Skip to main content

Efficient Uncertainty quantification For Optimization in Robust design of Industrial Applications (EUFORIA)

The EUFORIA project was initiated after an animated conversation between the companies NOESIS and NUMECA and VUB. The conclusion was that current state-of-the-art Computer-Aided Engineering (CAE) tools can only account for the uncertainties inherent to the processes in a very limited way, making the obtained global optima unreliable. The inclusion of an advanced and reliable uncertainty quantification in the CAE tools, coupled to an efficient methodology, would therefore be a major breakthrough for CAE, allowing industrial partners to design quicker and obtain better, cheaper and more robust (i.e. less uncertainty sensitive) products.

Euforia servers banner

1. Valorization objectives and the valorization approach of the project

The EUFORIA project was initiated after an animated conversation between the companies NOESIS and NUMECA and VUB. The conclusion was that current state-of-the-art Computer-Aided Engineering (CAE) tools can only account for the uncertainties inherent to the processes in a very limited way, making the obtained global optima unreliable. The inclusion of an advanced and reliable uncertainty quantification in the CAE tools, coupled to an efficient methodology, would therefore be a major breakthrough for CAE, allowing industrial partners to design quicker and obtain better, cheaper and more robust (i.e. less uncertainty sensitive) products.

The valorization deliverable of EUFORIA is, therefore, a computational methodology, capable of reaching a global robust optimum efficiently (EUFORIA RDO-methodology, computer-aided Robust Design Optimization methodology).  This is applicable for many design variables, in a multi-objective setting, while taking into account the (often many) relevant uncertainties.

To support the valorization, the consortium decided to involve SIRRIS in the EUFORIA project. SIRRIS is the collective technology center of the Belgian technological industry. Together with the Consortium partners they implemented the project valorization. To this end, a valorization Board was created that coordinated the dissemination and valorization.

The following major opportunities of our technology have been identified: 1) resulting in better, more efficient, more performant products or processes; 2) changing company know-how from “alchemy” to “science”; 3) preventing draining away of company know-how; 4) resulting in more environmentally friendly products; 5) decreasing the company’s product responsibility risk; 6) savings on R&D time and costs; 7) better defined product marketing strategies and marketing road maps; 8) reduced time-to-market path.

The above listed advantages and opportunities of applying our RDO-methodology within industry will result in higher sales, lower costs, stronger competitive advantages and thus increased profit.

The valorization objectives are twofold.

Firstly: vaporization of our RDO-methodology within the segment of Solution-and-Service-Providers (SSP). The SSP incorporated our innovation within their software solutions and consulting services, to sell those as a product to the end-users. This was done through Software License Agreements which will resulted in royalty fee incomes.

Secondly: valorization of our methodology to the segment of end-users. Our technology can be used in nearly every sector where products and processes have to be designed. Within our own Industrial Users Committee our partners belong already to 6 different sectors, ranging from the software sector to the industrial equipment sector, from the environmental industry towards the steel industry. The business cases and commitments of our industrial partners demonstrated the huge valorization potential of our EUFORIA RDO-methodology.

2. Scientific objectives and the research approach of the project

The final objective of this project was to develop an efficient methodology for the optimization of industrial processes under uncertainty.  The methodology enabled to construct/achieve robust designs, i.e., optimized designs that are as insensitive as possible to parameter fluctuations, to uncertainties in operating conditions, to manufacturing tolerances, etc. The methodology handled uncertainties in the model parameters, as well as uncertainties in the design variables. Hereby, the emphasis is on a large number of design variables and uncertainties. The inclusion of uncertainty quantification (UQ) in the design cycle requires combining the exploration of the design space (optimization) with exploration of the stochastic space and the development and use of accurate and efficient surrogate models.

The focus of the project is on flow related industrial processes, where a computational fluid dynamics (CFD) model is used as prediction tool. Such processes are the basis of many industrial applications related to energy production, HVAC (heating, ventilation and air conditioning), automotive, and aeronautics. They are of great interest to the Flemish industry. In addition, the methodology will be equally applicable to industrial processes described by other non-CFD models. This will allow applications for a whole range of processes in different engineering areas. The wide applicability of the envisaged project outcome is already evidenced by the members of the Users Committee and by the different valorization tracks put forward.

In order to achieve the project objectives, major new developments were needed to bridge the gap from the current scientific state-of-the-art to the required technological readiness level.  Breakthroughs were required with respect to the handling of large numbers of design variables and simultaneous uncertainties and with respect to the computational efficiency demands when addressing the short design cycles for industrial applications.

The first major scientific challenge was to improve the efficiency of the UQ methodology for a large number of uncertainties and this in combination with an expensive model (e.g. CFD).  This, so-called ‘curse of dimensionality’, has been handled in this project. A concerted action was undertaken, where non-intrusive techniques will be utilized in order to tackle large number of uncertainties and high-dimensional problems. In this context, modeling techniques for geometrical uncertainties will be evaluated. Furthermore, an innovative method for epistemic uncertainty quantification has been investigated. To this cause, the development and use of efficient surrogate models will be explored, both in the design space and stochastic space.

The second major challenge was to efficiently capture the global robust optimum in a multi-objective setting, which leads to a multidimensional Pareto front solution. This has been achieved by combining evolutionary algorithms (EA) with gradient-based methods. The EA ensures that the complete design space is explored and allows one to come in the neighborhood of the global optimum. Gradient-based methods subsequently ensure fast convergence to the global optimum. In the gradient-based methods, tailored adjoint formulations are used to efficiently compute the gradients within a UQ context. These aspects were covered in the project.

In the current project, the developed methodology has been applied to a class of flow driven processes, namely the robust design of heat exchangers. This topic was chosen since it is of interest to many Flemish companies, several of them being part of the Users Committee (Keppel Seghers, Vyncke, Bosal). Geometrical uncertainties have been investigated separately, due to their importance in industrial design.

The development of a database of test cases with prescribed uncertainties was a major scientific challenge. This was the subject of one of the work packages, where also an automatic tool for computations needed for the optimization and design procedure was developed.

Finally, supporting techniques have been developed that are indispensable for the successful realization of the project goals. We focused on crucial high-performance computing aspects.