Research On Algorithm Of AGV Fault Diagnosis Based-on Deep Learning

With the rapid development of science and technology and intelligent manufacturing,AG Vs are becoming more and more intelligent and are widely used in logistics,warehousing,manufacturing and automobile industries.In the event of AGV failure,AGVS may be damaged,causing huge economic losses in severe cases.Traditional fault diagnosis methods are difficult to establish accurate system model and fault diagnosis only focus on the current moment of system parameters,study the existing problems,this article studying AGV fault diagnosis algorithm based on depth study,introduce technology of deep learning,realize the real-time monitoring and accurate AGV malfunction discernment,timely troubleshooting for staff,ensure system running smoothly.The main research contents are as follows:(1)Through in-depth analysis of AGV common fault types and their manifestations and causes,finding that the relationship between fault types and manifestations is not one-to-one mapping.This clarifies the thinking framework of AGV system fault diagnosis based on deep learning,and completes the basic work of fault diagnosis based on deep learning,including data acquisition and data preprocessing.(2)Aiming at the problem that the convolutional neural network cannot diagnose the periodicity and trend of data during fault diagnosis,a multi-scale one-dimensional time-series deep residual network Multi-Scale Time ResNet is proposed.The network applies ResNet to the time series data,and adds convolution kernels of different scales to extract short-term and long-term data changes.The residual power short-circuit structure and deep structure are used to enhance the generalization ability and accuracy of the model respectively.Compared with the 1D convolutional neural network,experiments show that this network improves the F1 score of AGV fault diagnosis from 0.849 to 0.874,and the false positive rate and false negative rate are reduced by 3.3%and 1.7%,respectively.(3)On the long-sequence fault diagnosis data,a new type of EiReLU activation function is proposed for the problems of gradient dissipation and long-distance information loss in Long Short-Term Memory Network.Based on this,a Bi-directional Long Short-Term Memory Network is constructed,solving the problems of long-distance information loss.Compared with the sigmoid,tanh,and ReLU functions,experiments show that the EiReLU function has improved from 0.884 to 0.908 on the F1 score,while reducing the false positive rate and false negative rate by 2.9%and 3.8%,respectively.Finally,an experimental research platform for AGV fault diagnosis is established.Experimental research on the related research and theoretical analysis of AGV fault diagnosis.Experiments prove that the fault diagnosis method proposed in this paper has high practicability and accuracy in real environment.