Distinguishing between process upsets and sensor malfunctions using sensor redundancy

The ability to differentiate between process upsets and sensor malfunctions is crucial in the monitoring of a chemical process as different compensatory responses are required. Process upsets threaten the quality of the chemical product and may require immediate intervention by the process operator, while malfunctioning sensors can be replaced or compensated for mathematically. This paper addresses this problem and describes a new voting system procedure, based on probabilistic and empirical rules, for distinguishing between process upsets and sensor malfunctions. In contrast to traditional voting techniques which require the strict duplication of sensor elements, the redundant sensor voting system (RSVS) presented is rooted on the concept of state redundancy. Diagnosis employing state redundant sensors is based on the observation that process upsets are typically registered by a band of correlated sensors, while sensor malfunctions are commonly localized. According to the RSVS procedure, if the number of identified sensors equals or exceeds a predefined threshold value, a process disturbance is diagnosed. If the number of identified sensors is less than the threshold value, this is strong evidence of sensor malfunction. While situations can be envisioned in which the assumptions of RSVS are violated (e.g., disturbed sensors have not been correctly identified, the occurrence of localized process upsets), the bandwidth-based approach is reasonable for many types of processes, as is demonstrated for data collected on a liquid fed ceramic melter.