本文通过对石膏角砾岩进行不同围压下的常规三轴压缩试验，单轴以及不同围压下的三轴压缩蠕变试验，研究了石膏角砾岩的瞬时变形及破坏特征，在不同围压及应力水平下的蠕变特征等。研究结果表明，石膏角砾岩流变具有非线性特征。结合偏光显微镜和扫描电子显微镜观测分析，探讨了石膏角砾岩的蠕变机理。 在以上试验的基础上，基于统计损伤理论研究了石膏角砾岩瞬时损伤应力应变本构关系，提出统计损伤应考虑损伤门槛的影响，建立了考虑损伤门槛的统计损伤本构方程，理论计算结果与常规试验结果比较接近。根据石膏角砾岩的蠕变特性，建立了石膏角砾岩的流变组合模型；编写了拟牛顿法拟合程序，对流变模型参数进行拟合，分析了不同应力水平下模型参数变化的规律。在考虑损伤门槛的统计损伤及石膏角砾岩流变组合模型的基础上，应用应变等效原理，建立了考虑瞬时损伤的流变本构模型，理论计算结果与试验结果吻合较好，表明所提出的损伤流变模型能够描述石膏角砾岩的非线性流变。在Kachanov提出的一维蠕变损伤演化方程的基础上，建立了同时考虑瞬时损伤和蠕变损伤的流变本构模型，理论值与试验值基本一致，验证了提出的蠕变损伤演化方程是正确的，所建的损伤流变模型能够较好地描述石膏角砾岩的加速蠕变阶段。 基于人工神经网络在模拟非线性方面所具有的优势，用神经网络对石膏角砾岩的非线性流变特性进行尝试性研究。建立了分别加载下岩石流变的神经网络模型；以西原模型作为考题，验证了神经网络流变模型在描述岩石流变行为方面是可行并有效的。提出了“分段外推”的思想来提高神经网络模型对时间的外推性，同时，为了增加每段外推的稳定性，提出了“初次训练”的方法；以考题为例，验证了所提方法的优越性。对石膏角砾岩的非线性流变进行了模拟，结果显示神经网络流变模型拟合精度较高。 结合有限元法的基本原理，分析了粘弹塑性损伤流变模型和神经网络流变模型的粘性应变计算方法以及相应的有限元求解过程；采用ANSYS进行前处理，利用APDL语言提取计算所需的相关信息，用MATLAB编制了粘弹塑性损伤流变模型及神经网络流变模型的平面应变有限元计算程序。利用编制的程序对石膏角砾岩蠕变试验进行了有限元分析，有限元计算结果与蠕变试验结果吻合的较好，表明采用损伤理论和神经网络理论研究岩石的非线性流变是可行有效的，可应用于实际。
电加热器
In this paper, the instantaneous deformation and failure characteristics, creep characteristics of gypsum breccias under different confining pressures and stress levels are studied by means of conventional triaxial compression tests and uniaxial and triaxial compression creep tests. The results show that gypsum breccias can display the creep properties of nonlinearity and acceleration under certain stress level. On the basis of polarization microscope analysis and scanning electron microscope analysis, the creep mechanism of gypsum breccias is also discussed.Based on statistical damage theory, the instantaneous damage constitutive relationship of gypsum breccias is studied. In consideration of the impact of damage threshold, a statistical damage constitutional equation is established. The theoretical calculated results are basically close to conventional test results. In the light of creep characteristics of gypsum breccias, a multiple rheological mechanics model is set up. With Quasi-Newton fitting program written by the author, the rheological model parameters are fitted in accordance with creep test results and theirs variation regularities are analyzed under different stress levels. Based on the statistical damage evolution equation considering stress damage threshold and the multiple rheological model of gypsum breccias, the damage rheological constitutional model is established from hypothesis of strain equivalence. The calculated results of the model agree well with creep test results. This indicates that the damage rheological constitutional model put forward can effectively describe the nonlinear rheological property of gypsum breccias. On the basis of Kachanov’s one-dimensional creep damage evolution equation, the rheological constitutional model is established which considering instantaneous damage and creep damage. The good agreement of theoretical data with test data shows that the creep damage evolution equation put forward is rational and the damage rheological constitutional model established can describe the accelerated creep stage of gypsum breccias well.Because of artificial neural network’s good simulation of nonlinearity, the nonlinear rheological property of gypsum breccias is studied tentatively through neural network. The neural network model for rheology of rock is established under separate loading. Taking visco-elastoplastic model as an examination question, it is verified that the neural network rheological model is feasible and effective in describing rheological behavior of rock. In order to improve neural network rheological model’s extrapolation ability of time, the idea of "segment extrapolation" is put forward. And in order to raise extrapolation stability of each segment, the method of "primary training" is proposed.