High-Level Project Summary
MLSAT (Machine Learning for Safe Satellite Operations) project is aimed at creating software to support small satellite missions, extending the lifetime of the satellite in orbit. There are many facts of failure of satellite systems under the influence of space weather factors (geomagnetic disturbances, high-energy particle flows). We believe that we can prevent failures on small satellites by analyzing open telemetry data (obtained from the SatNOGS stations) and open space weather data (obtained from the DSCOVR). We are trying to predict and reduce the impact of space weather on satellite hardware, using data analysis and machine learning methods.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
Satellite Telemetry and Space Weather
Telemetry data is one of the main elements in the small spacecraft control (SSC) system, providing control of the state of its individual nodes and motion parameters. A large amount of telemetry data continuously coming from the SSC to receiving stations around the world, as well as information about the spacecraft that affects the parameters of satellite systems, requires the development of new methods and algorithms for processing the received data in order to reduce the number of failures in the operation of the SSC caused by negative space weather (SW) factors.
Timely prediction of malfunctions on small satellites is an integral part of their effective protection in the aggressive environment of near-Earth space. Some malfunctions, such as electrical breakdowns and failures in communication channels, can be predicted by measuring various parameters directly on board the spacecraft. Another approach to predicting malfunctions and failures on small satellites is a statistical one. A joint analysis of failures on satellites and variations in the parameters of the spacecraft makes it possible to establish the relationship between various states of the environment in which the device is located and the occurrence of emergency situations on it.
The combination of these methods will make it possible to comprehensively analyze the state of constellations of satellites that are similar in certain design characteristics, and, when jointly considering the geophysical situation, parameters on the satellite, its position in space, will make it possible to predict anomalies on board a particular spacecraft.
Serching for Anomalies and Their Causes
As part of the study:
- emerging anomalies in the data are analyzed, i.e. telemetry with non-nominal values of individual parameters, as well as the probable causes of such values;
- the degree of influence of space weather factors on the hardware systems of the small spacecraft was determined within the nominal values.
According to the Space Weather service of the SINP MSU, in the fourth quarter of 2020, ground stations recorded significant geomagnetic disturbances. Within the framework of this study, the emerging anomalous values in the telemetry data of the SiriusSat -1 small spacecraft in the specified period of time were analyzed. More than 323615 frames were received by uploading and decoding files from the database of the SatNOGS service .
Anomalies in the data were identified by setting boundary values for the telemetry parameters. The found data frames were fixed for further analysis. The next step was to compare the identified anomalous values with the CP indicators recorded in the same time interval. In particular, the values of planetary indices of geomagnetic activity ( Kp , Ap ), integral solar activity index (F10.7), geomagnetic activity index at low latitudes ( Dst ) were analyzed. The largest number of anomalies was identified in the data of the temperature sensors of the four batteries installed on the SSC (Table 1).
In the table, the anomalous values of the selected telemetry parameter of the SSC are highlighted in color, as well as the SP indices characterizing the disturbed geomagnetic situation ( Kp >=4, Dst <=-20). geomagnetic activity Dst . Below, we will use this index as the main characteristic of geomagnetic disturbances.
However, the influence of the spacecraft factors is not always expressed in the form of a failure or anomaly on board the small spacecraft. Heliophysical factors affect the telemetry parameters of the small spacecraft within the nominal values.
To determine the dependencies in telemetry and space weather data, we calculate the correlation matrix, which includes all available data without ranking them. The performed calculations indicate the actual absence of interrelations between changes in telemetry data and geomagnetic activity indices. However, it is known that spacecraft factors affect the small spacecraft differently in all parts of its orbit.
First of all, the assumption that the influence of cosmic factors is not the same with the side of the Earth illuminated by the Sun and not illuminated by the Sun requires verification. Let us calculate the coordinates of the location of the SSC for each time moment of the available telemetry data frames. Similarly, using the calculation of the correlation matrix, we will analyze the dependences of the change in telemetry and geomagnetic activity indices in the parts of the orbit illuminated and not illuminated by the Sun. The obtained calculations show, as before, a low correlation of the studied parameters, however, on the illuminated side, the absolute value of the correlation coefficient is much higher. Therefore, we will further study the telemetry data obtained while the SSC was in the parts of the orbit illuminated by the Sun.
The data presented above also indicate the presence of the influence of the SP factors mainly on the temperature indicators of the small spacecraft, such as t1_pw, t2_pw, t3_pw, t4_pw. Let us further explore the relationships in relation to the listed parameters.
As is known, near the magnetic poles of the planet, the lines of force of the Earth's magnetic field are closed and charged particles accumulate. This makes the polar regions a zone of particular risk for spacecraft. Let us calculate the correlation dependences of the data taking into account the geographic latitude of the small spacecraft in orbit.
The number of data frames is unevenly distributed over different latitudes and largely depends on the number of receiving stations at different points on the planet. Let's divide the available data set into approximately equal parts in accordance with the latitude (northern and southern) of the SSC in orbit and calculate the correlation coefficients for each resulting part.
The correlation index, close to and exceeding 0.3 in absolute value, which was calculat-ed for orbital segments located at a latitude of 51.5°-51.8°, indicates the presence of a rela-tionship. The number of elements in the sample confirms that the identified relationship is not random.
Unfortunately, the inclination of the SSC orbit of 51.8° does not allow a full assessment of the impact of the spacecraft on the SSC systems, but an exponentially changing trend to-wards strengthening the correlations between telemetry parameters and space weather factors at high latitudes is obvious.
Space Weather Prediction
In order to predict the occurrence of anomalies on board small spacecraft, data from The Deep Space Climate Observatory (DSCOVR) were used.
The forecast was carried out on the basis of a bidirectional LSTM model, based on every minute data, the ratio of the forecast range to the training data is 10k1.
The LSTM network works better with data that is in the range from 0 to 1, this interval can be transformed with the source data using the MinMaxScaler function from the sklearn library.
The time series was transformed into a matrix, where each row is a vector of the current value and a certain number of previous ones. The number of previous values in the vector was selected empirically during network training, 8 previous values were chosen as optimal. The time series matrix was compiled using the TimeseriesGenerator function from the keras library.
The bidirectional LSTM model contains two layers, a bidirectional layer with 264 neurons and an output layer. The model was built using the keras library.
The concept of “Operating Index” was introduced, which gives a recommendation to the operator on whether it is possible to work with the satellite in the current conditions or whether it is necessary to temporarily take the device into hibernation mode. Thus, it is proposed to solve the problem of the challenge, aimed at predicting the negative factors of space weather and reducing their impact on the hardware infrastructure.
The Operating Index takes values from 1 to 9 to evaluate the possibility of carrying out work on the satellite. 1-3 - work is possible, 4-6 - work is not recommended, 7-9 - it is necessary to put it into hibernation mode.
Software Prototype
The result of the work is a prototype of a system that allows you to visualize information about space weather. And also to warn operators of small spacecraft about the approaching negative factors of space weather.
Space Agency Data
In our project, we used data from The Deep Space Climate Observatory (DSCOVR), the National Oceanic and Atmospheric Administration (NOAA) Space Weather Prediction Center's principal asset for monitoring space weather and providing early warnings of solar events that could affect Earth.
In addition, we used data from the SatNOGS project, an open database maintained by volunteers from all over the world.
Hackathon Journey
Starting our journey, we (data analysts and programmers) chose a suitable challenge for a long time, in which we could fully apply our knowledge. Through a large list of different topics, we agreed on the idea of creating software that helps protect the important (and increasingly important) infrastructure in our lives associated with small satellites.
We chose this topic not by chance, our team members have experience with small satellites, and we are actively collaborating with teams that have successfully launched their cubesats. However, emerging failures and anomalies, as well as the search for a way to eliminate them, is an integral part of any "space" work. This prompted us to a rather creative interpretation of the challenge.
It was a really interesting experience for us. We look forward to new hackathons and new challenges from the Space Apps team.
References
SpaceX says a geomagnetic storm just doomed 40 Starlink internet satellites https://www.space.com/spacex-starlink-satellites-lost-geomagnetic-storm
Intelsat Loses Command of Galaxy 15 Satellite https://www.satellitetoday.com/broadcasting/2022/08/22/intelsat-loses-command-of-galaxy-15-satellite/
Benevolsky S.V., Mayorova V.I., Grishko D.A., Khanenya N.N. Analysis of telemetry from the Yubileiny spacecraft // Mechanical Engineering and Computer Technologies. – 2011. – no. 13. - P. 59.
Romanova N.V., Pilipenko V.A., Yagova N.V., Belov A.V. Statistical connection of the frequency of failures on geostationary satellites with flows of energetic electrons and protons // Space Research. - 2005. - T. 43, No. 3. - S. 186-193.
Pilipenko V., Yagova N., Romanova N., Allen J. Statistical relationships between satellite anomalies at geostationary orbit and high-energy particles // Advances in Space Research 37. - 2006. - P. 1192-1205.
Efitorov A.O., Myagkova I.N., Shirokiy V.R., Dolenko S.A. Dst-index prediction based on machine learning methods // Space Research. - 2018. - T. 56. - No. 6. - S. 420-428.
Camporeale E. The challenge of machine learning in space weather: Nowcasting and forecasting //Space Weather. - 2019. - T . 17. - no. 8. - S. 1166-1207.
Tags
#satellite #spaceweather #telemetry #machinelearning #ML #software