転移学習による熱源システム自動不具合検知・診断の学習コスト・不具合データベース削減

抄録

We investigated the transfer learning in the Automated Fault Detection and Diagnosis (AFDD) method for cooling plants using Convolutional Neural Networks (CNN) trained with Fault Database (FDB) created by simulations. The trade-off between the reduction of computation time and the degradation of diagnostic performance by transfer learning is demonstrated by a case study of two real systems.