A New In-situ Sensor Modeling Approach to Measurement Validation

Measurement error due to sensor degradation (fouling, mis-calibration etc.) is more difficult to identify compared to catastrophic sensor failure. Passive methods previously proposed for sensor-level monitoring are based on power spectrum or multi-scale analysis of sensor data. These methods have limitations caused by not accounting for various noise sources, and assumptions about sensor noise characteristics; thus resulting in false and missed alarms. In this paper, an on-line sensor fault detection scheme based on identification of sensor response characteristics is proposed and evaluated. We develop both robust passive and active in-situ techniques to identify sensor response characteristics that relate directly to its health. Using the identified sensor model, various kinds of sensor faults are quantified and mapped into the model parameters. A dynamic model-based estimator is proposed for data reconciliation. These ideas were experimentally validated using thermocouples, flow-meters, and resistance thermometric devices (RTD) on laboratory scale processes. The proposed approach was seen to accurately quantify the sensor model parameters and aid in measurement reconstruction. 1 Introduction Accurate measurements are necessary in providing controllers and operators with a view of the process status. Safe and profitable operations of a chemical plant rely heavily on the proper functioning of sensors. In process control, many (reportedly up to 60%) of the perceived malfunctions in a plant are believed to originate from the lack of credible measurements 1. Reliable measurement validation mechanisms are needed by operators and the control systems that use the information provided by the sensor for decision-making. Sensor faults can be classified into two broad types – hard (catastrophic) sensor failure that results in completely erroneous or no readings from the sensor; and sensor degradation (change in physical condition or calibration) that degenerates the performance of the sensor. In this work, we are primarily concerned with identifying and rectifying the latter type of sensor faults, though many of our methods are valid for the former as well. Sensor degradation can be due to a variety of circumstances such as fouling, mechanical damage, electrical malfunction, or incorrect calibration. Since sensors cannot be easily removed and tested for faults, it is desirable to develop in-situ measurement validation schemes. There is inherent isolation from process-level faults as such schemes are based on the sensor measurement alone. Process-model oriented sensor fault detection methods do not easily separate sensor faults from process problems. Furthermore, the disparity in time scales between the process and the measurement system is …