The Important Technologies In Research Of Satellite Fault Diagnosis Based On Data-Driven

Since the Soviet union launched the first artificial satellite in human history, with the expansion of the satellite function, and increasingly profound effects on People’s Daily lives, the number of satellites in space become more and more. On the other hand, a large number of new technologies, new equipments and new materials, make the satellite system presents the large-scale complicated trend, it is bound to be arranged on the satellite a lot of sensors to monitor the running state of those components to ensure the normal operation of the satellites.The sensors are widely used in the satellite components in the process of satellite running, and the massive, multi-source heterogeneous data, such as the abnormal flight records, report, control, information, warning, and a lot of telemetry data are quickly accumulated. How to make full use of these huge amounts of data to improve the reliability and security of the satellite, and making big data analytics has also become the focus in the satellite fault diagnosis. Combined with the background of the fault diagnosis process,and the current big data of sensors, we have studied four aspects the key technology of satellite fault diagnosis based on the data driven.1) The rapid detection technology of satellite’s abnormality based on the similarity comparison. Satellite structure is complex, high cost, expensive, and to ensure early detection abnormality as soon as possible, it is needed to study a rapid and accurate anomaly detection technology. Telemetry parameters of satellite data is mostly a typical time series model, almost all of the time series algorithm involves calculation sequence similarity, and satellite remote sensing data similarity comparison is one of the commonly used methods for anomaly detection. But this method is currently faced with how to find the anomalies quickly and accurately in the millions of data points of the satellite telemetry data. We proposed a parallel implementation mechanism of time series similarity comparison, the Map Reduce computing framework is used to improve the classic dynamic time warping(DTW) algorithm, and to improve the efficiency of satellite time series similarity search.2) The satellite time-series data discretization technique for knowledge discovery.The data-driven technique is a powerful tool which can extract useful information from a large complex database system. But its performance is largely dependent on the quality of the data, so the data preprocessing is a crucial step in the area of data driven. Satellite remote sensing data is usually continuous values, and many data mining algorithms can only deal with discrete data, therefore, before the satellite data mining, the continuous numerical attributes must be discretization, and translated them into a amount of discrete symbols. Based on the trend features of satellite time series data, we put forward the symbolization method of oriented knowledge discovery, which is used to mining the diagnosis knowledge from the telemetry data. The method can be used to improve the readability of the symbolic results, and also it is advantageous to the analysis of data mining in the late work.3) The satellite fault diagnosis rules acquisition technology based on the parallelization of Apriori algorithm. In order to realize effective health management of satellite,establishing and perfecting the satellite fault diagnosis knowledge(rules) library is very necessary. In the field of satellite fault diagnosis, the existing rules for diagnosis mostly come from domain expert summed up according to the experience or formula, but the failure mode of these rules cover is limited. The fault diagnosis for the inexperienced is powerless, and as the related satellite components changed, the rules must be changed or even rewrited. So the practical value of rules will be reduced. Existing rule acquisition techniques in the face of massive satellite telemetry parameter data, its performance is also a factor which has to be considered. We have improved the classic Apriori association rules algorithm, combining with the Hadoop cluster computing mode and the parallel algorithm, and giving the relevant fault set higher weights, and it can dig out the fault related rules more effectively.4) The satellite fault identification technology based on hybrid voting mechanism of support vector machine(SVM). Satellite is a large complex system, to accurately diagnose the fault category, the three serious problems are needed to considered. First, the parameters of the satellite are as many as thousands, but every failure can’t be related with all the parameters. So how to extract the characteristics of each fault corresponding parameters is the precondition of the satellite fault identification; Second, for so many satellite components, the corresponding fault categories number also are very big. How to determine the current exception corresponding fault from a large number of categories, traditional classification algorithms are facing a big challenge; Third, comparing to the normal data, the satellite fault data are very little. The existing classification methods requirement learning samples not only more but also high quality, but satellite system’s actual fault samples are difficult to meet this requirement. Due to the existence of a large number of parameters and fault types in satellite fault diagnosis, and the characteristic of less fault samples, we proposed a fault diagnosis technique based on mixed voting mechanism of support vector machine, to improve the accuracy of satellite fault identification.In summary, based on the background of massive telemetry parameters data constantly emerging, aiming at the present stage of the satellite reliability, security, and the challenge of stability requirement become higher and higher, we studied on the key technologies of satellite fault diagnosis for big data analysis. For the satellite anomalies found in time, establishing and improving the fault diagnosis knowledge library and accurately identify the satellite faults to ensure normal operation of the satellite, the research has important theory meaning and application value.