More information
Description
This project aims to develop a novel data-driven dynamic reliability assessment platform to improve predictive maintenance ability in complex cyberphysical systems (CPSs). This will be achieved by exactly identifying which degradation mechanism(s) are likely to cause an impending failure, and then highlighting the event to trigger for maintenance service or control operation. The expected outcomes are new methods and tools needed to leverage failure prognostics and prognostics-informed maintenance/control for making CPSs resilient with reduced levels of redundancy. This research will produce major advancements in extending core components’ life and durability in complex CPSs, bringing economic benefit for Polish industry. Research outcomes from this project will enable Polish companies to produce highly needed intelligent prognostics products at a reduced cost but in high quality. This will bolster their domestic and global competitiveness and bring economic benefits. Furthermore, the academic achievements and the patented technologies will expand Poland’s access to the global maintenance, repair, and overhaul markets, such as those in EU and USA. These benefits will be achieved by disseminating theoretical and experimental advances on predictive maintenance. This research will also raise the Poland''s profile in global maintenance, repair, and overhaul markets by offering training and education in intelligent prognostics technology.
Summary of project results
Prognostic and health management (PHM) of complex cyberphysical systems (CPSs) has always been a top priority for governments and engineering communities around the world. Typical complex CPSs including the power (battery) system in hybrid vehicles (HVs), offshore wind farms, and intelligent manufacturing systems, etc. However, during the service life, key components in the CPSs will be subject to functional degradations, leading to risky failures to the CPSs. The apparent shortcoming of existing techniques is that they simply detect the health condition of a CPS without considering the dynamic degradation process of the components. Generally, one should elaborate on the degradation mechanism(s) the CPS has experienced to identify the impending failure mode and whether an immediate action is required. The information from these diagnostic technigues does not provide much insight to the maintenance personnel and control electronics on what and when to fix or replace
There is a need to shift towards a new maintenance/control (M/C) paradigm, intelligent prognostics of failure modes. It transcends being merely a health prognostic system by exactly identifying which degradation mechanism(s)are likely to cause an impending failure, and then highlighting the event to trigger maintenance service and/or electronic control operation. To meet this new need in intelligent prognostics for CPSs, three challenging tasks should be addressed.
The first challenge is how to build high-fidelity multiphysics model to learn the system degradation mechanisms. Finite element methods are usually employed to establish the governing equations of the underlying electrochemistry and physics processes in a CPS. For high temporal and spatial resolutions, as required for a high-fidelity model, the model evaluation becomes computationally very expensive.
The second challenge is how to build health estimators capable of on-board degradation analysis. At present, the health diagnostics of CPSs is limited to estimation of current defects, but not considers generic indicators for remaining useful life prognostics. This health information needs to be enriched with information about the underlying degradation mechanism(s) to produce generic indicators for both diagnostics and prognostics.
The third challenge is how to predict CPS failures while considering multiple degradation mechanisms. Existing prognostic approaches for a general engineered system have been successful, in part, in predicting the performance degradation of the system. However, these approaches are mostly application-specific and are not robust across different applications, and are thus difficult to be directly applied for CPS prognostics.
The most important project deliverables are described as the following bullet points.
- A digital twin (DT)-driven approach to model the degradation mechanisms through estimation of the system degradation parameters
- A novel digital twin technique based on a Markov Chain Monte Carlo (MCMC)-based Bayesian updating for system degradation identification
- A novel ensemble deep learning neural network model, termed as CNN-BiLSTM-Attention model for system degradation prediction improvement
In this project, the principle investigator (PI) models the degradation mechanisms through estimation of the system degradation parameters. A digital twin (DT)-driven approach is proposed to accurately predict system fatigue life by establishing effective dual-information communication between a DT virtual model and a physical model of the research objective. The proposed DT virtual model consists of three modules (namely one crack tracking model, one high-precision approximating model and one dynamic Bayesian network (DBN) inference model) and operates in offline and online stages. The offline stage employs the extended finite element method (XFEM) to establish the crack tracking model, which will generate sufficient knowledge-known datasets to train the high-precision approximating model. In the online stage, the model parameters are updated by the DBN inference model based on the well-trained approximating model, where real-time information exchange from the physical model is performed. As a result, unexpected uncertainties of the model parameters can be significantly reduced.
Taking an offshore wind structure as an example, in the proposed DT method, the natural frequencies and mode shapes of the offshore wind structure, extracted both from the structural dynamic responses of a finite element (FE) model and the sensory measurements in a physical system, are used to construct the likelihood function of the Bayesian inference. Then, the posterior probability distribution function (PDF) of the uncertain parameters is derived under the assumption of uniform prior distribution, which is used to generate the Markov Chains by implementing the Metropolis-Hastings sampling. Hence, the most probable values of the uncertain parameters can be obtained from the Markov Chains for updating the FE model. A numerical study demonstrates high effectiveness of the proposed model updating method, even with high-level measurement noise and model uncertainty. Subsequently, a new ensemble-learning based prognostic approach is developed to identify the structural damage patterns. The research outcomes have been published in leading international journal of Reliability Engineering & System Safety.
This project produced new knowledge in reliability engineering and system safety (RESS) discipline from Information Science and Intelligent Manufacturing sectors. This project led to a new prognostic research direction using theoretical and experimental advances on predictive maintenance technology, and generated valuable knowledge in RESS discipline by enhancing the PHM ability of Poland and the Donor States against unaccepted failures in important physical systems. The achievements from this project have been published in top international journals and international conferences, generating huge impact on RESS discipline in global wide.
In recent five years, intelligent prognostics have been increasingly applied to a growing market of protecting complex CPSs such as HVs, battery management systems, aircraft engines, and wind turbines. Research outcomes from this project will enable Poland and the Donor States’ companies to produce highly needed intelligent prognostics products at a reduced cost but in high quality. This will bolster their domestic and global competitiveness and bring economic benefits. Furthermore, the academic achievements and the patented technologies will expand Poland and the Donor States’ access to the global markets, such as those in New Zealand, China, Japan and USA.