Stamping Aluminum Defect Detection System Based On Machine Vision

Aluminum plate defects directly affect product quality.As an important part of the processing production line,aluminum plate automatic defect detection technology has always been a research hotspot.In order to improve the detection accuracy and efficiency of aluminum plate defects,reduce production costs,and speed up the automation construction of production lines,developing an objective,efficient,accurate,low-cost,and easy-to-maintain defect detection system for stamping aluminum plates has high economic benefits and social value.With the improvement of image sensor and computer performance,the development of digital image processing technology and machine learning theory,machine vision detection system has been widely used in the industrial production.According to the current development of machine vision,the detection and classification algorithms of stamping aluminum plate defects are discussed in this paper.And an automatic defect detection and classification system is designed and implemented based on the requirements of real-time,accuracy and robustness.The research contents and achievements of this paper are as follows:1)In the production workshop,there are a lot of dust and debris that can not be removed in time.When they are attached to the surface of the aluminum plate,it will bring some noise similar to the defects in the image,which will affect the defect detection.rate.In order to reduce the negative impact of the noise and obtain the location information of the target area,a complete pre-processing scheme is proposed.First,combine with the fast median filter and mathematical morphology to shield the false defects caused by dust and debris.Then achieve image binarization while retaining details of the images as much as possible by using the maximum entropy threshold segmentation algorithm.Finally,mark the target location by image area analysis.2)Image feature description is an important basis for detect detection and classification.The quality of feature selection directly affects the results of defect detection.Therefore,the features that are generally required to be selected should have the characteristics of high sensitivity,strong independence,and fast operation.By analyzing the defect imaging characteristics of stamped aluminum plates,Hu moment,SITF algorithm,HOG algorithm and PHOG algorithm are selected for analysis and comparison.Aiming at the shortcomings of the traditional PHOG algorithm,an improved PHOG algorithm is propose by introducing the overlapping sampling mechanism,while obtaining spatial structure and gradient information.The algorithm describes the relationship between local pixels.3)Traditional computer vision can form a powerful and effective specific defect detector by combining various algorithms.However,each step needs to be constructed and analyzed,so the research and development cycle is relatively long,and the pertinence is strong and the scalability is weak.Deep learning provides powerful detectors in statistics,which can relatively easily train the models to detect defects without the requirment for tedious feature engineering.In this paper,DenseNet,which has strong resistance to network degradation,is taken as the research object to explore the feasibility of deep learning for defect detection of stamping aluminum plates,and to pave the way for the subsequent project realization.4)In order to complete building the hardware and realizing the software function of the machine vision system,the appropriate defect detection algorithm is selected based on comprehensive detection performance and implementation efficiency,and the hardware components are selected according to the cost performance and easy maintenance.This system is based on Visual Studio and OPENCV platform to finish software design,with a good human-computer interaction interface.It can help operators to achieve convenient and fast automatic detection of aluminum plate defects.The final system test results show that the average detection accuracy of the system for the defects on stamped aluminum plates is up to 97%,and the classification accuracy of each defect reaches is more than 96%.That means the designed system has high reliability and accuracy.