Neural network method for determining the oxide composition of a metal by the non-isothermal gas evolution curve

Oxide inclusions in steel can significantly alter its working properties. One of the methods for analyzing the composition of oxide inclusions in metal is Fractional Gas Analysis (FGA). As a result of applying FGA, the curves of dependence of the rate of CO release from a metal sample and of temperature on time are obtained. This paper proposes a neural network algorithm for further interpretation of the obtained dependencies. A fully connected neural network is used, which is trained to predict the rate of CO release from each type of oxide inclusions. It is required that the predicted individual dependencies satisfy the differential equations modeling the CO release process for a separate oxide inclusion. This requirement was built into the loss function using Physics-Informed Neural Networks (PINNs). The proposed algorithm was tested on samples of two metals of different composition. During testing, we demonstrated the effectiveness of practical application of the developed algorithm