基于支持向量机的声发射信号裂纹识别研究

In recent years, acoustics emission has been widely applied as a new QNDE technique in numerous industries including metal grinding. In contrast to other non-destructive testing methods, the major advantages of AE detection are its economy, high-speed and no interference with the normal production, especially suitable for the grinding process of high-precision and large-size metal components. Although acoustic emission technique has been developed in the detections for workpiece-wheel contact point and wheel wear, there is still no effective solving means for workpiece crack detection and evaluation, especially for the real-time monitoring and prediction on the cracks shape, size and location etc. Traditionally, the grind monitoring task is usually carried out by the operator. It depends on his experiences to judge whether abnormal situation happens. As a result of it, hidden danger will exist and it perhaps causes damage of tools and work pieces. Therefore, there is a urgent need to develop a grinding crack identification method which can give the real-time information of cracks on the surface or inside the workpiece so as to prevent undesirable consequences. This project is based on the research work about acoustic emission grinding theory by the proposer in recent years. It is aimed to carry out the following research programs:(1)to study the generation mechanism of acoustic emission during metal grinding processes, explain the energy changes of crack acoustic emission from the perspective of molecular material and mechanics, determine acoustic emission impact factors on microscopic material and macroscopic mechanics, respectively;(2)to build grinding acoustic emission eigenvectors based on energy band distributions and grinding environmental parameters, establish the mathematical model to characterize the acoustic emission energy changes with surface cracks;(3)to carry out the theoretical and experimental works on acoustic emission crack identification based on support vector machine;(4)to find out the optimal pattern recognition algorithms to quantitatively inverse the geometric dimensions of carcks;(5)to develop a system that can integrate mathematical modeling, online learning and recognition so as to achieve the efficient crack identification and forecast.

声发射作为一种新型的无损检测技术已经在诸多领域得到了非常广泛的应用。在高精度大尺寸金属工件磨削加工中，声发射技术已经在工件与砂轮接触状态监测研究和砂轮钝化监测研究等方面都取得了一定的进展。但在工件裂纹检测方面，特别是关于裂纹的形状、大小和位置等方面的实时监测预报，目前尚无有效的解决手段。本项目深入研究磨削裂纹声发射产生的机理，从分子材料学、力学的角度研究裂纹声发射能量的变化，确定声发射微观材料学影响因子和宏观力学影响因子；结合声发射能量频段分布和磨削环境参数，共同构建磨削声发射特征向量，建立表征磨削裂纹声发射能量变化的数学模型；采用支持向量机理论，开展对裂纹声发射信号识别的理论和实验研究；探讨利用磨削声发射特征向量反演量化裂纹几何尺寸的可能性，寻找最优模式识别算法；探索将磨削声发射数学建模、在线学习和辨识一体化，实现高效裂纹识别预报的方法。

声发射裂纹检测可在磨削过程中检测出工件表面或内部隐藏的裂纹，并对裂纹的大小和方位加以判断和识别，将大大提高工件的加工质量，防止事故的发生。在大型构件的磨削加工中，实现裂纹有效检测更可以减少巨大的材料浪费，对于提升加工效率具备重要意义。本项目首先通对磨削加工中声发射产生机理进行研究，建立多磨粒正常工件磨削模型和多磨粒有裂纹工件磨削模型，对裂纹产生过程中声发射能量变化加以分析，找出与工件裂纹声发射能量变化有关的微观材料学影响因子和宏观力学影响因子；对含裂纹弹性体中的超声非线性效应进行研究，对含裂纹弹性体的有效模量进行计算，分析了静态有效拉伸模量和静态有效压缩模量随裂纹密度变化的情况，从而确定超声非线性系数与裂纹密度及粗糙程度之间的定量关系，为其后采用超声非线性检测技术无损评估微裂纹导致工件材料早期性能退化奠定理论基础；分析了材料内部裂纹方向对声波传播特性的影响，可进一步预估材料内损伤的发展情况和可能的损伤结果；基于双数复小波变换提取声发射能量频段分布特征，建立由微观/宏观影响因子、能量频率分布和磨削环境参数共同作用下的声发射信号特征向量，可为后续的状态分类、模式识别以及磨削决策系统提供有效输入，减小系统冗余；研究裂纹模式识别新方法，并针对磨削裂纹数据提出支持向量机优化算法，可快速完成小样本学习和裂纹辨别分析，辨识准确率达到90%以上。本项目的研究成果可应用于实际磨削加工过程中工件裂纹的特征分类，提出了利用磨削声发射特征向量反演量化裂纹几何尺寸的可能性；本项目提出的模式识别算法，可将磨削声发射数学建模、在线学习和辨识一体化，提供了一种实现高效裂纹识别和预测的有效手段。