Certification by Analysis Uncertainty Quantification

Hi everyone,

The Boeing team has been examining the aero pressure reference time histories and has found large differences in pressure spread between sensor locations, making it difficult to fit a spatial surrogate for model-form uncertainty in the UQ4CbA challenge.

For example, the A1 comparison below shows substantial differences in 95% PI width between sensors at symmetric locations. The standard deviation maps in the attached PDF also show large differences between neighbouring sensors.

Removing each record's time-averaged mean also reveals repeated pressure histories. In Case A1, ports 7 and 37 share the same centred history, labelled "H49" in our analysis. Ports 16, 40, 49 and 51 share another, labelled "H08". The histories overlap within each group, and "H49" has substantially larger pressure variations than "H08".

Attachment(s)

pressure-summary_boeing.pdf   10.17 MB 1 version

Hi Alwin,

Regarding the AeroReference Data, the KHI team also found relatively large variations in the pressure time histories at several sensor locations. The attached PDF shows the time histories for all sensors, as well as the distributions of the time-averaged pressure and standard deviation.

Please find our feedback on your questions below:

• Are others observing the same variability and repeated histories? Is this reuse intentional, and how are source histories and amplitudes assigned to sensors and flight conditions?
  
  --> From the time histories, we can observe that the same variability pattern is applied to several sensor locations. I'm not sure whether it is intentional or not. 
  
  
• How are other teams interpreting this variation when estimating model-form uncertainty? Are spatial surrogates proving useful, or are bounds across sensor locations being used?
  
  --> Although the provided data were generated artificially, actual flight test data may exhibit unintended pressure variations due to sensor noise or other factors. Therefore, data quality screening may be necessary before assessing the model-form error. We felt that evaluating the model-form error based on the pressure at individual sensor locations would be difficult. Therefore, our team is considering evaluating the error in terms of integrated quantities, such as forces, rather than the pressure itself. However, this approach also has its own limitations and assumptions, and it remains unclear whether it is the most appropriate way to assess the error.
  
  
• How are those estimates being transferred to locations without reference data and to other flight conditions?
  
  --> Our team is currently considering how to estimate and extrapolate the error to locations without reference data and to other flight conditions, including how to handle such unintended variations, for example by using mean values only or moving-averaged values and excluding outliers.

Best regards,

Hidemasa

Attachment(s)

26KT019360_Review-of-AeroReference-Data.pdf   1.86 MB 1 version

Hi Alwin,

Thank you for reporting your finding. The noise I added onto the data is from actual flight test data I extracted from a project I worked on. It includes various factors that could cause degradation in sensor accuracy and reliability. The purpose was to replicate raw or minimally processed flight test data so I did not remove anomalies that I found. 

Please let me know if you would like more details.

Thank you.

-Seung (Paul) Yoo