复合材料螺栓连接强度预测中，深度学习黑箱模型虽然能够快速准确学习映射关系，但缺乏内在物理解释，难以理解模型的内部决策，可信度与普适性差。在融合复材物理规律约束与非线性辨识约束的基础上，提出一种多约束辨识的物理信息神经网络(Multi-Constrain Identification Physics-Informed Neural Network, MCI-PINN)。首先，物理规律的约束使用复材螺栓连接挤压强度工程估算公式；其次，非线性辨识约束以线性、多项式、幂函数、指数、对数等多种函数建立材料参数、力学参数、结构参数等与挤压强度的非线性关系，辨识具有最优精度的映射关系；然后，将物理规律约束与非线性辨识约束以损失函数的形式嵌入神经网络中指导模型训练。在案例验证中开展X850材料两种铺层的单钉连接挤压强度预测，分析结果表明，两种铺层挤压强度的预测误差指标MRE分别为1.24%、1.27%。MCI-PINN在离散程度预测、可解释性、泛化能力等方面，与ANN、PINN相比表现出优异性。
Although deep learning-based black-box models demonstrate high efficiency and accuracy in establishing input-output mappings for composite bolted joint strength prediction, their inherent lack of physical interpretability obscures model decision logic, ulti-mately compromising reliability and generalizability. Based on the fusion of physical law constraints of composite materials and nonlinear Identification constraints, a multi-constraint identification Physics-Informed Neural Network (Multi-Constrain Identifi-cation physics-informed neural Network, MCI-PINN) is proposed. Firstly, the constraints of physical laws apply the engineering estimation formula for the extrusion strength of composite material bolt connections; Secondly, the nonlinear identification con-straint establishes the nonlinear relationship between material parameters, mechanical parameters, structural parameters, etc. and extrusion strength through various functions such as linear, polynomial, power function, exponential, logarithmic, etc., and iden-tifies the mapping relationship with the optimal accuracy; Then, the physical law constraints and nonlinear identification con-straints are embedded in the neural network in the form of loss functions to guide the model training. In the case verification, the single-pin connection extrusion strength prediction of two layers of X850 material was carried out. The analysis results show that the prediction error index MRE of the extrusion strength of the two layers is 1.24% and 1.27% respectively. In terms of discrete-ness prediction, interpretability and generalization ability, MCI-PINN shows superiority compared with ANN and PINN.
[1]山美娟, 赵丽滨.复合材料螺栓连接的增强设计与分析研究进展[J].复合材料学报, 2023, 40(7):3771-3784
[2]董慧民, 李小刚, 马绪强, 等.聚合物基复合材料凸头螺栓连接研究进展[J].北京航空航天大学学报, 2023, 49(4):745-760
[3]豆广征, 刘峰, 王卓煜, 等.复合材料螺栓连接结构力学性能研究进展[J].新技术新工艺, 2022, (9):1-5[J].新技术新工艺, 2022, (9):1-5
[4]孙涛, 周金宇.碳纤维复合材料螺栓连接性能综述[J].现代制造工程, 2018, (9):154-160
[5]梁乐乐, 杨胜春, 宋贵宾, 等.复合材料机械连接强度分析研究[J].塑料工业, 2022, 50(3):90-94
[6]吴义韬, 姚卫星, 沈浩杰.复合材料宏观强度准则预测能力分析[J].复合材料学报, 2015, 32(3):864-873
[7]宁蕙, 谭志勇, 张宏宇.复合材料螺栓连接结构疲劳问题研究进展[J].强度与环境, 2023, 50(3):1-9
[8]TALAEI K T, OULD S H, KAABOUCH N.Deep learning: systematic review, models, challenges, and research directions[J].Neural Computing & Applications, 2023, 35(31):-
[9]闫海, 邓忠民.基于深度学习的短纤维增强聚氨酯复合材料性能预测[J].复合材料学报, 2019, 36(6):1413-1420
[10]刘子琦, 陈刚, SU Zhoucheng, 等.基于细观力学和机器学习的纤维增强复合材料热-力多物理性能预测[J].复合材料学报, 2025, Thermo-mechanical multiphysical properties prediction of fiber-reinforced composites based on micromechanics and machine learning:1-17
[11]彭凡, 邹司农, 任毅如.基于深度学习的复合材料螺栓连接失效预测[J].机械强度, 2023, 45(02):447-453
[12]Yang Z, Yu C H, Guo K, et al.End-to-end deep learn-ing method to predict complete strain and stress ten-sors for complex hierarchical composite microstruc-tures[J].Journal of the Mechanics and Physics of Sol-ids, 2021, 154(15):104506-
[13]REZA S, ANUJ K, MARYAM S.Adata-driven ap-proach to full-field nonlinear stress distribution and failure pattern prediction in composites using deep learning[J].ELSEVIER, 2022, 397:-
[14]ZHANG Y, TINO P, LEONARDIS A, et al.A Survey on Neural Network Interpretability[J].IEEE Transac-tions on Emerging Topics in Computational Intelli-gence, 2021, PP(99):1-17
[15]严如强, 商佐港, 王志颖, 等.可解释人工智能在工业智能诊断中的挑战和机遇: 先验赋能[J].机械工程学报, 2024, 60(12):1-20
[16]WANG W, THAI H T.A physics-informed neural net-work framework for laminated composite plates under bending[J].Thin-Walled Structures, 2025, 210(113014):-
[17]HEDAYATRASA S, FINK O, VAN P W, et al.k-space Physics-informed Neural Network (k-PINN) for Compressed Spectral Mapping and Efficient Inversion of Vibrations in Thin Composite Laminates[J].Me-chanical Systems and Signal Processing, 2025, 223(111920):-
[18]LINGHU J, GAO W, DONG H, et al.Higher-order multi-scale physics-informed neural network (HOMS-PINN) method and its convergence analysis for solving elastic problems of authentic composite materials[J].Journal of Computational and Applied Mathematics, 2025, 456:-
[19]LI Y, WAN D T, WANG Z, et al.Physics-constrained deep learning approach for solving inverse problems in composite laminated plates[J].Composite Structures, 2024, 348:-
[20]YAO L J.WANG J X, CHUAI M Y, et al. Physics-Informed Machine Learning for Loading History De-pendent Fatigue Delamination of Composite Laminates[J].Composites Part A: Applied Science and Manufac-turing, 2024, 187(108474):-
[21]ZHOU K, SUN H T, ENOS R, et al.Harnessing deep learning for physics informed prediction of composite strength with microstructural uncertainties[J].Compu-tational Materials Science, 2021, 197:-
[22]SHEN Y, HAN Z, LIANG Y, et al.Mesh reduction methods for thermoelasticity of laminated composite structures: Study on the B-spline based State Space Fi-nite Element Method and Physics-Informed Neural Networks[J].Engineering Analysis with Boundary El-ements, 2023, 156:475-487
[23]杜轲, 林志鹏, 吴文贤, 等.物理信息神经网络方法求解平面应力问题[J].固体力学学报, 2025, 46(02):230-243
[24]田松岩, 黄鑫格, 段焰辉, 等.面向流体力学的物理神经网络综述[J].计算机应用, 2024, 44(S1):133-141
[25]王飞, 党浩宁, 尚勇, 等.从传统数值方法到神经网络：偏微分方程数值解的演进与展望[J].现代应用物理, 2025, 16(01):41-66
[26]杨显昆, 郑锡涛, 成李南, 等.复合材料层合板单钉双剪连接挤压强度的一种工程估算方法[J].复合材料学报, 2012, 29(06):225-229
CJA