The purpose of this research was to evaluate the accuracy of two computational intelligence methods in the prediction of hardness and density of 3Y-PSZ/316L composites. For this reason, 3Y-PSZ/316L composites with different compositions were produced using spark plasma sintering (SPS) process. Hardness and density of the composites, as the mechanical and physical properties, were determined experimentally. To predict the values of the properties in the composites two computational intelligence methods including Support Vector Regression (SVR) and Adaptive Neuro-Fuzzy Inference System (ANFIS) models were used for the first time. Moreover, the performance and precision of the models in the prediction of the properties were compared together. The ANFIS was developed and validated by randomly limited experimental data separated into two sections, training and testing. Configuration of the ANFIS was determined by Hybrid learning method. To have a comparison between the obtained results and the experimental values, some statistical parameters such as mean relative error (MRE %), coefficient of determination (R2) and root mean squared error (RMSE) were used. The accuracy of both models was demonstrated by the difference between the experimental data and the predicted values with mean relative error less than 1.487%. It showed that both models are powerful tools for prediction of relative density and hardness of 3Y-PSZ/316L composites. However, the lower RMSE and MRE% for ANFIS showed that it offers much superior performance compared with the SVR model.

Keywords