Research on Satellite Autonomous Fault Detection and Recovery

Framework

Xunjia Li

1, a

, Xin Ma

1, b

, Tao Zhang

1, c

, Xiaodong Han

2, d

and Yajie Liu

1, e, *

College of System Engineering, National University of Defense Technology, Changsha 410073, China

Institute of Telecommunication Satellite, China Academy of Space Technology, Beijing 100081, China

Corresponding author: liuyajie@nudt.edu.cn

Keywords: Autonomy, Fault detection, Recovery, Task re-planning, Artificial intelligence.

Abstract: During the operation of the satellite in orbit, the operating state may change due to the fault. The impact of

satellite failures on satellites is enormous. In this paper, satellite fault detection and recovery is taken as the

research object, and an automatic fault detection and recovery framework (AFRF) for fault recognition and

recovery of satellite autonomous operation and task re-planning is proposed. Artificial intelligence methods

are used in the framework to implement on-board rapid diagnosis faults, autonomous fault repair, and

autonomic task re-planning. The experimental results show that the proposed framework can solve the

satellite fault problem well and has a good engineering application prospect.

1. INTRODUCTION

A satellite is a platform that runs in space and can

perform a variety of shooting, communication, and

navigation tasks. The normal operation of the

satellite is the basic condition for the successful

completion of the mission. Due to the influence of

various factors such as the environment and its own

equipment on the satellite, the fault may occur

during the operation of the satellite. The impact of

the fault on the satellite is serious. After the fault

occurs, the satellite will suspend the task being

performed and enter the satellite fault diagnosis and

recovery mode. After the satellite passes the fault

diagnosis and detection, it will resume normal

operation and can continue to perform tasks. A fault

self-detection diagnosis and recovery framework is

set up, and artificial intelligence is used to detect

various data in the satellite state to determine the

occurrence of the fault. After detecting the fault, the

satellite enters the emergency mode to achieve rapid

satellite state recovery and task re-planning, so that

the satellite can start performing new tasks in the

fault repair.

With the increasing complexity of satellite

systems, various sensors obtain massive telemetry

parameters and perform anomaly detection for

telemetry data. Currently, there are four methods for

anomaly detection of satellite telemetry data: manual

monitoring combined threshold-based method,

expert system-based method, model building method

based on expert experience and data driven method.

Telemetry data tends to be regular and periodic, and

the data will fluctuate within a large range. For these

characteristics, this paper focuses on data-driven

methods. The data-driven method does not require

prior knowledge and data distribution requirements,

and is highly scalable and can be detected in real

time for streaming data. Among many methods,

prediction-based methods are currently hot topics

(Yang, Y., & Hou, N, 2013). The prediction based

anomaly detection methods include ARMA

(Weizheng, L. I., & Qiao, M, 2014), LSSVM (Bing,

C., Gang, L., Hongzheng, F., & Li, A, 2014), and

RVM [4], dynamic Bayesian network (Yairi, T.,

Kawahara, Y., Fujimaki, R., Sato, Y., & Machida,

K, 2006), ANN (Sadeghi, B. H. M, 2000; Elman, J.

L, 1990) and so on. This paper mainly uses artificial

neural network algorithm.

Satellite mission planning research is also an

important part of the satellite field. Literature (Song,

Y., Huang, D., Zhou, Z., & Chen, Y, 2018) proposed

an autonomous satellite autonomous re-planning

method, and set up three task insertion algorithms.

In (He, Y, Xing, L., & Chen, Y, 2016), from the

perspective of software design, an automatic mission

planning software based on new satellite is designed.

Li, X., Ma, X., Zhang, T., Han, X. and Liu, Y.

Research on Satellite Autonomous Fault Detection and Recovery Framework.

DOI: 10.5220/0008385602930298

In Proceedings of 5th International Conference on Vehicle, Mechanical and Electrical Engineering (ICVMEE 2019), pages 293-298

ISBN: 978-989-758-412-1

293

Literature (Song, Y. J., Zhou, Z. Y., Zhang, Z. S.,

Yao, F., & Chen, Y. W, 2019) considers several

problems in satellite mission planning and proposes

a generalized solution framework including mobile

edge computation. Literature (Zheng, Z., Jian, G., &

Gill, E, 2018) studied the multi-star system and

designed a multi-star synergy. Literature (Song, Y. J.,

Zhang, Z. S., Sun, K., Yao, F., Chen, Y. W, 2019)

solved the problem of satellite data downlink

mission planning by using genetic algorithm. This

paper will consider both satellite fault detection and

satellite mission planning, and design an efficient

and versatile autonomous operation framework.

The structure of this paper is as follows. In the

second part, the satellite autonomous fault detection

and recovery framework proposed in this paper will

be introduced. The third part will adopt the effect of

an application scenario verification framework.

Finally, the conclusions and prospects of this paper

will be given.

2. AUTONOMOUS FAULT

DETECTION AND RECOVERY

FRAMEWORK(AFRF)

Using satellite autonomous fault detection and

recovery technology can quickly diagnose faults,

allowing satellites to return to normal operating

conditions and perform new tasks in a short period

of time. The autonomous fault detection and

recovery framework provides a solution for the

satellite to quickly detect and diagnose faults and

transition from an abnormal state to a normal

working state. First, we will analyze the functional

requirements of the autonomous fault detection and

recovery framework for satellites to meet

autonomous operation and fault recovery. After that,

we will give the overall structure of AFRF and give

the specific content of each module.

2.1 Functional Requirements

Combined with the characteristics of satellite

autonomous operation and execution tasks,

according to the high reliability and high stability

operation requirements of satellites, the functional

requirements for rapid fault detection, autonomous

fault recovery and on-board mission re-planning are

proposed for satellite autonomous fault detection

and recovery.

Fast fault detection the occurrence of the fault is

serious for the satellite, which will result in the

satellite not functioning properly or even the satellite

being scrapped. The satellite needs to establish a

complete fault detection mechanism, monitor the

satellite health status according to the data of each

sub-system on the star, and adjust the abnormal sub-

system to a normal level. When a fault occurs,

quickly locate the sub-system in which the fault

occurred based on the satellite's existing fault

template library and determine the specific fault

location.

Autonomous fault recovery Satellites will not

function properly after a satellite failure, which

requires a quick return of the satellite to its normal

state. The traditional way of using ground

commands for satellite fault repair has strict

requirements on satellite position, and the timeliness

of repair is not high. Autonomous fault repair allows

the satellite to initiate the fault repair process at the

same time as the fault occurs, and the satellite can be

quickly restored to normal operation in a short

period of time, and the task can be continued.

On-board task re-planning after the satellite

failure occurs, the original task will be aborted until

the fault is repaired. At this time, since the satellite

position is different from the position before the

fault, many tasks will not be executed. The satellite

should regenerate a new mission execution plan after

the fault is fixed and complete the subsequent tasks

according to the new scheme.

2.2 Overall Design

Throughout the framework, the main functions

completed are training and improvement of fault

detection models, fault detection and re-planning,

and fault repair. Among them, the training and

improvement of the fault detection model is carried

out on the ground, because the ground has more

sufficient computing resources than the satellite,

which can better support the training and

improvement process of the artificial intelligence

model. Fault detection and re-planning, fault repair

is performed on the satellite. Satellite self-execution

fault diagnosis and repair has the ability to quickly

detect faults, quickly diagnose faults and repair, and

ensure that the satellites return to normal operation

in a short time.

ICVMEE 2019 - 5th International Conference on Vehicle, Mechanical and Electrical Engineering

294

Anomaly detection

Fault template matching

Fault detection

Generate a new set of

tasks

Task replanning

Autonomous fault repair Satellite status update

Fault and repair data

downlink

Ground

Fault Detection and

Replanning

Fault Repair

Fault Detection

Model Training and

Improvement

Offline training model

and inject satellite

Further model training

Figure 1. AFRF overall framework.

The artificial intelligence model is trained and

injected into the satellite before the satellite starts to

run. After acquiring the satellite fault data, the

model is transmitted to the ground for retraining of

the model to improve the prediction accuracy. The

improved model is re-injected to the satellite when

the satellite has an available satellite

communications link.

After the satellite is injected into the satellite, the

operational status of each satellite system is

monitored according to the operational data of each

satellite subsystem, and the model is used to

determine whether there is a possibility of failure. If

it is determined according to the model that the

satellite has a possibility of failure, the fault is

automatically detected. After the satellite detects the

cause of the fault, the template library is used for

autonomous fault matching and autonomous repair

is performed. After the repair process is completed,

the satellite parameters are updated and the fault and

repair information is transmitted to the ground via

the star link.

After satellite autonomous fault repair, the

satellite returns to the normal state, and the

corresponding data is provided to the task planning

part. The task planning part updates the task status

according to the data, generates a new task set, and

performs task re-planning according to the new task

set.

Through the ground model training and learning

in the whole framework, the satellite independently

detects faults and repairs faults, and re-plans the

tasks to be performed according to the new mission

situation after the satellites resume normal operation.

Our proposed AFRF can detect faults and fix faults

more quickly, adjust the task execution plan in time,

and let the satellites detect the faults and repair

process after rapid detection, and continue to run

normally and complete various types of tasks.

2.3 Module Design

In AFRF, the core part is fault diagnosis and

identification, autonomous fault repair, and task re-

planning. In the following, the module design will

be carried out separately for these three parts.

2.3.1 Satellite Fault Diagnosis Module

Design

The satellite fault diagnosis module includes satellite

data anomaly detection and fault identification. The

data anomaly detection function uses artificial

intelligence tools to run on ground, run on the star,

and use new data to further improve the model's way

of running. The fault identification function uses the

satellite's existing fault template library for fault

analysis and identification based on the abnormality

found by the anomaly detection function, and

accurately identifies the satellite subsystem and the

specific fault cause in which the fault occurs in the

shortest possible time.

The module adopts neural network-based

prediction method. The algorithm is based on

Research on Satellite Autonomous Fault Detection and Recovery Framework

295

traditional time series prediction technology. It is

assumed that the telemetry data has temporal

correlation. The new data can be obtained by

historical data recursively by establishing a time

window model. The range of new data is passed.

The established model predicts the mean and

variance. When the real data arrives, if the real data

is in the interval, no abnormality occurs; otherwise,

an abnormality is considered.

Considering that BP neural network (Sadeghi, B.

H. M. 2000) has the disadvantages of slow learning

speed, limited network promotion ability and cannot

effectively process different historical time

information in data, Elman neural network (Elman, J.

L. 1990) overcomes the above shortcomings, and

Elman neural network will hide the previous

moment. The information in the layer-containing

unit is used as part of the input data of the unit at the

current time, so that the units in the network have a

memory function, and can be continuously updated

as the data changes, thereby better learning

information and rules in different time series. . It can

realize the modeling of static systems and the

mapping of dynamic systems. Its computing power

and network stability have obvious advantages over

BP neural networks.

2.3.2 Satellite Autonomous Fault Repair

Module Design

After the satellite fault diagnosis module finds and

detects the cause of the fault, the satellite performs

emergency treatment and satellite fault repair

according to the fault plan. The specific process of

fault repair is determined according to the severity

of the fault. After the fault repair is completed, the

satellite summarizes the fault information and the

repair result and then transmits it to the ground to

facilitate the ground control center to decide whether

to further repair the faulty satellite or take other

measures to ensure the normal operation of the

satellite in orbit.

2.3.3 Satellite Independent Task Re-

planning Module Design

The satellite returns to normal operation after the

fault is repaired and needs to perform new tasks. The

autonomous task re-planning module is used to add

the tasks that are not successfully executed and will

be executed in the future to the to-be-planned task

set, and use the on-orbit dynamic programming

algorithm to solve the task set and generate a new

task execution plan. The satellite performs the

corresponding tasks in the next phase based on the

results of the re-planning.

3. APPLICATION EXAMPLES

In order to verify the validity of the proposed AFRF

framework, we conducted a complete application

example experiment. First of all, in the process of

satellite operation, there may be a situation of

failure. The autonomic fault detection is performed

by using Elman neural network for telemetry data.

The detection result is shown in Figure 2. The blue

line is the real data, the red line is the prediction

data, and the two green lines are the upper and lower

bounds of the prediction interval respectively. The

fault occurs at about 1400 and 1700. It can be seen

from the figure that when there is no fault, the real

data is basically in the prediction interval. When the

fault occurs, the real data value is outside the

prediction interval, thus detecting the abnormality,

which proves the accuracy of the method.

After fault detection, the satellite uses the AFRF

framework for autonomous fault repair, and the task

can be re-executed by restoring the satellite to

normal state. We use three commonly used heuristic

rules for autonomous re-planning. The goal of the

plan is the completion rate of task re-planning. The

results are shown in Table 1.

Table 1. Task re-planning result.

Instance

Heuristic1

Heuristic2

Heuristic3

0.92

0.96

0.92

0.88

0.89

100

0.90

0.88

0.92

125

0.83

0.85

150

0.74

0.80

0.81

ICVMEE 2019 - 5th International Conference on Vehicle, Mechanical and Electrical Engineering

296

Figure 2. Fault detection result.

As can be seen from Table 1, the use of these

three heuristic rules can quickly achieve satellite

mission re-planning. After completing the re-

planning of the mission, the satellite performs the

mission according to the new mission, and the value

of the satellite is utilized. It can be seen from the

above experiments that the AFRF framework

proposed by us is effective for fault detection and

repair and task re-planning, and can well restore the

satellite to normal operation.

4. CONCLUSION

Satellite fault detection and repair is of great

significance for ensuring the normal operation of the

satellite and successfully completing the mission.

This paper takes satellite fault detection and repair

as the research object, analyzes the satellite fault

repair and re-planning process, and proposes a

satellite autonomous fault detection and recovery

framework. In our proposed satellite autonomous

fault detection and recovery framework, it includes

ground model training and improvement, on-board

autonomous fault detection and re-planning, and

satellite autonomous fault repair walking. We use

artificial intelligence methods in the framework to

detect satellite faults using artificial intelligence. The

task re-planning is implemented by a heuristic

algorithm after the satellite resumes normal

operation, allowing the satellite to function better.

In the following research, we will try our use of

the autonomous fault detection and repair

framework on the actual application platform. At the

same time, the verification of multiple artificial

intelligence models under the framework of

autonomous fault detection and repair, the selection

of the most appropriate model method is also the

next step.

REFERENCES

Bing, C., Gang, L., Hongzheng, F., & Li, A. (2014).

Method of satellite anomaly detection based on least

squares support vector machine. Computer

Measurement & Control.

Elman, J. L. (1990). Finding structure in time. Cognitive

Science, 14(2), 179-211.

He, Y., Xing, L., & Chen, Y. (2016). Software design of

autonomous mission planning for new imaging

satellite. International Conference on Space

Operations.

Sadeghi, B. H. M. (2000). A bp-neural network predictor

model for plastic injection molding process. Journal of

Materials Processing Technology, 103(3), 411-416.

Song, Y., Huang, D., Zhou, Z., & Chen, Y. (2018). An

emergency task autonomous planning method of agile

imaging satellite. Eurasip Journal on Image & Video

Processing, 2018(1), 29.

0 200 400 600 800 1000 1200 1400 1600 1800 2000

Time

-6.8

-6.6

-6.4

-6.2

-6

-5.8

-5.6

-5.4

-5.2

-5

Value

Abnormal detection results based on Elman neural network

True data

Prediction data

Interval bound

Research on Satellite Autonomous Fault Detection and Recovery Framework

297

Song, Y. J., Zhang, Z. S., Sun, K., Yao, F., Chen, Y. W.

(2019) A Heuristic Genetic Algorithm for Regional

Targets’ Small Satellite Image Downlink Scheduling

Problem, International Journal of Aerospace

Engineering, 2019, 1-13.

Song, Y. J., Zhou, Z. Y., Zhang, Z. S., Yao, F., & Chen, Y.

W. (2019). A framework involving mec: imaging

satellites mission planning. Neural Computing and

Applications, 1-12.

Weizheng, L. I., & Qiao, M. (2014). Fault detection for in-

orbit satellites using an adaptive prediction model.

Chinese Journal of Space Science, 34(2), 201-207.

Yairi, T., Kawahara, Y., Fujimaki, R., Sato, Y., &

Machida, K. (2006). Telemetry-mining: A machine

learning approach to anomaly detection and fault

diagnosis for space systems. IEEE International

Conference on Space Mission Challenges for

Information Technology. IEEE Computer Society.

Yang, Y., & Hou, N. (2013). Data series forecasting and

anomaly detection methods based on online least

squares support vector machine. Control Conference.

IEEE.

Zheng, Z., Jian, G., & Gill, E. (2018). Onboard

autonomous mission re-planning for multi-satellite

system. Acta Astronautica, 145, 28-43.

ICVMEE 2019 - 5th International Conference on Vehicle, Mechanical and Electrical Engineering

298