Smoke detectors are the most widely used fire detectors due to their high sensitivity. However, they have persistently faced issues with false alarms, known as nuisance alarms, as they cannot distinguish smoke particles, and their responsiveness varies depending on the particle size and concentration. Although technologies for distinguishing smoke particles have shown promising results, the hardware limitations of smoke detectors necessitate an intelligent approach to analyze scattering signals of various wavelengths and their temporal changes. In this paper, we propose a pipeline that can distinguish smoke particles based on scattering signals of various wavelengths as input. In the data extraction phase, we propose methods for extracting datasets from time series data. We propose a method that combines traditional approaches, early detection methods, and a Dynamic Time Warping technique that utilizes only the shape of the signal without preprocessing. In the learning model and classification phase, we present a method to select and compare various architectures and hyperparameters to create a model that achieves the best classification performance for time series data. We create datasets for six different targets in our presented sensor and smoke particle test environment and train classification models. Through performance comparisons, we identify architecture and parameter combinations that achieve up to 98.7% accuracy. Ablation studies under various conditions demonstrate the validity of the chosen architecture and the potential of other models.
30 June 2022 in International Journal of Fire Science and Engineering
Photoelectric smoke detectors, which operate by reacting to the scattering of light caused by particles entering the light path, are widely used and extremely sensitive. Owing to higher standards imposed by Underwriters Laboratories, researchers have begun analyzing the properties of smoke particles. In particular, several wavelengths are used to classify particles by their scattering reactivity. The performances of actual smoke detectors are limited by their hardware and price. Therefore, properties that can distinguish particle types in these limited conditions must be determined. In addition, algorithms for extracting valid data intervals from unstable scattering data must be developed. In this study, scattering intensity ratios for three wavelengths are derived via simulations of light scattering by particles. An upper cumulative sum is defined for the three wavelengths, and an index for the start of particle inflow is extracted. In addition, valid intervals are extracted based on the scattering intensity ratios and the moving variance of adjacent wavelengths, and the properties of each particle are defined using the extracted indexes. For verification, a data acquisition device that can obtain data using the three selected wavelengths (470, 525, and 850 nm) from two sensors is designed. Five types of fire sources and non-fire alarm sources are selected and used in a test chamber designed to generate particle data. After applying the algorithm, the data in the valid data intervals can be used to derive a sample mean scattering intensity ratio that is more constant than that of the overall data or the data processed using the CUSUM index. In addition, the fire sources have a higher sample mean scattering intensity ratio than water vapor, which is a non-fire alarm source. The scattering intensity ratios for smoke particles can be extracted in real time via a comparison with experimental results obtained from the selected sensors.
This paper proposes a simple, high-efficiency refractive index (RI) sensor, with a structure based on the planar lightwave circuit (PLC) probe type. The optical sensor has a 1 × 2 splitter structure with reference and sensing channels, each consisting of a U-shaped waveguide structure that is configured by connecting C bends. This design allows for the sensor device to have a probe structure wherein the surface interconnected with activity devices (i.e., an optical source and optical detector) is placed on one side. The reference channel is bent with a minimum optical loss, and the sensing channel has a bent structure, involving a C-bend waveguide with a maximum loss. The C-bend waveguide with a maximum loss is conformally aligned to have a trench structure with the same bending radius, designed to selectively expose the sidewall of the core layer. The local index contrast varies depending on the material in contact with the trench, resulting in a change in the optical output power of the waveguide. The sensitivity of the proposed sensor was 0 and 2070 μW/refractive index unit (RIU) for the reference and sensing channels, respectively, as the RI changed from 1.385 to 1.445 at a 1550 nm wavelength. These results suggest that the proposed structure enables efficient RI measurement through the use of a simple dip-type method.