{"title":"A reverse design method for cryocooler regenerator based on artificial neural network","authors":"Shanshan Li, Xiantong Chen, Hao Chen, Ziyang Wang","doi":"10.1016/j.cryogenics.2025.104053","DOIUrl":null,"url":null,"abstract":"<div><div>In conventional cryocooler regenerator design, researchers typically employ specialized software to traverse numerous parameter combinations to assess performance, comparing the results against predetermined requirements to determine the optimal structural dimensions and operating parameters. However, this approach suffers from low research and development efficiency, and the optimal solution may not be included in the tested cases. To address these issues, this study proposes a novel reverse design methodology that directly calculates the regenerator structural dimensions and operating parameters based on predefined target performance. The method centers around the deep learning artificial neural network (ANN) model as its core, effectively addressing the challenge of setting input and output parameters during the model’s learning process using the performance dataset obtained through forward calculation of the regenerator. Additionally, a wide-range and multi-step input parameter adjustment mechanism is incorporated, thereby ensuring the method’s adaptability and practical applicability. The optimized ANN model demonstrates promising results, with an average relative error of 2.36% for regenerator length prediction, 4.79% for cold end mass flow prediction, and 4.73% for coefficient of performance (COP) prediction. Extensive model accuracy assessments confirm the model’s predictive capabilities across various intervals of the predicted variables. By utilizing the rapid computational speed of deep learning models and their proficiency in parallel processing, the method proposed in this study dramatically shortens the regenerator design time to mere seconds, thereby introducing a pioneering approach for rapid and precise regenerator design.</div></div>","PeriodicalId":10812,"journal":{"name":"Cryogenics","volume":"148 ","pages":"Article 104053"},"PeriodicalIF":1.8000,"publicationDate":"2025-03-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Cryogenics","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0011227525000311","RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"PHYSICS, APPLIED","Score":null,"Total":0}
引用次数: 0
Abstract
In conventional cryocooler regenerator design, researchers typically employ specialized software to traverse numerous parameter combinations to assess performance, comparing the results against predetermined requirements to determine the optimal structural dimensions and operating parameters. However, this approach suffers from low research and development efficiency, and the optimal solution may not be included in the tested cases. To address these issues, this study proposes a novel reverse design methodology that directly calculates the regenerator structural dimensions and operating parameters based on predefined target performance. The method centers around the deep learning artificial neural network (ANN) model as its core, effectively addressing the challenge of setting input and output parameters during the model’s learning process using the performance dataset obtained through forward calculation of the regenerator. Additionally, a wide-range and multi-step input parameter adjustment mechanism is incorporated, thereby ensuring the method’s adaptability and practical applicability. The optimized ANN model demonstrates promising results, with an average relative error of 2.36% for regenerator length prediction, 4.79% for cold end mass flow prediction, and 4.73% for coefficient of performance (COP) prediction. Extensive model accuracy assessments confirm the model’s predictive capabilities across various intervals of the predicted variables. By utilizing the rapid computational speed of deep learning models and their proficiency in parallel processing, the method proposed in this study dramatically shortens the regenerator design time to mere seconds, thereby introducing a pioneering approach for rapid and precise regenerator design.
期刊介绍:
Cryogenics is the world''s leading journal focusing on all aspects of cryoengineering and cryogenics. Papers published in Cryogenics cover a wide variety of subjects in low temperature engineering and research. Among the areas covered are:
- Applications of superconductivity: magnets, electronics, devices
- Superconductors and their properties
- Properties of materials: metals, alloys, composites, polymers, insulations
- New applications of cryogenic technology to processes, devices, machinery
- Refrigeration and liquefaction technology
- Thermodynamics
- Fluid properties and fluid mechanics
- Heat transfer
- Thermometry and measurement science
- Cryogenics in medicine
- Cryoelectronics