论文部分内容阅读
基于传统的多向主元分析MPCA(multiway principal component analysis)常会导致误诊断,且对批过程难以保证在线状态监测和故障诊断的实时性,提出了一种基于特征子空间的滑动窗主元分析方法。在实时故障监测与诊断时,该方法采用适当大小的滑动窗逐步更新当前子数据空间,对当前子数据空间故障的识别通过依次计算其与基底库中各故障的匹配度来进行。这种方法克服了传统的MPCA不能处理非线性过程和实时性问题,并避免了MPCA在线应用时预报未来测量值带来的误差, 提高了批过程性能监测和故障诊断的准确性。
Based on the traditional multi-direction principal component analysis (MPCA), which leads to misdiagnosis, and the real-time nature of on-line state monitoring and fault diagnosis can not be guaranteed for batch process, a sliding window principal component analysis based on feature subspace method. In the real-time fault monitoring and diagnosis, this method updates the current sub-data space step by step by using a sliding window of appropriate size. The current sub-data space fault identification is performed by sequentially calculating the matching degree with each fault in the basement library. This method overcomes the traditional MPCA can not deal with the nonlinear process and real-time issues, and to avoid the MPCA online applications to predict future measurement error, improve the accuracy of batch process performance monitoring and fault diagnosis.