Predictive Maintenance of Manned Spacecraft Through Remaining Useful Life Estimation Technique

CHEN Runfeng, YANG Hong

Institute of Manned Space System Engineering, Beijing 100094

Abstract: Manned spacecraft pose challenges in terms of extremely high safety and reliability, and with the growth of system complexity and longer on-orbit operation time, the traditional management mode, such as monitoring the threshold of parameter passively, is dif ficult to meet the required safety standards. Predictive maintenance, which analyzes the system heath trend and estimates remaining useful life (RUL) to establish maintenance strategies ahead of time before failure occurs, is a new mode to approach maintenance tasks. Here, a predictive maintenance strategy for complex manned spacecraft is proposed based on the remaining useful life estimation technique. Firstly, a health index is established based on an abundance of telemetry data, reflecting the system’s current health state. Secondly, we map the health index to the remaining useful life through system degradation modelling, taking into consideration both the system’s stochastic deterioration and uncertainty. The maintenance and management strategies are then made based on the calculated distribution of RUL time. Finally, a case study on Chinese space station energy system predictive maintenance is presented.

Key words: remaining useful life, predictive maintenance, Chinese space station

1 INTRODUCTION

Manned spacecraft is a type of space vehicle that transports astronauts into space. Compared with other types of space vehicles, astronauts are taken as the core of the entire system and the astronauts’ safety is given the highest priority. Russia and the United States of America started their manned space flight programs in the early 1960s, built well-functioning manned spaceflight systems and made many achievements. All sixteen countries build the International Space Station based on the experiences of the Soyuz series and Skylab series. China started its manned space flight program in the late 1990s, and with the successes of the Shenzhou series and Tiangong series, China has become the third country with the ability to independently conduct manned space flights.

Currently, manned spacecraft management has a strong dependency on ground support, which typically consists a master control center, multiple receiving and monitoring stations,uplink antenna, communications networks and ground personnel. Once an on-orbit fault occurs, related experts analyze the causes based on the downlink telemetry data, make management strategies and uplink control commands to mitigate the effect of the fault on the spacecraft platform. This kind of groundbased management paradigm had been used over the past decades. However, as the complexity of manned spacecraft system increases and on-orbit operating time becomes longer,the costs of a ground-based operation will increase dramatically. Besides, on-orbit faults can put astronauts’ safety at risk, if core subsystems fail, such as energy subsystem, the space flight mission should be halted, and astronauts be returned to the ground as soon as possible. This will cause great economic loss and threaten the life of astronauts. So, faults should be dealt with quickly and ef ficiently, properly preventing the occurrence occurring.

Predictive maintenance technique is active in nature, which analyzes the condition of the system and predicts when and where maintenance should be performed[1]. By knowing which subsystem or device needs maintenance, complete management work can be better planned. This is particularly meaningful to manned spacecraft as the support ability of astronauts to perform the maintenance and spare parts of the needed devices are all constrained. What’s more, predictive maintenance can to some extent prevent unexpected failure from happening and thus improve the reliability and safety. There are several works on the predictive maintenance technique. JIANG Xiuhong proposed a reliability-based predictive maintenance method,which analyzed the degradation of the system through a reliability model and then determined the time and sequence of maintenance[2]. WANG Yiwei developed a cost-driven predictive maintenance policy for airframes by using a model-based prognostics method[3]. Panagiotis Aivaliotis calculated the RUL by the use of physical models and then scheduled the maintenance activities correspondingly[4]. FAN Hongdong proposed a cooperative predictive maintenance model for repairable systems with the incorporation of a hazard-rate function and effective age factors[5]. YOU Mingyi proposed an updated sequential predictive maintenance policy to determine online predictive maintenance[6]. From the work above, it can be seen that the research mainly consists of two parts: the prognostics part and the maintenance part. As the maintenance activity is naturally target-oriented, that is, the detailed maintenance strategies are made according to the characteristics of different targets,the core lies in prognostics, which is performed taking the real health condition of systems into consideration. In this paper, we explore a predictive maintenance strategy for complex manned spacecraft, paving the way for the development of a predictive maintenance paradigm for spacecraft.

2 PREDICTIVE MAINTENANCE

Predictive maintenance technique is designed to help determine the condition of in-service systems to predict when maintenance should be performed. The main promise of this kind of maintenance is to allow the scheduling of corrective maintenance and what’s more, to prevent unexpected equipment failures. Generally, the maintenance strategy is object-oriented which means different strategies are made according to the distinctive characteristics of systems. The key of predictive maintenance lies in the predicting information used to make strategies,namely, prognostics.

Prognostics is an engineering discipline focused on predicting when a system or a component will no longer perform its intended function, and this predicted time is called Remaining Useful Life (RUL). Prognostics has gained more and more attention in recent years, and RUL estimation technique is one hot subject in current research. RUL sometimes can be taken as a random variable, which depends on the system’s current operational condition, the working environment and usage wear-out condition. There are two major ways to estimate the system’s RUL. One is called model-based RUL estimation, and the other is called data-driven RUL estimation. The model-based method, just as its name implies, relies on the system’s degradation modeling. It attempts to incorporate physical models or experience models for the system into the RUL estimation. By comparing the actual working condition and the expected working condition through degradation models,the system’s condition trend in the future can be extrapolated out. The common model-based RUL methods include the dynamic process model method, physical model method, Kalman filter method[7], Extended Kalman filter method[8], particle filter method[9]and experience model method. These methods are usually utilized based on the availability of the mathematical model of system, which is mature for rotating mechanisms and vibrating mechanism systems, but it is hard to determine the degradation mathematical model for electronic systems or the complex multidisciplinary systems. Hence, the research and application of this type of methods are subject to limitations to some extent. The data-driven RUL method analyzes the sys-tem’s data sets, including the historical fault data, statistical data and real-time monitoring data, uses mathematical algorithms to extract the features of different faults and trends, then applies these algorithms to the real-time monitoring data sets to understand the inferences. The common data-driven methods include Artificial Neural Networks (ANN) based method[10],fuzzy system based method[11]and time series based method[12].These types of methods do not need the system’s mathematical model. The algorithms are built from the available data sets.For the development and application of sensors technology, it is easy to obtain the system’s operating data. So, the data-driven RUL method is friendlier and more feasible to start with. However, it also takes time and money to obtain the data that can reflect system’s inner condition directly, and the monitoring data sets are often of uncertainty and incompleteness in real cases,increasing the dif ficulty of developing prognostic algorithms.

3 FRAMEWORK AND PROCEDURE

The mode of maintenance for spacecraft is different from that of the ground system because of the harsh space environment, multiple working conditions, the constrained astronaut extravehicular activity time, device spare parts inventory etc.. Considering the safety concern and the limited resources in space, a predictive maintenance strategy is proposed to try to achieve an optimal result. First, establish the devices requirements for maintainability according to the system design, system Failure Modes and Effects Analysis (FMEA), Fault Tree Analysis (FTA) and reliability analysis. Then build up the devices RUL estimation methods respectively and find out the developing trend of health states.Thresholds are made based on ground reliability test data, which are used to determine whether devices need repair or not. If the device’s trending health state reveals that its performance will soon be unacceptable, a maintenance strategy then is made considering the available resources and the extent of urgency,that takes account of whether or not resources for repair are in place, and astronauts can engage the task at the appropriate time. If any needed resource is out of reach or the case is an emergency, the maintenance strategy should plan the supply of resources and adjust the work mode to mitigate the degradation process. This kind of maintenance strategy is an active preventive methodology, which can optimize the maintenance process and reduce the risk of failure. All of these are with one purpose of improving the reliability and safety. Figure 1 shows the frame-work of the proposed predictive maintenance scheme.

Figure 1 Predictive maintenance scheme framework for manned spacecraft

As for the complex multidisciplinary system, such as for manned spacecraft, it is quite dif ficult to establish the system’s mathematical model. When a manned spacecraft system runs in space, the con figured sensors collect the corresponding data and command & data handling subsystem forwards the data to the ground monitoring station. This telemetry data can reflect the health state of the system. So, it is more suitable to develop a data-driven scheme to establish the RUL. The key of datadriven RUL method is to extract features that can reflect the trending information, or some signs of faults. The extracted features are used as a health index (HI) of the system or device,and then a time series of health indices are formed by these HIs listed in time order. Using the time series prediction methods,such as linear/non-linear regression[13], the Auto-regressive Moving Average (ARMA) model[14], Arti ficial Neural Networks(ANN)[15], the developing trend of the system/device health state can be estimated and RUL calculated. Different devices have distinct characteristics, which brings about various degradation processes. The RUL estimation is closely related to the corresponding degradation process as it determines the threshold used to decide whether the performance is still accepted or not. In real cases, the system/device is affected by stochastic deterioration which usually accelerates the degradation process. In addition, there are some uncertainties, such as the inaccuracy of sensor data, noise, environment factors and degradation model itself, that need to be considered. Figure 2 shows the scheme of this data-driven method.

4 CASE STUDY

Figure 2 Procedures of data-driven RUL estimation method

The Chinese space station is an independently developed large-scale space laboratory program which will be launched into low Earth orbit and assembled in the near future. The con-struction of the station consists of three major modules, which are the Core Module named Tianhe, Experiment Module I named Wentian and Experiment Module II named Mengtian.The Shenzhou manned spacecraft will be used to carry astronauts to the station, and the Tianzhou cargo spacecraft will be served to resupply the station. Figure 3 is the diagram of the Chinese space station con figuration.

For this case study, we focus on the Electric Power System(EPS) of the station and explore the predictive maintenance mode. Each module of the station has its own EPS system, in which a pair of solar arrays serves as the primary power source,and batteries serve as the secondary power supply. For simpl ification, we have just taken the solar array and battery into consideration. When working in space, the station will experience sun period and eclipsed period repeatedly. On the assumption that degradation is a slow and continuous process with stochastic in fluences, the telemetry data can be processed segment by segment in one orbit length, and then feature extraction algorithms applied on each segment, as shown in Figure 4.

There will be voltage, current, temperature and other kinds of monitoring data from the solar array and battery, some common statistical characteristics in time domain can be used here, such as average amplitude, standard deviation and so on. Table 1 shows the details of the adopted feature extraction methods.

When the features are extracted, the similarity between each segment can be used as a health index. The threshold,where the health index crosses will then be taken as unacceptable, is ident ified by the ground reliability test data sets. As for the uncertainty consideration, stochastic process model can be used to make it up. Wiener process is suitable for degradation modeling in some cases, which can be represented asis the stochastic process,the de fined drift parameter, is a diffusion coef ficient, andis standard Brownian motion. These coef ficients should be identified using real on-orbit data.

Figure 3 Chinese space station configuration

Figure 4 The proposed telemetry data feature extraction method

Table 1 Details of the feature extraction methods

The EPS is vital to the spacecraft, by knowing when and where the maintenance should be done using this scheme, the safety of the spacecraft can be improved and costs of unexpected maintenance requirements can be prevented. Figure 5 shows the ideology of the predictive maintenance.

The steps for the proposed predictive maintenance are as follows:

Figure 5 The ideology of the predictive maintenance

1) Initialization.

2) Health Index extraction.

For each segment of telemetry data, the features in table 1 are calculated. The vector ofserves as the feature of the segment data, i.e.,is the feature ofpart. Then we can obtain the featuresfor the solar array andfor the battery. The health index is the similarity between each adjacent pair of features:

3) Remaining useful life estimation.

After all the health index is calculated, the RUL of the electric power system can be estimated through mapping the solar array’s HI and battery’s HI into the range of [0,1], where 0 means failure and 1 means completely new. We assume that the trend of RUL is a gradual reduction, so algorithms of linear/non-linear regression can be used. Here we add Wiener processto describe the uncertainty characteristic of degradation. The estimated RUL at time t can be expressed as:

3) Predictive maintenance policy establishment.

The predictive maintenance policy is made according to the RUL results. The most important thing is to guarantee the safety of the power system. On the basis of this prerequisite,the maintenance should achieve a trade-off between benefits for safety and costs.Letdenote the maintenance cost of solar array,denote the maintenance cost of battery,denote the remaining value of solar array, anddenote the remaining value of battery. When the power system finishes self-checking on orbit, theWhen the solar array or battery reaches its threshold of failure, theSo, theis proportion to the RUL. The predictive maintenance policy is the optimization of

5 CONCLUSIONS

In this paper, we have discussed the bene fits of predictive maintenance for manned spacecraft and show the developed methods. We use energy system on the Chinese space station as a conceptual demonstration to show how to apply the predictive maintenance ideology to the spacecraft, which helps to improve the reliability and safety. In the future, we will detail the methods and use on-orbit data to revise the models.