anomaly detection and equipment Health

Predict   

Empirical methods such as APR and machine learning model the expected performance variables that include the true process value plus any measurement uncertainty and operational variance. These empirical methods require limited knowledge about the process because the model will learn and infer them directly from the data. Typically, existing data from the data historian is used to create statistical relationships between relevant signals that can describe key performance, operational or reliability characteristics for key components. Once these relationships or prediction models are derived, they can be used to accurately estimate the predicted values of the various signals while the component is known to be operating normally.

Detect   

Anomaly detection identifies operating data that doesn't fit the equipment's normal behavior. The predicted signals at “normal” conditions are compared to the actual observed values in real-time by calculating the “residual” – the difference between the predicted normal values of the signals and the observed values of the signals. This residual should, over time, be statistically zero. If the residual begins to statistically deviate over time, this is a highly sensitive and accurate means for early detection of an anomaly that would lead to a failure.

Diagnose   

Once an anomaly is detected, the underlying problem must be diagnosed before it can be resolved. The Bayesian Belief Network (BBN) diagnostic method is a model-based, probabilistic approach that has proven to be very effective in diagnosing complex, multi-symptom faults.

The BBN diagnostic model determines the probability that a specific fault has occurred on the basis of the fault detector results. The design of the BBN diagnostic model captures the relationships between sensor residuals and the underlying anomalies that may have caused the problem. The BBN quantifies the belief that a failure mode is occurring based on the current state of the fault detectors. The BBN network shown above identifies fouled compressor rotor blades based on compressor mass flow, pressure ratio and discharge temperature deviations. 

Equipment Health

Proportional Hazard Model incorporates all three major types of information affecting estimated failure probability and remaining life. 

1. Design life based on operating hours and available condition-inducing measurements

2. Operational stressors based on count and duration of known stress events 

3. Sensor based condition indicators derived from deviations from expected behavior.

Equipment current health, reliability and remaining life are reported along with any recommended maintenance actions.