Incipient fault detection and isolation of sensors and field devices

The purpose of this research is to develop a robust fault detection and isolation method, for detecting faults in process sensors, actuators, controllers and other field devices. The approach to the solution to this problem is summarized below.

A novel approach for the validation of control system components and sensors was developed in this research. The process is composed of detecting a system anomaly, isolating the faulty component (such as sensors, actuators, and controllers), computing its deviation from expected value for a given system's normal condition, and finally reconstructing its output when applicable.

A variant of the Group Method of Data Handling (GMDH) was developed in this research for generating analytical redundancy from relationships among different system components. A rational function approximation was used for the data-driven modeling scheme. This analytical redundancy is necessary for detecting system anomalies and isolating faulty components.

A rule-base expert system was developed in order to isolate the faulty component. The rule-based was established from model-simulated data. A fuzzy-logic estimator was implemented to compute the magnitude of the loop component fault so that the operator or the controller might take corrective actions. This latter engine allows the system to be operated in a normal condition until the next scheduled shutdown, even if a critical component were detected as degrading.

The effectiveness of the method developed in this research was demonstrated through simulation and by implementation to an experimental control loop. The test loop consisted of a level control system, flow, pressure, level and temperature measuring sensors, motor-operated valves, and a pump. Commonly observed device faults were imposed in different system components such as pressure transmitters, pumps, and motor-operated valves.

This research has resulted in a framework for system component failure detection and isolation, allowing easy implementation of this method in any process control system (power plants, chemical industry, and other manufacturing industry). The technique would also aid the plant personnel in defining the minimal number of sensors to be installed in a process system, necessary for reliable component validation.