A robust distributed fault detection scheme for interconnected systems based on subspace identification technique

This paper proposes a robust data-driven distributed fault detection (FD) method for interconnected systems with stochastic noises, addressing the challenges posed by stochastic noises and the limitations of centralized designs on real industrial systems. The proposed method utilizes the process data from a sensor network to generate robust residual signals for FD. The method performs distributed calculations of residual signals and test statistics at each sensor node using the subspace identification technique and the average consensus algorithm. To ensure satisfactory detection performance and robustness against uncertainties caused by stochastic noises, the paper integrates the performance index and the Mahalanobis distance into the FD framework. Unlike existing FD methods that rely on the Mahalanobis distance, this study also explores improving detection performance through the consensus algorithm and performance index. It is worth noting that this method not only mitigates the negative effects of stochastic noises in FD, but also eliminates global communication costs and complex information interactions. The developed method is validated through an experimental study on a real traction drive system to assess its feasibility and effectiveness.