Abstract
Data-driven condition-based maintenance (CBM) can be an effective predictive maintenance strategy for components within complex systems with unknown dynamics, nonstationary vibration signatures or a lack of historical failure data. CBM strategies allow operators to maintain components based on their condition in lieu of traditional alternatives such as preventive or corrective strategies. In this paper, the authors present an outline of the CBM program and a field pilot study being conducted on the gearbox, a critical component in an automated cable-driven people mover (APM) system at Toronto’s Pearson airport. This CBM program utilizes a paired server-client “two-tier” configuration for fault detection and prognosis. At the first level, fault detection is performed in real-time using vibration data collected from accelerometers mounted on the APM gearbox. Time-domain condition indicators are extracted from the signals to establish the baseline condition of the system to detect faults in real-time. All tier one tasks are handled autonomously using a controller located on-site. In the second level pertaining to prognostics, these condition indicators are utilized for degradation modeling and subsequent remaining useful life (RUL) estimation using random coefficient and stochastic degradation models. Parameter estimation is undertaken using a hierarchical Bayesian approach. Degradation parameters and the RUL model are updated in a feedback loop using the collected degradation data. While the case study presented will primarily focus on a cable-driven APM gearbox, the underlying theory and the tools developed to undertake diagnostics and prognostics tasks are broadly applicable to a wide range of other civil and industrial applications.
Highlights
IntroductionIn the training phase, which only occurs once at the onset of the project, a representative sample of vibration data, collected through a set of accelerometers, is used as an input to generate the initial Gaussian mixture models (GMMs) and a degradation model (described in detail )
For this particular case study, several decisions in the condition-based maintenance (CBM) server-side routine involved expert intervention. It is the authors’ goal in the future to develop and refine the criteria necessary to automate these decisions as well. This automated cable-driven people mover (APM) system consists of a cable-driven, computer-controlled train that travels along a 1.5 km track over variable grade and connects three passenger stations
APM gearboxes typically present these main challenges for CBM: they are low-speed, highly non-stationary and complex systems with multi-path convolved vibration sources and often accompanied with scarce availability of historical failure data
Summary
In the training phase, which only occurs once at the onset of the project, a representative sample of vibration data, collected through a set of accelerometers, is used as an input to generate the initial GMM and a degradation model (described in detail ). From these models the initial failure thresholds and initial degradation path parameters are obtained. This is represented by Eq (1) in one dimension, where K denotes the number of mixture components, and !!, !!, !! denote the mean, standard deviation and weight of mixture component i. ! (1)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
