In recent years, with the rapid development of China’s automotive industry, the demand for automotive gears is increasing day by day. The quality inspection of automotive gears plays a certain role in the safe driving of automobiles. Currently, gear grinding technology has been widely used in the production of automotive gears, with many advantages such as low transmission noise, high transmission efficiency, and high service life. However, if improper heat treatment and gear grinding methods are used, grinding cracks may easily occur. Therefore, effective recognition of grinding crack images of automotive gears is of great significance for improving the transmission performance of automotive gears, improving the effect of speed change and torque conversion, and ensuring the safety of automotive driving.
Machine vision uses a CCD camera to convert the detected object into an image signal and transmit it to a dedicated image processing system. It is converted into a digital signal based on information such as pixel distribution, brightness, and color. The image processing system performs operations on digitized signals to extract target features, and outputs results according to preset permissibility and other conditions to achieve automatic recognition. Therefore, machine vision can reduce the impact of optical properties, improve image processing capabilities, and complete image recognition of automotive gear grinding cracks under multiple optical paths. Guo Zhongfeng et al. obtained automotive gear images through a machine vision system, used neural networks to classify the contours of the grayed, denoised, and binarized images, and used the least square principle to complete the measurement of automotive gear parameters. This method can effectively improve the accuracy of automotive gear image classification, but it requires a large amount of data calculation and is prone to generate redundancy. Xiao Wentao and Li Dengfeng introduced information entropy to improve adaptive median filters, improved Retinex algorithm to enhance the overall effect of the image, remove image mixed noise, and use wavelet transform coefficient modulus maxima to complete edge detection of automotive gear images. This method has good image denoising effect and strong image edge detection ability, but it takes a long time to identify and detect. Sun He uses the dual K cross validation method to improve the parameter optimization of the support vector regression algorithm, improve the edge distortion compensation effect of the tooth profile image, and improve the accuracy of visual measurement of automotive gears. This method has a good compensation effect for edge distortion of tooth images, but its recognition efficiency is low and it cannot recognize all tooth profile images in a specific time.
Therefore, in order to solve the problems of long detection time and low recognition rate in the above methods, a multi optical path image recognition method for automotive gear grinding cracks using machine vision is proposed.