Iridium-۱۹۲ backscatter spectrometry for corrosion detection: A hybrid MLP-based approach

This study presents a hybrid MLP-based approach for corrosion detection using Iridium-۱۹۲ backscatter spectrometry, validated through comprehensive MCNPX simulations. Five distinct corrosion types, sweet (FeCO۳), sour (FeS), microbiologically influenced (FeS, Fe۳O۴), atmospheric (Fe۲O۳), and chloride-induced (FeOCl, FeCl۲), were simulated at seven thickness fractions ranging from ۲۰% to ۸۰% corrosion layer coverage. For each corrosion type and percentage, precise MCNPX material cards were developed incorporating isotopic compositions (e.g., ۲۶۰۵۶ for iron, ۶۰۱۲ for carbon, ۸۰۱۶ for oxygen, ۱۶۰۳۲ for sulfur, and ۱۷۰۳۵ for chlorine) with corresponding densities ranging from ۴.۱۳ to ۷.۳۳ g/cm³. Backscatter gamma spectra were generated using an Iridium-۱۹۲ source, and key spectral features, including Compton edge shifts, peak-to-valley ratios, and total backscatter counts, were extracted as input features for a multilayer perceptron (MLP) neural network. The hybrid framework combines spectral preprocessing with a deep MLP architecture comprising four hidden layers (۱۲۸-۶۴-۳۲-۱۶ neurons) to classify corrosion type and estimate layer thickness. The model was trained and validated on ۳,۵۰۰ simulated spectra, achieving ۹۷.۲% classification accuracy across the five corrosion types and a root mean square error of ۰.۰۹ mm for thickness estimation. For individual corrosion types, the MLP achieved F۱-scores of ۰.۹۸ (sweet), ۰.۹۷ (sour), ۰.۹۶ (microbial), ۰.۹۷ (atmospheric), and ۰.۹۵ (chloride). Compared to conventional peak-ratio methods, the proposed approach improves detection sensitivity by ۲۸% and reduces false positives by ۱۸%. The results demonstrate that MLP-based analysis of MCNPX-simulated Iridium-۱۹۲ backscatter spectra provides a robust, automated tool for discriminating between corrosion mechanisms and quantifying material loss, suitable for field deployment in pipeline inspection and structural health monitoring.

کلیدواژه ها:

Iridium-۱۹۲ ، Backscatter spectrometry ، MLP ، Corrosion detection ، MCNPX simulation ، Neural Network