TY - JOUR
T1 - The advanced design of bioleaching process for metal recovery
T2 - A machine learning approach
AU - Mokarian, Parastou
AU - Bakhshayeshi, Ivan
AU - Taghikhah, Firouzeh
AU - Boroumand, Yasaman
AU - Erfani, Eila
AU - Razmjou, Amir
N1 - Publisher Copyright:
© 2022 Elsevier B.V.
PY - 2022/6/15
Y1 - 2022/6/15
N2 - Huge masses of waste containing hazardous but often valuable metals are being produced worldwide. Recycling these wastes is necessary because mineral deposits are being depleted rapidly, making mining more expensive, and recycling hazardous metal compounds reduce environmental contamination. Compared to conventional methods of metal recycling, such as pyrometallurgy and hydrometallurgy, bioleaching is simple, economical, environmentally friendly, and has low energy demand. Several parameters that affect metal recovery yield must be determined before bioleaching can be applied. However, finding the optimum conditions for bioleaching processes is complex and time-consuming; the use of machine learning (ML) techniques could avoid the need for repetitive and costly experiments. ML is a subfield of artificial intelligence, with proven accuracy, reliability, and cost-effectiveness in various applications. Evaluation of 40 regression-based ML algorithms determined that random forest regression returned the highest performance; it was used to design an efficient model for bioleaching processes. Experimental data (871 samples) was derived from 206 papers on metal bioleaching, and nine independent variables were chosen for ML analysis, including microorganism type, temperature, pulp density, cell nutrient, initial pH, used method, atomic number, particle size and density, and type of resources. Also, recovery rate as the target value was considered the dependent variable. The frequency distribution of all variables was obtained; the feature importance scores revealed that resources, particle size and density, temperature, and microorganisms were the most influential variables for estimation of recovery rate. The proposed model can predict metal recovery rates at 77% accuracy and save significantly on time and costs of laboratory experiments.
AB - Huge masses of waste containing hazardous but often valuable metals are being produced worldwide. Recycling these wastes is necessary because mineral deposits are being depleted rapidly, making mining more expensive, and recycling hazardous metal compounds reduce environmental contamination. Compared to conventional methods of metal recycling, such as pyrometallurgy and hydrometallurgy, bioleaching is simple, economical, environmentally friendly, and has low energy demand. Several parameters that affect metal recovery yield must be determined before bioleaching can be applied. However, finding the optimum conditions for bioleaching processes is complex and time-consuming; the use of machine learning (ML) techniques could avoid the need for repetitive and costly experiments. ML is a subfield of artificial intelligence, with proven accuracy, reliability, and cost-effectiveness in various applications. Evaluation of 40 regression-based ML algorithms determined that random forest regression returned the highest performance; it was used to design an efficient model for bioleaching processes. Experimental data (871 samples) was derived from 206 papers on metal bioleaching, and nine independent variables were chosen for ML analysis, including microorganism type, temperature, pulp density, cell nutrient, initial pH, used method, atomic number, particle size and density, and type of resources. Also, recovery rate as the target value was considered the dependent variable. The frequency distribution of all variables was obtained; the feature importance scores revealed that resources, particle size and density, temperature, and microorganisms were the most influential variables for estimation of recovery rate. The proposed model can predict metal recovery rates at 77% accuracy and save significantly on time and costs of laboratory experiments.
KW - Bioleaching
KW - Machine learning
KW - Metal recovery
KW - Parameters
KW - Random Forest Regression
UR - http://www.scopus.com/inward/record.url?scp=85127134224&partnerID=8YFLogxK
U2 - 10.1016/j.seppur.2022.120919
DO - 10.1016/j.seppur.2022.120919
M3 - Article
SN - 1383-5866
VL - 291
JO - Separation and Purification Technology
JF - Separation and Purification Technology
M1 - 120919
ER -