U.S. Air Force T&E Days 2009 10-12 February 2009, Albuquerque, New Mexico AIAA 2009-1727 Advanced Inertial Test Laboratory: Improving Low-Noise Testing of High-Accuracy Instruments Cleon H. Barker 1 and William G. Rock 2 746th Test Squadron, Holloman Air Force Base, New Mexico 88330 The 746th Test Squadron, Central Inertial GPS Test Facility, has a long legacy of testing high-accuracy inertial guidance and navigation components and systems. Located in the Tularosa Basin of New Mexico, away from urban and marine-generated seismic contamination, makes it an ideal location for performing low-noise characterization testing of high-accuracy instruments. The Advanced Inertial Test Laboratory (AITL) was designed and built in the late 1960s for testing high accuracy inertial components and systems for the Department of Defense Strategic Missile Development. In recent years, there has been an increased interest within the Department of Defense for development of high-accuracy inertial and pointing systems. In anticipation of meeting the demanding requirements for low-noise characterization of these systems, the 746th Test Squadron began an initiative to characterize the background seismic environment on the test pads in the AITL facility and to make improvements in the seismic environment by reducing the cultural contamination generated by the environmental support equipment necessary for creating a laboratory environment. To date, the seismic environment on the test pads has been characterized, and improvements have begun to reduce the seismic contamination. This paper will discuss the characterization that has been completed, the improvements that have been completed and the future efforts that are anticipated for making further improvements. Nomenclature g = normative acceleration of gravity on Earth (9.80665 m/s 2 ) I. Introduction he 746th Test Squadron, also known as the Central Inertial and GPS Test Facility (CIGTF), at Holloman Air T Force Base, New Mexico is home to the Advanced Inertial Test Laboratory (AITL), figure 1. AITL is the most seismically-quiet test facility in the United States. The seismic characteristics of AITL are due, in part, to its location in the Tularosa Basin, a naturally seismically-quiet part of the nation away from urban and oceanic-generated disturbances. AITL was designed and built in the late 1960s for testing high-accuracy inertial components and systems for the Department of Defense s intercontinental ballistic missile programs. This testing was conducted throughout the 1970s and into the mid-1980s. Since the mid-80s, AITL was utilized to test inertial systems which were placed on the Hubble Space Telescope. When AITL was constructed, the facility s environmental support equipment (ESE) were mounted on vibration isolators and housed in a separate building, the Mechanical Building (see figure 1), in order to minimize the effects its operation had on the test environment. A seismic study conducted in Dec 05 revealed considerable contamination of the seismic environment on the test pads inside the AITL facility. Further investigation revealed much of the contamination was caused by degradation of some of the isolators, the lack of isolators under new replacement ESE and modifications to hardware mounting structures resulting in parallel paths for transmissibility of vibrations around the isolators. Since the Dec 05 study, a new refrigerated air conditioning water chiller and a new, larger air compressor for the test cells thermostats and pneumatic actuators was installed. Because of the results of the Dec 05 study, the installation of the new ESE and the expectation of future testing of high accuracy inertial and pointing systems for the Department of Defense, the decision was made in the spring of 2007 to conduct a new study to 1 Consultant, Strategic Development, 1644 Vandergrift Road Holloman Air Force Base NM 883300. 2 Program Manager, Avionics Element, 1644 Vandergrift Road Holloman Air Force Base NM 883300. Copyright 2009 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
determine the most offensive ESE and the effectiveness of the passive isolation employed on each of the ESE. This study was conducted from Aug 07 to Mar 08. Figure 1. The Advanced Inertial Test Laboratory (AITL). During the course of the study, it became evident that the sensors, which were being used to record the seismic environment on the quietest test pad in AITL, were problematic. It was determined that the bandwidths of the same model sensor were not identical. Consequently, the outputs of the seismic sensors did not agree across the frequency band. This issue was somewhat mitigated by pairing up two sensors with similar bandwidths and disregarding the data at the higher frequencies where the bandwidths were different. While this provided more usable data in the context of the study, it was decided that a more robust solution was desirable. After the conclusion of the study, an effort was begun to determine the actual bandwidth for each seismic sensor and develop a mathematical model to compensate the scale factor for each of the sensors. II. Study Methods and Results The study was carried out in three phases. The first phase focused on determining the vibration generated by the newly installed air conditioning chiller and the air compressor, the second on the attenuation being provided by the passive isolators and, finally, measuring the seismic disturbance on the isolation pad in AITL from the transmitted vibrations through the ground. Two clusters of single-axis PCB Piezotronics 393B05 seismic accelerometers were used to record the vibrations generated by the chiller and air compressor and the vibration on the floor underneath the ESE. Our most sensitive seismometers could not be used for this recording, because they were physically saturated by the magnitude of the vibrations. Each cluster was comprised of three seismic accelerometers mounted orthogonally on an aluminum cube (figure 2). With only the chiller or air compressor operating and all other ESE turned off, data were recorded with one cluster placed on the ESE and the other on the concrete slab underneath. The recorded data from each sensor were converted into the frequency domain through a fast-fourier transform to produce a power spectral density (PSD) plot. A coherence plot was generated to determine the frequency content of the signals that were common to both sensor clusters. Next, a transfer function was plotted by dividing the data from the sensor under the unit by its parallel sensor mounted on the unit in the frequency domain. This then provided, in the frequency domain, a general assessment of how much attenuation was being provided by the passive isolators. Finally, the data were converted into units of acceleration (g) by taking the square root of the sum of the acceleration squared energy in the PSD. One plot displayed the cumulative acceleration recorded on the chiller or air compressor, and the other displayed the cumulative acceleration being felt on the floor underneath. The difference in the total cumulative accelerations was calculated in order to serve as a baseline for measuring the effectiveness of any future improvements that are made in isolation. This same procedure was repeated for each individual piece of ESE supporting the AITL facility.
Figure 2. PCB Piezotronics 393B05 Seismic Accelerometer Clusters. The second phase of the study focused on the effect that the ESE had on the seismic environment on the test pad in AITL s Room 13. The Room 13 test pad was chosen as the test area, because, of all the test pads in AITL, it provided the most isolation and was the location for the Hubble Space Telescope low-noise characterization gyro tests. For this part of the study, two different models of seismometers were used. The first model was a Geotech KS-2000 three axis seismometer. The KS-2000 is a single instrument with three single-axis seismometers mounted orthogonally inside its case. The second model was a Geotech S-510 single-axis seismometer. Using these instruments, two seismometer clusters were created. Each cluster consisted of one KS-2000 and three S-510s mounted to an aluminum plate. The S-510s were mounted orthogonally to each other and aligned with the seismometers inside the KS-2000. During testing, the two clusters were placed side by side on the test pad (figure 3). Figure 3. Two KS-2000 and S-510 Seismometer Clusters. Data from the Room 13 test pad were recorded under differing seismic conditions. First, data were collected with all of the ESE operating, followed by all of the ESE turned off. These measurements provided a baseline which indicated the best and worst possible seismic conditions on the test pad. The next series of measurements were conducted with a selected piece of ESE operating while all others were turned off. The individual pieces of ESE operated during this portion of the study were: an air compressor, a chilled water pump and several air handlers. Also, in addition to recording the disturbance generated by each individual air handler alone, all
air handlers were turned on simultaneously to record their combined contribution to the disturbance. The effects of the operation of air conditioning chiller on the isolation pad in Room 13 were also recorded. As before, the data collected were converted into the frequency domain through a fast-fourier transform and plotted as a PSD. Next, the coherence between two parallel, same model, seismometers was calculated in the frequency domain. A coherence of approximately one in a frequency range indicated the two sensors recorded true motion. A coherence of approximately zero indicates the data recorded in a particular frequency band was mostly random noise which is incoherent. Using the coherence function, the coherent power was calculated by taking the PSD from one seismometer and multiplying it by the coherence between the two parallel sensors. The resulting plot in the frequency domain then represented the true motion sensed by the seismometers. Because the units of PSD and Coherent Power (g 2 /Hz) are sometimes difficult to relate to, the power across the frequency band was summed and the square root taken of the sum yielding units in acceleration, g. The third phase, as mentioned previously, focused on measuring the vibration generated by the remaining ESE in the Mechanical Building and determining the attenuation provided by their respective passive isolators. The sensors, test methods, and analysis techniques were identical to those described in the first phase. The pieces of ESE evaluated in this manner were: each individual air handler, chilled water pump, ultra-clean air supply, and main commercial power transformer feeding the electrical requirements of the AITL complex. Once the study was completed, each ESE was ranked according to its contribution to the seismic disturbances recorded on the isolation pad in Room 13. Also from the study, it was determined that the disturbances generated by the ESE lie in the frequency band between 5 and 100 Hz. Additional information, concerning this study, can be found in the Seismic Characterization Study of the Advanced Inertial Test Laboratory (AITL) Report (CIGTF-SR- 08-01), dated 8 December 2008. III. Seismometer Calibration Method Vibrations important to inertial testing lie in the frequency band between 0.1 to 100 Hz. Motion below 0.1 Hz that concerns inertial testing is tilt or changes in the gravity vector. A different class of sensors is used to measure this motion and is readily available commercially. Seismometers with sensitivities down to 1 nano-g and an upper bandwidth of 100 Hz are not commercially available. Vibration sensors that operate between 1 and few thousand Hz have threshold sensitivity in the 1 micro-g range. Consequently, replacement with commercially-available seismometers was not an option. An attempt was made to have the seismometers in our inventory independently calibrated. The result of this effort was a certification that the manufacturers provided scale factors were accurate to within 10%. The manufacturer-supplied scale factor was a constant value for converting the signal voltage out of the sensor to values of acceleration, g, across its operating frequency band. What we required was a compensation algorithm to compensate for the high frequency roll off and, thereby, extend the upper frequency operating range. For the above reasons the 746th Test Squadron started an effort to develop a compensation model to extend the bandwidths of the KS-2000 and S-510 seismometers up to 100 Hz by performing our own calibration in house. The results from the study revealed the KS-2000s seismometers had good agreement below 40 Hz and the S-510s below 80 Hz. Therefore, our challenge was to generate a white random noise vibration between 20 and 100 Hz. To further complicate the effort, the vibration had to be less than 100 micro-g in magnitude to prevent the seismometers from physically saturating above 100 micro-g. The squadron has several shaker vibration systems used for environmental-type testing scenarios. These systems can generate precise controlled vibrations in the frequency range of 1 and 1 khz and accelerations from 0.1 to 85 g. One such shaker vibration system is the Ling B335 shaker (figure 4).
Figure 4. B335 Vibration System Mounted on Seismic Mass. The shaker was mounted on top of a concrete seismic mass which was 10 ft long by 5 ft wide and 3 ft thick. The concrete mass was supported on six isolating air bags for attenuating cultural disturbances that could potentially contaminate the test article preventing vibrations, generated by the shaker, from contaminating other test beds within the facility. The six air bags were connected in such a way that two air bags were connected to the same manifold that was pressurized to lift the seismic mass to a preset height. This configuration then created a three legged system (figure 5). In figure 5, the air bags are represented by the blue circles, the shaker by the orange and the mass by the gray. Figure 5. Top View Configuration of Shaker, Seismic Mass, & Air Bags. To create vertical vibrations of less than 100 micro-g, we hypothesized we would place the sensors on the concrete seismic mass supporting the vibration shaker and sense the vibrations that would be coupled into the concrete mass. To create horizontal vibrations, the shaker would be rotated about its trunion axis such that the shaker movement would be along a horizontal axis, thus generating horizontal vibrations in the seismic mass. For a reference standard to quantify the vibration profile of the concrete mass, we decided to use the PCB 393B05 seismic accelerometers mentioned above and shown in figure 2. The bandwidths of the seismic accelerometers were from 2 to several khz. The threshold sensitivities of a few micro g required the vibrations coupled into the seismic mass to be greater than 10 micro-g. The squadron has significant experience using the PCB seismic accelerometers and has built up a great deal of confidence in their ability to accurately record low-level seismic vibrations. The PCB used for this test effort was first calibrated against a calibrated accelerometer that was traceable to the National
10 0 PSD 10-2 S510 (V 2 /Hz) V 2 /Hz 10-4 10-6 10-8 g 2 /Hz 10-10 10-4 PSD 10-6 10-8 10-10 10-12 PCB (g 2 /Hz) 10-14 10-16 1.2 Coherence 1.0 Coherence 0.8 0.6 0.4 0.2 0.0 Frequency (hz) Figure 6. PSD and Coherence of Vibrations on Seismic Mass. Institute of Standards and Technology. The scale factor was checked and the linearity across the frequency band of 1 to 100 Hz was confirmed. Through experimentation, it was determined a 0.5 g, 5 and 100 Hz, white random noise excitation of the shaker, generated vibrations of 10 and 100 micro-g on the seismic mass. However, the motion on the seismic mass was no longer white, because of resonances and anti-resonances in the shaker mount, concrete seismic mass and air bag suspension (figure 6). This was not expected to be a problem, as there was excellent coherence between the PCB seismic accelerometer and the seismometer under test. We then generated transfer functions between the PCB seismic accelerometer and the KS-2000 or S-510 seismometers. Sample transfer functions are represented in both figures 7 and 8. Figure 7 is a typical transfer function with the gain displayed logarithmically. The output was seismometer data in volts, and the input was PCB data in g. Because the logarithmic gain plot didn t provide adequate resolution for our purpose, we re-plotted the transfer function with linear gain (figure 8). When the gain was viewed in linear units of volts/g, it was easier to observe the differences in the two sensor outputs.
10 4 Gain log gain 10 2 10 0 200 Phase 100 Degrees 0-100 -200 1.2 Coherence 1.0 Coherence 0.8 0.6 0.4 0.2 0.0 Frequency (hz) Figure 7. Transfer Function (Gain), Phase and Coherence.
PSD 10 0 PCB (g 2 /Hz) S510 (V 2 /Hz) 10-5 Power/Hz 10-10 10-15 600 Transfer Function (Linear Gain) 500 Scale Factor (v/g) 400 300 200 100 Frequency (hz) Figure 8. PSDs of S510, PCB and Linear Gain. Next a model was generated from each of the linear transfer functions between each seismometer and the seismic accelerometer using the method of least squares. Figure 9 presents the models for both SN1402 and SN1405 seismometers. The red trace is the recorded data used to calculate the model and the black trace is the best fit quartic model plotted on a logarithmic frequency scale. Both generated models appeared to fit the recorded data satisfactorily.
500 400 Model Data Sensor 1402 gain 300 200 100 600 500 Model Data Sensor 1405 gain 400 300 200 Frequency (Hz) Figure 9. Linear Gain and Model. Seismic data were collected on the seismic test pier in room 13 of AITL. The uncorrected seismic data using the manufacturers provided scale factors are presented in figure 10. The top traces are the coherent power from each seismometer, the center plot is the coherence between the two seismometers, and the bottom traces are the coherent cumulative acceleration from each seismometer. The coherence between the two sensors became very poor above 75 Hz. The cause of the loose of coherence could indicate a resonance in the seismometer mounting fixture or loss of significant seismic motion above this frequency.
g 2 /Hz Coherent Power 10-8 Sensor 1402 10-10 10-12 10-14 Sensor 1405 10-16 10-18 1.0 Coherence 0.8 g 2 /Hz 0.6 0.4 0.2 g 0.0 (x10-6 ) 50 40 30 20 Sensor 1402 Sensor 1405 Root Sum Acceleration 10 0 Frequency (hz) Figure 10. Uncorrected Coherent Power, Coherence, and Cumulative Root Sum Acceleration. The generated model, unique to each seismometer, was then applied to the recorded seismic data, and a determination of the improvement in cumulative sum acceleration agreement between two parallel seismometers was made. An example of this investigation is shown in figures 11, and 12. Figure 11 shows the same data after applying the proper correction model as a scaling factor to the coherent power and cumulative sum acceleration of the two sensors. If the correction model achieved the desired level of improvement, the two cumulative sum accelerations would overlay. Observing the two cumulative sum acceleration traces in both figures 10 and 11 doesn t provide the fidelity necessary to make a clear determination. Therefore, the uncorrected cumulative sum accelerations were differenced as were the corrected cumulative sum accelerations.
Coherent Power 10-8 Sensor 1402 10-10 Sensor 1405 g 2 /Hz 10-12 10-14 10-16 10-18 (x10-6 ) Root Sum Acceleration 40 Sensor 1402 30 Sensor 1405 g 20 10 0 Frequency (Hz) Figure 11. Corrected Coherent Power and Cumulative Room Sum Acceleration. Figure 12 then displays the two differenced cumulative sum accelerations overlaid on the same chart. The black trace is the uncorrected difference and the red trace is the corrected difference. If the correction models were ideal, the corrected difference would have been zero across the entire frequency band. As the result in figure 12 shows, the correction model employed improved the data but did not remove all of the error present.
(x10-6 ) 2 Corrected Uncorrected Cumulative Sum Difference 0-2 g -4-6 -8 Frequency (Hz) Figure 12. Uncorrected and Corrected Differenced Cumulative Root Sum Accelerations. In conclusion our attempts in modeling the transfer functions to calculate a correction model for each of the seismometers have been a limited success. While the data were improved with the correction models, the level of improvement falls short of our requirements and expectations. With the poor coherence above 75 Hz in the seismic data, we could not expect close agreement between the two parallel sensors above 75 Hz. Future efforts will include testing the plate the seismometers are mounted on to determine if there are resonances above 75 Hz which are compromising the coherence. If there are resonances, we will redesign the seismometer plate to remove the resonances. Also, we anticipate retesting in the vibration laboratory with the calibrated PCB. These new tests will be conducted with 5 to 150 Hz white noise excitation to the shaker. By extending the frequency to 150 Hz we hope to better define the transfer function at the higher frequencies. We may also attempt to perform a calibration test with the shaker control accelerometer mounted on the seismic mass the shaker is anchored to. If the shaker can be controlled with its signal sensor mounted on the seismic mass, the disturbances felt on the seismic mass should be considerably whitened. Future efforts will undoubtedly produce unexpected results and present new challenges for us to overcome.