Executive Summary : | Concrete's initial properties are crucial for its durability and strength, and monitoring these properties is essential for curing and hardening. Damage in concrete structures can occur due to aging, environmental factors, and excessive loading. Conventional methods are time-consuming and intrusive, but electromechanical impedance (EMI) testing has shown potential as a non-invasive and efficient method. Deep learning techniques like convolutional neural networks (CNN) can address this challenge by automatically analyzing EMI data and predicting concrete strength characteristics and damage.
This research proposes a novel smart sensing unit (SSU) using a multi-sensing technique for surface-bonded and embedded sensing. The SSU consists of a PZT patch, an adhesive layer, and a steel plate. The multi-sensing technique reduces collected data, but it presents challenges in handling extensive data. A physical model is employed to understand strength evolution and calculate equivalent stiffness using spring and damper elements. To develop accurate predictions of concrete strength, the study proposes a two-part method using deep-learning models with input parameters extracted from preprocessed data. The first part augments conductance signals corresponding to different concrete strengths, while the second part comprises a 2D-CNN-Bidirectional Long Short-Term Memory (BiLSTM) architecture consisting of a CNN layer, a BiLSTM layer, and an output layer. This fusion of neural network architectures can enhance early-age monitoring and damage assessment, leading to better structure maintenance and repair strategies. |