One Masked Model is All You Need for Sensor Fault Detection, Isolation and Accommodation

TL;DR

The paper addresses sensor FDIA in complex engineering systems by introducing a unified, self-supervised framework based on masked time-series modeling that can perform detection, isolation, and accommodation with a single, model-agnostic neural network. During training, random masks simulate faults, enabling the model to learn spatiotemporal sensor relationships and to handle faulty inputs without out-of-distribution issues. Key contributions include a formal masked modeling approach for time-series data, an online inference procedure with per-sensor residual thresholds, and empirical validation on both a public dataset and GE offshore-wind turbine data showing improved fault detection and substantial robustness gains with real-time feasibility. The work has practical impact by simplifying FDIA pipelines and enabling real-time corrective actions, with broad applicability to various sensor networks and systems.

Abstract

Accurate and reliable sensor measurements are critical for ensuring the safety and longevity of complex engineering systems such as wind turbines. In this paper, we propose a novel framework for sensor fault detection, isolation, and accommodation (FDIA) using masked models and self-supervised learning. Our proposed approach is a general time series modeling approach that can be applied to any neural network (NN) model capable of sequence modeling, and captures the complex spatio-temporal relationships among different sensors. During training, the proposed masked approach creates a random mask, which acts like a fault, for one or more sensors, making the training and inference task unified: finding the faulty sensors and correcting them. We validate our proposed technique on both a public dataset and a real-world dataset from GE offshore wind turbines, and demonstrate its effectiveness in detecting, diagnosing and correcting sensor faults. The masked model not only simplifies the overall FDIA pipeline, but also outperforms existing approaches. Our proposed technique has the potential to significantly improve the accuracy and reliability of sensor measurements in complex engineering systems in real-time, and could be applied to other types of sensors and engineering systems in the future. We believe that our proposed framework can contribute to the development of more efficient and effective FDIA techniques for a wide range of applications.

Figures (5)

• Figure 1: Comparison of existing sensor FDIA techniques with our proposed method.
• Figure 2: Training: Comparison of different sensor data modeling formulations. Superscripts indicate channel, subscripts indicate time step, and $f$ indicates a mapping or a model. Our proposed masked method (a) achieves multi-task learning by randomly masking one or more channels at each iteration, in contrast to existing techniques (b) Auto-regressive and (c) Auto-associative.
• Figure 3: ROC and PR curves for sensor bias in P1_PIT01 and P1_TIT01.
• Figure 4: Example of how our proposed sensor FDIA technique can be integrated into an existing engineering system to improve its robustness and reliability.
• Figure 5: Residual plots for the 4 different sensor fault cases versus the normal test set.