摘 要
作为钢铁工业的主要产品形式之一, 钢板已经成为汽车、机械制造、化工、航空航天和造船等工业不可缺少的原材料, 因此其表面质量的优劣将直接影响这些产品的最终性能和质量。在钢板制造过程中, 由于连铸钢坯、轧制设备、加工工艺等多方面的原因, 导致钢板表面出现裂纹、氧化皮、结疤、辊印、刮伤、空洞、针眼、折叠、表面分层等不同类型的表面缺陷, 这些缺陷不仅影响产品的外观, 更严重地降低了产品的抗腐蚀性、耐磨性和疲劳强度等性能。
神经网络理论作为一门新兴学科, 近年来被广泛应用于钢板的表面缺陷检测中, 但大部分都是基于BP 神经网络进行识别的, BP 网络的缺点是对于干扰量非常敏感、隐含层节点数需要经过多次尝试、学习速度慢且易陷入局部极小点。
本文使用RBF 神经网络( RBFNN) 对钢板缺陷数据集进行计算机仿真, 将缺陷识别结果与期望结果进行对比。 由于其结构简单, 每个隐含层节点映射函数是不同的非线性函数, 所以可以保证快速的学习速度, 并且是全局最优。
关键词：径向基函数（RBF ）；人工神经网络；钢板表面缺陷；缺陷识别
Abstract
As one of the major forms of production in steel industry，steel sheet has now become a kind of indispensable material for many industries like automobile ，mechanical ma nufacturing ，chemistry ，aviation ，astr0navigation ，ship building，etc ．Its surface quality affects the final production ’S perform ance and quality. In steel manufacturing process, because continuous casting billet, rolling equipment, processing technology and so on various causes, the steel plate surface occurrence crackle, descaling, scarring and roller and scratches, hollow, needle, folding, surface layer etc of different types of surface flaws that not only affect the product's appearance, the more severely reduced products corrosion resistance, abrasion resistance and fatigue strength of performance.
Neural network theory, as a new subject, widely used in recent years in the steel sheet surface defects detection, but mostly based on BP neural network to identify , BP network drawback is for interference quantity quite sensitive, hidden nodes several need after many attempts, the slow learning speed and easy into the local minimum points.
The paper uses RBF neural network (RBFNN) to recognize the fault dataset of iron sheets using computer simulations. The recognized results are compared with the expected outputs. Due to its simple structure, a variety of mapping functions, and rapid learning speed, RBFNN can achieve a global optimization.
Keywords: Radial Basis Function (RBF); Artificial Neural Network; Surface Defect; Defect Recognition
2