Use of open-source software is steadily increasing in various application areas, including industrial research and development. The more research-oriented an activity is, the more significant the benefits of openness are. In most cases, cost savings are not among the key benefits. The advantages of openness, in terms of cooperation and various ways of using software, are more important. Software is not just a precision tool. It can also form an infrastructure in which an activity is freely developed in different directions.
Use of open source markedly facilitates collaboration, as all parties can use the same software and everyone can immediately access the latest version through common version control. At best, new ideas and models of academic research are immediately available within companies, and software development is guided by the needs of research and industry. Anyone can implement new features and fix bugs in open-source code – there is no need to order the work from the owner of the software. However, even in the case of open source, it is important that one party monitors and controls the overall development.
Of course, openness brings challenges. The use and development of open-source software involves more work, and thus the path is not worth taking unless major benefits are to be expected.
What is Elmer?
Elmer is a multiphysical modelling software package based on the finite element method (FEM) – a tool that has been used by the industry for years [1, 2]. In the electromechanical industry ABB and Trafotek have been the pioneers in the use of the software in their R&D. Elmer's user base has grown, leading to the formation of the Elmer/EM cooperation network for Finnish industry and research. VTT coordinates this network. The work has been realised in the form of several research projects, the most significant of which was the SEMTEC project, funded by Tekes and industry, in 2015–2017. In addition to VTT and CSC, the project involved six companies (ABB, Ingersoll-Rand, Kone, Skanveir, Sulzer and Trafotek) and three universities (Aalto University, Tampere University of Technology and Lappeenranta University of Technology).
The electromechanical industry manufactures products such as electrical motors, generators and transformers. In Finland, the industry is largely based on fairly small production series and customised products. Product development must therefore be efficient, fast and inexpensive. This is best achieved through computer-based design and modelling. The sector is highly interested in electromagnetic, mechanical and thermodynamic phenomena and their modelling. Commercial software has not conquered the field in the same way as, say, software focused on structural mechanics. For example, heavy three-dimensional computation remain challenging for commercial software. Open source therefore offers highly attractive opportunities for meeting the needs of the electromechanical industry.
The industry and research sector's interest in Elmer aroused mainly due to its additivity and speed of computation. It is easy to add missing functions to Elmer, and to integrate the software with the company's own product development environment. Elmer also provides efficient parallel computing needed to harness the full capacity of workstations and computing clusters up to thousands of cores.
Computation of electrical motors and generators is particularly demanding numerically, because the presence of rotation and nonlinear materials require solution to a nonlinear system at tens of thousands of time-steps. Without parallel computing, a desired threedimensional electromagnetic computation of electrical machines would take weeks, or even months. This is far too slow for an efficient product development process. For this reason, electrical machine computation is traditionally limited to special cases, where two-dimensional computation is possible and computation times are just a few minutes.
The parallelisation of the finite element method is typically based on the division of the computational task into relatively independent sub-tasks (i.e., partitions) to be treated in parallel on dedicated computing cores. Elmer's design is based on parallelisation using the MPI standard. Both the assembly and solution of linear systems are executed in parallel. In the collaboration between VTT, CSC and ABB, Elmer's parallel performance was in particular enhanced for electrical motor computation. A characteristic feature there is the rotating interface between the rotor and stator that complicates the efficient parallelisation . In addition, when modelling electrical machines, the computation must often include an electrical circuit acting as a supply to a motor or a load for a generator. This has also proven challenging with respect to parallel performance; improvements have been made in this regard in recent years . VTT has in particular studied the simplification of the three-dimensional computation, to make it as fast as two-dimensional computation, but with a precision closer to that of three-dimensional models .
Coupled electromagnetics and mechanics problems
A key feature for the electromechanical industry is the software's ability to solve coupled problems in the fields of electromagnetics and mechanics. The interaction of the electromagnetic field with the material is usually represented using the Lorentz force or, more generally, by Maxwell's stress sensor. However, their direct use in mechanical coupling is challenging. The virtual work principle, based on energy functions, has been by far the most useful approach to such problems.
By matching the virtual transformations with the computational mesh, a simple representation was achieved in which the effect of the electromagnetic field presents as loads in the structural mechanical element solver. In this, a force vector, or generalised nodal force, is added to each nodal point of the computing network. The advantage of this approach is that the aforementioned loads are independent of the mechanical model used. In addition, the model works in exactly the same way in both two and three dimensions, facilitating its verification.
Within the framework of the SEMTEC project, generalised nodal forces were implemented in Elmer, based on cooperation between CSC and the Aalto University Electromechanics laboratory . Until recently, research into electromagnetic nodal forces has been fairly academic. However, with the Elmer implementation, such forces have been used to compute torque, vibration and noise in electrical machines, and the capacity of electromagnetic hoists [11, 12]. Without nodal forces, a force model would have had to be derived for each modelling task, but generalised nodal forces enable direct coupling with the mechanical task and they do not significantly increase the model's complexity.
Possibilities of tailoring the workflow
The key challenge with open-source software often lies in the lack of full-fledged user interfaces. Correspondingly, the ability to operate without a fixed user interface provides opportunities for tailored workflows. Whereas a commercial software interface tries to fulfil the full range of potential purposes, open software can be used to develop a simplified user interface for a certain, desired purpose. The result could be an interface through which it is possible to enter design parameters related to a company's product family and automate actual computations using a cloud service, for example. In such a case, modelling, computation and post-processing are integrated into the company's own product development environment; the system automatically creates a model from the given data, computes it and derives the desired quantities from the results. For customised user interfaces, most graphical work phases can be replaced with open libraries and automated scripts. This brings computation closer to product development and eliminates futile steps from the software control. Such a highly customised and automated modelling environment can provide a company with a major competitive advantage.
Open-source tools can also be used to create a traditional graphical workflow. Well-defined interfaces can be used to select the most suitable tools for different work phases. The Paraview software has been developed as the de-facto standard open-source tool for the post-processing of results; it is generally used for the processing of OpenFOAM results, for example, in addition to those of Elmer . There is no equivalently dominant tool for the creation of models. Tools are often selected according to user preferences. The open software most commonly used alongside Elmer are Gmsh  and Salome .
Guidance on all phases of workflow in a specific application area is an important element in the software's deployment. This is traditionally missing in open-source side. VTT has drawn up an extensive tutorial on using the Gmsh-Elmer-Paraview workflow in the two-dimensional modelling and simulation of induction motors .
The benefits of commercial software are undeniable particularly in pre-processing. Open-source software is often written as part of academic research, where there are no mechanisms for developing user interfaces that correspond to commercial ones. For this reason, it is often rational to combine commercial pre-processing with open-source solvers, thereby exploiting the strengths of both ecosystems.
While open source has undeniable advantages, various bottlenecks can prevent its application. Efficiency is the key issue in computationally heavy models; at worst, models become unusable if computation is too slow. Similarly, a lack of sufficient physical models can be a critical factor that determines whether software can be used in general. Finally, the usability of the software must be sufficient for the intended type of use, including applications to workflows. When all the critical success factors are in place, R&D activity based on open source can rise to the next level.
Janne Keränen, D.Sc. (Tech.), is a Senior Scientist and works as the Principal Investigator in the Electrical powertrains and storage team. He is leading VTT's scientific spearhead on Computational machine dynamics. Keränen is researching and developing various computational methods in electrical machine applications, particularly in the areas of electromagnetics and multiphysics. He coordinates Elmer/EM cooperation and also coordinated the SEMTEC project.
Juhani Kataja, D.Sc. (Tech.), is developing the Elmer software and computing models for electromagnetics. He also implements such models as an application specialist at CSC. He finished his doctoral dissertation at Aalto University in 2014 on the efficient use of boundary element methods in electromagnetic shape optimisation.
 M. Malinen, P. Råback, "Elmer finite element solver for multiphysics and multiscale problems", in (Eds.) I. Kondov, G. Sutmann, Multiscale Modelling Methods for Applications in Material Science, IAS Series, vol. 19, pp.101–113, Forschungszentrum Jülich, 2013.
 J. Keränen, J. Pippuri, M. Malinen, J. Ruokolainen, P. Råback, M. Lyly, K. Tammi, "Efficient Parallel 3-D Computation of Electrical Machines with Elmer", IEEE Transactions On Magnetics, 51(3), March 2015.
 E. Takala, E. Yurtesen, J. Westerholm, J. Ruokolainen, P. Råback, "Parallel simulations of inductive components with Elmer finite-element software in cluster environments", Electromagnetics, 36(3), pp. 167– 185, 2016.
 J. Keränen, P. Ponomarev, J. Pippuri, M. Lyly, J. Westerlund, "Parallel Performance of Multi-Slice Finite-Element Modelling of Skewed Electrical Machines", IEEE Transactions On Magnetics, 53(6), June 2017.
 P. Ponomarev, "Elmer FEM - Induction Machine Tutorial", VTT Report, 2017. https://doi.org/10.13140/RG.2.2.18599.75688
 S. Sathyan, A. Belahcen, J. Kataja, T. Vaimann, J. Sobra, "Computation of Stator Vibration of an Induction Motor using Nodal Magnetic Forces", 2016 XXII International Conference on Electrical Machines (ICEM), Lausanne, 2016, pp. 2198-2203.
 S. Sathyan, A. Belahcen, J. Kataja, F. Henrotte, A. Benabou, Y. Le Menach, "Computation of Magnetic Forces Using Degenerated Air-Gap Element", IEEE Transactions on Magnetics, 53(6), 2017.
 J. Keränen, P. Ponomarev, S. Sathyan, J. Kataja, A. Belahcen, "Magneto-structural simulation of an induction motor start-up using nodal magnetic forces in Elmer", Rakenteiden Mekaniikka, 50(3), pp. 296–299, 2017.