Source Direction Detection based on Stationary Electronic Nose System

Transcription

1 World Academy of Science, Engineering and Technology 5 8 Source Direction Detection based on Stationary Electronic Nose System Jie Cai, and David C. Levy Abstract Electronic nose (array of chemical sensors) are widely used in food industry and ollution control. Also it could be used to locate or detect the direction of the source of emission odors. Usually this task is erformed by electronic nose (ENose) cooerated with mobile vehicles, but when a source is instantaneous or surrounding is hard for vehicles to reach, roblem occurs. Thus a method for stationary ENose to detect the direction of the source and locate the source will be required. A novel method which uses the ratio between the resonses of different sensors as a discriminant to determine the direction of source in natural wind surroundings is resented in this aer. The result shows that the method is accurate and easily to be imlemented. This method could be also used in movably, as an otimized algorithm for robot tracking source location. Keywords Electronic nose, Nature wind situation, Source direction detection. T I. INTRODUCTION ODAY electronic nose are widely used in food(beverage) quality industry, ollution control and fire detection [-8]. The techniques for signal rocessing in ENose, including signal restoration, feature extraction and classification, are very well develoed [9-5]. Additionally, those techniques are able to give some other measurable information, such as concentration, which could be used to locate the emission source [6] or give the direction where the smell comes. Obviously the direction could be derived when the source is located. Normally, source localization is erformed by the cooeration of electronic nose and mobile vehicles since robotic techniques are getting mature [7-9]. However roblem could occurs during the secific situations, such as instantaneous emission and comlex landscae environment. In these cases, mobile sensors could not erform the task roerly. Thus, an interesting for the stationary sensors to locate the source and detect the direction of source arises. There are few ublications is about source localization with stationary sensors. Jorg Matthes rovided an aroach to locate the source and direct the source with satially distributed electronic noses [6], but it is costly to construct a such electronic noses network. In addition, Jorg s aroach only considered normal diffusion and advection, that Jorg s aroach is not able to solve a crosswind or wind situation. In this aer, a method of stationary sensors to detect the direction of emission source within natural wind surroundings is resented. In section II, the whole exerimental ENose system is introduced. A new aroach is resented in section III, which will be divided into two asects: () determine the direction of source with given axis of my ENose system by analysis the resonse of sensors in advection situation; () in crosswind situations the direction of source detection. Break wind cases will be discussed in section IV. Exerimental and resulted is resented in section V, afterwards the conclusion is drawn in section VI. II. SYSTEM CONSTRUCTION My electronic nose system is build u by four sensors which are searated by a square imermeable searator with four wings. Fig. shows the structure of the system, cross wind direction and seed (V Cross ) could be resented as the sum of x direction vector (V x ) and y direction vector (V y ). Angle α is used to denote the direction of crosswind. The shae of sensors is considered as a circle with radius.5cm. The dimensions of searator are cm X cm for square and each wing has included angle 5 with length. To make an assumtion, we treat the source is laced on an imermeable z= surface, thus the whole system is reduced the dimensions from to. Position of emission dot source is unknown. Angle θ resents the included angle of the axis of ENose system and connection line of emission dot source and center of ENose system. The direction of emission dot source is unknown, which is denoted byθ. Manuscrit received Setember 5, 6. Jie Cai is PhD student in School of Electrical and Information Engineering, University of Sydney, NSW 6, Australia (hone: ; jcai@ee.usyd.edu.au). David C Levy is Associate rofessor and head of school, School of Electrical and Information Engineering, NSW 6, Australia (hone: ; dlevy@ee.usyd.edu.au). Fig. Electronic nose system structure 8

2 World Academy of Science, Engineering and Technology 5 8 III. METHOD In this section, the method to determine the source direction is given in two different cases, advection case and crosswind case. The solution for the source direction angle θ is given in both cases. In crosswind case, solution of θ could be derived from the advection case, θ is resented as a function of wind direction α and eak oint angleθ. 5% +.69 e θ θ θ f ( θ) =.6 e +.95 e () roortion= sensor()/sensor sensor/sensor /sensor A. Determine the Direction of Source to Center of ENose System in Advection A series of simulation is done to analysis the relation between source direction angleθ and sensors resonses. Fig. shows the relations between angleθ and sensors resonses, which are considered in the advection situation with V x wind, randomly icked 5m/s. Resonses Relation between angle and Resonses of sensors Fig. Relations between angleθ and sensors resonses sensor sensor In Fig., resonses of different sensors are resented with different legends. Sensor has highest resonse. Resonse of sensor and are axis symmetry by angle=8. Resonse of sensor could be ignored. Additionally, from Fig. we could see that the resonses of sensors are a eriodic function of angle with eriod of π, when source rate and distance is treated as constants. With the rearation simulation and analysis, the method is established. After analysis of diffusion-advection equation and simulations, we found that source rate and distance of the source will not affect the ratio of resonses of different sensors. First we define that C, C, C, C is the measured concentration by sensor to sensor. Fig. shows that the curve C/C and C/C. It generated from the raw data of Fig., we could not distinguish the value for sensors when the source direction is between and by eyes in Fig., but Fig. shows us clearly the relationshi of sensors resonse between and. We could easily find that C/C and C/C are axis symmetry by angle=8, also the curve could be curvefit by Matlab. For C/C, the fitted equations could be resent in Gauss equation format: ercentage % 5% % 5% % 5% % Fig. Ratio of C/C and C/C Thus, the angleθ could be determined by calculating the ratio C/C and C/C, then if C/C is smaller than C/C, θ would between o and 8 o, otherwise, θ would between 8 o and 6 o. θ = f ( θ ) () Equation () is the function to calculate angleθ, which is the solution for source direction. B. Crosswind Case Direction Detection In this case, the crosswind with V x and V y is alied. We randomly icked a set of seed for V x and V y, which is 5 and resectively. A series of simulation is done to analysis this case. Fig. shows the relations between angle θ and sensors resonses in the crosswind situation Relation between angle and sensor resonses for crosswind sensor sensor Fig. Relations between angleθ and sensors resonses for crosswind case 8

3 World Academy of Science, Engineering and Technology 5 8 We found that after a simle movement along angle axis to make the eak oint starting the curve and connect the cut-off art to the end of curve a new figures shows in Fig. 5. After comare Fig. 5 and Fig., we can find that the resonses of sensors has similar attern in both case. Additionally, the ratio C/C and C/C in crosswind case has similar attern as Gauss distribution, but we do not need to make any movement for the curve along angle axis. The ratio for sensor and sensor also has symmetry axis θ =8. Real line in Fig. 7 resents the ratio C/C in crosswind case. The solution for angleθ is an inverse solve roblem. To solve the inverse roblem, firstly we assume wind direction is known, and then we can find out the roosed angle to reach the eak oint Relation between angle and sensor resonses for crosswind after movement along angle axis sensor sensor Fig. 5 Relations between angleθ and sensors resonses for crosswind case after movement along angle axis Fig. 6 shows the diagram for calculation scheme for the angle of eak oint. P is the roosed source osition; O is center of ENose system (origin); T is the intersection oint of PT and edge of searator, Q is intersection oint of PT and x axis. PT is the tangent with crosswind direction of sensor, which has equation as equation (). As TO, OQ, QT is known, by simly calculate Euclidean distance d = X () X x X X Here X X could be relaced by TO, OQ, and QT. Thus, β andγ can be calculated by law of cosines. Also the source osition P(x,y) could be rewritten as P( x, x tanθ ). Point P is on a straight line which denoted by equation (), also Δ OPQ obey all of laws for triangle. A set of equations could be established which looks like following: x + tanα x tanθ = R + + (5) d PT + d PO dot Cos( α θ ) = d PT d PO The second equation in equation set (5) is based on law of cosines. Thus, angle for eak oint θ can be calculated. Also, the set of the equation are easily to be imlemented by comuter language. After analysis, the eak oint is occurred when the source osition is on a line with sloe of crosswind direction α and also this line is the tangent of sensor. The tangent equation for crosswind direction α of the sensor circle could easily be given by following equations: x + tanα y = R + + where R is the radius of sensor. () Fig. 7 Comarison of C/C for advection and crosswind situation We have error at SSE:.99, R-square:.998, Adjusted R-square:.998 and RMSE:.58, which are accetable Fig. 6 Calculation of angle for eak oint Fig. 7 shows the curve for both advection and crosswind situation; obviously they all accord with Gauss distribution and have similar shae. To let f (θ ) be the function for ratio of sensor s resonse to sensor s resonse in advection, the 8

4 World Academy of Science, Engineering and Technology 5 8 function fc(θ ) in crosswind situation has the relation with direction of crosswind α and eak oint angleθ. fc ( θ) = f ( θ ( α θ )) (6) cosα Equation (6) is a curve fit function; Fig. 8 shows the fit result. TABLE I RESPONSES OF SENSORS FOR BOTH WIND BREAK CASE AND WIND BREAK FREE CASE free case sensor sensor Fig. 8 Curve fit for fact crosswind case of C/C by f c(θ ) In this case θ = f c ( θ ) is the way to obtain the direction angle. Based on the two cases above, we could easily calculate the reverse equation () by comuter. After the analysis of wind case, the exerimental result is discussed. IV. NATURAL WIND CASES In the natural wind cases, we are mainly considering the wind case, which means that wind s in a eriod then recovers as before. Table I resents the relation of sensor resonse and time in 5 seconds in both wind free case and wind case, where between.7s and.s wind s. According to Table I, in both cases, resonses of sensors stay steady after seconds. Thus we could treat the resonse of sensors during wind as a noise. To ignore the noise a set of new resonses can be derived. After filtering the noise, simly cut off the resonse signal between.8s and.5s, sensors resonses in wind case almost same as them in wind free case. Thus for wind case, we could consider it as a wind free case, by analyzing a steady resonses of sensors. case case after filtering noise Resonse Resonse sensor sensor sensor sensor V. EXPERIMENTS AND RESULTS Exeriments are done in both advection and crosswind cases. We set initial concentration of the source is 5 m, advection wind seed V x =5m/s, crosswind wind seed V x =m/s, V y =.8m/s. Time is set to seconds. We give a set of oint which has direction angleθ,, 6,, 8, and. The source is then laced on these oints one by one, each time start the exeriment after the sensors resonse recovers to environment. The result of calculated direction is listed in Table II. The calculated relative error shows in Table III, for the result of direction angle we use range of 6 as the style to resent the result, which avoids the unary oerator. According to Table III, we have highest error at 6 for both advection and crosswind case, that is because the Gauss fitting equation has highest error round 6, and curve fitting of eq5 also has higher error between 5 and. However, 5 8

5 World Academy of Science, Engineering and Technology 5 8 the mean relative error for both advection and crosswind cases are accetable. TABLE II EXPERIMENTAL RESULT AND SOURCE DIRECTION Real direction Advection Result ( ) Crosswind Result ( ) Real direction TABLE III EXPERIMENTAL RESULT ERRORS Advection Relative Error 6. Crosswind Relative Error 6.%.8%.5%.% 8.%.%.%.85% 6.55%.7% 6 ( ).%.% Mean Error.6%.% VI. CONCLUSION In this aer, a novel direction detection method bases on single electronic nose system is resented in nature wind situations. The method successfully solves the source direction detection roblem in advection, crosswind and wind cases. The exerimental results show that this method is accurate. Additionally, the otimization of this method needs to be done in otimizing the fitting equations for both advection and crosswind situation. To be the continuing research target, the method could be use to locate the odour source and navigate the smell tracking robotics. REFERENCES [] F. J. Acevedo, S. Maldonado, E. Dominguez, A. Narvaez, and F. Loez, "Probabilistic suort vector machines for multi-class alcohol identification," Sensors and Actuators B: Chemical, vol. In Press, Corrected Proof. [] H. Yu and J. Wang, "Discrimination of LongJing green-tea grade by electronic nose," Sensors and Actuators B: Chemical, vol. In Press, Corrected Proof. [] G. A. Bell and A. J. Watson, Tastes & Aromas, the chemical senses in science and industry. Sydney: University of New South Wales Press, 999. [] Y. Peng, P. Min, C. Yuquan, and L. Guang, "The recognition of Chinese sirits using electronic nose with dynamic method," resented at Engineering in Medicine and Biology Society. Proceedings of the rd Annual International Conference of the IEEE,. [5] K. Brudzewski, S. Osowski, T. Markiewicz, and J. Ulaczyk, "Classification of gasoline with sulement of bio-roducts by means of an electronic nose and SVM neural network," Sensors and Actuators B: Chemical, vol.,. 5-, 6. [6] T. Sobanski, A. Szczurek, K. Nitsch, B. W. Licznerski, and W. Radwan, "Electronic nose alied to automotive fuel qualification," Sensors and Actuators B: Chemical, vol. 6,. 7-, 6. [7] E. Scorsone, A. M. Pisanelli, and K. C. Persaud, "Develoment of an electronic nose for fire detection," Sensors and Actuators B: Chemical, vol. 6,. 55-6, 6. [8] S. Amuero and J. O. Bosset, "The electronic nose alied to dairy roducts: a review," Sensors and Actuators B: Chemical, vol. 9,. -,. [9] L. Carmel, "Electronic nose signal restoration--beyond the dynamic range limit," Sensors and Actuators B: Chemical, vol. 6,. 95-, 5. [] L. Carmel, N. Sever, D. Lancet, and D. Harel, "An enose algorithm for identifying chemicals and determining their concentration," Sensors and Actuators B: Chemical, vol. 9,. 77-8,. [] R. Haddad, L. Carmel, and D. Harel, "A feature extraction algorithm for multi-eak signals in electronic noses," Sensors and Actuators B: Chemical, vol. In Press, Corrected Proof. [] L. Carmel, S. Levy, D. Lancet, and D. Harel, "A feature extraction method for chemical sensors in electronic noses," Sensors and Actuators B: Chemical, vol. 9, ,. [] M. Padilla, I. Montoliu, A. Pardo, A. Perera, and S. Marco, "Feature extraction on three way enose signals," Sensors and Actuators B: Chemical, vol. 6,. 5-5, 6. [] C. Hong, R. A. Goubran, and T. Mussivand, "Imroving the classification accuracy in electronic noses using multi-dimensional combining (MDC)," resented at Sensors. Proceedings of IEEE,. [5] S. J. Qin and Z. J. Wu, "A new aroach to analyzing gas mixtures," Sensors and Actuators B: Chemical, vol. 8, ,. [6] J. Matthes, L. Groll, and H. B. Keller, "Source localization by satially distributed electronic noses for advection and diffusion," Signal Processing, IEEE Transactions on [see also Acoustics, Seech, and Signal Processing, IEEE Transactions on], vol. 5,. 7-79, 5. [7] A. Loutfi and S. Coradeschi, "Relying on an electronic nose for odor localization," resented at Virtual and Intelligent Measurement Systems. VIMS '. IEEE International Symosium on,. [8] L. Marques and A. T. De Almeida, "Electronic nose-based odour source localization," resented at Advanced Motion Control. Proceedings. 6th International Worksho on,. [9] W. Jatmiko, Y. Ikemoto, T. Matsuno, T. Fukuda, and K. Sekiyama, "Distributed odor source localization in dynamic environment," resented at Sensors, IEEE, 5. Jie CAI (98 resent) PhD. Student in School of Electrical and Information Engineering, University of Sydney. Receive his B.E. at in University of Science and Technology of China, M.IT. at 6 in University of Griffith. Now he is interesting in electronic nose system with source localization and signal rocessing. Daivd C Levy received his B Sc at 97, M Sc at 976, PhD at 995 in Electrical Engineering from University of Natal. Now he is Associate rofessor in School of Electrical and Information Engineering, University of Sydney. He is major research area is in electronic nose system, software engineering and Comonents, Architecture and Middleware. 85