Prediction Of Long-term Creep Life Of P91 Steel Based On Artificial Neural Network

P91 steel family is widely used in the main steam pipe,reheater and superheater of ultra-supercritical thermal power unit owing to its good high temperature creep resistance.Under such high temperature and high pressure environment of 600?,30MPa,the microstructure of these components will inevitably deteriorate during the service process,resulting in the degradation of service performance of the steel.In addition,the degradation of the steel reduces the residual life of the material and affects the production plan.Therefore,it is very important to evaluate the service condition of key equipment and accurately predict the creep life of materials.Some of the traditional creep life prediction method are too simple to obtain the accuracy prediction result.Others are too complex to calculate the parameters under various conditions.Artificial neural network?ANN?operates by imitating the mechanism of processing signal of human neural.Its multi-layer and multi-node structure make ANN have powerful ability of nonlinear fitting and extrapolation prediction.Since the artificial neural network has no rigid requirements for the selection of input and output,it has been widely used in the field of material failure prediction.In this paper,P91 steel is taken as the research object.BP artificial neural network is used to predict the long-term creep life of P91 steel at high temperature.The main research contents and results of this paper are as follows:?1?The uniaxial tensile test was carried out on a high temperature creep testing machine to obtain the mechanical properties of P91 steel at 600?,including tensile strength,elastic modulus,yield strength,etc.The creep rupture tests were carried out at 600?,140?,145,150,155,160,165,175,190 MPa respectively to obtain the high temperature creep rupture data with reference to the high temperature uniaxial tensile test results.According to the results of the creep rupture test,the high temperature creep damage samples were obtained by the interrupted creep test at 600?and 165 MPa.?2?By means of metallographic microscope,scanning electron microscope?SEM?,transmission electron microscope?TEM?,X-ray diffraction?XRD?and EDS spectrum,the microstructures of crept P91 steel were characterized and analyzed.According to the analysis results,the relationship between the stress and the change of free dislocation density,subgrain size,MX carbonitrides,M₂₃C₆ carbides and Laves phase size was established and modified.?3?A BP artificial neural network model for creep life prediction is established by taking temperature and stress of NIMS database as inputs and creep rupture time as output.Comparing the prediction results obtained by the traditional time temperature parameter method with BP artificial neural network,it is proved that the BP artificial neural network model has good law learning and extrapolation generalization ability.It is a better prediction method of long-term creep life than the traditional time temperature parameter method.?4?With the help of BP artificial neural network model,five kinds of quantitative values of microstructure including free dislocation density,subgrain size,MX carbonitrides size,M₂₃C₆carbides size,and Laves phase size are taken as inputs,and long-term creep life is taken as outputs.A BP artificial neural network model based on microstructure is established to predict creep life of P91 steel.Compared with the previous BP artificial neural network model,the microstructure-based BP artificial neural network model can even achieve accurate prediction of creep life when extrapolating the creep rupture time to 100000 h.