Improved driven pile design and construction control: A dynamic load test program of Illinois transportation bridge foundations with machine learning insights

Anderson, Andrew Christian

Permalink

https://hdl.handle.net/2142/114103 Copy

Description

Title

Improved driven pile design and construction control: A dynamic load test program of Illinois transportation bridge foundations with machine learning insights

Author(s)

Anderson, Andrew Christian

Issue Date

2021-12-06

Director of Research (if dissertation) or Advisor (if thesis)

Long, James

Doctoral Committee Chair(s)

Long, James

Committee Member(s)

• Mesri, Gholamreza
• Olson, Scott M
• Tutumluer, Erol

Department of Study

Civil & Environmental Eng

Discipline

Civil Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• geotechnical
• deep foundations
• dynamic testing
• machine learning

Abstract

Driven pile foundations are the most prevalent type of deep foundation system employed in the United States and in Illinois. A five-year dynamic load test field program was conducted on transportation highway bridge pile foundations to generate a database of measured axial pile capacity, pile setup, and driving induced pile stresses for Illinois soil conditions and construction practice. The database provides a performance baseline for existing methodologies, regional LRFD calibration, support for increased reliance on pile setup in pile design, and improved construction control for pile capacity and pile integrity verification. Pile foundations are typically composed of large quantities of relatively low-cost piles. Therefore, a reduction in the average required pile length to achieve nominal required bearing, required pile section size, frequency of pile splices, and size of cutoff lengths will increase foundation efficiency and result in significant cost savings. The primary objective of this dissertation, associated field research program, and resultant dynamic pile load test database is to increase the efficiency of pile design and improve construction control techniques for pile bridge foundations in Illinois. Pile efficiency is increased through 1) calibration of axial pile capacity methods for design, monitoring, and capacity verification 2) quantification and modeling of pile setup by pile and soil type, and 3) quantification and modeling of pile stresses to ensure pile integrity for Illinois soils. Pile design methods are improved through quantification of pile setup rate and magnitude for Illinois soil types (dominant soil texture along embedment length: clay, sand, mixed) and pile section (H-pile or closed ended shell) and development of respective correlation parameters. In addition to the evaluation of selected pile setup methods applied in common practice several modified formulas, new formula frameworks, and machine learning based approaches were examined. Application of separate resistance factors for pile setup and end-of-driving resistance are examined to differentiate the dissimilar magnitudes of uncertainty in these components of axial pile resistance. A log linear time based setup model is developed with additional terms to account for differences in observed setup correlated with average standard penetration test (SPT) blow count along the pile embedment length in Illinois soils. A rapid site evaluation tool is developed using a machine learning based approach to classify SPT borings for the occurrence or absence of significant setup relevant for design. Significant setup is defined as a change in total capacity greater than twenty percent. Dynamic formulas are reviewed, implicit setup is quantified by soil profile texture, and an explicit setup framework (ESF) is developed. Three databases of dynamic field tests are reviewed to quantify inherent setup in five common dynamic formulas for Illinois soils, Wisconsin soils, and national soils databases. Two new methods are developed for predicting pile induced driving stresses 1) wave equation calibrated mechanistic formula approach (Illinois Simplified Stress Method, SSF) and 2) a machine learning approach applying gradient boosted regression trees. The results from this research have been implemented by IDOT to successfully transition pile design methods for bridge foundations from allowable stress design (ASD) to load resistance factor design (LRFD) and support the application of database derived resistance factors larger than default code values specific to Illinois soils and construction practice.

Graduation Semester

2021-12

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/114103

Copyright and License Information

Copyright 2021 Andrew Anderson

Owning Collections