The permanent increase in size and complexity of the modern technical systems generates increasing demands on process control and monitoring, as well as on fault diagnosis (fault detection and isolation) as a specific monitoring task. The existing fault isolation algorithms have become computationally very challenging and most of them face the isolability problem.
The dual algorithm proposed in this thesis targets the isolability problem in order to improve the fault diagnosis performance metrics. The Weighted Sequential Fault Diagnosis algorithm improves both steps in fault diagnosis by using the information about the relative degrees of influence of process variables to modify the model structural matrices, yet without altering the model structures themselves. The Model-on-Demand algorithm for fault isolation utilizes a reverse modeling approach to modify the used model structures in order to directly improve the isolability property. The efficiency of the proposed algorithm has been successfully verified through its experimental application in two different large-scale technical systems.