基于混合型判别分析的工业过程监控及故障诊断

工业过程数据具有规模性大、复杂性高、变量多、关联性强等特点.如何从数据出发准确并快速地发现故障并处理, 保证过程高效运行意义重大.本文针对复杂的工业过程, 提出了一种多方法结合的混合型过程监控与故障诊断方法, 完成数据分类, 构建故障模型库, 故障在线诊断及可视化相关处理.首先通过常规主成分分析(Principal component analysis, PCA)方法对历史数据进行初筛, 区分出正常和故障信息, 然后利用聚类方法对故障数据集进行分类, 接着利用局部线性指数判别分析方法(Local linear exponential discriminant analysis, LLEDA)建立故障模型库进而进行故障诊断.本文将基于监督学习的LLEDA方法拓展到无监督学习, 便于复杂工业大量无标签数据的处理.最后利用典型的田纳西伊士曼(Tennessee Eastman, TE)过程对所提出的方法进行有效性验证.

Industrial Process Monitoring and Fault Diagnosis Based on Hybrid Discriminant Analysis

Industrial process data has the characteristics of large scale, high complexity, multivariable and strong correlation. There is great signiflcance on how to flnd out and deal with the fault from data accurately and quickly, which can help ensure the e–cient operation of the process. The paper proposed a hybrid multi-method process monitoring and fault diagnosis framework for the complex industrial process. The framework include the data classiflcation, model library establishment, timely diagnosis. Firstly, the historical data is simply screened by principal component analysis (PCA) methods to distinguish normal and fault information. Then the clustering method is used to classify the fault data set, and the fault model libraries are established by local linear exponential discriminant analysis (LLEDA) method. Finally, the fault diagnosis is carried out. The LLEDA method based on supervised learning is extended to unsupervised learning, which facilitates the processing of a large number of unlabeled data in complex industries. Finally, a typical Tennessee Eastman process is used to verify the efiectiveness of the proposed method.