Thickness Measurement of Metals Using Neural Network and Radioisotope Measurement-Modelling

The nuclear industry has undergone a significant technological revolution in recent years, with neural networks playing an increasingly important role. These advanced machine learning algorithms are now widely used to analyze nuclear data obtained from various sources, such as calculations, modeling, simulation, and nuclear sensors including gamma ray-based gauges, experimental tests, and non-destructive tests like radiography. Non-destructive measurement methods based on gamma rays passing through materials or processes have gained popularity due to their effectiveness and safety in the nuclear industry. These methods typically use radioactive radioisotope sources like cesium-137 and cobalt-60, along with scintillation detection units, to capture data related to fluid thickness, density, and flow. By applying neural networks, these measurements can be made even more accurate and reliable, ensuring optimal safety and efficiency in the nuclear sector. To investigate this further, a study was conducted to design the geometry of the transmission mode for radioisotope thickness measurement through simulation using the MCNPX Monte Carlo code. The model included three commonly used metals in industries: copper, iron, and aluminum, to determine their thickness. To create the dataset, the thickness of the samples was varied from 0.5 to 50 mm, with a step size of 0.5 mm, and the results of Monte Carlo calculations were recorded as pulse height in the detector. Artificial neural networks were trained based on radial basis functions (RBF) and multilayer perceptron (MLP) techniques using all the results obtained from different energies. After training, the networks were tested to predict the thickness of different metals. The results obtained from the two networks were compared for the output of the model. The study showed that the response of the MLP network was more satisfactory than that of the RBF network in this application. Overall, the study highlights the potential of neural networks in enhancing the accuracy and reliability of non-destructive measurements in the nuclear industry, particularly in predicting the thickness of different materials.