Study On Defect Detection Of Chip On PCB Before Reflow Based On Vision And Neural Network

In the process of electronic assembly,pre-reflow automated optical inspection machine is an important equipment to detect defects for chip on printed circuit boards.However,when using the existing pre-reflow automated optical inspection machine,if a bad product is produced,it is necessary to stop the machine,manually analyze the fault type,and manually adjust the equipment parameters.Failure to effectively combine the detection equipment with the surface mount machine to form an automatic feedback system.To realize this feedback system,the pre-reflow automated optical inspection machine is required to quantitatively output the deviation of the component position and attitude on the basis of detecting the type of defect.This puts forward new requirements for the pre-reflow automated optical inspection machine.This article combines vision technology and neural network technology,and proposes a set of defects detection scheme for chip on printed circuit boards before reflow.The scheme realizes the quantitative output of chip position and attitude deviation while detecting the defects of chip on printed circuit boards before reflow.The main work and innovations include the following parts:(1)Aiming at the problem that training neural network needs a large number of samples,this paper proposes a set of sample extraction and data enhancement scheme.The scheme answers the questions of how to obtain high level image,how to realize image registration,how to extract detection area and how to realize data enhancement.The fast generation from the initial image to the required data set of neural network is realized.(2)This paper proposes a method for detecting defects of chip on printed circuit boards before reflow based on global grayscale features and local texture features.This method achieves a defect classification capability of 99.81% accuracy and a pose calculation capability of 92.00% accuracy.When calculating the chip displacement deviation,this paper proposes an improved template matching method with higher accuracy and faster detection speed.This method improves the detection accuracy when the target element is at the edge of the image by calculating a more accurate mask;Through the detection,clustering,filtering and expansion of the linear features on the chip,the prior knowledge of the rotation angle of the chip is obtained,which significantly reduces the number of matching templates and improves the matching efficiency;(3)This paper proposes a detection method of chip defects on printed circuit boards before reflow based on classification and regression neural network.Under stricter standards,this method achieves a defect classification capability of 99.60% accuracy and a pose calculation capability of 99.98% accuracy;(4)This paper designs a prototype system for defect detection of chip on printed circuit boards before reflow.The system integrates the algorithms proposed in this paper,which verifies the feasibility of this scheme.