Permittivity Measurement in Multi-Phase Heterogeneous Concrete Using Evidential Regression Deep Network and...

Area of Science:

• Materials Science
• Electromagnetics
• Artificial Intelligence

Background:

• Permittivity measurements of concrete using high-frequency electromagnetic waves (HF-EMWs) are crucial for non-destructive testing.
• Existing methods face challenges with aleatory and epistemic uncertainties due to material heterogeneity and limited wave propagation knowledge.
• These uncertainties restrict the precision of current non-destructive testing techniques.

Purpose of the Study:

• To propose an evidential regression deep network for accurate permittivity measurements of concrete.
• To incorporate uncertainty quantification into the measurement process.
• To enhance the precision of non-destructive testing for concrete materials.

Main Methods:

• A finite-difference time-domain (FDTD) model simulates HF-EMW propagation in heterogeneous concrete, capturing aleatory uncertainties.
• A U-net-based model denoises HF-EMW signals, with differences quantifying measurement noise-induced uncertainty.
• A Dempster-Shafer theory (DST)-based evidential regression network computes permittivity, quantifying uncertainties using Gaussian random fuzzy numbers (GRFNs).

Main Results:

• The proposed method achieved a mean square error of 7.50% in permittivity measurements across four concrete types.
• A permittivity uncertainty value of 74.70% was quantified.
• The method successfully quantified measurement uncertainty using a GRFN-based belief measurement interval.

Conclusions:

• The evidential regression deep network effectively measures concrete permittivity with quantified uncertainty.
• This approach significantly improves the precision of non-destructive testing for concrete.
• The GRFN-based uncertainty quantification provides reliable belief measurement intervals.