Meaning
Diagnostic method that analyzes the difference between the observed behavior of a system and the output of a mathematical model. Technicians use residual monitoring to identify faults by looking for patterns in the errors that the model cannot explain. This approach is effective for finding hidden degradation in complex machinery where direct measurement of a component is impossible.
Error Generation
Subtracting the model prediction from the actual sensor reading produces a residual signal. When residual monitoring is applied to a healthy system, this signal consists mostly of random noise with a mean of zero. Any persistent bias or trend in the residual indicates that the physical system no longer matches the model.
Statistical Filtering
Filtering out the noise from the residual signal is necessary to avoid false alarms. Techniques like moving averages help highlight the underlying trend within the residual monitoring data. This ensures that the monitor only reacts to genuine changes in the system state rather than measurement fluctuations.
Isolation Logic
Pinpointing the location of a fault requires comparing data from multiple sensors. If only one residual shows a shift, the sensor itself is likely failing. When multiple residuals drift together, the problem usually lies in a shared physical component like a motor or a structural seal.
