UBC Theses and Dissertations
Fault detection and isolation using the local approach Cheng, Lechang
Fault detection and isolation (FDI) has become a crucial issue for industrial process monitoring in order to increase availability, reliability, and production safety. Model-based FDI methods rely on the mathematical model and input-output data of a process to perform detection. The local approach is a new model-based FDI method which aims to detect slight changes of parametric properties of a system. This thesis mainly addresses to the application of FDI using the local approach. Robustness with respect to model uncertainties is an important issue for the local approach. A new algorithm was proposed to recalculate threshold based on the original threshold and covariance matrix of the estimated parameters in order to reduce false alarms due to the estimation error of process parameters. A similar algorithm was also provided to recalculate threshold to reduce fault alarms due to regular parameter fluctuations. As fault detection algorithms are often applied to closed-loop data, closed-loop fault detection was also investigated. Two methods were proposed to deal with the relevance between system input and output data in closed-loop detection: the dimension reduction method and the indirect detection method. The dimension reduction method uses a linear transformation to reduce the dimension of the normalized residual so that the covariance matrix of the revised normalized residual has full rank. The indirect detection method uses the closed-loop model to calculate the primary residual and the normalized residual. By detecting the changes of the closed-loop parameters, the method also detects the changes of the open-loop parameters. Simulation results show that both of these methods can detect changes of every single parameters of a system. Industrial data from a cross-direction (CD) control system in a paper machine was also used to assess the applicability of the local approach. By dividing the CD databox into small sections, the sensitivity of the detection algorithm was improved and the algorithm successfully detected abrupt faults of a single actuator. However, incipient faults of a single actuator can not be detected due to noise and inaccuracy of the process model.
Item Citations and Data