Research On Health Monitoring Of Vertical Shaft Rigid Canal Channel Based On Convolutional Neural Network

With the development of industrial intelligence and the update and iteration of information technology,how to intelligently diagnose mechanical faults of complex systems has gradually become a research hotspot.As an interactive tool between the mine and the ground,the smooth and safe operation of the shaft hoisting system is the key to the safe production of the mine.This paper takes the shaft hoisting system as the research object.In the era of big data,the problems of fault diagnosis methods include too much reliance on manual work and too much input data,resulting in poor network training.In order to solve the problems such as difficulty in manual extraction of rigid tank channel fault diagnosis and poor diagnosis effect under variable working conditions,a fault diagnosis method of off-design bearing based on residual pyramid attention convolution neural network is proposed.Aiming at different application occasions and monitoring requirements,the vibration response signals generated by lifting vessels passing through the fault area are analyzed and tested,which proves the superiority and applicability of the method.Firstly,according to the structure and principle of the shaft hoisting system,the shaft hoisting system test bench was built,the three-dimensional model of the shaft hoisting system was established,and the vibration test response system was built,and the operation control system of the shaft rigid cage way is built.The installation position of the sensor is solved and analyzed by ANSYS,the sensor is installed in the best position,which is more convenient for subsequent data acquisition operations.The acceleration response signals generated when the lifting vessel is running under two types of faults,step excitation and joint gap,are collected and analyzed.Secondly,in view of the one-dimensional characteristics of the collected data,the traditional two-dimensional convolutional neural network will destroy the spatial structure of the signal,etc.,a one-dimensional convolutional neural network is proposed.It includes Dropout and Adam algorithms,automatically proposes data features,and the accuracy of one-dimensional convolutional neural network model is 72%.According to the experimental results such as the accuracy curve and the loss function curve of the model,the ordinary one-dimensional convolutional neural model has poor ability of over fitting and generalization.After that,the one-dimensional convolutional neural network is optimized and improved,and the residual pyramid neural network is proposed..The two-dimensional input of ResNet network model based on one-dimensional convolutional neural network is changed into one-dimensional input,and the one-dimensional signal of multiple working conditions is input into that attention convolution structure of the remain pyramid;The features of input data are absorbed by multi-scale features,and the feature learning ability is improved.According to the experimental results,the accuracy of the model is 92%,and the advantages of this method can be obtained from four comparative experiments.The T-SNE output of the network layer of the model is visualized,and the Gaussian white noise is added to the original data,which verifies that the model has good noise resistance.Finally,using PyQt5 development framework and MySQL database technology,a fault diagnosis system for rigid tank channel health monitoring is designed,which visualizes the original signal,and then selects the corresponding algorithm to be applied to the signal for detection.The system has designed five major system software function modules(login and registration module,user login and registration module,data import module,fault diagnosis module,history recording module)for the convenience of users Figure [59] Table [6] Reference [87]...