Document Type
Extended Abstract
Abstract
This study presents a field-validated deep learning model for real-time, in-situ concrete strength estimation using piezoelectric sensors and electro-mechanical impedance (EMI) signal processing. Addressing the limitations of prior research—namely, narrow datasets and poor field applicability—we constructed a diverse database comprising over 775 data points across 28 mix designs and 107 sensor deployments. A novel 1D Convolutional Neural Network (1DCNN) with a baseline correction mechanism was developed to analyze time-series EMI signals. Our model achieved a mean absolute error of 2.68 MPa and an R² of 0.96 in testing, outperforming conventional methods. Transfer learning techniques further enhanced adaptability to new mix designs and sensor types. Most notably, the model was successfully validated using field data from highway pavement projects, demonstrating its robustness and real-world applicability. This work establishes a new benchmark for AI-driven concrete strength monitoring, bridging lab innovation and practical deployment.
Keywords
Piezoelectric sensor, Deep learning, Electromechanical impedance, Concrete strength, Transfer learning.
DOI
10.5703/1288284318020
Real-Time Concrete Strength Estimation Using Electro-Mechanical Impedance and 1D CNN
This study presents a field-validated deep learning model for real-time, in-situ concrete strength estimation using piezoelectric sensors and electro-mechanical impedance (EMI) signal processing. Addressing the limitations of prior research—namely, narrow datasets and poor field applicability—we constructed a diverse database comprising over 775 data points across 28 mix designs and 107 sensor deployments. A novel 1D Convolutional Neural Network (1DCNN) with a baseline correction mechanism was developed to analyze time-series EMI signals. Our model achieved a mean absolute error of 2.68 MPa and an R² of 0.96 in testing, outperforming conventional methods. Transfer learning techniques further enhanced adaptability to new mix designs and sensor types. Most notably, the model was successfully validated using field data from highway pavement projects, demonstrating its robustness and real-world applicability. This work establishes a new benchmark for AI-driven concrete strength monitoring, bridging lab innovation and practical deployment.