Intelligent Engineering Systems through Artificial Neural Networks, Volume 16
58 A Three-Stage Structural Damage Detection Approach Using Artificial Neural Networks
Download citation file:
- Ris (Zotero)
- Reference Manager
As worldwide infrastructure has been aging quickly, long-term structural health monitoring (SHM) has been attracting significant attention as a potential tool to prevent catastrophic structure failures and improve bridge management. Strain based damage detection is highly attractive for long term on-line SHM because strain is easy to measure and is well understood by bridge engineers. This paper presents a novel strain based structural damage detection method using a new damage indicator, called the event-based Cumulative Live Load Strain Energy Density (CLLSED). This indicator is used to overcome the difficulties associated with the fact that the damage influence zone is typically very small. CLLSED enhances the sensitivity of strain measurements through mathematical integration, and its effectiveness in detecting small damages is shown through in-depth analysis and simulation.
In this work, an artificial neural network is used as the mathematical engine powering the mapping between the measurable structural response (CLLSED in this case) and the indeterminate features of structural damage (damage severity and damage location). Feedforward back-propagation supervised ANNs were trained to predict structural damage severity and location in three stages. During the first stage, an ANN was used to predict structural live load condition, which is represented as truck type and truck weight. Inputs to this ANN were the normalized CLLSED values that were derived from the strain data collected by multiple strain sensors. The goal of the second stage is to detect the occurrence of damage and to predict the associated damage severity. The truck type and truck weight obtained from the first stage along with the normalized CLLSED were used as the network input vector. During the final stage, the damage severity, truck type, truck weight, and the normalized CLLSED were combined as the input vector for damage location identification. Synthetic ANN analysis results are used to demonstrate the effectiveness of the proposed diagnosis strategy in this paper.