(66c) Risk-Based Fault Detection and Diagnosis

This paper presents a methodology to calculate process risk in combination with fault detection methods. This methodology aims to identify and screen the faults which are not safety concerns and also to dynamically update process risk at each sampling instant.

In general, fault detection techniques have been developed to detect operational faults that affect the control objectives of the process. Both model-based fault detection and history-based methods have been used for process fault detection and diagnosis. Although these techniques have some advantages in fault detection, none of these methods take into account the potential impact of the fault on the safety of process, personnel, and the environment; as such warnings can be generated for trivial changes in the process which do not have a substantial impact on process safety or operation. Thus, these methods often generate spurious alarms which can lead to alarm flooding in the control room.

To address this issue risk-based fault detection and diagnosis methods are proposed which could build upon both model-based methods such as Kalman filter and particle filter, and history-based methods such as principal component analysis.

This paper presents combination of dynamic risk assessment with fault detection methods. To analyze consequences, loss functions are also proposed to relate process deviations to economic losses. The consequence module consists of identification, quantification and integration of losses for a given scenario.

This methodology uses risk profile for real-time fault detection as well as for taking any real-time supervisory decisions to activate appropriate safety systems. The main benefits of this methodology are: improved safety, minimum interruption of operation, better alarm management and early warning system and higher availability of process.