Research And Development Of Silicon Ingot End Face Crack Detection System Based On Machine Vision

With the continuous growth of market demand for solar photovoltaic cells,it is becoming increasingly important to achieve high-quality and efficient automatic production of silicon ingots as raw materials.The phenomenon of hidden cracks on the end face of silicon ingots caused by the cutting process in the automatic production process of silicon ingots can seriously affect the machining efficiency of subsequent processes.Therefore,detecting the end face of the silicon ingot after the cutting process and removing the cracked silicon ingot from the automatic production line as soon as possible is an important step in achieving high-efficiency automatic production of the silicon ingot end face.At present,the main method for detecting and removing end face cracked silicon ingots is through manual observation and manual handling.This method not only increases labor costs but also has latency and danger.In order to further improve the efficiency and safety of automatic production of silicon ingots,relevant research has been conducted on automatic positioning algorithms for silicon ingot end faces and hidden crack defect detection algorithms based on machine vision technology.Finally,a set of automatic detection system for hidden cracks on the end face of silicon ingots was designed through the joint programming technology of C # and Halcon.The main research content is:(1)Aiming at the complex background of silicon ingot end faces,a two-step classification and fitting method was used to achieve automatic positioning of silicon ingot end faces.Firstly,use the Ramer algorithm to remove all line contours after Canny edge detection,and then complete the first classification by removing short contours based on the length of the lines to obtain long curved contours.Use the support vector machine algorithm for the second classification of the remaining contours to obtain the end face contour of the silicon ingot.The first fitting uses a random sampling consistency algorithm to fit a circle,and then the circle is used to reverse locate all possible points in the original contour on the edge of the silicon ingot end face.For the second fitting,the least square algorithm is used to fit the ellipse,and the complete silicon ingot end face image is obtained through the fitting ellipse,so as to realize the automatic positioning of silicon ingot end face.(2)An improved background difference algorithm is proposed to address the issues of uneven distribution of light on the end face of silicon ingots and large span of gray values.Starting with the gray value of the picture,the gray distribution of individual row and column pixels is analyzed.The first derivative and second derivative of row and column gray changes are used to establish the background model.By subtracting the preprocessed image from the background image,the interference caused by uneven light distribution is removed.Finally,the image segmentation is completed using the maximum inter class variance method.To address the issue of some normal areas being included in the segmented area,analyze the characteristics of the normal and hidden crack areas included.Select 11 feature values from each region based on their characteristics.Finally,the area is classified by a multi-layer perceptron classifier based on neural network to determine whether the end face of silicon ingot is cracked.(3)In response to the testing requirements,the selection of industrial cameras and industrial control computers in the hardware part has been carried out.Then,the overall architecture of the software system was analyzed,and the detection algorithm recognition module,data management module,real-time communication module,and human-machine interaction management module were designed.Finally,it is implemented through joint programming with C # and Halcon,with various functions concentrated in the upper computer detection system.