However,traditional fault diagnosis methods have problems such as comprehensive feature selection and information redundancy in feature extraction,which affect the accuracy of the final judgment and the ability to adapt to working conditions.In view of the above problems,this thesis has launched the research on the fault diagnosis method of the two key components of the inverter and the motor.The main research contents are as follows:(1)The structure and fault types of the hoist variable frequency control system are analyzed,the composition,working principle,signal response characteristics and fault types of the three-level inverter are analyzed in detail,and the working characteristics and fault types of the motor are explained,and signal analysis methods and fault diagnosis models are analyzed.(2)Based on the analysis of the three-phase output current signal of the three-level inverter,the Hilbert Huang transform method is studied,and the time-frequency domain signal analysis is performed on the three-phase output current signal.The Hilbert Huang transform method is used to extract the statistical features of the current signal samples,and the original feature set(Original feature set,OFS)is constructed.Aiming at the inconsistency and redundancy of feature expression in the original feature set,a sensitive feature selection method based on Fisher Score and the maximum information coefficient is proposed to evaluate the statistical features obtained by time-frequency analysis,and the random forest algorithm is used to evaluate the importance of features,and then to achieve the selection of sensitive features.(3)Aiming at the problem of incomplete feature information extraction in traditional feature extraction methods,inverter fault diagnosis method based on SE-Dense Net-ELM and multi-source feature fusion is proposed.The model uses SE-Dense Net to extract deep features from the original data set,and merge the selected sensitive statistical features;secondly,it uses the Local Fisher Discriminant Analysis algorithm to fuse and reduce the dimensionality of statistical features and depth features;Finally,the three-level inverter fault state recognition is realized by using the Extreme Learning Machine(ELM)classifier.Simulation and test bench data are used to carry out experiments,and the experimental results show that the accuracy of the proposed fault diagnosis model is 99.61% and 99.18%,respectively.Compared with the traditional experimental method,the accuracy of the proposed fault diagnosis is improved by 1.16% and 3.2% respectively,which further shows that the proposed inverter fault diagnosis method has a higher fault recognition accuracy.(4)Aiming at the problems of insufficient shallow structure feature learning ability,complex diagnosis process and low fault diagnosis accuracy in traditional motor fault diagnosis methods,a method based on Variational Automatic Encoder(AVE)and Random forest(RF)is proposed.Use AVE to automatically extract effective depth features from vibration signals,and use RF to diagnose and identify motor faults.Experiments with the data from the SQI-MFS motor test bench show that the proposed method has a fault identification accuracy rate of 97.53%.Compared with the traditional diagnosis method,the identification accuracy rate is increased by 3.71%.It is verified that the method has excellent deep feature extraction ability in feature extraction and high accuracy of motor fault recognition.The article has 54 figures,14 tables and 96 references.