DEFACTO project is fully aligned with the so-called Industry 4.0 effort through the development of digital twins for the Li-ion cell industry – Interview with UPM (Polytechnic University of Madrid)defacto
Fernando Varas, from UPM (Polytechnic University of Madrid), speaks in this interview about their main role in the DEFACTO Project, their expectations and the work they will carry out.
Q: What is the main role of your entity in DEFACTO?
A: As stated in the presentation, the main goal of the DEFACTO project is the development of multiphysics and multiscale models able to provide an in-depth understanding of cell materials, cell manufacturing processes and cell performance. Models developed by the different partners in this project account for all the complex phenomena involved in cell manufacturing and operation.
Unfortunately, the numerical solution of the resulting detailed models can be extremely expensive in terms of computational cost, thus requiring powerful computers or a very large computation time. This fact can seriously hinder the adoption of modelling tools resulting from the project. Our main contribution to DEFACTO project is focused on the acceleration of the numerical simulation. This is being done using model order reduction techniques.
More precisely, we are using these acceleration techniques in the framework of the analysis of a particular cell/battery design and also in connection with a cell/battery optimization (which is completely unaffordable without a significant acceleration of the numerical simulation of the cell performance for each combination of cell design parameters). But these model order reduction techniques are useful in other tasks too. In particular, model-based cell parameters identification techniques can significantly decrease the effort of experimental testing campaigns by reducing the number of tests to be performed. In collaboration with CIDETEC we are using reduced-order models to design more efficient experimental tests to identify, for instance, electrolyte properties.
Q: What are your expectations from the project?
A: Firstly, I would like to highlight that for us this project is a great opportunity to work with excellent research groups from Europe. We are learning very much from partners with quite different modelling backgrounds as well as partners with large experimental expertise.
On the other hand, I think DEFACTO project is fully aligned with the so-called Industry 4.0 effort through the development of digital twins for the Li-ion cell industry. These tools can prove essential concerning the goal of the European Commission to build a strong European battery industry, as promoted through the European Battery Alliance launched in October 2017.
In this sense, it is important to make advanced modelling tools easily available to European companies and laboratories. This is why DEFACTO project will deliver cell modelling software (including most of the detailed physics identified by our partners) released under an open-source software license. CIDETEC is doing a great job concerning the development of this piece of software, a task we are also contributing to. As previously mentioned, our main task will be to accelerate this numerical simulation software (using reduced-order model techniques) in order to provide European companies and laboratories with a fast, open-source modelling tool easy to integrate into the cell design cycle.
Q: Can you elaborate on the work you are doing/planning to do?
A: The basic idea behind the acceleration techniques we are implementing is rather simple. When the numerical simulation of a cell is done for the first time, the solution needs to be searched in a very large set of solutions. If, for instance, a finite element technique is used then any combination of numerical values for each variable at each node is allowed. For a 3D battery model, this means looking for a vector (containing values of the model fields at mesh nodes) with several millions of components. Thus a high computational burden is expected to solve this numerical model.
Once a cell has been simulated it is easily realized, by inspection, that model fields follow some particular (spatial) patterns. Thus looking for any combination of field values (the reason behind the high computational cost) seems rather inefficient. Instead, fields adopting the observed patterns could be searched. Since the observed patterns can be typically quite accurately reproduced using the (linear) combination of a few functions, the numerical simulation of the multiphysics model can be done using a very low number of unknowns (resulting in a very low computational cost). This inexpensive numerical model is called a reduced-order model.
Of course, there is still an important difficulty to overcome: those field patterns (allowing us to formulate our reduced-order model) need to be identified by solving the full model. Standard model order reduction techniques follow a preprocessed strategy: (a) the full model is first solved for a wide enough range of design and operation parameters, (b) patterns are extracted from these solutions, and (c) a reduced-order model is then prepared from these observed patterns. Any subsequent numerical simulation can be performed, in a very efficient way, using the reduced-order model. This approach can be very useful in some situations (for instance, following this technique a very accurate and inexpensive model can be integrated into a battery management system or BMS) but rather inefficient in many other cases.
The reason for the inefficiency of the preprocessed strategy is that the overall computational cost is larger than the direct numerical solution of the model since a large number of numerical simulations must be performed (in a previous, off-line phase) to identify observed patterns for a suitable range of design and operation parameters. As commented, an expensive offline phase can be afforded in some situations provided a very inexpensive online phase is facilitated (such as in BMS applications). In contrast, researching new battery designs involves the numerical solution of cells with quite different parameters and operating conditions. In this case, the computational cost associated with the simulation of an extraordinarily large set of cases (needed to cover all the possible configurations) become almost unaffordable and, in any case, quite inefficient.
The approach we adopt in the framework of DEFACTO project is the development of an adaptive reduced-order model. Following this approach, the expensive computation of a large set of cases to identify patterns for each physical variable is not needed (anyway, a few cases can be solved to improve our strategy). Physical variable patterns corresponding to the particular cell and operation conditions we are interested in can be identified instead if we solve the full numerical model for a short time. Thus we begin solving the full (expensive) model but once patterns have been identified we can switch to an (inexpensive) reduced-order model to continue the numerical simulation. In addition, we need to implement some checks to ensure our reduced-order model is accurate enough. Otherwise (since physical variable patterns may have evolved) we return to the full model to identify the new patterns and reconstruct our reduced-order model. Of course, the less time we need to run the full model the larger the acceleration of the numerical simulation. At this point of the project, we are designing a strategy to spend time as low as possible integrating the full model, in order to maximize the simulation acceleration.