Application of Faceted Classification to InService Records

1
Application of Faceted Classification to
In-Service Records

• Yee Mey Goh, Matt Giess, Dave Stewart and Chris
  McMahon
• IDMME Virtual Concept 2008
• 8-10th October 2008
• Beijing, China

2
Motivation

• What can we learn from our product in service?
• Expected and unexpected performance, maintenance
  and operational requirements
• Type of issues (design/use/operation-induced,
  etc.)
• Service information as a resource for learning
• What information is being collected from service
  and maintenance phase?
• How can we improve our ability to learn from it?
• A faceted classification approach applied to
  organise in-service records

3
Enumerative Classification
Size/weight
Cost
Pumping material
Pumped principle
Delivery pressure
Power Source
Volume flow rate
Manufacturer
4
Faceted Classification

• Associate documents with many categories that may
  themselves be organised into hierarchies.

Vibration analysis of gear pumps.
ACME pumps.Centrifugal coolant pump.2 bar
5
Generation of Classification Schemes

• Warrant is the authority by which a
  classification scheme may be generated
• User, scientific, literary
• User warrant
• Supporting the end user
• Scientific warrant
• As identified by domain experts
• Literary warrant
• According to the contents of the underpinning
  document corpus
• Reference Giess, M. D., Wild, P. J. Mcmahon,
  C. A. (2008) The generation of faceted
  classification schemes for use in the
  organisation of engineering design documents.
  International Journal of Information Management,
  DOI 10.1016/j.ijinfomgt.2007.10.001.

6
In-Service Records

• In-service issues database
• Contextual information engineer, operator, date
  etc.
• Free-text description that might describe the
  product, type of queries, operational environment
• Common concepts not readily identifiable
• Retrieval is a combination of context keyword
  search
• Export records for a selected number of years
  from Access database
• A few thousands of records

7
XML Documents

• Script into structured XML documents
• Each represent an instance of in-service query
  raised

8
Faceted Classification Schemes

• Controlled fields readily transferrable to
  classification schemes
• Product variant (facet) X, Y, Z (concept nodes)
• Operator (facet) A, B, C (concept nodes)
• etc.
• Additional facets are derived from text
  description field (based on literary warrant)
• Assembly/component, part number
• Function
• Failure mode (facet) corrosion, crack, wear
  (concept nodes)
• Operational phase
• Type of usage
• etc.

9
Waypoint Software System

• Automated classification using constraints/rules
• Classification schemes described in concept map
• Constraints/rules expressed under each concept
  node in the classification hierarchy
• Can be distilled from a subset of documents
• IF document contains Term Landing Gear AND Term
  Bush OR Term product SN XXXXX THEN document
  is relevant to concept Landing Gear Bush

10
Concept Map

• Associating XML documents with concept nodes
• Terms search within tagged fields
• Numerical manipulation possible
• Document structure enhances semantic meaning to
  classifying documents

11
Waypoint Interface
Inherited facets from database
Selecting nodes trims the tree and returns
documents that meet all constraints
Additional facets generated from manual
inspection of description.
Note Only a fraction of randomly selected
documents are classified due to confidentiality.
12
Returned Documents
13
Discussions

• Faceted classification applied to organisation of
  in-service records
• Expedite the retrieval of similar past issues
• Allow for repetitive systemic issues to be
  highlighted
• Currently subjective
• Description most useful but contents vary by
  engineer, by case, etc.
• No consistent way of referring to product,
  failure mode, operational environment etc.
• Common concepts not readily identifiable

14
Conclusions/Further Work

• Automated/semi-automated construction of faceted
  classification schemes
• Text mining to identify key concepts and phrases
  (word clusters, concept pairs heuristics)
• Evolving taxonomy and prescribe control
  vocabulary
• Identification of common queries hence
  discovering knowledge
• By browsing the tree
• Statistical ranking and data mining

15
Thank you.

• Please contact Dr Yee Mey Goh
• for questions.
• y.m.goh_at_bath.ac.uk