Proceedings VCM 2012 56 nâng cao chất lượng của thiết bị tích hợp INSGPS

Tuyển tập công trình Hội nghị Cơ điện tử toàn quốc lần thứ 6 411 Mã bài: 100 Improvement of the quality of integrated INS/GPS devices based on fuzzy logic Nâng cao chất lượng của thiết bị tích hợp INS/GPS trên cơ sở logic mờ PhD Student NGO Thanh Binh University of Transport and Communications (UTC), Hanoi, Vietnam e-Mail: ngobinh74@yahoo.com Abstract This paper presents a design with a new method using an electronic digital compass in integrated MEMS INS/GPS systems, applying for the field of management and supervision of moving objects. In the past, most of the IMUs usually used in the INS system were the single-axis sensors. Therefore, the calculation occurred many difficulties for the complicated PVT model. Moreover, designers needed to capture the values measured from the single-axis IMUs and then calculated for position of the center of mass INS. This caused large accumulated errors and reduced the calculating speed of the system. With the advent of i-Sensor MEMS 6- DOF (Analog Device, 2008), the calculation became simpler by eliminating the calculation of shifting the center coordinates of INS system. In September 2010, Mark Petovello (Inside GNSS) provided a model calculated in the discrete domain (z-Domain). Since then complex formulas to calculate the PVT model have been switching gradually to this method – the calculation method for PVA model. However, this method did not give the heading angle exactly and a solution to compensate for MEMS INS system errors flexibly. In comparison to previous systems, the integrated device used in my new design is an electronic digital compass, that can compensate for the calculated parameters of the MEMS INS system. Values measured by MEMS INS are limited by amplitude conditions. The calculations for this system are used in the discrete domain for the PVA model, using the compensation functions to process errors based on fuzzy control theory. This helps simplify the calculations and reduce the number of abnormal parameters, thereby improve the calculation speed and the accuracy of the integrated system. This method can be applied for integrated MEMS INS/GPS systems even in case of loss of GPS signal in short period of time. At this time, I have been making some tests and adjusting the fuzzy sets and rules step by step to limit position error. My target is to get the error value less than 10 meters when GPS signal is lost during 2 minutes. This is a new method, which has never been published in the previous designs. Tóm tắt Bài báo giới thiệu một thiết kế với một phương pháp mới sử dụng một la bàn điện tử trong hệ thống tích hợp MEMS INS/GPS, ứng dụng trong quản lý giám sát các đối tượng chuyển động. Trước đây hầu hết các IMU được sử dụng trong các hệ INS thường là những cảm biến đơn trục. Vì vậy, việc tính toán gặp rất nhiều khó khăn với mô hình PVT phức tạp. Thêm vào đó, người thiết kế cần phải lấy các giá trị đo đạc từ các IMU đơn trục rồi tính toán cho trọng tâm của khối INS. Điều này gây ra các sai số tích lũy lớn và làm giảm tốc độ tính toán của hệ thống. Với sự ra đời của i-Sensor 6-DOF MEMS (Analog Device, 2008), việc tính toán trở nên đơn giản hơn do loại bỏ được phần tính toán chuyển tọa độ cho trọng tâm hệ INS. Tháng 9 năm 2010, Mark Petovello (Inside GNSS) đưa ra mô hình tính toán trên miền rời rạc z-Domain. Kể từ đây các công thức tính toán phức tạp cho mô hình PVT dần được chuyển sang tính theo phương pháp này - phương pháp tính toán cho mô hình PVA. Tuy nhiên, phương pháp này vẫn chưa đưa ra được chính xác góc sai lệch ban đầu và phương pháp bù sai số một cách linh hoạt cho hệ MEMS INS. So với các hệ thống trước đây, thiết kế mới này của tôi sử dụng thêm thiết bị tích hợp là một la bàn điện tử số để bù cho các tham số tính toán của hệ MEMS INS. Các giá trị đo đạc bởi hệ MEMS INS được giới hạn bởi các điều kiện hạn biên. Phương pháp này tính toán cho hệ thống dựa trên miền rời rạc cho mô hình PVA, sử dụng các hàm hạn biên xử lý bù sai số trên cơ sở lý thuyết điều khiển mờ. Điều này giúp đơn giản trong tính toán, giảm bớt số lượng các tham số bất thường, qua đó nâng cao tốc độ tính toán và độ chính xác của hệ thống tích hợp. Phương pháp này có thể áp dụng cho các hệ thống tích hợp MEMS INS/GPS ngay cả trong trường hợp mất GPS trong thời gian ngắn. Tại thời điểm này, tôi đã và đang thực hiện một số thử nghiệm, điều chỉnh các tập mờ và các quy tắc theo từng bước để hạn chế lỗi vị trí. Mục tiêu của tôi là để có được giá trị sai số nhỏ hơn 10 mét khi mất tín hiệu GPS trong thời gian 2 phút. Đây là một phương pháp mới chưa từng được công bố trong những thiết kế trước đây. Key words Integrated INS/GPS, Kalman filter, Fuzzy logic control, Calculation in z-Domain 412 Ngo Thanh Binh VCM2012 Symbols Symbols Unit Meaning F, L constant matrices, which characterize the behavior of the model w(t) white noise process with a power spectral density Qc )(t  process noise  ,  ,  rad/s roll, pitch, yaw angles Abbreviations 6-DOF Six degrees of freedom KF Kalman Filter GPS Global Poistioning System GNSS Global Navigation Satellite System MEMS Micro-Electro Mechanical System IMU Inertial Measurement Unit INS Inertial Navigation System UAV Unmanned Aerial Vehicle 1. Introduction Inertial Navigation Systems (INS) are used for purposes of motion balance control, vibration measurement, calculation of acceleration, velocity and distance moved. The INS system based on Micro-Electro Mechanical Systems (MEMS) provides only information about the exact accelerations and tilt angles, from which we can calculate the values of the velocity and position of moving objects with high speed. However, these values are accurate within only a short time. Moreover, calculating the tilt angles (roll, pitch and yaw angles) from MEMS INS system helps control balance of moving objects. Global Positioning System (GPS) provides position and velocity information for a long time. It is not affected by gravity but gives out the results with low speed calculation and can identify only the heading angle but other angles. Monitoring control systems that use GPS only will have no effect when the GPS signal is lost. Thus, using GPS, Galileo, Glonass or Compass system integrated in controller systems is the way needed to be concerned so much. 2. Overview In the past, when bringing the integrated MEMS INS/GPS system into practice, many scientists not only encountered theory problems with the complex computation formulas leading to the fact that it was difficult for programming to meet the real-time, but also their works were limited by the technology because at that time there were only the single axis sensors. They had to use some single axis sensors integrated to make a 6-DOF (six degrees of freedom) sensor. Some commonly used kinds of sensors include IMU of Crossbow, Hibot, Honeywell, Analog Device, Watson, etc. In June 2007, Analog Device introduced INS ADIS 16350 integrated 6-DOF being one of the technological advances that helped the motion control problems using MEMS sensors widely develope with higher quality. Several other companies also introduced the 6-DOF sensor around this time. With powerful processors and calculating method in the z-domain, the 6-DOF sensors open a new solution for this problem. To determine the initial direction angle when the object starts moving or receives again GPS signal after losing in a short period of time, the system also needs to be integrated with a compass. In November 2008, Analog Device introduced ADIS 16405 sensor integrated with an electronic compass to determine the angle of the motion direction (heading angle). It helped develop more completely the motion control. To improve the accuracy of output signals, a filter is often used in integrated MEMS INS – GPS system, and most commonly is Kalman filter. The different forms of the Kalman filters such as KF (Kalman Filter), EKF (Extended Kalman Filter), UKF (Unscented Kalman Filter) give out the different calculation methods, mainly applied to model PVT (Position - Velocity - Time) with the combined MEMS INS/GPS system integrated with the other sensor systems. There are some typical calculation models presented by E-H. Shin (2005); Naser E-S. (2008); G. Lachapelle and M. Petovello (2008); J. Crittenden (2008); T. Takasu (2008), C. Goodall (2009); T. Li (2009); A. Soloviev (2010); M. Jew (2010); D. Sun (2010); B. Aminian (2011); A. Angrisano (2012), etc. Most of these calculation methods primarily use single axis sensors integrated to create a multi-axis sensor. Therefore, they must calculate and convert to get the position of the center of the sensor system and apply to the continuous models with the complex calculation formulas. However, when putting into practice, many authors use simple models and remove complex calculation components in the theoretical model. When the multi-axis sensors occurred with the given digital signals, the calculation methods became simpler, more accurate and higher speed. Some models are switched to calculate on the z-domain applying to ground equipments such as autonomous vehicles, robots, as well as the models applying to aircraft equipments or Unmanned Aerial Vehicle (UAV) systems. Tuyển tập công trình Hội nghị Cơ điện tử toàn quốc lần thứ 6 413 Mã bài: 100 3. Calculating PVA model in z-Domain 3.1 Amplitude conditions Most MEMS INS systems possess high accuracy in short time, but they will generate large errors when working independently for a long time due to the property of error accumulation. The cumulative errors will increase, so the amplitude errors will grow. However, these errors in every mutation can not be beyond threshold levels. In case of loss of GPS signal, the MEMS INS system works properly in a short period of time and then loses accuracy gradually. So, making the maximum levels for signals and making flexible rules for received data in each specific MEMS INS system is a matter need to be concerned. For vehicles, the maximum acceleration is calculated by using formula: ga   max , where 1   . With 1 max   , the condition 2 max /81.9 sma  is always satisfied. To reduce calculation, for moving objects on the ground, we limit acceleration under 2 max /81.9 sma  , and velocity under 300 km/h. So, knowing max v , in one step s T , we will calculate amplitude velocity max v (in fact max v ). Multiplying max v with s T we have the maximum amplitude moving distance max p . In a period of time when velocity and position values beyond max v and max p , we calculate with max v and max p . This method eliminates the calculation of abnormal data errors. It not only helps reduce the abnormal data needed to calculate but also increases the accuracy of the results because it does not need to handle false results exceeding with certain levels. 3.2 PVA Model PVA model in s-domain is described by Laplace operator (Fig.1) Fig. 1 State model in Laplace domain The form of linear time-invariant models is described with continuous-time state equation: )()()( t Lw t Fx t x    (1) where:    T tatvtptx )()()()(  are the initial conditions, with p(t), v(t), a(t).  Conditions of inertia x(0)~N(m(0),P(0), calculated from electronic compass and error  , are inertial values of error system )( fH  .  F and L are constant matrices, which characterize the behavior of the model  w(t) is a white noise process with a power spectral density Qc (survey of theoretical models). According to Mark Petovello (2010), the form of PVA model in s-Domain is: )( 1 0 0 )( )( )( 000 100 010 )( )( )( t ta tv tp ta tv tp                                               (2) We can obtain the continuous time state by solving the different inhomogeneous systems of above equations with following formula:    t tFFt dgexetx 0 )( )()0()(   (3) Where Ft e is the matrix exponential of F, and x(0) is the initial condition. 3.3 s-z transformation and calculation in z- Domain The typical approach to obtain the state according to (3) is expanded using a Taylor series expansion, with s T is a step, 3 I is identity matrix, as follows:  !2 2 3 s s FT FT FTIe S (4) The final formula calculated for PVA model often found in literature is: )()( 1 2 ]1[ ]1[ ]1[ 100 10 2 1 ][ ][ ][ 0 2 2  dv v na nv np T T T na nv np S T s s s                                                           (5) In signal processing there are many laws dealing with different discretion methods. According to IID method (Independent and Identically Distributed random variables), analog white noises and digitized white noise in convenient models )( fH  will be divided into 2 ranges of s f (two- sided bandwith) and values ][n  are calculated by )( fH  . The power spectrum of the noise sequence is constant in the bandwith        2 , 2 ff , 414 Ngo Thanh Binh VCM2012 the variance of ][n  will be 2 0 2 s FN    . Transforming to z-Domain with generic input x[n] = x(nT s ), and H a (f) to H(z), we can capture output signals y[n] in y[nT s ]. The output signal in the analog branch y(t) in formula (2) will be expended:     dfefXfHty ftj a  2 )()()( (6) Description of (6), we have:     dfefXfHnynTy S fnTj as  2 )()(][)( (7) According to Papoulis (1997), formula (7) will be written as:     dfeeHfXny SS fTjfTj  22 )()(][ (8) Compared (8) with (7), it leads to the following conditions for the simulation theorem if x(t) is a band-limited signal: 1. x BfwhenfX  0)( 2. xa fTj BfwhenfHeH S  )()( 2  We use transformed rules s sH a 1 )(  , with the transformation from the continuous domain to the discrete domain in accordance with the applicable laws as follows: Fig. 2 Transformed laws Fig. 3 Transformation from s-plane to z-plane Fig. 4 State model in z-Domain 3.4 Designing devices and developing applications In the field of railways, as we all know, at positions where the platform is weak, or ties of the rail track are broken, the train movement will fluctuate accordingly. In order to measure and monitor the downgraded rail ways, we can put measuring devices on the train to record fluctuation values such as tilt angles. The controllers will record data, calculate parameters, and transmit result to the central station. Based on the recorded data about the oscillation amplitude of the train, we can point out the state of the railway and maintenance requirement. Fig. 5 Diagram of management and monitoring system Tuyển tập công trình Hội nghị Cơ điện tử toàn quốc lần thứ 6 415 Mã bài: 100 Technical property of the method is to use sensor MEMS INS to identify acceleration and tilt angles compared to object’s coordinate axis. Oscillation of moving object will be computed and displayed in graph embedded in Google map. This process was combined with positioning algorithm to provide state of railway and suggest the maximum velocity allowing for moving trains on different rail tracks. For illustration purposes, I designed and created devices integrated GPS module ET333 (SiRF III) and MEMS INS ADIS 16354 with 32-bits microprocessor AVR32 UC3A programmed on IAR Embedded Workbench. These devices can measure automatically the oscillation, check the rail-way as well as position and supervise the running trains. Calculating in z-Domain with high speed processing microcontroller, we considered that errors )(t  is constant in the intervals ])1(:[ ss TnnT  . Finally, the formula calculated for PVA model in z-Domain is: ][ 2 6 ]1[ ]1[ ]1[ 100 10 2 1 ][ ][ ][ 2 3 2 n T T T na nv np T T T na nv np s s s s s s                                                             (9) Where: ][n  are system errors. In practice, we set the device, in which MEMS INS sensor is mounted, in a static state. The noises are examined by using stationary objects. The output parameters of noises are measured and saved in an error table over time. There are random parameters affected by the gravitational acceleration, the Coriolis Effect, random noise and other factors. For example, the Bias (from accelerator) and Drift (from gyro) of IMU, initial angle from north axis n  , the impact of temperature, as well as the impact of the magnetic and thermal hysteresis or vibration. These values ][n  multiply with step time (T s ) to generate scalar errors of acceleration and gyro. When calculating, we need to estimate these errors in a period of time, from that we access the error table and calculate corresponding offset values in each time based on fuzzy logic, according to (9). In fuzzy logic method, we define and calculate as follows:  Input signal Fig. 6 Fuzzy sets of velocity calculated by INS Fig. 7 Fuzzy sets of deflection angle between INS and Compass   Output signal Fig. 8 Coefficient of output signal   Rules based on CoM – Center of Maximum If Vel = VS And Change = Down Then Level = 2 If Vel = VS And Change = No Then Level = 3 If Vel = VS And Change = Up Then Level = 2 If Vel = S And Change = Down Then Level = 2 If Vel = S And Change = No Then Level = 3 If Vel = S And Change = Up Then Level = 2 If Vel = M And Change = Down Then Level = 2 If Vel = M And Change = No Then Level = 3 If Vel = M And Change = Up Then Level = 2 If Vel = F And Change = Down Then Level = 1 If Vel = F And Change = No Then Level = 2 If Vel = F And Change = Up Then Level = 1 If Vel = VF And Change = Down Then Level = 1 If Vel = VF And Change = No Then Level = 2 416 Ngo Thanh Binh VCM2012 If Vel = VF And Change = Up Then Level = 1 We programme to calculate velocity, deflection angle between INS and Compass to give out suitable levels. Offset values will be calculated by multiplying these levels with suitable data in errors table in period of time. Values of these levels, even the quantity and quality of the rules can be variant depending on actual process corresponding to each train, running routine, position of measurement devices on that train, etc. Fig. 9 Calculating offset parameters based on fuzzy logic Fig. 10 Real devices in the train and at the station Fig. 11 Applications in monitoring and management the running train Fig. 12 Oscillation angles Abilities of developing these devices are quite diverse, as in the management and supervision of ground vehicles, especially the transportation means with particular requirements, such as speed management and supervision of the journey road equipment, railway, etc. Tuyển tập công trình Hội nghị Cơ điện tử toàn quốc lần thứ 6 417 Mã bài: 100 4. Conclusion The fundamentals of previous problems of processing data for the classic model PVA is to use controller units in time-invariant range in s domain. The new algorithm is developed by calculating continuously in discrete domain in the condition of limitted error signals with suitable amplitudes. In addition, this new method determines the exact initial deflection angle to the north axis of the earth, and also indicates how to make data processing based on fuzzy theory in the case of loss of GPS signal. These are new in this method, not only help reduce abnormal data needed to calculate and smooth the orbit closest to the real trajectory without mutations of distances, but also give out position, velocity, and tilt angles of moving objects. Based on these parameters, we can give out the state of the railway and maintenance requirement. From that, we can point out the maximum velocity allowing for moving trains on different rail tracks. In the practice, I have been trying to adjust the fuzzy sets and rules to limit position error less than 10 meters when GPS signal is lost during 2 minutes. My target is to improve the quality of the positioning and management of moving objects, mainly for ground vehicle applications, so I do not mention the parameters of true attack  and sideslip  . In fact, the problems of monitoring and management of running train on rails do not mention these parameters. These factors just occurred when the train turned over from the rail. In this case, the parameters of angles are recorded in the memory of the equipment (like a black box) on the train and transmitted to devices at the host station. So, it does not need to include in the normal vibration calculation. These factors need to be studied further to fully develop supervision control problems and in order to apply for objects on the sea or flying objects. Development of this design offers a new approach in the research and provides wide applications to the navigation problems. Reference [1] Papoulis: Signal Analysis. McGraw-Hill, 1997 [2] S. Rönnbäck: Development of a INS/GPS navigation loop for an UAV. Lulea University of Technology, 2000 [3] E-H. Sin: Accuracy Improvement of low cost GPS/INS for land application. Calgary, 2001 [4] N.M. Faulkner; S.J. Cooper; P.A. Jeary: Integrated MEMS-GPS navigation systems. Plymounth, 2002 [5] K. J. Walchko: Low Cost Inertial Navigation: Learning to Integrate Noise and Find Your Way. University of Florida, 2002 [6] David H. Titterton; John L. Weston: Strapdown Inertial Navigation Technology - 2 nd Edition. The Institution of Electrical Engineers, 2004 [7] G. Welch; G. Bishop: An Introduction to the Kalman Filter. Chapel Hill, 2006 [8] B. Min and S. Gao: iSensor. Analog Device, 2008 [9] Ngo Thanh Binh: Development of smoother tool RTS in identification of moving objects. Transport and Communications Science Journal, No. 27, pp. 112-117., 2009 [10] Nguyen Thanh Hai: Research, design and manufacture of GPS devices for monitoring and management of road transport, railways scientific state technology - KC.06.02 program, "Research, development and application of advanced technologies in the production of key export products." Code KC.06.02/06-10. Branch No. 6., 2009 [11] Thanh Binh Ngo; Hung Lan Le; Thanh Hai Nguyen: Survey of Kalman Filters and Their Application in Signal Processing. International Conference on Artificial Intelligence and Computational Intelligence AICI'09, China, Vol.3, pp 335-339., 2009 [12] M. Potovello: A fully digital model for Kalman filter. Inside GNSS, 2010 [13] Hung Lan Le; Thanh Hai Nguyen; Quang Tuan Nguyen; Thanh Binh Ngo: Introduction of TBN Method in z-Domain Using in Signal processing, International Journal of UTC- MADI-SWJTU, No. 3. pp 13-21., 2011 [14] http://www.cs.unc.edu/~welch/kalman/ [15] http://www.lce.hut.fi/research/mm/ekfukf/ [16] http://www.insidegnss.com/ [17] http://plan.geomatics.ucalgary.ca/search_pubs_r esults.php Ngo Thanh Binh received Instrumentation Engineer degree in 1998 and M.Sc. degree in Instrumentation and Control in 2001 from Hanoi University of Science and Technology (HUST). From 1998 to 1999, he worked as programmer at the Centre for Automatic Engineering Research (nowadays: High Technology Centre), HUST. Since 1999 up to now, he has been 418 Ngo Thanh Binh VCM2012 working as lecturer at the Electronic Engineering Section, Faculty of Electrical and Electronic Engineering, University of Transport and Communications (UTC). His main research lines and teaching subjects include: Digital Signal Processing; Control and applications of Micro- Electro Mechanical Systems (MEMS) for mechatronic systems and embedded systems; Microprocessor and Microcontroller, Programming Languages (PLC Siemens, Assembly, C/C++ and C#), Technical Informatics (Matlab, Designing hardware using Altium Designer), Embedded Systems, RTOS (Real Time Operating Systems), Robotics. M.Sc. Ngo Thanh Binh has been doing his Ph.D. project at the UTC since Dec. 2007. . Improvement of the quality of integrated INS/GPS devices based on fuzzy logic Nâng cao chất lượng của thiết bị tích hợp INS/GPS trên cơ sở logic mờ PhD Student NGO Thanh Binh University of Transport. toán, giảm bớt số lượng các tham số bất thường, qua đó nâng cao tốc độ tính toán và độ chính xác của hệ thống tích hợp. Phương pháp này có thể áp dụng cho các hệ thống tích hợp MEMS INS/GPS. MEMS INS. So với các hệ thống trước đây, thiết kế mới này của tôi sử dụng thêm thiết bị tích hợp là một la bàn điện tử số để bù cho các tham số tính toán của hệ MEMS INS. Các giá trị đo đạc bởi