Li Ping, Feng Jie, Duan Shiwei
School of Management Science and Engineering, Anhui University of Technology, Maanshan, 243032, China.
School of Mechanical Engineering, Anhui University of Technology, Maanshan, 243032, China.
Sci Rep. 2025 Jul 17;15(1):25990. doi: 10.1038/s41598-025-08867-6.
Fracture damage in steel fiber concrete (SFRC) is a developmental process in which deformation and damage are coupled with each other. In order to accurately identify the high-temperature constitutive model taking into account the damage evolution, a high-temperature constitutive identification model using the Improved Whale Algorithm (IWOA) optimised Long Short-Term Memory (LSTM) neural network is presented. Firstly, the Laplace crossover operator strategy, the optimal neighbourhood perturbation strategy, the adaptive weighting strategy and the updating strategy of the variables helix position are introduced to solve the problems of the Whale Optimisation Algorithm (WOA) in relation to its slow convergence rate and its tendency to fall into the locally optimal solution. The supremacy of the IWOA has been demonstrated by comparing IWOA with WOA, Crown Porcupine Optimisation Algorithm (CPO), Butterfly Optimisation Algorithm (BOA) and Grey Wolf Optimisation Algorithm (GWO) in terms of optimisation search. Secondly, based on the experimental data, LSTM model, WOA-LSTM model and IWOA-LSTM model were established, where the MSE of IWOA-LSTM model was improved by 47.66% and 65.60% compared to WOA-LSTM model as well as LSTM model. Finally, the constitutive identification model of SFRC using the IWOA-LSTM model was applied to decouple the damage and plastic strain by the comparative analysis of the measured curves and the prediction curves without the damage, so that the damage and its evolution law of steel fiber concrete at different temperatures (T = 200 °C, T = 400 °C and T = 520 °C) were obtained. The degree of approximation between the IWOA-LSTM model's prediction and experimental data shows that the trained model has a high learning accuracy and good generalization capability, making it appropriate for use in structural engineering applications.
钢纤维混凝土（SFRC）中的断裂损伤是一个变形与损伤相互耦合的发展过程。为了准确识别考虑损伤演化的高温本构模型，提出了一种采用改进鲸鱼算法（IWOA）优化长短期记忆（LSTM）神经网络的高温本构识别模型。首先，引入拉普拉斯交叉算子策略、最优邻域扰动策略、自适应加权策略和变量螺旋位置更新策略，以解决鲸鱼优化算法（WOA）收敛速度慢和易陷入局部最优解的问题。通过将IWOA与WOA、皇冠豪猪优化算法（CPO）、蝴蝶优化算法（BOA）和灰狼优化算法（GWO）在优化搜索方面进行比较，证明了IWOA的优越性。其次，基于实验数据建立了LSTM模型、WOA-LSTM模型和IWOA-LSTM模型，其中IWOA-LSTM模型的均方误差（MSE）与WOA-LSTM模型和LSTM模型相比分别提高了47.66%和65.60%。最后，将采用IWOA-LSTM模型的SFRC本构识别模型应用于通过对实测曲线和无损预测曲线的对比分析来解耦损伤和塑性应变，从而得到不同温度（T = 200°C、T = 400°C和T = 520°C）下钢纤维混凝土的损伤及其演化规律。IWOA-LSTM模型预测与实验数据之间的近似程度表明，所训练的模型具有较高的学习精度和良好的泛化能力，适用于结构工程应用。
