Prediction of mechanical and structural properties of WC-Co cemented carbides from magnetic data: Comparison between machine learning and conventional linear regression

Based on experimental data and extensive experience, magnetic coercivity and saturation magnetization are traditionally used to estimate the microstructure and quality of cemented carbides, especially in the manufacturing industry. Using an artificial neural network (ANN), structural and mechanical properties of WC-Co elements can be predicted from magnetic data alone. The total field distribution, which is extracted from first-order-reversal-curves, serves as input data for the ANN. A collection of WC-Co pellet samples with a variety of powder compositions and processing parameters have been produced to cover a wide range of characteristic features for ANN training. It is shown that microstructural parameters such as mean WC grain size and mechanical properties such as hardness and fracture toughness can be derived with high accuracy. The prediction capability of the ANN is compared to the prediction capability of conventional linear regression. • Artificial neural networks are capable of predicting the mechanical and structural properties of WC-Co cemented carbides based on magnetic data only. • Comparison between machine learning and conventional linear regression. • The neural network model is trained using magnetic data from first-order reversal curves. • Predictions based on machine learning are superior to those based on linear regression.