How AI-powered solutions can help optimize smelters

Many of the commodities that affect our everyday lives are processed by smelters and fumer-furnace operations. These include products made from copper, nickel, and zinc, such as electrical wiring, kitchen appliances, and some types of batteries. Smelters are also used to produce precious metals, such as gold or silver, and animal feed, and to provide heat for dryers and roasters, which must be operated in concert for efficient production.

Despite the importance of smelters, operations are becoming increasingly challenging. First, assets are aging. Most smelters globally have been in operation for at least 20 years, resulting in higher maintenance requirements. In addition, many budgets don’t support building new ones. Second, feed quality is declining. Over the past decade, concentrate grades have declined while the presence of undesirable elements has increased. Third, experienced operators are nearing retirement, and companies are concerned about attracting and training the next generation of talent. Fourth, environmental, social, and governmental pressures on smelters are increasing. Communities are increasingly aware of the risks posed by industrial operations and subsequently hold companies to higher standards of responsibility. Finally, smelters are often squeezed at the middle of the value chain, as evidenced by declining treatment and refining charges over the past several years.

The convergence of improved collection and storage of historical data, advanced analytics, and domain-specific knowledge, combined with lean operations and change-management efforts, has created opportunities that were not possible even one or two years ago. And while recent innovations in advanced analytics and artificial intelligence (AI) have mostly been applied to mining sites and refineries, they can also benefit operations in smelters and furnaces, as decreased fuel consumption and heat-rate variability significantly impact profitability.

Yet smelter performance is frequently compromised, partly due to the age of the installations and partly because of feed variations. This article describes how AI-powered solutions can be deployed to improve long-term operational efficiency, providing real-time, dynamic adaptation to changing conditions and delivering improved metals production. Our estimates show that end-to-end smelter optimization can help companies achieve $5 billion to $11 billion globally in annual productivity gains.

Smelter operations: Sources of opportunity

There are more than 200 smelter-based metal refineries around the world. The typical operation consists of five primary activities: procurement and supply chain, raw-materials preparation, extractive metallurgy, metals refining, and sales (Exhibit 1).

We strive to provide individuals with disabilities equal access to our website. If you would like information about this content we will be happy to work with you. Please email us at: McKinsey_Website_Accessibility@mckinsey.com

A majority of the challenges—and opportunities—within the typical smelting value chain can be found in three areas.

• Raw-materials preparation. Extracting, combining, and blending raw materials is a crucial part of the value chain. Common challenges include poor visibility into feed-blend economics, high levels of equipment downtime, subpar control processes, and inefficient fuel consumption. However, new technologies can help bridge the value gap by aggregating various feed and assay data into a common data platform, optimizing feed composition and recycling, and refreshing closed-loop control logic to minimize energy use and ensure consistently high feed rates in dryers.
• Extractive metallurgy. Significant value is often left on the table during the actual smelting process. Reasons for this include a lack of standardized decision making across crews and shifts, poor real-time optimization of operations, and high process variability. Yet a combination of advanced analytics–based methodology and pyrometallurgical expertise can help make operating set point recommendations to reduce variation and improve metal recovery. There is also potential to analyze and refresh the control logic in advanced process controls (APCs) that increase the energy efficiency of slag-fuming furnaces.
• Metals refining. Smelting sometimes employs an electrolytic process, during which pure metals are formed by applying an electric current to impure metals. Otherwise known as “electrorefining,” this process can result in inconsistent recoveries, inefficient energy usage, and high equipment downtime. However, new technologies can help reduce unscheduled downtime by identifying top failure modes and the effects of equipment and building predictive models to increase high-fidelity alerts.

By deploying reliable, AI-powered optimization solutions, as well as improving control-loop logic with APCs, smelter and furnace operators can effectively plug these major sources of leakage in their value chain. However, doing so requires taking the right initial steps.

Getting started

The first step of an end-to-end smelter-optimization plan is conducting an initial opportunity and readiness assessment in which plant experts—alongside IT or operational-technology (OT) experts—evaluate data and sensor readiness, assess ways of decreasing bottlenecks, and evaluate control-loop design and utilization. This can help identify areas for improvement as well as target drivers to inform a complete top-down estimation of the value at stake.

Thus, smelter optimization requires a comprehensive view of the optimization stack (Exhibit 2). Initiatives that are narrowly focused on a single aspect of the optimization process may fail to deliver the full expected value.

We strive to provide individuals with disabilities equal access to our website. If you would like information about this content we will be happy to work with you. Please email us at: McKinsey_Website_Accessibility@mckinsey.com

1. Advisory models: Essentially optimized set points, advisory models enable the deployment of fit-for-purpose, cutting-edge technologies and state-of-the art tools, such as automated storage and retrieval systems. Examples include developing optimal feed-blend compositions to operate closer to metallurgical limits and inputting specific operating set points that consistently maximize metal recoveries.

2. Supervisory controls, such as APCs, and base-layer controls (BLCs): Process controls provide target set points for the underlying real-time control system to achieve the desired performance. Several assets in the value chain, such as dryers, roasters, grinding circuits, and secondary-slag-fuming furnaces are fitted and controlled through APCs. Significant value can be unlocked by optimizing and refreshing the control logic.

3. Sensors, instrumentation, and equipment: Model and optimization outputs are only as good as the measurements they are based on. Thus, specific protocols can be developed for cleaner data, and sensors can be calibrated to maintain the vitality of models and APCs. This begins with an end-to-end IT/OT review and assessment that includes sensor health as well as rules and advanced analytics–based methods to reduce unscheduled downtime.

In addition, successful organizations complement the stack with effective change management that builds foundational advanced-analytics capabilities while emphasizing lean operations and agile methods of working.

The following case examples illustrate how end-to-end smelter optimization can be implemented across the entire organization, from advisory models to operational strategy.

Case example 1: North America–based metals refinery

A lead and zinc refinery that processes complex, globally sourced feeds as well as a variety of recycled products employed an end-to-end optimization approach. In doing so, the refinery followed a typical process flow with traditional assets, such as feed plants, dryers, smelters, slag-fuming furnaces, and metal-purification units. The initial diagnostic phase identified value leakage in three main areas: lead recoveries, slag-fuming furnace operations, and the overall composition of recycled materials in the blend compositions.

Lead-recovery optimization. There was high variability in smelter recoveries, resulting in 4 to 6 percent residual lead in slag, likely the result of not fully understanding incoming feed and the impact of fuel changes on recoveries. Predictive models were applied to modulate the charge’s fueling state, taking into account feed assays and measuring process parameters to improve lead recovery. In parallel, predictive models were also built to estimate the impact on secondary process parameters, such as nitrogen oxide emissions and maximum furnace temperature, which must be controlled within the operating range.

Next, input tags with high feature importance ¹ 1. Scores assigned to input features are based on their usefulness. were selected from more than 7,000 data tags captured in the process historian. ² 2. Process historians, also referred to as “time-series databases,” enable storage and retrieval of measured values from sensors and equipment. The company identified process constraints and secondary variables that were affected by fueling and operating changes, resulting in the application of boundary conditions on allowed solution space. It also built a predictive model to search for optimal operations-fueling set points and performed rigorous testing and iterations over multiple weeks by fine-tuning the constraints to gain confidence and acceptance of the modeling and optimization framework (Exhibit 3). As a result, the company demonstrated a sustainable reduction of lead in slag, which is projected to deliver gains of $4 million to $6 million per year.

We strive to provide individuals with disabilities equal access to our website. If you would like information about this content we will be happy to work with you. Please email us at: McKinsey_Website_Accessibility@mckinsey.com

Feed-blend optimization and inventory control. Lack of on-time, accurate data for feed analyses, out-of-date metallurgical constraints, and limited visibility into operational or performance metrics (such as margin) led to suboptimal feed forecasts and charges that drove significant value leakage. Using a constrained optimization methodology, the company created a real-time data platform as well as a forecast-optimization model that considered inventory, metallurgy, margin, and process parameters. The model compared multiple forecast scenarios and selected the optimal forecast for a charge recommendation. Adherence to the forecast was measured as a key performance indicator, and the site delivered an annualized impact of approximately $6.3 million based on increased recoverable metals (Exhibit 4).

We strive to provide individuals with disabilities equal access to our website. If you would like information about this content we will be happy to work with you. Please email us at: McKinsey_Website_Accessibility@mckinsey.com

Optimal slag-fuming furnace operation. High-temperature slag from the lead smelter is evacuated periodically into a slag-fuming furnace for secondary recovery of high-value elements, such as zinc, germanium, and indium. Poor temperature control due to conservative APC logic ultimately resulted in suboptimal fumer performance. The company addressed this issue by conducting a control-loop diagnostic that tested flow controllers for levels of coal feed, air, and oxygen. Then they developed predictive models for slag bath temperature, oxygen-carbon ratio, and steam flow. The main objective was a rapid drop in slag bath temperature, which was achieved by using model parameter set points. The result of retuning the slag-fuming furnace’s APC was an annualized impact of $4 million, based on increased secondary recoveries.

Case example 2: Asia-based flash-furnace operation

A flash-furnace operation in Asia was already a high-performing asset but wanted to apply advanced analytics to help them achieve their next level of performance. The company had well-established lean principles, strong metallurgical practices, and high overall metals recovery. They also had well-defined “recipes” for processing different grades of incoming concentrate and, despite only having a modest degree of sensoring, realized meaningful improvements from the installation of an expert system on the main furnace two to three years prior.

That said, the company struggled with imperfect data and a lack of overall visibility into the performance of various steps in its operation, and it was increasingly struggling to preserve (as well as enhance) the knowledge experienced operators had built up over years of running the furnace. Advanced analytics gave this smelting operation the power to dramatically change the asset’s performance and economics.

The first step to implementing an advanced analytics–based optimization plan was mapping the causes of lost production across the value chain. The company identified a number of opportunities to improve operations and prioritized two areas to get started: improving furnace performance beyond what had been achieved with the expert system and predicting and optimizing the time needed to achieve target chemistry in the converter.

Improving furnace performance. The company developed two analytical solutions to improve furnace performance. First, they built a soft sensor to help operators understand the metal content of the slag at any point in time. Historically, an assay taken every two hours would tell operators how the furnace was performing. However, operators wouldn’t actually find out until well after the fact—oftentimes too late to react. The soft sensor would allow operators to understand how the furnace performs in real time, which would further enable them to tune it accordingly. Second, an advisory model based on AI was built to help operators determine the right set points for incoming feed chemistry and operating conditions. This model complemented the existing expert system and worked by recommending targets for critical plant variables (some of which were already under expert system control, while others were not). The insights from the model also helped focus the team on targeted equipment upgrades that would allow them to better control the operating parameters that proved critical to performance. Together, the insights from the AI model, on top of the expert system, would be able to unlock an additional 0.5 to 1.0 percent of metal production.

Predicting and optimizing converter time. The converter is a batch process that provides material for the flash furnace. For this particular company, it was also an operations bottleneck. In their optimization efforts, operators determined the importance of making sure the plant achieved the target chemistry in the most efficient way before passing the material forward to the flash furnace. The primary challenge was that they couldn’t predict how long it would take to achieve the target chemistry—there could be as much as 50 percent variation in batch time. Furthermore, it wasn’t clear which set points provided the most efficient way to get there, as feed conditions frequently changed. Because optimization was such a complex process, significant “tribal knowledge” had developed among operators over the course of their careers, yet not all operators agreed on the factors. Thus, the application of advanced analytics helped in two ways: First, it allowed operators to predict the batch duration to achieve target chemistry, better manage the cycle, prevent overprocessing, and plan for tapping. Second, an AI-based advisory system was able to recommend set points to further optimize the process, improve chemistry, and more quickly achieve target chemistry.

These case examples highlight challenges that are all too common for smelting operations all over the world: assets are aging, feedstock quality is deteriorating, experienced operators are retiring, and budgets are getting squeezed at the middle of the value chain. For these reasons, among others, operators must do everything they can to drive productivity. Harnessing the power of data, new process-control technologies, and advanced analytics are quick and low-capital ways to improve end-to-end smelter performance.