MIKA: Manager for Intelligent Knowledge
Access Toolkit for Engineering Knowledge
Discovery and Information Retrieval
Sequoia Andrade
1,2
and Dr. Hannah Walsh
2
1
HX5,
2
NASA Ames Research Center
• Repositories of narrative reports contain a wealth of information that can provide
insight during system development; however, they are often unstructured,
inconsistently filled out, or difficult to analyze at scale
• State-of-the-art natural language processing techniques that allow fine tuning on
domain-specific documents show promise in improving how these repositories
can be leveraged
• Packaging these techniques as a user-assistive tool enables efficient integration
into systems engineering practice at multiple junctions, demonstrated in Figure 1
• The Manager for Intelligent Knowledge Access (MIKA) toolkit provides
knowledge discovery and information retrieval capabilities for analyzing large
repositories of natural-language text
• The MIKA toolkit structure is shown in Figure 2 and includes three sub-
modules: utilities, knowledge discovery, and information retrieval
• Utilities: includes a data class that supports analysis in all MIKA capabilities, so
there is no extra setup to run different NLP techniques
• Knowledge discovery: for exploratory studies, understanding trends and
themes, and extracting hazards
• Includes: topic modeling, trend analysis, failure modes and effects analysis
(FMEA), Named entity recognition
• Information retrieval: extracting detailed descriptions of specific instances of a
hazard, trend, or theme via semantic search
PROBLEM & OBJECTIVE
CASE STUDY
Result #
Document
ID
Similarity
Relevant
Text Excerpt
Query
1: what components are vulnerable to fatigue crack
1
20121204
X63622
0
.746134
…
one of the first-stage compressor blades
had
fractured
due to fatigue cracking…
2
20141027
X62323
0
.728395
…
the crankshaft had fractured due to a fatigue
crack
that
had initiated at a fillet radius…
3
20160114
X73526
0
.715234
…
forward wing spar likely fractured due
to
compression
loading from wing loads…
(a) (b)
Figure 3: Average severity with standard deviations in (a) and frequency over time in (b)
Figure 2: MIKA toolkit architecture
Phase
Cause
Control Process
Recommendation
Bird strike
Maneuvering; Taxi
-from
Runway; Takeoff; Landing;
Other
at low, bird, surface, a, red
-
tailed, hawk, female
- flight, collision, with a, bird,
was, towed to the gate, a,
bird ingestion and
containment
see the public, enter, keep an, eye,
on, land
Midair collision
Maneuvering; Taxi
-from
Runway; Takeoff; Landing;
Approach; Emergency Descent
failure of both pilots to see
and avoid the other
-,
wingtip, mounted strobe
anticollision, lights, graphic,
remote communication
alter, course, pass well, not have
passed over, under
Figure 4: Risk matrix of selected hazards generated via MIKA trend analysis
• Applied MIKA toolkit to National Transportation Safety Board (NTSB) aviation
accident report dataset from 2011-2021 (16,914 documents)
• Knowledge Discovery:
• Topic modeling can create failure taxonomies from unstructured data, such as in
Table 3
• Figure 3a identifies loss of control in flight and midair collisions as most severe
• Figure 3b identifies loss of control on the ground and inflight as most frequent
• The risk matrix in Figure 4 identifies bird strikes, midair collisions, loss of control,
turbulence encounters, and ground collisions as high risk according to FAA
definitions
• MIKA-generated FMEA in Table 1 successfully extracts causes, failure modes,
effects, control processes, and recommendations from the set of documents
referring to the specified failure using named-entity recognition. In the first row, a
red tail hawk causes an in-flight collision with a bird resulting in substantial
damage. In the second row, both pilots failed to see and avoid each other
causing a midair collision resulting in impact to terrain.
• Information Retrieval:
• Example query and document retrieval results in Table 2 identify compressor
blades, the crankshaft, and the wing spar as components vulnerable to fatigue
crack.
Figure 1: MIKA usage in systems engineering process
MIKA TOOLKIT METHODS
Sequoia Andrade
Research Engineer
Sequoia.R.Andrade@nasa.gov
Dr. Hannah Walsh
Computer Engineer
Hannah.S.Walsh@nasa.gov
CONTACTS
Cause Narrative
Accident Narrative
# of Reports
fuel, oil, starvation, fuel starvation,
power fuel starvation, engine, loss
engine, loss engine power, engine
power, power
oil, engine, connecting, rod, connecting rod,
crankshaft, bearing, cylinder, revealed,
examination
33
fuel, tank, engine, fuel tank, power, tanks, pilot,
gallons, airplane, flight
62
CONCLUSIONS
• MIKA is intended to improve an organization’s ability to leverage knowledge
stored in documentation from past projects and accidents to reducing the space
of unanticipated hazards and problems.
• MIKA is designed to be integrated into the systems engineering process flexibly
at multiple junctions such that unanticipated problems can be most effectively
reduced through enhanced knowledge management.
Table 1: Subset of FMEA generated via MIKA named-entity recognition
Table 2: Example query and documents retrieved using semantic search
Table 3: Subset of taxonomy generated with MIKA topic modeling