Galileo Ferraris Contest
Comparing data-driven methodologies for the multi-physics simulation of traction electrical machines
Who was Galileo Ferraris
Operative Rules for GalFer’s Contest
Structure of the Contest
The contest aims to compare the accuracy and efficiency of data-driven procedures when applied to a dataset regarding electric motors for traction. The Contest has the following rules.
What contest will provide
- datasets on electrical machines will be made available on the GitHub website only to the research teams that will join the Contest;
- datasets are open according to Apache License 2.0;
- details about the technical contest (kind of motor, target performance, design variable space, etc.) are provided in the slides available on this GitHub website;
- two complete training datasets, produced according to the workflow described in Contest and referring to two kinds of motor, will be shared using .csv files containing, in each row, the variables defining motor configuration and motor performance. The number of configurations in each public dataset will be 4096; the third dataset, referring to a hidden motor, will be provided in a limited size of 256 configurations;
What research groups will provide
- research groups that intend to join the Contest will send a letter of intent (template and deadline provided in the following);
- research groups will provide procedures either in Matlab or in Python source code and a document, in the form of slides, explaining their procedures;
- research groups taking part in the Contest will have to train their procedures on the two public datasets and test them on the third dataset;
- organizers will re-run the procedures to assess results and computational effort for the training and evaluation phase;
- the focus of the Contest will be on the multi-criteria evaluation capabilities of the data-driven procedures, thus the target of the procedures will be the evaluation of the Pareto Front in the performance space;
- the quality of the results will be judged on the third dataset referring to the hidden motor. The reference Pareto Front (rPF) will be created by Contest organizers by running a dataset with a large number of configurations;
- Results will be evaluated on the basis of the following parameters:
- accuracy: distance from the rPF, Generational Distance (GD), Reverse generational distance (RGD), Spacing (S), and Error ratio (ER);
- efficiency: number of tunable parameters of the model, computational effort of the training process, and computational effort for the evaluation of Pareto Front.
- unless previously defined with Contest organizers, the data and procedures provided by participants will be considered as a public disclosure.
Awards
The number of awards and their economic content will be defined as soon as the sponsorship collection is completed.
Any research group can attend the Contest, submit their results, and have them ranked according to the previously defined criteria. However, awards with economic rewards will be reserved for academic teams formed as defined in the following.
Timeline
- the datasets will be available in September 2024;
- research groups planning to take part in the Contest are invited to submit their intention to join the Contest using the letter of intent following the template provided by mailing it to the address galfercontest@gmail.com, at last by November 1st, 2024;
- sharing of the preliminary results from the research teams and first informal evaluation, by January 15th, 2025;
- publication of the partial dataset on the third machine by January 31st, 2025;
- final results and procedures must be submitted by the research teams by April 15th, 2025;
- award ceremony will be held in a dedicated free workshop during the COMPUMAG 2025 Conference in Naples, June 22-26, 2025.