Search

import sys, os
sys.path.append(os.path.join("..",".."))
import numpy as np
import pandas as pd
from sentence_transformers import SentenceTransformer
from mika.utils import Data
from mika.ir import search

Prepare Data: NTSB

os.chdir('../')
os.chdir('../')
ntsb_filepath = os.path.join("data/NTSB/ntsb_full.csv")
ntsb_text_columns = ['narr_cause', 'narr_accf'] # narrative accident cause and narrative accident final
ntsb_document_id_col = 'ev_id'
ntsb_database_name = 'NTSB'
ntsb_data = Data()
ntsb_data.load(ntsb_filepath, preprocessed=False, id_col=ntsb_document_id_col, text_columns=ntsb_text_columns.copy(), name=ntsb_database_name, load_kwargs={'dtype':str})
ntsb_data.prepare_data(create_ids=False, combine_columns=ntsb_text_columns.copy(), remove_incomplete_rows=False)
# filter to years of interest
years_of_interest = ['2011', '2012', '2013', '2014', '2015', '2016', '2017', '2018', '2019', '2020', '2021']
ntsb_data.data_df = ntsb_data.data_df.loc[ntsb_data.data_df['ev_year'].isin(years_of_interest)].drop_duplicates(subset='ev_id', keep="last").reset_index(drop=True) # keep the last record, this one has the phase of flight and mishap right before the accident
ntsb_data._Data__update_ids()

Combining Columns…: 100%|██████████| 196535/196535 [00:02<00:00, 75855.79it/s]

data preparation:  0.04 minutes

Perform Search

# fine tuned model
model = SentenceTransformer("NASA-AIML/MIKA_Custom_IR")
#model = SentenceTransformer('all-distilroberta-v1') #uncomment to use pretrained model
ir_ntsb = search('narr_cause', ntsb_data, model)
embeddings_path = os.path.join('data', 'NTSB', 'ntsb_sentence_embeddings_finetune.npy')
ir_ntsb.get_sentence_embeddings(embeddings_path) # uncomment this if the embeddings do not yet exist
#ir_ntsb.load_sentence_embeddings(embeddings_path) # uncomment this if you wish to load sentence embeddings that already exist

queries = ['what components are vulnerable to fatigue crack', 'what are the consequences of a fuel leak', 'what are the risks of low visibility']
for query in queries:
    print(query)
    display(ir_ntsb.run_search(query, return_k=10, rank_k=10))

what components are vulnerable to fatigue crack

	top_hit_doc	top_hit_scores	top_hit_text
0	20121204X63622	0.699882	The total loss of engine power as a result of ...
1	20120703X63653	0.633137	An undetected fatigue crack on the outer surfa...
2	20180404X13226	0.603839	Extensive fatigue cracking in the left-wing ma...
3	20161027X90950	0.600091	A No. 1 engine fire caused by a fuel manifold ...
4	20110613X64905	0.599796	A quality control failure in the Sungear, Inc....
5	20140921X93622	0.598661	The fatigue fracture of the No. 3 exhaust valv...
6	20110315X35559	0.591568	The fatigue fracture and subsequent failure of...
7	20170607X22245	0.582699	The PW4056 engine lost power due to the fatigu...
8	20161216X91529	0.562310	A fatigue crack of the crankshaft at the No. 4...
9	20130524X83830	0.560072	Failure of the right inboard foreflap outboard...

what are the consequences of a fuel leak

	top_hit_doc	top_hit_scores	top_hit_text
0	20110729X01918	0.596457	The loss of engine power due to fuel contamina...
1	20140608X03928	0.595382	The loosening of a B-nut in the fuel system wh...
2	20211004104034	0.592871	A total loss of engine power due to fuel conta...
3	20110923X51306	0.592871	A total loss of engine power due to fuel conta...
4	20150421X33923	0.591596	The partial loss of engine power due to contam...
5	20110621X20741	0.585873	Fuel leaking from the fuel flow transmitter th...
6	20151005X24854	0.582271	The total loss of engine power due to contamin...
7	20150325X74504	0.582222	Loss of engine power due to fuel contamination...
8	20180918X41159	0.582222	Loss of engine power due to fuel contamination...
9	20200601X20239	0.575583	A total loss of engine power due to a fuel lea...

what are the risks of low visibility

	top_hit_doc	top_hit_scores	top_hit_text
0	20150110X84136	0.471297	The non-instrument rated pilot’s decision to d...
1	20180319X90457	0.445092	The pilot's failure to see and avoid power lin...
2	20170114X83817	0.441732	The pilot's failure to see and avoid power lin...
3	20210909103825	0.434227	The pilot's failure to maintain clearance from...
4	20110401X40531	0.431236	The pilot’s inadequate visual lookout and fail...
5	20131002X40410	0.431074	The pilot’s distracted attention and failure t...
6	20150903X31635	0.426771	The pilot’s inadequate visual lookout and fail...
7	20150514X61132	0.425973	The pilot's failure to see and avoid power lin...
8	20130517X02521	0.425482	The pilot's failure to maintain clearance from...
9	20140430X30514	0.425437	The pilots’ failure to maintain clearance from...