ChillGuard:Intelligence against extreme cold
WINTERGUARD AI predicts indoor cold-risk conditions and provides intelligent safety guidance for elderly people living alone.
- Problem, approach and result explained
- Expert reviewed· 6 reviews
—
Environmental Monitoring — Real-time indoor temperature, humidity and pressure monitoring.
—
TCN-Based Forecasting — Time-series model integrated to predict changing indoor environmental conditions.
—
Intelligent Risk Guidance — Risk assessment and preventive recommendations based on sensor readings and predicted trends.
Overview
The problem
Imagine an elderly person living alone during a cold winter night.
As the temperature inside their room gradually decreases, they may be sleeping, resting or unable to recognise how cold the environment has become. If the situation continues, they may need assistance, but nobody might know that they are at risk.
Traditional environmental monitoring systems mainly display current temperature and humidity. They do not necessarily understand how environmental conditions are changing or provide timely, actionable guidance.
We identified an important question:
What if a safety system could recognise a developing cold-exposure risk, predict what might happen next and guide an elderly person before the situation becomes critical?
This became the motivation behind WINTERGUARD AI.
Our goal is to develop an affordable, intelligent and predictive safety system for elderly people living alone during extreme winter conditions.
The approach
We designed WINTERGUARD AI around the Arduino UNO Q dual-processor architecture. Our system combines environmental sensing, edge computing, time-series forecasting and intelligent recommendations. For environmental monitoring, we integrated the BME680, BME280 and DS18B20 sensors to collect indoor temperature, humidity, atmospheric pressure and gas-resistance measurements. The STM32U585 microcontroller handles real-time sensor acquisition and hardware control. The Qualcomm Dragonwing QRB2210 processor runs Linux and our Python-based application. One of the key technologies in our project is a Temporal Convolutional Network (TCN). Unlike a traditional monitoring system that only evaluates the current sensor reading, our TCN analyses historical sensor sequences to forecast future environmental conditions. The predicted values are then combined with live sensor readings and exposure duration in our intelligent risk assessment engine. Based on the assessed risk, our recommendation engine generates appropriate safety guidance. For example, if the system identifies a continuing downward temperature trend, it can recommend wearing warm clothing, using a blanket or moving to a warmer room. Our architecture also supports local alerts, safety-control actions and caregiver notifications through a FastAPI backend. We developed the HEATGUARD monitoring dashboard interface to display sensor readings, predictive information, risk classifications, recommendations and system status. Our central idea is simple: Don't just detect the cold. Predict the risk and guide the person towards safety.
The result
Through WINTERGUARD AI, we have developed an integrated prototype combining environmental monitoring, Arduino UNO Q processing, TCN-based time-series forecasting and intelligent risk assessment.
The system collects readings from multiple environmental sensors and makes them available to our processing application. The integrated TCN model is used to forecast future environmental trends, while our risk engine evaluates current and predicted conditions against configured safety thresholds.
The recommendation engine converts these assessments into understandable safety guidance.
Our monitoring interface brings these components together by displaying environmental readings, system information, predictive insights and risk status.
For example, consider a simulated winter scenario in which indoor temperature falls to 15.4°C. The system can evaluate the reading alongside exposure duration, classify the configured risk level and generate a cold-protection recommendation. Depending on the implemented hardware and communication connections, it can also trigger local alerts and notify a caregiver.
The project demonstrates how edge computing and predictive machine learning can be combined to create a more proactive elderly safety solution.
Reflection
Building WINTERGUARD AI taught us that developing an intelligent safety system involves much more than connecting sensors and displaying readings.
One of our biggest challenges was integrating hardware and software across the Arduino UNO Q's dual-processor architecture. We needed to understand how the STM32U585 and QRB2210 communicate and how real-time sensor data can be transferred into a Linux-based machine-learning application.
Integrating the TCN model also introduced important challenges involving historical data, preprocessing, time-series windows and prediction reliability.
We also realised that predicting an environmental change is not the same as confirming an emergency. A reliable safety system must consider exposure duration, occupancy, sensor reliability and false alarms before taking action.
If we continue developing this project, our priorities will include improving forecasting accuracy, validating recommendations using real-world winter data, strengthening fall and inactivity detection, improving offline operation and testing reliable caregiver communication.
Our biggest takeaway is that technology should not simply report a problem. It should help people understand what is happening, what they can do and when they may need assistance. WINTERGUARD AI is our step towards making elderly safety more predictive, responsive and accessible.
Submitted to Heat Intelligence at the Edge
- Arduino board used?
Yes
- Edge AI used?
Yes
- AI/ML model used
Time-series forecasting model for predicting environmental heat conditions over 30, 60, and 120 minutes, combined with a rule-based heat-risk engine and a location-specific digital twin. The exact ML model will be selected and validated using collected sensor data.
- Sensors used
BME688 environmental sensor, black-globe temperature probe, air-speed sensor, and optional VEML7700 ambient-light sensor. The LU90614 infrared temperature sensor may be used for supplementary surface-temperature monitoring.
- Camera used
No dedicated camera is required in the current design.
- Modulino used
None planned in the current design.
- Problem being solved
Industrial and workplace heat exposure can cause heat stress, reduce worker safety, and increase the risk of heat-related illness. HeatGuard Edge AI aims to monitor environmental heat conditions across multiple zones, predict rising heat risk, and trigger timely protective actions, even without internet connectivity.
- AI model training platform
Python-based model development and training using tools such as Google Colab or a local Python environment, with NumPy, pandas, and scikit-learn or TensorFlow as appropriate. The trained model will be optimized and deployed for local edge inference.
- Arduino UNO Q features used
Qualcomm Dragonwing QRB2210 processor for Linux-based computing, AI inference, forecasting, and the local dashboard. STM32U585 microcontroller for sensor acquisition, watchdog functions, status indicators, and deterministic control. Local data storage using SQLite. Local web dashboard and predictive multi-zone heat map. Sensor-network communication, real-time alarms, and offline-first operation.
- Deployment method
Deploy the application locally on Arduino UNO Q. Connect multiple sensor nodes through wired RS-485 or a suitable local interface. Run data validation, heat-risk assessment, forecasting, dashboard services, and safety logic locally. Cloud synchronization is optional.
- Key innovation
A location-specific, multi-zone thermal digital twin combined with Edge AI forecasting and offline-first safety monitoring. The system aims to anticipate heat-risk changes, assess sensor reliability, and initiate controlled protective actions with feedback verification, without depending on cloud connectivity.
- Solution summary
HeatGuard Edge AI is an Arduino UNO Q-powered industrial heat-stress monitoring and early-warning system. It combines environmental sensing, a location-specific digital twin, AI-based short-term forecasting, and safety-rule-based risk assessment to monitor multiple workplace zones. It provides a local dashboard, predictive heat map, audible and visual alarms, and approved fan or ventilation control. Its offline-first architecture supports continued monitoring and safety functions during internet outages.
- Report (PDF)
- Report (PDF)PDF
- What's working now?
WINTERGUARD AI is an intelligent, predictive elderly safety system designed to protect senior citizens living alone during extreme winter conditions. Built using the Arduino UNO Q dual-processor architecture, the system integrates BME680, BME280 and DS18B20 sensors to monitor indoor environmental conditions. The STM32U585 handles real-time sensor acquisition, while the Qualcomm Dragonwing QRB2210 runs the Linux-based Python application and integrated Temporal Convolutional Network (TCN) for time-series forecasting. The TCN analyses historical environmental readings to predict future temperature and humidity trends. These predictions are combined with real-time sensor data and exposure duration to classify environmental risk. Based on the detected risk, the recommendation engine generates preventive safety guidance. The system is designed to provide local alerts, automatic safety responses and caregiver notifications through a FastAPI backend. A web-based dashboard provides sensor monitoring, predictive insights, risk status, recommendations and historical information. Our goal is to move beyond traditional monitoring by predicting developing cold-related environmental risks and helping elderly people receive assistance before conditions become critical.
AI tools used
AI tools were used as development assistants for technical research, understanding Arduino UNO Q architecture, debugging sensor integration, exploring TCN model implementation, improving Python code, and refining project documentation. All hardware integration, implementation and testing decisions were reviewed by the team.