Research On Surface Quality Detection And Processing Performance Evaluation System For Clean Cutting

The future cutting technology will inevitably meet the requirements of green and environmentally harmonious clean machining.The basic phenomena and mechanisms during the machining process under clean cutting conditions will undergo significant changes,and the research on related detection technologies has not yet been improved.This study focuses on the national key research and development project:Comprehensive Performance Evaluation and Testing Technology for Clean Cutting(2018YFB2002205),which proposes clean cutting methods such as high-speed dry cutting,low-temperature cutting,and micro lubrication cutting.Starting from the machining surface quality detection methods and methods created by cutting,it addresses the technical shortcomings of existing detection methods such as cumbersome processes and low intelligence,based on deep learning methods and machine vision technology,the construction of surface roughness detection models and surface defect detection models is studied.Based on the research of the existing cutting database in the research group,the system functions are enriched and expanded,and a comprehensive evaluation system for clean cutting surface quality is designed and developed to achieve surface roughness and surface defect detection functions,as well as cutting data management and sharing of cutting test data Process calculation and material processability evaluation function.The specific research work of this article is as follows:A surface roughness detection model was constructed based on convolutional neural network and image repair algorithm.High speed dry milling experiment was carried out with superalloy GH4169 as the research object.Surface image of workpiece was obtained by laser scanning microscope and roughness was measured.Roughness detection data set was established.The convolutional neural network(CNN)classification model was constructed for roughness detection,and its detection performance for clean surface images and surface images with chips was tested.The results showed that the average relative error values of the two detection methods were 3.5%and 14.0%,respectively,and the average detection time of a single image was 0.28 s.In order to solve the problem of chip affecting the detection accuracy,the research of anti-chip interference detection based on image repair is carried out.The CBAM Res-Unet semantic segmentation model was built based on the integrated attention mechanism and residual structure of Unet model to realize the location and segmentation of chip region.PConv-Net was used to construct the image repair model to realize the chip area repair.CBAM Res-Unet,PConv-Net and CNN models were integrated to form an anti-chip interference detection framework.For surface images covered with chips,the relative error of the framework was controlled within 6.8%,the average relative error value was 3.6%,and the average detection time of a single image was 0.79 s.Construct a surface defect detection model based on Rep-CA-YOLOv5s target detection algorithm.Taking GH4169 workpiece processed by high-speed dry milling as the research object,an industrial camera was used to collect surface images of the workpiece,and a defect detection dataset containing pit and scratch defect categories was established.Using the constructed dataset to train and test the YOLOv5s network,the results show that the average detection accuracy,recall rate,and average accuracy of the model are 92.2%,83.0%,and 91.7%,respectively.The model has low detection accuracy for long scratch defects.Rep-CAYOLOv5s network is constructed by introducing RepVGG module and coordinate attention module into YOLOv5s network,and network regression loss function is constructed using SloU evaluation index to improve model detection performance.The experimental results show that the Rep-CA-YOLOv5s model improves the detection ability for scratch defects,and its detection accuracy,recall rate,and average accuracy are increased by 1.5%,7.2%,and 3.5%,respectively.Research on reasoning speed improvement based on channel pruning and filter pruning expansion models,adding BN layers to the loss function respectively γ The L1 regular constraint of the scaling factor and convolution kernel weight w implements sparse training of the model,and selects the optimal sparse model as the pruning model based on the model detection accuracy and sparsity degree.Set different pruning ratios for pruning experiments,and the results show that for the Rep-CA-YOLOv5s model,the filter pruning method has a greater improvement in the compression strength and reasoning speed of the model under the same pruning ratio.When the pruning rate is 50%,the optimal detection model is obtained through the filter pruning method,with a mAP value of 94.2%and an average inference time of 4.3 ms for a single image.A comprehensive evaluation system for clean cutting surface quality was designed and developed.According to the analysis and evaluation system requirements of high-speed dry cutting and other clean cutting machining scenarios,the functions realized by the evaluation system and the data support of the requirements were determined,and the relevant cutting data information was collected and sorted out.The concept structure and logic structure of the database are designed,and the cutting database is constructed to provide data support for the evaluation system.A material machinability evaluation model was constructed based on processing cost,environmental impact and processing quality.The evaluation system adopts B/S architecture,and uses Django framework and Python assembly language to complete the development of the evaluation system.The roughness detection model and surface defect detection model are deployed to the evaluation system.Finally,the system realizes the functions of cutting data management,material machinability evaluation and workpiece surface quality detection and evaluation.