Physics-informed extreme learning machine (PIELM) for consolidation around an expanded cylindrical cavity

Area of Science:

• Geotechnical Engineering
• Computational Mechanics
• Soil Science

Background:

• Cavity expansion significantly impacts soil consolidation.
• Traditional physics-informed neural networks (PINNs) can be computationally intensive.
• Accurate analysis of excess pore water pressure dissipation is crucial.

Purpose of the Study:

• To introduce a physics-informed extreme learning machine (PIELM) for analyzing soil consolidation post-cavity expansion.
• To develop an efficient computational tool for geotechnical analysis.
• To establish a method for interpreting consolidation coefficients from CPTU data.

Main Methods:

• The PIELM replaces deep neural networks with a single-hidden-layer extreme learning machine (ELM).
• Exact stress invariants define initial conditions for excess water pressure.
• A loss vector combining governing, initial, and boundary conditions is optimized using the least squares method.

Main Results:

• The PIELM demonstrates accurate predictions for consolidation analysis after cavity expansion.
• Excess water pressure dissipation is shown to be highly dependent on its initial distribution and soil mechanics.
• The PIELM provides an efficient means to interpret consolidation coefficients.

Conclusions:

• The proposed PIELM is an effective and efficient tool for analyzing soil consolidation.
• This approach offers a novel method for deriving consolidation coefficients from CPTU measurements.
• Understanding initial excess water pressure distribution is key to predicting consolidation behavior.