MODELING BASELINE METAL RECOVERY IN ORDER TO MEASURE THE IMPACT OF MAJOR CONCENTRATOR PROCESS IMPROVEMENTS USING ADVANCED ANALYTICS - SME Annual Meeting

The interactions of mineralogy, chemistry and mechanics in an ore concentrator are highly complex, so it can be particularly hard to measure the impact of a specific plant change. This paper describes a method for modeling a baseline metal recovery so that the value of a major process improvement can be measured. This method only considers variables that cannot be optimized, such as mineralogy or equipment availability. Linear regression, random forests, extreme gradient boosting (XGB), and simple two-layer neural network models of process value were tested as candidate baseline model types to assess the impact on throughput and recovery in a copper concentrator. XGB performed the best in terms of accuracy for both throughput and recovery models. Using an XGB model of nonoptimizable process variables provides an accurate, dynamic baseline that can be compared with continuous production data to determine the value added by major concentrator improvements in real time.