A Cascaded Classification-Regression Framework for Shear Strength Prediction of Cold-Formed Steel Screw Connections.

Area of Science:

• Structural Engineering
• Materials Science
• Computational Mechanics

Background:

• Existing AISI S100 provisions lack strength equations for screw shear and net section fracture in cold-formed steel (CFS) connections.
• Traditional machine learning (ML) models struggle with imbalanced datasets, hindering accurate prediction of minority failure modes.

Purpose of the Study:

• To propose and evaluate a cascaded ML framework for predicting failure modes and strengths in CFS screw connections.
• To develop a novel Probability-Weighted Cascade (PW-C) strategy to mitigate error propagation from misclassification.

Main Methods:

• A cascaded ML framework involving failure mode classification followed by mode-specific regression was developed.
• Two cascade strategies, Hard Classification Cascade (HC-C) and Probability-Weighted Cascade (PW-C), were evaluated.
• The PW-C model's performance was benchmarked against direct regression and existing AISI S100 provisions, with resistance factor calibration via LRFD.

Main Results:

• The proposed cascaded models significantly outperformed the direct regression model.
• PW-C improved R-squared for screw shear from 0.765 to 0.933 and for net section fracture from 0.784 to 0.912.
• PW-C effectively captured screw group effects and extended coverage to unaddressed failure modes, validated against 564 tests.

Conclusions:

• The PW-C model offers a practical, data-driven approach for enhancing CFS connection design codes.
• A reliability analysis yielded an overall resistance factor (phi_c) of 0.64 for the PW-C model.
• A recommended divisor of 1.15 is proposed for integrating the PW-C model into the AISI design framework.