最小二乘支持向量机构造的函数链接型神经网络在滚动轴承故障诊断中的应用

Application of Functional Link Artificial Neural Networks Constructed With Least Squares Support Vector Machine in Fault Diagnosis of Rolling Bearings

提出一种用最小二乘支持向量机(least squares support vector machine,LS-SVM)构造函数链接型神经网络(functional link artificial neural networks,FLANN)的滚动轴承故障诊断系统。介绍了相关原理和具体算法,并给出了滚动轴承故障诊断系统模型。首先,采用LS-SVM模型核函数代替常规FLANN模型的扩展函数,避免了扩展函数选择的任意性;其次,利用LS-SVM学习模型得到FLANN权重系数,避免了BP方法多次迭代寻优存在的耗时长、局部极小及迭代设置初值依赖经验等不足;最后,构造了多层LS-SVM-FLANN结构,对多类滚动轴承故障进行诊断。具体实验表明,用LS-SVM构造FLANN的滚动轴承故障识别系统精度高、鲁棒性好、实现简单。

A diagnosis system for rolling beating fault is presented based on functional link artificial neural networks （FLANN） inverse system constructed with least squares-support vector machine （LS-SVM）. The principle and algorithms are introduced and the system diagnosis model for rolling beating fault is presented. First, the model avoids the randomness for choosing external function because it replaces the external function of common FLANN function with the kernel function of LS-SVM. Second, the FLANN model weight coefficients can be obtained with LS-SVM learning model, avoiding the disadvantages embodied by the BP method such as long time- consumption, local minimum, and dependence on experience while setting initial values. Last, a multi-layer LS-SVM-FLANN structure is built to diagnose the long-time consumption rolling beating faults. The result of experiment shows that the diagnosing method for rolling beating fault with LS-SVM-FLANN has the characteristics of high precision, strong robustness, and is easy to realize.