Shared Learning from Procemin Geomet 2022
In the mining industry, thickener process control and stability have always been important aspects of plant operations, particularly for metals. With the increasing pressure towards more efficient and sustainable processes, and with increasing water scarcity in many regions of the world there is now an unprecedented drive towards innovation in the field of real-time process control.
Thickening is a solid-liquid separation process where gravity is used as a preponderant factor for material separation, based on differences in density. The process of sedimentation and consolidation can be explained by phenomenological* theory, and numerical simulations based on it showing a good correlation with laboratory results.
However, the industrial setting involves significantly higher levels of complexity and uncertainty than laboratory work. Even modelling on longer timescales for design purposes is challenging, while predicting the real-time operational performance of a thickener is more difficult still.
Thickener Modelling Approach
Our success uses a method based on a machine learning (ML) process model combined with an explainable and real-time configurable optimization engine. The process model is a physics-inspired Bayesian* deep neural network that is trained from historical data to predict the performance of the thickener in dependence of the future control settings.
The model is paired with an optimizer engine that uses a reward-based evaluation of possible thickener trajectories to recommend control settings that take into account a holistic view of the thickener performance.
The reward is broken down into partial rewards reflecting different aspects of the thickener operation. These are configurable in real-time and can also be evaluated to explain the reasoning behind the recommendations.
In the paper, you can also find more about the modelling and optimization methodology as well as the outcomes of a field test in a copper mine in Chile. The test was conducted on paste tailings thickeners and the application demonstrated its ability to successfully improve and stabilize the operation of these dewatering circuits.
The Procemin-Geomet conference was very interesting and emphasized the increasingly strong introduction of digital transformation and data management for optimizations associated with mineral processing and Geometallurgy. Digital transformation is more and more established as a transversal tool in the mining production chain.
It’s Time for Digitalization
Stimulating digitization to improve business performance, sustainability and reduce risk is part of Intellisense.io‘s core values. Through the use of Scientific AI and ML, we have built digital applications to facilitate decision-making processes in one of the harshest environments. The new machine learning models developed by Intellisense.io have the capacity to learn complex non-linear dynamics, making precise predictions possible. By doing so, our apps can drive process improvements, identify and remove bottlenecks, increase asset availability, reliability and utilization even for a single operating unit (like thickeners, stockpiles, grinding, flotation, heap leaching, pumping, etc) or work in tandem to optimize overall mine-to-market efficiency.
*The phenomenological theory of sedimentation describes a flocculated suspension as a mixture of the solid and the fluid as two superimposed continuous media.
*Bayesian deep learning refers to probabilistic deep learning that is based on the Bayes theorem. Bayesian methods combine prior “expert” information and the likelihood of data to produce posterior distributions, which can represent different uncertainties in the model.
*Scientific AI is the fusion of physics models with machine learning techniques. It was developed by Intellisense.io to further increase net metal production through the use of our AI solutions.