Investigating the possibility of predicting iron recovery in iron ore processing plants based on feed grade using artificial intelligence

Message:
Article Type:
Research/Original Article (دارای رتبه معتبر)
Abstract:
Objective

The aim of processing iron ore in processing plants is to achieve a product with an appropriate grade and maximum iron recovery. The amount of iron recovery in processing systems depends on several parameters, and its determination through weighing and laboratory tests is time-consuming and costly. With the development of the use of artificial intelligence for predicting and optimizing the performance of industrial systems, it seems that this technology could address many of the issues faced in mineral processing industries, including iron ore processing plants. Therefore, the objective of this research is to assess the feasibility of using artificial intelligence to predict iron recovery based on iron (Fe) and iron oxide (FeO) grades as the first step towards developing the application of this technology in the mining industry.

Materials and methods

 For this study, daily data on the Fe and FeO grades in the feed as well as iron recovery from the Central Iron Ore Concentrate Plant, which includes two production lines (Choghart and Sechahun), were collected. Iron recovery modeling was performed using two neural network models: MLP (Multilayer Perceptron Neural Network) and CFNN (Cascade Forward Neural Network). In this modeling, the Fe and FeO grades of the feed were treated as the model inputs, while iron recovery was considered the output.

Results

The results showed that both models performed relatively similarly, but CFNN exhibited better statistical parameters. The R² value for the CFNN model was obtained as 0.831 for the Choghart production line and 0.837 for the Sechahun line, while the RMSE for these models was calculated as 1.655 and 1.823, respectively. The analysis indicated that the CFNN model could confidently predict iron recovery with a relative error of less than 5% at a 95% confidence level for both production lines.

Conclusions

Using Fe and FeO grades alone as inputs for the models cannot lead to a comprehensive model that can replace conventional calculations. Therefore, the influence of other effective parameters will be thoroughly identified in this study. Additionally, sensitivity analysis revealed a direct relationship between iron recovery and both input parameters, with the Fe grade having a greater impact on iron recovery. The results of this study show that using artificial intelligence to predict iron recovery is very promising. By increasing model accuracy through the addition of data and input parameters, it is possible to develop models that can reduce the costs and time required for grade assessment in the plant.

Language:
Persian
Published:
Journal of separation science and engineering, Volume:16 Issue: 2, 2024
Pages:
92 to 107
https://www.magiran.com/p2836239