Yi, Chen-PeiChen-PeiYiTsai, I-HaurI-HaurTsaiChou, Po-HuanPo-HuanChouChung, Wei-DerWei-DerChungSHIH-CHIN YANG2026-04-142026-04-1420259798331541309https://www.scopus.com/record/display.uri?eid=2-s2.0-105030344949&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/737170This paper presents a novel motor bearing fault diagnosis method based on the artificial intelligent based convolutional conditional autoencoder (CCAE). Conventional fault diagnosis methods typically require the fault reflected harmonic signatures among different failures. The proposed CCAE only requires normal phase current signals. It significantly reduces the complexity of bearing fault diagnosis process. In this paper, a one-dimensional (1-D) CCAE based on normal current signals is proposed and trained to create a single-class model. By using this unsupervised classification, the proposed CCAE can detect the anomaly bearing fault. In addition, a Durenberger-style state observer is proposed to estimate the motor speed and load torque conditions and to combine the state observer with CCAE, the bearing fault can be identified under different load and speed conditions. Based on experimental results, it is shown that the proposed CCAE with state observer can effectively detect bearing anomaly for different faults. These faults might not be detected based on conventional spectrum analysis of fault harmonic signatures.falseanomaly detectionautoencoder (AE)Motor current signature analysis (MCSA)unsupervised learningconference paper10.1109/ECCE58356.2025.112604002-s2.0-105030344949