Sensorless Downhole Fluid Submergence Estimation for Progressive Cavity Pumps Using an Extended Kalman Filter

Imagine you are trying to fill a giant, deep bathtub with a hose, but you can't see inside the tub. You only have a gauge on the hose outside that tells you how hard the water is pushing. If the tub is full, the water pushes back hard against the hose. If the tub is empty, the hose has nothing to push against and spins freely. This is the basic puzzle engineers face with oil wells. They need to know how much oil is sitting right above the pump deep underground (called "submergence") to keep the machine running smoothly. If the pump runs dry, it gets hot and breaks, costing millions to fix. Usually, to see the oil level, they have to lower a special microphone down the hole to listen to the sound of the liquid, or install expensive sensors that often break in the harsh, hot, sandy environment.

This paper tackles that problem by asking: "Can we guess the oil level just by listening to the pump's motor from the surface?" The researchers use a clever math trick called an "Extended Kalman Filter." Think of this filter as a super-smart detective that looks at clues the motor gives off—like how much electricity it's using and how much torque (twisting force) it feels—and uses those clues to guess what's happening deep underground. It's like guessing how heavy a backpack someone is wearing just by watching how fast they are walking and how much they are sweating, without ever seeing the backpack. The goal is to replace expensive, breakable sensors with a "virtual sensor" that lives in a computer, saving money and keeping the oil flowing.

The Story of the Invisible Oil Level

Deep underground, oil wells are like giant, vertical pipes stretching miles into the earth. To get the thick, sticky oil (often called "heavy oil") out, companies use a machine called a Progressive Cavity Pump (PCP). Imagine a giant corkscrew spinning inside a rubber tube. As the screw turns, it scoops up oil and pushes it up to the surface. But here's the catch: if the pump spins when there isn't enough oil right above it, the rubber tube starts rubbing against the metal screw with no lubrication. This is called "running dry." It's like trying to run a car engine without oil; the parts get hot, melt, and the machine dies. When this happens, the company has to send a massive rig to the well to fix it, which costs a fortune and stops production for weeks.

For a long time, the only way to know if the oil level was getting too low was to send a technician down to the well with a special acoustic device. This device sends a sound wave down the pipe and listens for the echo to measure the distance to the oil. But this is slow, expensive, and only gives a snapshot in time. It's like checking the gas gauge in your car only once a month; you might run out of gas in between. Other methods tried to use flow meters on the surface, but in heavy oil, these meters get clogged and unreliable.

The Detective's New Trick

This paper introduces a new way to solve the mystery without ever looking inside the pipe. The researchers built a "virtual sensor" using a computer algorithm called an Extended Kalman Filter (EKF). Instead of needing a physical sensor deep in the well, this system looks at the motor that spins the pump from the surface. It watches two main things: how fast the motor is spinning and how much "twisting force" (torque) the motor is working against.

The logic is simple but powerful: If the pump is lifting a lot of heavy oil, the motor has to work hard, and the torque is high. If the oil level drops and the pump starts lifting less fluid (or just air), the motor works easier, and the torque drops. The EKF acts like a super-smart calculator that constantly updates its guess of the oil level based on these changes. It's like a video game character that predicts where the enemy is hiding based on the sound of their footsteps, even if you can't see them.

What They Found

The team tested this idea on a real oil well called "Well X," located offshore in Trinidad. They hooked up their new computer system to the existing equipment and let it run while the pump was working. To see if their "virtual sensor" was telling the truth, they compared its guesses against the old-school acoustic measurements (the ones taken by technicians with microphones).

The results were impressive. The virtual sensor guessed the oil level with an average error of just 54.3 meters (178.16 feet). Considering the well is 1,420.4 meters (4,660 feet) deep, that's an accuracy of about 3.82%. The paper notes that this is a new benchmark for this kind of "sensorless" guessing. It's accurate enough to tell operators when the oil level is getting dangerously low, giving them time to slow down the pump before it runs dry.

The Safety Net

The system also has a built-in "panic button." The researchers realized that if the pump runs completely dry, the math can get confused because the rules change when there's no liquid. So, they added a safety rule: if the motor's twisting force drops to a very low level (meaning it's just spinning in air), the computer immediately stops guessing and screams "PUMP-OFF!" It sets the oil level to zero and alerts the operator. This prevents the computer from making up fake numbers when the machine is in trouble. They tested this safety feature using old data from a time the pump actually failed, and it worked perfectly, catching the problem before the machine could be damaged.

Saving Money and Time

Why does this matter? The paper does the math on how much money this could save a company. They looked at a fleet of 24 wells. By using this new system instead of the old, expensive proprietary equipment, they could save on hardware costs and monthly subscription fees.

• Revenue Boost: By knowing the exact oil level, they can run the pump at the perfect speed to get more oil out without breaking it. They estimate this could add about 5% more production, which equals roughly USD $7,726,320.00 in extra revenue per year for the whole fleet.
• Disaster Avoidance: If they can stop just one pump from running dry and breaking, they save the cost of a "workover" (a major repair job). That single saved repair is worth about USD $3,000,000.00 per well.
• Labor Savings: They don't need to send technicians to take acoustic measurements every month anymore. They can do it less often, saving about 1,920 labor hours a year.

The Catch and The Future

The paper is very honest about what it didn't do. They didn't let the computer automatically change the pump's speed on the live well yet. Why? Because if the computer makes a mistake and speeds up the pump too fast, it could snap the metal rod string inside the well, causing a disaster. So, for now, the system acts as a "watchdog." It tells the human operators, "Hey, the oil is getting low, you should slow it down," but the human has to press the button.

They also noted that the system updates every five minutes. While this is good for steady oil flow, it might be too slow if the oil flow changes very suddenly. And because the math uses some fixed numbers (like how thick the oil is), if the oil gets much thicker or the well changes over time, the system might need to be "recalibrated" by a human.

The Bottom Line

This paper proves that you don't need a broken-down sensor deep in the mud to know what's happening in an oil well. By using a clever math detective (the EKF) and listening to the motor's effort, you can guess the oil level with surprising accuracy. It's a cheaper, safer, and smarter way to keep the oil flowing, potentially saving millions of dollars and preventing catastrophic failures. While it's not fully automatic yet, it's a giant step toward making oil wells "smart" and self-aware.