|
iSYS R&D(Intelligent SYStem Reliability & Design Laboratory) |
||
|
Home | Professor
|
People
| Research | Publications | Resources |
||
|
|
|
|
|
Research Prognosis & Health Management(PHM) |
It consists of three research topics: 1) Data
Mining and Knowledge Extraction, 2) Predictive Modeling and 3) Risk-based
Design Optimization. 1) Data Mining and Knowledge ExtractionMotivation:
A big challenge in engineering decision-making is
that it demands on adequate data to support the decisions, but in a real world,
considerable kinds of data, especially subjective data (e.g., expert
knowledge, customer survey) and certain testing data (i.e., failure data
for nuclear plant or aircraft), are
very difficult to obtain and in a limited amount. Additionally, due to the
insufficiency of the data, treated results are always in a large uncertainty.
This makes engineering decision-making more difficult. Therefore, the
objective of this research is to find an effective way to mine for subjective and insufficient data, extract statistical knowledge of those data, and provide valuable information for
engineering decision-making. Project
Highlight: ·
A model has
been developed to elicit subjective data, and then to find out the relevant
objective data to help decision-making. This model contains four sub models:
Decision Decomposition Model, Subjective Data Model, Measured Data Model and
Decision Model.
·
This model
has been applied to a cell-phone case study. Relevant
Publications: 1. 2) Predictive Modeling
Motivation: Nowadays, people are more and more relying on
predictive models for components and system design considering cost, time limitation, planning, and
other technical and economic factors. Accurate prediction of system
performance is critical to produce reliable products. The existence of
uncertainties makes it difficult to predict the system performance accurately
and efficiently. Furthermore, epistemic uncertainty due to lack of data is
inevitable challenges in this area. Project
Highlight: • Eigenvector Dimension Reduction (EDR) method
is proposed for accurate and efficient uncertainty propagation analysis. EDR
method is able to handle correlated/uncorrelated variables, arbitrary
symmetric/asymmetric input distributions. A probability density function
(PDF) of system response is constructed in the method. • An adaptive dimension-reduction-based polynomial
chaos expansion (DR_PCE) method is proposed for accurate and efficient uncertainty
propagation analysis while remove some limitations in the EDR method. DR_PCE method is
able to handle nonlinear correlated variables, arbitrary input distributions,
and predict multi-mode response PDF. • Bayesian updating technique
(conjugate/non-conjugate) is employed to deal with epistemic uncertainty.
Reliability function is proposed using Bayesian theory with both aleatory and
epistemic uncertainties. EDR method is incorporated for efficiency
improvement for Bayesian reliability analysis. • A generic framework to characterize a random field (spatial
variability) structure is
proposed for reliability analysis and design. Proper consideration of the random field is quite
significant to variability in system performances in many engineering
applications, especially, geometry-sensitive failures (e.g., buckling, fillet
failure), small-scale applications in which tolerance control is more
challenging. Relevant
Publications: 1. Youn B.D., Xi Z., and Wang P.F., “Eigenvector Dimension-Reduction (EDR) Method
for Sensitivity-Free Uncertainty Quantification,” Structural
Multidisciplinary Optimization, v37, n1, 2008 2. Wang, P., Youn, B.D., Xi Z., and Artemis, Kloess, "Bayesian Reliability
Analysis with Subjective, Insufficient, and Evolving Data Sets,” Accepted, Journal
of Mechanical Design, ASME, 2008;
(ASME - DAC Best Paper 2008). 3. Hu C. and Youn
B.D., “Advances in Polynomial Chaos Expansion for Structural Reliability Analysis
and Design,” Probabilistic Engineering Mechanics, Submitted, 2008 4. Xi Z. and Youn
B.D., “An Effective Random Field Characterization for Probability Analysis
and Design,” Structural Multidisciplinary
Optimization, Submitted, 2009 3)
Risk-based Design Optimization Motivation: Traditional risk-based design optimization
(RBDO) is haunted with lack of accuracy, efficiency, and stability. Sampling
method is too expensive to be applicable in reality. Expansion method can only
provide moderate accuracy while requiring sensitivity. MPP-based method
generates better accuracy than expansion method with the aid of sensitivity.
However, relatively large error is expected with high nonlinear response or
multiple MPPs. Response surface method is suffered from the curse of
dimensionality. Recently proposed dimension reduction (DR) method is not able
to maintain high accuracy without sacrificing efficiency. All these obstacles
prevent a promising RBDO method in real engineering application. Project
Highlight: • Eigenvector Dimension Reduction (EDR) method
is proposed for sensitivity-free reliability analysis with high accuracy and
efficiency. • An approximate response surface facilitates design sensitivity
calculation where the response surface is constructed using the eigenvector
samples from the EDR method. Thus, sensitivity analysis for design
optimization becomes very efficient and simple. • The proposed RBDO methodology has a single-loop structure. One
EDR execution evaluates a set of quality (objective) and reliability
(constraint) functions. Each evaluation in one EDR execution can be computed
independently by a parallel computing power. RBDO can be far more efficient. Relevant
Publications: 1. Youn B.D., Xi Z., and Wang P.F.,
“Eigenvector Dimension-Reduction (EDR) Method for Sensitivity-Free
Uncertainty Quantification,” Structural Multidisciplinary Optimization,
v37, n1, 2008 2. Youn B.D. and Xi Z., “Reliability-based Robust Design Optimization Using
the Eigenvector Dimension Reduction (EDR) Method”, Structural
Multidisciplinary Optimization, v37, n5, 2009 |
|
|
|
|
|
