The degradation of infrastructure assets can compromise their future safety, resilience, and performance. While degradation may be mitigated through proper maintenance (inspection, overhaul, renewal, etc.), planning future interventions is crucial for optimally balancing failure risks (e.g. safety) and maintenance costs, and for optimally timing large capital expenditures (e.g. renewal). Yet, pursuing traditional reactive/scheduled maintenance based on standard approaches can lead to poor maintenance regimes due to the large uncertainties in the in situ asset behaviour (e.g. due to differences between “as-built” and design, changes in loadings/operating environment/building use). Also, asset owners/operators struggle to translate the large amount of complex data acquired into actionable information as well as information to improve the future design. The Infrastructure Health Monitoring and Predictive Maintenance Theme will work with the other themes of the RIIS Hub to address the challenges in infrastructure monitoring and predictive and preventive maintenance through innovative research and technologies in the area of SHM. The theme will focus on developing and implementing new technology and research innovation for the following sub-themes:
- Performance monitoring and health status evaluation
- Analytics for degradation prediction and remaining service life
-
4.1 Performance monitoring and health status evaluation
Introduction
It is important to identify whether an infrastructure system behaves as intended and safely under external loads. Infrastructure systems can be simultaneously subjected to a combination of external loads (e.g., vehicular load, wind load, earthquake, collision, water flows, and temperature and humidity fluctuations).
The challenge is to monitor all external loads, their effects and accurately predict the structural response.
Innovation
This stream will develop automated performance evaluation approaches based on smart data collection (with Subtheme 2.1) and data analytics and machine learning (with Subthemes 2.2 & 3.2), and predictive computational modelling (with Subtheme 3.1) to effectively correlate the response of structures under complex external loads, as compared to the limits specified in the corresponding standards.
Different models will be developed for analysing and predicting structural responses at multiple points under different types of loads.
Traditional sensors (e.g., accelerometers, weight-in-motion, pressure sensors, anemometers) will be optimally integrated with advanced sensor technologies (e.g., laser-based, machine- vision based, image-based), unmanned aerial vehicle (UAV) and unmanned ground vehicle (UGV) techniques to improve the accuracy and effectiveness of monitoring systems. To render the monitoring systems responsive and understandable for end-users, comprehensive solutions for data collection and management will be developed (with Subtheme 2.1).
In addition, synthetic approaches based on advanced computational mechanics, inverse analysis and augmented ML to enable comprehensively real-time mapping of the relationship between the structural response and complex state of loading.
Outcome
New data extraction techniques to correlate in real-time the behaviour of infrastructure systems under various loads, towards reliable estimation of as-is capacities of infrastructure systems.
-
4.2: Analytics for degradation prediction and remaining service life
Introduction
The prediction of degradation is essential to understanding future failure probabilities and implementing pre-emptive maintenance. The key technical challenge is to develop methods that can predict key degradation mechanisms (e.g. corrosion, erosion, crack propagation) in the presence of variable environmental conditions.
The predictions can then be used to update (degrade) relevant parameters in the numerical models to predict future failure probabilities and risks. Most existing methods for degradation prediction (also known as prognosis or prognostics) are either simplified physical models or are purely data-driven.
These approaches have important drawbacks:
i) they are overly reliant on large degradation/condition data sets, which rarely exist for of critical infrastructure;
ii) they extrapolate poorly to new operating environments and/or new assets; and
iii) simplified engineering models provide point predictions, which do not properly accommodate uncertainty.
Innovation
To address these challenges, new degradation prediction analytics will be developed in this subtheme that synthesise augmented machine learning/statistical approaches (with Subtheme 3.2) with established physical degradation models (e.g. erosive wear models), advanced computational modelling (with Subtheme 3.1), while exploiting data from Digital Twins (Subtheme 5.1).
Moreover, new parameter identification and condition prediction algorithms will be developed for infrastructure assets under the Bayesian reliability paradigm.
While this paradigm has found success in other applications where failure data is sparse (e.g. Nuclear), it has yet to be widely adopted for infrastructure assets.
Importantly, the new Bayesian methods will provide additional mechanisms for including engineering knowledge and mitigate the over-confidence of predictions based on machine learning approaches by providing a more complete and flexible characterization of uncertainty.
Outcome
Degradation analytics for the prediction of key degradation modes with a characterisation of uncertainty.