Exploratory Data Analysis Approaches to Reliability: Some New Directions

1
Exploratory Data Analysis Approaches to
Reliability Some New Directions

• Chris McCollin
• Cornel Bunea
• Maria Ramalhoto

2
Chris McCollinThe Nottingham Trent University

• Involved in Reliability since 1976
• Worked as a reliability engineer for 3 major
  aerospace companies
• Consultancies/Training Nuclear, Rail, Commercial
• Involved with QRE paper for Engineering Council
  for last 8 years
• RSS representative for BSI
• ENBIS Reliability Website coordinator

3
Areas of Common SME Problems

• Effect of short-term management outlook on
  reliability
• Lack of time, manpower for analysis and
  improvement
• Lack of expertise, resources
• OEM dependency, meeting requirements only
• Lack of knowledge retention

4
FMEA/FRACAS Comparison
5
Problem Solving Requirements

• Structured approach, easy to use, computer/web
  based
• Developing hypotheses to answer (inter-related)
  problems over life cycle of a product(s)
• Using past/present information across diverse
  databases
• Central storage access on-line

6
Flowchart of Problem Solving Database
7
Problem Solving Procedure

• 1. Layout of Scenario
• Problem environment, conditions, flowcharts, etc
• 2. State Null Hypothesis
• Consider Problem effect across all interfacing
  levels leading to possible (multiple) causes
  (flowchart). Address complexity of problem,
  whether there is more than one. Alternative
  hypotheses listed. Costing issues addressed.

8

• 3. Analysis Flowchart
• Failure records, statistical flowcharts
  alternative research methodologies identified
• 4. Risks
• Previous works, arguments, risk assessments
  available?
• 5. Problem solving tools, Working model
• Physical assumptions, background theory design
  equations, physics, use of C and E (appropriate
  method), 5 Whys, brainstorming, Fault tree, FMEA,
  MORT, Design of Experiments, etc

9

• 6. Bias, Rejection criteria for null hypothesis
• 7. Hard Collection, analysis
• Analysis Questionnaires, Engineering analysis
  e.g. materials test, statistical analysis, etc

10

• 8.Conclusions Accept/reject hypothesis based on
  model/assumption/bias or change hypothesis (go to
  step 2)
• 9. Recommendations for corrective action Change
  to schedules, procedures, Poka-Yoke devices, etc.
  Standardisation.
• 10. Feedback/Feed forward
• To next problem, to database for dissemination
  and comment

11
Job Description of Facilitator

• Aids the problem solving activity
• knowledge and experience of the problem solving
  approach, team dynamics
• knowledge of what expertise is required for a
  particular problem and who can provide it
  (available from personnel files)
• has the ability to aid incorporation of diverse
  knowledge
• can mediate in issues arising from differing
  viewpoints
• suggest methods of solution (qualitative and/or
  quantitative)
• provide guidance of the holistic view of the
  company strategic plan.

12
No Fault Found (NFF)

• Reason not been installed on the aircraft and
  since the classification Missing did not exist
  in the failure definitions inventory (because
  Missing was not a failure category) the nearest
  most appropriate category was NFF. In this case,
  NFF is a misleading classification because it may
  indicate that a failure did not exist in the
  first place.
• We should stratify the problem by disseminating
  our data into more appropriate categories and
  discuss them individually.

13
No Fault Found
Plenty of Reasons No classification for what has
been found Replace everything (saves
time) Interdependencies between systems, e.g.
common power supplies Loading Working at limits
of operation Intermittent Wiring faults Ground
test conditions cannot reproduce latent defect
14
Example Hypothesis

• Aircraft operating, external temperatures and
  vibration affecting systems
• Time lag of thermal shocks 10º a minute in
  chamber but system takes longer
• Rise in temperature causes expansions effects
  on interconnections (transistors pnp,npn
  solder DC wetting may create micro-cracks
• (DC wetting is passing of DC current over dry
  joint creates an increase in heat, resulting in
  the joint melting back together) cannot locate
  fault

15
Continued

• Possibly surfaces become more elastic, cracks
  open quicker over time allowing contamination
• Cracks will close again, only long term exposure
  to adverse conditions may produce identifiable
  failure
• Road Surface testing, DOE (long term effects),
  FEA/Thermal effects, compatible materials, HALT,
  Simulation within CAD of thermal/vibration
  effects

16
Step 1.

• The environment, the operating conditions and
  the problem and associated inter-relationships
  should be outlined in sketch form (e.g. an
  Affinity diagram) to highlight areas where a
  possible solution may lie. Flowcharts, diagrams,
  previous analyses should be made available
  (preferably on-line).

17
Step 3. Structure

•  Approaches to identifying structure can be split
  into two separate areas where extra explanatory
  information is available and where it is not.

18
Multivariate Data Analysis Flowchart

• Description of physical and functional system
• Check for missing or corrupt data
• Discriminant analysis
• Multivariate analyses for determining structure -
  PCA, correspondence, cluster, correlation,
  distance measures, etc
• EDA
• Modelling time metric data - time series, PHM,
  PIM, GLIM, regression

19
Data Analysis

• Hypothesis 1 The stratum of a number of sockets
  is homogeneous. The alternatives are that times
  are clustered (non-independence) and/or
  inhomogeneous
• Hypothesis 2 The processes are independent
  against clustering (process identified as
  colored)
• Hypothesis 3 The colored process is stationary
• Hypothesis 4 The process is color blind
  competing risk
• Hypothesis 5 The process is stationary competing
  risk
• Hypothesis 6 The process is renewal competing
  risk
• Hypothesis 7 The process is Poisson competing
  risk and under the alternative hypothesis, H1
  Renewal process.

20
(No Transcript)
21
Step 5

• A repository of tools should be kept with
  examples of how they may be used in conjunction
  with each other. The repository may contain
  examples of the 7 quality tools, the 7 new
  quality tools, brainstorming, Management
  Oversight and Risk Tree (MORT), Failure Modes and
  Effects Analysis (FMEA), radar charts, etc.

22
(No Transcript)
23

• The Pro-Enbis project is supported by funding
  under the European Commission's Fifth Framework
  'Growth' Programme via the Thematic Network
  "Pro-ENBIS" contract reference
  G6RT-CT-2001-05059.
• The authors (i.e., Pro-ENBIS) are solely
  responsible for the content and it does not
  represent the opinion of the Community, the
  Community is not responsible for any use that
  might be made of data therein