Implementing Machine learning algorithms in the prediction of aluminum quality during recycling

Achieving high-quality aluminum casting hinges significantly on melt preparation, influencing the occurrence of diverse casting defects. While traditional methods face challenges in assessing melt quality online due to time and energy constraints, recent innovations present a practical evaluation system. This advancement holds promise for improving final product quality and operational efficiency by minimizing suboptimal molten material and subsequent scrap. This research explores the integration of machine learning (ML) methodologies to revolutionize melt quality assessment. By establishing correlations between rapidly obtainable thermodynamic data and critical melt parameters, predictive models will enable estimations of chemical composition and solidification processes. Leveraging fundamental melt composition data and cooling regimes, these models aim to predict diverse solidified part characteristics, aiding process engineers. This project aims to elucidate crucial connections between cooling curves and solidified part characteristics using predictive models, thus advancing melt quality control. Integration of various ML techniques such as simple linear regression, Multiple Linear Regression, Lasso regression, ElasticNet regression, Ridge Regression, Polynomial, SVR, Randomforest, and Decision tree represents a significant step forward, promising rapid, real-time estimation of melt quality parameters. For all models cross validation was performed to ensure about the reported results. Results obtained from the evaluation of the mentioned models highlighted the higher accuracy of the Randomforest model followed by the ElasticNet regression model as the best candidates to be used for the prediction of magnesium content during aluminum alloys preparation.

Achieving high-quality aluminum casting hinges significantly on melt preparation, influencing the occurrence of diverse casting defects. While traditional methods face challenges in assessing melt quality online due to time and energy constraints, recent innovations present a practical evaluation system. This advancement holds promise for improving final product quality and operational efficiency by minimizing suboptimal molten material and subsequent scrap. This research explores the integration of machine learning (ML) methodologies to revolutionize melt quality assessment. By establishing correlations between rapidly obtainable thermodynamic data and critical melt parameters, predictive models will enable estimations of chemical composition and solidification processes. Leveraging fundamental melt composition data and cooling regimes, these models aim to predict diverse solidified part characteristics, aiding process engineers. This project aims to elucidate crucial connections between cooling curves and solidified part characteristics using predictive models, thus advancing melt quality control. Integration of various ML techniques such as simple linear regression, Multiple Linear Regression, Lasso regression, ElasticNet regression, Ridge Regression, Polynomial, SVR, Randomforest, and Decision tree represents a significant step forward, promising rapid, real-time estimation of melt quality parameters. For all models cross validation was performed to ensure about the reported results. Results obtained from evaluation of the mentioned models highlighted the higher accuracy of the Randomforest model followed by the ElasticNet regression model as the best candidates to be used for the prediction of magnesium content during aluminum alloys preparation.