Model Predictive Kettle Control for Energy Savings

ISEF Category: Robotics and Intelligent Machines

Subcategory: Control Theory  ·  Difficulty: Advanced  ·  Setup: School Lab  ·  Time: 1 to 2 Months

The Hook

Your kettle does not have to work like a light switch. A smarter controller can learn when to push hard, then ease off before it overshoots. That can save energy and still hit a target temperature faster. This project turns a kitchen appliance into a control systems testbed.

What Is It?

Model predictive control, or MPC, is a way for a computer to plan ahead. Instead of reacting only to the current temperature, it predicts how the kettle will warm up over the next few steps and chooses the best heater command now. Think of it like steering a bike while looking a few feet ahead on the path, not just at the front wheel.

Bang-bang control is the simple version. The heater is either fully on or fully off. That can work, but it often causes overshoot, which means the water gets hotter than your target, or it wastes energy by cycling too often. MPC tries to balance speed, energy use, and smooth control at the same time.

For your project, you build a math model of the kettle, then test how well the controller predicts and tracks temperature. The key idea is not the kettle itself. The key idea is the control rule. You measure how different rules change heating time, energy use, and temperature error.

Why This Is a Good Topic

This topic works well because you can test a real control strategy with clear numbers. You can measure time to target, overshoot, and energy use, then compare MPC against bang-bang control. That makes the project easy to judge and easy to improve. It also connects to real problems in appliances, battery systems, and industrial temperature control.

Research Questions

• How does a model-predictive controller change energy use compared with bang-bang control for the same kettle and setpoint? ?
• What is the effect of different prediction horizons on overshoot and settling time? ?
• Does adding a penalty for heater switching reduce temperature oscillation without slowing the response too much? ?
• To what extent does water starting temperature change the controller’s tracking error? ?
• Which model structure predicts kettle temperature more accurately, a simple first-order model or a model with heat-loss terms? ?
• How does sensor placement affect the measured performance of the controller? ?

Basic Materials

• Household electric kettle with exposed or measurable temperature response.
• K-type thermocouple probe with readout or data logger.
• Digital multimeter with current or power measurement function.
• Stopwatch or phone timer.
• Notebook for logging setpoints, response times, and energy data.
• Computer for model fitting and controller simulation.
• Safety gloves or heat-resistant tools for handling hot equipment.

Advanced Materials

• Household electric kettle with accessible power measurement.
• K-type thermocouple and thermocouple amplifier or DAQ interface.
• Inline power meter or wattmeter with data export.
• Microcontroller or relay board for control switching, if your setup allows it safely.
• Laptop with Python, do-mpc, CasADi, NumPy, SciPy, and Matplotlib.
• Data acquisition hardware for synchronized temperature and power logging.
• Insulation materials for testing model assumptions about heat loss.

Software & Tools

• Python: Fits the kettle model, runs simulations, and compares controller performance.
• do-mpc: Builds and tests model-predictive control strategies.
• CasADi: Solves the optimization problem behind MPC.
• NumPy: Organizes sensor data and performs numerical calculations.
• Matplotlib: Plots temperature trajectories, control actions, and energy curves.

Experiment Steps

1. Define the control goal, the setpoint range, and the performance metrics you will compare.
2. Build a simple thermal model of the kettle from measured heating data and heat-loss behavior.
3. Choose the control strategies you will test, such as bang-bang control and MPC, and match them on the same setup.
4. Plan the sensor and logging system so temperature, power, and time data stay synchronized.
5. Decide how you will judge success, using overshoot, settling time, tracking error, and energy use.
6. Design repeat trials across different starting temperatures or water volumes so you can test whether the controller still works well under changed conditions.

Common Pitfalls

• Using a temperature sensor that touches the heater or kettle wall, which makes the water reading look faster or hotter than it really is.
• Comparing control methods with different starting water temperatures, which makes one controller look better for the wrong reason.
• Ignoring power logging, which leaves you unable to prove that one method used less energy.
• Building a model from only one trial, which makes the MPC predictions too tuned to one run.
• Letting room drafts, lid position, or water volume vary between trials, which adds heat-loss noise that hides the controller effect.

What Makes This Competitive

A stronger project goes past a simple comparison chart. You can test more than one model, then show which model predicts the kettle best under changing start conditions. You can also use a harder metric, like total energy to reach target within a tight error band, instead of only reporting peak temperature. If you add uncertainty analysis, repeated trials, and a clear reason your controller generalizes better, the project looks much more like real control engineering.

Project Variations

• Test how the controller performs on different kettle fill levels, then compare whether one model still predicts heat loss well.
• Replace temperature-only control with a multi-objective controller that also limits heater switching and energy spikes.
• Compare MPC against proportional control, bang-bang control, and a hand-tuned on-off threshold to see which method gives the best tradeoff.

Learn More

• MIT OpenCourseWare, Control Systems: search the course materials for feedback control, state-space models, and stability basics.
• MIT OpenCourseWare, Model Predictive Control: search for lecture notes and assignments on constrained control and optimization.
• do-mpc Documentation: read the free examples and API guide for building MPC models in Python.
• CasADi Documentation: find the tutorials on nonlinear optimization and automatic differentiation.
• NOAA Physics and Energy resources: use background material on heat transfer, measurement, and energy accounting.
• PubMed: search for review articles on thermal control, appliance energy efficiency, and sensor-based temperature regulation.