Multiphase Flow Monitoring with Sparse Instrumentation: A Physics-Based Approach

Field instrumentation is expensive, maintenance-intensive, and inevitably incomplete. In a typical midstream gathering network, operators are working with 30–50% of measurement points that are either absent, failed, or producing unreliable data. The standard response—wait for the next maintenance window, flag the node as N/A, or estimate using a simple average—leaves significant operational blind spots.

Physics-based state estimation offers an alternative: infer the unmeasured variables from the observed ones using the governing equations that constrain the system.

How State Estimation Works

A multiphase flow network can be described as a graph: nodes represent junction points, separators, wellheads, and compressors; edges represent pipe segments with associated flow equations.

At every node, conservation of mass applies: what flows in must equal what flows out (accounting for any accumulation or withdrawal). Along every edge, the pressure drop equation constrains the relationship between flowrate, fluid properties, pipe geometry, and the pressures at each end.

These constraints form a system of equations. When some variables are known (measured) and others are unknown (unmeasured), we can solve for the unknown values—provided the system is sufficiently constrained. This is fundamentally the same approach used in electrical network analysis and structural mechanics.

The Challenge of Multiphase Flow

What makes oil and gas networks more complex than a water distribution system is the presence of multiple phases: gas, liquid hydrocarbon, and water flowing simultaneously through the same pipes.

Multiphase flow introduces additional equations: slip between phases, phase holdup calculations, pressure gradient contributions from each phase's hydrostatic and friction effects. The fluid properties—density, viscosity, GOR, water cut—all affect the flow equations and must be accounted for.

The physics model must be rich enough to capture these effects without being so computationally expensive that it can't run in real-time. This is where reduced-order modeling becomes essential.

Building the Network State Estimator

The key steps in building a physics-based state estimator for a gathering network are:

1. Network characterization. Map the topology: which nodes connect to which, what are the pipe geometries, what are the approximate fluid properties for each segment. This data is typically available from engineering drawings and production history.

2. Physics model selection. Choose the appropriate flow correlation for each segment (Hagedorn-Brown, Beggs-Brill, or a simplified slip model depending on the pipe geometry and flow regime). The goal is the minimum physics needed, not the most complex available.

3. Measurement integration. Ingest available sensor data—flowmeters, pressure transmitters, temperature sensors—and account for measurement uncertainty. Not all sensors are equally reliable.

4. State estimation solve. Run the physics-constrained optimization: find the network state (all node pressures, all segment flowrates) that is most consistent with both the governing equations and the available measurements. This is typically a weighted least squares problem.

5. Uncertainty quantification. Propagate measurement uncertainty through the physics model to produce confidence intervals on estimated values. An estimated flowrate of 1.2 Mstb/d ± 0.15 Mstb/d is more operationally useful than a point estimate.

Real-Time Operation

Once built, the state estimator runs continuously—updating every scan cycle as new sensor data arrives. The output is a complete network state: every node pressure, every segment flowrate, every phase split at every junction.

When the estimated state is inconsistent with the physics model beyond a defined tolerance, the system flags an anomaly. This anomaly detection is inherently physics-based: the system isn't comparing today's reading to yesterday's, it's checking whether today's readings are consistent with the laws governing the network. This approach detects novel failure modes that statistical anomaly detection would miss.

What Operators See

Rather than a SCADA screen with N/A for 45% of the tags, operators see a complete network picture. Unmeasured nodes are clearly indicated as estimated, with confidence intervals. Anomalies are highlighted with physics-based explanations of which segments are implicated.

The result is operational clarity: a complete, consistent, physically-valid picture of what is happening in the network at any given moment—regardless of how many instruments are working.

---

WellBeyond.ai has built multiphase flow state estimators for gathering networks across upstream and midstream operations. Contact us to discuss your network.