马高, 覃春雄, 王瑶. 基于知识和数据联合驱动的全灌浆套筒失效模式预测与解释[J]. 工程力学. DOI: 10.6052/j.issn.1000-4750.2022.05.0477
引用本文: 马高, 覃春雄, 王瑶. 基于知识和数据联合驱动的全灌浆套筒失效模式预测与解释[J]. 工程力学. DOI: 10.6052/j.issn.1000-4750.2022.05.0477
MA Gao, QIN Chun-xiong, WANG Yao. PREDICTION AND INTERPRETATION OF FAILURE MODES OF GROUTED SLEEVE BY COMBINED KNOWLEDGE-DRIVEN AND DATA-DRIVEN METHODOLOGY[J]. Engineering Mechanics. DOI: 10.6052/j.issn.1000-4750.2022.05.0477
Citation: MA Gao, QIN Chun-xiong, WANG Yao. PREDICTION AND INTERPRETATION OF FAILURE MODES OF GROUTED SLEEVE BY COMBINED KNOWLEDGE-DRIVEN AND DATA-DRIVEN METHODOLOGY[J]. Engineering Mechanics. DOI: 10.6052/j.issn.1000-4750.2022.05.0477

基于知识和数据联合驱动的全灌浆套筒失效模式预测与解释

PREDICTION AND INTERPRETATION OF FAILURE MODES OF GROUTED SLEEVE BY COMBINED KNOWLEDGE-DRIVEN AND DATA-DRIVEN METHODOLOGY

  • 摘要: 为精准预测全灌浆套筒失效模式(钢筋拉断、钢筋拔出),该文提出一种基于知识和数据联合驱动的全灌浆套筒失效模式预测方法。根据348例全灌浆套筒试验数据,通过机器学习回归算法与规范公式分别得到全灌浆套筒灌浆连接承载力预测值与钢筋极限抗拉承载力计算值,并基于领域知识建立两者与失效模式关系的特征参数,将其与试验数据组成新的数据库。基于新的数据库,通过机器学习分类算法对全灌浆套筒失效模式自动预测。研究结果表明,随机森林算法对全灌浆套筒灌浆连接承载力与失效模式均具有最好的预测效果。相比其他全灌浆套筒失效模式预测方法,该文提出的知识和数据联合驱动预测方法更具竞争力,总体分类准确率可达到94.3%。结合Shapley additive explanations (SHAP)方法和Partial dependent plot (PDP)方法对预测结果进行解释,审视各参数对失效模式的影响。SHAP与PDP的解释结果都表明钢筋锚固长度与灌浆料抗压强度是对失效模式影响最大的两个试验参数,其中PDP解释结果还给出两者影响失效模式转化的临界值,分别为6.01 d (d为钢筋直径)、80.87 MPa。

     

    Abstract: In order to accurately predict the failure mode (rebar fracture and rebar pull-out) of grouted sleeves, a method driven by knowledge and data is proposed. Based on the test data of 348 cases of monotonically stretched grouted sleeves, the predicted value for grouting connection bearing capacity of grouted sleeve and the calculated value for ultimate tensile bearing capacity of rebar are obtained through machine learning regression algorithm and code formula respectively. The characteristic parameters of the relationship between them and failure mode are established based on domain knowledge, which is combined with the test data to form a new database. The failure mode is automatically predicted by using the new database and machine learning classification algorithm. The results show that the random forest algorithm has the best prediction effect on the bearing capacity and the failure mode. Compared with other prediction methods of the failure mode, the method driven by knowledge and data is more competitive in this paper. The general classification accuracy can reach 94.3%. The prediction results are interpreted by combining the methods of shapley additive explanations (SHAP) and partial dependent plot (PDP), and the influence of various parameters on failure modes is examined. Both the interpretive results of SHAP and PDP indicate that the anchorage length of rebar and the compressive strength of grouting material are the two test parameters that have the greatest influence on the failure mode. The interpretive results of PDP also give their critical values affecting the transformation of failure modes, which are 6.01d (d is the diameter of rebar) and 80.87 MPa, respectively.

     

/

返回文章
返回