A Data-Driven Framework for Fault Diagnostics in Gearbox Health Monitoring Under Non-Stationary Conditions

Monitoring gearbox health is essential in industrial systems, where undetected faults can result in costly downtime and severe equipment damage. While vibration-based diagnostics are widely utilized for fault detection, analyzing large-scale, non-stationary vibration signals remains a computational challenge, particularly in real-time and resource-constrained environments. This paper presents Data-Driven Synchrosqueezing-based Signal Transformation (DSST), a novel time-frequency method that integrates synchrosqueezing transform (SST) with structured downsampling in both time and frequency domains. DSST significantly reduces computational and memory demands, while preserving high-resolution representations of fault-related features such as gear meshing frequency sidebands and their harmonics. In contrast to prior SST variants, DSST emphasizes diagnostic interpretability, invertibility, and compatibility with data-driven learning models, making it suitable for deployment in modern condition monitoring frameworks. Experimental results on non-stationary gearbox vibration data demonstrate that DSST achieves comparable diagnostic accuracy to conventional SST methods, with substantial gains in processing efficiency—thereby supporting scalable, real-time industrial health monitoring. Unlike existing downsampling-based SST methods, DSST is designed as a diagnostic component within a scalable, data-driven framework, supporting real-time analysis, signal reconstruction, and downstream machine learning integration.