Smart Fridge Spoilage Prediction Project

ISEF Category: Embedded Systems

Subcategory: Internet of Things  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A fridge can look cold and still have warm spots that speed up spoilage. That means two cartons on different shelves can age at different rates, even if they sit in the same appliance. If you can map those zones, you can predict risk before food goes bad. That turns a simple fridge into a live data system.

What Is It?

A digital twin is a software copy of a real object or system. In this project, the object is a refrigerator. Your twin watches what the fridge does, then uses that data to estimate what is happening inside each shelf zone.

The key signals are thermal pixel data from an MLX90640 sensor and door-open events. The thermal sensor acts like a low-resolution heat camera. It does not give you a perfect photo, but it can show cooler and warmer regions. Door-open events matter because every opening brings in warm air and changes the temperature near each shelf.

Your model can combine those signals to estimate spoilage risk. Think of it like a weather forecast for food. You are not directly measuring bacteria with the sensor. You are predicting the conditions that help food age faster, then comparing those predictions to actual storage outcomes or known freshness markers.

Why This Is a Good Topic

This topic works well because it has a clear input, a clear output, and real-world value. You can test whether thermal patterns and door usage predict faster warming in specific shelf zones, then turn that into a risk model. A student can learn sensor calibration, data fusion, time-series analysis, and model validation without needing a medical or industrial lab. The project also connects to food waste, energy use, and smarter home devices.

Research Questions

• How does shelf position affect the temperature stability measured by an MLX90640 sensor in a refrigerator?
• What is the effect of door-open frequency on predicted spoilage risk for each shelf zone?
• Does combining thermal pixel data with door-open events improve risk prediction compared with door-open events alone?
• To what extent can a simple model distinguish fast-warming zones from slow-warming zones inside a fridge?
• Which shelf zones show the largest mismatch between average fridge temperature and local thermal hot spots?
• What is the effect of loading the fridge with different food arrangements on predicted spoilage risk?

Basic Materials

• Raspberry Pi Zero or similar single-board computer.
• MLX90640 thermal camera module.
• Door-open sensor, such as a magnetic reed switch or Hall effect sensor.
• MicroSD card with power supply.
• Breadboard and jumper wires.
• Stable refrigerator or mini fridge for testing.
• Notebook or spreadsheet for logging observations.
• Thermometer or temperature probe for validation checks.

Advanced Materials

• Raspberry Pi Zero with GPIO accessories.
• MLX90640 thermal camera module with mounting hardware.
• Magnetic reed switch or Hall effect sensor for door state logging.
• Additional temperature probes for comparison points.
• Microcontroller or data logger for synchronized timestamps.
• Reference thermometer or calibrated temperature probe.
• Optional humidity sensor for extended modeling.
• Food-safe mock loads or inert thermal masses for controlled trials.

Software & Tools

• Python: Handles sensor logging, cleaning, feature extraction, and model building.
• Pandas: Organizes time-stamped sensor and door-event data into usable tables.
• NumPy: Supports numerical calculations for temperature trends and features.
• Matplotlib: Plots thermal patterns, door cycles, and predicted risk over time.
• ImageJ: Helps inspect thermal frames and compare pixel patterns visually.

Experiment Steps

1. Define the shelf zones you want to compare and decide what counts as spoilage risk in your model.
2. Set up synchronized logging for thermal frames and door-open events so every data point lines up in time.
3. Build a baseline map of normal temperature behavior in an empty fridge and in a loaded fridge.
4. Choose features that capture both heat patterns and user behavior, then plan how you will combine them into one prediction score.
5. Design a validation plan that compares predicted risk against a real freshness proxy or controlled storage outcome.
6. Decide how you will test whether your model works better than a simple average-temperature rule.

Common Pitfalls

• Mounting the thermal sensor in a spot that sees the door seal or exterior wall, which creates fake hot spots.
• Logging door-open events without matching timestamps to thermal frames, which breaks the data fusion step.
• Treating the whole fridge as one zone, which hides shelf-level differences and weakens the model.
• Using a validation proxy that changes for reasons unrelated to temperature, which makes the results hard to trust.
• Skipping a baseline run, which makes normal fridge behavior look like a treatment effect.

What Makes This Competitive

A strong version of this project goes beyond a simple temperature monitor. You would compare shelf-level zones, test more than one prediction method, and validate the model against a real freshness proxy or controlled storage pattern. Strong entries also quantify uncertainty, check false alarms, and show that the digital twin still works when the fridge is packed, opened often, or rearranged. That kind of careful modeling makes the work feel like an engineering system, not just a gadget.

Project Variations

• Use a mini fridge instead of a full-size fridge to see whether smaller volumes create stronger zone differences.
• Replace the thermal camera with a lower-cost sensor array and compare how much prediction quality drops.
• Add humidity as a second environmental signal and test whether it improves spoilage-risk estimates.

Learn More

• MIT OpenCourseWare: Search for courses on embedded systems, sensor networks, and data acquisition to build your hardware and logging plan.
• NASA Earthdata: Use remote-sensing and time-series tutorials to practice handling sensor streams and spatial patterns.
• NIH PubMed: Search for review articles on food spoilage, temperature abuse, and cold-chain monitoring.
• USDA FoodKeeper: Check storage guidance and typical shelf-life ranges for common foods.
• IEEE Xplore: Search for papers on smart refrigerators, digital twins, and IoT food monitoring, then read abstracts and methods sections.