AI-based condition monitoring on mechanical systems using multibody dynamics models

Abstract

Real-time monitoring of mechanical systems via vibration measurements allows for detection of faults in them and facilitates their predictive maintenance. Use of Artificial Intelligence (AI) in damage detection can provide an automated means for Condition Monitoring (CM) of the system while also allowing for characterization of the various health states when labeled damaged state training datasets are available. In this work, a novel CM framework using Convolutional Neural Networks (CNNs) is presented for damage detection and identification on mechanical systems, applied on an elevator door rail, using simulated data from a Multibody Dynamics (MBD) model of the physical system. First, an optimal MBD model of the actual structure is constructed by means of black box optimization, using a small number of initial healthy state measurements. Several of the most common nonlinear contact force models are examined before integrating the best suited one to the MBD model. Damaged states of the system are then simulated by modeling various fault mechanisms on the optimal MBD model, allowing for generation of labeled simulated training datasets for the CNNs. Uncertainty is introduced to the models by sampling their key parameters from a Gaussian distribution, thus considering the real system’s inherent uncertainty. The trained CNNs’ robustness and accuracy are then validated by accurately classifying faults on the physical system, proving the proposed damage detection method’s generalization capabilities and highlighting its potential. The proposed methodology may find application in mitigating the problem of data scarcity of damaged state responses which is present in most mechanical systems and can provide a means for more efficient predictive maintenance. Successful implementation of the proposed framework adds to the existing AI based damage detection and identification methodologies by allowing for extension of the method to industrial systems.