This study proposes an innovative configuration for thickness measurement based on X-ray transmission, intending to improve the precision of measuring thin aluminum sheets. In this configuration, an activation sample containing Zr, Sb, and Ba elements is irradiated by 59.54 keV gamma rays emitted from three 241Am radioactive sources with a total activity of 1.78 GBq. Subsequently, the activation sample emits fluorescent X-rays at energy levels of 15.78 keV (Zr-Kα1,2), 17.67 keV (Zr-Kβ1), 26.36 keV (Sb-Kα1,2), 29.73 keV (Sb-Kβ1), 32.2 keV (Ba-Kα1,2), and 36.38 keV (Ba-Kβ1). These X-rays are collimated into a narrow beam, which then penetrates through an absorbing sample, and is ultimately recorded by a Si(Li) detector. Two different approaches are investigated to determine the thickness of absorbing samples including Calibration Curve Fitting (CCF) and Artificial Neural Network (ANN). The CCF approach requires constructing linear calibration curves for establishing the relationship between lnR (R is the ratio of the peak areas in measurements with and without absorbing sample) and the thickness of the absorbing sample at each X-ray energy level. The sample thickness is then determined by calculating a weighted average of the measured thicknesses associated with all analyzed energy levels. This approach requires in-depth understanding of radiation physics and proficiency in X-ray spectrum analysis. Meanwhile, the ANN approach uses raw spectra obtained by the Si(Li) detector to predict the thickness of aluminum sheets, facilitating analysis without requiring human intervention. The reliability of these approaches is evaluated through experimental measurements on aluminum sheets with thicknesses ranging from 0.064 cm to 1.074 cm. The results indicate that using X-rays with many different energies leads to superior accuracy in thickness measurements compared to using X-rays with a single energy. Besides, both the CCF and ANN approaches yield relative deviations of less than 3% between the predicted and reference thicknesses. It is important to emphasize that the ANN approach represents a promising solution for automated analysis without the intervention of experts.
Read full abstract