# Application

# Underlying idea

This project aims to provide more specific, accurate and timely decision support in operation of safety-critical systems, by combining physics-based modelling with data-driven machine learning and probabilistic uncertainty assessment. The underlying idea is to combine well-established and robust physics-based full order models (FOM), that are made effective by reduced order modelling (ROM), and use of probabilistic data-driven models to both increase the accuracy as well as focus simulation efforts where the information gained produce the most value with respect to the relevant decision context.

Uncertainty in behaviour of safety-critical systems leads to conservatism in operating envelopes
Current scientific understanding and engineering practices are often based on modelling of first principle physics and causal reasoning. Meanwhile, we see rapid advances in modern machine learning techniques and self-learning autonomous systems, which are mainly experience-based and data-driven. This offers new opportunities for improving the status quo of engineering practices

Digitalization and autonomy are increasingly being explored and implemented across many industries. For industries where the value of what is produced is high, where the negative impact of a disrupted service is significant, or where operational deviations could be dangerous, it is of utter importance that control strategies and critical decisions are based on the best possible scientific understanding and all relevant experience. Better understanding (less uncertainty) of the behaviour of safety-critical systems will reduce the need for conservatism in operating envelopes and thus increase value creation.

Providing high-fidelity results in real time using todays methods is a computational challenge
Physics-based and data-driven methods have the same objective; To make sure that the control of-, and decisions related to-, the systems under study are optimal and safe (see Figure 1 for illustration). However, these approaches are fundamentally different, with different strengths and weaknesses.

Physics-based models seek to simulate key mechanisms behind a system’s behavior, yet they are inevitably imperfect idealizations of reality, and often computationally heavy. On the other hand, data-driven models are based on uncertain observations of a limited experience base, but may offer fast answers. This limit the scope where data-driven models can be used, as well as the accuracy and timely response of physics-based methods. On the other hand, data-driven models are able to capture non-idealized behavior of complex systems, while physics-based models are able to extrapolate system behavior outside the current experience base.

RaPiD-models Figure 1. The main idea of RaPiD: Faster, better and safer decisions based on physics and data

This project will combine physics-based and data-driven models for high-fidelity real-time support Thus, to be able to provide high-fidelity insights for better decision support in complex and high-risk systems, it is essential to alleviate the deficiencies of both data- and physics-based model by capturing the complementary advantages of both. This project focus on establishing basic methodology to consistently combine data-driven and physics-based models.

This will enable reduced cost and more efficient use of high-fidelity models across many resource intensive industries, as well as better use of observed data and operational experience. Better control of complex and value creating systems enable both large savings by reduced down-time and costly disruption of service as well as increased value creation based on more optimal production.

# Level of innovation

The main innovation of the project is to enable real-time use of computationally expensive high-fidelity results by combining physics-based and data-driven methods in a way that ensures uncertainty and risk is acceptable.

Combining physics-based and data-driven models for high-fidelity real-time support
Creating high-fidelity insights for better real-time decision-support in complex and high-risk systems requires innovative and new ways of combining data-driven and physics-based modelling, beyond current approaches (e.g. hybrid modelling, multi-fidelity modelling, etc). Integrating these domains in a way that ensures both that the computational speed and the overall uncertainty is acceptable is the biggest research challenge of the project. This enable the asset owners to make faster and better safety-critical decisions and reduce (overly conservative) restrictions on operating envelopes, maintenance and inspection intervals, etc.

Increasing uptime, more optimal production, and avoiding safety critical incidents, if only by a few percent, has a huge value for the asset owners.

“Conventional finite-elements simulations appear to be conservative compared with what we observe from sensor data. However, models based on sensor data do not capture safety-critical behaviour which we have not yet experienced. In effect, neither model is able to give proper decision support and we are left with conservatism to offset the uncertainty”.

Paraphrased statement from asset owner operating on the NCS

Several innovate elements will be part of the project to realise the overall innovation:

  1. Development of new methods for better computational efficiency (such as Reduced order modeling)
  2. Develop new approaches for fast approximations and confident predictions away from simulated scenarios that include relevant uncertainties.
  3. A novel approach for UQ (uncertainty quantification) by integrating uncertainty propagated through physical models with uncertainty represented by probabilistic machine learning (Gaussian Processes) from sensor data.
  4. A new way of applying Design of Experiments (DoE) for effective selection of relevant simulation scenarios that reduces uncertainty where it matters the most and thus reduce overall simulation efforts and time.
  5. Knowledge sharing to facility uptake, through development of a Guideline

Facilitate development of predictive Digital Twins
Digital twins are still an immature technology in most industries. The concept of a virtual representation of a physical object or system which uses real-time data, simulation capabilities, and other sources to enable understanding, learning, reasoning, and dynamic recalibration for improved decision making across its lifecycle (see Rasheed et al. (2020) for more details), is highly desired, but not yet fully attainable. The proposed integrated modelling approach herein will facilitate transformative research and enable Digital Twin technology that will change the paradigm to a two-way communication between computational methods and real-world assets and data as illustrated in Figure 1.

A novel innovation of the present project is to provide high level of physical realism for analysis tools used in the operational phase that can enable development of predictive Digital Twins. New paradigms for product lifecycle management may then be established as illustrated by the use cases included in the proposal: (a) to extend operational limits of offshore workover drilling, (b) to facilitate significant reduction of levelized cost of energy produced by offshore floating wind parks, (c) to significantly improve structural health monitoring for increased safety and reduced maintenance costs of road bridges.

# Need for research

Data-driven decision support and physics-based decision support are two fields of engineering that is currently separated both by methodologies, capabilities, and culture. Although both fields try to optimize the outcome of a decision (e.g. the output produced by how a system is operated), they differ in how the problem is approached and tackled.

Where data-driven approaches try to generalize a systems behaviour based on massive amounts of experience data, the physics-based approach tries to understand the causal behaviour of a system based on knowledge about first principle laws of physics, causal relations, and logical deduction. Data-driven approaches are thus limited by the experience available, i.e. it cannot reliably predict how the system will behave when a scenario sufficiently different from its experience-base occur. This might lead to dangerous decisions being made in the event of less likely, but highly consequential, events arise. Contrasting this, physics-based methods are well suited to explore new scenarios that have not been experienced yet, and are thus a valuable tool when designing and implementing novel technology and systems in new environments. The downside is that physics-based high-fidelity models are often extremely computationally demanding, as well as the infinite possible realizations of a system is not possible to explore fully, and thus not applicable for real-time decision support.

To be able to capture the complementary advantages of both approaches, a concerted research effort is needed to understand how and when the different approaches can be used to both enhance the result of each other, as well as inform improvements to the individual methods. Research is also needed to understand how to integrate results from one approach into another, and how relevant uncertainties and assumptions need to be handled when carried over from one domain to another.

A review article addressing the science of wind energy were just published by Veers et al. (2019) and they focus on three grand challenges, where this proposal directly addresses the second one:

  • Aerodynamics, structural dynamics, and offshore wind hydrodynamics of enlarged wind turbines.

# Objectives

Main objective:

Develop and document the methodologies and technologies needed to consistently combine physics-based and data-driven models to alleviate the deficiencies of both by capturing their complementary advantages.

Secondary objectives:

  • Increase computational efficiency of advanced modelling tools by reduced-order modelling.
  • Develop hybrid analysis and modeling, i.e., combine physical models with data-driven models.
  • Reduce computational demand and increase safety by effective selection of relevant simulation scenarios based on probabilistic machine learning and risk aware objective functions.
  • Tailoring and demonstrating the integrated modelling approach to the selected use cases.


  • Methods and open source software module for reduced order modeling.
  • Methods and open source software module for integrating physical models with data-driven models.
  • Guideline on problem evaluation and understanding, and subsequent relevant selection of methodologies and technologies.
  • Publication in open access journals and dissemination to relevant Norwegian industries