Odour concentrations prediction for the rubber industry based on the key odorous substances screening and machine...

Rubber products industry is a significant contributor to odour emissions in the world which can cause sensorial impact to surrounding communities. The odour concentration obtained by the artificial olfactometry is used as an important evaluation index for quantitative unpleasant odours. However, the artificial olfactometry has problems such as strong subjectivity, poor stability and timeliness, which has made researchers always have doubts about the test results. In this study, it is a new method that a machine learning based method for measuring odour concentration. 22 Chinese rubber products enterprises were statistically analyzed, with severe complaints in the past five years. The mixing and vulcanization, the most polluted stages, were dissected to obtain the characteristics of odour pollution. Subsequently, a prioritization model constructed based on odor thresholds, detection rates, and concentrations was employed to screen for key odorants. The screening results were further supplemented and verified via GCO olfaction, and 58 types of key odorants were ultimately identified. And combining the measured odour concentration, a random forest regression model was chosen from 5 models to establish and optimize the best odour concentration prediction model with R² value of 0.92. In addition, a qualitative method based on gas chromatography-ion mobility spectrometry was developed to improve the efficiency of odour concentration measurement. The principal component dimension reduction and a Bayesian model were employed to establish the relationship between the ion mobility response values of substances and the odour concentration, with R² value of 0.87, thus creating an on-site rapid measurement method for odour concentration.