A critical review of artificial intelligence in mineral concentration. (November 2022)

• Record Type: Journal Article
• Authors: Gomez-Flores, Allan; Ilyas, Sadia; Heyes, Graeme W.; Kim, Hyunjung
• Is Part Of: Minerals engineering. Volume 189(2022)
• Publication Date: 2022-11
• Subjects: Artificial intelligence -- Mineral concentration -- Gravity separation -- Density separation -- Magnetic separation -- Sensor–based sorting (SBS); Mines and mineral resources -- Periodicals; Ressources minérales -- Périodiques; Mines and mineral resources; Periodicals; Electronic journals; 622
• DOI: 10.1016/j.mineng.2022.107884
• Languages: English
• ISSNs: 0892-6875

Abstract

Graphical abstract: Highlights: Artificial intelligence applications in density, gravity, magnetic, and sensor-based separations were reviewed. Historical evolution of artificial intelligence was provided for the applications. Supervised, semisupervised, and unsupervised models were assessed for the applications. Kind and wicked learning environments were assessed for the applications. Abstract: Although various articles have reviewed the application of artificial intelligence (AI) in froth flotation (summarized in this article), other unit operations for mineral concentration in mineral processing have not been reviewed. Thus, this article reviews AI application in various unit operations for mineral concentration. Because unit operations for mineral concentration deal with yields not necessarily linearly correlated with input variables, subsequent yield prediction using AI can add value to their control. The current applications of AI have neglected fundamental variables (e.g., particle agglomeration, particle magnetic susceptibility, particle wettability, particle surface charge, and particle Hamaker constant) as inputs for prediction. Instrumentation and industrial simplicity have hindered the consideration of those variables because validation is required. There are kind learning (repeated patterns and high accuracy measurements) and wicked learning (continuously novel patterns and noise in measurements) environments, which are suitable and challenging for machine learning, respectively. Kind learning environments were largely used for the applications of AI. Furthermore, flow can be captured by AI (e.g., neural networks) to attempt to control drag and mixing using synthetic jet type actuators in equipment (shaking tables, fluidized beds, or vessels). Thus, future applications of AI should consider these points.