轴承品质在线检测算法研究与实现

发布时间：2018-11-20 20:35

【摘要】：轴承是机械行业中非常重要且应用十分广泛的转动部件，其生产批量大，精度要求高。在轴承的生产和使用中，为了保证轴承产品生产和使用正常，需对半成品或者成品轴承进行检测。目前大多数轴承生产厂家采用接触式检查方法即人工检测方法。该方法检测速度不仅慢，而且检测者主观因素会影响检测结果，影响工件质量，尤其是表面质量，，在大规模的自动化生产中存在弊端，对处于工作环境中轴承部件的检测也同样不利。针对上述问题，本文旨在研究一种非接触式的轴承品质检测方法，即基于图像理论的轴承品质检测，这种方法不仅避免了接触式检测的弊端，而且由于其能自动检测，非人工干预，具有高速、高精度、自动等特点，符合当今社会大生产的需求。目前，已有的一些基于产品图像缺陷检测方法概括起来，有两大类：第一类是基于图像产品的灰度信息判断产品的好坏，这类方法简单的用单阈值法将产品与缺陷信息分割，但可能会丢失部分缺陷信息；第二类是基于图像产品的纹理信息判断产品的好坏，这类方法在检测速度和对缺陷信息聚类方面存在一些不足。结合轴承图像自身特点，本文利用最小二乘法和轴承相关参数等先验知识快速定位分割轴承。针对单阈值和多阈值算法在轴承检测上的不足，提出了一种基于多次OSTU算法的轴承检测方法，它很好地解决了前述两种算法的弊端，采用八连通域法对处理后的轴承图像进行缺陷提取。研究利用不变矩和relief算法提取并筛选特征，减少在实际应用中纹理特征提取的数量，使运算速度加快，并利用BP人工神经网络对提取出来的纹理特征信息进行聚类分析，结果证明了该方法的有效性。最后给出了轴承检测系统的硬件结构组成，并实现了检测程序的设计和编制。
[Abstract]:Bearing is a very important and widely used rotating component in mechanical industry. In the production and use of bearings, in order to ensure the normal production and use of bearing products, semi-finished or finished bearings should be tested. At present, most bearing manufacturers use contact inspection method, that is, manual inspection method. The detection speed of this method is not only slow, but also the subjective factors of the examiner will affect the inspection result and the quality of the workpiece, especially the surface quality. The detection of bearing parts in the working environment is equally unfavorable. In view of the above problems, this paper aims to study a non-contact bearing quality detection method, that is, bearing quality detection based on image theory. This method not only avoids the disadvantages of contact inspection, but also can detect the bearing quality automatically. Non-manual intervention, with high-speed, high-precision, automatic and other characteristics, in line with the needs of large-scale production in today's society. At present, some existing defect detection methods based on product image are summarized. There are two kinds of methods: the first is based on the grayscale information of image product to judge the quality of product, this kind of method simply uses single threshold method to segment the product and defect information. However, some defect information may be lost; The second kind is based on the texture information of image products to judge the quality of products. This method has some shortcomings in detecting speed and clustering defect information. According to the characteristics of bearing image, this paper uses the prior knowledge such as least square method and relative parameters of bearing to locate and segment the bearing quickly. Aiming at the shortcomings of single threshold and multi-threshold algorithms in bearing detection, a bearing detection method based on multiple OSTU algorithm is proposed, which solves the disadvantages of the two algorithms well. The eight connected region method is used to extract the defect of the processed bearing image. In order to reduce the number of texture feature extraction in practical application and speed up the operation, the invariant moment and relief algorithm are used to extract and filter features, and BP artificial neural network is used to cluster the extracted texture feature information. The results show that the method is effective. Finally, the hardware structure of the bearing detection system is given, and the test program is designed and compiled.
【学位授予单位】：江南大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH133.3;TP274

【参考文献】