Preliminary Design of Tall Buildings 2015 · and response parameters of tall buildings directly...
1
CTBUH 2015 New York Conference See Proceedings: Page XXXX Length of Shear Wall along Y Floor Area Cross-sectional Area of Shear Walls Cross-sectional Area of Columns No. of Stories Length along X Length along Y Length of Shear Wall along X Base Shear Roof Drift Cross-sectional Area of Column Cross- sectional Area of Shear Walls Thickness of Shear Walls Floor Area 1.78 4.19 7.05 5.19 7.64 1.90 3.64 6.83 5.44 7.11 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 ) c e s ( d o i r e p l a r u t a N Testing set of building Actual value of Time period Simulated value of time period B2 B3 B5 B4 B1 0.028 0.016 0.023 0.029 0.012 0.03 0.02 0.02 0.03 0.01 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 1 2 3 4 5 f o a e r a o t a e r a n m u l o c f o o i t a R r o o l f r e w o t Testing set of building Actual value Simulated value B2 B3 B5 B4 B1 0.0049 0.026 0.021 0.016 0.038 0.004 0.03 0.02 0.02 0.04 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 1 2 3 4 5 r e w o t o t l l a w e r o c l a t o t f o o i t a R floor Testing set of building Actual Simulated B2 B3 B5 B4 B1 11.46 14.11 12.99 14.11 14.11 11.92 14.63 12.36 13.64 13.21 0 2 4 6 8 10 12 14 16 1 2 3 4 5 p o t t a a e r a r o o l f t i n u r e p t h g i e W ) 2 m / N K ( l e v e l m u i d o p f o Testing set of building Actual Value Simulated value B2 B3 B5 B4 B1 Input Input 1 I1 H1 H2 H3 H4 I2 I3 Input 2 Input 3 Hidden Output Output node Hidden nodes Input nodes The ANN Data Collection Data Pre-Processing Development of Neural Networks Training, Testing and Sensitivity Analysis of Neural Networks Extraction of data from drawings and Selection of input and output parameters A Database Architectural & Structural Drawings and Analysis/Design Results of Case Study Buildings. Normalization of input and output variables using z-score normalization process Selection of best networks after training and performance evaluation by comparison of test set results with actual outputs. Appropriate number of hidden layers, hidden layers nodes Splitting dataset into two categories, Testing and Training Initialize the network weights and bias Create Neural Network MLP Select the appropriate activation function Network training based on BP algorithm Preliminary Design of Tall Buildings Using Artificial Neural Networks Author(s): Lila Khatiwada Naveed Anwar Thaung Htut Aung Jose A. Sy Properly trained Artificial Neural Networks (ANN) based on data of existing buildings can provide a quick alternative for preliminary design and response estimation of new buildings, using basic architectural parameters. Highlights The data collected from architectural drawings and structural design from code based design process of 38 buildings were used as inputs and target for each networks. The input data were taken from the architectural drawings. The target data were taken from the structural design results. A total of 14 input variables and a single target variable were chosen and trained for each network. The performance of networks with different architecture was evaluated in terms of mean square error (MSE) between actual output and target of data set. The best network was chosen with least MSE and highest correlation to training, testing and validation set. The following developed neural network models were employed to simulate the key design and response parameters of tall buildings directly from architectural parameters. a. Model assessment for natural period of building (sec) b. Model assessment for ratio of area of column by area of floor at top of podium level. c. Model assessment for ratio of area of Core wall to area of floor at top of podium level d. Model assessment for ratio of weight per unit floor area of tower (KN/m2) e. Models assessment for ratio of weight of building per unit volume of tower (KN/m3). f. Models assessment for thickness of Shear wall (m) g. Models assessment for maximum story drift ratio h. Models assessment for ratio of base shear to total weight of building. Abstract This study presents the outcome of an artificial neural-network based approach to directly determine design parameters based on experience gained from previously designed buildings, using both code-based and performance-based approaches. Artificial neural network models are trained to determine structural design indicators from architectural parameters. The proposed approach can not only provide means for quick estimation of design output, but can also provide a sanity check on code-based design output as well as performance- based design results. • Artificial Neural Networks with supervised learning can be used in preliminary design of tall buildings to reliably predict the key structural parameters from architectural drawings. • Basic structural design and response parameters of new buildings are estimated from developed ANN models with reasonable accuracy which can not only be used for starting design to determine appropriate member sizes but also to quickly check the results of detailed structural design. • Results of eight developed ANN models for 14 input variables, trained on data extracted from 40 buildings showed a reasonably accurate predicting capacity for new buildings, when compared with results obtained from detailed structural design. • Earthquake and wind loading parameters can also be added and can be utilized in simplified performance assessment, if sufficient building data for different seismic zones and exposure conditions is available. Figure 5: Comparison of actual and predicted values of natural period Figure 6: Comparison of the ratio of total area of columns to floor area of tower Figure 7: Comparison of ratio of total area of core wall to floor area with predicted values Figure 8: Comparison of actual and predicted values of weight per unit floor area Figure 3: The overall methodology and development of Artificial Neural Network (ANN) models with supervised learning Figure 4: Using the developed networks for new buildings Conclusions and recommendations Figure 2: An Overview N1 N2 Inputs in the form of Architectural Parameters Some Other Inputs: • Ratio of total tower height to length of tower along both directions • Aspect ratio of tower plan • Ratio of tower height to length of core wall in both directions • Ratio of length of core wall in x direction to length in y direction Preliminary Design and Response Estimation Natural Time Periods Maximum Roof Drift Column Reinforcement Ratio Shear Wall Reinforcement Ratio Thickness of Shear Wall Area of Columns Floor Area Area of Shear Walls Floor Area Weight of Building Total Volume Maximum Base Shear Total Weight Outputs for Performance- based Design Outputs for Code-based Design Building Response Weight of Building Cumulative Floor Area Natural Time Periods Initial Sizing Columns and Shear Walls Reinforcement Ratio Extraction of Inputs Output for Code-based Design Output for Performance-based Design Architectural Plans & Data Structural Design (Code-based) Performance-based Design N1 N2 Figure 1: What is ANN
Transcript of Preliminary Design of Tall Buildings 2015 · and response parameters of tall buildings directly...
CTBUH 2015New York Conference
See Proceedings: Page XXXX
Length of Shear Wall along Y
Floor Area
Cross-sectional Area of Shear Walls
Cross-sectional Area of Columns
No. o
f Stor
ies
Length along X
Length along Y
Length of Shear Wall along X
Base Shear
Roof Drift
Cross-sectional Area of Column
Cross-sectional Area of Shear Walls
Thickness of Shear Walls
Floor Area
1.78
4.19
7.05
5.19
7.64
1.90
3.64
6.83
5.44
7.11
0
1
2
3
4
5
6
7
8
9
1 2 3 4 5
)ces( doirep larutaN
Testing set of building
Actual value of Time periodSimulated value of time period
B2 B3 B5B4B1
0.028
0.016
0.023
0.029
0.012
0.03
0.02
0.02
0.03
0.01
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
1 2 3 4 5
fo aera ot aera nmuloc fo oitaR
roolf rewot
Testing set of building
Actual value Simulated value
B2 B3 B5B4B1
0.0049
0.026
0.021
0.016
0.038
0.004
0.03
0.02
0.02
0.04
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
1 2 3 4 5
rewot ot lla
w eroc latot fo oitaR
floor
Testing set of building
Actual Simulated
B2 B3 B5B4B1
11.46
14.1112.99
14.11 14.11
11.92
14.63
12.3613.64 13.21
0
2
4
6
8
10
12
14
16
1 2 3 4 5
pot ta aera roolf tinu rep thgieW
)2m/
NK(level muidop fo
Testing set of building
Actual Value Simulated value
B2 B3 B5B4B1
Input
Input 1 I1
H1
H2
H3
H4
I2
I3
Input 2
Input 3
Hidden
Output
Output node
Hidden nodes
Input nodes
The ANN
Data Collection
Data Pre-Processing
Development of Neural Networks
Training, Testing and Sensitivity
Analysis of Neural Networks
Extraction of data from drawings and Selection
of input and output parameters
A Database Architectural & Structural Drawings and Analysis/Design Results of
Case Study Buildings.
Normalization of input and output variables using z-score normalization process
Selection of best networks after training and performance evaluation by comparison of test set results with actual outputs.
Appropriate number of hidden layers, hidden layers nodes
Splitting dataset into two categories, Testing and Training
Initialize the network weights and bias
Create Neural Network MLP
Select the appropriate activation function
Network training based on BP algorithm
Preliminary Design of Tall Buildings Using Artificial Neural Networks
Author(s): Lila Khatiwada Naveed Anwar Thaung Htut Aung Jose A. Sy
Properly trained Artificial Neural Networks (ANN) based on data of existing buildings can provide a quick alternative for preliminary design and response estimation of new buildings, using basic architectural parameters.
HighlightsThe data collected from architectural drawings and structural design from code based design process of 38 buildings were used as inputs and target for each networks. The input data were taken from the architectural drawings. The target data were taken from the structural design results. A total of 14 input variables and a single target variable were chosen and trained for each network. The performance of networks with different architecture was evaluated in terms of mean square error (MSE) between actual output and target of data set. The best network was chosen with least MSE and highest correlation to training, testing and validation set. The following developed neural network models were employed to simulate the key design and response parameters of tall buildings directly from architectural parameters.
a. Model assessment for natural period of building (sec)
b. Model assessment for ratio of area of column by area of floor at top of podium level.
c. Model assessment for ratio of area of Core wall to area of floor at top of podium level
d. Model assessment for ratio of weight per unit floor area of tower (KN/m2)
e. Models assessment for ratio of weight of building per unit volume of tower (KN/m3).
f. Models assessment for thickness of Shear wall (m)
g. Models assessment for maximum story drift ratio
h. Models assessment for ratio of base shear to total weight of building.
AbstractThis study presents the outcome of an artificial neural-network based approach to directly determine design parameters based on experience gained from previously designed buildings, using both code-based and performance-based approaches. Artificial neural network models are trained to determine structural design indicators from architectural parameters. The proposed approach can not only provide means for quick estimation of design output, but can also provide a sanity check on code-based design output as well as performance-based design results.
•Artificial Neural Networks with supervised learning can be used in preliminary design of tall buildings to reliably predict the
key structural parameters from architectural drawings.
•Basic structural design and response parameters of new buildings are estimated from developed ANN models with reasonable
accuracy which can not only be used for starting design to determine appropriate member sizes but also to quickly check the
results of detailed structural design.
•Results of eight developed ANN models for 14 input variables, trained on data extracted from 40 buildings showed a reasonably
accurate predicting capacity for new buildings, when compared with results obtained from detailed structural design.
•Earthquake and wind loading parameters can also be added and can be utilized in simplified performance
assessment, if sufficient building data for different seismic zones and exposure conditions is
available.
Figure 5: Comparison of actual and predicted values of natural period
Figure 6: Comparison of the ratio of total area of columns to floor area of tower
Figure 7: Comparison of ratio of total area of core wall to floor area with predicted values
Figure 8: Comparison of actual and predicted values of weight per unit floor area
Figure 3: The overall methodology and development of Artificial Neural Network (ANN) models with supervised learning
Figure 4: Using the developed networks for new buildings
Conclusions and recommendations
Figure 2: An Overview
N1
N2
Inputs in the form of Architectural Parameters
Some Other Inputs:
• Ratio of total tower height to length of tower along both directions
• Aspect ratio of tower plan
• Ratio of tower height to length of core wall in both directions
• Ratio of length of core wall in x direction to length in y direction
Preliminary Design and Response Estimation
Natural Time Periods
Maximum Roof Drift
Column Reinforcement Ratio
Shear Wall Reinforcement Ratio
Thickness of Shear Wall
Area of Columns Floor Area
Area of Shear Walls Floor Area
Weight of Building Total Volume
Maximum Base Shear Total Weight
Outputs for Performance-based Design
Outputs for Code-based Design
Building Response
Weight of Building Cumulative Floor Area Natural Time Periods
Initial Sizing
Columns and Shear Walls Reinforcement Ratio
Extraction of Inputs
Output for Code-based Design
Output for Performance-based
Design
Architectural Plans & Data
Structural Design (Code-based)
Performance-based Design
N1
N2
Figure 1: What is ANN
