6685

6236

كارشناسي ارشد

خاك و پي

اصفهان: دانشگاه صنعتي اصفهان، دانشكده عمران

1390

يازده، 135ص.: مصور، جدول، نمودار

ص.ع. به فارسي و انگليسي

محمدعلي روشن ضمير

مازيار پالهنگ

17/2/91

حميد هاشم الحسيني، بهروز كوشا

1396/10/06

كتابنامه

عمران

مهندسي عمران

ID6236

به فارسي و انگليسي: قابل رويت در نسخه ديجيتالي

Prediction of Pile Axial Bearing Capacity Using Artificial eural etwork Based on Ensemble Learning Seyed Mehdi Moosavi Sm moosavi@cv iut ac ir September 21 2011 Department of Civil Engineering Isfahan University of Technology Isfahan 84156 83111 IranDegree M Sc Language FarsiMohammad Ali Rowshanzamirmohamali@cc iut ac irMazyar Palhangpalhang@cc iut ac irAbstract The cone penetration test CPT is one of the most common of in situ tests used for pile designapplications This is because the test probe simulates a model pile The measured cone resistance qcand sleeve friction qs are used to calculate respectively the pile toe and shaft resistance Theartificial neural network modeling has been used by geotechnical researchers as a complementarytool for examining the pile load test results along with CPT during the past decade Till nowvarious neural network models have been considered and the obtained results have been comparedwith each other but no unique model capable of properly predicting accurate results based onlimited training data has been introduced The artificial neural network modeling requires access tothe right set of data Thus a comprehensive set of data including fifty eight case studies on variouspiles from different parts of world was collected The data base included the results of load piletests the soil profile of the pile sites the corresponding CPT results and the piles geometry Theback propagation artificial neural network modeling was employed to predict the piles overall axialcapacities using the piles effective cross section the embedment length of piles and mean valuesof qc and qs from the CPT logs The early stopping and multiple verification techniques were usedto avoid of common problems in the neural networks such as overfitting in the modeling of variousfunctions Moreover to fix the unwanted effects of the input parameters dimensions on the networkproperties the pre processing and post processing techniques were applied respectively on thenetwork input and output data Applying the suggestions of previous researchers on properlymodeling the neural networks made it possible to easily build the required neural network modeland reduced the try and error time for achieving the optimum network As the purpose of this studywas to evaluate the capability of neural network model in predicting the pile capacity the sameinput parameters were used as those in the pile capacity computations in the conventional methodssuch as Schmertmann Notingham De Ruiter Beringen European and direct CPT methods The values of correlation coefficient between the predicted and the measured results of the pilesbearing capacity R2 were found to be 88 2 73 3 65 4 respectively in these conventionalmethods while the single model of neural network yielded the R2 value of 88 in comparison tothe values of 91 and 95 from ensemble learning methods of Bagging and Boostingrespectively These results clearly indicated that the ensemble learning techniques are moresuccessful than the single neural network and could predict the pile axial capacity with high levelof accuracy even with limited training data These results showed also a great advantage of theseapproaches in comparison with conventional methods Keywords Pile axial bearing capacity Artificial neural network pile loading test Cone penetrationtest CPT Ensemble learning

محمدعلي روشن ضمير

مازيار پالهنگ

حميد هاشم الحسيني، بهروز كوشا