63rd International Astronautical Congress, Naples, Italy. Copyright ©2012 by the International Astronautical Federation. All rights reserved.
IAC-12,E2,2,8,x16326 Page 1 of 9
SOLID-BORNE SOUND MEASUREMENT FOR THE INDEPENDENT EVENT DETECTION
Andreas Leonhard WinhardUniversity of the German Federal Armed Forces, Germany, email@example.com
Failure detection and analysis is a very important aspect of unmanned and therefore remotely operated spaceflight missions. A dedicated failure detection system for every mechanical part would significantly increase the weight and complexity of a spacecraft. The goal of the SOMID (Solid-borne Sound Measurement for the Independent Event Detection) experiment is to develop, build and flight test a failure detection system which monitors multiple components onboard a spacecraft regarding its correct function or causes of a possible malfunction. As every kind of mechanical event on a spacecraft induces micro-vibrations into its structure, solid-borne sound has been chosen as the source of information for SOMID. The experiment measures these micro-vibrations with piezo-based accelerometers and then uses the created data to detect and analyze various events. All data is stored and compared with laboratory data of nominally functioning mechanical components in post-flight data evaluation. Events created by malfunctioning components show significant changes in the characteristic frequency spectra when compared with those created by nominally functioning components. The experiment design includes three accelerometers on the module bulkhead and one accelerometer on the module wall for reference measurements. Two electro-magnetic valves and one servomechanism are used to create specific mechanical events during the flight. The created data is stored on a flash data storage device. The SOMID experiment has been flight tested on the REXUS-12 (Rocket Experiments for University Students) sounding rocket on the 19th of March 2012. The experiment was operational from nine minutes and 30 seconds before Lift-Off until payload impact. It did not only record the mechanical events created by its own actuators, but also solid-borne sound created by mechanical components of two other experiments onboard the rocket as well as the rocket’s service and recovery system. The primary goal of this flight test was the confirmation of the method of failure detection via solid-borne sound measurements. However in flight an anomaly in the recovery system occurred causing a crash of the payload section. This led to the addition of the secondary mission goal of identifying, isolating and analyzing the non-nominal event that caused the failure. Together with MoRaBa (Mobile Raketen Basis, Mobile Rocket Base) of the DLR (Deutsches Zentrum für Luft- und Raumfahrt, German Aerospace Center) the SOMID Team was able to locate a possible point in time for the non-nominally performed event in the acceleration data recorded by the experiment. SOMID is currently developed into a more compact and lightweight system for failure detection on launch vehicles and spacecraft.
I.I Scientific BackgroundEvery kind of mechanical event on a space vehicle induces micro vibrations into its structure. Instead of considering these micro vibrations as a disturbing factor, the experiment will use them as a source of information for detecting and analyzing various events on the space vehicle. These events include actuation of valves and servomechanism, unfolding of solar panels, activation of attitude control systems etc. A characteristic frequency spectrum of the vibrations can be identified for each of these events. Comparison of the created data with laboratory data of nominally functioning components then allows the identification of malfunctions.
SOMID is based on a laboratory experiment conducted by D. Wilde and D. Obst at the University of the German Federal Armed Forces in September 2010.
The MIRIAM-1 mission (Main Inflated Re-entry Into the Atmosphere Mission Test) failed partly due to the malfunction of an onboard valve responsible for the inflation of the re-entry balloon. The goal was to find an independent source of information for failure detection. This experiment confirmed that correctly functioning and malfunctioning mechanical components induce vibrations with different characteristic frequencies and amplitudes into the structure of spacecraft and that these changes can be measured and detected.
The next step was to evaluate the ability of the principle of solid-borne sound measurement to detect malfunctions in flight. Since all measurements up to this point were gathered in a low noise laboratory environment the flight test showed in which phases of the flight the system can be used. A sounding rocket was found to be the best platform for the flight test because it is able to simulate multiple possible applications for the system. During the ascent of the rocket the experiment is subjected to vibration spectra also present on any other carrier rocket. The following exospheric coast phase simulates the orbiting phase of a satellite. During re-entry the system is tested for its ability to work under atmospheric influences.
I.II Experimental ObjectivesThe primary objective of the experiment is to confirm the method of detecting, identifying and analyzing the mechanical events created by SOMID’s onboard actuators as well as by components of other experiments on the rocket by measuring the solid-borne sound in flight. Operation of the actuators and solid-borne sound recording during the entire mission allows the team to judge the system’s ability to detect malfunctions in the different phases of the flight and therefore find possible applications on launch vehicles and satellites.
The secondary objective after the failure of the rocket’s recovery system is the identification, isolation and analysis of the non-nominal event that caused the malfunction.
I.III Experiment OverviewThe experiment consists of 4 accelerometers which measure the vibrations induced into the structure of the experiment. These vibrations might be events during different phases of the flight (e.g. separation of the rocket motor), the aerodynamic loads of the rocket or vibrations that are induced by other experiments. In order to fulfill the scientific objectives additional sources of vibrations are mounted onto the module bulkhead which create body-sound at specified times during flight. These actuators include two electro-magnetic valves and one servomechanism. These devices are actuated beginning with Lift-Off and the actuations will follow a predefined pattern to create events in the different phases of the flight (rocket motor burn, atmospheric ascent, sub-orbital coast phase, re-entry). The data is saved on a flash data storage. A microcontroller controls the actuators and the data handling as well as the connection to the rocket’s service system.
After the flight the signal is converted using Fast Fourier transform routines to visualize the frequency domain. This allows a precise analysis of the changes in the frequency spectra, when compared to a reference measurement.
II. EXPERIMENT DESCRIPTION
II.I Experiment SetupThe experiment uses a module which is 355 mm in diameter and 120 mm in height. All components except one accelerometer are mounted on the experiment bulkhead. The additional accelerometer is mounted on the module wall. The three accelerometers on the bulkhead are placed close to the two valves and the servomechanism for maximum measurement accuracy.
The electronics include a microcontroller unit, a power supply unit and four analog-to-digital-converter units including signal conditioners for each of the four channels. The microcontroller unit and the power supply unit are located in the center of the bulkhead. The analog-to-digital-converter units are placed between the microcontroller and the sensors.
A D-Sub 15 connector is used as an interface to the rocket’s service system for power supply, Start-of-Experiment- and Lift-Off Signals as well as the experiment downlink.
The downlink sends an experiment status signal to the ground station for monitoring purposes. Fig. I shows an overview over the experiment Setup.
II.II Mechanical DesignFor transferring high frequency and low amplitude vibrations a tight connection is extremely important for the servomechanism, the valves and the accelerometers. Laboratory tests resulted in the use of standard metric threads in conjunction with Loctite for maximum transfer. Furthermore a main priority was to keep the number and overall mass of components mounted on the bulkhead as low as possible to avoid high attenuation levels. Fig. II shows a CAD-drawing of the mechanical experiment design.
Fig. II: Mechanical Experiment Design
(1) signal conditioner
(5) power supply unit
(6) microcontroller unit
(7) analog-to-digital-converter unit
(8) D-Sub 15 service system interface
The mechanical design follows a modular approach which allows for easy replacement of parts in case of malfunctions. All components are easily accessible from the top. The overall weight for the entire system including the actuators and the module bulkhead is 2.2 kg.
The accelerometers chosen are PCB Piezotronics JM352C68. These are piezo-based accelerometers converting the acceleration input signal into a voltage output signal and offer a sensitivity of 100 mV/g as well as linearity in acceleration measurements in a frequency range from 0.5 Hz to 20 kHz. Texas Instruments ADS8513 analog-to-digital-converters with a 16-bit resolution, a 40 kSPS sampling rate and a measurement range of +/-5 V are used. This allows for a +/-50 g range with an accuracy of 3x10-3. The characteristic frequencies of the vibrations induced by the mechanical components showed at around 15 kHz in the laboratory experiments and lie therefore well under the maximum recordable frequency of 20 kHz limited by the linearity of the accelerometers and the maximum sampling rate of the analog-to-digital-converters.
All four channels are operated via one SPI interface which significantly reduces processor capacity needed to handle the data stream. The flash data storage is connected to the microcontroller via a second SPI interface. A USART interface is used downlink. The microcontroller employed is an STM32F103RET6 from ST Microsystems.
II.IV Experiment Timeline
After Lift-Off a certain switching sequence, which includes operation of single actuators as well as various combinations of the three actuators, is triggered in different phases of the flight. Table I specifies at which times and in what phases of the flight this sequence is triggered. The times are chosen to determine in which phases of the flight the system can be used for failure detection.
Start of switching sequence
[T+s] Flight phase
5 Rocket motor burn time
30 Atmospheric ascent
71 Exospheric ascent
100 Suborbital coast phase
143 Suborbital coast phase with with back- ground noise from other experiments before apogee
165 Suborbital coast phase with background noise from other experiments after apogee
190 Start of atmospheric re-entry
210 Hypersonic re-entry
230 Supersonic descent
251 Subsonic descent
Table I: Actuator switching sequence timeline
II:IV Data Evaluation PlanThe raw acceleration data is not suitable for classifying events as nominal or non-nominal. The vibrations created by the switching process of the valves equal a drumbeat with the following oscillations of the various eigenfrequencies of the experiment’s bulkhead. Due to the relatively high attenuation levels of the bulkhead the amplitudes of the vibrations fall under the maximum resolution of the measurement system after about 1 ms. This timespan equals 40 samples at a sampling rate of 40 kSPS. As a result the amplitude of the vibrations was deemed not suitable for a reliable classification of events.
The method chosen was the analysis of the frequency spectra of the solid-borne sound. The frequency spectrum of an event subject to investigation is gained by converting it from a function of time into a function of frequency. This is done with the Fourier transform (FT). The Fourier transform F(jω) of a function of time f(t) is defined be the following equation which is applicable to any integrable function:
Just like a chord in music, the acceleration data gathered by the experiment consists of various superimposed frequencies. For instance, a chord made up of only three main frequencies displayed as a function of amplitude versus time will appear as a plot of undefined frequency as shown in Fig. III.
Fig. III: Amplitude – Time function of a chord
When transformed into a function of frequency the three frequencies or tones it consists of will appear as discrete peaks in the frequency spectrum which allows analysis of the chord regarding the frequencies. The spectrum of a three tone chord is displayed in Fig. IV.
Fig. IV: Amplitude – Frequency function of a chord
Unlike the audio-signal in Fig. III, which is continuous over time and in amplitude, the acceleration data recorded possesses discrete time values and is quantized due to digitalization. It is therefore not integrable. The Fourier transform X(ejΩ) of a sequence of discrete values x[n] is defined as:
where Ω = ωt
This so-called DTFT (Discrete Time Fourier transform) is not a new transform, but the application of the continuous Fourier transform on a sequence of discrete signals. It allows the illustration of the frequency spectra of a signal that has been digitized.
However the DTFT is only of limited practical use because it summates over an infinite number of values. The DFT (Discrete Fourier transform) is an approximation for the DTFT and summates over a number of values N. This means that the created frequency spectrum is only valid for the observed part of the signal covered by N. This timespan is called window. The following equation defines the DFT:
where m is the number of spectral lines
The FFT (Fast Fourier transform) used for the data evaluation of the experiment is only an efficient algorithm for the calculation of a DFT.
All Fourier transforms described so far only calculate the frequency spectrum for the entire signal (FT, DTFT) or parts of the signal (DFT, FFT). However the interesting part is the development of the characteristic frequencies over time. The solution to this problem is called STFT (Short-time Fourier transform). STFT is simply a number of transforms performed with the window moving over the signal that is subjected to the transform. This means that many transforms are created at different points in time. Every transform IAC-12,E2,2,8,x16326 Page 5 of 9
creates a plot of the frequency spectrum inside the window at a specific point in time. These spectra are then put together like slices to create a 3D-Plot which shows the changing spectra over time. These plots are called spectrograms and are very commonly found in audio signal analysis. Fig. V shows an example of such a 3D-Plot. In this particular case the frequency of the signal starts at 600 Hz and rises to a frequency of 1500 Hz over a time of 10 s. This kind of spectrogram was chosen as the primary method of classifying events detected by SOMID.