Performance Evaluation of Artificial Neural Network Methods for Moisture Content Determination in Lightweight Foamed Concrete Using a Microwave Sensor

This article investigates the use of artificial neural networks (ANNs) combined with a microwave sensor for the noninvasive determination of moisture content in lightweight foamed concrete (LFC). LFC's dielectric properties undergo significant changes with varying moisture levels, making it suitable for analysis using microwave sensor. Using an open-ended coaxial sensor, the study explores three ANN training algorithms: Levenberg-Marquardt (LM), Bayesian regularization (BR), and scaled conjugate gradient (SCG). Through systematic experimentation, optimal ANN configurations are identified, with neuron counts ranging from 28 to 53 and epochs from 223 to 375. The ANN models achieve high regression (R) values, averaging 0.97 across all datasets. The highest regression values were achieved with BR (0.9736 for the five-input and 0.9735 for epsilon(r)*), followed closely by LM. SCG performed slightly lower, especially in the three-input datasets. These findings highlight the reliability of BR and LM across different input configurations. Sensitivity analysis reveals that different training methods prioritize distinct features: S11 phase for LM, S11 magnitude for BR, and epsilon(')(r) for SCG. Targeting these key parameters for each method can enhance model accuracy and efficiency in practical applications, ensuring more reliable predictions with better resource allocation. Overall, the findings demonstrate that ANNs coupled with microwave sensor offer a robust, real-time solution for nondestructive moisture content monitoring. The research underscores the effectiveness and adaptability of ANNs for moisture content monitoring in LFC.