水化学溶液下灰岩力学特性及神经网络模拟研究

Study on mechanical behavior of limestone and simulation using neural network model under different water-chemical environment

基于水化学溶液浸泡的饱和灰岩三轴压缩试验结果,分析了影响灰岩变形及强度等力学特性的主要因素,认为随溶液酸碱度增加,弹性模量及峰值强度均有降低的趋势;相同pH值,Na2SO4溶液、蒸馏水和CaCl2溶液对灰岩强度的弱化程度依次降低;随着围压的增加,岩石强度逐渐增大,岩石到达峰值强度时的轴向变形也越来越大,塑性变形明显增加。考虑应力路径和化学溶液的影响,利用遗传算法优化神经网络结构,建立了进化神经网络本构模型。通过样本学习模型能较好地描述化学环境下灰岩的力学性能,5种不同化学环境下的力学试验验证了进化神经网络模型模拟结果的可靠性,该方法可以推广到化学环境下其他岩石的力学试验模拟。

Based on triaxial compression tests of saturated limestone soaked by water-chemical solution, main factors affecting mechanical characteristics of limestone such as pH value, ion concentration and confining pressure are analyzed. The tests show that the elastic modulus and the peak strength of saturated limestone trend to decrease with the increase of acidity-alkalinity for the same solution with the same mol concentration; strength of limestone soaked in Na2SO4 solution, distilled water and CaCl2 solution with the same pH value decreases according to priority; the axial strain when the peak strength reaches, the strength of limestone and plastic deformation increase markedly with the increase of confining pressure. Considering the effect of the chemical solution and the stress path for mechanical property of limestone, an implicit modeling method describing mechanical property of limestone using evolutionary neural network is proposed. The neural network constitutive model trained by learning samples, tested by testing samples, and whose structure are searched by genetic algorithm, can well describe the mechanical properties of the limestone and can be used to simulate the mechanical tests in the similar condition. Correction of simulation is proved by 5 mechanical tests under different lab conditions.