Majority voting for fault tolerance in microelectromechanical systems sensor arrays

Abstract

This paper presents a practical method to enhance the fault tolerance of microelectromechanical systems (MEMS) inertial measurement unit arrays by combining simple local plausibility tests with robust group criteria (median) and majority voting. Unlike approaches that rely on extended Kalman filtering at the low level, the proposed algorithm operates without an EKF, using hysteresis and a finite state machine for deterministic channel isolation and recovery. The final estimate is formed by robust aggregators (median or trimmed mean), reducing sensitivity to both isolated outliers and long-term failures. The method was evaluated exclusively in software simulators using Monte Carlo experiments across diverse motion profiles, sensor noise models, and injected failure scenarios, including bias jumps, drifts, scale errors, “sticking” effects, impulsive outliers, frame losses, and inter-channel noise correlations. Performance metrics include detection probability, false alarm rate, detection latency, recovery, estimate stability, and computational cost. Results demonstrate high detection probability of single faults with low false alarms and controlled latency in small arrays (N=3–7). Robust aggregators provide better resilience against mixed error modes and noise correlation compared to the arithmetic mean, while embedded hysteresis prevents “chattering” near thresholds. The algorithm’s computational complexity is linear in the number of sensors per sample, making it suitable for real-time embedded platforms. Limitations concern systematic errors and highly correlated multiple failures, which reduce the informativeness of majority voting. Future work includes adaptive thresholding based on noise estimates, support for multiple simultaneous failures, integration with lightweight complementary filters, and experimental validation on a turntable and unmanned aerial vehicle flight data.