GOALI: Development of a Novel Subspace Approach to the Monitoring and Control of Chemical Processes

一个通用的理论框架，故障检测和诊断将发展。主要内容包括故障识别、故障重建和选择合适的模型维度。提出了一种新的子空间控制方法，可以实时对故障或干扰进行校正。故障检测将通过监测两个检测指标来完成。平方预测误差指数检测残差子空间的显著变异性，Hotelling T2指数表明超出历史极限的操作。提出了一种确定最佳故障构造主成分数的客观方法。将使用多元统计方法，包括主成分分析和偏最小二乘法对数据进行分析，并将测量空间划分为正态子空间和异常子空间。由于高达40%的制造成本花费在维护和故障情况管理上，因此该项目将具有潜在的应用于一般工业重要性问题。如果成功，该项目将实现以下四个目标：(1)通过故障重建和识别开发一种新的根本原因分析方法；(2)为确定主成分数量提供客观、完善的标准；(3)为子空间控制和监测提供综合方法；(4)建立故障检测、识别、重建和确定主成分数量的基本理论。

A general theoretical framework for fault detection and diagnosis will be developed. The main elements include fault identification, fault reconstruction, and selection of the appropriate model dimensions. A novel subspace control approach that can provide corrections to faults or disturbances in real time will be developed. Fault detection will be accomplished by monitoring two detection indices. The squared prediction error index detects significant variability in the residual subspace, and the Hotelling T2 index indicates operation beyond the historical limits. An objective approach to determining the number of principal components for best fault construction is proposed. Multivariate statistical methods, including principal component analysis and partial least squares, will be used to analyze the data and partition the measurement space into a normal and an abnormal subspace. Because up to 40% of manufacturing costs are spent on maintenance and fault situation management, this project will have potential application to a problem of general industrial importance. If successful, the project will accomplish the following four goals: (1) develop a novel root-cause analysis approach by fault reconstruction and identification, (2) provide an objective, well-established criterion for determining the number of principal components, (3) provide an integrated approach for subspace control and monitoring, and (4) establish a fundamental theory for fault detection, identification, reconstruction, and determination of the number of principal components.