Position Sensor Fault Prognostic using Data Driven Approach

Abstract

Resolver is a widely used in the feedback loop of the PM traction drive to find exact rotary position of the permanent magnet. In real systems, position error is caused by various factors such as amplitude imbalance, imperfect quadrature, inductive harmonics, reference phase shift, excitation signal distortion or other disturbance signals. This has influence on motor torque production. So, it is crucial to monitor resolver performance so that failed sensor can be easily replaced. This also benefits supply chain to keep the parts ready.This paper demonstrates monitoring the health of the resolver sensor using a data driven approach. The algorithm developed is not only capable of classifying faulty/ healthy resolver, but it can also show the amount of degradation in the resolver sensor. The state-of-the-art developed neural network model is trained on the robust database covering all possible resolver degradations, partial and complete failures. This model is developed on a complete synthetic data tapped from the Simulink model and it is further optimized for the accuracy and size. The algorithm was initially tested on the standalone open-loop resolver model which later extended for the closed-loop version. It also supports commanded mode of prognostics which can detect and classify possible harness faults of the resolver sensor. The proposed algorithm has shown high confidence when it is tested offline on the actual hardware data.