Research On Surface Defect Detection For The Parts Based On Machine Vision

Defect inspection is the important means to guarantee the safe use of parts. Thetraditional detection is completed by manpower which with big workload andsusceptible to effect by personnel subjective factors, so it is difficult to ensure theefficiency and precision detection. While the machine vision inspection can improvethe production efficiency and automation level.Based on common axial parts and gear parts, the main research are consist ofbackground image segmentation, parts surface defect extraction, defect’s featureparameters extraction, classification and identification.This paper firstly put forward the background image segmentation of dualstructure element morphology based on detailed analysis of parts. This method applyboth linear and fractal structure elements for opening and closing the operations. Thealgorithm combines the advantages of both linear and fractal structure elements thatthe improved algorithm can better eliminate the image background, and get the clearand complete parts which showed by the theoretical analysis and simulationexperiments.A new segment algorithm has proposed for detect the defect on the surface of thepart in this paper. The wavelet transform theory is applied in the Otsu algorithm. Weuse the method of morphology to remove the background of the image, after that wechoose single-deck wavelet coefficient to decompose the image. Then we reconstructthe low-frequency component of the decomposed image, so as to remove theredundancy information and noise. At last, we use Otsu threshold segment to extractthe defect. The experiment proves that, our algorithm is better than the classic Otsualgorithm. It has the advantage of the high accuracy for segmentation and good noiseimmunity. In addition, we have taken the after-treatment for the extracted defect. Themethod of morphology is applied for filling the cavern defect. The algorithm forCanny edge detecting based wavelet transformation is proposed. It not only removesthe affect of the noise, but also detects the edge which has defect.The paper proposes a method of defect feature parametric detecting. We design asuitable BP neural network sorter for this paper. First, we extract23kinds of textureand geometric features. Then we propose the part quality grade evaluation standard. Inorder to reduce the compute complexity, we use PCA to reduce the dimensions of thefeature parametric and build the corresponding database. With the feature parametricas the BP neural network’s input, we design a BP neural network sorter of singlehidden lever structure. Then we choose905test samples to validate the recognizer.The results show that the recognition rate is86.2%. At last, the paper makes a total design for the defect detecting system. Thekeynote is to design the detect system software. The total software system includesinput module, defect post-processing module, defect classify module and part qualitygrade evaluation module. Then we discuss the function of the sub-module included inevery module. Through the total design of the system, we classify the defect of thepart and evaluate the quality grade. The experiment shows that, this system has gooddetecting performance.