Table 5.
Methods for suppressing drift of sensors.
| Ref No. | Author (year) | Gas Sensors/e-Nose System | Sampling Type | Gases to Be Detected | Interference Source | Data Processing Method | Effects |
|---|---|---|---|---|---|---|---|
| [53] | T. Artursson, et al. (2000) | The e-nose contains two arrays of 10 MOSFET sensors, and two arrays of MOS sensors containing 10 and 9 sensors respectively. | Pump suction | Hydrogen, ammonia, ethanol, ethene | Sensor drift | PCA, PLS, CC | Root mean square errors: less than 10 ppm. |
| [54] | M. Padilla, et al. (2010) | 17 conductive polymer gas sensors | Pump suction | Ammonia, propanoic acid and n-butanol | Sensor drift | OSC, CC | OSC is a suitable method for drift correction in a longer time. |
| [55] | A.C. Romain, et al. (2010) | TGS822, TGS880, TGS842, TGS2610, TGS2620, TGS2180 | Pump suction | Print house odor and compost odor | Sensor drift | Multivariate array correction, univariate sensor correction, signal pre-processing | F criterion: 33; 56, 26, 18 for: no correction, correction by univariate multiplicative factor, correction by PLS: correction by PCA. |
| [56] | L. Zhang, et al. (2013) | TGS2602, TGS2620, TGS2201 | Diffusion sampling | HCHO, CO, C6H6, C7H8, NH3, NO2 | Sensor drift | PSR and RBF neural network | RMSEP: less than 0.005 |
| [57] | L. Zhang, et al. (2016) | TGS2602, TGS2620, TGS2201 | Diffusion sampling | HCHO, CO, C6H6, C7H8, NH3, NO2 | Sensor drift | ARMA and Kalman filter models | RMSEP: TGS2602: 0.004, TGS2201A: 0.0039, TGS2201B: 0.0134 |
| [58] | M. Holmberg et al. (1997) | 10 MOSFET, 4 Tagu-chi and 1 CO2 monitor | Pump suction | 1-propanol, 2-propanol, 1-butanol, 2-butanol | Sensor drift | Adaptive estimation algorithm, recursive least squares algorithm | Recognition rate: static model: 85%; recursive model 91%. |
| [59] | S. Marco, et al. (1997) | TGS822, TGS813, TGS815, TGS812a, TGS812b, TGS812 | Pump suction | H2, CO, CO2 and CH4 | Sensor drift | Adaptive SOM | Recognition rate: higher than 97%. |
| [60] | M. Zuppa, et al. (2004) | 32 conducting polymer gas sensors (A32S) | Pump suction | Acetonitrile, methanol, propanol, acetone and butanol | Sensor drift | mSOM neural network | Recognition rate: 97.2%. |
| [61] | S. De Vito, et al. (2012) | Five semiconductors | Pump suction | Six single coffee varieties and 8 blends | Sensor drift | A boosting-like approach to semi- supervised learning (SSL) | Recognition rate: higher than 92.5%. |
| [61] | S. De Vito, et al. (2012) | 5 Metal Oxides (MOX) sensors, temperature and Relative Humidity (RH) sensors | Pump suction | CO, Benzene, NMHC, NOx, NO2 | Sensor drift | SSL, SSL-based adaptive strategy | Mean absolute error: performance gain of 11.5%. |
| [62] | Q. Liu, et al. (2014) | 16 metal-oxide gas sensors | Pump suction | Acetone, acetaldehyde ethanol, ethylene, ammonia, toluene | Sensor drift | comgfk (combination of weighted geodesic flow kernels), comgfk-ml (comgfk with manifold regularization) | The comgfk-ml can effectively handle sensor drift. |
| [63] | L. Zhang, et al. (2015) | 16 metal-oxide gas sensors | Pump suction | Acetone, acetaldehyde ethanol, ethylene, ammonia and toluene | Sensor drift | DAELM | Average recognition rate: Setting 1: 91.86%; Setting 2: 91.82%. |
| [48] | K. Yan and D. Zhang. (2016) | 16 metal-oxide gas sensors | Pump suction | Acetone, acetaldehyde ethanol, ethylene, ammonia and toluene | Sensor drift | TCTL, SEMI | Average recognition rate: 87.6. |
| [64] | K. Yan and D. Zhang. (2016) | 16 metal-oxide gas sensors | Pump suction | Acetone, acetaldehyde ethanol, ethylene, ammonia and toluene | Sensor drift | DCAE, the basic DCAE (DCAE-basic), DCAE with correction layer (DCAE-CL) | Average recognition rate: DCAE-basic: 92.59%±0.61; DCAE-CL: 93.21%±0.52 |
| TGS4161, TGS822, TGS826, WSP2111, SP3S-AQ2, GSBT11, TGS2610-D00, TGS2600-TM, TGS2602-TM, WSP2111-TM, HTG3515CH | Pump suction | Diabetes, chronical kidney disease, cardiopathy, lung cancer, breast cancer | Sensor drift | DCAE, the basic DCAE (DCAE-basic), DCAE with correction layer (DCAE-CL) | Average recognition rate: DCAE-basic: 81.84% ± 0.67; DCAE-CL: 84.13 ± 0.82. | ||
| [65] | S. AlMaskari, et al. (2014) | 16 metal-oxide gas sensors | Pump suction | Acetone, acetaldehyde ethanol, ethylene, ammonia and toluene | Sensor drift | Kernel fuzzy C-means clustering and KFSVM | Average recognition rate: 82.18% |
| [66] | A. Vergara, et al. (2012) | 16 metal-oxide gas sensors | Pump suction | Acetone, acetaldehyde ethanol, ethylene, ammonia and toluene | Sensor drift | Ensemble of classifiers (weighted combination of SVM) | The classifier ensembles were better than baseline classifiers. |