Short Paper Proc. of Int. Colloquiums on Computer Electronics Electrical Mechanical and Civil 2011
An Improved Smart Pressure Measuring Technique Santhosh K V1, B K Roy2 1
Department of Electrical Engineering, National Institute of Technology, Silchar, India Email: kv.santhu@gmail.com 2 Department of Electrical Engineering, National Institute of Technology, Silchar, India Email: bkr_nits@yahoo.co.in aggravates the situation when there is change in environmental conditions. Since the output of a CPS is dependent on applied pressure as well as temperature, when the ambient temperature changes frequently, the situation becomes very complicated [1], [2]. In [1], linearization of CPS was reported using ANN. The output of CPS is made independent of measurement temperature along with linearization of CPS and was reported in [2]. To overcome the above difficulties, an improved smart pressure measuring technique been proposed in this paper which uses artificial neural network to train the system to obtain linearity and make the output independent of thickness, ME of diaphragm and the temperature variation. The paper is organised as follows: after introduction in Section-I, a brief description on CPS model is given in SectionII. The output of the CPS is capacitance; a brief discussion on data conversion i.e. a timer circuit and a Frequency to voltage converter is discussed in Section-III. Section-IV deals with the problem statement followed by proposed solution in Section-V Finally, result and conclusion is given in SectionVI.
Abstract— This paper aims to design a smart pressure measuring instrument. The objectives of this work is to eliminate the nonlinearity in the Capacitance Pressure Sensor (CPS), the interference of Modulus of Elasticity (ME) and the thickness of diaphragm used in the CPS and to compensate the temperature effect on the output of the instrument. The capacitance output of the CPS is converted to frequency with the timer circuit followed by a frequency to voltage converter which converts the available frequency to voltage. An Artificial Neural Network (ANN) block is added in cascade to Frequency to Voltage converter. This arrangement helps to linearise the overall system and make it independent of ME, the thickness of diaphragm used in the CPS and the effect of temperature. Since the proposed pressure measuring instrument produces output independent of physical properties of diaphragm and effect of temperature, thus the present work avoid the requirement of repeated calibration every time the diaphragm is replaced or there is a change in temperature. An improvement of the similar earlier reported work is proposed in the present paper by designing the pressure measuring system independent of temperature change also. Index Terms — Artificial neural networks, Capacitance Pressure Sensor, Sensor Modelling, Temperature compensation.
II. CAPACITANCE PRESSURE SENSOR
I. INTRODUCTION
CPS uses a thin diaphragm, usually metal or metal-coated quartz, as one plate of a capacitor. The diaphragm is exposed to the process pressure on one side. Changes in pressure cause it to deflect and change the capacitance which is proportional to the applied pressure. Fig 1 shows the model of the CPS [2], [3], in press [11].
With the steam age came the demand for pressure measuring instruments. Bourdon tubes or bellows, where mechanical displacements were transferred to an indicating pointer were the first pressure instruments, and are still in use today. Pressure metrology is the technology of transducing pressure into an electrical quantity. Normally, a diaphragm construction is used with strain gauges either bonded to, or diffused into it, acting as resistive elements. Under the pressure-induced strain, the resistive values change. In capacitive technology, the pressure diaphragm is one plate of a capacitor that changes its value since pressureinduced displacement. Among all, capacitive technologies play an increasingly important role in the fields of industrial and automotive sensors because of its low power consumption and high sensitivity. However, its highly nonlinear response characteristics give rise to several difficulties. To overcome the difficulties faced due to the nonlinear response characteristics of the CPS, dependency of output on ME and thickness of the diaphragm, several techniques have been suggested, but these are tedious and time consuming. Further the process of calibration needs to be repeated everytime the wear out diaphragm is replaced. The problem of nonlinear response characteristics of a CPS further © 2011 ACEEE DOI: 02.CEMC.2011.01. 515
Figure 1. Capacitance Pressure Sensor
Assuming a circular diaphragm, the elastic deflection ‘h’ at distance ‘r’ from the centre of the diaphragm restrained around its circumference under pressure ‘P’ is given by
m 109
(1)
Short Paper Proc. of Int. Colloquiums on Computer Electronics Electrical Mechanical and Civil 2011
Variation of Modulus of Elasticity with temperature [5] can be given by (2) (2) Where E (t) = Modulus of Elasticity at temperature to C E (25) =Modulus of Elasticity at temperature 25oC t = Temperature in o C
Figure 3. 555 as an Astable Multivibrator
Using (2) in (1), the equation for the deformation caused due to the pressure applied can be found under different temperatures. The effective capacitance of the chamber may be expressed as
F
Frequency to voltage converter (F-V) is a circuit which converts the given frequency into voltage and is given by Vout = (VS * C1 * R1) * f
Volts
(7)
Here LM331 is wired as a frequency to voltage converter [7] as shown in Fig 4.
(3)
By using the above equations (1) and (3), the capacitance can be found [1], [3], [4] as
Capacitance is also a function of temperature [6], which can be given by the (5) C (t) = C (to) (1+á (t–to) + â (t-to)2) F (5) o Where C (t) = capacitance at temperature t C C (to) = capacitance at temperature to o C á, â = constants Using (4) and (5) the effective capacitance change for the pressure applied for different values of temperature can be found.
Figure 4. LM331 Frequency to voltage converter
III. DATA CONVERSION CIRCUIT
IV. PROBLEM STATEMENT
The block diagram representation of the proposed instrument is given in Fig 2.
In this section characteristic of CPS is simulated to understand the difficulties associated with the available measuring scheme. For this purpose, simulation is carried out with two different thicknesses of diaphragm. These are d1 = 5 mm and d2 = 6 mm. Further two different ME’s of diaphragm are considered. These are ME1 = 70 GPa and ME2 = 150 GPa. For different temperatures like t1 = 25oC, t2 = 50o C, t3 = 75oC and t4 = 100oC are used to find the output capacitance of CPS with respect to various values of input pressure considering a particular thickness, ME of diaphragm and temperature. These output capacitances are used as input of timer circuit and by using (6) intermittent output frequencies are generated. Finally voltage signals are produced by using (7). The MATLAB environment is used of and the following characteristics are found
Figure 2. Block diagram
Timer circuit consist of a 555 IC connected in astable mode [12] as shown in Fig 3. This circuit generates a train of pulses whose frequency is given by.
f=
Hz
(6)
From (6) it’s clear that the frequency of the timer depends on the capacitance (CPS). Now the input Pressure applied on the CPS is converted to frequency. This frequency is further converted to voltage. © 2011 ACEEE DOI: 02.CEMC.2011.01. 515
110
Short Paper Proc. of Int. Colloquiums on Computer Electronics Electrical Mechanical and Civil 2011
Figure 8. Frequency to voltage converter outputs for variation of pressure with diaphragm thickness of 6 mm and ME 150GPa (at 25 o C).
Figure 5. Frequency to voltage converter outputs for variation of pressure with diaphragm thickness of 5 mm and ME 70GPa (at 25 o C).
Fig 5, Fig 6, Fig 7 and Fig 8 shows the variation of voltage with the change in input pressure considering different values of thickness, ME of diaphragm and temperature. It has been observed from the above graphs (Fig 5, Fig 6, Fig 7 and Fig 8) that the output from the frequency to voltage converter circuit has a non linear relation. Datasheet of CPS suggests that the input range of 10% to 60% of full scale is used in practice as linear range. The output voltage also varies with the change in thickness, ME of diaphragm and temperature. These are the reasons which have made the user to go for calibration techniques using some circuits. These conventional techniques have a drawback that its time consuming and need to be calibrated every time a diaphragm is changed in the system, variation of environment conditions like temperature and the use is restricted only to a portion of full scale. To overcome these drawbacks, this paper makes an attempt to design a pressure measuring technique incorporating intelligence to produce linear output and to make the system independent of thickness, ME of diaphragm and temperature using the concept of artificial neural network.
Figure 6. Frequency to voltage converter outputs for variation of pressure with diaphragm thickness of 6 mm and ME 70GPa (at 25 o C).
V. PROBLEM SOLUTION The drawbacks discussed in the earlier section are overcome by introducing an Artificial Neural Network (ANN) model is cascaded after frequency to voltage converter. This model is designed using the neural network toolbox of MATLAB. The first step in developing a neural network is to create a database. The output voltage of the system for the change in pressure, thickness, ME of diaphragm and temperature is stored in one matrix; which forms the input matrix for the ANN model. The output matrix would be the target matrix consisting of data having a linear relation with the pressure and independent of all other parameters.
Figure 7. Frequency to voltage converter outputs for variation of pressure with diaphragm thickness of 5 mm and ME 150GPa (at 25 o C).
Š 2011 ACEEE DOI: 02.CEMC.2011.01. 515
111
Short Paper Proc. of Int. Colloquiums on Computer Electronics Electrical Mechanical and Civil 2011 First the data is initialized: training base (65%), test base (15%), validation base (20%), number of layers and neurons, type of the transfer functions, number of iteration and estimate threshold. The neural network is configured with the initialization as shown in table 1 now the weights of the network are varied to match the target. Mean Squared Error (MSE) is the average squared difference between outputs and targets. Lower values are better. Zero means no error. Regression (R) values measure the correlation between outputs and targets. An R value of 1 means a close relationship, 0 a random relationship. With these details the network is trained to achieve its desired target. The following are the details of the neural network.
A: Structure Of Neural Network Model:
Figure 9.Neural Network Architecture
The Fig. 9 shows the architecture of ANN model so found for the optimal system condition. The ANN model considered here is Multilayer Perceptron (MLP), having an input layer, output layer and 6 hidden layers, with each of the hidden layer consisting of 7, 7, 8, 8, 9 and 7 neurons respectively.
TABLE I
SUMMARIZES THE REQUIRE DATA FOR TRAINING.
B: Training With the help of simulated data the neural network is trained with architecture as shown in Fig 9. The process of varying the weights to achieve the output is called training. The neural network algorithm uses back propagation neural network trained by ant colony optimization [8], [9], [10] To satisfy the linearity property between input pressure and output voltage signal, the target graph is considered as shown in Fig 10.
Figure 10. Target graph
C: Algorithm For Trainng The pseudo code of the algorithm can be given by: Initialize the weights in the network (randomly) Do For each data_pair in the training set O = neural-net-output (network, data); forward pass T = target for input data Calculate, error = (T - O) at the output units Compute delta_wh for all weights from output layer to hidden layer; backward pass Repeat forward and backward pass Update the weights in the network; ACO_MetaHeuristic aco: procedure ACO_MetaHeuristic while (not_termination) generateSolutions() daemonActions() pheromoneUpdate() end while end procedure Return the network Š 2011 ACEEE DOI: 02.CEMC.2011.01. 515
VI. RESULT AND CONCLUSION The proposed ANN is trained, validated and tested with the simulated data. Once the training is over, the system with CPS along with other modules in cascade as shown in Fig 2. is subjected to various test inputs corresponding to different pressure at a particular ME, thickness of the diaphragm and temperature all within the specified range. For testing 112
Short Paper Proc. of Int. Colloquiums on Computer Electronics Electrical Mechanical and Civil 2011
purposes the range of pressure is considered from 0 to 2.5 MPa, the range of ME is 70 GPa to 150 GPa, range of thickness is 5 mm to 6 mm and temperature ranges 0oC to 100oC. The outputs of system with ANN are noted corresponding to various input pressure at different values of ME, thickness of diaphragm and temperature. The input output result is plotted and is shown in Fig 11. The output graph is matching the target graph as shown in Fig 10.
Measurement noise is not considered in the present work. Performance of proposed measuring technique in presence of measurement noise will be taken up in future. An embedded system will be attempted incorporating the design technique to make if suitable for practical application. REFERNCES [1] Jagdish Chandra Patra, Alex C. Kot, Ganapati Panda, “An Intelligent Pressure Sensor Using Neural Networks”. IEEE transactions on instrumentation and measurement, vol.49, pp 829-834, August 2000. [2] Jagdish C. Patra, Ganapati Panda, “ANN-based intelligent pressure sensor in noisy environment”. IEEE Measurement, vol. 23, pp 229-238, December. 1998. [3] Lyons, J. L., “The Designer’s Handbook of PI-Pressure-Sensing Devices”, Van Nostrand Reinhold, New York (1980) [4] Neubert, H. K. P. “Instrument Transducers: an Introduction to their Performance and Design”, Clarendon Press, Oxford (1975) [5] AS4100 Fire Provision “ANSI Standards” [6] N Krotkov and M D Klionskii, “Evaluation of the temperature characteristic of reference capacitors”. Springer. pp 52-56 September 1996. [7] LM331 Datasheet “National Semiconductor Corporation” 2006. [8] Jeng-Bin Li, Yun-Kung Chung, “A Novel Back propagation Neural Network Training Algorithm Designed by an Ant Colony Optimization” IEEE/PES Transmission and Distribution Conference & Exhibition: Asia and Pacific Dalian, China 2005 [9] L. Bianchi, L.M. Gambardella, M.Dorigo “An ant colony optimization approach to the probabilistic travelling salesman problem” . In Proceedings of PPSN-VII, Seventh Inter17 national Conference on Parallel Problem Solving from Nature, Lecture Notes in Computer Science. Springer Verlag, Berlin, Germany, 2002 [10] Stuart Russell and Peter Norvig. “Artificial Intelligence A Modern Approach” [11] Santhosh K V, B K Roy, “An Intelligent Pressure Measuring Instrument” accepted for International conference ICIC 11, conducted by PSG Technology, Coimbatore, India. “in press” [12] IC 555 datasheet “National Semiconductor Corporation” 2006.
Figure 11. Response of the system
It is evident from the Fig 11, that the proposed measuring system discussed has incorporated intelligence to the CPS; it has increased the linearity range of the CPS. Also the output is independent of the ME, thickness of diaphragm and temperature. Thus if the diaphragm is replaced by another diaphragm having different thickness and ME, the system does not require any calibration, Similarly if there is a change in environment conditions like change in temperature the system need not be calibrated to give the accurate reading. The present paper is compared with the similar reported works in [1],[2]. In the present paper, output of CPS is made independent of thickness and ME of diaphragm which is substantive improvement over the earlier works.
© 2011 ACEEE DOI: 02.CEMC.2011.01. 515
113