We presented a collaborative privacy-preserving algorithm for power defect detection involving multiple parties. With increasing concerns regarding data privacy in the power industry, ensuring user privacy has become a critical issue in power defect detection. To address this challenge, this paper proposed an algorithm that combines multi-party collaboration and homomorphic encryption to achieve efficient power defect detection while safeguarding data privacy. First and foremost, we employ a distributed data encryption approach, wherein power-related data is distributed among various entities, including power companies, energy providers, and consumers. By utilizing homomorphic encryption techniques, the confidentiality and integrity of the data are ensured during transmission and storage. Next, we introduced the concept of federated learning to enable collaborative model training. Each party conducts model training in an encrypted state and computes encrypted gradients. These encrypted gradients are securely aggregated on a central server using homomorphic encryption, resulting in updated global model parameters. This design ensures data privacy while facilitating centralized model training and parameter sharing. Finally, secure result inference is achieved through homomorphic encryption. The central server decrypts and distributes the encrypted global model parameters to the respective parties. Each party then locally decrypts the received parameters and performs power defect detection. This process guarantees the privacy and security of the detection results.