03F SIW 019

On the prediction of anomalous events onboard spacecraft: 

On the prediction of anomalous events onboard spacecraft Dr. Venkat V S S Sastry and Maj. Timothy M P Mountford Engineering Systems Department Cranfield University, RMCS Shrivenham Wilts. SN6 8LA, UK Sastry@rmcs.cranfield.ac.uk

Motivation: 

Motivation Space craft constantly exposed to various risks caused by space weather Space weather refers to magnetic, electro magnetic or particulate activity Anomalous behaviour can be disruptive Source:www.exploratorium.edu/ solarmax/whatis.html

What is space weather?: 

What is space weather? "conditions on the Sun and in the solar wind, magnetosphere, ionosphere and thermosphere that can influence the performance and reliability of space-borne and ground-based technological systems and can endanger human life or health." Source:www.noaanews.noaa.gov/ stories/s80b.htm

Space weather: some important domains: 

Space weather: some important domains From State of the art of space weather modelling and proposed ESA strategy, Technical Note SPEE-WP310-TN-1.2, Issue 1.2 October 12, 1998 (Page 8)

What is an anomaly?: 

What is an anomaly? An anomaly is any form of unexpected action or activity on board spacecraft Typical anomalies Surface charging Deep dieletric or bulk charging Single event upsets Spacecraft orientation effects Solar radio frequency interference Source:http://www.sec.noaa.gov/SatOps/

Distribution of anomalies: 

Distribution of anomalies Source: Furton Inc.

Anomaly Data: 

Anomaly Data Anomaly database for SKYNET 4 spacecraft (Davidson, 2002) Covers 12 year period 1990-2001 Total anomalies: 1300 471 of them are associated with ESD

Environmental data: 

Environmental data Energy particle (electron flux) data (>2MeV) Dst (Disturbance storm) index Geomagnetic data Kp Kp – weighted average of K index values from several observatories on the earth’s surface Use of Kp is further supported by Swedish studies (Andersson, 2000)

Sample data for 1993: 

Sample data for 1993 Electron flux data Electron flux data, Kp with anomaly data

Neural network models: 

Neural network models Model 1 Three inputs – electron flux data from GOES5 and GOES6 and Kp Two outputs – anomaly, nonanomaly Training data Dataset for 1993 -- 48 anomalies and 100 samples randomly chosen from 8712 non-anomalies Ten hidden layer neurons Transfer function: log-sigmoid function

Performance of the network: 

Performance of the network Consider misclassifications on the complete dataset G : anomaly , B : non-anomaly

Sample results – Model1: 

Sample results – Model1 48 anomalies Good classification on anomalous events Poor performance on non-anomalous events Run 1 Run 2

Model 2 – point estimates: 

Model 2 – point estimates Divide dataset for the year into blocks of given length Associate cumulative sum of environmental inputs to Corresponding sum for anomalous events Note – more than one anomaly may occur in a given block Using block length 60 for the year 1993 42 blocks associated with anomalous events 104 blocks associated with non-anomalous events

Sample results for Model 2: 

Sample results for Model 2 Able to learn non-anomalies (83%) better than Anomalies (69%) Performance of block sizes of 120 and 240 degraded substantially Point estimates of time histories were abandoned

Kp index: 

Kp index Kp INDEX. A 3-hourly planetary geomagnetic index of activity generated in Gottingen, Germany, based on the K INDEX from 12 or 13 stations distributed around the world. Source: http://www.sec.noaa.gov/info/glossary.html#KCORONA

Model 3: 

Model 3 Consider environmental data Kp only Associate an anomaly with a fixed length of the data series Training set – dataset for 1993 Test set – dataset for the period 10 – 19 March 1994 consists of four anomalies on Skynet and 13 ESD anomalies on other spacecraft

Sample results for Model 3: 

Sample results for Model 3 Dataset: year 1993 Dataset: 10-19 March 1994

Anomaly predictions for 10-19 March 1994: 

Anomaly predictions for 10-19 March 1994

Recorded anomalies: 

Recorded anomalies

Summary: 

Summary Space weather prediction is relevant Available space environmental data can be used to improve prediction of anomalies Data series associated with geomagnetic activity, Kp is found to be a better predictor Neural network models require further analysis