Abstract
Oil refinery and petrochemical plants are of crucial importance to enhance world economy. Besides the related considerable economic impact of this industries, the population and governments pose towards a great attention which boosted the necessity of developing effective tools for the control of the associated environmental pressures, without hindering economy growth. Among the environmental aspects of key relevance in the refinery sector, air quality and atmospheric pollution in terms of odour emissions, especially in the last years, have become a priority concern in the management of the plants. The monitoring and control of the odour emissions are thus needed in order to prevent potential impact and avoid complaints from the exposed population. Nowadays, senso-instrumental methods by using Instrumental Odour Emissions Systems (IOMSs) represents the most attractive tool for the monitoring of environmental odours, allowing the possibility to obtain real-time and continuous information’s, to support the decision-making process and for the implementation of proactive approach. No universally accepted regulation or standardized procedure exist at present and limited data are available in the scientific and technical literature with reference to these systems for the odour monitoring control in the oil refinery plants. In the CEN/TC264 ‘Air Quality’ standardization committee, a specific working group (WG41) is currently working to produce a validation procedure for the IOMSs. The research presents and discusses an advanced instrumental odour monitoring system device useful for the continuous characterization of the odour emissions in complex oil refinery and petrochemical plants. The influence of the use of different feature extraction methods in the odour monitoring model elaboration, in terms of classification, is analysed. Results demonstrate the existence of considerable differences in terms of correct classification percentage, in the use of different feature extraction methods. The importance and usefulness of having a fully-developed flexible system that allows to select and compare automatically different settings options, as the different feature extraction methods, in order to guarantee the best training model, is highlighted. Results demonstrated the high reliability of the system in recognizing artificial gas samples, with characteristics similar to the emissions from refinery plants.
