Aidan P. LaBella

Predictive maintenance systems have the potential to significantly reduce costs for maintaining aircraft fleets as well as provide improved safety by detecting maintenance issues before they come severe. However, the development of such systems has been limited due to a lack of publicly labeled multivariate time series (MTS) sensor data. MTS classification has advanced greatly over the past decade, but there is a lack of sufficiently challenging benchmarks for new methods. This work introduces the NGAFID Maintenance Classification (NGAFID-MC) dataset as a novel benchmark in terms of difficulty, number of samples, and sequence length. NGAFID-MC consists of over 7,500 labeled flights, representing over 11,500 hours of per second flight data recorder readings of 23 sensor parameters. Using this benchmark, we demonstrate that Recurrent Neural Network (RNN) methods are not well suited for capturing temporally distant relationships and propose a new architecture called Convolutional Multiheaded Self Attention (Conv-MHSA) that achieves greater classification performance at greater computational efficiency. We also demonstrate that image inspired augmentations of cutout, mixup, and cutmix, can be used to reduce overfitting and improve generalization in MTS classification. Our best trained models have been incorporated back into the NGAFID to allow users to potentially detect flights that require maintenance as well as provide feedback to further expand and refine the NGAFID-MC dataset.

This work presents a redesign of the National General Aviation Flight Information Database (NGAFID) to allow swift ingestion of flight data and calculation of flight safety events, as it is rapidly growing and currently housing over 750,000 hours of flight data. This redesign reduced memory and storage usage by a factor of 5, as well as reducing flight data import and event calculation time from over 3 minutes to 0.3 seconds by using a per-parameter compressed data representation in the database and a new pipelined data ingestion strategy. In addition to this, four new advanced flight safety event calculations were developed for proximity, self-defined glide path deviation, stall and loss of control in flight (LOC-I). A time and location bounded flight matching strategy was used to calculate proximity events for flights in on average 3.2 seconds per flight in the full 660,000 flight database. The stall and LOC-I event calculations show strong improvements upon prior work in the area on both a test flight flown for accurate stall time measurements and a historical sample of data independently annotated by two subject matter experts. The enhancements to the stall and LOC-I event calculations resulted in an increase in stall event detection from 23.1% to 83.1% in the historical data while reducing false negatives by a factor of 3. As the leading cause of aviation accidents worldwide is LOC-I, it is an especially important issue for flight safety monitoring programs, especially as other existing flight data monitoring (FDM) software lacks these new stall, LOC-I and proximity event detection capabilities.