Rotational Angle Sensor Diagnosis Using Learned Reference Values

The system performs preliminary learning to establish individual reference values for each sensor before actual diagnosis operations. During a learning period, the system accumulates sum-of-squares data from sine and cosine signals and determines a reference value specific to each sensor's characteristics. This preliminary characterization allows subsequent diagnosis to use tighter, more accurate thresholds based on individual sensor behavior rather than broad population-based ranges.

The system changes the diagnostic parameter from fixed universal thresholds to dynamic reference values with associated standard deviations. Instead of using static upper and lower thresholds for all sensors, the system maintains individual reference values and calculates diagnosis thresholds as reference value ± k×standard deviation, where k is a diagnosis coefficient. This parameter transformation enables accurate diagnosis while accounting for individual sensor variations.

The system performs self-characterization during an initial learning period without requiring external calibration equipment or manual intervention. Each sensor automatically generates its own reference value and standard deviation by processing its own sine and cosine signals during normal operation. This self-service approach eliminates the need for complex external calibration systems while still achieving individualized accurate diagnosis thresholds.

The learning mechanism serves multiple functions: it establishes reference values for diagnosis, characterizes sensor performance, and adapts to individual sensor variations. The same signal processing circuitry used for normal sensor operation is also used to generate diagnostic parameters, eliminating the need for separate calibration hardware or procedures. This multi-functionality reduces overall system complexity while improving diagnosis accuracy.