A convolutional neural network (CNN) was developed to recognize sprinkler activation based on long-wave infrared (LWIR) images, creating a nonintrusive, real-time model for detecting sprinkler activation. Training data were taken from ten large-scale fire tests with storage heights ranging between 4.6 m and 13.7 m and ceiling heights ranging between 6.1 m and 15.2 m. A sample of 25,000 LWIR images was extracted from the fire tests, split 70/30 between training/testing data. To prevent overfitting, the images were randomly reversed and cropped. The time required to train the model was reduced by 96% through GPU computing. The overall accuracy of the model was 99.7% for both pendent and upright sprinklers. The methodology described in this study can be generalized and applied to other image classification problems.
Adequate fire protection of distilled spirits stored in oak barrels requires understanding the failure mode of these barrels, including quantifying the leak rate. In this study, the use of a custom-calibrated, long-wave microbolometer camera is demonstrated to seek new protection methods for rack-stored distilled spirits. Individual oak barrels ranging between 200 L and 500 L filled with 75%/25% ethanol/water were exposed to both propane gas fires and pure ethanol pool fires. The IR camera was used to see through the smoke and flames showing the location of the leaks. The increase in HRR due to the leaked content was measured using gas calorimetry of the combustion products. This study showed that barrels leaked at a rate of approximately 4-8 lpm, resulting in heat release rates ranging between 1.2 and 2.4 MW. These numbers are confirmed by the quantitative measurements of gaseous H2O and CO¬2 in the exhaust. Surface temperature of the exposed oak could reach temperatures up to 750ºC.
Long-wave infrared (LWIR) cameras provide the unique ability to see through smoke and condensed water vapor. However, soot generated inside the flame does attenuate the LWIR signal. This work focuses on gas flame attenuation effects of LWIR signals originating from a blackbody. The experimental setup consists of time averaged, laboratory-scale turbulent diffusion flames with heat release rates set at 5 kW, 10 kW, and 15 kW. Propylene and ethylene were used as fuel, providing two different soot yields. A 30 cm by 30 cm blackbody was used with maximum surface temperatures set to 600°C. Both instantaneous and time-averaged blackbody temperature profiles through the flame were measured using a LWIR microbolometer camera (7.5–14 μm). Flame intermittency was quantified by color segmenting visible images. The experiments showed that low blackbody temperatures were significantly affected by the presence of the flame. At 600°C, the effect of flame absorption matches the emitted radiation from the flame itself. Using data obtained at various blackbody temperatures, the flame transmittance was obtained using a Generalized Reduced Gradient optimization method. The transmittance was lower for propylene flames compared to ethylene flames. Ethylene flames were shown to have higher temperatures. Using the values for flame radiance and transmissivity, the total averaged radiance of the flame plus the blackbody could be reproduced with 1% accuracy.
The development of fire modeling tools capable of predicting large-scale fire phenomena is of great value to the fire science community. To this end, FM Global has developed an open-source CFD fire simulation code, FireFOAM. The accuracy of this code relies fundamentally on high-quality experimental validation data. However, at larger scales, detailed measurements of local quantities (e.g., surface temperatures) needed for model validation are difficult to obtain. Often, the information obtained from large-scale fire tests is limited to the global heat release rates (HRR) or point temperature or heat flux measurements from embedded thermocouples or heat flux gauges, respectively. The present study addresses this limitation by introducing IR thermographic measurements in a three- and a five-tier-high rack storage scenario. IR temperatures are compared against modeled results. The tested and modeled cases represent realistic industrial warehouse fire scenarios. The rack-stored commodity consisted of corrugated paperboard boxes wrapped around a steel cubic liners, placed on top of a hardwood pallet. The global heat release rate was measured using a 20- MW fire products collector located inside FM Global’s Fire Technology Laboratory. An in-house calibrated microbolometer IR camera was used to obtain two-dimensional temperature measurements on the fuel surfaces and on the surfaces inside the flue spaces. Maximum temperatures up to 1200 K were observed on the external surfaces of the test array. Inside the flue spaces between pallet loads, temperatures up to 1400 K were measured. The modeled fire spread results match well fire spread shown in the IR thermographic images. The peak modeled surface temperatures obtained inside some of the horizontal flue spaces were ~1400K, which agreed well with the peak temperatures seen by the IR camera. The effect of the flames present between the surfaces of interest and the IR camera only contribute to about 50 K increase in measured temperature due to the limited flame emissive power with low soot concentration in the long-wave IR regime. This study shows the capability of IR cameras to obtain high resolution temperature measurements in large-scale fire scenarios, which enhances existing large-scale model validation data set.
Quantitative temperature measurements of large-scale fires are of key interest to FM Global’s researchers and engineers. In this study, the effectiveness of extending an uncooled fixed-integration-time IR camera’s temperature range via a reduced aperture was investigated. The corresponding calibration of the focal plane array (FPA) was performed, in situ, by investigating spatially resolved radiance levels with and without the aperture present. In-the-field calibration results were compared and validated using a blackbody (ε = 1) source. The effect of reduced radiant intensity on the noise equivalent temperature difference (NETD) was investigated over a wide temperature range. This study shows the effective temperature extension of a fixed-integration-time (microbolometer) IR camera from 1200ºF (650ºC) to 2192ºF (1200ºC), making this camera particularly suitable for studying fires. The temperature extension was accomplished at low cost without changing the integration time of the focal plane array (FPA), removing the camera’s lens, or by using a neutral density (ND) filter.
Proceedings Volume Editor (2)
This will count as one of your downloads.
You will have access to both the presentation and article (if available).