Practical Implementation of Multiple Faults in a Coupled-Tank System: Verified by Model-Based Fault Detection Methods

In order to prevent system failures, fault detection and performance assessment are always essential factors. This article investigates the practical implementation of common faults in a two-tank networked control system, one of the most widely utilized liquid-level control systems in the food industries, petrochemicals, and many more. The system was modeled and controlled, and then, faults were conducted on the system, including leakage, sedimentation, sensor bias, and pump stuck, which covered a wide range of fault types. After that, model-based fault detection methods, such as unknown input observer (UIO), parity, and factorization methods, were employed to evaluate fault implementation. By thresholding residuals of fault detection methods, the factorization method shows the most effective performance concerning false alarm rates (FAR), missed alarm rates (MAR), and detection delays (DD). Using the UIO is a suitable substitute, and parity is only appropriate for announcing the fault start time.