TEMPERATURE EFFECT ON ACCELEROMETERS FOR ROBOTICS POSITION SENSORS

Transcription

1 TEMPERATURE EFFECT ON ACCELEROMETERS FOR ROBOTICS POSITION SENSORS CASSANDRA EGGETT JEFFREY LUTSEY DAVE PAKULA FERNANDO ZUMBADO MAY 2001 ME 224 PROFESSOR ESPINOSA NORTHWESTERN UNIVERSITY

2 TABLE OF CONTENTS INTRODUCTION 2 THEORY 2 Figure 1: Accelerometer modeled 2 Strain-Gage 3 Figure 2: Cross-Section of piezoresistive sensor 3 Accelerometer Configuration 3 Description of Experimental Accelerometer 3 Operation of Experimental Accelerometer 3 Figure 3: Joint motion for angular acceleration 4 Effect of Temperature on Accelerometer 4 INSTRUMENTATION 5 Experimental Setup 5 Figure 4: Joint specific setup 5 Figure 5: System wide setup 6 ANALYSIS 6 Comments on Methods Used 6 Supplementary Cantilever Method 7 Thermistor Methods 7 CONCLUSION 8 REFERENCES 9 APPENDIX 10 APPENDIX A: Illustrations 10 Figure A.1 ICSensors, model Figure A.2 ICSensors, model 3140, circuit diag 10 APPENDIX B: Specifications 11 Specification B.1 RRC model 807i 11 Specification B.2 ICSensors model AUTHOR BIOGRAPHIES 12 NSF GRANT PROPOSAL DOCUMENTATION 1

3 INTRODUCTION Much of today s industry relies strongly on a multitude of sensors in order to direct the actions of its machine components. One of the most important of such devices is the position sensor, providing data necessary for precise motions in a number of critical applications, ranging from repetitive object manipulation on an assembly line to repairing a solar panel on the International Space Station. Such position sensors can be created out of a number of different sensor and circuit combinations including laser encoders, torque sensors, strain gages, and accelerometers. The accelerometer, though requiring a significant amount of signal processing, provides a high level of accuracy in a very small package. Complex machines and robots function in many different environments, and are, thus, required to maintain their precision as the surrounding atmosphere alters (e.g, humidity, temperature, and pressure changes). The most significant environmental impact on the functioning of accelerometers is temperature fluctuation. Due to the extreme temperatures accelerometers can be subjected through a robotics application, ranging from the exceptionally hot (e.g. nuclear power plants) to the exceptionally cold (e.g. space applications), a comprehensive understanding of temperature s effect on accelerometer functioning is required. The recent paper Experimental Robot Postion Sensor Fault Tolerance Using Accelerometers and Joint Torque Sensors discusses the application of a piezoresistive accelerometer. Using the aforementioned manuscript as a premise and foundation, a method is proposed to correct for the temperature effects in the accelerometers, and thus, maintain measurement consistency in varying environments. THEORY Accelerometers provide a method of position sensing that has the advantage of a very small package as well as minimal power consumption. Accelerometers are a fraction of the size of other position sensors, such as optical encoders or torque sensors. Additionally, they can be mounted on an arm, rather than the joint itself. In general, accelerometers sense acceleration by using a proof mass on which external acceleration can act. By measuring the displacement of this proof mass, the force exerted can be calculated. An accelerometer, in theory, is composed of the following sections depicted in Figure 1: a proof mass, a spring, a damping constant and a hard limiter [5]. By equating the force generated by acceleration to the force necessary to deflect the spring the acceleration can be calculated: F = m a = k x mass (1) where k is the stiffness coefficient of the spring and x mass is the displacement of the proof mass relative to the accelerometer frame. In practice, the stiffness constant corresponds to the material stiffness (Silicon for the current application). Figure 1. Accelerometer modeled 2

4 The spring also serves to return the proof mass to the intial position after the acceleration disappears [5]. Silicon wafer micromachined accelerometers have a proof mass that is supported by bending beams. These beams act as the spring element for the proof mass. controls overrange oscillations, [6]. The lower glass plate and the overrange stop constitute the hard motion limiters depicted in Figure 1. The strain gauge is positioned at the back of the support beam of the proof mass since it was determined through finite element modeling that the maximum stress concentration is located at this point [6]. Strain-Gage Accelerometer Configuration Description of Experimental Accelerometer Modern accelerometers are composed of a micromachined silicon wafer enclosed by two protective plates. As illustrated, the enclosure plates can be 7740 Pyrex glass wafer that are anodically bonded to the silicon wafer. Notice the cavity isotropically etched in the enclosure plate (only lower plate shown in Figure 2) [6]. This allows motion of the silicon proof mass necessary for acceleration value calculations as shown in (1). The sensor implemented in Aldridge and Juang s Experimental Robot Position Fault Tolerance is the ICSensors, model A simple Wheatstone bridge (APPENDIX, Figure A.2) uses the chip s behavior to momentum changes in relating accelerations of the body, on which it is attached, to slight fluctuations in voltage through the device s circuit. Used for lowacceleration applications, the 002 dash number was utilized, rated for accelerations of ± 2g, a sensitivity of 1 Volt/g, and operating temperatures of 20 C to 85 C [4]. Operation of Experimental Accelerometer Figure 2: Cross-section of piezoresistive sensor The silicon wafer is micromachined using photolithography and etching. The basic process includes the growth of a silicon oxide later patterned to the desired configuration. Post patterning, KOH etching yields the desired structure. This process is repeated for the other side of the wafer to release the proof mass. A second wafer is bonded to the proof mass structure wafer to prevent the mass from deviating from a specific range. Such a process The accelerometer outputs a voltage that is linearly related to the acceleration. This signal is then fed into a processor which continually evaluates the data, and outputs the acceleration. With a perfect accelerometer, the output could be integrated twice to obtain the precise position of the joint; noise and vibration error in this method, however, lead to poor signal-to-noise ratios and imperfect position output signals. A new method to obtain joint position from Cartesian accelerometers without integration was recently developed by Aldridge and Juang in order to produce greater location precision [2]. 3

5 The theory behind this method is based on a complicated algorithm that numerically calculates the joint position using kinematics. In order to reduce noise, very small accelerations are taken to be zero. Although this assumption appears to introduce more error, the cutoff for readings is set, such that the lowest realistic acceleration for a specific joint will be picked up, while smaller values will be ignored. This method is able to determine position with sufficient accuracy to determine robot arm location. Joint under observation α d Figure 3. Joint motion for angular acceleration a The accelerometers are placed on the component of the robot that is directly controlled by the joint to be measured. By analyzing the Cartesian accelerations, the position of the joint itself can be determined. In order to calculate this position, the accelerometer must be fixed at a known distance from the joint, sufficient enough to cause an acceleration to be easily perceived by the sensors. Also, since simple accelerometers only measure acceleration in one direction, it is necessary that they are oriented to sense the tangential component of the acceleration. (see Figure 3) Using this method, the angular acceleration of the joint can be determined using simple kinematics. α = a d where α is the angular acceleration, a is measured tangential acceleration, and d is the distance from the joint to the accelerometer. The acceleration of the joint can then be run through the algorithm to determine a precise position of the joint. The addition of more accelerometers, with different orientations, can measure the position of the robot arm in its different degrees of freedom. Effect of Temperature on Accelerometer Exposure to extreme temperatures or rapid transition within a temperature-variant environment is inevitable in the robotics applications previously considered. As a result, the accelerometer utilized in this application must be immune to temperature changes or include a compensation parameter to account for the material property changes with temperature. The former assumption would not be valid since most materials experience changes in their properties with temperature variation. Thus we shall explore the second approach. Fast transient temperature changes may cause spurious signals on the sensor output due to thermal expansion of the metal parts. When using poorly designed or specified sensors these spurious outputs can trigger false alarms [6]. Low frequency 4

6 accelerometers and sensors with poor strain sensitivity are much more susceptible to transient temperature effects. Thus, having a component in the accelerometer setup that measures this temperature variation is helpful in compensating for the discrepancies. The piezoresistor placed on the supporting beam has a nominal resistance. Upon deformation of the beam, the piezoresistor will change resistance as a function of the deformation. However, temperature effects can cause deformation in the Si beam and thus alter the resistance reading. When the ambient temperature changes there are four effects that change and influence the signal R/R from the strain-gage: 1. The gage factor Sg 2. Gage elongation ( l/l = α T) 3. Beam elongation ( l/l = β T) 4. Resistance of gage changes ( R/R= γ T) These terms combine to give a change in resistance of the gage that can be expressed as R R T = ( (2) β α) S g T + γγ T The thermal coefficient of expansion for the gage is α, β denotes the thermal coefficient of expansion of the beam and γ is the temperature coefficient of resistivity of the gage [3]. INSTRUMENTATION centimeters long, 7 degrees of freedom, and a 10 kg payload (manufacturer specifications shown in Specification B.1 in APPENDIX). Relevant in this study are either 2 or 3 of the 7 total degrees of freedom, depending on the specific setup prescribed below. Two specific setups were implemented to test the validity of the backup position-sensing technique using two varying methods. The first, called the joint specific setup, involves the measurement of two motions the shoulder pitch and elbow pitch (two degrees of freedom). Figure 4 indicates the placement of the two joints and the motions being mapped. This setup involves two computer processors with VME bus-to-bus connection with the sensors on the RRC robot arm one as the torque information processor, and one as the accelerometer processor. Three orthogonally-mounted accelerometers (i.e. one triaxial accelerometer) sense acceleration in their respective axes at each of the joints labeled. Shoulder pitch Elbow pitch Experimental Setup The testing apparatus in Aldridge and Juang s investigation includes a Robotic Research (RRC) manipulator, model 807i; ICSensors piezoresistive accelerometers, model 3140; and computer processors with VME bus interfaces. The RRC simulates a simplified human arm, including shoulder, elbow and wrist, of approximately 80 Figure 4: Joint specific setup 5

7 A second setup, called the system wide setup, shown in Figure 5, monitors the rotational motions of shoulder pitch, elbow pitch, and elbow roll (three degrees of freedom). A triaxial 3140 accelerometer is attached on each of the three joints. This configuration is being tested because of its ease in mounting, rather than having the most advantageous position sensing locations. Here, three processors are used, one for torque sensing, one for acceleration sensing, and the last for iterating the system of dynamics equations to attain the robot s position. and lowest cost accelerometer, with a range of ± 2g, was ideal for the application. The small size also made easier the mounting of the accelerometers, ensuring no interference with the robot arm s motion. If this experiment were to be implemented, the orthogonal sensors, in reality, would need to be modified and mounted internally, for the size and weight are still too large for the space and nuclear applications being considered. ANALYSIS Figure 5: System wide setup Shoulder pitch Elbow pitch Elbow roll Comments on Method Used A couple of notes must be mentioned, regarding the accuracy of the results, after using the aforementioned experimental method. The signals outputted are insufficiently noisy and still required two low-pass filters. Also, the motor parameters were not specified or available, and, thus, an inferior qualitative relationship was established to ascertain its characteristics. Also, noted in a similar experiment, the measurements are severelydependant upon the preloading of the sensor s screw. It is recommended that (1) the sensor is mounted flat on a machined aluminum block, and (2) the screws are tightened to 6 foot-pounds per inch of screw length [3]. ICSensors model 3140 (Figure A.1 and Specification B.2 in APPENDIX) was chosen in this experiment for a number of reasons. This model offers 99% linearity, slight signal conditioning, and anti-aliasing, improving the signal output. The piezoresistive nature of the model ensured measurements for variant and constant accelerations. Also, because the robot arm in the test experiences (1) small accelerations and (2) small temperature fluctuations, the smallest Found in the Aldridge and Juang [1] paper, strengths and limitations for each of the positioning methods. The joint specific methods provides for a more predictable and computationally simple solution for only one or two joints. On the other hand, the system wide approach provides a much more flexible setup. This technique allows for position sensoring for a wider range of motions. The cost for greater 6

8 versatility comes in with computational complexity and less positioning information on each joint. Of prime importance, the effect of temperature on the accelerometer measurements must be considered. The ± 2g range ICSensors accelerometers were found to have a drop of 0.03 g with an increase in temperature of 10 o C. With a 5-V supply to the Wheatstone bridge, that 0.03 g drop corresponds to a sensitivity drop of 0.3 mv, and,thus, a less effective position sensor. Although the temperature fluctuations were small, the effects were substantial, and the equivalent, more costly temperaturedependent 3140 model offered by ICSensors may be required for different applications. Additions and modifications to the experimental methods discussed may further aid in improving measurement consistency in environments of temperature variation. Discussed below are a couple of potential temperature-compensation techniques. Supplementary Cantilever Method A different approach to temperature compensation is to include an additional cantilever beam. This cantilever would not be attached to the proof mass. As a result, the strain that this beam would experience is a due to temperature variation. A piezoresistor would be placed just as in the cantilevers attached to the proof mass. Therefore, the strain due to the temperature change in the piezoresistor is governed by equation (2). During the micromachining process, this beam would be created close to one of the proof mass beams to ensure that the temperature is similar to that of the supporting beams. For measuring the strain the beam is subjected, as a result of the proof mass acceleration, the additional beam would serve as the correction factor. The strain of the supporting beams includes the proof mass displacement and the strain caused temperature variation. Subtracting the measured free beam resistance from the supporting beam resistance cancels out the temperature variation. This configuration enables the accelerometer to opperate in varying temperature conditions and still provide an accurate measure of acceleration. Thermistor Methods It may be possible to compensate for drastic temperature changes so that the accelerometer position sensing system could be used in more extreme environments. Two remedies have been proposed that could correct the acceleration readings, thereby improving location accuracy of the robot. The error in the acceleration values, recorded outside the operating temperature range, is caused by thermal expansion of the silicon accelerometer. As temperature increases, the piezoelectric strain gauge detects this deformation but processes the data as if an additional acceleration had been applied to the sensor. This deviation could easily be corrected by using a thermistor embedded into the silicon wafer contained the accelerometer. Since thermal expansion is a linear function, the measured acceleration is biased by a factor proportional to the temperature of the silicon. With the thermistor tracking this temperature, a small addition to the processor programming could compensate for the environmental temperature and keep the error within acceptable limits. 7

9 This system could be improved further by using two thermistors in the accelerometer to determine a temperature gradient on the silicon. This gradient could be used to extrapolate the precise temperature of the cantilever arm on the sensor. Although, it would require more processor time, this method could help control inaccuracies even in environments with extreme temperature fluctuations. possible compensations for the thermal variance were proposed. An initial suggested approach consisted of incorporating thermistors into the silicon components of the accelerometer. However, size constrains in the accelerometer packaging and additional data processing time render the solution difficult to implement. CONCLUSION Environmental temperature changes can adversely affect the behavior of accelerometers. Thermal expansion of the fundamental components of this device leads to inaccurate acceleration readings. In particular, robotic applications requiring position sensing are severely limited in their operating temperature. As a result, compensation for thermal effects is necessary to reduce accelerometer sensing error in extreme temperature conditions. An analysis of the Aldrige and Juang [1] research reveals the potential deficiency with respect to temperature fluctuation in accelerometer operation. After scrutinizing the experimental methodology, A second correction method utilizes the microfabrication of an additional cantilever beam, instrumented with a piezoresistor, to nullify the temperature effect on the accelerometer output. The inclusion of this component does not drastically affect the fabrication process, packaging, or data processing time. Thus, this solution is considered optimal for the studied application. 8

10 REFERENCES [1] Aldridge, H.A. and Juang, J-N. Experimental Robot Postion Sensor Fault Tolerance Using Accelerometers and Joint Torque Sensors. NASA Technical Memerandum pp, March [2] Dalley, James W. Instrumentation for Engineering Measurements, 2 nd edition. John Wiley & Sons, Inc., USA, pp , [3] Evans, John R. The Design and Performance of a Low-cost Strong-motion Sensor Using the ICS-3028 Micromachined Accelerometer. U. S. Geological Survey. Open File Report, , 19 pp., [4] ICSensors website. Model [5] Kovacs, Gregory T.A. Micromachined Transducers Sourcebook. WCB/McGraw-Hill, USA, pp , [6] Wilcoxon sensors (Frequently Asked Questions site). 9

11 APPENDIX A: Illustrations Figure A.1 ICSensor, model 3140 Figure A.2 ICSensor, model 3140, circuit diagram 10

12 APPENDIX B: Specifications Specification B.1 RRC model 807i Specification B.2 ICSensors, model

13 AUTHOR BIOGRAPHIES Eggett, Cassandra Q. Cassandra Eggett is a mechanical engineering major with a focus in automotive design and will be graduating June of Cassandra has experience in the field of robotics and sensors from working in the Laboratory of Intelligent Mechanical Systems at Northwestern University for over a year. In the future Cassandra would like to be involved with the systems design aspect of the stock cars on the NASCAR circuit. Lutsey, Jeffrey K. Presently a fourth-year mechanical engineering student at Northwestern University, Mr. Lutsey plans to graduate June In academia as well as his cooperative education experience with Honeywell International, his focus has lied in thermodynamics, control systems, and propulsion. Potential career vocations include alternative energy research and development and astronautics. Pakula, David A. Mr. Pakula is an undergraduate at Northwestern University, majoring in mechanical engineering with a focus on design. He is expecting to graduate in June, His past experience with robotics include the Northwestern University Design Competition and sensor testing in ME 224 Experimental Engineering. He aspires to work for a small engineering design firm. Zumbado, Fernando A. A third year mechanical engineering student at Northwestern University, Mr. Zumbado plans to graduate in June He is enrolled in the cooperative education program with Johnson Space Center (Houston) where he worked in a haptic feedback response project for the Robotics Department (ER4). His focus on robotics has lead him to participate in Design Competition Potential career vocations include robotics/automation research and development. 12