validation gaining confidence in simulation darre odeleye ceng mimeche
TRANSCRIPT
Simulation and Modelling 2016 Conference Tuesday 13 and Wednesday 14 September 2016, Birmingham
Thinktank, Birmingham Science Museum Millennium Point Birmingham West Midlands B4 7XG
VALIDATION: GAINING CONFIDENCE
IN SIMULATION A Program Office perspective
Distinction must be made between
the verification and validation of the
key components that constitute a
simulation
• Verification - Does the code implement the physics of the
problem correctly and does the solution generated
compared favourably to exact analytical results ?
• Validation – Does the actual simulation agree with physical
reality? Is the level of uncertainty and error acceptable ?
From a Program Office perspective Validation
assessment generates credibility for decision
makers
Simulation verification may include analysis of:
• Discretization strategy,
• Application of boundary conditions,
• Grid convergence criteria,
• Handling non-linearity,
• Iterative convergence,
• Numerical stability ( relaxation factors etc.)
D.Odeleye CEng MIMechE 4
–Validation is the process of determining the degree to which a model is an accurate representation of the real world from the perspective of the intended use of the model.
–High confidence simulations offers the promise of developing higher quality products with fewer resources in less time
D.Odeleye CEng MIMechE 5
• Generating the right data, not a lot of
data
• Understanding what is appropriate and
required for validating models
• Assessing the influencing factors that
need to be considered and tested
teg
D.Odeleye CEng MIMechE 7
D.Odeleye CEng MIMechE 8
• What does the Program need to
know and when? – This defines the problem and what CAE methods that can be
used
• Lean product creation concepts
require – Compatibility with Quality, Cost& Delivery before design maturity
– Front loading using all design options before commitment
– No late changes
D.Odeleye CEng MIMechE 9
• Generating the right data, not a lot of
data
• Understanding what is appropriate and
required for validating models
• Assessing the influencing factors that
need to be considered and tested
teg
Correlate and compare the
components of simulation in
order to validate models i.e.
– Invest upfront resources to generate the most accurate inputs
for Simulation
– Compare results from different Simulation tools,
– Compare real world measurements to simulated results,
– Compare “controlled” test environment ( i.e. wind tunnel)
results to simulation results.
D.Odeleye CEng MIMechE 11
Extend legacy/baseline physical
tests to generate comprehensive
model input parameters
– Typical durability testing focuses on “ has it broken yet” .
providing little information on optimisation and distance from
failure modes
– Testing to failure provides quantitative data i.e. time to failure,
cycles to failure etc. and damage mode provides insight into
distance from failure mode and performance degradation
supports better predictive simulations.
D.Odeleye CEng MIMechE 12
D.Odeleye CEng MIMechE 13
Bogey Testing
Testing to Failure
Degradation testing
Trade off between different test schemes and implications
Delivery Cost Quality
Long test duration impact on engineering sign off timing
High Cost
Comprehensive data
Shortest test duration Most likely to meet program timing
Lowest relative Cost. Results of test cannot be used To predict performance with sufficient confidence
D.Odeleye CEng MIMechE 14
Bogey test
Test to failure
D.Odeleye CEng MIMechE 15
Predicted life Crankshaft #1
Predicted life for Crankshaft #3 ( revised #2 @125PS)
Predicted life for Crankshaft #3 ( revised #2 @150PS)
Crankshaft #1 Crankshaft #2 Crankshaft #3 ( revised #2) Crankshaft #3 @ 150PS
D.Odeleye CEng MIMechE
Original design is the purple part and the
green part is the revised part with the
width of web 6 increased
17
•Based on sampling at the extremes of either product strength, or level of
induced key stresses.
i.e. Selection of weakest parts, worst clearances, etc.
•If Input data is based on the weakest part test data, all stronger parts would
also pass the test ( for a given test cycle).
D.Odeleye CEng MIMechE 18
D.Odeleye CEng MIMechE 19
D.Odeleye CEng MIMechE 20
D.Odeleye CEng MIMechE 21
D.Odeleye CEng MIMechE 22
D.Odeleye CEng MIMechE 23
D.Odeleye CEng MIMechE 24
D.Odeleye CEng MIMechE 25
• Generating the right data, not a lot of
data
• Understanding what is appropriate and
required for validating models
• Assessing the influencing factors that
need to be considered and tested
teg
Conduct studies to identify
sources of
– Variation in required simulation output, what input parameters
have what effect on the required results ?
– Conduct sensitivity analysis of input parameters ,
– Classify influencing factors and identify main effects
• Validate input data from experimental tests, does it
make engineering sense
D.Odeleye CEng MIMechE 27
D.Odeleye CEng MIMechE 28
D.Odeleye CEng MIMechE 29
Inputs CAD, Mesh
generation, space ( F.E, Volume,
elements) and time ( stability constraints)
discretization
Parameters, Boundary
Conditions.
Solver
Settings i.e. mesh refinement vs
solution convergence time
Different Simulation tools, iterative solver
Output
Simulation Solutions
Experimental data ( controlled noise
environment), real world data
Compare components , conduct sensitivity analysis
and Parametric studies to validate simulations
Simulation components
Confidence in Simulation is gained
by demonstrating value added to
– Quality – credible data necessary to make decisions at each stage of the
product development , fidelity of predictions over time.
– Cost – return on investment versus e.g. cost of testing prototypes
– Delivery – are simulations conducted and results available exactly when
needed ?
D.Odeleye CEng MIMechE 31
Further considerations and
opportunities for increasing
confidence • Faster post processing
• Use of virtual reality & augmented reality,
• Real time simulation ( running simulations on the ‘fly’ ),
• Cloud computing
• Artificial intelligence ( using techniques such as evolutionary
computations, artificial neural networks, fuzzy systems , general
machine learning and data mining methods
• Simulating Electric vehicles and EV powertrains
• Simulating Autonomous vehicle performance.
D.Odeleye CEng MIMechE 32
• Electric/hybrid vehicles present the next opportunities for the use
of simulation to further compress product development
timescales.
D.Odeleye CEng MIMechE 33
DISCUSS
The better the Validation the better
the prediction consequently the
more confidence customers have
in ongoing and future Simulations
D.Odeleye CEng MIMechE 35