Presenter Information

Category

Applied

Description

Structural Health Monitoring is a critical tool for analyzing infrastructure resilience. However, measuring dynamic external loads, such as wind, traffic, or seismic activity, remains a major engineering challenge. In order to address this “Inverse Problem” in structural dynamics, this study proposes a deep learning-based approach for structural load identification. To precisely reconstruct unknown force histories from vibration responses, a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architecture is created. Convolutional layers can filter raw data, and LSTM takes the refined features from the CNN and reconstructs the time-history force distributions. Combined, these two methods create an effective model that can handle the demands of structural load identification. While the developed model is validated with a three-story building example, the results demonstrate its capability to accurately predict the dynamic ground truth force. A synthetic training dataset comprising acceleration, velocity, and displacement under random impact scenarios is created by using the Newmark-Beta method. The primary objective is to predict unknown impact forces from the vibrational responses. By combining CNNs for feature extraction and LSTMs for time-history prediction, the proposed model successfully maps known structural responses to unknown ground truth forces, accurately capturing even zero-force scenarios. Various improvements were implemented to scale the model’s prediction capabilities. While the current model has been built and validated for the simple building example, the core architecture is designed to adapt to a broader range of infrastructures. Reconfiguring current algorithms and structural hyperparameters to apply this method to train models on complex structures with higher degrees of freedom could provide a scalable path towards a tool capable of identifying structural irregularities across different engineering systems.

Share

COinS
 
Apr 21st, 11:30 AM Apr 21st, 12:00 PM

Dynamic Structural Load Identification

Applied

Structural Health Monitoring is a critical tool for analyzing infrastructure resilience. However, measuring dynamic external loads, such as wind, traffic, or seismic activity, remains a major engineering challenge. In order to address this “Inverse Problem” in structural dynamics, this study proposes a deep learning-based approach for structural load identification. To precisely reconstruct unknown force histories from vibration responses, a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architecture is created. Convolutional layers can filter raw data, and LSTM takes the refined features from the CNN and reconstructs the time-history force distributions. Combined, these two methods create an effective model that can handle the demands of structural load identification. While the developed model is validated with a three-story building example, the results demonstrate its capability to accurately predict the dynamic ground truth force. A synthetic training dataset comprising acceleration, velocity, and displacement under random impact scenarios is created by using the Newmark-Beta method. The primary objective is to predict unknown impact forces from the vibrational responses. By combining CNNs for feature extraction and LSTMs for time-history prediction, the proposed model successfully maps known structural responses to unknown ground truth forces, accurately capturing even zero-force scenarios. Various improvements were implemented to scale the model’s prediction capabilities. While the current model has been built and validated for the simple building example, the core architecture is designed to adapt to a broader range of infrastructures. Reconfiguring current algorithms and structural hyperparameters to apply this method to train models on complex structures with higher degrees of freedom could provide a scalable path towards a tool capable of identifying structural irregularities across different engineering systems.

 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.