Reproduce “Explainable AI for Engineering Design: A Unified Approach of Systems Engineering and Component- Based Deep Learning Demonstrated by Energy- Efficient Building Design” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Philipp Geyer, Manav Mahan Singh, Xia Chen
A Component-Based Machine Learning (CBML) approach unifies deep learning with systems engineering principles to enhance generalization and intrinsic explainability in engineering design, as demonstrated by energy-efficient building design. This method consistently reduced prediction errors, achieving a Mean Absolute Percentage Error (MAPE) of 3.76% on random building shapes compared to 5.88% for monolithic models, and enabled direct interpretation of internal model states as physical engineering quantities.
Key insights
- Aligning the machine learning model structure with the hierarchical decomposition of engineering systems significantly improves generalization to structurally novel designs.
- Embedding interpretability directly into the model architecture through physically meaningful component interfaces provides quantitative, white-box explanations of internal processes.
- Data-driven insights from CBML, such as those derived from sensitivity analysis and local decision trees, consistently cohere with established domain knowledge and physical simulations, fostering trust.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.