16 IEEE Intenational Confeence on Multisenso Fusion and Integation fo Intelligent Systems (MFI 16) Kongesshaus Baden-Baden, Gemany, Sep. 19-1, 16 A Thee-Axis Magnetic Senso Aay System fo Pemanent Magnet Tacking* Houde Dai, Wanan Yang, Xuke Xia, Shijian Su, and Kui Ma Abstact Oientations and positions of a pemanent magnet can be acquied via a thee-axis magnetic senso aay system. Thus, the position and oientation of an object embedded with the pemanent magnet can be tacked in eal-time. This magnet tacking system can be adapted to medical and industial applications. How to detect the weak magnetic signal and the senso fusion algoithm ae vitally impotant fo the accuacy pefomance of the tacking system. In this study, we pesent a novel magnet tacking system based on nine thee-axis digital magnetic sensos, instead of analog output, with the valid tacking space of.5m.5m.m. Befoe calibation, the system achieved aveage localization eo at 6.6 and.98. This system has the advantages such as low powe consumption, potable, and convenient configuation via a SPI bus. I. INTRODUCTION Oientation and localization method is the key technique in the applications such as obotics, medical devices, and industy puposes [1]. Thee ae vaious positioning and navigation technologies such as GPS (global positioning system), ulta-sound, inetial sensing, adio fequency (Ulta Wideband /UWB, Radio Fequency Identification /RFID, Wieless Fidelity /WIFI, and etc.), electomagnetics, vision and optical techniques []. Each of these technologies has its advantage and disadvantages. Electomagnetic techniques have the elative supeioity fo the tacking in a small but close space [1]. Fo example, a wieless endoscope capsule can move and photogaph the entie intestine. Howeve, it is difficult to tack the oientation and position of the capsule endoscope when it is going though the stomach and gastointestinal (GI). ANKON Co., Ltd. China developed a magnetically contolled capsule endoscope system, which consists of a cubic magnetic senso (Honeywell HMC5843) aay and can pecisely tack and contol the magnet-based capsule endoscope duing the complete examination of the stomach and GI tact. Thus, the pefomance of the capsule endoscope is geatly impoved and this novel capsule endoscope system has been widely adopted by hospitals in China. Thee is no detailed infomation fo the tacking accuacy [3,13]. Detection of magnetic object motions has been widely investigated. Schlagete et al. developed a tacking system using 16 hall sensos and a pemanent in eal-time up to 5Hz [4]. Chao Hu et al. have developed a seies of magnet tacking systems [5, 6]. They used Honeywell HMC153 and HMC143 to build plane o cubic magnetic senso aays. A seies of obust tacking algoithms wee tested. The HMC153-based plane senso aay pefomed aveage localization eo at 3.3 and the aveage oientation eo at 3., while the HMC143-based plane senso aay pefomed aveage localization eo at.1 and the aveage oientation eo at.1. Howeve, HMC153 and HMC143 ae analog sensos, and the hadwae of the tacking system includes multiplexe, compensation cicuit, amplifie, analog-to-digital convete (ADC), and othe cicuit pats. And noises could influence the detected magnetic signal acoss the hadwae. Theefoe, the newly developed mico-electo-mechanical system (MEMS) magnetic sensos, which have digital outputs, should be adopted [7]. With the digital magnetic sensos, the stuctue of the hadwae will be vey simplified and the signal-to-noise will be impoved geatly. One limitation of the pemanent magnet tacking system is its limited tacking ange because of the apid attenuation of the magnet induced signal [8, 9]. In the fist step of this study, a.5m.5m plane senso aay was designed. The majo concen is to impove the tacking accuacy and simplify the system stuctue of the pemanent magnet tacking system. In this pape, Section II descibes the hadwae and pototypical ealization of the tacking system. Section III descibes the mathematical model fo the magnetic field induced by a pemanent magnet. Section V pesents the tacking algoithms and expeiment esults. II. PROTOTYPICAL REALIZATION Magnet SPI+Senso Select ARM Contolle UART-USB GUI *Reseach suppoted by the Chinese Academy of Sciences unde Pojects CAS-ITRI154, 165 (the Chinese Academy of Sciences /ITRI coopeation pogam) and Poject YZ151 (Reseach equipment development poject of the Chinese Academy of Sciences). Houde Dai, Xuke Xia, Shijian Su, and Kui Ma ae with the Quanzhou Institute of Equipment Manufactuing, Haixi Institutes, Chinese Academy of Sciences, Jinjiang, Fujian, 36 China (Tel: +86 595 3635373; e-mail: dhd@fjism.ac.cn). Wanan Yang is with the School of Compute and Infomation, Yibin Univesity, China (e-mail: ywaly@16.com). Figue 1. System diagam of the magnet tacking system Fig. 1 shows the system diagam of the pemanent magnet tacking system. The nine magnet sensos (senso aay) detected the induced magnetic field intensity when a pemanent magnet moves on the upside of the senso aay. The senso data wee acquied via SPI (seial peipheal 476
inteface) inteface and tansmitted into compute via a USB (univesal seial bus) to UART (univesal asynchonous eceive/tansmitte) inteface. Then the tacking algoithms wee pefomed in the compute and the calculated paametes wee displayed on the GUI (gaphical use inteface). The calculated paametes ae 3-dimentional position and 3-dimentional oientation values. Fig. shows the simulation of the magnetic field intensity induced by a pemanent magnet (NdFeB), while Fig. 3 shows its magnetic flux density. Fig. 4 pesents the photo of such pemanent magnet. AK991 was chosen in this poject fo its lage measuement ange and elative high esolution. TABLE I. COMPARATIVE SPECIFICATIONS OF MAGNET SENSORS Senso Range LSB a Powe Bus Package AK991, 3-axis,.15 1.7-3.6V; IIC e BGA b AKM ±4.9mT ut 1mA@1Hz /SPI f MMC3416XPJ, 3-axis,.5 1.6-3.6V; IIC BGA Mems ±1.6mT ut.14ma@7hz HMC5983, 3-axis,..16-3.6V; IIC QFN c Honeywell ±.8mT ut.1ma@7.5hz /SPI BMM15, X,Y:.3 1.6-3.6V; IIC Ball Bosch ±1.3mT; ut.5ma@hz /SPI pitch HSCDTD8A, Alps Z:±.5mT 3-axis, ±.4mT.15 ut 1.8-3.6V; 1mA@1Hz IIC /SPI LGA d a. LSB (Least significant bit) means esolution; b. BGA epesents ball gid aay; c. QFN epesents quad flat no-lead package; d. LGA denotes land gid aay; e. IIC denotes inte-integated cicuit; f. SPI means seial peipheal inteface. The infomation was acquied based on the datasheets of these senso chips. AK991 is thee-axis electonic compass chip based on Hall-effect sensing. Fig. 5 depicts the system diagam of AK991, and Fig. 6 shows photo of a magnet senso boad. Figue. Magnetic induction intensity emulation of a single pemanent magnet ( = 6, h = 3) with ANSYS softwae (Ansys, Inc., USA) 3-axis Hall Senso MUX Amplifie Senso Dive ADC Signal Pocessing Inteface,Logic &Registe IIC/ SPI Figue 5. Block diagam of AK991. Figue 6. Photo of a magnet senso boad Figue 3. Magnetic flux density emulation of a single pemanent magnet ( =6,h=3) with ANSYS softwae Figue 4. Photo of a single pemanent magnet ( = 8, h = 3) A. Selection of Sensos The fist step in implementing the magnet tacking system was to select the appopiate sensos. The pimay featues of seveal available majo magnetic sensos and IMUs ae listed in Table I. These paametes wee taken fom the elative datasheets. These sensos ae poduced by the manufactues such as AKM, Bosch, Honeywell, Bosch, and Alps, espectively. The sensos in Table I ae digital sensos. The dominant paametes ae measuement ange and output data esolution. High measuement ange denotes that the senso is not easily satuated when the senso closes to a pemanent magnet. Geat output data esolution means lage tacking dimension. B. Pototype Realization With new digital magnetic sensos (AK991) and the new system stuctue as shown in Fig. 1 and Fig. 7, a novel pototype was built up. The dimensions of the senso aay, which consisted of nine magnetic sensos, was.5m.5m. A USB cable was used to connect the ARM contolle boad to the compute. The powe supply is 5V and the cuent is 87 ma, while the senso aay just consumes about 7mA of cuent. The digital output magnetic sensos wee configued and ead via a SPI bus. Figue 7. Pototype ealization of the magnet tacking system. 477
III. MATHEMATICAL MODEL AND ALGORITHM A. Mathematical Model Figue 8. dipole model fo a pemament magnet in this study The pemanent magnet in this study woks as a dipole [11]. Thus as Fig. 8 shows, the magnetic field (B) aound the dipole can be expessed as: v v v M 3( H P ) P H T B B e B e B e ( ), (1) x x y y z z 5 3 4 whee μ denotes the elative pemeability of the medium; μ means the ai magnetic pemeability; M T epesents the magnetic intensity constant of the magnetic dipole; P denotes vecto defining a spatial point (x l, y l, z l ) T with espect to the magnetic dipole;h means a vecto (m, n, p) T defining the diection of the dipole; means a scala defining of the length of P. With opeation as B l P l H, when the senso amount is N, with l-th senso located in (x l, y l, z l ) T, simplified it as: bp cn cm ap an bm B B B B y B z B z B x B x B y lx ly lz lz y x z y x l m n p, () whee (m, n, p) T and (a, b, c) T denote the thee-axis oientation and thee-axis position espectively. When the cylinde-shaped pemanent magnet moves along its axis, B l ae invaiant. Thus, only 5-dimentional paametes of the tacking object (pemanent magnet) can be calculated. But a constaint can be acquied: m n p 1. (3) B. Algoithms The diffeences between the calculated values (B calc ) based on the mathematical model [()] and the measued values (B meas ) ae descibed as follows: N 3 m ( x a ) n( y b) p ( z c).( x a ) m x lx T 5 3, (4) l 1 N 3 m ( x a ) n( y b) p ( z c).( y b) n y ly T 5 3, (5) l 1 N 3 m ( x a ) n( y b) p ( z c).( z c) p z lz T 5 3. (6) l 1 l-th Senso whee μ denotes μ μ M T /4π. Theefoe, the total diffeence in all sensos can be descibed as: N ( xl ) ( xl ) ( yl ) ( yl ) ( zl ) ( zl ) [( ) ( ) ( ) ]. (7) meas calc meas calc meas calc l 1 E B B B B B B The algoithm is to seach fo the minimum solution of (7) as the esult and should meet the equiements such as eal-time calculation and obustness to the noises in the data. Table II shows the coonly use of the algoithms to solve nonlinea equations such as (7). By compaison, this poject adopts Levenbug-Maquadt algoithm [, 6]. TABLE II. COMPARISON OF TRACKING ALGORITHMS Algoithm Eo Speed Requie fo Initial Guess Powell s algoithm lage fast odinay DIRECT; Multilevel coodinate seach Levenbug-Maquadt method small slow odinay odinay fast low C. Calibation Calibation is necessay fo the paametes elated to the sensitivity, positions and oientation of each senso [11]. Then this magnetic tacking system can achieve highe accuacy. IV. EXPERIMENTS Fig. 9 shows the GUI of the eal-time magnet tacking system. It displays the tacking paametes in eal time, with the two/thee-dimensional plots of the tacked magnet s tajectoy. Figue 9. Inteface fo the magnet tacking system The magnet tacking system was evaluated befoe calibations. Fig. 1 and Fig. 11 show position eos and oientation eos espectively. The position eo o oientation eo wee the diffeences between the eal value and algoithm output. 478
degee degee degee Diection Eos (deg) Position Eos () Results show the total position eos (absolute mean) anged fom to 15, while the oientation eos anged fom to 6. 15 1 5-5 -1-15 4 6 8 1 1 14 16 Sample Numbe Figue 1. Position eos x y z total 8 8 78 1 3 4 5 6 7 8 88 86 Position Vaiation in X-axis Sample Numbe 84 1 3 4 5 6 7 8 46 44 Position Vaiation in Y-axis Sample Numbe Position Vaiation in Z-axis 1 8 6 4 - -4 x y z total 4 1 3 4 5 6 7 8 Figue 13. Plot of the position dift duing the 85-second peiod () Fig. 14 shows the oientation dift duing 85 seconds. -8-1 Sample Numbe Oientation Vaiaton in X-axis -6 4 6 8 1 1 14 16 Sample Numbe Figue 11. Oientation eos As Fig. 1 shows, 5 samples ae evenly placed at the plane of 3 3. The tested values and eal values of the X/Y-axis positions ae maked in the table. 15 1 D Plot Real Tested -1 1 3 4 5 6 7 8 Sample Numbe 8 6 4 1 3 4 5 6 7 8 Sample Numbe -76 Oientation Vaiaton in Y-axis Oientation Vaiaton in Z-axis 5-78 -5-1 -15-15 -1-5 5 1 15 Figue 1. X/Y-axis position plot () Fig.13 and Fig. 14 show the tacking stability of this system, while the difts of location and oientation duing 85 seconds wee ecoded when a pemanent magnet was fixed in the point (X=8.3, Y=85.5, Z=4.4). -8 1 3 4 5 6 7 8 Sample Numbe Figue 14. Plot of the oientation dift duing the 85-second peiod (degee) Table III shows the absolute mean and oot-mean-squae values of the position eos and oientation eos. The dominant eason why the tacking eos ae bigge than pevious studies is that the calibations ae not caied out befoe the measuements. TABLE III. POSITION AND ORITENTATION ERRORS Absolute Mean Root Mean Squae X 3.1 4.48 Y 3.94 5.13 479
Absolute Mean Root Mean Squae Z.7 3.9 Total 6.6 7.48 1.8.35 1.76.59.14.1 Total.98 3.5 V. CONCLUSION At pesent, eal-time tacking systems based on a pemanent magnet wee too cumbesome and had high powe consumption. In this study, a potable tacking system was built up based on the digital magnet sensos. The system stuctue was geatly simplified with low powe consumption (35mW) and convenient configuation via SPI bus. The total position eo and oientation eo (absolute mean) of this tacking system wee 6.6 and.98 espectively. The next step of this study is the calibations of the magnet tacking system, which can impove the tacking accuacy lagely. In the next pototype vesion, an ARM contolle pefoms the signal pocessing and the tacking paametes can be tansmitted to othe device wielessly. Afte futhe validation, this system can be utilized in medical and industial applications [1]. [8] Y. Baell and H. W. L. Naus, Detection and localisation of magnetic objects, IET Science Measuement Technology, vol. 1, no. 5, pp. 45 54, 7. [9] M. Li, S. Song, C. Hu, D. M. Chen and M. Q.-H. Meng, A novel method of 6-DoF electomagnetic navigation system fo sugical obot, in Poc. 8th Wold Congess on Intelligent Contol and Automation, pp. 163 167, Jul. 1. [1] S. Song, C. Hu, M. Li, W. Yang and M. Q. -H. Meng, Real time algoithm fo magnet's localization in capsule endoscope, 9 IEEE Intenational Confeence on Automation and Logistics, pp. 3-35, Aug. 9. [11] C. Hu, M. Q. -H. Meng and M. Mandal, The Calibation of 3-Axis Magnetic Senso Aay System fo Tacking Wieless Capsule Endoscope, 6 IEEE/RSJ Intenational Confeence on Intelligent Robots and Systems, pp. 16-167, Oct. 6. [1] W. Yang, Y. He, Z. Wu, C. Hu, Y. X. Zou and F. Q. Qin, Investigation fo Movement of Spial-type Capsule Endoscope in the Pocine Lage Intestine, 15 IEEE Intenational Confeence on Infomation and Automation, pp. 563-566, Aug. 15. [13] Z. Liao, X. Hou, E. Q. Lin-Hu, J. Q. Sheng, Z. Z. Ge, et al., Jiang Accuacy of Magnetically Contolled Capsule Endoscopy, Compaed With Conventional Gastoscopy, in Detection of Gastic Diseases, Clinical Gastoenteology and Hepatology, vol. 14, no. 6, pp. 1-36, May 16. ACKNOWLEDGMENT This wok is suppoted by the Chinese Academy of Sciences unde Pojects CAS-ITRI154, 165 (the Chinese Academy of Sciences /ITRI coopeation pogam) and Poject YZ151 (Reseach equipment development poject of the Chinese Academy of Sciences). REFERENCES [1] C. Hu, S. Song, X. J. Wang, M. Q. -H. Meng and B. P. Li, A Novel Positioning and Oientation System Based on Thee-Axis Magnetic Coils, IEEE Tansactions on Magnetics, vol. 48, no. 7, pp. 11-19, July, 1. [] C. Hu, Y. P. Ren, X. H. You, W. A. Yang, S. Song and S. Xiang, Locating Inta-Body Capsule Object by Thee Magnet Sensing System, IEEE Sensos Council, vol. PP, no. 99, pp. 153-437X, Ap. 16. [3] NaviCam magnetically contolled capsule endoscope system fom ANKON Technologies Co., Ltd., Available (16): www.ankoninc.com.cn/ [4] V. Schlagete and R. Popovic, Tacking system using 16 hall sensos and a pemanent, Sensos and Actuatos APhycial, vol.9, no.1, pp:37-4, Aug. 1 [5] M. Li, C. Hu, S. Song and H. D. Dai, Detection of Weak Magnetic Signal fo Magnetic, 9 IEEE Intenational Confeence on Automation and Logistics, pp. 9-95, Aug. 9. [6] S. Song, C. Hu, B. P. Li, X. X. Li and M. Q. -H. Meng, An Electomagnetic Localization and Oientation Method Based on Rotating Magnetic Dipole, IEEE Tansactions on Magnetics, vol. 49, no. 3, pp. 174-177, Aug. 1. [7] D. K. Shaeffe, MEMS inetial sensos: A tutoial oveview, IEEE Counications Magazine, vol. 51, no.4, pp. 1 19, 13. 48