Graph-Assisted Domain Adaptation for Machine Fault Label Propagation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing unsupervised domain adaptation techniques fail to effectively address significant distribution shifts and disjoint sub-graphs in machine fault diagnosis, particularly when data is limited and unlabeled, leading to poor performance in label propagation.

Innovation Solution

The Graph Assisted Unsupervised Domain Adaptation (GA-UDA) technique minimizes class-wise maximum mean discrepancy to transform data into a shared feature space and iteratively updates a joint graph using optimized projection matrices, enabling label propagation across domains.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a fixed graph is used for label propagation in domain adaptation, then label propagation works when domain discrepancy is small, but the graph becomes disjoint with no edge connectivity when domain discrepancy is large

Engineering Contradiction:
Improvelabel propagation accuracyVSAvoidgraph connectivity under large domain discrepancy
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent applies dynamics by making the graph structure adaptive rather than fixed. The graph connectivity is dynamically adjusted based on the domain discrepancy between source and target domains. When domain discrepancy is large, the graph automatically reduces connectivity to avoid incorrect label propagation, while maintaining connectivity when domains are similar. This is achieved through a domain discrepancy-aware graph construction mechanism that adaptively controls edge existence based on feature similarity and domain alignment metrics.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If existing unsupervised domain adaptation techniques are used, then marginal distribution alignment is addressed, but conditional distribution discrepancy is ignored leading to poor performance

Engineering Contradiction:
Improvemarginal distribution alignmentVSAvoidconditional distribution alignment
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent applies segmentation by separating the distribution alignment task into two distinct components: marginal distribution alignment and conditional distribution alignment. The marginal alignment handles the overall data distribution between domains, while the conditional alignment specifically addresses the class-conditional distributions. This segmentation allows each component to be optimized independently with appropriate metrics, ensuring both aspects are properly addressed rather than treating them as a single unified problem.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If manual labeling is performed for machine fault diagnosis, then accurate labels are obtained, but the process is time consuming and economically not viable

Engineering Contradiction:
Improvelabel accuracyVSAvoidlabeling time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies self-service by enabling the system to automatically generate labels for the target domain without human intervention. Through unsupervised domain adaptation and graph-based label propagation, the system self-generates accurate labels by leveraging source domain knowledge and the structural relationships in the data. This eliminates the need for time-consuming manual labeling while maintaining label accuracy, as the system serves itself by transferring knowledge across domains.

Inventive Principle:
Principle #25Self-service

4Quantity of substance

If limited data is available for training, then training cost is reduced, but training effectiveness deteriorates due to insufficient data

Engineering Contradiction:
Improvetraining data volumeVSAvoidmodel performance
Core Design Contradiction:
Quantity of substanceVSReliability

Solution Approach 1:

The patent applies universality by enabling a single source domain model to serve multiple target domains through domain adaptation. Instead of requiring separate training data for each target domain, the adapted model universally applies to multiple target domains by learning domain-invariant features and relationships. This multi-functionality allows the model to generalize across different machines and operating conditions, effectively multiplying the utility of limited training data across multiple applications.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4488892B1Methods and systems for graph assisted unsupervised domain adaptation for machine fault diagnosis
Publication Date: 2026.01.14 TATA CONSULTANCY SERVICES LTD
  • EP4488892B1 patent drawingFigure 1
  • EP4488892B1 patent drawingFigure 2
  • EP4488892B1 patent drawingFigure 3A

AI summary

The disclosure generally relates to methods and systems for graph assisted unsupervised domain adaptation for machine fault diagnosis. The present disclosure solves the technical problems in the art using a Graph Assisted Unsupervised Domain Adaptation (GA-UDA) technique for the machine fault diagnosis. The GA-UDA technique carries out the domain adaptation in two stages. In the first stage, a Class-wise maximum mean discrepancy (CMMD) loss is minimized to transform the data from both source and target domains to a shared feature space. In the second stage, the augmented transformed (projected) data from both the source and the target domains are utilized to construct a joint graph. Subsequently, the labels of target domain data are estimated through label propagation over the joint graph. The GA-UDA technique of the present disclosure helps in addressing significant distribution shift between the two domains.