Flight Data Analysis

Question 1

What is OFDM?

Answer 1

• OFDM (operational flight data monitoring) is the pro-active use of recorded flight data from routine operations in order to improve aviation safety.
• Airlines typically setup routines that trigger reports when safety relevant events occur
• In the industry, monitoring operational flight safety involves a lot of manual work:
  – get data from flights with events
  – visualize time series of that flight
  – discuss impact and relevance (also with pilots)
  – trigger SMS process to improve operations

Question 2

How do airlines define safety events in FDM?

Answer 2

Events consider the exceedance of parameters beyond Standard Operating Procedures (SOP)
- low/medium/high Limit

Question 3

What errors can you find in QAR data?

Answer 3

• outliners
• constant part
• empty
• wrapping around
• missing sign bit
• corrupt part

Question 4

How to deal with QAR errors?

Answer 4

outliners/corrupt/constant:
- remove values with jumps far away from rolling average
- remove values outside of physically feasable range
Empty:
- calculate using other parameters (e.g. vert. speed from flight path angle)
wrapping/missing sign:
- add discrete offset on large jumps to unroll the value
- force value to be within a given range (radio height at T/O)
combined approach
trajectory reconstruction

Question 5

What are difficulties when defining timepoints?

Answer 5

Robust Definitions
- Needs to work for data errors from many different flights and airlines
- Which threshold value to use?
- Minimize chance for erroneous detections
Decision Logics
- Taking into account different ways to detect a timepoint
- Find indications which one to select
- Quality criteria
Different available variables for different airlines or aircraft types
- Older B737 do not record vertical speed
- 1 or 2 radio altitude sensors / ILS deviations / other redundant sensors

You must do a trade off between automation and measurement quality

Question 6

How do you estimate unrecorded parameters?

Answer 6

• estimated by employing System Identification method (F8.19) along with variables measured by aircraft sensors
• input-model-output

Question 7

What are some distribution types?

Answer 7

F8.24
Weibull: General purpose reliability distribution for modeling material strength, times-to-failure of electronic and mechanical components, equipment or systems
Exponential: Commonly used for components or systems exhibiting a constant failure rate
Normal: Commonly used for general reliability analysis, time-to-failure of simple electronic and mechanical components, modeling uncertain system parameters, random errors

Question 8

How to select the most appropriate type?

distribution fitting

Answer 8

F8.26
- using specific goodness of fit test
- visually comparing the fittet and the empirical distribution
Negative Logarithm Likelihood (NLL), Bayesian Inference Criteria (BIC)…

Question 9

What is the most appropriate kind of visualization in different cases?

Answer 9

Two common types of data in QAR context
- Statistical Related Data
# Mean
# Standard deviation
# Distribution of a specific value of a parameter
-> Charts:Histogram or pie
- Time Series Data
# Parameter history e.g. air speed during flight, engine N1 during flight, etc.
-> Plots: 2-Dimensional plot
Animation: Animation in 2 or 3-Dimensional