Case 1 - ENVISAT Gyroscope Monitoring: Case Summary

Size: px

Start display at page:

Download "Case 1 - ENVISAT Gyroscope Monitoring: Case Summary"

Geoffrey Ashley Warner
5 years ago
Views:

Code FUZZY_134_005_1-0 Edition 1-0 Date 22.03.02 Customer ESOC-ESA: European Space Agency Ref.

1 Code FUZZY_134_005_1-0 Edition 1-0 Date Customer ESOC-ESA: European Space Agency Ref. Customer AO/1-3874/01/D/HK Fuzzy Logic for Mission Control Processes Case 1 - ENVISAT Gyroscope Monitoring: Case Summary GTD was the prime contractor in this project

2 Cod. FUZZY_134_005_1-0 Ed. 1-0 Date Pag. : 2/ 6 1 INTRODUCTION AND OBJECTIVES This work describes a fuzzy expert system specifically designed for monitoring and diagnosing ENVISAT-1 satellite gyroscopes. This system, the ENVISAT Gyroscope Monitor, was jointly developed by GTD + and UNINOVA* for the European Space Agency project Fuzzy Logic for Mission Control Processes. Two main goals were defined for this system. One was to have a diagnostic tool capable of issuing intermediate warning alarms several days or weeks before a failure is detected by the satellite on-board software, in addition to the currently used Out of Limits 1 (OOL) alarms. The second objective was to automate some of the analysis done by the Flight Control Team. The ENVISAT Gyroscope Monitor architecture is composed of two main components: a database, used to store the satellite telemetry data of the gyroscopes; the expert system itself, containing the domain knowledge formalized in a set of fuzzy rules and the inference engine. The database was designed using decision-support concepts namely from data warehousing area ([1], [2]). The fuzzy expert system is capable of analyzing ENVISAT-1 gyroscopes telemetry data and issuing warnings and alarms of future failures. Two basic types of situations regarding gyroscope alarms exist. Either the domain experts can conclude from the data that a specific gyroscope is faulty or the behaviour cannot be directly associated with one gyroscope and a system level warning is raised. The flight control engineers can then check the status of other sensors and decide what is the defective component. In this paper, we describe in more detail how we compute the gyroscope specific diagnostic. The system also computes a data quality level that is measured with a fuzzy rule model similar to the ones used for gyroscope and system level diagnostic. 1.1 MOTIVATIONS Successor of the ERS-2 satellite, ENVISAT-1 ([3]) is the new component of the European Remote Sensing (ERS) program, a long-term program of environmental remote sensing of the Earth. The ERS initiative is targeted at collecting highly reliable data on the entire planet to be used in scientific studies of critical environmental problems. ENVISAT-1 uses four two-axis dry tuned gyroscopes in its Inertial Measurement Unit (attitude control determination and control topics can be found in [4]). This unit is redundant (there are more sensitive axis and gyroscopes than necessary) to provide for coherency checking of outputs and specially to supply backup gyros in case of failure. The orientation of the satellite, termed attitude, needs to be very accurate (solar array pointed to the sun, environmental sensors pointed to the earth, sensitive parts protected from solar wind, etc). A severe failure in the Inertial Measurement Unit is then critical because in the worst case it could mean the end of the satellite's life. Flight control engineers have learnt from experience during past missions how to recognize gyroscope progressive degradation through specific data transformation, data analysis and engineering judgment. A solution to automate the data analysis and diagnostic of ENVISAT gyroscopes was required to support the Flight Control Team during routine operation. This system will play an advisory role while the flight control engineers remain responsible to declare the gyroscope faulty and to ensure that the correct actions are implemented. They currently use three main fault-detection techniques for monitoring: In orbit autonomous surveillance algorithms. These are automatic checks performed by the on-board computer. Out of Limits alarms issued by the Mission Control System based on spacecraft telemetry data. On the ground off-line checks of the gyroscope telemetry data, performed by the Flight Control Team. These techniques have their own advantages and disadvantages for the diagnostic. They are targeted for shortterm detection and avoidance of false alarms and so they will not trigger any intermediate level alarms. In addition, they are defined as a set of rigid rules where each rules either is satisfied or not. A gyroscope can stay below a limit, but the actual pattern of the gyroscope data is known to be a sign of future problems. 1 Out of Limits Alarms are simple crisp rules. A range or a fixed value is defined for a parameter. If the parameter goes outside the range or above the value then an Out of Limit is issued to the spacecraft controller.

3 Cod. FUZZY_134_005_1-0 Ed. 1-0 Date Pag. : 3/ 6 This led to our understanding of the problem as well suited for a fuzzy expert approach namely because the number of occurrences of bad events (e.g. number of similar failures that have happened in the past) is very limited and the knowledge to detect an anomaly is mainly with experts without some form of knowledge engineering. No data is available to automatically induce new rules and no detailed mathematical model exists to describe the gyroscope. To solve the gyroscope diagnostic problem, the development focused on two important aspects. First, the extraction of knowledge from experts and the respective supporting of the monitoring and decision process. Second, the development of a fuzzy expert system that fuzzifies crisp rules and performs the inference. The fuzzy expert system supplies a final composite score of gyroscope health status that will generate various alarm levels. Figure 1 shows a summary of the full process of producing a diagnostic. Raw inputs (from TM data) Note: Different time windows IDVA outputs Random Drift Random Noise Gyro Mode! " # $ Transform raw data into derived variables for diagnostic process Estimate time series Knowledge base Transform crisp inputs into fuzzy sets using me mbership functions Fuzz ificat ion Using the faultdetection model (expressed in a set of rules), infer the diagnostic Inference Engine Figure 1 - Fuzzy fault-detection Process. Transform the fuzzy outputs of the model into a crisp alarm level Preprocessing Defuzzification Gyroscope alarm level 2 SYSTEM DESIGN & DEVELOPMENT 2.1 GYROSCOPE DIAGNOSTIC VARIABLES The strategy to diagnose a gyroscope uses previous experts knowledge and manufacturer specifications. In summary, the power spectral density (PSD) of the gyroscope output signal has certain properties, specifically a nominal PSD should have certain energy values and show a known phenomenon named hunting frequency. The hunting frequency acts like a signature of the gyroscope and qualitative changes to it can be related with specific mechanical failures. This approach is only possible when the gyroscope is in a specific mode of operation that corresponds to the nominal mode of operation and that of the highest precision. Gyroscope diagnostic variables are all based on the PSD of each gyroscope output (named IDVA 2 ). As shown in Figure 2 the two sets of variables are denoted hunting noise and random noise. In addition, Figure 2 defines the crisp rules that are going to be used in the fuzzy expert system for the alarm levels. These variables and rules were provided by the manufacturer of the gyroscopes. Gyro PSD monitoring Analysis of hunting noise energy and frequency Analysis of gyro noise level No more hunting Hunting frequency change > 0.15Hz Hunting frequency energy Hunting noise energy increased by 3 and random noise not increased by 3 Random noise increased by 3 Alert 1 Warning Alert 2 Figure 2 Gyroscope variables and rules. The first set of variables has four classes per gyroscope, which are further divided in two axes. We use the PSD of each IDVA variables to determine the hunting frequency or noise level, i.e. a hunting frequency variable is 2 IDVA stands for Integrateur Digital de Vitesse Angulaire or Angular Speed Digital Integrator. The IDVA signal is an 8Hz snapshot of the contents of a simple digital counter. This counter is incremented or decremented when the gyroscope senses a change in the sensitive axis.

4 Cod. FUZZY_134_005_1-0 Ed. 1-0 Date Pag. : 4/ 6 obtained from the power spectral density (PSD) for each axis gyroscope output (IDVA). Further, the PSD will be represented by the variable PSD (1:2048), where each beam has the frequency 4/2048 Hz. One variable H i,j (where i represents the gyroscope and j the axis) to check if there is, or not, a hunting frequency; a second variable HV i,j for the hunting frequency value; a third variable HC i,j for the amplitude of the PSD coefficient of the hunting frequency and a fourth variable HN i,j for the energy in the band f hunting ±0.35Hz. Table 1 depicts the variables for one gyroscope. Variables H i,1 H i,2 HV i,1 HV i,2 HC i,1 HC i,2 HN i,1 HN i,2 Description Existence of hunting frequency Deviation of hunting frequency value Amplitude of the PSD coefficient of the hunting frequency Energy in the band f hunting ±0.35Hz Table 1 The variables for the gyroscope "i" The second set of variables relates with the random noise. We consider three variables for each axis gyroscope output and each variable is a function of the energy in three different bands. The first band is defined in the interval [0.01Hz; 1 Hz], the second in [1 Hz; 2Hz], and the third in [2Hz, 4Hz]. Variables RN1 i,j RN2 i,j RN3 i,j Description The random noise in the first band The random noise in the second band The random noise in the third band Table 2 The variables for the gyroscope "i" 2.2 FUZZY EXPERT SYSTEM In the fuzzy expert system, we started by fuzzifying all the variables described above. We defined these variables as linguistic variables with different labels. For instance deviation for the hunting frequency, f, is given by 4 f Hunting = arg max PSD() i f ref 2048 i 1536 i 2048 The respective fuzzy linguistic variable, for the deviation of the hunting frequency HV i,j, has two labels {Tolerable, Not tolerable}. Figure 3 depicts the fuzzy label Tolerable and "Not tolerable" is obtained by the negation operation (not). Experts supplied all the parameters used in the fuzzy sets construction. µ ( f ) 1 Tolerable α 1 β 1 0 β 2 α 2 f Figure 3 Example of a fuzzy set used for the linguistic label the deviation of the hunting frequency from the nominal reference value is tolerable. There are 4 composite rules that use the linguistic labels defined for each gyroscope variable. The fuzzy faultdetection rules are based in Figure 2 and they are the following:

5 Cod. FUZZY_134_005_1-0 Ed. 1-0 Date Pag. : 5/ 6 If (H i1 is "Exist" or H i2 is "Exist") and HV i1 is "Tolerable" and HV i2 is "Tolerable" and HC i1 is "Tolerable" and HC i2 is "Tolerable" and HN i1 is "Tolerable" and HN i2 is "Tolerable" and RN1 i1 is "Tolerable" and RN2 i1 is "Tolerable" and RN3 i1 is "Tolerable" RN1 i2 is "Tolerable" and RN2 i2 is "Tolerable" and RN3 i2 is "Tolerable" then Alarm is "Not warning" If (H i1 is "Exist" or H i2 is "Exist") and HV i1 is "Tolerable" and HV i2 is "Tolerable" and HC i1 is "Tolerable" and HC i2 is "Tolerable" and RN1 i1 is "Tolerable" and RN2 i1 is "Tolerable" and RN3 i1 is "Tolerable" RN1 i2 is "Tolerable" and RN2 i2 is "Tolerable" and RN3 i2 is "Tolerable" and (HN i1 is "Not tolerable" or HN i2 is "Not tolerable") then Alarm is "Warning" If (H i1 is "Exist" or H i2 is "Exist") and HV i1 is "Tolerable" and HV i2 is "Tolerable" and HC i1 is "Tolerable" and HC i2 is "Tolerable" and (RN1 i1 is "Not tolerable" or RN2 i1 is "Not tolerable" or RN3 i1 is "Not tolerable" or RN1 i2 is "Not tolerable" or RN2 i2 is "Not tolerable" or RN3 i2 is "Not tolerable") then Alarm is "Alert2" If H i1 is "Does not exist" and H i2 is "Does not exist" or HV i1 is "Not tolerable" or HV i2 is "Not tolerable" or HC i1 is "Not tolerable" or HC i2 is "Not tolerable" then Alarm is "Alert1" The validation of the tool will be carried out in two phases: the first, so-called pre-flight validation will be based on data made available during on-ground satellite test sessions. It will be used to fine-tune the diagnostic knowledge of the fuzzy logic inference engine, in term of membership function and rules definition. The second phase, will take place after the launch of ENVISAT, currently scheduled for the beginning of March 2002, with the objective to verify the overall performance of the diagnostic tool. 3 ENVISAT GYROSCOPE MONITOR ARCHITECTURE A context diagram of the ENVISAT Gyroscope Monitor is shown in figure 4. Satellite ENVISAT Gyroscope Monitor Application Software Spacecraft Controller Ground Station Mission Control System SPEVAL Extraction Transformation and Loading ENVISAT Gyroscope Monitor ODS System Engineer ENVISAT Gyroscope Monitor Figure 4 - Context Diagram of the ENVISAT Gyroscope Monitor. The development methodology made extensive use of evolutionary prototype concepts and high user involvement as proposed in the Dynamic System Design Methodology [5]. The system consists of an Operational Data Store that keeps the telemetry data along with other information (e.g. fuzzy models, reference values, calibration factors) and the Application Software, which is the software that the end-users interact with to look at diagnostics and analysis. The Mission Control System (MCS) controls the satellite in real time receiving telemetry and sending back telecommands. Amongst several functionalities, the MCS generates Out of Limit alarms on variables defined by spacecraft controllers. SPEVAL is a software tool that keeps a record of all the data generated by the MCS. SPEVAL is used as the data source for our system.

Cod. FUZZY_134_005_1-0 Ed. 1-0 Date 22.03.02 Pag. : 6/ 6 4 CONCLUSIONS In this work, we described a fault detection model for gyroscopes based on a fuzzy expert system solution.

The fuzzy expert system approach was considered an adequate and correct option given the availability of previous experience, the qualitative nature of the diagnostic process and the inexistence of

6 Cod. FUZZY_134_005_1-0 Ed. 1-0 Date Pag. : 6/ 6 4 CONCLUSIONS In this work, we described a fault detection model for gyroscopes based on a fuzzy expert system solution. It is expected that this model improve the current operational processes of monitoring the gyroscopes by detecting earlier on potential failures. The fuzzy expert system approach was considered an adequate and correct option given the availability of previous experience, the qualitative nature of the diagnostic process and the inexistence of historical data on gyroscope failures. The model presented here is now embedded in a decision-support system, the ENVISAT Gyroscope Monitor, specifically designed for monitoring and diagnosing ENVISAT satellite gyroscopes. The system also includes other fuzzy expert systems: one computes a data quality level and another one issues generic system level alarms (when a problem cannot be attributed to a specific gyroscope). The expected benefits of the fuzzy logic approach are demonstrated with the operational experience of the diagnostic tool. Gyroscopes having a slow degradation processes is the key assumption here (they don't stop functioning in a short-time), and the evolution of the diagnostic provided by the current model will detect such degradation, improving on the currently simplistic alarms produced by the available systems (simple crisp rules). Figure 5 - Snapshots of the end-user human machine interface. The first window shows the diagnostic evolution of all four gyroscopes (usually only two are turned on) and the system level diagnostic. All of the former have data quality levels shown for the spacecraft controller to crosscheck the validity of a warning. The second window shows the detailed explanation for a diagnostic. It plots the same diagnostic evolution information for a specific gyroscope but also writes the fuzzy rules selected by the inference process along with the input variables values. 1. Inmon, W.H., Building the Operational Data Store. 2nd ed. 1999: John Wiley & Sons. 2. Kimball, R., et al., The Data Warehouse Lifecycle Toolkit : Expert Methods for Designing, Developing, and Deploying Data Warehouses. 1998: John Wiley & Sons. 3. ESA, ENVISAT-1 Mission & System Summary. 1998, European Space Agency. 4. Wertz, J.R., ed. Spacecraft attitude determination and control. Astrophysics and space science library. Vol , Kluwer Academic Publishers Stapleton, J., DSDM Dynamic system development method - The method in practice, ed. D. consortium. Vol : Addison-Wesley. 163.

Autonomous and Autonomic Systems: With Applications to NASA Intelligent Spacecraft Operations and Exploration Systems

Walt Truszkowski, Harold L. Hallock, Christopher Rouff, Jay Karlin, James Rash, Mike Hinchey, and Roy Sterritt Autonomous and Autonomic Systems: With Applications to NASA Intelligent Spacecraft Operations