Application of frequency characteristic extraction in increasing the accuracy of X-ray based thickness gauges used for aluminum alloys employing GMDH neural network

• Published in: Applied radiation and isotopes 2024-06, Vol.208, p.111310-111310, Article 111310
• Main Authors: Mayet, Abdulilah Mohammad, Thafasal Ijyas, V P, Raja, M. Ramkumar, Muqeet, Mohammed Abdul, Shukla, Neeraj Kumar
• Format: Article
• Language: English
• Subjects: Alloy; Aluminum layer; Artificial neural networks; Feature extraction; GMDH; Thickness meter; X-ray-based system
• ISSN: 0969-8043; 1872-9800
• DOI: 10.1016/j.apradiso.2024.111310

Radiation-based gauges have been widely utilized in the industry as a dependable, non-destructive method of measuring metal layer thickness. It is only possible to trust the conventional radiation thickness meter when the material's composition is known in advance. Thickness measurement errors are to be anticipated in contexts like rolled metal factories, where the real component of the material could diverge greatly from the stated composition. An X-ray-based device was suggested in this study to measure aluminum sheet thickness and identify the type of its alloys. Transmission and backscattered X-ray energy were recorded using two sodium iodide detectors while a 150 kV X-ray tube in the described detection system was operated. Aluminum layers of varying thicknesses (2–45 mm) and alloys (1050, 3105, 5052, and 6061) were simulated to be placed between the X-ray source and the transmission detector. The development of radiation-based systems used the MCNP code as a very powerful framework to imitate the detecting architecture and the spectra acquired by the detectors. The recorded signals were transferred to the frequency domain using the Fourier transform, and the frequency characteristics were extracted from them. Two GMDH neural networks were trained using these characteristics: one to identify the alloy type and another to determine the aluminum layer's thickness. The classifier network had a 92.2% success rate in identifying the alloy type, while the predictive network had a 1.9% error rate in determining the thickness of the aluminum layer. By extracting important characteristics and using powerful neural networks, this study was able to improve the precision with which aluminum layer thickness was measured and correctly identify the alloy type. The suggested method is used to determine the thickness of aluminum and its alloy sheets and may also be applied to other metals. •An X-ray-based device was suggested to measure aluminum sheet thickness and identify the type of its alloys.•Transmission and backscattered X-ray energy were recorded using two NaI detectors.•The recorded signals were transferred to the frequency domain using the Fourier transform.•Two trained GMDH neural networks determined the alloy type and the aluminum layer's thickness.