IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 58, NO. 9, SEPTEMBER 2009 3253 Smart Sensors Network for Air Quality Monitoring Applications Octavian A. Postolache, Senior Member, IEEE, J. M. Dias Pereira, Senior Member, IEEE, and P. M. B. Silva Girão, Senior Member, IEEE Abstract—This paper presents a network for indoor and out- door air quality monitoring. Each node is installed in a different room and includes tin dioxide sensor arrays connected to an acquisition and control system. The nodes are hardwired or wire- lessly connected to a central monitoring unit. To increase the gas concentration measurement accuracy and to prevent false alarms, two gas sensor influence quantities, i.e., temperature and humidity, are also measured. Advanced processing based on multiple-input– single-output neural networks is implemented at the network sensing nodes to obtain temperature and humidity compensated gas concentration values. Anomalous operation of the network sensing nodes and power consumption are also discussed. Index Terms—Air quality (AirQ), embedded Web server, neural network, wireless networks. I. I NTRODUCTION A IR supplies us with oxygen that is essential for our bodies to live. Air is 99.9% nitrogen, oxygen, water vapor, and inert gases. Human activities can release substances into the air, some of which can cause problems for humans, plants, and animals. Air quality can be expressed by the concentration of several pollutants such as carbon monoxide (CO), sulphur dioxide, nitrogen dioxide, and ozone. The threshold values specified by the European Environment Agency [1] for these pollutants are 10, 350, 40, and 120 μg/m 3 , respectively. Pollution also needs to be considered inside our homes, offices, and schools. Some of these pollutants can be created by indoor activities such as smoking and cooking. Generally, in industrialized countries, the population spends about 80%–90% of time inside buildings and is therefore exposed to harmful indoor pollutants. Indoor air quality is generally assessed by separately measuring CO, temperature, and humidity [2]. This information, even if fused, is insufficient to allow a good characterization of indoor air quality. The development of wireless local area network (WLAN; IEEE802.11X) technology and the marketing of low-cost ac- Manuscript received July 3, 2007; revised October 21, 2008. Current version published August 12, 2009. This work was supported by Programa Operacional para a Sociedade da Informação through Project SFRH/BPD/11549/2002. O. A. Postolache and J. M. Dias Pereira are with the Instituto de Teleco- municações, Lisbon 1049-001, Portugal, and also with the Escola Superior de Tecnologia de Setúbal, Instituto Politécnico de Setúbal, Estefanilha 2910-761 Setúbal, Portugal (e-mail: poctav@alfa.ist.utl.pt; joseper@est.ips.pt). P. M. B. Silva Girão is with the Instituto de Telecomunicações, Lisbon 1049- 001, Portugal (e-mail: psgirao@ist.utl.pt). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TIM.2009.2022372 cess points (APs; e.g., Linksys WAP11), wireless network adapters (CardBus; e.g., D-Link DWL-G650+), and wireless bridges (e.g., DWL-810+) creates the possibility of implement- ing indoor/outdoor air quality monitoring networks character- ized by high flexibility, modularity, and low cost. Tin oxide sensors (e.g., Figaro, Nemoto [3]) are inexpensive and fair selective gas sensors. To overcome some of their limitations such as cross sensitivities [4], [5] and a temperature and humidity dependence behavior [6], appropriate sensor data processing is required. The aim of this work is to present a Wi-Fi indoor—outdoor air quality monitoring network that combines the capabilities of tin oxide sensors with advanced sensor data processing based on multilayer perceptron neural networks for an accurate measurement of air quality and for the detection of air pollution events and of sensors’ abnormal operation. II. DIRECT AND I NVERSE MODELING OF THE SENSORS’CHARACTERISTICS The sensors’ nonlinearity requires the utilization of direct and inverse modeling for sensor calibration and on-line mea- surement phase [7]. For the particular case of tin oxide gas sensors TGS800, TGS822, TGS842, and TGS203, the sensors’ response is strongly dependent on parameters such as tempera- ture, humidity, and cross influence of the other gases. For prac- tical and economic reasons, the number of calibration points is very low, and thus, a neural network (multilayer perceptron architecture), which is a global approximator of multivariable characteristics [8], was used in this paper. Polynomial modeling is another solution for multivariable characteristics modeling. Representative of this type of solution is the polynomial model that is a part of the IEEE1451.2 standard for smart sensors particularly related to smart sensors correction engine imple- mentation [9]. The method represents an interesting solution. However, it requires a large set of data (i.e., a higher number of calibration points compared with a neuronal network model) for polynomial model coefficients calculation [10], i.e., D(1)  i=0 D(2)  j=0 ··· D(n)  p=0 C i,j,...,p [X 1 −H 1 ] j [X 2 −H 2 ] j ··· [X n −H n ] p (1) where X n are the input variables to the sensor characteristic block, H n are the offsets to the input variables, and the D(k) represents the degree of the input X k , i.e., the highest power to which [X k − H k ] is raised in any term of the multinomial. The 0018-9456/$26.00 © 2009 IEEE Authorized licensed use limited to: UNIVERSIDADE TECNICA DE LISBOA. Downloaded on November 11, 2009 at 09:57 from IEEE Xplore. Restrictions apply.