All Issue

2026 Vol.3, Issue 1

Research Article

Performance Verification of Data-Driven AI Predictors Against the ACI 408R-03 Mean Bond-Strength Equations

ACI 408R-03 평균 부착강도식 대비 데이터 구동형 AI 예측모델의 성능 검증

Ibodullaev Akhrorbek Fayzullo Ugli1
,
Sangwoo Park2
,
Jinho Ahn3
,
Hyundo Yun4
,
Sangyoung Han5*
아흐로르벡 이보둘루에프1
,
박 상우2
,
안 진호3
,
윤 현도4
,
한 상영5*
1Graduate Student, Department of ICT Integrated Ocean Smart City Engineering, Dong-A University, Busan, 49315, Rep. of Korea
2Professor, Department of Civil Engineering and Environmental Sciences, Korea Military Academy, Seoul, 01805, Rep. of Korea
3Assistant Professor, Department of Civil Engineering and Environmental Sciences, Korea Military Academy, Seoul, 01805, Rep. of Korea
4Professor, Department of Architectural Engineering, Chungnam National University, Daejeon, 34134, Rep. of Korea
5Assistant Professor, Department of ICT Integrated Ocean Smart City Engineering, Dong-A University, Busan, 49315, Rep. of Korea
1동아대학교 ICT융합해양스마트시티공학과 대학원생
2육군사관학교 토목·환경학과 교수
3육군사관학교 토목·환경학과 조교수
4충남대학교 건축공학과 교수
5동아대학교 ICT융합해양스마트시티공학과 조교수
*Corresponding Author
31 January 2026. pp. 1-14
PDF XML Citation
Abstract

This study develops and evaluates data-driven AI surrogate models to reduce uncertainty in predicting bond capacity for reinforcement development and lap splices, which is a critical input for protective (blast/impact) design of reinforced concrete (RC) members. Using the ACI 408 experimental database, the proposed models are intended to complement or replace the ACI 408R-03 mean bond-strength equations. The target output is defined as the total force transferred by reinforcement at bond-related failure (Tb = Ab·fs). After data preprocessing, the dataset is split into 80% training and 20% testing, and 5-fold cross-validation with Optuna-based hyperparameter optimization is applied. Six regression algorithms (ANN, RF, GBR, XGBoost, DT, and MLR) are compared under a consistent learning validation protocol. The results indicate that ANN, boosting models (GBR and XGBoost), and an ensemble model (RF) yield tighter clustering around the 1:1 line in prediction–experiment scatter plots, whereas DT and MLR show comparatively lower accuracy. These findings support that bond-capacity prediction is a multivariate nonlinear problem and that data-driven nonlinear learners can complement fixed closed-form equations. In addition, SHAP-based pilot explanations are used to interpret feature contributions and their directions, demonstrating that contribution patterns may differ by algorithm and highlighting the value of interpretability for checking physical plausibility and mitigating overfitting. Because the models are trained and validated within the ACI 408 data range, caution is required for extrapolative applications; future work should include external validation and practical guidelines covering the applicability domain, feature definitions, and data-quality criteria.

Keywords
reinforced concrete
blast and impact load
ACI 408
bond strength
AI

본 연구는 철근콘크리트(RC) 구조물의 방호(폭발·충격) 설계에서 핵심 입력인 정착 및 겹침이음 부착강도 예측의 불확실성을 저감하기 위해, ACI 408 실험 데이터베이스를 기반으로 ACI 408R-03 평균 부착강도 예측식을 대체·보완할 수 있는 데이터 구동형 AI 대체 모델을 구축하고 성능을 비교·검증하였다. 목표 출력은 부착파괴 시 철근이 전달하는 총 힘(Texp = Ab·fs)로 정의하였으며, 데이터 전처리 후 학습 80%, 시험 20% 분할과 5-fold 교차검증, Optuna 기반 하이퍼파라미터 최적화를 적용하였다. ANN, RF, GBR, XGBoost, DT, MLR 등 6개 회귀 알고리즘을 동일한 학습–검증 프로토콜로 비교한 결과, ANN과 부스팅 계열(GBR, XGBoost) 및 앙상블(RF) 모델이 예측–실험 산포에서 1:1 선 주변에 데이터가 상대적으로 밀집하는 높은 적합도를 보였으며, DT와 MLR은 상대적으로 낮은 성능을 나타냈다. 이는 부착강도 예측이 다변수 비선형 문제임을 뒷받침하며, 데이터 기반 비선형 학습기가 고정된 경험식의 한계를 보완할 수 있음을 시사한다. 또한 SHAP 기반 파일럿 해석을 통해 주요 변수의 기여도와 영향 방향을 확인하고, 알고리즘에 따라 기여 패턴이 달라질 수 있음을 보여 과대적합을 점검할 수 있는 해석가능성의 필요성을 시사하였다. 본 모델은 ACI 408 데이터 범위 내에서 학습·검증되었으므로 외삽 조건 적용에는 주의가 필요하며, 향후 외부 데이터셋 기반의 외부 검증과 적용 범위·입력 변수 정의·데이터 품질 기준을 포함하는 실무적 가이드라인 마련이 요구된다.

키워드
철근콘크리트
폭발·충격
ACI 408
부착강도
AI
References
1

ACI Committee 408 (2003) Bond and Development of Straight Reinforcing Bars in Tension (ACI 408R-03). Farmington Hills, MI; American Concrete Institute (ACI).

2

Akiba, T., Sano, S., Yanase, T., Ohta, T., and Koyama, M. (2019) Optuna: A Next-Generation Hyperparameter Optimization Framework. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD ’19), Anchorage, AK, 4 August 2019. New York, NY: Association for Computing Machinery. 2623-2631.

10.1145/3292500.3330701
3

Breiman, L. (2001) Random Forests. Machine Learning 45(1), 5-32.

10.1023/A:1010933404324
4

Breiman, L., Friedman, J. H., Olshen, R. A., and Stone, C. J. (1984) Classification and Regression Trees. Belmont, CA: Wadsworth International Group.

5

Canbay, E. and Frosch, R. J. (2006) Design of Lap-Spliced Bars: Is Simplification Possible?. ACI Structural Journal 103(3), 443-451.

10.14359/15323
6

Chen, T. and Guestrin, C. (2016) XGBoost: A Scalable Tree Boosting System. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD ’16), San Francisco, CA, 13 August 2016. New York, NY: Association for Computing Machinery. 785-794.

10.1145/2939672.2939785
7

Friedman, J. H. (2001) Greedy Function Approximation: A Gradient Boosting Machine. The Annals of Statistics 29(5), 1189-1232.

10.1214/aos/1013203451
8

Lee, M. and Kwak, H.-G. (2018) Blast and Impact Analyses of RC Beams Considering Bond-Slip Effect and Loading History of Constituent Materials. International Journal of Concrete Structures and Materials 12, 32.

10.1186/s40069-018-0244-9
9

Montgomery, D. C., Peck, E. A., and Vining, G. G. (2021) Introduction to Linear Regression Analysis (5th ed.). Hoboken, NJ: John Wiley & Sons.

10

Orangun, C. O., Jirsa, J. O., and Breen, J. E. (1977) A Reevaluation of Test Data on Development Length and Splices. ACI Journal Proceedings 74(3), 114-122.

10.14359/10993
11

Rumelhart, D. E., Hinton, G. E., and Williams, R. J. (1986) Learning Representations by Back-Propagating Errors. Nature 323(6088), 533-536.

10.1038/323533a0
12

Zuo, J. and Darwin, D. (2000) Splice Strength of Conventional and High Relative Rib Area Bars in Normal and High-Strength Concrete. ACI Structural Journal 97(4), 630-641.

10.14359/7428
Information
  • Publisher :Korea Protective Facility Institute
  • Publisher(Ko) :한국방호시설학회
  • Journal Title :Protective Facility
  • Journal Title(Ko) :방호시설
  • Volume : 3
  • No :1
  • Pages :1-14
  • Received Date : 2025-12-09
  • Revised Date : 2025-12-20
  • Accepted Date : 2025-12-24
  • DOI :https://doi.org/10.23310/PF.2026.3.1.001
Journal Informaiton Protective Facility Protective Facility
Online Submission & Review System
  • KISTI Current Status
  • KISTI Cited-by
  • KCI 문헌 유사도 검사
  • orcid
  • open access
  • ccl
Journal Informaiton Journal Informaiton - close