Five decades into flow simulations, three decades into transient flow simulations, and two decades into flow simulations online, the question remains: Is there a solution for predictive and practical transient flow simulations in an operational real-time setting? We believe there is – hybrid technology.
While the history of practical flow simulations and their wider application within the oil and gas industry is about half-a-century old, the foundational concepts date back to Ancient Greece. The seed of what later became fluid mechanics was planted by Archimedes around year 250 BC. While most famous for the ‘Law of buoyancy,’ commonly known as Archimedes’ principle, he also introduced fundamental concepts within fluid mechanics by postulating: “If fluid parts are continuous and uniformly distributed, then that of them which is the least compressed is driven along by that which is more compressed” . In other words, Archimedes postulated pressure-driven fluid flow.
Two millennia later, the mathematical foundation of fluid mechanics was laid down by Leonhard Euler and Joseph-Louis Lagrange. Then in 1822, Claude-Louis Navier expanded on this work and incorporated viscosity, leading up to what we now know as the Navier–Stokes equations. The equations constitute a rigorous description of fluid flow, but the devil is in the details and analytic solutions can only be obtained for simple systems. More practical applications – flow simulations – were still 150 years away, awaiting computers to enter the stage. With increasing access to more and more powerful computers, computational fluid dynamics (CFD) became commonplace in for example aeronautics and automotive industry, but the techniques were not practically transferrable to multiphase flow in pipes on larger scale, not even today.
Focusing on oil and gas applications, there is a number of complicating factors. Pressure and temperature may vary significantly along the length of a pipe and with them fluid properties. Further, there are phase transitions such as evaporation and condensation, dispersions, liquid droplets carried in the gas phase, mechanical and thermal interactions between the fluids and the pipe, and much more. Already, the challenges at hand span across multiple disciplines, e.g., fluid mechanics and thermodynamics. Still, we have not addressed aspects of flow in pipeline networks, time-dependent aspects of the flow, or the fact that the characteristics (flow regimes) change dramatically with changing flow conditions.
A saving grace for fluid flow in pipes is that the fluid motion primarily is in the axial direction of the pipe. Effects in the cross section are of limited interest and can often be treated in an averaged manner, resulting in one-dimensional so-called hydraulic models. While simpler, a hydraulic model is by no means trivial for multiphase flow. Yet again, the devil is in the details as there is a need to create models to address the challenging phenomena mentioned above. This leads to a framework of bespoke models not only in terms of the different phenomena but also in terms of varying flow conditions.
These models disregard any time-dependent variations in the flow, i.e., each simulation estimates the conditions in a pipe under the assumption that no conditions or properties change with time. Despite this simplification, the models offer fundamental insights into multiphase flow and are commonly used in all phases from concept engineering through to operations. With often fast and easy-to-use workflows, steady-state simulations have reached a broad adoption across domains.
In the 1970s, computers enabled the first wave of multiphase flow simulations. The first models introduced belong to a family called steady-state models.
While the oil and gas industry adopted steady-state models in the 70s, development of transient models, i.e., models that incorporate time-dependent changes in the flow, gained interest in a different energy sector, namely nuclear power. Models were developed for multiphase flow of water and steam in nuclear reactors. In the 1980s, work started to adopt these models to oil and gas applications. The added dimension of time increases the complexity by orders of magnitude as compared to steady-state models. As a consequence, transient simulations have not reached an as broad adoption across domains but are mainly limited to smaller groups of expert users. Nonetheless, transient multiphase flow simulations changed the oil and gas industry forever. Up until the late 80s, every offshore development had required its own platform. With transient multiphase flow simulations at hand, it was possible to perform the comprehensive design assessments and risk analyses required to safely move from individual platforms to subsea multiphase flowlines. Following the first successful subsea installations on the Norwegian Continental Shelf in the late 80s, the technology has redefined development concepts and helped push the limits further, and deeper all over the world.
So far, we have considered flow simulations for design of installations and better operational practices. Around the turn of the millennium, the interest grew around the logical next step; to apply existing engineering software in online applications connected to live measurements, to mimic the multiphase flow at real-time. The adoption to online environments is, however, not without challenges. Engineering applications require flow simulations which deliver reasonable results under almost any circumstances. The solutions rely on numerous model components, and which one to apply depends on the physical conditions. In operations, predictability is key. To achieve the best performance possible calibration is required, but the numerous model components and their interdependencies make calibration difficult. On top of that comes the challenge to adopt engineering software to meet the demands on performance and maintainability associated with online deployment.
Over the first two decades of the 21st century, improvements in computer storage and network connectivity allowed operators to collect and store more production data than ever before. This lead flow simulations in a new direction and introduced machine learning and so-called data driven models. With historical information about a producing asset available, computers can create (train) models to predict complex phenomena without detailed knowledge of the underlying physics. No matter how complex the phenomenon, the resulting data-driven model is always fast to execute and easy to deploy. However, data-driven models are no silver bullet but come with challenges of their own. For example, training of data-driven models requires large amounts of high-quality data. Furthermore, since the models are based purely on historical data, they have very limited capability to extrapolate beyond the range of historically known conditions.
Five decades into flow simulations, three decades into transient flow simulations, and two decades into flow simulations online, the question remains: Is there a solution for predictive and practical transient flow simulations in an operational real-time setting? We believe there is – hybrid technology. Hybrid solutions combine first-principle physics with machine learning. First-principle physics is a necessity to address operational conditions that have not occurred before. It is also required to gain a complete understanding of the entire flowing system from inlet to outlet. However, the first-principles flow simulation software must be robust, fast, easy to calibrate, and easy to deploy to meet the demands of real-time operations. Combine this with the ability of machine learning models to make the most out of available data, and you have a solution that can meet your expectations. In other words, it is about performant and autonomous flow simulations that self-adjust to changing conditions, deliver the best possible predictions, and enable easy access to always up-to-date simulation models for consumption to analyze and optimize your operations.
 G.A. Tokaty. A History and Philosophy of Fluid Mechanics. ISBN 0-486-68103-3. 1994.