Machine learning approach to model and predict the densification and mechanical properties of Inconel 718 prepared by spark plasma sintering

This study employs a combined multi-layer perceptron-random forest algorithm (MLP-RF) model to predict the densification degree and tensile strength of Inconel 718 processed via spark plasma sintering (SPS). Key parameters (sintering temperature, pressure, heating rate, dwelling time, and electrical modes) were analyzed. A 5-10-10-2 neuron architecture optimized via mean square error (MSE) and average error (AE) evaluations achieved high predictive accuracy, validated by experimental data. Performance metrics, including root mean square error (RMSE) and mean absolute percentage error (MAPE), confirmed model robustness, with a correlation coefficient (R) of 94.557 %. Finally, the sintering experiment was conducted in an environment with a vacuum level of 10⁻⁴ Pa. Experimental verification highlighted the influence of electrical parameters on densification and mechanical properties. This machine learning framework enables efficient optimization of SPS process parameters for enhanced material performance.