Mapping Algorithm Adjusts Real-World Data for Simulation-Trained Neural Networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Neural network models trained using simulators for floating production platforms and other systems require frequent retraining due to discrepancies between simulated and real-world data, leading to increased costs and downtime.

Innovation Solution

A method that compares real-world data to simulation data and adjusts it using a threshold metric to maintain similarity, allowing the neural network model to operate with adjusted data that corresponds to the simulation data, thereby preserving decision logic and reducing the need for continuous retraining.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If the neural network model is trained using simulation data, then the learning process is greatly accelerated and the model is available from the beginning of operations, but the simulation accuracy must be verified in parallel with retraining the model, adding costs and time

Engineering Contradiction:
Improvelearning process speedVSAvoidmodel verification and retraining time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent creates a mapping between real-world data and simulation data, allowing the neural network to operate with simulated data that copies the structure and relationships of real data without requiring continuous verification against actual measurements. This copying approach maintains model accuracy while eliminating repeated retraining cycles.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent introduces a data mapping mechanism as an intermediary layer between real-world measurements and the neural network model. This mapping transforms real data into simulation-compatible format, allowing the model to operate with simulated inputs while indirectly representing real-world conditions, thus avoiding direct continuous verification.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If the neural network model is retrained frequently to maintain accuracy with real-world conditions, then the prediction accuracy is improved, but the costs and downtime increase

Engineering Contradiction:
Improveprediction accuracyVSAvoiddowntime and costs
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent uses simulation data as a copy of real-world data structures, allowing the neural network to maintain prediction accuracy by operating with simulated data that preserves the relationships and patterns of actual operational data without requiring frequent retraining on new real-world measurements.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent transforms real-world data parameters into simulation-equivalent parameters through a mapping function. This parameter transformation allows the model to maintain accuracy by working with transformed parameters that retain the essential relationships of the original data without requiring model retraining.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If real-world data is used directly for model operation, then the model reflects actual conditions, but the model requires considerable time to accumulate sufficient training data and is not available during that period

Engineering Contradiction:
Improvemodel reflects actual conditionsVSAvoiddata accumulation time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary action by creating a comprehensive mapping between real-world data and simulation data before the neural network needs to operate. This pre-established mapping allows the model to immediately use simulation data that already encodes the relationships of real-world conditions, eliminating the data accumulation delay.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent creates a copied representation of real-world data relationships through simulation data and mapping functions. This copying allows the neural network to operate immediately with simulated data that preserves the essential patterns and relationships of actual operational data, without waiting for sufficient real data to accumulate.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3802310B1Continuous learning of simulation trained deep neural network model for floating production platforms, vessels and other floating systems.
Publication Date: 2024.09.04 TECHNIP ENERGIES FRANCE SAS
  • EP3802310B1 patent drawingFigure 1
  • EP3802310B1 patent drawingFigure 2
  • EP3802310B1 patent drawingFigure 3

AI summary

The present invention provides a system and method of side-stepping the need to retrain neural network model after initially trained using a simulator by comparing real-world data to data predicted by the simulator for the same inputs, and developing a mapping correlation that adjusts real world data toward the simulation data. Thus, the decision logic developed in the simulation-trained model is preserved and continues to operate in an altered reality. A threshold metric of similarity can be initially provided into the mapping algorithm, which automatically adjusts real world data to adjusted data corresponding to the simulation data for operating the neural network model when the metric of similarity between the real world data and the simulation data exceeds the threshold metric. Updated learning can continue as desired, working in the background as conditions are monitored.