Federated Product Inspection Models for Heterogeneous Data

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing federated learning (FL) systems face insufficient model accuracy due to multimodal data distributions from clients, particularly in industrial settings where multiple data sources like sensors on production machines are used, leading to poor performance.

Innovation Solution

The method creates subsets of clients with similar data distributions by forming 'sub-clients' or cohorts, using similarity analysis based on statistical characteristics, and trains personalized local models within these groups, enhancing model accuracy by ensuring similar clients contribute to their own models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If federated learning is applied to clients with data from multiple underlying data distributions, then data protection is maintained, but model accuracy deteriorates due to multimodal data distributions

Engineering Contradiction:
Improvemodel accuracyVSAvoiddata distribution diversity
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent segments the client population into multiple sub-clients based on their data distribution characteristics using clustering algorithms. Each sub-client group shares similar data distributions, allowing models trained within each group to achieve higher accuracy while still maintaining the federated learning framework's data protection benefits.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements local model personalization where each client receives a customized model adapted to its specific data distribution characteristics. This is achieved by training models on locally captured data subsets and fine-tuning global models with client-specific data, ensuring local optimality while maintaining overall system coordination.

Inventive Principle:
Principle #3Local quality

2Adaptability or versatility

If a global FL model is trained with data from multiple factories, then system-wide coverage is improved, but model performance deteriorates due to heterogeneity in production lines

Engineering Contradiction:
Improvesystem-wide coverageVSAvoidmodel performance
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent segments the heterogeneous global data into multiple homogeneous clusters representing different production line types, factories, or operational conditions. Each cluster forms a sub-client group that trains specialized models, preventing the performance degradation that would result from training a single global model on highly heterogeneous data.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a hierarchical dimension to the federated learning architecture, organizing clients into multiple levels (factories, production lines, devices). This multi-level clustering approach allows the system to maintain both broad coverage and specialized performance by operating at different granularities simultaneously.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Reliability

If single global model is used for all clients, then system complexity is reduced, but model accuracy deteriorates due to inability to capture client-specific characteristics

Engineering Contradiction:
Improvemodel accuracyVSAvoidmodel architecture complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the model training process into multiple stages: global model training followed by local model personalization. This segmentation allows the system to maintain a simple global model structure while adding complexity only where needed for local customization, achieving high accuracy without overwhelming system complexity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent merges the advantages of both simple global models and complex local models by combining a pretrained global model with client-specific fine-tuned components. This merging approach achieves high accuracy through local adaptation while leveraging the simplicity and generalization capability of the global model.

Inventive Principle:
Principle #5Merging (Combining)

4Reliability

If client data is used directly for model training, then model accuracy is improved, but data protection is compromised due to transmission of sensitive information

Engineering Contradiction:
Improvemodel accuracyVSAvoiddata privacy risk
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent extracts only the necessary model parameters (weights and gradients) from the training process while leaving the sensitive raw data localized at each client. This extraction approach allows model improvement through data utilization without transmitting or storing actual client data, maintaining both accuracy and privacy.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces encrypted communication channels and secure aggregation protocols as intermediaries between clients and the central server. These intermediaries enable the transmission of model updates while protecting the underlying data, allowing accurate model training without compromising data privacy.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4645162A1Method and system for inspecting or operating a product
Publication Date: 2025.11.05 SIEMENS AG
  • EP4645162A1 patent drawingFigure 1
  • EP4645162A1 patent drawingFigure 2a)~2f)
  • EP4645162A1 patent drawing

AI summary

A computer-implemented procedure for inspecting or operating a first product (TD1), and the following steps are performed: a) transmitting a provided global model (GM) of the first product (TD1) or its operation from the server (S) to the first client (C1) and second client (C2), b) capturing a first data set (DS1) of the first product (TD1) or its operation with first product parameters of the first product (TD1) as respective first subsets (C1.1-C1.3), and generating and training respective first local models (LM1.1-LM1.3) with the respective first subsets (C1.1-C1.3), c) capturing a respective second data set (DS2) of the respective second product (TD2) or its operation with respective second product parameters of the respective second product (TD2) as respective second subsets (C2.1-C2.3), and generating and training respective second local models (LM2.1-LM2.3) with the respective second Subsets (C2.1-C2.3), d) Performing a similarity analysis between the models of the first and the second local models (LM1.1-LM1.3, LM2.1-LM2.3), and assigning these similar models to a common model group (G1-G3), e) Training a global detail model (M1) based on the model of the common model group (G1-G3) and transferring it to the first client (C1), f) Applying the detail model (M1) to the first product (D1) for inspection or operation by the first client (C1).