A cross-condition neural network feature decoupling method and system

By employing a condition-driven feature decoupling method and utilizing nonlinear spatial mapping and dynamic projection mechanisms, the instability and generalization ability of equipment state recognition models under cross-condition conditions are addressed. This enables the expression of state features that remain unchanged under the operating conditions, thereby improving recognition accuracy and robustness.

CN122432634APending Publication Date: 2026-07-21FUZHOU ZHONGGONG CLOUD INTELLIGENT MFG TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
FUZHOU ZHONGGONG CLOUD INTELLIGENT MFG TECH CO LTD
Filing Date
2026-04-28
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing technologies suffer from decreased recognition accuracy and insufficient generalization ability in equipment status recognition models under cross-operating conditions, making it difficult to effectively separate status information from operating condition information. This results in insufficient stability and generalization ability of the model under complex multi-operating conditions.

Method used

By constructing a condition-driven feature modulation and dynamic decoupling mechanism, and using a nonlinear spatial mapping function and a dynamic projection mechanism, the decoupling of state features and condition information is achieved. Joint optimization training is performed using the feature consistency loss function and the state discrimination loss function to generate condition-invariant state feature components.

Benefits of technology

It improves the accuracy, robustness, and generalization ability of equipment status recognition under various working conditions, and enhances the convergence efficiency and training stability of the model under complex multi-working conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122432634A_ABST
    Figure CN122432634A_ABST
Patent Text Reader

Abstract

The application provides a cross-working-condition neural network feature decoupling method and system, relates to the technical field of artificial intelligence and industrial equipment state monitoring, and acquires multi-source time sequence data under different working conditions; extracts initial state features through a neural network model containing a state feature coding branch, a working condition feature coding branch and a feature space generation branch; modulates and transforms the features by using a nonlinear space mapping function driven by working condition features, and decomposes the features into a first subspace feature component which is invariant to working conditions and a second subspace feature component which carries working condition information through a dynamic projection mechanism; constructs a feature consistency loss function and a state discrimination loss function, and performs joint optimization training, so that the first subspace feature component meets the distribution consistency constraint; and finally is used for equipment state recognition or performance evaluation. The application realizes effective decoupling of working condition factors and state features, and improves the accuracy and stability of cross-working-condition recognition.
Need to check novelty before this filing date? Find Prior Art