Professional Documents
Culture Documents
Degree
School of Civil Engineering
• Design of RBFNN
RBFNN prediction and • Training, testing and validation
verification • Prediction & comparison of Fragility Curves
• Discussion and Conclusion
Part-1 BACKGROUND AND
SIGNIFICANCE
Late
1900’s
1.1BACKGROUND OF THE RESEARCH
Source of earthquake(M)
Material
Uncertainty(E&Fy)
Capacity Uncertainty
Geometric
Uncertainty(L,W,H)
BEAM COLUMN
2. 2 Uniform Design Method(UDM-sampling method)
UDM is selected having 10 different sample numbers and 8 factors with 10 levels Un(nm) =
SPACE Better
FILLIN UDM Represent
G ativeness
Fewer
Number of
runs
2. 3 Selection of Ground Motion Parameter
Data taken from the official website of the PEER ground motion database (http://ngawest2.berkeley.edu) is used.
An example below:
coverage
50 samples
Sa
Sd
E
Sa vs T plot shows Sd vs T plot shows the Distribution of
the peak acceleration
PGA
peak displacement for earthquake in
for SDOF system at SDOF system at different Magnitude(M) and
M&R
different periods. periods. Source to site distance.
(R)
2. 3 Selection of Ground Motion Parameter
User- BEAM189
Friendly
10 different
steel models MASS21
2. 4 Modeling and Analysis Of Structure
Modal Analysis Results
Dimension
Sample Length Breadth Height
To find Natural Frequency and Time Period(Tn)
1 4m 4m 2.65m
Assist in the selection of variety of earthquake data.
Element Selection Mode
Component Element Type Density(kg/m3) 1 1.203 1.269 1.329 1.531 1.586 1.947 2.023
7850+(added 2 1.075 1.098 1.192 1.285 1.459 1.625 1.815
Beam BEAM189 mass/volume)
3 0.881 0.895 0.966 1.038 1.189 1.300 1.479
7850kg/m3
Column BEAM189 4 1.386 1.386 1.474 1.489 1.654 1.654 1.880
Elasto-plastic behavior(Bilinear
Isotropic Hardening)
Part-3 TIME HISTORY ANALYSIS AND
FRAGILITY CURVE
3.1 Time History Analysis Of Structure
Load Steps and settings in time history analysis
Results obtained from analysis for one member
Maximum deformation obtained at the end of analysis(NLEPEQ(mm/mm)
• Global Damage Index() & Weight() 62 1.12E-02 2.19E+07 0.025 1.15E+06
0.01 0.001
0.01 0.001
; Dtotal= 0.376
Dtotal= 0.376
FLOOR DAMAGE INDEX DISTRIBUTION
Overall damage
= 0.627
𝑆𝑎
𝑆𝑎
BOTH EARTHQUAKES AND STRUCTURES are DIFFERENT
FLOOR Max Inter-storey Drift Ratio(MIDR)
𝑆𝑎
𝑆𝑎
BOTH EARTHQUAKES AND STRUCTURES are DIFFERENT
3.3 Cloud Analysis
My results for Sd
Fragility Curves for 4 damage level
3.4 Fragility Curve based on cloud analysis
SOFT COMPUTING
MACHINE LEARNING
ARTIFICIA
L NEURAL
NETWORK
Why is RBFNN chosen??
Saves
computing
Self Organizing time High
Capability efficiency
Adaptive Cheap
Learning
Parallel
Processing Reliable
4.1 Design of RBFNN
10 10 ground
Steel frame X motion
sample = 100 sample
data
50%
45% training
validation
5% testing
Common Neural Network Architecture
4-Different RBFNN
Structures Considered
Number of
Spread
Neurons
RBFN RBFNN-1 1.4 24
RBFN
N-1 RBFNN-2 0.7 18
N-2
RBFNN-3 10 21
RBFNN-4 0.4 26
MSE =
RBFNN-1 RBFNN-3
RBFNN-2 RBFNN-4
4.4. DISCUSSION AND CONCLUSION
• Discussion
Comparision of median and beta with reference results
Minimum difference on
Slight Moderate Extensive Collapse
results from RBFNN
Median Beta Median Beta Median Beta Median Beta
and FEA is in RBFNN-2.
0.198 0.601 1.008
Best = RBFNN-2
RBFNN- FEA 0.760 0.760 0.760 - -
1 0.073 0.411 0.931
Difference in median =
NN 0.619 0.619 0.619 - -
0.022
RBFNN- FEA 0.028 0.414 0.163 0.414 0.371 0.414 - -
2
Difference in beta =
NN 0.006 0.378 0.092 0.378 0.347 0.378 - -
0.036
RBFNN- FEA 0.214 0.68 0.6 0.68 0.969 0.68 - -
3 NN 0.076 0.64 0.416 0.64 0.917 0.64 - -
RBFNN- FEA 0.196 0.981 0.593 0.981 0.993 0.981 - -
4 NN 0.088 0.677 0.437 0.677 0.923 0.677 - -
3. DISCUSSION AND CONCLUSION