An Intelligent Machine Learning Framework for Infrastructure Fault Detection via Multi-Source Sensor Fusion
Contributors
Sindhusaranyabalraj
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
Heterogeneous sensing arrays present in critical civil and industrial infrastructures produce copious data, requiring complex processing. While data fusion improves the representation of the SHM data state, automated fault detection under noisy operating conditions is non-trivial. The current practice of single-sensor and threshold based approaches fail to distinguish between operational noise and degradation.
This paper proposes an end-to-end Machine Learning infrastructure fault detection approach that uses multi-source feature level fusion. The framework implements a pipeline that includes hierarchical feature acquisition, selection (Mutual Information, Recursive Feature Elimination) and classification (Normal/Warning/Critical) using an ensemble of trees and deep learning algorithms.
A comparative architecture analysis is conducted to evaluate the performance of two ensemble tree techniques (Random Forest, XGBoost) and two deep learning algorithms (CNN, LSTM) for condition monitoring. Operating on high-dimensional fused vectors, the framework achieves good diagnostic accuracy while maintaining low latency for operationalization at the edge.