[jhnn]

Dear PHM North America 2026 Data Challenge Organizing Team,

I am currently working with the dataset for the PHM 2026 North America Challenge and wanted to bring a potential data anomaly to the community’s and organizers’ attention.

While analyzing the extended low-frequency data for Gear A, I noticed that the very first file in Run 1 is exactly identical to the first file in Run 2.

Given this specific overlap, I strongly suspect that there might be further duplicated files hidden elsewhere in the dataset, perhaps across other runs or even different gears.

Could the organizers clarify if this is an intentional part of the dataset (for example, representing a shared baseline, calibration, or initial state) or if it might be an accidental duplication during the data export? Also, has any other team noticed similar overlaps in different parts of the provided data?

I just wanted to flag this early on, as these duplicates could potentially impact feature extraction and model training for all of us. Looking forward to your feedback!

Best regards,
ReichWiens

[thejbills]

Hi team ReichWeins,

We’ll check that out and get back to you.

[thejbills]

Team RechWiens,

Just looked at all of the data in detail, and you have indeed found that there are redundant files. We failed to notice that before. The files are redundant in the folder, but the file names are still correct.

Exp-A_HDF5_Run-1/Run1 has data for run 1 and run 2
Exp-A_HDF5_Run-2/Run2 also has data for run 1 and run 2

Please use the file names to determine which run they are associated with. This might be pervasive across the dataset.

• This reply was modified 1 week, 6 days ago by thejbills.

[jhnn]

Gear A should only have 5 runs, because there are only photos up to run 5. According to the sensor data, the extended low-frequency data goes up to run 9 (see the “run 5” folder).

[thejbills]

We’re going to get on a call with the data owner to discuss this data inconsistency.

[jhnn]

Thank you.

[Author]

Posts