May 19, 2011
School of Engineering, Coimbatore
Earthquakes happen all the time. The harm they can cause, ranges from minor to disastrous. In the case of a massive earthquake like the recent 9.0 cataclysmic Japan trembler, both property and lives were destroyed.
Technology to reduce damage to buildings that house people is essential to protect society, when earthquakes unexpectedly strike.
Design and construction based on sophisticated modeling techniques that determine how much earthquake-induced force a structure can withstand, can help.
Artificial Neural Networks (ANN) are used to ascertain this calculation. A data processing system that mimics biological neural networks (interconnected neurons in the brain), ANN is being used in structural engineering since the late ’80s.
This promising technique was researched by Dr. K. M. Mini, Department of Civil Engineering at Amrita’s Coimbatore campus, for her doctoral dissertation. She utilized ANN to model, analyze and assess the efficacy of infilled frames subject to lateral loading, during earthquakes and high winds.
The work will be now be available as a research monograph titled Analysis and Assessment of Behaviour of Infilled Frames Using Artificial Neural Networks.
Published by Germany-based Lambert Academic Publishers, with ISBN # 978-3-8443-1068-9, the monograph is priced at 79 Euros.
“An infilled frame is a composite structural system comprised of steel or reinforced concrete,” explained Dr. Mini. “The frame is strengthened by stone, brick or concrete panels that fill planar voids between beams and columns. Infilled frames aid in the resistance of lateral forces and horizontal loads caused by earthquakes, also offering additional rigidity which improves structural performance by reducing lateral deflection.”
Analyzing multistory infilled frames is complex and tedious. The normal analytical method only measures the structure of the bare frame, not the infill portion. To accurately ascertain structural behavior, the infill portion must also be assessed. Mathematics alone cannot help, modeling is also necessary.
The traditional limited modeling approach utilizes an empirical equation followed by a regression analysis. These mathematical methods analyze experimental data to determine unknown coefficients.
“ANN emerged as an alternative analytical method because it didn’t need to establish a mathematical relationship to use available field results for clustering, mapping and classifying input and output,” explained Dr. Mini. “Among ANN’s available networks, Back Propagation Network (BPN) is used today to solve most structural engineering problems.”
In her research, Dr. Mini modeled the behavior of infilled frames by adopting a multilayer feed-forward network with back propagation learning. Reinforced concrete and steel frames were analyzed. A parametric study was used to define the optimum architecture of the network.
After testing all patterns, Dr. Mini concluded that trained networks were able to satisfactorily predict failure loads and displacement of both R.C. and steel infilled frames.
