In high-temperature metallurgical processes, such as converter operations in platinum smelting, operators often lack real-time visibility into critical process variables like slag basicity, furnace temperature, and specific oxygen content. These parameters are either measured infrequently via lab analysis or estimated manually, limiting timely intervention and making process optimisation difficult.
To address this, Opti-Num initiated a multi-year programme in partnership with a major platinum producer. The goal was to develop and deploy soft sensors—virtual models that estimate unmeasured but vital variables using real-time plant data—to enhance operational control, reduce instability, and support digital transformation in converter operations.
Key challenges included deploying models in harsh, high-variability environments, maintaining model adaptability as plant conditions evolved, and ensuring outputs were interpretable and trusted by operations teams.
Opti-Num developed a suite of soft sensor models using a hybrid modelling framework—combining first-principles process knowledge with advanced machine learning techniques. The models were designed to run live, continuously estimating values for:
Data Integration and Model Design
Time-aligned, high-resolution operational data was used as model input, supported by lab-based chemistry for calibration. Bayesian regression techniques enabled the incorporation of uncertainty and physical constraints into model design.
Deployment and Validation
The models were validated against lab benchmarks and operator observations, then deployed into the client’s operational environment. Model outputs were visualised and monitored in real-time to support stability-focused decision-making.
Enablement and Sustainability
Beyond model delivery, Opti-Num trained internal engineers to maintain, update, and extend the models—ensuring long-term sustainability of the solution and reducing dependency on external support.
The soft sensors were fully embedded into converter operations, delivering measurable technical and operational impact:
The models now serve as foundational components for advanced process control and digital optimisation initiatives in metallurgical operations.
The project progressed through four structured phases:
Each phase built toward increased accuracy, impact, and independence, ensuring long-term value from the modelling effort.
Opti-Num brought together deep technical and domain knowledge across:
.png)
.png)
.png)
