Research On Wear Of Diamond Saw Wire Based On Machine Vision

In recent years,with the rapid development of industries such as photovoltaics and chips relying on brittle materials,the importance of slicing and processing of brittle materials has become increasingly prominent.Diamond wire saw cutting technology is a major cutting method,with diamond wire as the cutting tool.The wear of the wire saw is mainly manifested by the detachment of abrasive grains,which directly affects the surface quality of the workpiece and the processing efficiency.This article aims to detect diamond abrasive grains through machine vision to obtain the law of changes in the number of abrasive grains during the cutting process,thereby judging the wear of the wire saw.Based on the predictive model,it provides cutting time guidance recommendations.The specific work content is as follows:Firstly,according to the characteristics of diamond wire and abrasive grains,hardware such as a global camera,telecentric lens,and point light source are selected to build a detection platform that can capture clear images of diamond abrasive grains.A total of 2000 images were collected during the cutting process.Abrasive grains with complete morphology or white color were classified as effective abrasive grains,while those flattened or detached from the wire saw with black morphology were classified as ineffective abrasive grains.This created a dataset for diamond abrasive grains.Secondly,traditional visual detection and deep learning-based detection methods were used to detect diamond abrasive grains,and a comparative analysis was conducted.Haclon software was used to select suitable operators and adjust parameters for image enhancement,image segmentation,morphological processing,and feature extraction to achieve diamond abrasive grain detection of designated images.However,due to the diverse morphology of diamond abrasive grains,traditional visual detection is difficult to adapt to complex and varied abrasive grain morphology,making it difficult to detect all abrasive grains in the images,with low robustness.Then,multiple deep learning detection models were used to detect abrasive grains.It was found that YOLOv5 model had more advantages than other models,with an accuracy of79.8%,recall rate of 81.1%,and m AP comprehensive index of 84.8%.The inference speed per image was only 0.008 s,and the detection effect and performance were significantly better than traditional visual detection.For the diamond abrasive grain dataset,the YOLOv5 network model was optimized by adding a coordinate attention mechanism and replacing bottleneck layer modules,resulting in an increase in accuracy by 1.7%,recall rate by 3.7%,and m AP by 3.2%,further improving the detection effect on diamond abrasive grains.Finally,the improved YOLOv5 model was connected to the DeepSORT multi-object tracking algorithm to count the number of effective and ineffective abrasive grains within the sampled section of the wire saw after every 60 minutes of cutting.The number of effective abrasive grains decreased by 30% to 50% after 60 minutes of cutting,and by 40% to 70%after two hours of cutting.The number of ineffective abrasive grains increased first and then decreased with increasing cutting time,and the rate of decrease in abrasive grains was related to the line speed and feed speed.A multilayer perceptron neural network model was built to predict the cutting time at which the number of abrasive grains decreases to the set value based on the line speed,feed speed,and the amount of decrease in effective abrasive grains,providing guidance recommendations for diamond wire saw processing time.