Research On Straightening Process Forecast Simulation System Based On Deep Learning

Straightening machine is the core equipment in metal processing and manufacturing,which controls the quality of metal products.The intelligent straightening process model forecast and simulation system is the key to realize the intelligent straightening machine.The defects of metal sheet are often related to the uneven distribution of residual stress,so the size and distribution characteristics of residual stress are very important for setting the subsequent processing parameters of metal sheet.This paper focuses on the key parameter of residual stress distribution feature of sheet metal,designs a network model of residual stress distribution feature prediction based on deep learning,optimizes the existing straightening process,and designs a simulation system based on deep learning optimization.This paper conducts the following research topics:(1)A deep learning-based feature prediction method of metal plate residual stress distribution is proposed.Using the lightweight convolutional neural network to find the relationship between the characteristics of metal plate shape and residual stress distribution.Collect metal plate shape data through line structure light or laser camera,and extract subarea plate shape data as input data for neural network training.The distribution value of the residual stress in each subregion was used as the residual stress detection equipment to construct the sample data set of the experiment.(2)Research on predicting network model of residual stress distribution based on deep learning.Using MobileNet V2 to build the subject of the network model,the global maximum pooling layer and the full connection layer of the network structure are improved to realize the predicted output of the residual stress distribution value of the region.By introducing the coordinate attention module,enhancing the attention to the detailed features of the plate shape,and adding the RFB multi-scale fusion module,the extraction ability of the model to form the global plate features is improved,and the model prediction accuracy is further improved.(3)Research on the forecast and simulation model of straightening process based on deep learning.Based on the residual stress distribution characteristics obtained by the plate-shaped residual stress distribution prediction module,the bending roll model and the left-right tilt model in the straightening process are optimized,and the prediction simulation model of the straightening process based on the residual stress distribution characteristics is constructed.(4)On the basis of the above research,a straightening process prediction simulation system based on deep learning optimization is designed and implemented,which verifies the feasibility of the proposed model.