In wind power generation, equipment anomalies and cyber threats impose substantial operational and economic risks. While multi-party data sharing among turbine manufacturers, operators, and analytics providers could improve anomaly detection, privacy issues restrict collaboration. This paper proposes FHEPdM (Fully Homomorphic Encryption-enabled Secure Collaborative Anomaly Detection Architecture), a novel framework enabling privacy-preserving multi-party predictive maintenance without revealing sensitive data. Utilizing fully homomorphic encryption (FHE), it trains and infers a lightweight logistic regression model on encrypted Supervisory Control and Data Acquisition (SCADA) data for end-to-end confidentiality. Integrated with a hybrid Elliptic Curve Diffie-Hellman-fully homomorphic encryption (ECDH-FHE) key management system, it reduces latency while upholding 128-bit security. Evaluation on the Wind Turbine SCADA Dataset yields 0.979 accuracy, 0.52 membership inference attack success, and 17-minute training, delivering a practical, scalable solution for secure industrial cooperation in wind systems.
