Eventos Anais de eventos
COBEM 2023
27th International Congress of Mechanical Engineering
RELIABILITY ESTIMATION OF CRACK PROPAGATION IN A ROTATING MACHINE SHAFT USING BAYESIAN DEEP LEARNING
Submission Author:
Matheus de Moraes , SP
Co-Authors:
Matheus de Moraes, Douglas Kohatsu, Joao P. Dias, Helio Fiori de Castro
Presenter: Helio Fiori de Castro
doi://10.26678/ABCM.COBEM2023.COB2023-1451
Abstract
Rotating machines, such as transmissions and turbines, are critical components in a wide range of industrial and power generation systems. These machines prognostics and health monitoring studies are important for ensuring their reliable and efficient operation. These studies involve monitoring various parameters, such as vibration, lubricant oil debris, and temperature to detect indications of degradation or imminent failure. By identifying potential issues early, maintenance and repairs can be scheduled before they become critical, preventing unexpected downtime and costly repairs. Additionally, the analysis of data collected from machines over time allows the identification of patterns and trends that can be used to improve their design and operation. The main goal of this study is to investigate and evaluate some techniques for assessing the reliability and remaining useful life (RUL) of a computational model of a rotating machine. To achieve this goal, the following research questions were proposed: is it possible to use data-driven techniques such as Bayesian Neural Networks (BNNs), First Order Reliability Method (FORM), and Second Order Reliability Method (SORM), to estimate the reliability and the RUL of a rotating machine? To answer this question a computational model of a breathing transversal crack that accounts for the uncertainties and the stochastic modeling of crack propagation was developed. First, several BNNs were designed and trained, and their performance was evaluated in the task of measuring the crack length. Next, FORM and SORM were applied to the model to calculate the reliability of the mechanical system. Finally, a crack propagation model based on the Paris Law was employed to estimate the RUL of the rotating machine. Preliminary results showed that the BNNs were able to diagnose the crack length with satisfactory accuracy, whereas FORM and SORM were able to estimate the reliability of the system. This framework showed a potential for test in physical applications.
Keywords
Bayesian neural networks, remaining useful life prediction, reliability, Rotordynamics, Crack propagation
