1 June 2006
Data Doesn't Wait
Air force engine test facility develops seamless search and retrieval process
By Edward W. Dorrell, Jr., Ronald L. Turner, and Dr. Donald J. Malloy
Traditional database capabilities used to exist only for steady-state data with no transient capabilities because of the immense amount acquired under transient conditions. But engineers at the Arnold Engineering Development Center (AEDC) engine test facility (ETF) developed a way to access database information from thousands of unique sensors located throughout the test facility and test article. The new method allows them to retrieve time-dependent transient test data in seconds as opposed to hours or days.
A graphical user interface seamlessly queries, retrieves, and analyzes more time-dependent transient test data than before. Engineers acquire data and record it in the data acquisition system, then transfer it to the data processing system. The processing system time-merges the data and calculates engine performance parameters. The processed data transmits to the local analysis system, remote customer systems, and the test data management system. The data simultaneously transfer to the test data management system for archiving. Steady-state data automatically store in a steady-state database during processing.
The ETF supports test and evaluation of aero-propulsion systems for advanced aircraft, missiles, satellites, and space vehicles using 13 altitude simulation turbine engine test cells, three sea-level turbine engine test cells, and four altitude simulation rocket motor test cells. Simulated flight tests conducted over a wide range of Mach numbers and altitudes provide the data at precisely controlled conditions required to determine operational characteristics of aero-propulsion systems.
Test information comes from instrumentation installed in and around the engine. The installed instruments measure forces, pressures, temperatures, vibrations, flows, and various test-dependent measurements. Engineers can acquire up to 1 million samples of data per second on the data systems in the AEDC propulsion test cells.
Turbine engine testing of propulsion systems, such as the F119 engine, which powers the F/A-22 aircraft, provides data to evaluate gas turbine engine operability, performance, aeromechanical structural behavior, mission simulations, engine-inlet compatibility, ice accretion, altitude performance, and fault detection/accommodation.
An F135 engine installed in Test Cell J-2 gets support from the white overhead thrust stand. Air at simulated altitude conditions flows from right to left, exiting the convergent-divergent nozzle into the exhaust diffuser. You could install screens in front of the engine to simulate the airflow characteristics of the aircraft inlet. The engine throttle operates as the air supply compressor; and exhauster systems regulate to simulate flight conditions with controlled airflow rate, temperature, and pressure as specified in the test plan. A snap occurs when the throttle position snaps from one position to another. This corresponds to what happens in flight when the pilot jerks the throttle from one position to another as he might do in a dogfight. An accel starts at a low throttle position and gradually increases to a higher position whereas a decel starts at a high throttle position and decreases to a lower position.
A bodie is a series of snaps (up or down) with predetermined dwell times between the snaps. For large engine test programs associated with future generation fighter aircraft, 3000+ engine operation hours should reach completion in the system design and development phase. This could result in as much as 40 terabytes of transient, time-dependent test data and thousands of test events to undergo database, query, and retrieval. This is an order of magnitude more data than the AEDC accumulated from all turbine engine tests in 2001.
An aero-propulsion test period usually lasts from 12 to 24 hours, during which the engine operates at various simulated test conditions and could acquire up to 100 steady-state and 100 transient data points. It will acquire steady-state data points when the test article and facility are at stable operating conditions and contain around 10,000 measured and calculated parameters. It will acquire transient data points during transient engine operation. These points may contain several thousand measured and calculated parameters processed at a rate of 100 samples per second (or more) for nearly two minutes.
In querying, searching, and retrieving events in the high-speed transient data, we met constraints in the design of the statistical database and associated process. Because we scheduled the aero-propulsion data analysis and processing systems to update within the next 36 months, we needed to demonstrate the process to search and retrieve data quickly, using the existing database software on the existing data analysis and processing systems with an online storage capability of 1TB.
We got the idea to create an additional statistical database to supplement the transient aero-propulsion test data by reviewing techniques from flight test centers to database and query time-dependent flight test data. Based on this review, we began calculating and storing the average, minimum, maximum, and standard deviation of each measured and calculated parameter for each one-second interval of data acquired. We also evaluated four more statistics to determine their utility relative to storing, querying, and retrieving events in high-speed transient test data.
The first of the four additional statistics we evaluated was F, used herein to determine whether selected parameters have reached a steady-state condition. F is the result of doing a statistical F-test on a discrete set of data. F determines the significance of the quadratic and linear terms of the least-squares quadratic fit of a set of data. If we determine these two terms are insignificant, we assume the more appropriate fit should be a horizontal straight line; hence, we assume the data is steady state. If Q0 and Q2 compute as the sums of the squares of the residuals for the horizontal straight-line fit and the quadratic fit, respectively, then F = (N - 2)(Q0 - Q2)/(2Q2), where N is the number of points in the set.
The second statistic was the slope of the first-order, least-squares fit of the data in the one-second interval. The slope calculates as a byproduct of the F test using orthogonal polynomials. The slope is useful in determining whether the data are truly flat or are part of an accel, decel, or snap.
The third statistic was a simple, yes or no to indicate whether the data were flat. We determined this by using the F, the slope, and the standard deviation. The fourth and final statistic was an indicator of whether the data for the parameter and time are valid. The data validation process is highly automated, using software applications that validate test data in real time or near real time.
We performed initial evaluations to determine the effectiveness of storing, querying, and retrieving using the archived high-speed data files and all eight statistics stored in the online database. For a 4000-parameter database with eight statistics, we can store 32,000 pieces of information for every second of data we acquire. For 20 sample-per-second data (80,000 samples of data per second), there is a compression factor of only 2.5:1.
Subsequent evaluations revealed one-second average data points provided enough information and accuracy to meet the search criteria while the other seven statistics added little. Eliminating statistics associated with minimum and maximum values does not preclude the capability to identify minimums and maximums over a one-second interval. We assumed we could address these limitations using secondary calculations in the data reduction program. The ability to quickly retrieve and analyze high-speed, time-dependent data also eliminated the need for some of the statistics.
The primary reason for using only the average was to minimize the online disk space required. The enormity of the data required a highly compressed summary database that still provided information to identify events in the high-speed, time-dependent data. Replacing this and similar data with the one-second average produces adequate data characterization and accuracy for the search technique and results in a compression ratio of 1000:1. However, while the average is adequate to characterize most parameters for search and retrieval purposes, the recorded data are often still necessary for in-depth analysis. See additional statistics we used to more fully characterize the test maneuver below.
One of the key requirements of the query capability is to be able to find the transient data points for any maneuver performed. A user should be able to identify all the transient data points when the throttle snaps from 15 deg (idle power) to 100 deg (military power). The search capability should be able to locate the transient time-dependent data, display a plot of the data, and allow the user to analyze any parameters associated with the maneuver. Unfortunately, you might not see full characterization of the throttle maneuver or have it readily available as part of the recorded test data. Therefore, you need to identify the maneuvers from the throttle position as a function of time. For this reason, you should compute additional statistics from engine throttle position to better characterize typical engine maneuvers.
Using a computer to determine whether or not a throttle maneuver has transpired is a difficult process. One way is to curve fit the data. If the data always conformed to some functional form, say three hinged straight lines, you could fit it to that functional form and easily solve the problem. However, bodies can consist of any number of lines. A single transient data point can be a series of maneuvers, and the user may want to search for data points that contain a combination of maneuvers. Also, the amount of data acquired can vary widely from one transient data point to the next. Because you can model snaps, accels, and decels with three-hinged straight lines, and bodies consist of a series of three-hinged straight lines, we decided to fit the data for each nominally one-second interval with three-hinged straight lines and use the results to identify maneuvers.
The product of the research provides an automated capability to query and retrieve information from terabytes of high-speed, time-dependent, transient test data. You can also easily retrieve test events such as snaps, accels, decels, and bodies. Indexes in a database, much like an index in a book, are pointers to the data they reference. The next generation system will include software and hardware enhancements to accommodate the increasing demands for computational resources, data throughput, data storage, and seamless data access by other systems and processes. We'll consider developing automated capabilities to store, query, search, and retrieve events in high-speed transient data.
Indexes make the difficult task of finding information almost trivial. Search time using the current generation database software (circa 1990) can be logarithmic for indexed parameters and linear for unindexed parameters. The typical database we used contains 4,447 measured and calculated transient parameters and 335 GB of data, and indexes test and data point numbers to rapidly retrieve and analyze selected data points. The effective compression ratio is 62:1. We've tailored the database software for use in aero-propulsion applications and provided the capability to search on any of the measured and calculated parameters.
About the Authors
Edward W. Dorrell, Jr., Ronald L. Turner, and Dr. Donald J. Malloy are with the Aerospace Testing Alliance at the Arnold Air Force Base in Tennessee.
The research reported herein was performed by the Arnold Engineering Development Center (AEDC), Air Force Material Command. Work and analysis for this research were performed by personnel of Aerospace Testing Alliance, the operations, maintenance, information management, and support contractor for AEDC. Further reproduction is authorized to satisfy needs of the U. S. Government.
Development of Inorganic Fluorescent Coatings for High Temperature Aerospace Applications - 49th IIS 2003
ISA Sensors Community
ISA Aerospace Industries Division
Network, Organize Thyself