Signal and data processing are essential elements in electronic noses as well as in most chemical sensing instruments. The multivariate responses obtained by chemical sensor arrays require signal and data processing to carry out the fundamental tasks of odor identification (classification), concentration estimation (regression) and grouping of similar odors (clustering). In the last decade, important advances have shown that proper processing can improve the robustness of the instruments against diverse perturbations, namely, environmental variables, background changes, drift, etc. This article reviews the advances made in recent years in signal and data processing for machine olfaction and chemical sensing. Figure 3. Evolution of the pattern of response of a SB31 sensor from FIS Japan when the heated voltage is scanned due to changes in the gas composition (mixture of CO and CH4).One of the problems that affect sensor array instruments is the lack of reproducibility in the sensor signals when technologically identical devices are exposed to a chemical under the exact same conditions. Figure 4 shows the voltage waveforms of 11 temperature modulated Metal Oxide Sensors. It is obvious that the pattern of response is slightly different for both analytes, but the waveform scatter is also remarkable, which poses problems during sensor replacement and calibration transfer scenarios (see section 3).

Inherent variability of chemical sensors makes it necessary to calibrate chemical detection systems individually. This shortcoming has traditionally limited usability of systems based on Metal Oxide gas sensor arrays and prevented mass-production for some applications. Here, aiming at exploring calibration transfer between chemical sensor arrays, we exposed five twin 8-sensor detection units to different concentration levels of Ethanol, Ethylene, CO, or Methane. First, we built calibration models using data acquired with a master unit. Second, to explore the transferability of the calibration models, we used Direct Standardization to map the signals of a slave unit to the space of the master unit in calibration. In particular, we evaluated the transferability of the calibration models to other detection units, and within the same unit measuring days apart. Our results show that signals acquired with one unit can be successfully mapped to the space of a reference unit. Hence, calibration models trained with a master unit can be extended to slave units using a reduced number of transfer samples, diminishing thereby calibration costs. Similarly, signals of a sensing unit can be transformed to match sensor behavior in the past to mitigate drift effects. Therefore, the proposed methodology can reduce calibration costs in mass-production and delay recalibrations due to sensor aging. Acquired dataset is made publicly available.Only recently calibration transfer techniques have been used in regression tasks. In con-54 trast to classification tasks, regression is a more challenging problem, but also offers a more 55 sensitive measure of the quality of the calibration transfer. Lei Zhang et al. presented a 56 methodology for on-line calibration transfer [22]. They built six twin units: a master unit 57 and five slave units. Each unit was composed of four MOX gas sensors along with tempera-58 ture and humidity sensors. They fit univariate linear regression curves between each of the 59 slave units and the master unit to transform the signals acquired with slave units to the space 60of the master unit. Although the units were exposed to formaldehyde, benzene and toluene, 61 3 only the former was used as reference for calibration transfer. Their results show that a sim-62 ple homogeneous linear transformation provides good signal mapping between sensing units. 63In another study by Deshmukh et al., the authors proposed calibration transfer between two 64 chemical sensor arrays by means of box-behnken design and robust regression [23]. Two 65 twin systems with six MOX gas sensors each were built and tested simultaneously. Artificial 66 neural network models were built with the master unit to predict the concentration of four 67 compounds relevant for the paper industry: hydrogen sulfide, methyl mercaptan, dimethyl 68 disulphide, and dimethyl sulphide. The authors showed that the calibration model developed 69 for the master system, built upon 100 calibration samples, can be transferred to the slave unit 70 using a smaller s...

No abstract

Shifts in working temperature are an important issue that prevents the successful transfer of calibration models from one chemical instrument to another. This effect is of special relevance when working with gas sensor arrays modulated in temperature. In this paper, we study the use of multivariate techniques to transfer the calibration model from a temperature modulated gas sensor array to another when a global change of temperature occurs. To do so, we built 12 identical master sensor arrays composed of three different types of commercial Figaro sensors and acquired a dataset of sensor responses to three pure substances (ethanol, acetone and butanone) dosed at 7 concentrations. The master arrays are then shifted in temperature (from-50 to 50°C, T = 10°C) and considered as slave arrays. Data correction is performed for an increasing number of transfer samples with 4 different calibration transfer techniques: Direct Standardization, Piece-wise Direct Standardization, Orthogonal Signal Correction and Generalized Least Squares Weighting. In order to evaluate the performance of the calibration transfer, we compare the Root Mean Square Error of Prediction (RMSEP) of master and slave arrays, for each instrument correction. Best results are obtained from Piece-wise Direct standardization, which exhibits the lower RMSEP values after correction for the smaller number of transfer samples.