Research On Aircraft Engine Fault Detection Algorithm Based On Multivariate Time-series Data

The flight safety of aircraft directly affects the economic of airlines. How to effectively monitor the flight parameters and quickly make failure detection for aircraft have been one important issue for airlines now. At present, because many domestic airlines use Quick Acces s Recorder to record various flight parameters during the flight, so it is more and more import ant to determine whether the aircraft engine has broken down by analyzing the QAR data. Thi s paper mainly study the failure detection of aircraft engine. For the problems that the large da ta volume and complex structure of aircraft engine and different QAR data volume of each fli ght, this paper propose an adaptive segmentation algorithm based on the important points. The algorithm firstly automatically generate adaptive number of important points for data with di fferent lengths to split the QAR data and realize data compression. Then use the time series an omaly pattern detection algorithm based on local density to do anomaly detection for aircraft engine data and measure the similarity between the abnormal sequences detected and standard fault model, which can determine the aircraft fault. In this paper, the main contents are as foll ows:1. Used the time series anomaly detection algorithm to detect aircraft engine fault. In order to detect the aircraft engine fault combining the characteristics of QAR data,the first step is to preprocess QAR data. It mainly divided into two aspects:(1)Smoothing the noise data.(2) Extracting cruise phase data from QAR data.2. Segmented QAR data to achieve data compression. In view of the situation that the lengths of QAR data caused by different flight time is not equal, this paper propose an adaptive segmentation algorithm based on the important points. Comparing with the traditional segmentation algorithm based on the important points, the improved algorithm can automatically generate appropriate number important points to segment the QAR data and model the segmented subsequence by the average value and length of subsequence.3. In order to detect abnormal data of QAR data, this paper use the time series anomaly pattern detection algorithm based on local density to make an anomaly detection for the aircraft engine data. The method can reduce the computational complexity and find the abnormal Sub-sequence accurately, Then extracted the main features of abnormal sequence, Finally, Measured the similarity between known standard fault model and detected abnormal sequences to determine the aircraft engine fault.