[BUGFIX] Fix stuck sensor detection only applying to last column

[misi9170]

Addresses issue raised in #175.

The stuck sensor indices were only being returned for the last column checked, as opposed to a union of the indices over all columns. This PR addresses that issue.

To do

• [X] Address fundamental issue and ensure that indices are returned for any stuck sensor (not just final column)
• [x] Check that examples still run as expected
• [x] Add test for this (can base on @DanielKerrigan 's example script)
• [x] to discuss: do we want to return a separate list of indices for each column provided, or simply a union of all indices that contain a stuck sensor fault in at least one column (the current expected behavior)? I decided to add an option to do this, defaulting to False, which when True will instead return a dictionary containing a numpy array of stuck sensor indices for each column passed.

@DanielKerrigan 's script below now produces the expected results

import pandas as pd
from flasc.turbine_analysis.find_sensor_faults import find_sensor_stuck_faults

df = pd.DataFrame({"a": [0, 0, 0, 1, 2, 3], "b": [4, 5, 6, 7, 7, 7]})

results_a = find_sensor_stuck_faults(df, columns=["a"], ti=0, plot_figures=False)
print("columns=['a']:", results_a)

results_b = find_sensor_stuck_faults(df, columns=["b"], ti=0, plot_figures=False)
print("columns=['b']:", results_b)

results_ab = find_sensor_stuck_faults(df, columns=["a", "b"], ti=0, plot_figures=False)
print("columns=['a', 'b']:", results_ab)

results_ba = find_sensor_stuck_faults(df, columns=["b", "a"], ti=0, plot_figures=False)
print("columns=['b', 'a']:", results_ba)

# Output:
# columns=['a']: [0 1 2]
# columns=['b']: [3 4 5]
# columns=['a', 'b']: [0 1 2 3 4 5]
# columns=['b', 'a']: [0 1 2 3 4 5]

[misi9170]

Two of the examples (that I'm aware of) use find_stuck_sensor_faults(), called within ws_pow_filtering.py's filter_by_sensor_stuck_faults():

• examples_artificial_data/01_raw_data_processing/00_filter_ws_power_curves.ipynb
• examples_smarteole/03_filter_ws_power_curves.ipynb In these, find_stuck_sensor_faults() is being called separately for each turbine in a for loop; however, the turbine's wd and ws columns are checked simultaneously, in that order, so the bug meant that only the stuck ws indices were being returned. After the bugfix, there is a slight change (increase) to the data removed by the stuck sensor filter for examples_artificial_data/01_raw_data_processing/00_filter_ws_power_curves.ipynb; and there is no (or minimal) change to the data removed by the stuck sensor filter in examples_smarteole/03_filter_ws_power_curves.ipynb (presumably, there were few/no instances of stuck wd readings). Out of caution, I have rerun any subsequent notebooks after these examples so that they are consistent with any changes to the data made.

[misi9170]

Thanks @paulf81 . I realized I hadn't pushed the changes to the examples (not to the running code, just to the notebook output), so I'll do that now and then merge.