Machine Learning-Driven Sensitivity Analysis for a 2-Layer Printed Circuit Board Inductive Motor Position Sensor

Area of Science:

• Automotive Engineering
• Electrical Engineering
• Sensor Technology

Background:

• Traction motor control in electrified powertrains relies on precise motor position sensors.
• Increasingly compact motor designs create significant packaging constraints for these sensors.

Purpose of the Study:

• To present a novel two-layer (2L) printed circuit board (PCB) routing strategy for inductive motor position sensors in space-limited applications.
• To analyze the accuracy and identify dominant error sources in sensor performance under varying installation conditions.

Main Methods:

• Fabrication and testing of a prototype sensor with a comprehensive design of experiments (625 combinations of offsets, tilt, and airgap).
• Accuracy analysis using Fourier series and development of an Extreme Gradient Boosting (XGBoost) model (R² = 0.9951) for error prediction.
• Application of SHapley Additive exPlanations (SHAP) to identify key error drivers and their interactions.

Main Results:

• Sensor accuracy remained within ±1 electrical degree across all tested installation tolerances.
• DC offset and magnitude mismatches were identified as dominant error contributors due to routing modifications.
• Tilt angle and Y-offset were found to be the primary drivers of accuracy degradation, with a strong coupled influence (interaction intensity = 0.9581).

Conclusions:

• The developed 2L PCB routing strategy effectively maintains high accuracy for inductive motor position sensors in restricted spaces.
• The study highlights the non-linear nature of installation-induced accuracy degradation and provides a framework for sensor layout optimization and tolerance establishment.