The knowledge and expertise gained over the past century by members of the superconducting community have demonstrated that effectively addressing the challenges encountered by this technology frequently necessitates the integration of other disruptive technologies into the realm of superconductivity.
An article accepted for publication in the journal Superconductor Science and Technology demonstrates the application of artificial intelligence (AI), machine learning (ML), deep learning (DL), and big data (BD) in enhancing the development, production, operation, identification of faults, and monitoring of the condition of superconducting technology.
Background
Over the last century, superconductivity enthusiasts had anticipated its complete integration into diverse industrial sectors. This sense of anticipation stems from recognizing that conventional alternatives lack the competitive advantages inherent in superconductors and superconducting technology. These advantages include stronger magnetic field strength, lighter weight, enhanced current carrying capacity, smaller size, reduced losses, improved efficiency, and higher power density.
Study: Revolutionizing Superconductivity: Harnessing AI, Machine Learning, Deep Learning, and Big Data. Image Credit: everything possible /Shutterstock
However, the progress in commercializing superconducting devices has been hindered by various technical obstacles, including high initial purchasing costs, currently high levelized energy costs, and difficulties in the supply chain related to producing low-cost, high-temperature superconducting (HTS) materials. Moreover, these devices are often implemented at the demonstrator level rather than in real sizes, which introduces uncertainties regarding the technological limitations in manufacturing commercial-scale devices.
The current study discusses numerous significant problems regarding the usage of superconducting devices in crucial domains. The first concern pertains to the reliability of superconducting devices, particularly in relation to the occurrence of quench phenomenon. The second concern relates to the challenges the cryogenic environment poses and the need for a relatively inefficient cooling system.
Significance of the results
The present study identifies that AI techniques play a crucial role in the domain of superconductivity by offering optimal design solutions for a range of superconducting devices, such as flux pumps, magnets, electric machinery, and other large-scale apparatus.
The study also illustrates the significance of superconducting magnets in the fusion industry. It claims that in the foreseeable future, compressed nuclear fusion-based power plants will be capable of producing economically viable electrical energy without causing any pollution. The second generation of fusion power plants after the construction of ITER is envisaged as demonstration power plants (DEMO), which are massive, high-field, steady-state tokamaks.
The most important DEMO parts have traditionally been designed using a blend of logical and finite part approaches in a serial trial-and-error way. However, these trial and error methods are costly in terms of time and calculation. Swarm-based optimization and artificial neural networks are examples of AI techniques that have been developed to speed up computation and locate optimal design configurations more effectively. They guarantee convergence to an ideal resolution to the winding pack arrangement and greatly minimize the computational time needed for magneto-structural analysis. The DEMO configuration of 2019 ENEA, comprising 16 toroidal field coils made with a wind-and-react method, has successfully used the proposed technology.
It was observed that AI could consider design variables that are both continuous (real-valued) and categorical (discrete). The utilization of these techniques results in a notable reduction in the computational time necessary for conducting magneto-structural analysis. Furthermore, they ensure the attainment of an optimal solution for the configuration of the winding pack, with convergence being guaranteed.
Furthermore, AI methods can be used to develop surrogate models, which facilitate the utilization of various optimization methods to identify the most optimal solution(s). Lastly, AI techniques possess the capability to understand and consider the interdependencies among various inputs comprehensively. Hence, the suggested AI techniques can be employed to analyze superconducting equipment during the modeling phase, functioning as a readily adaptable code.
Conclusion
This paper presents the latest advancements and pioneering research endeavors in artificial intelligence applied to superconductivity. These efforts encompass a wide range of applications, including the utilization of AI techniques in the design process as well as the creation of novel methods for monitoring the condition of superconductors and superconducting devices and systems.
According to the author, there is hesitancy among major industries to adopt new technologies when a well-established conventional alternative is already widely demonstrated. However, the author hopes that this study inspires researchers in the superconductivity community, urging them to acknowledge and utilize the potential of AI and big data methodologies to explore novel prospects in the realm of superconducting materials and technologies.
Machine Learning Powering Breakthroughs in Climate Forecasting and Modeling