Act Polytechnic Hungric Vol. 0, No. 3, 03 Compenstion of the Impct of Disturing Fctors on Gs Sensor Chrcteristics Zvezditz Nenov, Georgi Dimchev Technicl University of Grovo Deprtment of Electricl Engineering 4, H. Dimitr Str., Grovo 5300, Bulgri nenov@tug.g; gdimchev@tug.g Astrct: Methods for gs control hve een extensively developed for the monitoring of ir qulity, for gs lek control, for the development of electronic nose systems, etc. Metl oxide gs sensors hve een widely used in prticulr. However, prt from chnges in the controlled gs concentrtion, chnges in their prmeters lso depend on mient conditions. The min impct comes from temperture nd humidity. Therefore, the compenstion of these disturnces is importnt for incresing the ccurcy of concentrtion mesurements of the controlled gses nd the reliility of control. The present pper proposes method for compensting the impct of temperture nd humidity on gs sensor chrcteristics using rtificil neurl networks. This compenstion method is pplied to the control of methne concentrtion y gs sensors TGS83 nd TGS6. The results otined confirm the pplicility of this method. Keywords: compenstion; gs sensors; disturing fctors; rtificil neurl networks Introduction Gs systems re widely used for monitoring outdoor nd indoor ir qulity, in gs lek control systems, in the chemicl industry, in the development nd implementtion of electronic nose systems, etc. [-5]. The control of ir prmeters is importnt for the protection of the environment nd humn helth, s well s for providing sfe working conditions. Gs pollution cn spred over wide re in short time, nd therefore methods nd equipment for its mesurement nd monitoring re eing extensively developed. A wide rnge of gs sensors [6-0] hve een designed, including metl oxide gs sensors. Different kinds of metl oxides such s SnO, ZnO, Fe O 3, WO 3, Co 3 O 4, etc. [- 6] re used s sensing mterils. Their operting principle is sed on incresing the conductivity of the surfce film of the sensitive element when the test gs is dsored. Depending on the composition of the surfce film, the sensor responds 97
Z. Nenov et l. Compenstion of the Impct of Disturing Fctors on Gs Sensors Chrcteristics to different gses such s cron monoxide, cron dioxide, ethnol, methne, propne, mmoni, hydrogen sulfide, hydrogen, etc. [6-]. Metl oxide gs sensors hve high sensitivity, low cost nd short response time. However, their chrcteristics re influenced y vrious mient prmeters which ct s disturing fctors in gs control. Temperture nd humidity hve mjor impct mong these fctors [7-0, 7, 8]. To enhnce the mesurement ccurcy nd reliility of control, compensting the impct of disturing fctors on gs sensors is of prime importnce. A method for compensting the impct of mient temperture nd humidity on gs sensor chrcteristics y using rtificil neuron networks (ANN) is proposed in this pper. The method is sed on three-dimensionl pproximtion of the gs sensor chrcteristics employing n ANN. The method is pplied to the control of methne concentrtion with gs sensors TGS83 nd TGS6, nd the results of tht implementtion re shown. ANN Compenstion Method For metl oxide gs sensors, the input quntity is the unknown concentrtion, Conc, of the gs eing controlled, which leds to chnge in the output quntity of the sensor - its resistnce, Rs. The mient fctors, temperture, t, nd reltive humidity, RH, which ct s disturing fctors, lso hve n effect on this resistnce (Fig. ). t RH Conc Gs sensor Rs Figure Input quntity Conc nd disturing fctors t nd RH for gs sensors Sensor mnufcturers usully report gs sensor chrcteristics s the sensor resistnce rtio, Rs / Ro, under vrious gs concentrtions nd mient conditions, i.е.: Rs f Conc, t, RH Ro, () where Rs is sensor resistnce, nd Ro is resistnce for referent concentrtion, temperture nd humidity. 98
Act Polytechnic Hungric Vol. 0, No. 3, 03 However, these chrcteristics re usully given only for some vlues of the disturing fctors Rs Ro f i Conc t RH const i, i () i=,,, n. It should e noted tht in prctice it is difficult to clirte the gs sensor, nd the impct of disturing fctors is usully given only for chrcteristics t fixed concentrtion. Rs Ro f i t RH Concconst i, i (3) i=,,, n. If in the ppliction of gs sensors the operting chrcteristic is chosen for fixed t nd RH (most commonly t 0С/65%RH), this inevitly leds to mesurement errors due to chnges in mience. In order to tke into considertion the impct of t nd RH, it is necessry to pproximte the trnsformtion function of the sensor nd use it in pplictions. Bsed on eqution (), this should e three-dimensionl pproximtion. Difficulties rise owing to the gret nonlinerity of chrcteristics () (3). Additionlly, s ws mentioned, the sensor chrcteristics usully cnnot e given t uniform points tht cn e used in function pproximtion. A theoreticl method for polynomil pproximtion of multivrile sensor chrcteristic ws proposed in [8]. In its prcticl ppliction only the compenstion of the impct of humidity, RH, on gs sensor chrcteristics is shown. However, introducing second disturing fctor (such s temperture), would sustntilly increse the numer of equtions nd coefficients used. Artificil neurl networks cn lso e employed for solving different prolems with mny input prmeters. It is very common to use ANN for gs nd odor recognition, the clssifiction of products, the control of environmentl prmeters, etc. [9-3]. In [4] n ANN-sed virtul compenstor for correcting the effect of disturing vrile in trnsducers is proposed. Tht method is pplied to strin-guge trnsducer-sed pressure mesurement system. The correction is crried out y nonliner two-dimensionl rtificil neurl networksed inverse model of the trnsducer. ANN hs lso een used for twodimensionl pproximtion of humidity sensor chrcteristics in order to compenste for the impct of one fctor - temperture [5]. This pproch is shown to chieve the highest ccurcy for nonliner trnsformtion function compred to polynomil nd interpoltion methods. The compenstion for temperture effects in gs sensors vi ANN is reported in [6]. 99
Z. Nenov et l. Compenstion of the Impct of Disturing Fctors on Gs Sensors Chrcteristics The ANN-sed method proposed in this pper ims to compenste for the impct on gs sensors of oth mient temperture nd humidity through threedimensionl pproximtion of the gs sensor chrcteristics. The method is implemented in two stges: trining of the ANN, nd rel mesurement nd control of gs concentrtion. In the trining stge, the clirtion chrcteristics given y the mnufcturers re used. Input prmeters for the ANN re: the gs sensor resistnce rtio, Rs / Ro, mient temperture, t, nd reltive humidity, RH, nd n output prmeter the concentrtion, Conc, of the respective gs. The points of trining cn e complemented, whenever possile, y functionl pproximtion of chrcteristics () nd offsets sed on (3). As result of the ANN trining, three-dimensionl pproximtion of the sensor chrcteristics is performed with reltionships of the type Conc f Rs / Ro, t, RH, W,,, (4) where W, nd re ANN prmeters. In the stge of rel mesurement nd control, in ddition to mesuring the gs sensor prmeter, it is necessry to mesure temperture seprtely y mens of temperture sensor nd ir humidity y mens of humidity sensor. A schemtic digrm of the method implementtion for one gs is shown in Fig.. Amience Gs Sensor Rs/Ro Temperture Sensor t, C W ANN Concentrtion Conc, ppm Humidity Sensor RH, % Figure Schemtic digrm of the implementtion of the ANN compenstion method for one gs On the sis of the pproximtion reltionships otined (eqution (4)) the mesured gs concentrtion is determined. The chnges in mient temperture nd humidity re tken into ccount, nd therefore, their impct on gs sensor chrcteristics is compensted for. The method cn lso e employed for higher order pproximtion for greter numer of disturing fctors. 00
Act Polytechnic Hungric Vol. 0, No. 3, 03 3 Results nd Discussion The method is pplied to compenste for the impct of temperture nd humidity on gs sensors TGS83 nd TGS6 for the control of methne concentrtion. Sensor chrcteristics () nd (3) t 000 ppm nd 5000 ppm, respectively, given y mnufcturers hve een used [7]. For sensor TGS83, Ro is the gs sensor resistnce for the referent concentrtion of 000 ppm nd 0 C/65%RH. According to experimentl chrcteristics [7], in Rs / Ro f Conc of the sensor for given t logrithmic scle, the chrcteristics nd RH re stright lines nd cn e represented y n eqution of the form y. x, (5) 0 where y lgrs / Ro, x lgconc. These chrcteristics re prllel stright lines; i.е., coefficient is constnt nd cn e determined y ny of the experimentl reltionship Rs / Ro f Conc for t const nd RH const. Vritions in temperture nd reltive humidity led only to chnge in the offset 0 of these chrcteristics. This offset hs een clculted on the sis of chrcteristics (3) t the reference concentrtion for temperture vrition in the rnge of -0 С +40 С nd reltive humidity in the rnge of 0 00%RH [7]. Thus, the fmily of chrcteristics re otined nlyticlly t vrious tempertures in the rnge of -0 С +40 С nd fixed humidities of 0, 0, 40, 65 nd 00%RH. Fig. 3 presents this fmily of chrcteristics for sensor TGS83 t 65%RH, showing the impct of temperture. Figure 3 Anlyticlly otined chrcteristics for sensor TGS83 for temperture vrition nd 65%RH 0
Z. Nenov et l. Compenstion of the Impct of Disturing Fctors on Gs Sensors Chrcteristics Similrly, sed on the experimentl chrcteristics [7] for sensor TGS6, the fmilies of chrcteristics hve een otined t tempertures in the rnge of - 0С to 40С nd for fixed vlues of reltive humidity of 35, 50, 65 nd 95%RH. For this sensor, Ro is the resistnce t referent concentrtion of 5000 ppm nd 0С/65%RH. Fig. 4 presents the fmily of nlyticlly otined chrcteristics for sensor TGS6 for temperture vrition nd 65%RH. Figure 4 Anlyticlly otined chrcteristics for sensor TGS6 for temperture vrition nd 65%RH The experimentl chrcteristics, prt from those t 0С/65%RH, nd the whole set of nlyticlly otined chrcteristics for ech sensor TGS83 nd TGS6 is used for ANN trining. The experimentl chrcteristics t 0С/65%RH re used for checking the ccurcy of the proposed method. Experiments with vrious lgorithms hve een crried out for the ANN trining. The est convergence for the smllest numer of neurons is otined in trining with the LM lgorithm (Levenerg-Mrqurdt ck propgtion). The otined ANN with ck propgtion of error hs three lyers: two hidden (input, intermedite) nd one output lyer. The first lyer consists of three neurons, one for ech input quntity, the second lyer is mde up of seven neurons, nd the third lyer hs one neuron (Fig. 5). In oth the first nd second lyers the trnsfer functions of neurons f nd f 3 re sigmoidl, nd in the third lyer f it is liner. The neurl network hs the following form 3 3,,, 3 Y f LW f LW f IW p, (6) where Y Conc, p Rs / Ro, p t, p3 RH. 0
Act Polytechnic Hungric Vol. 0, No. 3, 03 p Rs / Ro p Temperture p 3 Humidity iw Hidden lyer,,, iw 3,3 3,, lw, Hidden lyer 3 f lw 7 7, 7,3, 3, lw, Output lyer 3, lw,7 3 f Y Concentrtion 3 3, 3 f ( IW p ) f ( LW ) Y f ( LW ) Figure 5 f 3 ANN for pproximting the gs sensor chrcteristics Fig. 6 shows the results from the output of trined neurl networks for sensors TGS83 nd TGS6 nd surfces with different humidity levels re illustrted. Thus, on the sis of three-dimensionl pproximtion of sensor chrcteristics resulting from ANN trining, the vlue of methne concentrtion cn e otined when the impct of temperture nd reltive humidity is compensted for. Fig. 7 shows the chrcteristics otined y ANN, illustrting the joint impct of mient temperture nd humidity on the sensors resistnce rtio, Rs / Ro, t concentrtion of 000ppm for sensor TGS83 nd 5000ppm for sensor TGS6. 03
Z. Nenov et l. Compenstion of the Impct of Disturing Fctors on Gs Sensors Chrcteristics Figure 6 Results from the output of the trined neurl network: а) for sensor TGS83; ) for sensor TGS6 04
Act Polytechnic Hungric Vol. 0, No. 3, 03 Figure 7 Impct of mient temperture nd humidity on sensors resistnce rtio Rs/Ro: а) of sensor TGS83 t referent concentrtion of 000ppm; ) of sensor TGS6 t referent concentrtion of 5000ppm 05
Z. Nenov et l. Compenstion of the Impct of Disturing Fctors on Gs Sensors Chrcteristics The lgorithm for compensting for the impct of temperture nd humidity on gs sensors redings y mens of ANN in the process of gs control is shown in Fig. 8. Strt Mesuring current temperture Mesuring current reltive humidity Mesuring current gs sensor resistnce Processing of mesurements y ANN Otining compensted vlue for the gs concentrtion End Figure 8 ANN compenstion lgorithm in gs sensors To estimte the error which occurs if there is no compenstion for temperture nd humidity on sensor chrcteristics, the solute error Conc t, RH Conc Conct, RH (7) nd normlized error Conc t, RH n t, RH.00%, (8) Concmx Concmin re clculted, where Conc is the concentrtion sed on the operting chrcteristic without compenstion; Conc t, RH is the rel concentrtion corresponding to chrcteristics given the vrition in temperture nd reltive humidity; nd Concmx Conc min is the rnge of concentrtion vrition for ech sensor. The errors occurring for 000ppm nd 000 ppm when using the sic chrcteristics of sensors t 0С/65%RH, without tking into ccount the vrition in temperture nd in reltive humidity, re shown in Tles nd respectively. 06
Act Polytechnic Hungric Vol. 0, No. 3, 03 Tle Normlized error for sensor TGS83 without compenstion when using the sic chrcteristic t 0С/65%RH t C /%RH Conc, n t, RH, ppm % -0 C / 0%RH 000-8.7-0 C / 0%RH 000-60.0 40 C/00%RH 000 3.7 40 C/00%RH 000 7.3 Tle Normlized error for sensor TGS6 without compenstion when using the sic chrcteristic t 0С/65%RH t C /%RH Conc, n t, RH, ppm % -0 C / 0%RH 000-6.5-0 C / 0%RH 000-3.5 40 C/00%RH 000 5.9 40 C/00%RH 000.9 These results confirm the necessity of compensting for the impct of temperture nd reltive humidity. Using the trined neurl network, the vlues Conc ANN of methne concentrtion hve een otined t vrious vlues of resistnce, temperture nd humidity. The solute error is determined sed on these vlues Conc ConcANN Conc (9) nd the normlized error of the ANN method is Conc n.00%, (0) Conc mx Conc min where Conc ANN is the concentrtion vlue, determined using the trined neurl network; Conc is the respective rel concentrtion vlue from the sic experimentl chrcteristics which hve not tken prt in trining; nd Concmx Conc min is the rnge of concentrtion vrition for ech sensor. The experimentl sic chrcteristics, which hve not een used in trining, nd those otined y ANN for the two sensors re shown in Fig. 9. Fig. 0 gives grphic presenttion of normlized errors (eqution (0)) when employing the proposed method for compenstion y ANN. 07
Z. Nenov et l. Compenstion of the Impct of Disturing Fctors on Gs Sensors Chrcteristics Figure 9 Experimentl sic chrcteristics nd chrcteristics otined y ANN Figure 0 Normlized errors when implementing the ANN-compenstion method The otined results show tht the normlized error of the ANN method is in the rnge of -0.05% to +0.35% for sensor TGS83, nd -0.% to +0.3% for sensor TGS6, which confirms the effectiveness of the implementtion of the proposed ANN compenstion method. 08
Act Polytechnic Hungric Vol. 0, No. 3, 03 Conclusions On the sis of the reserch conducted, the following conclusions cn e drwn: mient temperture nd humidity hve sustntil impct on metl oxide gs sensor chrcteristics; method is proposed for compensting for the impct of temperture nd humidity on gs sensors y using ANN for three-dimensionl pproximtion of their chrcteristics; to compenste for the impct of mient conditions on gs sensors of type TGS83 nd TGS6 for the mesurement nd control of methne, trined ANN with ck propgtion of error with two hidden (input nd intermedite) nd one output lyers hs een otined; the three-dimensionl pproximtion of gs sensor chrcteristics y the trined neurl network llows the impct of temperture nd humidity to e compensted for nd the normlized error is from -0.05% to +0.35% for sensor TGS83 nd from -0.% to +0.3% for sensor TGS6. the proposed method of compensting for the impct of disturing fctors y ANN cn lso e used for other types of sensors, s well s for performing higher order pproximtion with greter numer of disturing fctors. References [] G. F. Fine, L. M. Cvngh, A. Afonj nd R. Binions: Metl Oxide Semi- Conductor Gs Sensors in Environmentl Monitoring, Sensors 00, 0, pp. 5469-550 [] M. Fleischer, M. Lehmnn: Solid Stte Gs Sensors - Industril Appliction, Springer-Verlg Berlin Heidelerg, 0, p. 69 [3] K. Arshk, E. Moore, G. M. Lyons, J. Hrris nd S. Clifford: A Review of Gs Sensors Employed in Electronic Nose Applictions, Sensor Review, Vol. 4, Numer, 004, pp. 8-98 [4] A. D. Wilson nd M. Bietto: Applictions nd Advnces in Electronic- Nose Technologies, Sensors, 009, 9, pp. 5099-548 [5] S. Zmpolli, I. Elmi, F. Ahmed, M. Pssini, G. C. Crdinli, S. Nicoletti, L. Dori: An Electronic Nose Bsed on Solid Stte Sensor Arrys for Low- Cost Indoor Air Qulity Monitoring Applictions, Sensors nd Actutors B 0, 004, pp. 39-46 [6] T. Nenov, P. Pnteleev: Gs Sensors for Environmentl Monitoring. Automtic&Informtics, 00, No., pp. 6-9 [7] FIGARO Engineering Inc. Products - Gs Sensors (www.figro.co.jp/en/product/) 09
Z. Nenov et l. Compenstion of the Impct of Disturing Fctors on Gs Sensors Chrcteristics [8] SYNKERA Technologies Inc. Products (www.synker.com) [9] ev Technologies. Products (www.ev.com) [0] Sencer. Products (www.sencer.com) [] N. Brsn nd U. Weimr: Understnding the Fundmentl Principles of Metl Oxide-sed Gs Sensors; the Exmple of CO Sensing with SnO Sensors in the Presence of Humidity, J. Phys.: Condens. Mtter, 003, 5, pp. 83-839 [] K. Shimizu, I. Chinzei, H. Nishiym, S. Kkimoto, S. Sugy, W. Mtsutni, A. Stsum: Doped-Vndium Oxides s Sensing Mterils for High Temperture Opertive Selective Ammoni Gs Sensors. Sensors nd Actutors B 4, 009, pp. 40-46 [3] N. Hn, L. Chi, Q. Wng, Y. Tin, P. Deng, Y. Chen: Evluting the Doping Effect of Fe, Ti nd Sn on Gs Sensing Property of ZnO, Sensors nd Actutors B 47, 00, pp. 55-530 [4] Ch.-Y. Lin, Y.-Y. Fng, Ch.-W. Lin, J. J. Tunney, K.-Ch. Ho: Friction of NO x Gs Sensors Using In O 3 ZnO Composite Films, Sensors nd Actutors B 46, 00, pp. 8-34 [5] Z. Jing, Z. Guo, B. Sun, Y. Ji, M. Li, J. Liu: Highly Sensitive nd Selective Butnone Sensors Bsed on Cerium-doped SnO Thin Films, Sensors nd Actutors B 45, 00, pp. 667-673 [6] G. Korotcenkov, B. K. Cho: Thin Film SnO -sed Gs Sensors: Film Thickness Influence, Sensors nd Actutors B 4, 009, pp. 3-330 [7] Ch. Wng, L. Yin, L. Zhng, D. Xing nd R. Go: Metl Oxide Gs Sensors: Sensitivity nd Influencing Fctors, Sensors 00, 0, pp. 088-06 [8] J. Jniczek: Approximtion of Multivrile Sensor Chrcteristic Applied to the Cron Monoxide Sensor, Mesurement, 00, 43, pp. 5-8 [9] E. L. Hines, J. W. Grdner: An Artificil Neurl Emultor for n Odor Sensor Arry, Sensors nd Actutors B, 8-9, 994, pp. 66-664 [0] W. Ping, X. Jun: A Novel Recognition Method for Electronic Nose Using Artificil Neurl Network nd Fuzzy Recognition, Sensors nd Actutors B 37, 996, pp. 69-74 [] H.-K. Hong, Ch. H. Kwon, S.-R. Kim, D. H. Yun, K. Lee, Y. K. Sung: Portle Electronic Nose System with Gs Sensor Arry nd Artificil Neurl Network, Sensors nd Actutors B 66, 000, pp. 49-5 [] D. Luo, H. G. Hosseini, J. R. Stewrt: Appliction of ANN with Extrcted Prmeters from n Electronic Nose in Cigrette Brnd Identifiction, Sensors nd Actutors B 99, 004, pp. 53-57 0
Act Polytechnic Hungric Vol. 0, No. 3, 03 [3] S. Osowski, T. H. Linh, K. Brudzewski: Neuro-Fuzzy Network for Flvor Recognition nd Clssifiction, IEEE Trnsctions on Instrumenttion nd Mesurement, Vol. 53, No. 3, June 004, pp. 638-644 [4] A. P. Singh, S. Kumr, T. S. Kml: Virtul Compenstor for Correcting the Disturing Vrile Effect in Trnsducers, Sensors nd Actutors A 6, 004, pp. -9 [5] T. Nenov, S. Ivnov: Lineriztion of Chrcteristic of Reltive Humidity Sensor nd Compenstion of Temperture Impct, Sensors nd Mterils, 007, Vol. 8, No., pp. 95-06 [6] W. Ho, X. Li, M. Zhng: Appliction of RBF Neurl Network to Temperture Compenstion of Gs Sensor, Proceeding CSSE '08 Proceedings of the 008 Interntionl Conference on Computer Science nd Softwre Engineering - CSSE '08, Vol. 04, pp. 839-84