AI-based non-destructive testing framework for predicting dielectric properties of wood without direct microwave measurement
Nondestructive Testing and Evaluation, 2025 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Publication Date: 2025
- Doi Number: 10.1080/10589759.2025.2548343
- Journal Name: Nondestructive Testing and Evaluation
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, Applied Science & Technology Source, Aquatic Science & Fisheries Abstracts (ASFA), Communication Abstracts, Compendex, INSPEC, Metadex, Civil Engineering Abstracts
- Keywords: complex dielectric constant, deep learning, loss tangent, Material characterisation, natural wood, non-destructive testing
- Akdeniz University Affiliated: Yes
Abstract
Electrical properties of materials are critical to signal transmission and device performance. These properties are described by complex permittivity (ε*), comprising real (ε′) and imaginary (ε′) components, with loss tangent (ε′′/ε′) indicating energy dissipation as heat. Conventional methods for determining these parameters require precise sample dimensions, specialised instrumentation and stable environmentalconditions, making them costly and time-consuming. This study proposes a Deep Neural Network (DNN) framework as a non-destructive testing (NDT) method to predict dielectric properties of natural hardwood and softwood commonly used in Turkey”s wood industry. Data were obtained with waveguides, coaxial cables and a vector network analyser (VNA) using the microwave scattering technique. DNN predicts dielectric parameters based on frequency, density and wood type. Model performance was first evaluated using train, validation and test data, achieving mean absolute error (MAE), mean squared error (MSE) and root mean squared error (RMSE) values of 0.72, 1.82 and 1.08 respectively on prediction ε*. Validation on 2nd round of samples which contained new densities from different samples yielded errors of 1.76, 4.71 and 1.88. Results show strong alignment between predicted and measured values, confirming the DNN”s effectiveness for rapid, cost-efficient material characterization and qualty control in the wood-based products industry.