Statistical Discrepancy
Recursive estimation loops rely on statistical discrepancies to compare sensor data against a theoretical model. The innovation residual functions as the primary feedback signal in algorithms like the Kalman filter. It isolates the new information contained in the latest data packet.
Filter Performance
Analysis of the zero-mean property and the covariance of the innovation residual reveals how well the model matches the physical process. A consistent bias in this value indicates a failure in sensor calibration or an unmodeled disturbance.
Fault Detection
Monitoring the magnitude of the innovation residual allows an integration system to identify sensor failures or environmental outliers. When the value exceeds a statistical threshold (often defined by the Mahalanobis distance), the filter rejects the measurement to protect the integrity of the state estimate. This validation step prevents erratic data from corrupting the navigation or control solution.
In systems with high redundancy, the residual identifies which specific component has drifted out of tolerance.
Gain Adjustment
Weighting factors within the observer logic depend on the magnitude of these differences. A small innovation residual suggests that the model is accurate, while a large value forces the system to rely more heavily on the sensor input to correct the state.