ACCELEROMETER BASED ATTITUDE ESTIMATING DEVICE

Transcription

1 Proceedings of the 2004/2005 Spring Multi-Disciplinary Engineering Design Conference Kate Gleason College of Engineering Rochester Institute of Technology Rochester, New York May 13, 2005 Project Number: ACCELEROMETER BASED ATTITUDE ESTIMATING DEVICE Robert Wible Mechanical Engineer / Team Leader Royce Abel Mechanical Engineer / Chief Engineer Brandon Roth Mechanical Engineer Erin Long Mechanical Engineer ABSTRACT This paper details the design and development of an accelerometer based attitude estimation device. A custom circuit board and test stand were fabricated to support a revolutionary theory for attitude estimation based on local gravitational vectors derived from accelerometer measurements. NOMENCLATURE DAQ GPS INS PCB PCI PWM RIT VI INTRODUCTION Data Acquisition Global Positioning System Inertial Navigation System Printed Circuit Board Peripheral Component Interconnect Pulse Width Modulator Rochester Institute of Technology Virtual Instrument Inertial Navigation Systems are used for a multitude of applications, including aircraft, ground units, autonomous robots, etc. Current commercial systems contain gyroscopes and accelerometers to measure angular velocity and linear acceleration, respectively. Modern practice uses Euler kinematic relationships defined by the following to produce attitude angle estimations: & θ = & φ = ψ& = ~ ~ q cos( φ ) r sin( φ ) ~ ~ ~ p + [ r cos( φ ) + q sin( φ )] tan( θ ) ~ ~ ~ [ r cos( φ ) + q sin( φ )] sec( θ ) (1) where θ, φ, and ψ are the Euler angles, and q, p, r are pitch, roll, and yaw rates, respectively. Complexities arise in this method when errors occur in the accelerometer and gyro readings, such as drift effects. When a double integration is performed, the effect of such errors is propagated exponentially. To resolve the issue of incorporating expensive hardware components to minimize errors, scientists have researched accelerometer schematics in keeping with a goal of a low-cost system [1, 2, 3]. A new concept in Fig. 1 has been proposed by Dr. Agamemnon Crassidis, an associate professor at RIT, encompassing a circular arc of accelerometers that measure local gravitational vectors and redefines the attitude initial conditions to determine the pitch angle, roll angle, and yaw angle Rochester Institute of Technology

2 Proceedings of KGCOE 2005 Multi-Disciplinary Engineering Design Conference Page 2 rate gyro mounting point sz 1 2 Z X s X linear accelerometers Figure 1. Proposed device schematic for estimating attitude in a two-dimensional space The primary goal of this project focused on fabricating a prototype accelerometer based INS with a gyroscope for determining initial conditions and a corresponding test stand to validate the proposed theory in only the pitch angle. The resulting unit will be integrated with an autonomous robot to aid in traversing a non-gps friendly 3D environment as a future work concept. DESIGN PROCEDURE The attitude estimation device was developed according to a stringent concurrent design procedure encompassing the following gates: 1. Needs Assessment 2. Concept Development 3. Feasibility Assessment 4. Preliminary Design and Analysis 5. Detailed Design 6. Prototype Fabrication 7. Testing 8. Final Design Each gate was vital in the design procedure, enabling the group to focus on the requirements set at hand by the sponsor. The board and test stand had to withstand a force of 10 gees to simulate a dynamic maneuver. The test stand and board each had their own requirements, which will be discussed in further detail. ACCELEROMETER BOARD DEVELOPMENT +/- 2.0 gee capability. A total of 26 resistors and 14 capacitors were also incorporated into the design. The resistors defined the PWM properties and the resistors were for incorporating on chip filtering. The accelerometers will output a reading every 10ms. The board is a four-layer PCB that includes a ground and power plane. Working from the base of the board to the top, there will be a printed wires/power/ground/printed wires layout and electrical components. The ground plane was placed between the power and components to try to reduce any type of noise that may occur from local fields. There were many attempts to try to reduce any noise to the accelerometers. There was a 5V regulator to maintain constant input. After this state a 0.1 nf capacitor connected the ground and power plane to reduce the rippling effects. There is also an optional 2 nd filtering capacitor that wasn t included in the final build that could be used to filter more noise out of the power supply. This connects the 9V input and ground plane. This noise didn t seem to show and the capacitor wasn t included. A final noise minimization technique implemented was to make sure that the printed wire tracks made less then or equal to 45 deg angles. Doing this made sure that the signals had a clear path on the tracks and didn t interfere with themselves. After all these precautions the team concluded that as much noise is out of the board as possible. For the gyroscopic component, the GyroPAK 3 was chosen from O-Navi, LLC. The component not only included a gyroscope but additional filtering which was ideal in maintaining the low noise input requirement. See Fig. 2 for the finalized board layout. Additional requirements were placed on the actual circuitry to fabricate the device. The board itself had to have the ability to mount on an already existing autonomous robot in the sponsor s possession. Several software design packages were explored to produce the board. The team decided upon PCB123 s software package to construct the intricate design. PCB123 was chosen because of its unique ability to rotate the accelerometer components on the board, corresponding to the design intent in Fig. 1. The board itself consists of numerous components necessary for operation. Analog Devices ADXL202J accelerometers were chosen for their PWM output and Figure 2. Finalized PCB layout Included on the board are 4 mounting holes that will be used to mount it to the test and as well as mount it to the autonomous robot as discussed before. The mounting of the hardware was assisted by the SMT Lab in the CIMS building. There assistance was critical to the proper layout of the accelerometers. Fig. 3 displays the fully populated PCB. Paper Number 05011

3 Proceedings of the 2005 KGCOE Multi-Disciplinary Engineering Design Conference Page 3 INTEGRATION & TESTING Once the test stand and accelerometer board were completed, integration of the units commenced. The first task was successfully mounting and aligning the board with the test stand. The addition of a flat on the rotating shaft allowed it to be locked into position, and a locating bolt in the mounting ring allowed for fixing the zero deg angle of the accelerometer board. Figure 5 shows those mechanisms. Figure 3. Populated prototype PCB TEST STAND DEVELOPMENT The test stand was originally designed as a platform that would have a single rotating shaft and a potentiometer to gather a truth reading of the rotational displacement. As the design progressed, adjustable mounts were added to allow for leveling adjustments to the stand in case it was placed on a non-flat surface. The initial intent was to incorporate a potentiometer for the truth reading. Further background research yielded the inaccuracies of lowcost potentiometers, thus forcing the team to explore other venues. A digital optical encoder was the finalized component. The encoder has division every 0.5 deg but with LabVIEW software it can be estimated to 0.25 deg, which falls well within the 1º requirement imposed by the sponsor at this iteration. The hollow rotating shaft is housed in three bearing stands to provide stability while traveling. The bearings are slightly offset, as the shaft was not perfectly cylindrical. Also, two exit holes were drilled into the shaft to allow clearance for the electrical wiring connecting the accelerometer board to the data acquisitioning device. (Fig. 4) Figure 4. Hollow rotating shaft Figure 5. Zeroing mechanisms With the hardware now zeroed out to the local environment, the electrical components needed to be zeroed. The capacitors that define the filtering for the output of the PWM for output of each axis have a tolerance of +-10%. This resulted in a constant error for each accelerometer someplace between this value. This was achieved by setting the board to the downward (0 degree) position and setting each accelerometer to the expected value that that should be read out. Implementation technique will be explained later in this section. With the accelerometers now zeroed out the readings are available to be acquired and saved for post processing to determine the angle based on the accelerometers and then compare that to the actual readings based on the encoder. The LabVIEW file is too large and complicated to actually show a screenshot here. Therefore, a thorough explanation of the process gone through will be discussed. With the most recent version of the LabVIEW software, the developers included a new collection of tools as well as a new data type called dynamic data. This dynamic data stems from the waveform type that was used before and were primarily build to be used with the new set of tools mentioned above which are called VI Assistants. As a side note, many of these VI assistants will end up using the waveform data type in the sub vis, so the total reason for this new data type, which does make some processes that would have been simple in the past more difficult. These VI Assistants will be used extensively in the acquisitions of the readings. Copyright 2005 by Rochester Institute of Technology

4 Proceedings of KGCOE 2005 Multi-Disciplinary Engineering Design Conference Page 4 There are two PCI boards that will be used to acquire the data. The first board is a multifunctional DAQ card and the second is an 8 port counter card. The multifunctional card will be used to acquire the data from the accelerometers as well as the gyro and the counter card will be used to keep the index of the encoder. Therefore, initially in the block diagram, two VI Assistants (DAQ Assistant) are implemented. The first acquire the data from the accelerometers and gyroscope at a rate of 20000Hz every 2000 samples. This allows retrieval of an excellent waveform of the PWM. The counter card just outputs the current count of the encoder. The counter card is actually looking at both the high side and the low side of the encoder pulses to half the resolution of the encoder from.5 deg to.25 deg. The rest of this discussion will be broken into three sections: the manipulation of the accelerometer readings, gyroscope, and rotary encoder. Once the accelerometer reading is demuxed from the gyroscope it is sent through Timing and Transition VI Assistant to output of the duty cycle readings. These signals are then filtered using the Filter VI Assistant to calculate a moving average smoothing function. These duty cycle values are then added to a correction factor to take out the capacitor tolerance errors discussed before. These zeroed signals are then sent through a simple function that relates the duty cycle value to the relative gee reading. These are then the final readings that are saved. Some notes about this process. First, the duty cycle VI assistant is very complex. Because of resistor tolerance that determine the actual timing of the accelerometer output, each waveform would have a different time step if each value were converted to an actual reading. To get around this, the Timing and Measurements VI Assistant actually stores the last 10 readings in a queue, determines the duty cycle for each, averages them, and then determines a common time step for each accelerometer. Therefore, since each of the accelerometers in this system were reading a rate of every ~10ms. After going through this block the readings were now 0.1 seconds. This simple complication was very discouraging because it was not documented in the LabVIEW software and resulted in fewer readings then was preferred. It was decided that the current system would stay as is, and if more readings were desired then the resistors and capacitors would be changed at a later late. To calibrate the accelerometers in LabVIEW is based on a simple process. The addition term that is mentioned above is actually an array of waveforms that are created from a set of constants that can be set on the Front Panel of the vi. These values are made into waveforms built with the same sample rate as the accelerometers and then just added into the readings. So, the process needed for this to work is as followed: variable, simple vector, waveform creation, waveform array, and then finally conversion into dynamic waveform to be added into accelerometer readings. To determine the values to go into the front panel boxes is by comparing the current value which is displayed by using a Sample Compression VI Assistant (5 sample Mean) and calibration values from theoretical calculations. The gyroscope readings are much simpler to process. After the data is demuxed from the accelerometer data, it is sent through a Resample VI Assistant. This is because initially this was retrieved with a time increment of sec. This rate is too fast when compared to the accelerometer data. The time step was re-sampled to 0.1 seconds. The modified data was then sent through a similar moving average filter as the accelerometers and recorded to file. The rotary encoder data is sampled differently then the others components. It does not have a standard time step or sample frequency because there is no internal clock built into the board. Therefore, the data is just recorded each time the counter is checked by the software. Since the time increments were inconsistent, this data was sent through a similar resample block to do a linear interpretation to determine the values at.1 increments. This data was then recorded to file. This data is then examined with MATLAB to determine the pitch reading and compare it to the encoder. The below figures show sample outputs that result from running the LabVIEW vi that was developed. Figure 6 is the accelerometer data shown for a +90 to 0 deg rotation at a fairly constant rate. Figure 7 shows the full run from the previous segment of the rotary encoder. Figure 6. Accelerometer outputs for a +90 to 0 input maneuver Paper Number 05011

5 Proceedings of the 2005 KGCOE Multi-Disciplinary Engineering Design Conference Page 5 Figure 9. Simulated accelerometer readings for a +45 input maneuver Figure 7. Rotary Encoder outputs for a +90 to 0 input maneuver ANALYSIS The accelerometer readings and a sampling time rate are outputted into an Excel file in gees and seconds, respectively. Figure 8 shows accelerometer readings for a + 45 input maneuver at a sampling rate of 0.1 seconds. Both figures exhibit the same behavior for the accelerometer readings, with the prototype data generated at a faster rate. Effects from noise are apparent in the unsteadiness of the data collected from the prototype. Once the GyroPak3 has been fully integrated, those measurements will be incorporated into the simulation to induce a dynamic bias and to produce a theta estimate value that will be compared with the truth model from the digital encoder readings. CONCLUSIONS The data gathered from the prototype replicates the trends predicted by the theoretical simulation. The accelerometer device fulfilled the requirements set by the sponsor, with a resolution of 7.5 that can be improved upon with additional accelerometers and incorporating a mounting design that would fit on the autonomous robot. ACKNOWLEDGMENTS Figure 8. Accelerometer readings for a +45 input maneuver Figure 9 demonstrates the simulations results for a + 45 input maneuver utilizing the following equation to simulate the individual accelerometer measurements: ga 2 [ gaz, cg qxz q z Z ]*cosθ Z = & (2) Z, i + The work here was performed and completed thanks to the support of Dr. Agamemnon Crassidis and Dr. Wayne Walter. The team would also like to thank Professor Wellin for allowing the use of his lab and equipment for testing purposes. REFERENCES [1] Q. Wang, M. Ding, and P. Zhao, A New Scheme of Non-gyro Inertial Measurement Unit for Estimating Angular Velocity, IEEE 29 th Annual Conference Industrial Electronics Society, vol. 2, pp , [2] C.-W. Tan, S. Park, K. Mostov, and P. Varaiya, Design of Gyroscope-Free Navigation Systems, IEEE Intelligent Transportation Systems ConferenceProceedings, 2001 pp [3] K. Mostov, A. Soloviev, and T.-K. Koo, Initial Attitude Determination and Correction of Gyro-Free INS Angular Orientation on the Basis of GPS Linear Navigation Parameters, Intelligent Transportation System, USA: 1997 IEEE, pp Copyright 2005 by Rochester Institute of Technology

6 Proceedings of KGCOE 2005 Multi-Disciplinary Engineering Design Conference Page 6 Paper Number 05011