[an error occurred while processing this directive]

Coal Engineering ›› 2026, Vol. 58 ›› Issue (7): 203-209.doi: 10.11799/ce202607025

Previous Articles     Next Articles

Research on constitutive model for plastic flow of coal samples based on machine learning#br#

#br#   

  • Received:2026-01-24 Revised:2026-06-14 Online:2026-07-15 Published:2026-08-03
  • Contact: shan zhongwang E-mail:wangzhongshan2026@163.com

Abstract:

To address the issue that coal pillars in the Guotun Coal Mine are prone to plastic flow after excavation-induced yielding, this study aims to reveal the plastic deformation laws and mechanical response characteristics of coal samples under post-peak cyclic loading conditions. Cyclic loading–unloading tests were conducted on coal samples under different confining pressures and loading paths to systematically analyze the axial stress–strain response characteristics after shear yielding. Based on the experimental data, a constitutive model describing the complex plastic behavior of coal samples was constructed by leveraging the excellent generalization capability of machine learning methods. Dynamic elastic modulus and damage evolution parameters were incorporated into the model to characterize the early-stage compaction and failure processes of coal samples observed in the tests. The results show that as the confining pressure increases, the peak strength of coal samples increases from 37.69 MPa to 68.36 MPa. Furthermore, the post-peak residual strength, plastic deformation capacity, and overall deformation resistance all exhibit an increasing trend. During the cyclic loading–unloading stages, the stress–strain curves display a pronounced hysteresis effect. The areas of the first to third hysteresis loops decrease successively, with cumulative attenuation rates ranging from 10.56% to 68.45%, indicating a gradual reduction in plastic dissipation energy. As the confining pressure increases from 2 MPa to 8 MPa, the total hysteresis loop area increases from 0.051 to 0.153, demonstrating a significant enhancement in plastic deformation capacity. Based on the experimental data, a machine learning constitutive model was constructed. The maximum relative error between model predictions and experimental results is 5.97%, validating the rationality of the model. The proposed model can efficiently solve the plastic behavior of coal samples and effectively describe their complex plastic mechanical characteristics, thereby providing a theoretical basis for the stability analysis of coal pillars under similar conditions.

CLC Number: