Artificial neural network in the measurement of environmental odours by e-nose

Odour measurement plays a crucial role in environmental odour management. Continuous odour measurement systems are promoted to keep the situation always under control, such as being able to adopt the most suitable mitigation measures in real time to avoid odour complaints and impacts. Electronic Nose (eNose) represents currently the instrument of having the highest future developing potential to guarantee continuous odour measurements. To use an eNose, a training phase is however mandatory, which has the scope to create the Odour Monitoring Model (OMM) that is able to identify the presence of odour, the different odour classes and the quantification of the odorous stimuly. Statistical or biological inspired measurement techniques are applied to create the optimum OMM. The study presents and discusses the elaboration of an Artificial Neural Network (ANN) technique to recognize environmental odour with eNose. The proposed system was architected on a feed-forward neural network with Bayesian Regularization algorithm using Matlab R2017a software. The elaborated ANN was tested and validated using the seedOA eNose, realized by the Sanitary Environmental Engineering Division (SEED) of the Department of Civil Engineering of the University of Salerno (Italy). Tests were carried out analyzing odour samples collected at a large Wastewater Treatment Plant (WWTP). The comparison between the Odour Monitoring Model (OMM) elaborated through the proposed ANN system and the traditional statistical techniques, such as the Partial Least Square (PLS) and the Linear Discriminant Analysis (LDA), is also discussed. Results shown the efficiency of the elaborated ANN to identify the different odour classes and predict the odour concentration in terms of OUm-3. The artificial neural network shows smaller Root Mean Squared Errors (RMSE) and greater coefficient of determination (R2) as compared to the traditional statistical methods. The main advantages of neural networks are their adaptability in terms of learning, self-organization, training and noise-tolerance.