A Smart Multimodal Innovative Model For Marine Environmental Monitoring

Hydrocarbon pollution represents one of the most serious issues for the health and entirety of the extremely fragile marine ecosystem, thus, the strategies for its monitoring have been grown in number and complexity in the last decades. Therefore, the realization of systems able to detect the presence of pollutants in the marine environment has become extremely complex, involving different figures and integrated know-hows. This paper presents an innovative model for the real-time assessment of pollutants on sea surface based on a network of autonomous underwater vehicles (AUVs), also able to sail the sea surface, equipped with sensors, capable of detecting volatile organic compounds (VOCs) produced by hydrocarbons. In particular, within this context, an AUV equipped with an E-Nose-like system is proposed, with the sensors employed that were characterized both on laboratory bench and at sea. The results obtained confirmed the feasibility of the approach proposed as well as a good reliability of the data acquired, confirming the likely employment of this system within an integrated marine monitoring tool. INTRODUCTION The Mediterranean Sea is almost completely surrounded by land, covering an approximate area of 2.5 million Km 2 , connected to the Atlantic Ocean by the Strait of Gibraltar, and representing 1% of the world ocean surface. Its average depth is around 1,500 m, with the deepest point located in the Ionian Sea, between Greece and Italy, 5,267 m under the sea surface (Barale 2008). It represents an extremely fragile and vulnerable ecosystem, being its waters slowly renewable, thus making it rather sensitive to all kinds of pollutants, especially when coming from commercial traffic of the big tankers, industrial and tourism activities (Er-Raioui et al. 2009). Pollution is more intense in coastal areas, where highly anthropized urban settlements are mainly located and maritime traffic is more concentrated. In order to preserve the integrity of this complex ecosystem, several protected areas have been created in the last decades. Such as the “Pelagos Sanctuary”, located in the Corso-Provencal Ligurian Basin, considered among the main feeding and reproductive areas for a number of cetaceans in the Mediterranean (Forcada et al. 1995; Notarbartolo di Sciara et al. 2003; Azzellino et al. 2012), and consistent with the EU Habitat Directive – Annex IV, having the aim to preserve, protect and improve the quality of the environment. In the last decades, in particular, petroleum pollution has become a matter of serious environmental concern Proceedings 29th European Conference on Modelling and Simulation ©ECMS Valeri M. Mladenov, Petia Georgieva, Grisha Spasov, Galidiya Petrova (Editors) ISBN: 978-0-9932440-0-1 / ISBN: 978-0-9932440-1-8 (CD) in the Mediterranean Sea and all over the world, with petroleum hydrocarbons (gasoline, kerosene, fuel oil, etc.) known to enter the marine environment through spills or leaks, as well as accidents (Mille et al. 2007; Zrafi et al. 2013). Thus, in the above cited areas, the monitoring of such pollutants is extremely critical to properly preserve the environment and the safety of animal species. To date, a number of different approaches have been employed to accomplish this difficult mission, including synthetic aperture radar (SAR) imaging and analysis (Alpers and Huhnerfuss 1989; Topouzelis et al. 2007), hyperspectral and thermal imaging (van der Meer and de Jong 2001), hydrodynamic mathematical modeling (Martins et al. 2001), and chemical sensors for electronic nose-like systems (Bourgeois and Stuetz 2002; Sobanski et al. 2006; Tonacci et al. 2015). Each of these approaches accounts for some relatively critical issues, including low detection capability during particular weather conditions (i.e. low and high wind speeds for SAR), or during particular parts of the day (i.e. at night). Thus, the employment of an innovative approach, based on the signals produced by electrochemical sensors in presence of hydrocarbons or other potentially dangerous pollutants, could represent an important development and/or a useful complement to the traditional monitoring methods, in order to improve their performances in such key-tasks for marine environment preservation. In order to perform a proper analysis of the environmental pollution state, the electrochemical sensors system, whose functioning is based on the Electronic Nose technologies, has been placed into an Autonomous Underwater Vehicle (AUV), sailing on the water‟s surface, and able to ride out customized or preloaded missions depending on the user‟s needs. Thus, the aim of this paper is to display a smart system, based on the technologies of Electronic Nose, able to dynamically monitor the presence of pollutants (particularly hydrocarbons) on the sea surface. The system proposed could be employed, together with traditional methods, for a complete and exhaustive analysis of the marine pollution caused by hydrocarbons. MATERIALS AND METHODS The main part of the smart system based on the ENose technology has been composed by an array of sensors, placed with a radial symmetry into a cylindrical flow chamber manufactured in polymeric material. The sensors employed for this application have been photoionization detectors, whose driving force relies on a vacuum ultra-violet radiation capable of ionizing volatile organic compounds (VOCs) contained in the air overhanging seawater. Such ionized particles were then detected by a proper electronics, placed inside (a pair of electrodes) and outside (electronic board) the sensor and producing an output signal somehow proportional to the concentration of VOCs present in the air. One of the main advantages of this system lied in the fact that such sensors were not responsive to major air components, thus not producing spurious signals possibly due to such a similar contamination of the sample drawn. The air inlet and outlet have been represented by a smart system of tubes, micropumps and valves, able to sample a given amount of air through an aspiration cone within a single analysis cycle, normally lasting 6 minutes overall (1 minute of air intake and 5 minutes for purging the system), and then releasing the air analyzed by an air outlet conceived to be connected to a hose external to the AUV. The integration within the AUV was another pivotal step that was performed in the design phase of the system described. The choice made was in favor of the modularity of the system, and the “E-Nose” payload was designed in the way it could be integrated but also to be detached from the remainder of the AUV when not necessary. Thus, all the cables and electronic components were connected with the remaining section of the vehicle, in order to ensure the possibility to be externally supplied and to reduce as much as possible the eventual inconvenience due to the displacement of the internal parts of the payload. The AUV (Figure 1) is composed of a central body, with an area reserved for a payload (320 mm in length, 110 mm in diameter) – the E-Nose for this purpose – with the possibility to be integrated with the external module when needed. The present system was reduced in dimensions and weight when compared to a similar system previously described (Tonacci et al. 2015). Figure 1: 3D rendering of the AUV with the Electronic Nose system The sampling and sensor chamber were resized in order to be placed into the payload compartment (113.6 mm 3 of volume for the sampling chamber, 127.2 mm 3 for the sensor chamber), while the pumps were dimensioned to be able to carry on the minimum air flow necessary for the VOCs analysis (the diameters of the hoses were 4 mm) (Figure 2). Figure 2: 3D rendering of the Electronic Nose system payload The control, sampling and analysis sections were placed within the payload compartment, together with the battery pack (14.4 V @ 5000 mAh), while the only part interfaced with the exterior was represented by the two tubes for air inlet and outlet, linked to a pillar able to keep them high enough to detect VOCs produced by hydrocarbons and, in the same time, to prevent the water from entering the sampler and damaging the overall system. The overall Electronic Nose system layout is displayed in Figure 3, with the electric and fluid dynamic connections between its different parts (sampling chamber, sensorized chamber, electrovalve and pumps). Figure 3: Layout of the Electronic Nose system Within the AUV, the engines were placed afore and astern. Data acquired by the E-Nose system were analyzed through two Artificial Neural Networks (ANNs) relying on Kohonen Self Organizing Map (KSOM) structure, based on clustering upon centroid distance (distance for each point and maximum distance from each cluster‟s ellipse centroid). RESULTS The system proposed was at first bench-tested in order to assess the responses provided by the sensors and to evaluate the optimal air flow to maximize the signal-tonoise ratio produced by the detectors employed. This latter analysis produced a clear result that confirmed the 2 l/min of air flow as the optimal air flow rate for the present application. The responses of the sensors in presence of hydrocarbons‟ dilutions was evaluated in terms of minimum detectable quantity of the various substances analyzed. In particular, a concentration of 100 ppm of the different hydrocarbons employed (gasoline, kerosene, diesel fuel and crude oil, often considered as the most frequently present compounds in polluted sea, according to Mille et al. 2007 and Zrafi et al. 2013) was clearly detected by the system. After this initial characterization, the system, integrated into the AUV, was deployed into seawater and used for marine monitoring purposes. The data gathered during these missions, together with the ones at bench, were employed to train two ANNs. The first of these ANNs aimed to classify the stimuli detected according to three different levels of warning („low‟, „medium‟ and „high‟, depending on the intensity of the stimulus concentration, corresponding to different amounts of hydrocarbons‟ VOCs present in the air). The ANN designed for this purpose was of the type KSOM, and was composed of 100 neurons (10 x 10) (Figure 4). Figure 4: The first ANN realized to discriminate stimuli depending on pollutants concentration This ANN had three inputs, corresponding to the maximum output of each of the three piD sensors (Bronze, Silver and Black piD), determined to be the most relevant features among all the data extracted from the E-Nose system after a Principal Component Analysis (PCA). The performances of this network were satisfying, with a good 80.76% of correct classification of the stimuli into the three classes above stated, against only 9.62% of misclassification (Figure 5). The most common misclassification was related to the difference between category 1 and 2 (low vs. medium concentration) minimally affecting, therefore, high concentrations that were the most important category of stimuli to be correctly detected in our purpose. Figure 5: Confusion matrix of the first ANN Similarly, a second ANN was designed and implemented, with somewhat comparable characteristics in terms of overall number of neurons. In this case, anyway, the outputs of the network were four, corresponding to the four types of hydrocarbons employed for the sensors‟ characterization (gasoline, kerosene, diesel fuel and crude oil, as above stated). The aim of this second ANN was to correctly identify the hydrocarbon detected based on the output signal features of the sensor network. In this case, after a PCA, three inputs were identified as the most significant, represented by the outputs of each of the three piDs in terms of peak time (tup). The task was correctly accomplished in 66.67% of cases, with 11.11% of misclassification, a slightly higher value when compared to the first task due to the different, and undoubtedly more complicated, analysis required in this latter case, thus resulting, anyway, in a satisfying overall outcome. Indeed, in this case, no common misclassifications were identified, probably because of the high complexity of the task required to the network.