Performance and neural modeling of a compost-based biofilter treating a gas-phase mixture of benzene and xylene

Biofilter (BF) has been regarded as a versatile gas treatment technology for removing volatile organic compounds (VOCs) from contaminated gas streams. In order for BF to be utilized in the industrial setting, it is essential to conduct research aimed at removing VOC mixtures under different inlet loading conditions, i.e. as a function of the gas flow rate and inlet VOC concentrations. The main aim of this study was to apply artificial neural networks (ANN) and determine the relationship between flow rate (FR), pressure drop (PD), inlet concentration (C), and removal efficiency (RE) in the BF treating gas-phase benzene and xylene mixtures. The ANN model was trained and tested to assess the removal efficiency of benzene (REB) and xylene (REX) under the influence of different FR, PD and C. The model's performance was assessed using a cross-validation method. The REb varied from 20% to >60%, while the REx varied from 10% to 70% during the different experimental phases of BF operation. The causal index (CI) technique was used to determine the sensitivity of the input parameters on the output variables. The ANN model with a topology of 4-4-2 performed the best in terms of predicting the RE profiles of both the pollutants. Furthermore, the effect was more pronounced for xylene because an increase in the benzene concentration reduced xylene removal (CI = -25.7170) more severely than benzene removal. An increase in the xylene concentration had a marginally positive effect on the benzene removal (CI = +0.1178).

Automated summary

Biofilter (BF) is a versatile gas treatment technology for removing volatile organic compounds (VOCs) from contaminated gas streams. This study applied artificial neural networks (ANN) to determine the relationship between flow rate, pressure drop, inlet concentration, and removal efficiency in the BF treating gas-phase benzene and xylene mixtures. The model's performance was evaluated using a cross-validation method, and the results showed that the ANN model with a topology of 4-4-2 performed the best in predicting the removal efficiency of both pollutants.