Sensor Fault Diagnosis and Data Reconstruction in Mechanical Systems: A Comprehensive Review

As mechanical systems become increasingly intelligent and automated, sensors have evolved into indispensable components that serve as the perceptual infrastructure for monitoring and control. However, unacceptable deviations in sensor outputs can severely compromise data reliability, leading to incorrect monitoring decisions and potential risks to system safety. While traditional fault diagnosis research has predominantly focused on component-level failures, sensor faults have received comparatively less systematic attention despite their critical impact on system performance. This paper presents a comprehensive review of sensor fault diagnosis and data reconstruction methodologies in mechanical monitoring systems. The mechanisms of sensor faults are first analyzed from a structural perspective, and common fault signal patterns are summarized. Existing diagnosis approaches are then systematically categorized into three main paradigms: physics-based modeling, signal processing, and machine learning methods. Based on this taxonomy, recent advances in fault detection are reviewed, followed by a focused discussion on sensor fault isolation, particularly in distinguishing sensor faults from component faults. Furthermore, data reconstruction strategies for faulty sensors are classified according to temporal and spatial redundancy exploitation, with their assumptions, advantages, and limitations critically evaluated. Finally, challenges and future research directions are discussed to provide insights into the development of trustworthy and intelligent sensor systems.