Extraction Method Of Arc Magnet Surface Defects Based On Chaotic Characteristic

Arc magnet surface’s defects such as wear scar, crack, chamfer and other complexdefects were generated in the production process. Those defects would seriously affect theperformance, service life and safety of the motor. Now, the manual visual detection’sefficiency is low and the rejection rate is unreliable. At the present time, the study ofmachine vision in the field of defects detection gradually becomes hot at both domestic andforeign. Most of the methods are constructed on the basis of the stable and orderly data,those are not available for the variable sizes, complex background, and even faint signalsof the defects. This thesis presents an extraction and judgment method of the image basedon the characteristic information of the arc magnet surface’s defects. With the method ofcombining image processing, chaotic time series analysis, neural network and the use ofOpenCV, a test of extraction of the magnet surface’s defects was conducted, the result isaccurate. This extraction method is innovative.The paper mainly includes three parts below: image preprocessing, chaotic time seriesanalysis and defects extraction.First, the region of the arc magnet was dug out from the source image with houghtransformation and perspective transformation， and image filtering and histogramtransform is applied on the image output.Then a new method was introduced to generate the image’s time series. Throughanalyzing the Pearson correlation coefficients of the gray values of the neighboring tworows (columns) in calculating domain gray image, the time series is generated. The C-Cmethod is adopted to calculate the time series after denoising and smoothing; and thechaotic characteristic parameters, such as embedding dimension, time delay, shannonentropy and average period of the time series are obtained. Then the Wolf method is usedto obtain the largest Lyapunov exponent of the sequence. After that, various chaoticparameters of the sample are used to train and test the neural network, the defect arcmagnets are extracted according to their defect characteristics. Experiment was carried outto verify the proposed method, and the experiment results show that this algorithm leads agood classification of the defects.Finally, we extract the arc magnet surface’s defects based on the Open SourceComputer Vision Library. Aiming at the feature of wear scar and chamfer we propose anew method to calculate the threshold，and crack uses the adaptive threshold method. At last, the program returns the information of the defect’s area and position and the defectswas extracted accurately.