Stress detection model of 7065 aluminum alloy thick plate based on deep neural network
Aiming at the problem of measuring error of the traditional ultrasonic stress detection method on the aluminum alloy thick plate under different tensile ratio and different temperature conditions in industrial production,a stress detection model based on the dendritic neural network stress prediction model and the traditional ultrasonic detection method under different tensile rates and different temperature conditions was proposed.The improved GSA-GRNN was used to compensate the temperature of the stress detection model.In this study,7065 aluminum alloy thick plate produced by Nannan Aluminum Company was taken as the research object,and a constant temperature tank was used to provide a constant temperature environment for ultrasonic testing.Ultrasonic testing was carried out on 7065 aluminum alloy thick plate under different tensile rates and different temperatures,and the sound time difference and tensile rate were taken as input parameters and the stress as output parameters.A stress detection model based on dendritic neural network was created.Then,the output of the stress detection model was taken as input,and the improved GSA-GRNN was used to build a temperature compensation model to compensate the stress detection model.The results are shown as follows.The root-mean-square error and correlation coefficient of the detection model integrating the traditional ultrasonic sound time difference are 0.84636 and 0.99743.Compared with other neural network models,this model can have better accuracy.After temperature compensation,the stress root-mean-square error and correlation coefficient of the model can reach 0.78848 and 0.99844,respectively.The accuracy of the model has been further improved,which proves that the data-driven neural network integrated with traditional ultrasonic detection can effectively reduce the detection error,save the time of manual stress calculation of traditional detection methods,and improve the detection efficiency.The research results can provide further optimization reference for stress detection model based on data-driven.
stress detectiondendritic neural networkparticle swarm optimizationgravity search algorithmacoustic time difference
万方数据
维普
