A method for calibration of a particle concentration measurement system with artificial intelligence

In history, many particle concentration measurement instruments operate by determining the mass concentration (e.g., filtering and mass determination). Therefore, the laws and rules concerning air pollution today are primarily based on the principle of mass concentration. The mass-(or volume-) concentration means how much of the particle mass (in percentage) from the total mass of the particles is in a defined size range. However, modern particle concentration measurement devices, in particular, optical devices, primarily operate by determining the number concentration (e.g., light scattering or light blockage). The number concentration indicates how many particles (in percentage) of the total number of particles are in a defined size range. These devices have a much higher sampling rate and resolution. In order to meet the legal requirements based on the mass concentration, the conversion from the number concentration to the mass concentration is necessary. Technically, such conversion would be possible assuming that all the particles have the same material density and are completely spherical in shape. However, both are not the case in most practical applications. The shape of particles usually deviates from the sphere, and their material density can differ in different size classes. In addition, humidity has an influence on the size of the particles. In this work, a machine learning approach for the calibration of particle concentration measurement system is proposed for a relatively stable particle composition from applications such as street fine dust. The novelty lies in the conversion rules from the number concentration to the mass concentration by means of a neural network. For this, the neural network must be trained with a reference device, which primarily determines the mass concentration. The proposed method enables wider adoption of optical devices for particle concentration measurement and the results are comparable with the mass concentration, used for laws and rules.