The Study On Algorithm And Implementation Of Bearing Quality Online Detection

Bearings is very important and widely used in machinery industry.It needs prodution volume and high precision.In bearing production and use,It is often detected in order to ensure that it can be used properly.Most bearing manufacturers adapt contacting inspection methods that is manual detection methods.The method is slow and easy vulnerable to detect by subiective factors. there is damage to the bearing’s quality, especially in surface quality.So it is not suitable for large-scale automated production and not conducive to the bearing components of detection in the work environment.Based on the above issues, this paper aims to study a non-contact detection of bearing quality. On the quality of the image detection theory, this method not only avoids the disadvantages of contact detection, and automatic detection because of its nature, non-labor intervention, with high-speed, high precision, automatic and so on.It Meets the needs of today’s society of mass production.At present, some of the defect detection methods based on the product image summed up in two categories:the first is based on gray-scale image information to determine product quality of the product, these methods simply use a single threshold value method and defect detection, but may lose some defect information; second category is based on the product image texture information to determine product quality, speed and type of method in the detection of defects in the information on the cluster, there are some deficiencies.Combined the characteristics of the bearing images, use the least squares method and bearing parameters prior knowledge to quickly locate the split bearing. Based on the detection of of repeatedly OSTU algorithm, a good solution to the above two algorithms for a single threshold and multi-threshold algorithm in the detection of defects, drawbacks, and eight-connected domain law bearing on the treatment image defects extract. Use of moment invariants and the relief algorithm to extract and filter characteristics to reduce the number of texture feature extraction in practical applications, so that the computing speed, and finally using BP artificial neural network on the texture feature information extracted from the cluster analysis, the results show the reliability of the method. Finally, bearing detection system hardware structure, and the development of detection’s procedure.