Structural Soundness Assessment With Explainable Decision Trees
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining the degree of soundness of structures like bridges and tunnels lack transparency, making it difficult for inspection engineers to understand the grounds for automatic determination results, thereby complicating the verification of their accuracy.
Innovation Solution
An information processing apparatus that determines the degree of soundness by using defective-state information and specs-and-conditions information, and outputs grounds for the determination, employing a decision tree-based machine learning model to facilitate understanding and verification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automatic determination of degree of soundness is performed using machine learning methods, then productivity is improved by reducing inspection engineer burden, but reliability deteriorates due to lack of transparency in determination grounds
Solution Approach 1:
The patent introduces an explanation generation unit that acts as an intermediary between the machine learning determination system and the inspection engineer. This unit generates human-understandable explanations (such as decision trees or rule-based justifications) that bridge the gap between automated determination and human verification, thereby maintaining both productivity gains and determination credibility
2Measurement precision
If complex machine learning models are used for automatic determination, then measurement precision is improved, but device complexity increases making verification difficult
Solution Approach 1:
The patent creates a simplified copy or representation of the complex machine learning model's decision-making process. Instead of presenting the actual complex model, it generates an explanatory model (such as a decision tree or set of rules) that mirrors the determination logic in a human-comprehensible format, allowing verification without exposing the underlying complexity
3Productivity
If automatic determination is implemented without explanation output, then productivity is improved, but loss of information occurs regarding determination grounds
Solution Approach 1:
The patent segments the determination output into two distinct parts: the determination result itself and the explanation of grounds. This segmentation allows the system to efficiently produce rapid determinations while separately generating comprehensive explanatory information, ensuring neither productivity nor information completeness is compromised
Data Source
AI summary
Based on a defective state of a structure, a degree of soundness that indicates how much the structure is sound is determined. Information about a defective state of the structure that provides grounds for this determination is outputted.