The Challenge
In mineral processing, hydrocyclones are used to classify slurry and play a critical role in plant stability and throughput. Monitoring the number of active hydrocyclones in a given cluster is essential for effective control, yet direct sensor data is often unreliable or unavailable due to hardware faults or signal loss.
A leading mining company partnered with Opti-Num to develop a virtual sensor—or "soft sensor"— capable of estimating the number of active hydrocyclones based on standard process variables such as pressure, feed flow rate, and slurry density. The challenge was to model this hidden system state with sufficient accuracy and interpretability, using a hybrid modelling approach that blends engineering principles with data-driven techniques.
The biggest challenge was training the model using plant data that was often inaccurate or inconsistent. Technical uncertainties included variability in the data, sensitivity to measurement errors, and ensuring the derived model aligned with known physical relationships.
The Solution
Opti-Num implemented a hybrid modelling strategy designed to infer the number of active hydrocyclones from indirect measurements. This involved combining physical process understanding with advanced statistical methods to deliver a reliable, interpretable model.
- Model Structure Identification - The model assumed a parabolic relationship between pressure, density, flow, and the number of active cyclones—derived from first principles in mineral processing.
- Bayesian Linear Regression - Applied to historical process data to estimate model coefficients, capturing uncertainty and incorporating prior domain knowledge, leveraging the team’s machine learning and statistical experience.
- Monte Carlo Parameter Tuning - To refine key parameters like system coefficients, Monte Carlo simulations were applied to produce probabilistic estimates that could handle noisy or incomplete datasets.
- Soft Sensor Inference - Once validated for pressure estimation, the model could be inverted to estimate the number of operating hydro-cyclones, presenting a feasible approach forreal-time soft sensing from live plant data.
- Visualisation and Validation - The solution included a suite of visual tools to compare measured and predicted values, identify model drift, and quantify alignment between theoretical and empirical results.
The Outcome
While the hydrocyclone model was not deployed, the project delivered significant value in deepening process understanding and developing a visualisation tool that provides detailed insight into hydrocyclone performance. The validated modelling approach offered engineers a clearer view of process dynamics, even in the absence of reliable sensor data, and highlighted opportunities to improve operational visibility and future control reliability.
- Enhanced Process Understanding – The modelling and visualisation framework provided actionable insight into hydrocyclone behaviour, supporting better diagnostic and operational decisions.
- Operational Insight Without Sensors – Virtual estimates revealed how plant performance could be monitored effectively even when physical sensor data was unreliable.
- Scalable Modelling Framework – The approach can be adapted for other soft-sensing and process analysis challenges within mineral processing.
- Foundation for Future Control Integration – The outcomes lay the groundwork for future development of soft sensors and potential integration with Advanced Process Control strategies.
- Human + Data Synergy – The project demonstrated the value of combining engineering expertise with data-driven modelling to unlock new insights into complex process operations.
.png)
.png)
.png)
