Research And Implementation Of Intelligent Solder Joint Defect Detection Algorithm

With the development of welding technology, the miniaturization of component brings that the count of components per square centimeter has increased sharply. It brings great challenges for the PCB quality testing. Pads defect detection is one of the detection circuit boards for ensuring the quality of the content is important. Simply rely on manual testing cannot meet the demand of industrialization. In this context, a method of low cost and high detection accuracy has very important significance.This paper aims to provide a solution of obtaining the circuit board pictures by high-definition digital camera under the condition of general light and test the PCB solder. The handling of the circuit board can be divided into three parts. One is the PCB image preprocessing. The PCB tilt and shape distortion is corrected in this part. The second part is the acquisition of the solder joint images. The third part is the extraction of solder joint features and the solder joints are classified by neural network.Circuit board pretreatment aims at using the Hough transform detection center position of circuit board location hole and comparing it with the actual position of PCB location holes to obtain the correction of the circuit tilt and shape distortion. The mapping function of PCB image is obtained by the relation of the image coordinates of the location hole and the PCB coordinates of that.To extract the solder joint, firstly, the location of solder joints and shapes are able to be obtained by reading and parsing gerber file and they are mapped to image coordinates by the mapping function. There will be an error of 3-5 pixels. So the scope of the solder joints must be expanded and the most likely position of solder joint will be found by the algorithm of template match in the image.The feature of solder joint is extracted by using the discrete cosine transform, the discrete wavelet transformation and the log Gabor transformation. The DCT samples and analyzes the images in frequency domain. The wavelet transform and the log Gabor transform can sample and analyze the images in both spatial domain and frequency domain. The three kinds of algorithm extract the features of solder joint image in different aspects. This paper tests and compares the performance of the classifier in the three algorithms. So the paper concludes that the log Gabor transforms has the better performance.As this paper describes, the detection system is able to obtain better detection result under the condition of low cost and achieve high detection accuracy and high testing speed. It provides a feasible implementation for industrial solder joint detection system.