A fault diagnosis method based on federated learning of dynamic weighted voting

By employing a federated learning approach combining dynamic weighted voting and adversarial learning, this study addresses the fault diagnosis challenges posed by heterogeneous data distribution and scarce labels in multi-client scenarios. This improves the accuracy of fault diagnosis and the adaptability of the model for critical equipment such as compressors, and is applicable to fault diagnosis of hydraulic systems in petroleum hydrocracking processes.

CN121167378BActive Publication Date: 2026-06-09NANJING TECH UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Patents(China)
Current Assignee / Owner
NANJING TECH UNIV
Filing Date
2025-08-14
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

In industrial production, especially in the hydrocracking process of petroleum, fault diagnosis of key dynamic equipment such as compressors faces challenges such as data privacy constraints, noise suppression, and decoupling analysis of multi-dimensional fault features. Traditional federated learning methods are difficult to effectively solve the problems of heterogeneous data distribution and scarce target labels in multi-client scenarios, resulting in insufficient model adaptability.

Method used

A federated learning method based on dynamic weighted voting is adopted. By calculating the feature similarity between the target client and multiple source clients, the weights are dynamically adjusted. Combined with adversarial learning and transfer learning, the prediction accuracy and generalization ability of the fault diagnosis model are optimized.

Benefits of technology

It improves the prediction accuracy and model adaptability of the target client in small sample scenarios, reduces the dependence on a large amount of local data, enhances the accuracy and reliability of fault diagnosis, and improves the data fusion efficiency in multi-client scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121167378B_ABST
    Figure CN121167378B_ABST
Patent Text Reader

Abstract

A fault diagnosis method based on federated learning with dynamic weighted voting is presented. In a production control system, similar devices are considered clients, controlled by a server. Clients include newly accessed devices from the target client and existing devices from the source client. A fault diagnosis model is deployed on each client. Test samples are input into the local diagnostic model to obtain fault classification results. The construction steps of the fault diagnosis network model include: sample data distribution and model initialization; feature extraction and adversarial learning; feature comparison and pseudo-label generation; incremental transfer learning and optimization. This invention overcomes the sensitivity of traditional methods to data heterogeneity and noise. This method not only reduces the target client's dependence on large amounts of local data but also improves the prediction accuracy and generalization ability of the target client in small-sample learning scenarios.
Need to check novelty before this filing date? Find Prior Art