Real-Time Single Particle Classification Based On Polarized Light Scattering Characterization And Support Vector Machine

Aerosol particles refer to gaseous scatterers composed of solid or liquid particles suspended in gaseous medium.Their geometric size distribution is wide,generally several hundred microns to nanometers.Because aerosol particles are ubiquitous in real life,the physical and chemical characteristics of aerosol particles are closely related to climate environment and human health.In addition,in industrial production processes(such as medicine,food,cosmetics,etc.),aerosol size,shape,concentration and other characteristics will affect its biological efficacy,life and other indicators.Therefore,how to conduct quantitative and qualitative analysis of particulate matter with the help of emerging sensor technologies is crucial for air quality and human settlement environment monitoring and industrial process control.Optical measurement technology based on light scattering theory is a common online nondestructive testing technology in industry.This method obtains the information of particle size and concentration by analyzing the light scattering characteristics of particles.However,the scattered light intensity depends not only on the particle size,but also on the refractive index of the particle(which is usually unknown),which not only affects the accuracy of particle size detection,but also limits the application of this method in qualitative classification and identification of particulate matter in principle.Optical polarization measurement is sensitive to material refractive index and can effectively improve the information dimension,so it is expected to realize the detection and classification of aerosol single particle fingerprint spectrum.According to the above application requirements and research background,this paper carried out the research of single particle classification based on polarized light scattering measurement and support vector machine.Through theoretical simulation,experimental testing,result analysis and optimization,the on-line dynamic measurement and identification of various particles are realized.The research contents mainly include:(1)Aiming at the three common aerosol particles(PSL,Silica,Carbon)in daily life,this paper firstly simulated the relationship between the scattering intensity of single particles and their particle size,refractive index,illumination wavelength,and acquisition Angle based on the Mi scattering theory.Then,the feasibility of using the polarization ratio of the scattered light to classify the single Angle scattered signal acquisition mode and the basis of Angle optimization selection were studied.In addition,in order to further improve the classification accuracy,the feasibility of particle classification by using the forward and backward ratio of scattering and the polarization ratio in the case of double-angle scattering signal collection was simulated,which provides theoretical support for the construction and testing of the subsequent experimental equipment in this paper.(2)According to the previous theoretical analysis results,the portable sensing device based on single Angle and double Angle scattering signal acquisition was designed and manufactured.Firstly,the actual effects of different polarization splitter components in the collection of single particle scattering polarization components were compared,and the collection scheme of scattering light polarization components was determined.Then,3D printing technology was used to manufacture the mechanical structure of the sensor.The optimized sheath flow design not only improved the detection accuracy,but also avoided the pollution of particles on the surface of the optical element during the detection.In addition,we completed the design and customization of laser lighting module and photoelectric signal acquisition and amplification module,and designed and processed an aspherical lens that can shape Gaussian beams into one-dimensional flat-top beams.This avoided particle size inversion errors caused by uneven energy distribution when Gaussian beams were used for lighting,and significantly improved particle size detection accuracy.The acquisition of weak scattered signal of single particle was realized.Finally,the aerosol single particle generating device and sensor testing device were built.(3)Experiments verified the online dynamic classification scheme of particles in the paper.The aerodynamic characteristics of particles based on the pulse width of light scattered signal were studied.The scattered signals of three different kinds of particles with different sizes were tested experimentally.Combined with the supervised learning-based support vector machine algorithm,the accuracy of particle classification by polarization ratio in single Angle scattered signal acquisition mode and the accuracy of particle classification by forward and backward ratio and polarization ratio in double Angle scattered signal acquisition mode were analyzed.It was found that the classification accuracy of single particle scattering signals can reach more than 90% when the forward/backward ratio and polarization ratio were used for classification,which,as far as the authors know,was significantly higher than the classification accuracy of single particle based on light scattering polarization measurement reported in the literature.Finally,in order to meet the requirement of Intrinsic safety of sensors in specific industrial fields,this paper developed and constructed a particle online detection and identification device based on dual-wavelength laser illumination and modulation trigger acquisition mode.The device had a promising application prospect in the pharmaceutical industry involving toxic,flammable and other energy sensitive particles.