| Power transformer plays the role of voltage transformation and power distribution in the circuit.It is the key component of electrical equipment.Its quality is directly related to the reliability and safety of power system.At present,the defect detection of power transformers in factories mainly depends on manual sorting.This method is not only poor in accuracy and efficiency,but also affected by the degree of fatigue and technical experience of workers.Combining with machine vision technology to inspect the appearance quality of power transformer can improve the accuracy of defect identification,shorten the detection time and work for a long time.Therefore,in this paper,a power transformer defect detection system based on machine vision is developed according to the characteristics of power transformer defects.The main research work of this paper is as follows:(1)Analyze the appearance defect characteristics of power transformer,draw up the design requirements of the project,formulate the overall working scheme of machine vision inspection system,research and design the image acquisition hardware platform,and construct the geometric transformation relationship between the world and pixel coordinates.(2)The process of image preprocessing is studied.The Freeman chain code is used to extract the contour of binary image.The minimum circumscribed rectangle and affine transformation are combined to realize the skew correction of the image.The median filter is used for smooth denoising.In the stage of image enhancement,an improved gamma transform algorithm is proposed,which introduces an automatically adjusted transformation coefficient,and classifies different initial image gray values and light intensity to achieve adaptive image enhancement.(3)The process of image threshold segmentation is studied.An improved Otsu algorithm combined with two-dimensional Renyi entropy for twice threshold search is proposed.The processing time of the algorithm is reduced by limiting the traversal range of gray value and accelerating the calculation of integral image.The experimental results show that the improved algorithm has more advantages in segmentation accuracy and processing speed.According to the defect characteristics of power transformer,the corresponding detection algorithm is designed.Through the open operation and contour detection,the printed code area is extracted,and then the character image is segmented by combining the vertical projection method,and the optimized normalization algorithm is used to normalize the image.Finally,the convolution neural network is constructed for character recognition to realize the defect detection of printing code leakage and spurting Through the specific threshold segmentation and contour area comparison,the defect detection of qualified label is realized,and the scratch spot defect detection is realized by extracting the contour of connected domain and judging the defect shape.(4)The upper computer software platform of the detection system is designed and implemented.The camera calibration,defect detection,information query and other functions are developed.Corner extraction and calibration calculation are carried out for different orientation calibration board images,and the internal and external parameters of the camera are solved,and the world pixel coordinate transformation equation and pixel equivalent are obtained.The experimental results show that the overall detection accuracy of the system is 93.84%,and the average time is 2.213 s,which meets the design requirements of the system. |
