Machine Learning and Threshold Analysis of Sensor Data for Cooktop Fire Prevention

Cooking equipment is involved in nearly half of home fires in the United States, with cooktop fires the leading cause of deaths and injuries in cooking-related fires. Researchers from NIST’s Engineering Laboratory and SED have evaluated 16 electrochemical, optical, temperature and humidity sensors, placed in the cooktop exhaust duct for use in predicting and preventing cooktop ignition. The sensors were evaluated in a series of 60 experiments conducted in a mock kitchen. Experiments covered a broad range of conditions, including both unattended cooking and normal cooking scenarios, where 39 experiments led to auto-ignition. The experiments involve a variety of cooking oils and foods and were conducted using either an electric coil cooktop, gas-fueled cooktop, or electric oven.  

The sensor data collected in the experiments were used in two types of analysis, threshold analysis and neural-network analysis, to estimate the performance of sensors for predicting ignition and ignoring normal cooking conditions. The combined information from multiple sensors was evaluated using sensor ratios in the threshold analysis, and in the neural-network models using selected pairs of sensor inputs. The best performing sensor, consistently across both threshold and machine learning analyses, was the sensor measuring volatile organic compounds. This sensor was also included in all the best performing sensor ratios and sensor pairs considered. Details of this collaborative work are published in Fire Technology.  To facilitate further research and algorithm developments, the complete time series data for this project was also published recently by NIST scientists.