Big Benefits
Can physics-informed AI improve the way industrial companies control their operations? Geminus is determined to make that happen and is showing promise as it moves beyond pilots to scaling. In the real world of production, we try to make good decisions and automate control based on a massive volume of data from many sources in different formats. Making the right decisions in near real-time can tremendously impact revenues and costs. Geminus is working to achieve this by applying its physics-based AI to automate optimization for industrial processes.
Geminus’ early successes include:
- 40% reduction of energy usage in the water distribution network
- Eliminated flaring in natural gas networks by adjusting network settings to account for overpressures
- Increasing oil production by 10% by optimizing submersible pump control across a network of wells (resulting in tens of millions of dollars in value across six wells in a single day)
Physics-Informed AI
What Geminus is doing differently is fusing physics-based models with process data sets. They use specialized deep learning and engineering-specific generative AI techniques. The resulting AI models deliver high-precision inferencing and quantified uncertainty bounds, which are critical for control decisions.
Simulation is a proven way to improve operations, but it typically isn’t fast or robust enough to control operations in real time. Insights derived from process data can be challenging to get and lack an understanding of physical behavior. Combining the value of simulation with process data and AI can create a faster cycle time for predictive analytics. Shifting optimization to real-time operations increases the value by accelerating the path to profitability and optimization. For example, it can detect how to lower energy usage, reduce emissions, and improve process efficiency.
Geminus says their approach has many advantages, including developing and deploying high-precision AI models faster and more accurately. Another benefit is training models with far less data than is typically needed. As a result, Geminus’ approach offers the ability to model extremely large systems (think enterprise level) with minimal compute resources. These cases might be prohibitively expensive using traditional AI techniques because of long training cycles and compute times. Geminus has productized its approaches in a software platform for building, deploying, and maintaining models, so when equipment or process specifics change, a production company can update the model in a few hours.
Geminus’ technical leaders are Chief Scientist Karthik Durasamy, who specializes in the emerging field of scientific foundational models, and Chief AI Scientist Alex Gorodetsky, an expert in computational autonomy. By combining academic breakthroughs with business operational expertise, Geminus aims to accelerate autonomous industrial operations in industries such as energy, chemicals, space, and electricity distribution and storage.
Full, Secure AI Platform
Building out the foundations for this complex blend of simulation and AI has been a significant undertaking. Geminus offers a full stack solution, available natively on the cloud, edge, or on-premise. They have made navigating the industrial sector’s unique security and data sovereignty challenges a significant focus. Geminus has built technology to train models securely on a combination of the customer’s data, sensor readings, and simulation models. Additionally, Geminus’ approach ensures the privacy of client data. Data is never used to pre-train models shared outside a customer’s environment.
Physics-Based Digital Twin
The starting point for most implementations is a physics-based digital twin. The platform offers a path to model predictive control and eventually to autonomous self-optimizing and self-healing systems. Fully autonomous systems may be a way off for most companies; most of Geminus’ current applications have humans in the loop. Yet, once the system comes online, it can automate controls to make timely process corrections to optimize production or correct deviations without as much need for human intervention.
Productize and Scale
AI use cases are often easy to identify but typically difficult to deploy and scale beyond an initial pilot. Geminus has had success with several specific applications of its technology. Now, they are at a stage of packaging these into applications for specific industry use cases. The early offerings include:
- Energy Network Optimizer for piping networks of energy suppliers in water, natural gas, and oil industries. These companies typically have good simulation and process data to feed into the Geminus AI model.
- Flow assurance is another area of oil & gas where Geminus has an application.
By mid-2025, Geminus expects to have applications covering the entire oil and gas industry spectrum, from reservoir issues to surface production and downstream refining. They have also built a systems framework to connect all those AI models. This will enable enterprise-level optimization of an operator’s most important assets.
Strong Partner in SLB
Naturally, selling to large companies in the energy industry appears challenging for a small company like Geminus. Oil and gas and utilities companies have been working on AI for years, many with limited success at high costs. As a result, they tend not to trust claims of significant or fast benefits from AI projects. Fortunately, Geminus has a strong reseller partner in SLB (formerly Schlumberger), one of the most trusted companies in the energy-related industries.
Most oil and gas companies already have contracts with Schlumberger, so trying this new approach to AI has gained some traction. Together with Schlumberger, Geminus has developed case studies and white papers explaining their successes and the benefits their joint pilot projects delivered.
Ongoing Insights
Geminus will seek similarly strong and credible partners in other industry segments, such as chemicals, materials, and utilities. This business model could enable the company to grow revenue faster than the rate it adds employees. The company’s goal is to focus its spending on R&D more than commercial go-to-market.
Thank you, Joe Vacca, Wendy Buck, and Greg Fallon, for getting us up to speed. We look forward to regular updates on Geminus’ progress in the market!