| description abstract | Abstract. Achieving the desired hardness is critical for optimal tribological performance of functionally graded materials (FGMs). The numerical and finite element-based methods for estimating hardness are computationally intensive; also, the fabrication of FGMs and the process of their hardness evaluation are often difficult and time-consuming. The present study introduces data-driven predictive models using machine learning (ML) techniques to accurately estimate the maximum hardness specifically for centrifugally cast ceramic-reinforced aluminum FGMs. The relationship between various input parameters and FGMs' hardness is established through the artificial neural network (ANN), support vector regressor (SVR), and decision tree regressor (DTR) models. The dataset encompasses an inclusive set of 15 input features, such as Al alloy type, including ceramic reinforcement, particle content, particle size, and other parameters during the processing conditions. Initial correlation between the input parameters was examined through the Pearson correlation coefficient. Then the GridSearchCV technique was used for identifying the optimal hyperparameters for different ML models. The predicted hardness for all ML models closely matched the experimental results in testing, achieving accuracy R2 > 90%. Additionally, 8 wt% TiB2–Al7075 FGM was fabricated and examined for microstructure, hardness, and wear behavior. Moreover, the developed ML models were validated using the hardness of fabricated FGM and data from the literature. In the validation process, the decision tree regressor 3 achieved the highest accuracy R2 of 98.14% and mean squared error 0.03602, mean average error 0.16692, followed by decision tree regressor 2 and artificial neural network 1 model. Moreover, in a comparative study, the developed FGM's wear performance was explored, and its outer surface demonstrated 73.1% and 45.8% higher wear resistance at 5.04 m/s speed and 10N load, respectively, compared to stir-cast composites tested under similar test conditions. Further, the morphological analysis of worn surfaces exhibited smooth, mild delamination at 2.52 m/s speed, while they were worsened by high material removal, with pits observed under 40 N load. This underscores the hardness and its effect on improving wear performance in materials. The ML models offered predictive capability for the estimation of hardness before rigorous experimentation. | |
