Early fault detection in gearboxes via dynamic principal component analysis-driven multivariate statistical process control.

Abstract

Early detection of gearbox failure is essential due to their critical role in industrial operations. Therefore, effective condition monitoring techniques are required to identify incipient deviations in operational behaviour. Therefore, this study proposes a dynamic principal component analysis methodology, integrated within a multivariate statistical process control framework, to detect progressive failures in spur gearboxes from vibration signals. The signal is segmented into sub-windows and characterised using condition indicators in the time and frequency domains. Diagnosis is based on Hotelling's T2 statistic and the squared prediction error, which define statistical control limits to discriminate between normal and failure conditions. Empirical validation uses an experimental dataset covering combinations of load, speed, and failure severity. The results demonstrate high sensitivity to progressive degradation and accurate early-stage detection, supporting the multivariate statistical process control approach with dynamic principal component analysis as an effective tool for diagnosis and predictive maintenance in high-criticality industrial environments.