Featured
Boosting productivity in heavy industries using AI
OptimusAI from QuantumBlack, AI by McKinsey
Following our series of articles exploring QuantumBlack’s product solutions for AI transformation, we now embark on a new series focused on the application of AI within different industry sectors. The series will take an in-depth look at a host of industries where our AI suite of products is solving global business challenges.
Over the years, QuantumBlack, AI by McKinsey has helped organizations reinvent themselves to achieve accelerated, sustainable, and inclusive growth with AI. In QuantumBlack Labs, the R&D innovation hub within QuantumBlack, we use our colleagues’ collective experience to develop suites of tools and assets to facilitate their projects.
This article describes OptimusAI from QuantumBlack, which targets efficiency in processing plants, using AI to learn plant behavior from operational data, and make recommendations to steer the plant to optimum productivity potential.
Processing plant complexity challenges
Processing industries face unrelenting pressure to boost productivity. Plant efficiency is crucial in the face of increasing costs for raw materials and energy, rising sustainability targets, and fierce competition. Across the energy and materials sectors, companies strive to optimize their production assets by unlocking extra capacity, improving energy efficiency, and enhancing product quality.
Traditionally, management of these complex processing plants relied heavily on the operators’ experience to adapt to varying conditions. Over the last 5–10 years, industry leaders have started supplementing this human experience with AI experience: using the available operational data, machine learning models can be trained to capture plant behavior and assist operators in their decision-making. By analogy, a driver who uses a real-time navigation app can find the best route and avoid traffic; the app does not replace the driver but helps them make better decisions.
How can AI optimize production plants in real time?
AI can be trained on historical operations data to build a ‘digital twin’ of the production process: a digital replica that simulates aspects of production. It’s constructed by translating operations logic and historical data into a simulation engine, which is then validated against past production data to ensure high accuracy.
This approach enables:
- Set point optimization: Digital twin simulations help find the control set points which optimize the production plant’s output and efficiency.
- Integrated value chain planning: Digital twins help in planning operations, e.g., finding the best feedstock blend to maximize materials processed, improve end product quality, and minimize waste.
- Condition monitoring and preventative maintenance: Digital twins help identify and analyze failure points within the production process before they occur, thus preventing production failures and minimizing downtime.
- Design optimization: Digital twins can validate specific design-based decisions, using AI/ML to run multiple simulations of potential designs.
Introducing QuantumBlack OptimusAI
OptimusAI acts as an AI copilot for plant operators and engineers, guiding on unlocking plant productivity gains and freeing up time for value-adding activities.
OptimusAI is a complex AI system that combines:
- Learning from multimodal data inputs, like operational data, laboratory data, maintenance logs, and cameras images.
- Capturing first principles, including physics, chemistry, material and energy balances, and kinetics.
- Integration with existing process control logics.
- Optimization within operational limits using metaheuristic algorithms that take into account physical, chemical, safety boundaries of the system. This is aligned to engineering and process control principles, and incorporates the control room operators’ needs.
The portfolio of OptimusAI functionalities includes:
- Set point optimization, described in depth below.
- Value chain optimization for planning and scheduling.
- Control Room Advisor, a customizable user interface for operator control rooms, including a GPT copilot to interact in natural language with the underlying data and AI solutions.
Set point optimization
Developing a set point advisory solution follows a four-step process:
- Ingest and explore: The platform takes data from multiple unit operations in the plant. To overcome critical data gaps, soft sensors can be used to predict important variables. The system uses the data to distill insights into plant performance.
- Model: Plant behavior is modeled through a smart combination of ‘first principles’ engineering models (e.g., mass/energy balances, kinetics) and machine learning algorithms (e.g., multi-layer perceptrons, gradient-boosted decision trees). Various explainable AI methods (e.g., SHAP, PDP) are available to investigate main drivers of plant performance.
- Optimize: Proprietary, metaheuristic optimization algorithms, designed specifically for heavy industry problems, find the optimal operating parameters for a given shift.
- Visualize: The operator can accept or comment on a recommendation and monitor the resulting impact. An integrated GPT copilot allows operators to ask clarifying questions on the recommendations, visualize data, or search through manuals and procedures.
Ensuring sustained impact
AI only goes so far if the impact isn’t sustained for the long-term. Hence, it is critical to drive user adoption and monitor the quality of the solution:
- During the AI transformation, operator comments are logged and trigger QuantumBlack’s data science team to review the recommendation engine.
- The integrated GPT copilot enables operators to ask clarifying questions and provides all relevant knowledge and data at the operator’s fingertips.
- Data and model quality are continuously monitored, triggering automatic alerts or model retraining. QuantumBlack’s LiveOps team can help companies longer-term in maintaining and upgrading their deployed AI solutions.
No two plants are the same, which is why scaling machine learning solutions across a production network of similar, yet intrinsically different sites, is a common industry challenge.
OptimusAI has solved this issue through its modular architecture and configurable data pipelines; sites can reuse pre-built code components and focus efforts on configuring them to their specific requirements. This has enabled us to add value and unlock the potential of processing plants across all production sectors.
Monitoring and maintenance by QuantumBlack’s LiveOps team ensures long-term impact from deployed models and AI-driven solutions.
Impact of OptimusAI
OptimusAI delivers more impact, faster, with less risk. The platform has an unquestionable track record, with more than 100 effective use cases across mining, oil and gas, power generation, chemicals and many other processing industries.
These projects have consistently demonstrated the following benefits:
- Increased capacity: OptimusAI typically unlocks between 10–15 percent productivity gains through data-driven operational adjustments.
- Improved energy efficiency: The platform helps to address decarbonization goals, reducing energy consumption by 5–10%.
- Enhanced product quality: OptimusAI helps to reduce quality losses, by minimizing off-spec grades or product defects.
Case study: OptimusAI to reduce carbon emissions
As we described in a recent case study, Vistra is the largest competitive power producer in the US and operates power plants in 12 states with a capacity of more than 39,000 megawatts of electricity, enough to power nearly 20m homes.
Vistra is committed to reducing emissions by 60 percent by 2030 and achieving net-zero emissions by 2050. Vistra identified the ‘heat rate’ as a key focus. To support this journey, Vistra partnered with QuantumBlack to use OptimusAI to enhance efficiency in energy production via a machine-learning-based Heat Rate Optimizer for power plant operations.
The initial deployment improved the Martin Lake Power Plant’s efficiency by 2 percent, saving $4.5M annually and cutting 340,000 tons of carbon emissions — the equivalent of taking 66,000 cars off the road. Vistra then rolled out the technology over 67 power generation units across 26 plants, saving more than $60 million within a year, helping the firm abate approximately 1.6m tons of carbon a year. That equates to 10 percent of its remaining 2030 carbon reduction commitment.
Case study: Transforming mining through AI
Our case study about a leading global mining company described how it was facing a dilemma. To maintain growth in its Americas copper operations, Freeport-McMoRan needed to increase production, but with mature mines and aging technology, associated costs would be significant. The business believed the answer lay in improving operations with artificial intelligence, but it lacked the technology skills and capabilities to design and deploy AI at scale.
The QuantumBlack OptimusAI team created a digital roadmap using AI and agile work methods to increase productivity at every step of Freeport’s processes, while collaborating at a single aging mine in Bagdad, Arizona. This soon resulted in a 10 percent increase in production outputs. The additional copper production the company is projected to unlock over five years, equating to one new processing facility, without needing to spend up to $2 billion to do so.
Understanding the AI prerequisites
For processing plant leaders considering introducing QuantumBlack OptimusAI, the necessary barriers to entry are minimal:
- Robust data collection systems: The algorithms use plant data to evaluate improvement potential and drivers of performance.
- Open-minded teams: The impact of our work is only as strong as the team’s drive to embrace change.
Summary
QuantumBlack OptimusAI brings the power of AI into processing plant control rooms by combining cutting-edge data science with operations and IT expertise for end-to-end optimization and long-lasting productivity gains. Working alongside plant operators, OptimusAI offers real-time recommendations and warns of potential challenges ahead of time. Across a variety of industry sectors, OptimusAI has delivered significant, scaled, and sustained productivity improvements using the power of AI.
To learn more about what QuantumBlack OptimusAI can do for you, please email optimusai@mckinsey.com
Thanks to all who contributed to this article: Milan Korbel, Sean Buckley, Stijn De Bruyne, Jun Yoon, Zhe Sha, Sripriya Verma, Manishankar Panda, Xavier Morin, Nimit Patel, Marc Solomon, Rory Walsh, Jo Stichbury, Joanna Sych & Antidote Communications.
