Rule-based determination of effective strain for externally bonded CFRP on beams through hyperbox machine learning modeling

• Main Author: Chua, Alvin B.
• Format: text
• Language: English
• Published: Animo Repository 2022
• Subjects: Carbon fiber-reinforced plastics—Testing; Civil Engineering
• Online Access: https://animorepository.dlsu.edu.ph/etdm_civ/18; https://animorepository.dlsu.edu.ph/cgi/viewcontent.cgi?article=1017&context=etdm_civ
• Institution: De La Salle University

Fiber-reinforced polymers (FRPs) are innovative materials used for the local retrofitting of concrete structures. Carbon FRPs are the most predominant type in such applications accounting for their high strength, remarkable durability, minimal weight, and relative ease in installation. Concerning strengthening of concrete structures, CFRPs can be used for flexural, axial, and shear strengthening. Among these applications, shear design for CFRP utilization is the most critical as it often deals with brittle failure. The shear mechanics of concrete is still an emerging topic due to the numerous parameters involved, many of which are still unknown. Most research about externally bonded FRPs on beams focuses on determining the shear capacity contribution of FRPs, in which a parameter called the effective strain is often used. The effective strain is often limited by the governing failure mode (typically debonding). This study determines the shear capacity sufficiency of externally bonded (side-bonded and U-wrapped) CFRPs on beams using rule-based models obtained by hyperbox machine learning modeling. The database used for this contains at least 500 data points from about 70 experimental programs. An analysis of the data collected identified 7 parameters that are influential in determining the FRP shear contribution, 𝑉𝑓. A total of 8 rule-based models have been obtained for the two CFRP configurations. The S-bonded CFRP model obtained its highest accuracy at 65.38% while the U-wrapped CFRP model obtained 100.0% accuracy during validation. For the best performing models of the two configurations, false-positive occurrences have been minimized (i.e., 𝜔𝑉=0), which are ideal for conservative design purposes. The best models for each CFRP configuration were also compared with fib 14 and ACI 440.2 design guidelines. The rule for S-bonded CFRP was 78% and 65% accurate against fib 14 and ACI 440.2, respectively, while the rule for U-wrapped CFRP was 72% and 76% accurate. This study shows that, although shear mechanics of composite systems have not yet been fully understood, rule-based decision models can be used as a guide especially when brittle failure modes are sought to be minimized.