CPCQ 2015, 31 Mars - 02 Avril 2015
MODELLING OF SMART THIN FILM THERMAL-CONDUCTIVITY HUMIDITY SENSOR USING ANN
Karim Ferroudjia, b*, Nabil Benoudjitb, and Fayçal Medjilia
a Thin Films and Application Unit (U.D.C.M.A)-Sétif . Welding and NDT Research Centre (CSC), BP 64 Cheraga, Algeria
b Laboratoire d’Automatique Avancée et d’Analyse des Systèmes, Université de Batna, Algeria
* email : [email protected]
Abstract
The aim of this paper is to achieve a modeling design of a smart thermal-conductivity based humidity sensor (THS) using neuronal networks. Since the variation in temperature can strongly affect the sensitivity of the humidity sensor leading to a nonlinear response of the applied humidity, our mission is twofold: (i) accurately express the sensor using Radial Basis Function Neuronal Network (RBFNN); (ii) propose an element of correction whose goal is to correct the nonlinear response and eliminate the temperature effect. The paper proposes a smart sensor which incorporates Artificial Intelligence into the thermal-conductivity based humidity sensor, it exploits the RBFNN capability to provide compensation and self-calibration which is verified by simulations results.
Keywords : Humidity Sensor, Thermal-conductivity, Artificial intelligence, Neuronal Network, ANN modeling, Corrector model, RBFNN.
1. Introduction
Humidity sensors based CMOS technology has seen an explosion of scientific interest in the past two decades [1]. This paper deals with thermal conductivity based humidity sensor THS. We investigated the possibility of modeling a THS and its corrector using Radial Basis Function Neuronal Network (RBFNN). The goal is to: (i) accurately express the sensor using RBFNN, (ii) incorporate an element of correction in order to correct the nonlinear response and eliminate the temperature effect to provide accurate readout of the applied humidity. Thus, humidity readout is carried out using two RBFNNS. The first RBFNN provides calibrated response characteristics. The second RBFNN provides accurate humidity readout. The evaluation was performed using hold-out-set cross validation technique to avoid overfitting and assure statistical validity of the results [2].
2. Thermal-Conductivity Humidity Sensor
A cross-sectional view of the thermal conductivity- based humidity sensor structure is shown in Fig. 1. It contains two diodes: One acts as the sensor diode, while the other acts as the reference diode [3]. Fig. 2 illustrates the output voltage the humidity sensitivity of the THS. The output voltage measured as the humidity was increased from 20 % R.H to 90
% R.H at constant temperatures of 20°C, 30°C, and 40°C [3].
3. Radial-Basis Function Neural Network Model
Radial-Basis Function Neural Networks can be used for a wide range of applications mainly due to the fact they can approximate any regular function and their training is faster compared to the multilayer perceptron (MLP) [4-6]. The input nodes in the input layer are equal to the dimension of the input vector. The optimal number of neurons as well as the spread of the RBF gaussian are determined using cross validation technique.
Fig. 1: Cross-sectional view of the thermal- conductivity-based humidity sensor.
Fig. 2 :Outputs of the humidity sensor for temperature: 20 °C, 30 °C, and 40 °C.
1
The general block diagram of the proposed ANN model is shown in Fig. 3. The input data are first collected using the experimental results [3]. The dataset is arranged as (H, T, Vdz), where H is the humidity applied to THS, T is the temperature, and Vdz is the THS response. After that, we randomly divide the dataset into three subsets training set, validation set, and test set. The test set is used only for the assessment of the model selected by the cross-validation technique, while the validation set is used to tune the RBFNN parameters.
The comparison between the experimental dataset and that obtained after the training indicates that our model expresses accurately the response variation of the THS. Fig. 4 illustrats the model performances obtained at fixed temperatures 20°C, 30°C, and 40°C. Another measure of how well the neural network has fit the data is the regression plot (refer to Fig.5). Here the regression is plotted across Test samples. The regression plot shows the actual network outputs plotted in terms of the associated target values. If the network has learned to fit the data well, the linear fit to this output- target relationship should closely intersect the bottom-left and top-right corners of the plot. The R value is an indication of the relationship between the outputs and targets. If R = 1, this indicates that there is an exact linear relationship between outputs and targets. If R is close to zero, then there is no linear relationship between outputs and targets.
4. CORRECTOR Model
We propose an element of correction in order to correct the nonlinear response and eliminate the temperature effect. Fig. 6 shows general block diagram of the proposed smart humidity sensor. The CORRECTOR model may be used in cascade with the humidity sensor to
compensate for the adverse effects on the THS output due to the nonlinear response characteristics and the variations with ambient temperature.
RBFNN model (THS) T
H CORRECTOR
Vdz VH
Fig. 6: Block diagram of the proposed smart humidity sensor.
Fig. 3 : Block diagram of the proposed RBFNN model.
Testing data Training
data
Selected model
Training the selected model
Output Results Training different
models
Training stage
Testing stage
Appropriate parameters
Input data
10 20 30 40 50 60 70 80 90 100
2.5 3 3.5 4 4.5 5
Humidity %
Output voltage [V]
Experimental at 20°C RBF model at 20°C Experimental at 30°C RBF model at 30°C Experimental at 40°C RBF model at 40°C
3 3.5 4 4.5
3 3.5 4 4.5
Target
Output ~= 1*Target + -0.0012
R=0.99996
Data Fit Y = T
Fig. 4 : RBFNN model performance obtained at fixed temperatures 20, 30 and 40°C.
Fig. 5 : Plot linear regression between the network outputs and the corresponding targets.
2
5. Results & Discussion
Fig. 7 shows the delivered voltage Vdz at the THS output and the delivered voltage VH at the smart sensor output. The simulations were carried out at fixed humidity of 40% and 80%, when temperature is varying within the range of 20 to 40°C. The results show that our proposed smart sensor eliminates the temperature effect in comparison to the THS response where the temperature affect strongly the sensitivity of the humidity sensor leading to a nonlinear response of the applied humidity.
This can be explained by the fact that incorporating an element of correction can correct the nonlinear response and eliminate the temperature effect and provide accurate readout of the applied humidity.
6. Conclusion
In this paper we propose a smart humidity sensor which incorporates Artificial Intelligence into the thermal-conductivity based humidity sensor. We exploit the RBFNN capability to provide compensation and self-calibration by incorporating an element of correction in order to eliminate the temperature effect. The simulation results illustrate that the proposed smart sensor performs quite satisfactorily irrespective of any change in temperature. The proposed ANN model basically consists of two RBFNNs. The first RBFNN is used to provide the estimated response characteristics and its dependence on temperature. The second RBFNN (CORRECTOR) provides the accurate humidity readout and compensate the nonlinearity in sensor characteristics due to the temperature effect. Thus the proposed model can accurately estimate the humidity over a range of temperature 20°C to 40°C.
References
[1] Silverthorne, S.V.; Watson, C.W.; Baxter, R.D. Integrated Relative Humidity Sensor, Technique Digest. In Proceedings of IEEE Solid-state Sensor and Actuator Workshop, Hilton Head Island, SC, USA, 6–9 June,(1998), pp. 67-71.
[2] Breiman, L., and Spector, P. "Submodel selection and evaluation in regression: The x-random case". International Statistical Review 60, (1992), pp.291–319.
[3] Burak Okcan, Tayfun Akin, A Low-Power Robust Humidity Sensor in a Standard CMOS Process, IEEE Transactions on Electron Devices, Vol. 54, (2007), pp. 3071-3078.
[4] S. Haykin, "Neural networks: a comprehensive foundation," 1999.
[5] R.O. Duda, P.E. Hart, and D.G. Stork. Pattern Classification. John Wiley & Sons, New York, 2 edition, 2001.
[6] N. Benoudjit, K. Ferroudji, M. Bahaz, A. Bouakaz; "In vitro microemboli classification using neural network models and RF signals," Ultrasonics 51, (2011), pp. 247–252.
20 22 24 26 28 30 32 34 36 38 40
2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4
Temperature [°C]
Output voltage [V]
Humidity=40%
Vdz
VH CORRECTOR
20 22 24 26 28 30 32 34 36 38 40
2 2.5 3 3.5 4 4.5 5 5.5
Temperature [°C]
Output voltage [V]
Humidity=80%
Vdz
VH CORRECTOR
(a) (b)
Fig. 7 : Temperature effect on the response of the THS and the CRRECTOR at fixed humidity (a) 40% (b) 80%.
3
