HUMS ARCHITECTURE DESIGN

1
HUMS ARCHITECTURE DESIGN
TECHNIQUES FOR DESIGN, DESIGN AUTOMATION
EVALUATION
Advisor Dr.Ravi Mukkamala Student Parthiban
Sundaram
This work was supported in part by NASA LaRC,
VA.
2
Presentation Outline

• Introduction
• Design Issues Challenges
• HUMS Design Automation
• Conclusions

3
MAP

• Introduction
• Design Issues Challenges
• HUMS Design Automation
• Conclusions

4
Design Automation

• Design Automation is a natural step and is the
  last stage in the evolution of any system
• Successful codification of the best practices
  helps
• to avoid costly mistakes
• to automate problem-solving
• HUMS domains are well-established and several
  implemented systems have been in existence for a
  long period
• Some advantages of automation
• Minimization of cost,
• Productivity improvement,
• Production of optimal innovative designs, etc.

5
Design Automation

• This work attempts to capitalize on the knowledge
  developed from years of experience in
  engineering, system design and operation of the
  HUMS systems in order to economically produce the
  most optimal and domain-specific designs.
• Two specific objectives of the current work
• To produce the most optimal design for a specific
  domain economically
• To allow reconfiguration of system components at
  pre-installation stage as well as
  post-installation stage

6
Automation Process

• The automation process involves selection of
  solutions from a large design space as well as
  pure synthesis of designs
• The process is not an absolute AI approach,
  however it uses a knowledge-based system that
  epitomizes a specific HUMS domain
• The environment variables, design parameters,
  assumptions behind design techniques may be made
  to signify the domain of interest

7
Users Involved

• Participants involved in the process
• End-Users,
• Domain Experts,
• Requirements Analyst,
• System Developers
• System developers use the design automation system

8
Basic Design Approaches

• Bottom-up Approach
• Less number of searches
• How to make system-level decisions at
  component-level?
• Top-down Approach
• More number of searches
• An extensive design database required
• Hybrid Approach

9
MAP

• Introduction
• Design Issues Challenges
• HUMS Design Automation
• Conclusions

10
Capturing Requirements Data

• All details (quantitative and qualitative)
  pertinent for the design process must be captured
• Must support a format that allows qualitative
  analyses also
• Data and units must be generic enough to allow
  selection from among component instances of
  different types
• Whenever data is not accurate, the associated
  degree of uncertainty must be specified

11
System Decomposition

• Considering entire system designs increases
  complexity
• To reduce complexity, the target system is
  decomposed into several manageable layers
• System decomposition must
• Enable quick and easy selection of best
  components
• Ensure the attainment of quality attributes

12
Design Process

• Compare the requirements data against the
  parameter values
• Simple functions (Direct comparisons)
• Complex functions (Logical Expressions)
• Perform Cost-benefit analysis
• Perform Trade-off analyses
• Use a criteria list or design guidelines in the
  selection of components
• Search the designs database for a fitting solution

13
Design Process (Contd.)

• Use Engineering/formal techniques or Quantitative
  Design techniques
• Use Qualitative Design techniques
• Selection of designs based on future usage
  profile
• Identify and Use heuristics to solve complex
  design problems

14
Design Evaluation

• Design must be evaluated at every stage for the
  required quality attributes
• Techniques
• Using evaluation metrics at
• Component level
• System level
• Building the pareto-optimal set
• Performing heuristic-based analysis

15
Design Specification

• Architecture is explained in top-down
• fashion
• High-level system description
• Low-level system description
• Characteristics of auto-designer output
• Design justifications are provided at each stage
• Graphical as well as textual forms

16
MAP

• Introduction
• Design Issues Challenges
• HUMS Design Automation
• Conclusions

17
System Model
Design Specification
Component library
Evaluator
Design Explorer
Requirements
Design Space
18
Design Technique

• Quality attributes are often in conflict with
  each other and hence trade-off analyses are
  pertinent in design
• Designers depend on several design guidelines or
  principles to guide their design work
• The design explorer is provided with a set of
  design classes called base models that guarantee
  a selected few (usually one) quality attribute
• These base models tend to skew the final design
  and hence the explorer must merge different
  models to develop a suitable design

19
Flat Model (Performance Model)
TL
 
Ni
N2
N3
N1

FORMULAS LT TBL TTL NBLTTR A (1-PBL)
NBL (1-PNT) NMM(1-PTL) S (CAP NBL)/NBL
20
Cluster Model (Performance Model)
 
Reduction of processing time
21
Pyramid Model (Scalability Model)
 
FORMULAS LT TTL TBL TIN (NIN
NBL)TTR A (1-PBL) NBL(1-PTL) (1-PNT)
NMM(1-PIN) NIN S (CAPIII-NBL)/NBL
22
Gossip Model (Availability Model)
 
FORMULAS A (1-PBL NRM) NBL (1-PNT) NMM
(1-PTL NRM) LT TBL TTL NBLTTR
NRMTGS S (CAP NBL)/NBL
23
Merging Designs

• If there are n base models, then Explorer has two
  approaches
• Try all n! combinations possible till a
  satisfactory design results
• Merge the models in any order that minimizes the
  number of combinations
• Since merger of different designs is not
  straight-forward, we provide another set of
  specifications called merge specifications

24
Merging Designs (Contd.)

• These merge specifications will enable the
  explorer to combine any combination of base
  models, thus leading to
• nC2 nC3 nC4 nCn-1 nCn designs
• The merge specifications must consider the impact
  of different quality attributes that conflict
  with one another. E.g.
• Increasing number of levels gives better
  scalability but poorer performance
• Increasing number of replica managers increases
  availability but decreases performance, etc.

25
Merging Flat Model Pyramid Model
TL
Ni
N2
N3
N1

26
Merging Flat Model Pyramid Model
FORMULAS Levels - 1, LL 1 TM
Nodes at last level IN
-1TM For III 1 to IN do NIN III
TM LT TTL TBL TIN
(NIN NBL)TTR A (1-PBL) NBL(1-PTL)
(1-PNT) NMM(1-PIN) NIN S
(CAPIII-NBL)/NBL Next III
27
Trade-off Analyses

• Every iteration of merging of base models results
  in structural changes
• These structural changes impact the several
  quality attributes involved in different ways
• The trade-off is achieved in an interactive
  manner, i.e., the user controls the merging so as
  to achieve the desired benefits
• These trade-off analyses can also be automated if
  additional information is stored

28
Trade-off Analyses (Contd.)
29
Trade-off Analyses (Contd.)
30
Trade-off Analyses (Contd.)
31
Trade-off Analyses (Contd.)
32
Nodes Number, Position Configuration

• Factors considered
• Bandwidth requirements
• Buffer requirements
• Sensor, Transducer Topology
• Processing Speed
• Storage requirements

33
Nodes Number, Position Configuration

• Bandwidth-based decisions
• A Greedy algorithm is used to assign sensors to
  transducers and transducers to nodes
• Buffer-space limitations
• Not all sensors that were mapped to a transducer
  based on bandwidth considerations can be
  supported
• The buffer space limits number of sensors
  supported

34
Nodes Number, Position Configuration

• Buffer-space limitations (Contd.)
• This problem is related to periodicity of
  producers

Fig A Periodic Producer
Fig B Aperiodic Producer
35
Nodes Number, Position Configuration

• Sensor, Transducer Topology
• The position of a consumer is affected by the
  distance over which a producer can communicate
• Signal attenuation is an important factor

Fig Sample Sensor Topology
36
Nodes Number, Position Configuration

• Node Configuration
• Node configuration could be determined based on
  existing benchmarks (if any) or based on domain
  specific information

Fig Relationship between TPS and File size
37
Weight Heuristics
Sensor
S3
S4
S5
S2
S1 (5 lbs 0.5 lbs 100 lbs)

• Heuristic
• Select the sensor if the sum of its weight and
  the weights of the lightest instances of the
  other components is below the specified limit

38
Link Decision Tree
Avoid channel failure
Reliability
Minimize garbled data
39
MAP

• Introduction
• Design Issues Challenges
• HUMS Design Automation
• Conclusions

40
Conclusions

• Techniques for design, design automation and
  evaluation have been developed
• The design automation technique can automatically
  produce n! designs for n base models
• Base models have been greatly simplified for
  proof of concept study. The models must be made
  sophisticated to represent real time behavior
• The number of base models considered were minimal
  and hence the suite of models needs more models
  that can help realize the benefit of automation
• The mean values used for calculations and the
  benchmarks used have to represent domain
  information

41
Questions