Thank you very much for your question. Regarding the computational time, the used Tsfel library calculated the initial 1600 features within a few minutes on a standard modern laptop. Implementing it into the pipeline took time to get used to, since the documentation may not be as thorough as it could be.
Manual feature engineering on the other hand is more labor intensive since the interviews that we held with the process experts consisted of:
a) Determining important process values (e.g. temperature)
b) Selecting corresponding sensors
c) Individually analyzing each sensor’s time series data manually
d) Manually implementing the engineering of the features
I hope this gives you an idea of computational time and costs. Let me know if you have further questions.
Mr. Mende, thank you for the prensentation, It’s exciting to see results like these in a challenging field as the glass industry! How did you apply sensor fusion in feature engineering steps to pair each data because each sensor may not have same sampling rate?
Great presentation!
I have a question about the computational time and costs of implementing manual feature engineering rather than applying automatic methods.
Thank you! Congratulation again for your work!
Dear Federica,
Thank you very much for your question. Regarding the computational time, the used Tsfel library calculated the initial 1600 features within a few minutes on a standard modern laptop. Implementing it into the pipeline took time to get used to, since the documentation may not be as thorough as it could be.
Manual feature engineering on the other hand is more labor intensive since the interviews that we held with the process experts consisted of:
a) Determining important process values (e.g. temperature)
b) Selecting corresponding sensors
c) Individually analyzing each sensor’s time series data manually
d) Manually implementing the engineering of the features
I hope this gives you an idea of computational time and costs. Let me know if you have further questions.
Best regards,
Hendrik Mende
Mr. Mende, thank you for the prensentation, It’s exciting to see results like these in a challenging field as the glass industry! How did you apply sensor fusion in feature engineering steps to pair each data because each sensor may not have same sampling rate?