Generative knowledge-based transfer learning for few-shot health condition estimation

Abstract In the field of high-end manufacturing, it is valuable to study few-shot health condition estimation. Although transfer learning and other methods have effectively improved ability learning, they still cannot solve lack prior knowledge. this paper, by combining data enhancement, knowledge reasoning, a generative knowledge-based model proposed achieve First, with effectiveness enhancement on machine novel batch monotonic adversarial network (BM-GAN) designed for generation, which can problem insufficient generate simulated training data. Second, performance advantages belief rule base (BRB) method combines expert obtain generalized BRB then fine-tunes real dedicated model. Third, through uniform sampling NASA lithium battery simulating conditions, transfer-belief (GT-BRB) in paper verified be feasible estimation improves accuracy approximately 17.3%.