Impact Reliability Analysis Of Pressure Vessel Based On Bayesian Optimized Neural Network

Lubricating oil cooler is a typical pressure vessel.As an auxiliary device of power system,the safe operation of power system will be directly affected by its impact reliability.Many test results show that in addition to the impact load,the size error and material property changes during the design and manufacturing process will have a significant impact on the impact response of the equipment.Therefore,this paper comprehensively considered many uncertain factors such as load size,geometric size and material parameters,and experiments,numerical simulation and programming were used to analyze the shock resistance reliability and shock resistance reliability parameter sensitivity of pressure vessels,so as to provide theoretical basis and technical guidance for the design of pressure vessels.The main contents are as follows:(1)The quasi-static mechanical properties of Q345 R steel were measured by WDW-20 universal testing machine,and the basic mechanical parameters such as elastic modulus,static yield strength and ultimate tensile strength of Q345 R steel were obtained.The dynamic mechanical properties of Q345 R steel at four high strain rates were analyzed by Split Hopkinson Pressure Bar(SHPB)test system.It was found that the instantaneous stress of the material changed significantly with the increase of strain rate.Based on the analysis results of quasistatic test and SHPB dynamic test,the Johnson-Cook constitutive model of Q345 R was obtained,which laid a foundation for the numerical simulation.(2)Based on the BV043-85 standard,the pressure vessel was loaded,and the impact response characteristics of each part of the pressure vessel were studied.The maximum stress position and stress time history curve of each part of the pressure vessel under impact load were obtained.The variation law of pressure vessel stress with time under impact load was analyzed,and the weak position of each part of the pressure vessel under impact load was determined.On this basis,a series of impact conditions were set to load the pressure vessel,the capacity under impact load was studied,the impact resistance value of the pressure vessel was determined,and the foundation for subsequent impact resistance reliability analysis was laid.(3)Aiming at the problems that surrogate model method based on traditional BP neural network had inadequate fitting accuracy,deficiency of computational efficiency,too long adjustment time,and considering the complexity of load conditions and research objects,this paper put forward an improved surrogate model method that uniform design(UD)method and synthetic minority over-sampling technique(SMOTE)method were used to improve the utilization efficiency of sample points,and Bayesian optimization(BO)algorithm was used to optimize BP neural network.Firstly,the feasibility and superiority of the proposed method were verified by two classical cases.Secondly,the reliability of each part of the pressure vessel under different impact loads was studied by this method,and the change of reliability evaluation index with impact load was given,and the change trend was analyzed.Finally,the influence degree of each random variable on the impact resistance reliability of pressure vessel was studied and compared,and the parameters with the great influence degree were determined,which provided theoretical basis and technical guidance for pressure vessel design.