American Association Of State Highway And Transportation Officials

Special Committee On Research And Innovation

FY2023 NCHRP PROBLEM STATEMENT TEMPLATE

Problem Number: 2023-E-01

Problem Title

Development Of Machine Learning Algorithms For Geotechnical Applications And Assembly Of A Quality Database Of Deep Foundation Load Tests

Background Information And Need For Research

Machine Learning (ML) Is An Approach For Developing Empirical Relationships That Allows Computer Systems To Learn From Data Without Knowing The Laws That Govern The Phenomenon From Which The Data Were Obtained And Collected. Nowadays, With The Availability Of Previously Unseen Amounts Of Data, Cloud Storing Capabilities, And The Continuously Increasing Computing Power, ML-Based Techniques Have Become A Suitable Tool For A Number Of Applications In Many Areas, Including Geotechnical Engineering. Despite The Increasing Interest In ML Within The Engineering Community, Applications Within The Realm Of Geotechnical Engineering In NCHRP And TRR Libraries Are Scarce.

The Major Resource Needed To Develop Predicting Tools Using ML Are Data. State Departments Of Transportation (Dots) Manage Large Amounts Of Geotechnical Data Associated With Public Infrastructure. These Data Represent An Asset For Applications Of ML To Classical Geotechnical Problems Such As The Bearing Capacity Of Deep And Shallow Foundations, Settlement Estimation, Landslide Susceptibility, Liquefaction Susceptibility, And Others. However, The Lack Of A Standard Format Or Structure Limits The Usefulness Of Those Databases. In This Regard, The Efforts Of The Data Interchange For Geotechnical And Geoenvironmental Specialists (DIGGS) Project Is Relevant Since It Provides A Standard Format To Store And Share Geotechnical Data. The DIGGS Protocols Should Be Followed For Assembling Databases. Thus, Later Development And Updates Of Statistical Regressions, Reliability-Based Calibration, And ML Applications Use The Same Basis.

Quality And Complete Deep Foundation Load Test Databases, Among Other Databases, Are Needed By Researchers In Performing Reliability-Based Calibrations And By Practitioners In Evaluating And Improving The Geotechnical Design. Datasets Regarding Deep Foundation Load Tests Are Typically Generated When Initial Site Characterization Information Is Combined With Foundation Performance Testing To Provide A Comprehensive And Complete Record Of Site Character, Predicted Performance, And Measured Performance Of Foundation Systems. Complete Load Testing Datasets Include Three Types Of Information: (1) Soil/Rock Information (Geotechnical Site Characterization Data, Foundation Design Parameters, SPT Hammer Energy, Sample Extraction, And Lab Testing, CPT, DMT, PMT, Etc.). (2) Test Foundation Information (Location, Layout, Dimensions, Type, Construction And Installation Information, And Quality Control); And (3) Load Testing Information: Test Procedure, Duration, Rates Of Loading, Test Data, And Results)

Literature Search Summary

Artificial Intelligence (AI) And Machine Learning (ML) Techniques Have Been Used In A Number Of Disciplines Within Science, Engineering, Economics, And In Many Topics Of Public Interest, Such As Healthcare, Transportation, And Finance. Several Applications Of ML To Classical Geotechnical Problems Such As Bearing Capacity Prediction, Settlement Estimation, The Load-Settlement Response Of Pile Foundations, Liquefaction Susceptibility, And Landslide Susceptibility Can Be Found In The Literature.

Three Factors Have Helped Lead To The Growing Popularity Of AI And ML: (1) Availability Of Data, (2) Increases In Computer Power, And (3) Maturity Of The Algorithms Used To Process The Data. ML Toolboxes And User-Friendly Software Are Now Available To Engineers To Train Their Own Models And Make Predictions From Their Collected Data. Numerous ML Techniques Have Been Developed In The Past Several Decades, Theoretically And Practically. Among All These Techniques, Tree-Based Methods, ANN, And Svms Have Been Used For Studying Geotechnical Engineering Problems. Several Applications And Reviews About ML In Geotechnical Engineering Can Be Found In The Following List:

Aguilar, V., Wu, H., & Montgomery, J. (2020). Machine Learning In Foundation Design And More. TRB Centennial Circular, 103.

Ahmed, A., S. Khan, S. Hossain, T. Sadigov, And P. Bhandari. Safety Prediction Model For Reinforced Highway Slope Using A Machine Learning Method. Transportation Research Record. Https://Doi.Org/10.1177/0361198120924415.

Breiman, L., J. H. Friedman, R. A. Olshen, And C. J. Stone. Classification And Regression Trees. Belmont, CA: Wadsworth. International Group, Vol. 432, 1984, Pp. 151–166.

Goetz, J. N., A. Brenning, H. Petschko, And P. Leopold. Evaluating Machine Learning And Statistical Prediction Techniques For Landslide Susceptibility Modeling. Computers & Geosciences, Vol. 81, 2015, Pp. 1–11. Https://Doi.Org/10.1016/J.Cageo.2015.04.007.

Goh, A. T. C., And S. H. Goh. Support Vector Machines: Their Use In Geotechnical Engineering As Illustrated Using Seismic Liquefaction Data. Computers And Geotechnics, Vol. 34, No. 5, 2007, Pp. 410–421. Https://Doi.Org/10.1016/J.Compgeo.2007.06.001.

Harandizadeh, H., M. M. Toufigh, And V. Toufigh. Different Neural Networks And Modal Tree Method For Predicting Ultimate Bearing Capacity Of Piles. 2018, P. 18.

Khan, S. Z., S. Suman, M. Pavani, And S. K. Das. Prediction Of The Residual Strength Of Clay Using Functional Networks. Geoscience Frontiers, Vol. 7, No. 1, 2016, Pp. 67–74. Https://Doi.Org/10.1016/J.Gsf.2014.12.008.

Kordjazi, A., F. P. Nejad, And M. B. Jaksa. Prediction Of Load-Carrying Capacity Of Piles Using A Support Vector Machine And Improved Data Collection. 2016, P. 9.

Kordjazi, A., F. Pooya Nejad, And M. B. Jaksa. Prediction Of Ultimate Axial Load-Carrying Capacity Of Piles Using A Support Vector Machine Based On CPT Data. Computers And Geotechnics, Vol. 55, 2014, Pp. 91–102. Https://Doi.Org/10.1016/J.Compgeo.2013.08.001.

Momeni, E., R. Nazir, D. Jahed Armaghani, And H. Maizir. Application Of Artificial Neural Network For Predicting Shaft And Tip Resistances Of Concrete Piles. Earth Sciences Research Journal, Vol. 19, No. 1, 2015, Pp. 85–93. Https://Doi.Org/10.15446/Esrj.V19n1.38712.

Samui, P. Support Vector Machine Applied To Settlement Of Shallow Foundations On Cohesionless Soils. Computers And Geotechnics, Vol. 35, No. 3, 2008, Pp. 419–427. Https://Doi.Org/10.1016/J.Compgeo.2007.06.014.

Samui, P. Vector Machine Techniques For Modeling Of Seismic Liquefaction Data. Ain Shams Engineering Journal, Vol. 5, No. 2, 2014, Pp. 355–360. Https://Doi.Org/10.1016/J.Asej.2013.12.004.

Shahin, M. A. Load–Settlement Modeling Of Axially Loaded Steel Driven Piles Using CPT-Based Recurrent Neural Networks. Soils And Foundations, Vol. 54, No. 3, 2014, Pp. 515–522. Https://Doi.Org/10.1016/J.Sandf.2014.04.015.

Shahin, M. A. State-Of-The-Art Review Of Some Artificial Intelligence Applications In Pile Foundations. Geoscience Frontiers, Vol. 7, No. 1, 2016, Pp. 33–44. Https://Doi.Org/10.1016/J.Gsf.2014.10.002.

Song, Y.-Y., And L. U. Ying. Decision Tree Methods: Applications For Classification And Prediction. Shanghai Archives Of Psychiatry, Vol. 27, No. 2, 2015, P. 130.

Tarawneh, B. Pipe Pile Setup: Database And Prediction Model Using Artificial Neural Network. Soils And Foundations, Vol. 53, No. 4, 2013, Pp. 607–615. Https://Doi.Org/10.1016/J.Sandf.2013.06.011.

Tarawneh, B. Predicting Standard Penetration Test N-Value From Cone Penetration Test Data Using Artificial Neural Networks. Geoscience Frontiers, Vol. 8, No. 1, 2017, Pp. 199–204. Https://Doi.Org/10.1016/J.Gsf.2016.02.003.

The Federal Highway Administration (FHWA) Has Led The Development Of The Deep Foundation Load Test Database (DLFTD) (Version 1.0 2007) That Led The Development Of The First Generation Of AASHTO LRFD Factors In The NCHRP Report 507. In 2017, The FHWA Released The Updated FHWA Deep Foundation Load Test Database (DFLTD V.2). The DFLTD V.2 Replaces The Previous DFLTD (V.1) And Also Adds New Information On Over 150 Load Tests On Large Diameter Open-End Piles. In Addition, Several State Dots And Researchers Have Developed And Are Now Developing Their Own Foundation Load Test Databases (E.G., Florida, Iowa, Louisiana, California, Nevada (Https://Trid.Trb.Org/View/1393456), New Mexico And Illinois,).

Additional Research Efforts Include:

Load And Resistance Factor Design (LRFD) Pile Driving Project -Phase Two Study (Paikowsky Et Al. 2014).

Developing A Resistance Factor For Mn/DOT’s Pile Driving Formula, (S. Paikowsky Et Al. 2009)

Capacities And Resistance Factors For Driven Piling In Illinois, (J. Long, A. Anderson, W. Kramer, 2014)

Improved Design For Driven Piles Based On A Pile Load Test Program In Illinois: Phase 2, (J. Long, A. Anderson, 2014)

Application Of LRFD Geotechnical Principles For Pile Supported Bridges In Oregon: PHASE 1, Final Report, OTREC-TT-09-0, (Trevor D. Smith, Peter Dusicka, Portland State University, 2009)

Updating Florida Department Of Transportation's (FDOT) Pile/Shaft Design Procedures Based On CPT & DTP Data, (D. Bloomquist, M. Mcvay, Zhihong Hu, 2007)

Published 2007

Improving Agreement Between Static Method And Dynamic Formula For Driven Cast-In-Place Piles In Wisconsin, (J. Long, 2013)

Evaluation Of FHWA Pile Design Method Against The FHWA Deep Foundation Load Test Database Version 2.0, (Nikolaos Machairas, Gregory A. Highley, M. Iskander, 2018), Transportation Research Record.

Databases For Other Subjects Are To Be Explored By Surveying Academia, State Dots, Practitioners And Consultants, Federal And Local Agencies. The International Society For Soil Mechanics And Geotechnical Engineering Lists Publicly Available Databases (See Https://Www.Issmge.Org/Committees/Technical-Committees/Impact-On-Society/Risk).

Research Objective

The Objective Of This Research Is To Collect And Assemble Quality And Complete National And/Or Regional Axial Load Test Databases In The USA For Small And Large Diameter Deep Foundations (Driven Piles And Drilled Shafts) From The State Departments Of Transportation, The Federal Highway Administration, And Other Sources. This Effort Should Be In Agreement With The Finalization Of The DIGGS Protocol For Load Test Data Entry.

The Databases Will Be Used For Developing Software That Uses Artificial Neural Networks, Support-Vector Machines, Random Forests, And Other ML-Based Techniques That The Research Team Sees Fit For Predicting Deep Foundation Axial Capacity. Additionally, A Wide Search For Available Geotechnical Databases Should Be Performed, And ML-Based Algorithms Should Be Developed For Classical Geotechnical Applications.

Specific Tasks Of The Research To Accomplish The Main Objective Include:

Task 1. Survey Of State Dots For Deep Foundation Load Testing Information

Task 2. Identify And Locate Other Deep Foundation Load Testing Sources

Task 3. Prove Recommendations For The Completion Of The DIGGS Protocol For Load Test Data

Task 4. Develop A Criterion For Completeness And Accuracy Such That The Load Tests Can Be Classified On These Quality Measures

Task 5. Assemble A Quality And Complete Database Which Contains The Load Tests Collected In Task1 And Task 3, Which Meet The Criterion Developed In Task 4.

Task 6. Characterize And Summarize The Information From The Quality And Complete Load Tests Assembled In The Previous Tasks.

Task 7. Develop Software That Uses ML Algorithms For Predicting Deep Foundation Axial Capacity (For Instance:

Artificial Neural Networks, Support-Vector Machines, Random Forests, Bayesian ML, And Other ML-Based Techniques That The Research Team Sees Fit)

Task 8. Explore Among State Dots And Other Sources For Databases That Can Be Assembled To Study Relevant Geotechnical Problems.

Task 9. Develop And Train ML Algorithms Applicable To The Data Collected In Task 8.

Urgency And Potential Benefits

There Is Knowledge And Technology With Tremendous Potential That The Discipline Is Not Currently Taken Advantage Of. ML Can Help Solve Complex Problems In Geotechnical Engineering, As Well As Provide Engineers With New Tools For The Design And Evaluation Of Foundations.

Furthermore, Relatively Few Reliability-Based Resistance Factors For Deep Foundations Have Been Developed By AASHTO And State Dots. This Is Likely Due To The Expense Associated With A Comprehensive Load Testing Program Or A Lack Of Data. Therefore, A Quality And Complete Database Is Most Warranted.

Quality And Complete Databases Would Allow For The Development Of ML Algorithms And Reliability-Based Calibrations, Which Ultimately Lead To Efficient And Safe Geotechnical And Foundation Designs.

Implementation Considerations

The Research Can Be Implemented Through Collaboration Between National And State Transportation Agencies (AASHTO, FHWA, State Dots) And Professional Organizations (ASCE, DFI, ADSC, And PDCA) To Aid In The Expansion Of The Database With Additional Datasets And To Explore For New Relevant Databases.

As Currently, ML Algorithms Are Not In Use By Public Agencies Nor Practitioners, The Implementation Of The Research Results, As New Tools For Analysis Or Design Based On ML, Might Require Application Examples To Facilitate The Transition To These New Modern Alternatives.

List The AASHTO Committee(S) And/Or Council(S) – And Any Other Organization – That Might Be Interested In The Research Results And Could Help Support Implementation.

• Organization, Contact Person, Phone Number And Email Address

TRB Committee On Foundation Of Bridges And Other Structures, (AKG 70)

TRB Committee On Geotechnical Instrumentation And Modeling Committee, (AKG60)

TRB Committee On Soil And Rock Properties And Site Characterization Committee, (AKG20)

California Department Of Transportation (Caltrans)

Recommended Research Funding And Research Period

Research Funding Recommended Funding For The Project Is $600,000.-

Research Period: It Is Estimated That 36 Months Will Be Required To Perform The Research.

Problem Statement Author(S): For Each Author, Provide Their Name, Affiliation, Email Address And Phone.

Dr. Victor Aguilar, Universidad San Sebastián, Concepción, Chile

Phone: +569 76117673, E-Mail: Victor.Aguilar@Uss.Cl

Dr. Naser Abu-Hejleh, P.E., FHWA, US Department Of Transportation,

E-Mail: Naser.Abu-Hejleh@Dot.Gov

Potential Panel Members: For Each Panel Member, Provide Their Name, Affiliation, Email Address And Phone.

• Organization, Contact Person, Phone, And Email Address

If This Problem Statement Is Submitted By An AASHTO Committee Or Council, Please Recommend Committee Or Council Members As Potential Panel Members.

• Member Name, State, AASHTO Committee Or Council, Phone, And Email Address

Person Submitting The Problem Statement: Name, Affiliation, Email Address And Phone.

Provide Contact Information For The Individual Submitting This Problem Statement.

• Name Of Individual : Sharid K. Amiri

• Phone Number : (949)-371-3817

• Email Address: Sharid.Amiri@Dot.Ca.Gov

• Affiliation : California Department Of Transportation (Caltrans)