Mineral Processing Data
The primary sources of data in mineral processing plants are collected from Process Control Systems, environmental monitoring systems, and SCADA systems. Advanced sensors and smart devices are present in most of the plant equipment, such as crushers, mills, conveyor belts, flotation cells, and critical components like pumps and pipelines. Most of the data generated are a measurement and a timestamp together, such as energy consumption (kWh) and throughput (t/h).
Mineral processing plants are an exceptional environment for the application of advanced analytics technologies and methods. A wide range of applications is discussed in the literature. Not surprisingly, the quality of the data used for training and testing the models is a critical element. Process mechanics and dynamics are very complex, and it is vastly challenging to take reliable, accurate, or direct measurements of specific process variables [40]. In the perfect world, training data should cover all operating regimes, dynamic plant behavior, have a high resolution, be correctly labeled, and be enough size for machine learning techniques. Mineral processing operations typically produce a massive amount of data. However, often it comes at low resolution (e.g., ore grade measurements, which demand laboratory analysis), with different sampling intervals, containing missing values and noise, and is unlabeled.
According to Ref. [41], a supervisory control platform should be able to perform validation and reconciliation of process data, identify operation and instrumentation issues, and coordinate local control loops under a holistic strategy. Most of these essential capabilities should be defined in the plant design stages, but the major operational parameters usually are only assigned once the process facilities are running. Moreover, a shortage of technical skills is often identified in all managerial and operational levels regarding the comprehension of the modeling and operational aspect of nonlinear and complex processes in dynamic operating conditions [41,42].
The flotation process is considered as one of the most complex and data- sensitive steps in the mineral processing field. Multiple parameters are measured by existing instrumentation such as ore composition, flow rates, and some ore specific properties (e.g., density, pH, pulp levels, and particle size). However, some essential properties such as liberation degree, surface chemistry, bubble size distribution, bubble loading remain challenging to measure and infer [42,43,44].
A set of artificial intelligence techniques have been successfully applied in the creation of smart operational systems, providing a better alternative with a higher tolerance for imprecision, uncertainty, and partial truth. Some frequently used examples are fuzzy logic, artificial neural networks, GAs, support vector machines, decision trees, and hybrids of these methods [45].
By applying some data quality concepts discussed in this chapter, mining organizations can significantly improve their efficiency in all activities from pit to port and support effective, fact-based decision-making.
