Predicting explosive spalling of concrete containing polypropylene (PP) fibers at elevated temperature is a challenging problem. It is difficult for traditional FEM or DEM methods to tackle this problem at the moment due to unclear mechanism of PP fibers in mitigating explosive spalling and difficulty to measure instantaneous hot permeability of concrete. This paper describes development of two Artificial Neural Network (ANN) models for assessment of explosive spalling risk of concrete. One model (ANN1) was concrete mix-based, the other model (ANN2) was concrete strength-based. A total of 306 and 300 test records collected from literature were used to train ANN1 and ANN2, respectively. Twenty groups of heated tests were conducted on high performance concrete and ultra-high performance concrete containing PP fibers to validate ANN1 and ANN2. The two ANN models were successfully trained and validated, with a prediction accuracy of 100% and 90% for ANN1 and ANN2, respectively. Excellent prediction performance demonstrated the feasibility of ANN models for predicting explosive spalling tendency of PP fiber reinforced concrete.

预测在高温下含有聚丙烯（PP）纤维的混凝土的爆炸剥落是一个难题。目前，由于PP纤维在减轻爆炸剥落中的机理尚不清楚，并且难以测量混凝土的瞬时热导率，因此传统的FEM或DEM方法目前难以解决此问题。本文介绍了两种用于评估混凝土爆炸剥落风险的人工神经网络（ANN）模型的开发。一个模型（ANN1）基于混凝土混合料，另一个模型（ANN2）基于混凝土强度。从文献收集的总共306条和300条测试记录分别用于训练ANN1和ANN2。在包含PP纤维的高性能混凝土和超高性能混凝土上进行了20组加热试验，以验证ANN1和ANN2。这两个ANN模型已成功训练和验证，对于ANN1和ANN2的预测准确性分别为100％和90％。出色的预测性能证明了ANN模型可预测PP纤维增强混凝土的爆炸剥落趋势。