Research On QAR Data Mining Algorithm Based On Gray System Theory

The reliability of the aircraft flight directly affects the airline's safety and economic benefits of the aircraft, how to monitor the aircraft flight parameters and aircraft systems for rapid fault diagnosis is the main issue of the airlines. Before the aircraft system failure, by monitoring the corresponding flight data parameters and analyzing to determine whether there is a hidden fault in the system, and thus provide the correct information for aircraft maintenance, so that quickly and accurately locate and exclude fault when a fault occurs,reduce accidents rate.Quick access recorder(Quick Access Recorder, QAR) records the stream data of the flight status parameters. Namely QAR data belong to the high complexity of time-series data.It seriously affects the efficiency if it is directly handled, so as to reduce the complexity of data, efficiently retrieve data and determine whether the QAR data is failure data. In this paper, time series data segmentation algorithm is adopted, QAR data as an object of study for its data mining. By dividing the QAR data into non-overlapping sub-sequences to reduce the amount of data, and locate the fault data series. Extracting useful rules for people decisions and then using gray system theory approach for the fault data sequence to determine the specific type of fault. This paper completes the following contents:1. Given the high dimensionality and big data volume of the QAR data, and the time of the aircraft sudden fault generally is short, the QAR recorder records less fault data, while the relationship between the values of each attribute is fuzzy. Then introducing time series fuzzy segmentation to process QAR data and dividing it into non-overlapping sub-sequences. The selection of the segmentation number becomes key issues. By introducing heuristic bottom-up algorithm and PCA similarity criterion automatically select the optimal number of segmentation, so as to detect QAR abnormal sequence.2. The grey process of QAR fault data. Gray generation processing sub fault data sequence, to obtain a consistent dimension. Combining gray system theory has better applicability to small sample characteristics. Thus gray correlation degree is calculated between the sequence to be detected and standard sequence.3. Improved gray correlation degree for engine fault diagnosis. In view of the engine each property parameters for the contribution of engine fault is different, through the use of improved gray correlation method can improve the fault recognition rate. Proposing changes to gray correlation coefficient is calculated as weight of gray correlation degree, which is weighted calculation. Finally, the fault pattern recognition and the determination of thediagnosis results are presented according to the principle of maximum correlation degree.