Maintenance Status Prediction Using Neural Network and MCDA

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing predictive maintenance techniques face challenges in accurately predicting the maintenance status of complex real-world devices, due to limitations in handling complexity and interpretability.

Innovation Solution

A computer-implemented method using a neural network to train a Multi-criteria Decision-Aid (MCDA) sorting model, which takes maintenance-related physical and/or functional data as input to predict the maintenance status of a real-world device.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If data-based techniques are used to predict failures, then the ability to handle complex systems is improved, but the interpretation and validation of predictions become challenging

Engineering Contradiction:
Improveability to handle complex systemsVSAvoidinterpretation and validation of predictions
Core Design Contradiction:
Adaptability or versatilityVSLoss of information

Solution Approach 1:

The patent introduces an intermediary layer between the neural network and the MCDA sorting model. The neural network processes complex maintenance data and generates predictions, which are then fed into the MCDA sorting model that applies interpretable rules and criteria. This intermediary structure allows the system to handle complex systems through the neural network while maintaining interpretability through the rule-based MCDA model, thus resolving the contradiction between handling complexity and maintaining interpretability.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Ease of manufacture

If model-based techniques are used to predict component failure, then simplicity and ease of application are improved, but the accuracy is limited due to assumptions about stochastic processes

Engineering Contradiction:
Improvesimplicity and ease of applicationVSAvoidprediction accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent merges model-based techniques (MCDA sorting model with predefined criteria and rules) with data-based techniques (neural network trained on historical maintenance data). The MCDA model provides simplicity and ease of application through its structured approach, while the neural network enhances accuracy by learning complex patterns from data. This combination resolves the contradiction between simplicity and accuracy by integrating the strengths of both approaches.

Inventive Principle:
Principle #5Merging (Combining)

3Loss of information

If hybrid techniques are used to combine model-based and data-based approaches, then a compromise between accuracy and interpretability is achieved, but the ability to handle complexity and accuracy are still limited

Engineering Contradiction:
ImproveinterpretabilityVSAvoidability to handle complexity
Core Design Contradiction:
Loss of informationVSAdaptability or versatility

Solution Approach 1:

The patent implements a dynamic architecture where the neural network can adapt to complex maintenance scenarios by learning from historical data, while the MCDA sorting model dynamically applies relevant criteria based on the specific maintenance situation. The system can adjust its behavior depending on the input data characteristics, allowing it to handle greater complexity while maintaining interpretability through the dynamic interaction between the data-driven neural network and the rule-based MCDA model.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP4553701A1Predicting a maintenance status of a real-world device
Publication Date: 2025.05.14 DASSAULT SYSTEMES SA
  • EP4553701A1 patent drawingFigure 1~2
  • EP4553701A1 patent drawingFigure 3
  • EP4553701A1 patent drawingFigure 4

AI summary

The disclosure notably relates to a computer-implemented method for predicting a maintenance status of a real-world device. The method comprises providing a dataset. The dataset includes data describing historical real-world maintenance events and properties of devices of a same type as the real-world device. The method further comprises training, based on the dataset, a neural network to predict parameters of a MCDA sorting model. The MCDA sorting model is configured to take as input at least one time measurement of maintenance-related physical and/or functional data of the real-world device and to output a prediction of a maintenance status of the real-world device.