Available online at www.sciencedirect.com Sensors and Actuators B 131 (2008) 230–235 Characterisation of humidity dependence of a metal oxide semiconductor sensor array using partial least squares Jae Ho Sohn a,∗ , Michael Atzeni a , Les Zeller a , Giovanni Pioggia b a Sustainable Intensive Systems, Department of Primary Industries and Fisheries, 203 Tor Street, Queensland Government 4350, Australia b Interdepartmental Research Centre “E. Piaggio”, University of Pisa, Italy Received 7 September 2007; received in revised form 7 November 2007; accepted 7 November 2007 Available online 21 November 2007 Abstract Metal oxide semiconductor (MOS) sensors are a class of chemical sensors that have potential for being a practical core sensor module for an electronic nose system in various environmental monitoring applications. However, the responses of these sensors may be affected by changes in humidity and this must be taken into consideration when developing calibration models. This paper characterises the humidity dependence of a sensor array which consists of 12 MOS sensors. The results were used to develop calibration models using partial least squares (PLS). Effects of humidity on the response of the sensor array and predictive ability of partial least squares are discussed. It is shown that partial least squares can provide proper calibration models to compensate for effects caused by changes in humidity. Crown Copyright © 2007 Published by Elsevier B.V. All rights reserved. Keywords: Partial least squares; Metal oxide semiconductor sensor; Calibration; Electronic nose 1. Introduction An electronic nose system largely depends on an array of chemical, organic and optical sensors which collect chemi- cal data from the odorous air at the headspace of a sample. When appropriate chemical sensors are exposed to a sample, each sensor produces a characteristic response dependent upon the chemical interactions between the sample and the sensor. The data collected from the sensor array for a particular sam- ple can be interpreted as a pattern of responses, or fingerprint of that sample. Differences in the patterns can be correlated with differences in perceived sample odour. Samples with sim- ilar odours generally give rise to similar patterns, and samples with different odours show differences in their patterns. Using automated pattern recognition algorithms, patterns of differ- ent samples can be compared. A library of patterns can be gathered and stored in a computer database, such that data from test samples can be compared to the library and classified [1]. ∗ Corresponding author. Tel.: +61 7 4688 1117; fax: +61 7 4688 1192. E-mail address: jaeho.sohn@dpi.qld.gov.au (J.H. Sohn). A variety of different sensor technologies are used in sensor array systems [2]. Some of the most common are metal oxide semiconductor (MOS) sensors, conducting polymers (CP), bulk acoustic wave (BAW), surface acoustic wave (SAW) sensors , metal oxide semiconductor field effect transistor (MOSFET) and carbon black polymer sensors (CBPS). Such sensors can be divided into two main classes: hot (MOS, MOSFET) and cold (CP, SAW, BAW, CBPS). The former operate at high tem- peratures and are considered to be less sensitive to moisture with less carry-over residuals from one measurement to another [3]. Most studies on electronic nose systems published so far are based on MOS, and organic CP sensors [4]. These two tech- nologies rely on changes in resistance (or conductance) due to adsorption of gases. When a voltage is applied across the electrodes, a current passes through the metal oxide or poly- mer surface. The interaction of volatile compounds with the surface alters the electron flow in the system, and hence, the resistance of the sensor. The mechanisms are however very dif- ferent. The MOS sensors, working at high temperature, typically 200–600 ◦ C, are based on a combustion principle [4]. Most MOS sensors are made of tin oxide (SnO 2 ). MOS gas sensors were first demonstrated in the early 1960s. Since 0925-4005/$ – see front matter. Crown Copyright © 2007 Published by Elsevier B.V. All rights reserved. doi:10.1016/j.snb.2007.11.009