Particle Size Prediction Method for Fire or Nuisance Event using a Simple Photoelectric Detector

Abstract This study demonstrates that a simple smoke detector with minimal components can be used to estimate the particle size of fires or nuisance incidents. Machine learning (ML) models were created using scattered light data. Various test materials such as wood, cotton, polyurethane foam, cigarette, N-heptane, printed circuit board (PCB), paraffin, polyalpha olefin (PAO), di-ethyl hexyl-sebacate (DEHS), plaster powder, and cement dust were used in the experiments. The proposed prediction method was tested against completely unknown particles of cigarette, PAO, PCB, and plaster powder, which were not used in the training. The particle size prediction capability of forward, backward, and side scattering of light at 980 nm was investigated using ML models with time correlation function (TCF) data. The prediction errors of the best ML model for particle median sizes ranged from 0.4% to 35.1%. Traditional simple smoke detectors using the proposed methodologies can measure the median particle size and volume concentration, thereby effectively suppressing false alarms.