Feature-level interpretability in transfer learning-based chiller fault diagnosis

doi:10.1016/j.buildenv.2025.113527

Feature-level interpretability in transfer learning-based chiller fault diagnosis

2025-08-08

0

PRE

AI

DOI：10.1016/j.buildenv.2025.113527

Original

Original request for help

Save

Abstract

En 中文

• Proposes feature-level interpretability framework for cross-condition FDD in chillers. • Validates the robustness of LRP, IG, SHAP via 10 randomized trials. • Enhances interpretation clarity by increasing data volumes in chiller diagnostics. • Achieves 25% higher accuracy than baseline model through transfer learning. • Verifies that transfer learning preserves discriminative features across domains.

Keywords:

feature-level interpretability

cross-condition fault detection and diagnosis

LRP

IG

SHAP

Feature-level interpretability in transfer learning-based chiller fault diagnosis

Abstract

Journal

Organization