A method for matching device fault description and maintenance measures semantics

By using a frozen contrastive learning encoder and a lightweight linear drift map, combined with dynamic causal masks and a fault simulator, the problem of semantic matching between unstructured fault descriptions and maintenance measures is solved, achieving high-precision and robust semantic matching and reducing the risk of mismatches.

CN122389846APending Publication Date: 2026-07-14WUHAN UNIV OF SCI & TECH +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Applications(China)
Current Assignee / Owner
WUHAN UNIV OF SCI & TECH
Filing Date
2026-03-18
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively understand the deep semantic relationships between unstructured fault descriptions and maintenance measures, and lack semantic consistency verification, leading to mismatches and secondary risks.

Method used

A frozen contrastive learning encoder is used to unify the encoding of fault and maintenance semantics. By combining a lightweight linear drift mapping and a dynamic causal masking strategy, the alignment and consistency verification of fault and maintenance semantics are achieved. A fault simulator is used for inverse state deduction.

Benefits of technology

It improves the semantic matching accuracy and robustness of fault description and maintenance measures, reduces the risk of mismatch, and enhances the reliability and adaptability of the semantic matching system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122389846A_ABST
    Figure CN122389846A_ABST
Patent Text Reader

Abstract

The present application relates to the technical field of equipment intelligent operation and maintenance, and particularly relates to a device fault description and maintenance measure semantic matching method, comprising the following steps: S1, inputting fault text and maintenance corpus into a same frozen encoder to generate a latent vector, and outputting a drift key vector after linear drift of the fault vector through a light matrix; S2, taking the drift key vector as a query to retrieve k candidate measures and aftereffects, splicing the candidate measures and aftereffects into a segment and applying a mask to generate a masked measure list; S3, inputting each masked measure into a frozen simulator to generate a predicted state, calculating a complementary overlap degree of the predicted state with an original fault, and retaining those less than a threshold value. Through latent space semantic alignment, dynamic causal mask driven by working conditions, and complementary overlap verification based on a fault simulator, the present application realizes high correlation, low conflict and explainable consistent semantic matching between unstructured fault descriptions and maintenance measures.
Need to check novelty before this filing date? Find Prior Art