Spacecraft Anomaly Analysis and Prediction System - SAAPS

The project will result in the development of the Spacecraft Anomaly Analysis and Prediction System (SAAPS). The components of SAAPS are illustrated in the figure below and will consist of

Data base and data base tools

The data base will be developed with the users in mind. This includes the identification of spacecraft operators that are interested and willing to put their anomaly data into the data base. The data base will first be set up at IRF-STL and at the end of the contract be delivered to a server at ESTEC. The tools will be implemented in HTML and Java code to be accessible through a web browser. Privileged users will be able to add new satellite anomaly data to the data base. This is important to keep the data base up to date. Others can either operate the software to analyse their data base and submit a request for permanently include their data set.

Solar-terrestrial data, such as solar wind data and magnetospheric indices, will also be included in this study. The data base tools will enable access to the data that are publicly available on the internet, such as the different world data centres, OMNI data base, and ACE solar wind data.

Anomaly analysis module

Based on the data in the data base it is possible to analyse the probable cause of the anomalies. From a Web browser the user will be able to apply the anomaly analysis module on the data base. Functions that should be included are calculations of correlations between different parameters and superposed epoch analysis. It should also be possible to define different selection criteria like anomaly type, local time, geographical position, time of year, and so on. The results will be either graphical output or tables.

Anomaly prediction module

Finally real time predictions will be made with the developed AI models based on different solar-terrestrial parameters and space environment data. A preliminary model for the prediction of spacecraft anomalies based on Kp (Wu et al., 1998) will be available early in the project. During the project the prediction module will then be extended to include more of the space weather parameters, and a second prototype will be available during 1999.

The extended data base will make it possible to further explore the relation between non-local environment data, such as Kp or solar wind data. The inclusion of more satellites will improve the model estimation, e.g. the training of the neural networks, and it will put higher confidence on the statistics of the model testing.

As the space environment is a region of many different sources and particles a satellite anomaly index (SAI) could serve as a useful parameter to the spacecraft operators and other users. The SAI will be available in real time and might be composed of several indices which take into account different effects and use different input parameters.

Initially the SAI will be a general index in the sense that the model that calculates the index is developed from this project. The aim is that in the prediction module there will also be a dedicated user defined model that can be trained on data that the user specifies. This will enable the user to have a dedicated satellite anomaly index to predict anomalies for e.g. one specific satellite or a selection of satellites.

To complement the analysis module it will be examined if models similar to the SAI predictions can be developed to make predictions of the charged particle environment at geosynchronous orbit from solar wind data and magnetospheric indices. Such predictions will help in the understanding of the predicted anomalies.

Implementation for real-time predictions

The final product from this study should be a working system for real-time analysis and predictions of satellite anomalies. The predictions are to be made on two time scales: hours in advance, and days in advance. The daily predictions are of lower confidence and possibly summarised in a single SAI. The hourly predictions will give a more detailed picture of the space environment. The program will automatically collect the relevant data, make appropriate normalisation and feed it into the models for predictions. The results will be formatted so that the predictions can be viewed using a Web browser.