KOOPMAN-Luenberger Observer Design for Nonlinear Systems with Application to the Monitoring of a Latent Thermal Energy Storage
This paper proposes a KOOPMAN-based Linear State Observer (KOOPMAN-LSO) framework that lifts nonlinear dynamics into a higher-dimensional space to enable efficient linear observer design, demonstrating its effectiveness in accurately estimating unmeasured internal states of a latent thermal energy storage system with high fidelity under limited sensing conditions.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are trying to guess the weather inside a giant, sealed, magical jar. You can't see inside, and you can't stick a thermometer in there without breaking the jar. All you have is a tiny window on the outside where you can feel the temperature of the air blowing out. If the jar is simple, like a bucket of water, you could probably guess what's happening inside just by feeling that breeze. But what if the jar contains a mysterious, shape-shifting substance that freezes and melts, changing its personality and behavior depending on how hot or cold it gets? That's the puzzle scientists face with "latent thermal energy storage" systems—giant batteries that store heat by melting materials like salt or wax. These systems are incredibly efficient but notoriously tricky to monitor because their insides are hidden, and their behavior is wildly nonlinear, meaning a small change in input can cause a huge, unpredictable shift in temperature.
To solve this, engineers usually try to build complex mathematical models that mimic the physics of the melting process, or they rely on expensive, messy networks of sensors. But there's a smarter, more modern way: using data to find hidden patterns. Think of it like taking a blurry, chaotic photo of a storm and using a special filter to turn it into a clear, straight-line graph that predicts the next gust of wind. This paper introduces a clever new "filter" called the KOOPMAN-based Linear State Observer. It's a method that takes a messy, nonlinear system and "lifts" it into a higher dimension where the chaos looks like simple, straight lines. Once the chaos is straightened out, the team can use simple, reliable math (the kind used in basic engineering) to guess the hidden temperatures inside the jar just by looking at the air blowing out. It's like turning a tangled ball of yarn into a straight line so you can easily pull out the end you need.
The researchers, led by Mustapha Habib and his team, applied this idea to a real-world heat storage unit filled with a special salt-based material called SP31. This material acts like a thermal sponge, soaking up heat as it melts and releasing it as it freezes. The challenge was that they only had one sensor measuring the temperature of the water leaving the system, yet they needed to know the temperature of the material deep inside. They built a "digital twin" of the system using a technique called eDMDc, which learns how the system behaves by watching it run through charging and discharging cycles. Instead of trying to model the complex physics of melting directly, they used "physics-informed" math tricks—functions that look like the way melting happens—to lift the data into that higher-dimensional space where the rules become simple and linear.
Once the system was "lifted," they attached a "Luenberger Observer," which is essentially a smart guesser that constantly corrects its own predictions. Imagine a student taking a test who knows the answer key is slightly different from their guess; every time they get a new clue (the outlet temperature), they adjust their answer to get closer to the truth. The team tuned this guesser using a method called LQR, which acts like a volume knob, balancing how fast the guesser learns against how much it trusts the noisy sensor data. They tested this setup in a lab with a 200-liter tank, running it through various scenarios with different water flow rates (2, 4, and 8 liters per minute) and temperatures.
The results were impressive. When the system was running at a steady pace, the observer could predict the temperature of the water leaving the tank with incredible precision, missing by as little as 0.0819 °C. More importantly, it successfully reconstructed the hidden temperatures of the melting material inside, keeping the average error below 1.0 °C for most sensors. However, the paper is honest about its limits. When the water flow was cranked up to 8 liters per minute, or when the system suddenly switched from heating to cooling, the "guessing" took a little longer to catch up, and the errors grew slightly larger, reaching up to 2.3072 °C for some specific spots inside the tank. The authors explain that this happens because some parts of the tank are just too far away from the single sensor to be seen clearly, a bit like trying to hear a whisper from the back of a noisy room.
Crucially, the paper argues against the idea that you need a perfect, physics-heavy model or a sensor in every nook and cranny to monitor these systems. It also suggests that while other methods like the Extended Kalman Filter exist, they can be computationally heavy and harder to tune. This new approach offers a "middle ground": it's not a perfect crystal ball that sees everything instantly, but it's a robust, computationally efficient tool that turns a messy, nonlinear problem into a manageable linear one. The authors suggest that while this method works well for steady conditions, future work might combine it with other data-driven techniques to handle even wider ranges of operating conditions. For now, it stands as a promising way to keep an eye on the hidden heart of thermal batteries without needing to drill holes in them.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.