System breakdowns in modern industrial environments can result in millions of dollars in lost time and productivity, and even the loss of life and property. In the utilities industry — where the continuous operation of coolant pumps is essential — the breakdown of a single pump can result in a loss of as much as $10 million in downtime.
An early-warning system, called the Multivariate State Estimation Technique (MSET), was developed to monitor the performance of sensors, equipment, and plant processes in an industrial environment. A highly sensitive, highly accurate tool, MSET monitors the operation of any process that uses multiple sensors, detecting and alerting users of potential problems.
MSET consists of a suite of statistically based pattern recognition modules. It detects and identifies malfunctions that may occur in process sensors, components, or control systems; or changes in process operating conditions. The modules interact to provide users with the information needed for the safe, reliable, and economical operation of a process by detecting, locating, and identifying very subtle changes that could lead to future problems well in advance of actual equipment degradation.
Since it provides continuous calibration validation for all sensors, MSET offers a technical basis for reducing instrument calibration requirements. It can also help users determine when it is appropriate to continue or extend operation of certain components, or to schedule corrective actions such as sensor replacement or recalibration, or component adjustment.
MSET uses an ultra-sensitive Sequential Probability Ratio Test (SPRT) to discern sensor or system anomalies at the earliest possible time. MSET’s unique capabilities make it better than conventional approaches such as neural networks in terms of sensitivity, reliability, and computational efficiency.
To use MSET, the user first collects sensor readings (via a digital acquisition system) to characterize the normal operating state of the system. MSET automatically selects an optimal subset of these data and uses it to “train” the system to recognize normal behavior. During monitoring, MSET generates an accurate estimate of what each signal should be based on the latest set of sensor readings and the previously learned correlations among them. Then, SPRT analyzes the difference between this state estimate and the measurement, and quickly detects and alerts the smallest developing faults. If an abnormal condition is detected, the initial diagnostic step identifies the cause as either a sensor degradation or an operational change in the process. When a sensor fault is identified, MSET uses the estimated value of the signal to provide an extremely precise “virtual sensor” that can be used to fully replace the function of the faulted sensor.