Comparative Between Artificial Neural Networks and Multiple Linear Regression for the Estimation of Mechanical Properties in Cast Aluminum

Abstract

Metallic materials are composed of elements with defined chemical composition, and their intrinsic atomic arrangement confers them a distinctive crystalline structure. Thus, it is relevant to study of metallic materials, specifically cast aluminum alloys, whose physical and mechanical properties depend inherently on their chemical composition. Regarding the importance of mechanical properties, such as hardness, elastic modulus and ultimate tensile strength in optimizing industrial performance, it becomes essential to employ robust methods for their estimation. This study examines the computational estimation of mechanical properties from the chemical composition of various cast aluminum alloys. Two estimation modeling approaches were employed: Artificial Neural Networks (ANNs) and Multiple Linear Regression (MLR). Model performance was evaluated using three statistical metrics: Mean Absolute Error (MAE), which measures the average magnitude of errors; Root Mean Square Error (RMSE), which emphasizes larger error; and Mean Absolute Percentage Error (MAPE), which evaluates the percentage error relative to observed values. The results revealed that the ANN model exhibited superior estimation accuracy across all metrics when compared to the MLR approach. Specifically, the ANN model achieved lower values of MAE and RMSE, indicating more precise estimations and a significantly reduced MAPE, demonstrating improved reliability in estimating mechanical properties. These finding underscore the potential of ANNs as a more effective tool for estimating the mechanical performance of cast aluminum alloys based on their chemical composition. Additionally, the estimation capacity of both models was externally validated using experimental data reported in the literature, enhancing the reliability of the findings.