Research On The Technology Of Character Recognition In Piston Cavity And Defect Detection Of Piston External Surface

With the rapid development of the automobile engine manufacturing industry,the standards for precision and quality in engine piston manufacturing are becoming higher and higher.In order to keep the quality of the piston before leaving the factory,piston quality inspection has become a key part of piston manufacturing.At present,the mainstream piston quality inspection methods are tested by naked eye or contact.But these methods have some disadvantages,such as large subjectivity,low detection efficiency and unintuitive display.Therefore,how to achieve low-cost,high-efficiency,high-precision and non-contact piston quality inspection has become a concern of the piston manufacturing industry.According to the requirements of a piston manufacturing company in Shandong,our research group conducted a series of research on high-precision dimensional measurement and defect detection of engine pistons based on machine vision.At present,the research group has realized the key technical analysis of high-precision dimensional measurement.On the basis of that,we conduct research on key inspection tasks such as defects detection of the piston external surface and character recognition in piston cavity,design a piston mechanical station,and eventually design and develop a high-precision dimensional measurement and defect detection software system for piston.The main work is as follows:1.A character recognition method in piston cavity based on Faster R-CNN and character sequence prior knowledge library is designed.We consider that the characters in the piston cavity are protruding upward and the metal being easily reflective,we design a ring light source and an imaging device for the piston cavity to collect the characters images and make a dataset of the characters in the piston cavity.Because the background of the piston cavity is noisy,we use Gaussian filtering and morphological operations to suppress the noise then obtain a relatively clean image of the piston cavity.According to the characteristic that the character aspect ratio is not definite,we use Faster R-CNN with appropriate parameters to train the dataset to obtain the character recognition model.The character types and position information detected by the character recognition model are used to form the character sequences in piston cavity.However,due to the problem of data imbalance in the dataset samples,the recognition model trained only by Faster R-CNN has a high error rate.In order to solve this problem,we construct and use a prior character knowledge library of piston cavity character sequences to correct the recognition results of Faster R-CNN according to the highly similar characteristics of the character sequences of the piston cavity of the same models.This method effectively improves the accuracy of the character recognition of the piston cavity by combining the deep learning network and the prior character knowledge library of piston cavity character sequences.2.A defect detection method based on the comparison of grayscale features is designed for external surface defects detection of piston.The detection of external surface defects can be divided into outside surface defect detection and top surface defect detection.Because the images of the external surface of the same type of piston collected under the same light and shooting angle are similar,we propose a method that uses the outer contour registration of the piston to obtain the same area on the image space of the standard piston and the piston to be tested.And then we perform grayscale feature comparison on the same area to detect defects.Firstly,under a specific and constant shooting environment,the standard pistons of different models are photographed to build a template library,which mainly includes images of the piston side surface images at different viewing angles and the piston top surface images at the same viewing angles.Secondly,the corresponding template piston image is picked according to the shooting angle and shooting position information of the piston image to be detected,and the piston outer contour registration is performed using the Scale Invariant Feature Transform(SIFT)algorithm to determine the same area in space of the two images.Finally,we use the sliding window to traverse the same area after registration of the two images at the same time,and extract the grayscale features in the window for comparison to determine whether there is a defect in the window of the image to be detected.This method can accurately find the same area of the piston to be detected and the template piston by using SIFT outer contour registration,so that the result of gray comparison is more reliable.3.The hardware and software involved in the piston high-precision dimensional measurement and defect detection are analyzed and designed.In terms of hardware,according to the specific contents of each test item of the project,the mechanical stations used for piston testing were analyzed and designed,and finally two mechanical stations were obtained.These two mechanical stations can meet the requirements of all test items and save production costs.In terms of software,a detailed analysis of the system user's various needs is made first.Then,the system design and interface design are performed according to the detailed requirements obtained from the analysis.Finally,the piston high-precision dimensional measurement and defect detection system are realized by coding,and the existing detection algorithm is integrate into the system.The research results of this thesis combined with the actual mechanical station can realize the character recognition in piston cavity and the defect detection of piston external surface.We build a software system platform which can effectively improve the efficiency of various piston quality inspection items.