一种基于非线性主成分回归的过程监测及故障诊断方法

近年来，统计过程监测方法在多变量过程监测领域得到广泛应用，但对于存在显著 非线性的过程，这类方法的性能往往不尽人意，而神经网络在处理非线性问题上具有卓越的 优势．本文将多变量投影方法和径向基神经网络良好的逼近能力结合起来，提出了一种基于 嵌入径向基网络的非线性主成分回归算法的过程监测及故障诊断方法．在三水箱实验装置上 进行的实验结果说明该方法确实能够有效地实现过程监测、快速地检测并诊断出故障状态．

PROCESS MONITORING AND FAULT DIAGNOSIS BASED ON A NONLINEAR PRINCIPAL COMPONENT REGRESSION METHOD

Statistical process monitoring (SPM) is widely used to improve process operation performance and produce consistently high-quality products for multivariate processes. However, existed SPM methods are frustrated by the dramatic nonlinear feature of some processes. Neural network is the most common tool dealing with nonlinear problems. This, a new process monitoring and fault diagnosis method is presented based on nonlinear principal component regress (NLPCR) algorithm integrated with radial basis function (RBF) network, which will evidently combine the advantages of multivariate projecting technology and the excellent approximation of RBF network. The results of the experiment indicate that the method is capable of monitoring the whole process operation performance, detecting the occurrence of abnormal conditions and then giving correct diagnosis result quickly and accurately.