Machine learning approach for revealing the nickel grade and recovery optimization in reduction process of laterite ores

The growing demand for ferronickel has led many researchers to continuously explore ferronickel making processes over the past decade. With the increasing demand for energy-efficient processes, the selective reduction of laterite nickel ore, followed by ferronickel metal separation, has become the main focus. Researchers have studied and published many process variables or features to date. Among the various process variables, achieving high-grade and nickel recovery is the primary goal. This research attempts to take a machine learning approach to find the right model for process optimization. Data sets were collected from published studies that feature nickel ore grade (Ni and Fe content) and reduction process (temperature, additives, reductant, etc.), while prediction models for Ni grade and recovery were formed from four types of regression algorithms. Thus, the obtained models provide comprehensive information and summarize the research results developed over the past decade, highlighting the influence of each feature on the target. Furthermore, the machine learning approach can expedite the process of achieving the target Ni grade and recovery. The model with the random forest regression algorithm was chosen because it can predict Ni grade and recovery well, as evidenced by the R² training values of 0.95 and 0.97 when predicting Ni grade and recovery, respectively.