An evaluation of machine learning and artificial intelligence models for predicting the flotation behavior of fine high-ash coal

Abstract

In this study, five different machine learning (ML) and artificial intelligence (AI) models: random forest (RF), artificial neural networks (ANN), the adaptive neuro-fuzzy inference system (ANFIS), Mamdani fuzzy logic (MFL) and a hybrid neural fuzzy inference system (HyFIS) were employed to predict the flotation behavior of fine high ash coal in the presence of a novel “hybrid” ash depressant consisting of polyacrylamide chains grafted onto aluminium hydroxide nanoparticles: Al(OH)3-PAM (Al-PAM). A total of 51 flotation tests were conducted on coal samples with 38% ash-content and a P80 of approximately 49 μm. Different influencing variables of coal flotation including polymer dosage, pH, polymer conditioning time, sodium metasilicate dosage (commercial dispersant), and the impeller speed were used as inputs for the models. The combustible recovery and ash content of coal reported to the concentrate were used as response variables (outputs). For AI model development, 80% of the total data was used for training phase and 20% was used for testing phase. Coefficient of determination (R2) and root-mean-square error (RMSE) were used as performance indicators of the models. The MFL model showed the best accuracy for the prediction of the combustible recoveries and the froth ash contents for this specific feed. However, in case of any significant change in the characteristics of the feed, these models would have to be re-trained using the data obtained through further physical experimentation and/or process model simulations. Moreover as these models are trained on laboratory scale data, these are only good for the predictions at laboratory scale.