Theoretical Study And Forecasting On Postfailure Of Coal And Rock

The dissertation has done a systematic study on the coal and cock materials destoies and results forecasting.The paper mainly contains below items: Firstly,the paper considers that strain localization is beginning and guide of coal and cock's destroy.Coal and cock strain localization are discussed in one or two dimensional space under uniaxial stress condition by strain gradient theory.Under one dimensional space, analytic expressions of strain plasticity and the strain localization band width are obtained. Under two dimensional space,the paper sets up plastic strain gradient plasticity increment constructure model of rock based on adding rock internal length parameter in the yielding function.And then, analytic expressions between the strain localization band width and it's inclination angle in plane strain is presented based on von Mises' yielding criterion .Conditions of maximal value and minimal value of the strain localization band width and critical angle of rock destroy are obtained . The conclusions in plane stress and in Drucker-Prager 's yielding criterion are similar with plane strain in von Mises' yielding criterion.And then,the paper sets up rate dependent and gradient-dependent constitutive models due to be short of gradient-dependent constitutive model.Interaction between Different internal length scales for strain localization analysis in two and four orders mixture models at one and two dimensions. Strain localization band width between one and two dimensions is discussed once again.The results forecasting of coal and cock materials destory has been done by BP neural network and application of multilayer perceptrons based on genetic simulated annealing algorithm . Training a feedforward networks by an error backpropagation algorithm is slow and lost in local minima and paralysis.The paper proposes a new asynchronous iterative algorithm and asynchronous iterative algorithm with relaxation factor .The paper points out that asynchronous iterative algorithm with relaxation factor is but good speed and also avoids local minima and paralysis. Their` validities are demonstrated by simulation results. At same time,the paper uses genetic simulated annealing algorithm to decide the best structures and weights in the multilayer perceptrons and give a new method that gets more fitting primitive populationa from BP algorithm. And then, rockburst's mathematical model is put forwards by BP network with three layers structure in adopting 1999—2000's field rockburst datas of Huafeng coal mine of Xinwen's mining industry company . Average error between the network forecast value and testing groups'datas using above two methods are small, the result verified the methods is successsful. At last ,the paper uses Matlab 6.1 to design the rockburst forecasting system of Huafeng coal mine based on above models. The introduce of calculating methods , the calculating evaluation, training systems and predicting systems and so on have been designed . For the purpose of fitting practical cases, the system provide twelve calculating methods based on derivative and inderivative calculating methods in the system to train neural network and prediction.