| “Zero-defect manufacturing” has become the development direction of product manufacture. Online detection based on machine vision as a full product testing emerging technology, due to its intelligent, non-contact, fast, accurate characteristic, has been recognized and valued by many enterprises. However, the complication in the identification of machine vision algorithm of precision analysis has seriously affected the machine vision’s wide application. Therefore, this thesis proposed a optimization algorithm of sheet parts hole detection for the parameter selection of visual online detecting hole algorithm.Sheet part hole detection is one of the most frequently features detected. The current algorithm transformed by Hough is the most classic hole detection algorithm. It has been made as an internal standard algorithm in OpenCV, including six undetermined parameters. For different parts, the parameters must be adjusted to meet the needs of the detection accuracy. However, there has not yet any mature method and rule to choose the parameters of detecting circle algorithm. Therefore, firstly, this thesis used genetic algorithm to build a single-objective optimization model for optimization of Hough circle transforming algorithm parameters. On this basis, this thesis further applied the non-dominant Sorting Genetic Algorithm(NSGA-II) to establish a multi-objective optimization model. Last, to calculate the optimal detection hole parameters by the model.Firstly, the test platform of online detection system based on machine vision was designed and built through the selection of the hardware system and software system in this thesis. According to the accuracy and vision detected sheet parts required, visual test bench, industrial cameras, lenses, lighting, calibration plate and other equipments were selected. The software detection system was developed based VS2010 software development platform, OpenCV and HALCON technology. Then, this thesis analyzed the six undetermined parameters in Hough circle detection algorithm, selected one of the six parameters which has the most significant effect on detection results as a decision variable, applied the difference between the results of the circle detection and the measured size as the fitness function. According to the above conditions, a single-objective optimization model of Hough circle transforming parameters was established. Last, the optimal parameter was calculated by genetic algorithm. The single-objective optimization model based on GA can only meet a part of testing requirements. We need to profoundly study the parameters optimization algorithm to meet the requirements of multiple detection. So a multi-objective optimization model based on NSGA-II algorithm was proposed in this thesis. This model selected the minimum resolution image of the accumulator as the decision variable, and increased the fitness function to large diameter hole deviation, small hole diameter deviation and the deviation of sum of hole. Then the NSGA-II algorithm was used to optimize decision variable. Finally, the results of parameter optimization were applied to the case detection of the rear bumper beam. The detection results proved that the optimized parameters could meet the testing requirements. |
PDF Full Text Request