Neurosymbolic Gas Turbine Component Design Under Constraints

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing design processes for gas turbine engine components, particularly cooling structures, are limited by rule-based systems that fail to adequately account for performance criteria, and data-driven approaches like convolutional neural networks lack assurance of constraint compliance.

Innovation Solution

An automated design method utilizing a neural function-learning based module to input desired performance parameters, a design realization module to generate physically realizable designs, and an evaluation module to assess performance, iteratively optimizing the design through a combination of rule-based and neural function-learning modules.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If rule-based systems are used for design, then constraint compliance is ensured, but performance optimization is insufficient

Engineering Contradiction:
Improveconstraint complianceVSAvoidperformance optimization
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent combines rule-based systems with data-driven neural networks into a hybrid neurosymbolic framework. The rule-based module ensures constraint compliance while the neural network module optimizes performance, allowing both functions to operate simultaneously rather than choosing one over the other.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces a hybrid neurosymbolic module as an intermediary between the rule-based constraint checker and the data-driven performance optimizer. This intermediary coordinates the outputs of both modules, integrating constraint satisfaction with performance optimization in a unified design process.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If data-driven approaches like convolutional neural networks are used, then performance optimization is improved, but constraint compliance assurance is lost

Engineering Contradiction:
Improveperformance optimizationVSAvoidconstraint compliance
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent merges data-driven neural network approaches with rule-based constraint checking systems. The neural network provides performance optimization while the rule-based module simultaneously ensures constraint compliance, creating a hybrid system that delivers both benefits.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent implements feedback mechanisms where the rule-based module continuously validates the outputs of the neural network module. When constraint violations are detected, the system provides feedback to adjust the neural network's design proposals, ensuring ongoing compliance while maintaining performance optimization.

Inventive Principle:
Principle #23Feedback

3Manufacturing precision

If automated design exploration evaluates many possible designs, then design quality is improved, but computational time increases

Engineering Contradiction:
Improvedesign qualityVSAvoidcomputational time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by using the rule-based module to pre-filter design options before they are evaluated by the neural network. This preliminary constraint checking eliminates infeasible designs early in the process, reducing the number of designs that require computationally intensive evaluation while maintaining design quality.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent segments the design evaluation process into two distinct stages: a rule-based constraint satisfaction stage and a neural network performance optimization stage. This segmentation allows each module to focus on its strength, improving overall efficiency while maintaining high design quality through comprehensive evaluation.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4582996A1Automated design exploration through neurosymbolic reasoning
Publication Date: 2025.07.09 RTX CORP
  • EP4582996A1 patent drawingFigure 1A
  • EP4582996A1 patent drawingFigure 1B~1C
  • EP4582996A1 patent drawingFigure 2A

AI summary

A method of designing a component (120) includes the steps of 1) inputting operator desired performance (114) of a final component (120) into a neural function-learning based module (115), and utilizing the neural function-learning based module (115) to reach desired structural features for the component (120), 2) generating a physically realizable design in a design realization module (112; 212) based on the structural features and 3) evaluating the physically realizable design in an evaluation module (117; 217; 156), and utilizing evaluation module output to determine performance results for the physically realizable design. A system (138; 238) is also disclosed.