Data Fusion of Multiple Sensing Techniques

Abstract

This paper investigates the use of data fusion from multiple sensing methods for structural damage diagnosis. In order to monitor structures effectively, multiple sensing techniques are preferable compared to a single sensing technique. However, one of the challenges in information fusion is heterogeneity of the data obtained from different types of sensing techniques. The heterogeneity is caused by different techniques producing data in different formats, such as time series signals, images, and point measurements. This paper investigates a Bayesian network approach to handle this challenge. Structural health monitoring of concrete structures is used to illustrate the data fusion methodology. The objective is to diagnose the presence of alkali-silica reaction (ASR) in a concrete slab. Three types of techniques are used: infrared thermography (IR), nonlinear impact resonance acoustic spectroscopy (NIRAS), and vibro-acoustic modulation (VAM). Vibration signals are generated from NIRAS and VAM, while IR generates thermal images over time. NIRAS and VAM can qualitatively identify the existence of ASR, whereas IR can locate and quantify the area of ASR damage. The Bayesian network is used to fuse the information from multiple techniques and diagnose the ASR (presence, location and extent), and also to quantify the uncertainty in the diagnosis result