MEMS Multi-sensor System for Flight Testing

Transcription

1 Invited Paper MEMS Multi-sensor System for Flight Testing Minas H. Tanielian The Boeing Company, Phantom Works P.O. BOX 3999, MS 3W-80 Seattle, WA (253) , ABSTRACT We have developed a MEMS-based system, which provides a multitude of "smart" pressure sensors at the point-of-use, through integration of the sensors and their corresponding electronics. This multi-sensor system is easy to install and operate and can provide an order of magnitude more accurate pressure measurements than the current technology. To accomplish this goal we developed a standard network interface protocol capable of handling large numbers of sensors, a low-profile, high accuracy MEMS pressure sensor, and innovative packaging techniques, including packaging of the MEMS sensor and its associated electronics in ASIC format on a high density multichip module substrate. Two different prototypes were built to validate the pressure belt concept. Prototype versions were flown on Boeing , 737-BBJ, , and F-18E aircraft. Measurements obtained during flight were compared results obtained by a reference MEMS sensor in a conventional package and connected to a pressure tube. The data taken by our modules were as good or better than the reference values. Keywords: MEMS, MCM, COB, smart sensors, flight test, multisensor integration, pressure belt, electronic packaging 1. INTRODUCTION One of the methods used to assess the efficiency of a new airplane wing is the measurement of the dynamic pressure distribution over the whole wing area. This measurement provides the aerodynamicists the load distribution and thus a measure of the design goals. The current technology used in flight loads testing utilizes an extensive network of thin plastic tubes, which are routed to remotely located pressure detectors using a pneumatic control system. This system is labor-intensive to install, expensive to operate, and does not have sufficient accuracy for future designs. Our goal was to reduce the installation costs by more than a factor of five, to replace the pneumatic control system by a corresponding electronic version, and improve the system accuracy by more than a factor of ten. These objectives were to be achieved using a minimum of wiring and interconnections and in a very thin packaging format, so as not to disturb the boundary layer flow characteristics over the areas of interest on the wing. This latter requirement translated to a total thickness of 0.1" for the measurement system. The thickness requirement forced us to use micro-electro-mechanical system (MEMS) technology for the sensor elements and bare silicon devices for the electronics, all integrated on a thin, silicon multi-chip module (MCM). In the past two years we built prototypes of this system and have tested various versions both in the laboratory and in the field, using aircraft manufactured by The Boeing Company. Our work has demonstrated, for the first time, a bus-based multi-sensor network, which employs at each measurement node MEMS sensors and highly integrated electronics in an aerospace application. In the following sections we discuss in detail the overall system architecture, the electronics used, the packaging technology, and flight testing of the prototype pressure belt hardware. 2. SYSTEM ARCHITECTURE Early on in this program, it was decided that the architecture of the system had to be modular. This would allow the pressure belt to be made of smaller units, which are easier to build, test, and handle. This modularity is clearly evident in Figure 1. The pressure belt is made of a number of segments, each segment having as many as six "smart" transducer modules on it. A "smart" transducer module includes the sensing capability (MEMS or otherwise), the corresponding electronics for signal conditioning and processing, a correction engine, and a digital bus interface. The "smart" transducer modules are referred to as Transducer to Bus Interface Modules (TBIM), a term derived by the IEEE P terminology. The IEEE P is a proposed standard for a "smart" transducer bus and currently it is in its final 120 MEMS Components and Applications for Industry, Automobiles, Aerospace, and Communication, Henry Helvajian, Siegfried W. Janson, Franz Lärmer, Editors, Proceedings of SPIE Vol (2001) 2001 SPIE X/01/$15.00

2 Field bus 2 in Connection Segment Inter-segment connection Figure 1 : The architecture of the MEMS pressure belt used for flight-testing. definition stage. Our architecture is based on an early version of this standard. An advantage of a modular system is that one can fabricate pressure belts in varying lengths, without having to make specialized segments for different wing sizes or for different locations on a particular wing. Furthermore, the use of the digital bus allows one to take readings in closely spaced intervals when needed Le., at the leading edge of the wing or at larger spacings, as for instance at the trailing end of the wing where the pressure gradients become less pronounced. In our case this can be easily accomplished by leaving certain module locations on a specific segment or segments unpopulated. Each TBIM has a MEMS sensor which is used for measuring pressure and temperature, an analog ASIC, a digital ASIC, an EEPROM, a SRAM and 37 passive components (resistors and capacitors). The TBIM provides the following functions: Signal conditioning for a number of different types of transducers such as piezoelectric, piezoresistive, variable capacitance, and strain gauge with 4 automatically selectable anti-aliasing filters Programmable sampling rate Analog and digital (12 bit) output Correction of digital output Sensitivity, linearization, zero compensation Transducer electronic data sheet (TEDS) Comprehensive self test Digital Bus Interface Our baseline pressure belt design accommodates up to 127 TBIMs on a single Network Capable Application Processor (NCAP). The number of TBIMs used in any particular implementation is a function of the bus clock, sampling frequency, the number of measurements taken per unit time, and the number of data channels used. These parameters are variable and can be adjusted depending on need. The NCAP is not part of the pressure belt that resides on the wing surface, although it may be in close proximity. All in all we would typically use anywhere from 10 to 20 NCAP for a typical loads survey, based on the size of the wing and the type of data sought. The NCAPs are connected to each other via a Fieldbus, like Ethernet, which is connected to a flight test computer inside the cabin of the airplane. 3. HARDWARE DEVELOPMENT APPROACH The overall approach was to design and build prototype versions of all the hardware prior to designing the integrated versions. This allowed for the use of more conventional hardware, such as circuit boards, which easily accommodated Proc. SPIE Vol

3 cuts and jumpers. It also allowed for changing of various components (for instance, capacitors or resistors) in an easy fashion. Having this capability is very important, especially for the analog design. These brass-board versions of the hardware provided a platform for the development and debugging of the system software. This was important because we would have at our disposal "known good" software when it came time to debug the highly integrated version of the hardware. However, this meant that the brass-board hardware was subject to the same limitations as the miniaturized version, such as ASIC library availability. The result was the development of two types of prototype hardware, prior to the final version, one that had the form and fit of the final version and was primarily used to evaluate the packaging technology, the materials, and fabrication processes in a realistic environment (flight) and one that had the full functionality, without the right form factor, to be used in the laboratory. This approach allowed separate testing of each configuration, thus reducing risk in the program. The laboratory version was designed in two passes. In the first one, the functions of the analog and digital ASIC were done on separate boards and the function of the TBIM in yet another board. In the second pass, the analog and digital boards were turned into ASICs and the packaged dice were used to build the TBIM. In the final version, this second pass laboratory version was reduced to a multichip module (MCM) using bare chips interconnected on a silicon substrate. Figure 2: Prototype PWB version of the analog ASIC and its die form " The two versions of the analog ASIC are shown in Figure 2. Some of the functions done in the analog ASIC are: (1) programmable gain, (2) instrumentation amplifier, (3) binary programmable gain non-inverting amplifier, (4) setting gain, (5) programmable dc offset adjustment, (6) programmable current excitation, (7) self-test, and (8) analog antialiasing filtering. Figure 3: Photographs of the digital board as well as the digital ASIC " " The digital ASIC shown in Figure 3 consists of an microcontroller, a programmable digital filter engine, a multinomial correction engine and a high-speed transmitterheceiver interface. The two engines and the high speed 122 Proc. SPIE Vol. 4559

4 interface are realized in three different FPGAs. The high speed bus interface FPGA allows the TBIM to communicate with the TBC (Transducer to Bus Controller) over a digital transducer bus and allows synchronization and control over the local ADC sampling process. The receiver circuit performs hardware address filtering for all packets and hardware command decoding for specific commands. This allows deterministic time response to specific commands such as the Trigger command. The receiver also allows messages which are not decoded by the hardware to be queued up for interpretation by the associated microcontroller. A calibration using prototype versions of the TBIM and unpackaged MEMS devices mounted on a Si substrate, similar to the final version of the pressure belt was done in the laboratory. The calibration incorporated all the functional elements of the ASIC designs, such as the correction engine, along with the design intended for the pressure belt segment. We used automatic test equipment and a Labview application to control the test program. The test performed 21-point calibrations over 10 temperature settings from 0.5 psia to 15 psia and temperatures from -50 C to +50"C. A surface fit was calculated using the 190 data points collected from the test. This fit represented the type of polynomial, which is used in the correction engine of the digital ASIC, having the capability to provide output in engineering units. What we found was that residuals from this surface fit do not exceed 0.04% in the 15 psia range, well within the margin needed by our application. 4. PACKAGING TECHNOLOGY 4.1 MEMS device and packaging The MEMS sensor used for this program was a redesigned 8515 Endevco pressure transducer made in two segments. One segment has an etched Si membrane (with a thickness of - 1 pm) that senses the pressure. This is achieved by forming on the backside of the diaphragm, a fully active Wheatstone bridge of doped Si resistors by ion implantation. Since Si is piezoresistive these resistors form the strain sensing element of the transducer. The diaphragm bearing segment is attached to a second piece of Si, which has a blind cavity etched in it and four feedthroughs that go from one side to the other. The feedthroughs are used to reach the backside of the Si membrane, where the piezoresistive resistors are routed to. These feedthroughs are then metallized to form bonding pads on the transducer that can be attached to the MCM substrate using flip chip. The two Si pieces are joined together in a hermetic fashion using glass sealing. The sealing is done in high vacuum, so that the sealed vacuum cavity is used as an absolute pressure reference. A schematic drawing of the sensor is shown in Figure 4. Top view f sensing membrane section) Cross-sectional view (along the AA line) Vacuum cavity Figui *e 4: Schematic representation of the redesigned MEMS pressure sensor. 4.2 Multichip module packaging An MCM is used to integrate all the functions of the TBIM on a Si substrate. The choice of the Si MCM was motivated by two important requirements: (a) the total thickness of the pressure belt had to be less than 0.1 inches in height from the skin of the airplane to the top of the belt and (b) from the requirement to be able to maintain a measurement accuracy Proc. SPIE Vol

5 of 0.1 % over the whole temperature range of application. The Si MCM provided a thin, stiff substrate that matched the thermal expansion coefficient of the MEMS device. It also allowed for high density bare chip packaging and the formation of embedded passive components such as resistors, capacitors, and inductors. Two MCM versions were built, a prototype analog version and the final digital version. Both the prototype version of the MCM and the final version are shown in Figure 5. The prototype version was used to debug the packaging issues related to the pressure belt in flight applications. It did not have the full functionality of the TBIM, however, it had the form and fit of the final digital module. It also allowed us to test the packaging technology for attaching the MCM to the pressure belt segment and the segment to the airplane wing. This was important because our whole packaging strategy was to make sure the MEMS device was not subjected to any spurious strains due to the flexing of the wing, since such strains could introduce errors in the pressure measurement. The MEMS device was isolated from the rest of the module using a thin glass ring and the flip chip connections at its base were covered by a low stress silicone gel. The top of this glass ring was capped by a perforated protective cover to protect the MEMS device from direct impact damage. Finally, in order to protect the devices from the environment, protective coatings were employed. The ICs were protected using a glob-top epoxy followed by a silicone coating that covered the whole module. Figure 5: The two multicliip iiiuuules developed in the project. (a) prototype version, (b) final version. On the left is the prototype version used for flight testing. All the materials and processes used in its fabrication and assembly are identical to the final version, shown on the right. The protective cap on the final version has not been installed in the picture so that one can see the MEMS device resident inside the glass ring. Also, the final version does not have the protective coatings so that all the devices can be clearly seen. 4.3 Segment packaging There are up to six MCMs attached to a polymeric tape to form the pressure belt segment. A number of segments are then connected to each other to form a pressure belt. Early in the program, the tape material for the pressure belt segments was based on TAB (Tape Automated Bonding) technology. However, as the complexity of the circuitry on the segment grew, it became apparent that a TAB-based tape would be extremely expensive and would only be available from a limited number of suppliers. For this reason we decided to migrate our design to a flex circuit tape. Flex circuits with two layers are widely available at reasonable cost from many suppliers. The down side of this choice was a looser manufacturing tolerance. This was deemed acceptable, however, even though it made segment assembly a bit more 124 Proc. SPIE Vol. 4559

6 complicated. The flex circuit-based segment has a length of inches and contains 6 MCM locations. As explained earlier, all locations do not have to be populated at all times. This fact also provides built-in fault-tolerance in the pressure belt design. The flex circuit tape consists of two in thick (1 02) copper layers on both sides of a in thick polyimide support with dielectric overlays (0.002 in thickness including adhesive, if any) on both sides. The top layer contains landing sites for all the components, including the power regulators, bus circuit elements, and discretes. It also contains interconnection tab features at both ends of the segment unit. The bottom conductor is used for routing. Vias are formed that connect from the top to the bottom of the segment using conventional flex circuit manufacturing processes, such as drilling and plating. Additional redundant vias are incorporated for power and ground. The overall stack is laser cut to create openings in the flex circuit with a tolerance of less than 2 mil in reference to its corresponding MCM unit center. The sprocket holes located at the edges are used for mechanical alignment during the assembly process. They are also used as a reference mark for locating the MEMS sensors in regards to the airplane wing. All copper surfaces are protected with either Sn or a NiJAu layer, which also makes the surface solderable. A photograph of an assembled segment is shown in Figure 6. Figure 6: A photograph of an assembled pressure belt segment. After the pressure belt is attached to the airplane, its surface is covered by an aerodynamically shaped fairing. The fairing is used so that the presence of the pressure belt does not create any turbulence, which could disturb the measurement we are trying to make. The use of the fairing was motivated by work done at NASA by Richard Hanley on the accuracy of a pressure measurement when the transducer is somewhat below the measurement boundary.") In addition, the fairing provides a means for handling the belt and provides extra protection for the various components. More detailed descriptions of the various aspects of the packaging technology used have been reported el~ewhere.'~' 5. FLIGHT TESTS The validation of the packaging approach used, and the whole approach in general, was carried out through a series of flight tests. This allowed for testing the various prototypes in a realistic environment and for identification of failure mechanisms. The first flight test of a prototype belt was done on a airplane, which was being tested for its landing gear performance. The pressure belt segment was attached close to landing gear, since this allowed for easy access to electrical and pneumatic interconnections. The pressure belt was installed side-by-side with a pressure tube, which mimicked the conventional systems used for flight loads surveys. The reference sensor used for this test used a Honeywell PPT system, which has been rated to have an accuracy of better than 0.1%. However, the difference between the conventional system and the system under development is that the conventional system requires routing of the pressure signal over a long skinny tube (20-30 feet long) to the measurement location, whereas our system routes an electrical signal instead. Thus, to compare the two systems we had to choose a steady state flight condition (cruise), where very little dynamical change occurred. In Figure 7 we plot the percentage difference between the measurement of the reference sensor and the MEMS sensors resident on the various MCMs on a segment. As one can see, the difference in all the measurements between the prototype modules and the reference are roughly within If: 0.1% of each other. Proc. SPIE Vol

7 Figure 7: The pressure reading difference between the MEMS sensors on the prototype belt and the reference. Each point is an average of 30 samples. The second opportunity to conduct a flight test was on a hybrid 737 aircraft called the Boeing Business Jet (BBJ). The BBJ was undergoing flight tests at the time to evaluate the performance of the winglets at the end of the conventional wings. In this instance, the prototype MCMs were installed on the front slats of the airplane. A special fairing had to be designed for this situation. The installation is shown in Figure 8. Figure 8: The sensor modules installed on the forward slat. The light colored tape protects the reference ports and the MEMS access holes on the fairings. The conventional tubing belts can be seen at midspan and on the winglets. 126 Proc. SPIE Vol. 4559

8 The flight loads survey was conducted in Mesa, Arizona. Just prior to the flight to Mesa, the MEMS sensor located nearest to the flush pressure ports (reference) failed during a pre-flight test. The cause of the failure was human error. The remaining MEMS modules operated perfectly through all the flight conditions conducted during the loads survey. However, this unfortunate accident prevented us from directly comparing measurements taken from two adjacent devices, as was done in the 757 flight test. In Figure 9 we plot the local speed along the slat cord of the BBJ wing during cruise conditions calculated using Computational Fluid Dynamics. As one can clearly see the location where our sensors were placed at x/c = 0.35, the local speed is supersonic. This information was important because we wanted to evaluate the pressure belt design on supersonic military aircraft, too. The calibration of the sensor modules was checked prior to the flight tests and after they had concluded. The difference in the readings when compared to a reference was 5 0.1%. L Q) E S z c 0 Q I Slat X/C Figure 9: Calculated air speed as a function of location (cord fraction) on the BBJ wing. The location of the MEMS sensors on the wing is marked by a vertical line, which corresponds to a local speed of 1.3 Mach. At the end of this flight testing program, when the airplane was landing at Boeing Field in Seattle, both functional MEMS devices failed. An investigation of this incident revealed that the moment of failure corresponded to a severe rainstorm the airplane encountered. A microscopic study showed that the sensor diaphragm had cracked. This led to the development of the protective cap that protects the sensor from this cause of failure. The fact that this was the only failure that took place was verified by replacing the MEMS sensor. The reworked module functioned flawlessly, More tests were conducted on a aircraft. Here the main goal was to ascertain the long term reliability of the belt. The purpose of this test was to verify that a module could survive under rather harsh conditions in the field for extended periods of time. The total test duration was over 3 months. This period of time is significantly longer that the expected flight loads testing, which typically is concluded in about two weeks. The total number of flights this module experienced was 59, with multiple takeoffs and landings during each flight. The module was airborne for a total of 11 1 hours and 46 minutes. Some of the aircraft test conditions included brake testing, crosswind landings, high altitude stalls, wind up turns, Dutch rolls, and multiple aggressive hardover airplane maneuvers, common in basic certification test programs. All in all, although there were some shifts in the calibration curves at low temperatures, the module was functional and the difference at 10 C or higher was within 0.5% of the original calibration curve, obtained 8 months earlier. The design was also checked against vibration. Random vibration in 3-axes, with 41g peak values and an rms value of 14g was conducted. The calibration was checked both before and after the testing period to verify that there was no damage to the module. These data are shown in Figure 10. Proc. SPIE Vol

9 C 0 z -0.1 '5 al U " " g g " g g " g " " g g g g g g g g g g g g g g g awlied Dsia Figure 10: Sensor calibration curve before and after 3-axes vibration testing (15 midaxis). The first flight demonstration on a military airplane took place at the Patuxent River Naval Air Warfare Center on an F-18E3 aircraft undergoing flight tests. The airplane on which this test was done is shown in Figure 11. The pressure belt was attached to one of the airplane pylons. This location provided easy access to the instrumentation bay. The airplane underwent a variety of flight tests in the course of over two months. In all cases we found very close agreement between the our pressure belt and the reference sensor, which was a conventional version of the sensor used on our pressure belt and known to have an accuracy of 0.05% of full scale or better in the 0-15 psia range. Figure 11: Installation of the pressure belt on a F-18E3 at Patuxent River Naval Station. Right, the installation location, is shown by the arrow. A close comparison of the reference sensor (Endevco 8515C-15) and one of our MEMS Belt sensor readings in Figure 12 reveals that even though we were using essentially the same sensor, our packaging had much better dynamic response than the conventionally packaged part, which was connected at the end of a tube. This is evident by the "peaks" and "valleys" in Figure 12. The reference pressure readings are more filtered than the directly exposed sensors on the pressure belt. This verified the soundness of our packaging approach. 128 Proc. SPIE Vol. 4559

10 Time (seconds) Figure 12: A comparison between our MEMS Belt design and the reference sensor on the F-18. The offset between the two sensors is a calibration artifact. 6. CONCLUSIONS We have demonstrated the concept of a MEMS multisensor belt for flight test applications, both in the laboratory and in the field. This system is versatile in that it can use a variety of sensor types, such as pressure sensors, accelerometers, strain gauges, and variable capacitance sensors. The system can be customized for a number of applications where a large number of sensors are employed, while keeping the number of interconnections to an absolute minimum. The system developed uses an early version of the IEEE P standard but it is forward compatible with the final version, by adding an interface chip. Finally, we have shown that this new system has better dynamic performance on the system level than conventional sensor installations. ACKNOWLEDGEMENTS The author would like to acknowledge the efforts of a large team of people who contributed to this project at The Boeing Co, Endevco Corp., and at Georgia Tech. Special thanks to Namsoo Kim, Mark Holland, Hung Mach, Lee Eccles, Larry Malchodi, John Stice, Dave Smith, Jean Nielsen, and Mark Chisa of Boeing, Alex Karolys, Fernando Gen-Kuong, Ron Poff, and Craig Evensen of Endevco, and Prof. C.P. Wong and Jiali Wu of Georgia Tech. This project was funded in part by DARPA/AFRL Agreement F REFERENCES 1. For more information on the IEEE P standard see the web site 2. R. Hanley, "The Effects Of Transducer Flushness On Fluctuating Surface Pressure Measurements" in AIM 2"* Aero- Acoustics Conference, Hampton, VA, March 24-26, I N.P. Kim, M.J. Holland, M.H. Tanielian, and R. Poff, "MEMS Sensor MCM Assembly with TAB Carrier", Proceedings of 50* IEEE Electronic Components and Technology Conference, May N.P. Kim, K. Li, D.J. Kovach, C-P. Chien, and M.H. Tanielian, in IMAPS 99, Oct. 1999, p N.P. Kim, N. Amirgulyan, C-P. Chien, and M.H. Tanielian, IAMPS Oct Proc. SPIE Vol