Surface defect detection of machined parts based on machining texture direction

Abstract

Compared with detecting the regular texture of fabrics and prints, the detection of the processed texture on the surface of mechanical parts is more difficult. To quickly and accurately detect defects caused by abnormal machining of the surface of metal parts, a one-shot machine-vision method based on a texture orientation histogram is proposed. An improved Mean-C local threshold method is proposed to solve the problem of difficulty in extracting surface texture. Using the minimum enclosing rectangle, the skeleton texture is extracted from the enhanced image obtained by the improved Mean-C local threshold. The statistical information from the histogram is used to pre-process the texture direction, and then a novel angle region growth method proposed in this paper is used to search the main texture cluster and the abnormal texture cluster of the part, so as to realize the product quality detection. Experimental results show that this method is highly targeted for the detection of surface texture defects caused by abnormal processing, which is equivalent to the average performance of a multi-angle illumination detection system, but much faster. This detection method has high detection efficiency, high accuracy, and strong robustness, and can meet the requirements of industrial detection.