QAR Data Outlier Detection And Fault Location Algorithm Research

At present, the airlines at home and abroad are all using QAR(Quick Access Recorder) data for the monitoring of the aircraft when they are in the flight state. Because of the continuous increase in flights and the requirements running on schedule of aircraft at the same time, based on the QAR data analysis to find out the hidden dangers that may exist is becoming more and more important. To achieve the purpose of outlier detection for aircraft temporal QAR data to find the fault whether exist or not, combined with the normal trend of main attributes of aircraft in flight, and carries on the analysis in view of QAR data, we mainly study to find abnormal data in aircraft during flight through the way of outlier detection and fault location, and we can detect with the main attributes of fault occurs, then use the algorithm to find the outliers, which could effectively help the aircraft maintenance personnel to detect and adjust of aircraft components. The study mainly completed the following work:1) Research and implement on QAR data outlier detection algorithm and locate the fault. At the first stage of the algorithm, the QAR data in flight are classified according to the data flow partition, and through cluster analysis of data blocks in the data, the data is replaced with mean reference point after clustering; in the second stage, according to mean reference point,we use the fitting find trends of properties during the flight, and according to the each mean reference point to the fitting curve or surface distance determines whether it is outliers, all data represented in the cluster are considered as outliers, and find the occurrence time according to the outlier serial number. These data can be determined whether it's fault by the engineers according to experience or fault model. So these data could provide reliable data for aircraft maintenance.2) In order to reduce the influence of the initial clustering centers on the final effect, the traditional K-Means clustering algorithm is improved to cluster the QAR data. After dividing the data according to the time sequence, calculate weight of attribute in accordance with the changes in each attribute range, and finally through the weight of each attribute calculate the weight of data, then sort the weight to select the closest initial cluster center, which optimizes the clustering effect of algorithm, and reduces the number of iterations of clustering.3) Using the least square method to fit the mean reference point. At the same time, according to the definition of outlier detection based on distance, we can determine whether the mean reference points is outlier or not, and we put forward the definition of outlier factor, which is suitable for QAR data outlier detection, and according to the outlier factor we define outlier data. We consider the points which exceed the setting threshold as outliers, and the approximate time failure occurred can be find at the same time.