Neural Network-Based Reliability Robust Design Of Mechanical Components

The safety and reliability of structures is one of the major objectives of mechanical structure design. In structural design, uncertain factors are inherent. If the randomness of parameters, such as material quality, load and geometry size and so on, is in neglect, the safety of structure will not be ensured enough, and the product quality will be influenced as well. However, when reliability design and optimization design are united organically, it can help designer to design adaptive tolerance of mechanical structure and control the effect of random parameters to structural safety. Thus, forecasted performance of structures in this way will be more safe and economical, and match the practical case than before. So research of Reliability-based Design Optimization (RBDO) is very necessary and important.The author focuses on the research of mechanical part reliability based on NN in the paper, and the NN with Back Propagation (BP) algorithm is adopted, because which is the most mature and be used widely in practical engineering. In the present study, the use of NN is motivated by the approximation concepts inherent in reliability analysis. The applying of NN brings a new method to structural reliability, and then some difficult problems could be resolved effectively by NN.It is difficult to give out the explicit expression of the ultimate state function in practical engineering, as well as, the defect problem of the stochastic finite element method (SFEM) and the response surface method (RSM), for these case, based on the finite element method (FEM), neural network technique (NN), the reliability-based optimization design theory, the reliability-based sensitivity technique and the robust design approach, a neural network method for reliability-based robust design is proposed in this paper. In this paper, the uniform design method (UDM) is used to gain the sample and design proposal for finite analysis, then the numerical solution of maximum stress is obtained by FEM accurately, establish the NN model of the performance function with the sample data as inputs, then the relationship between the maximum stress and sample data is obtained with NN. Afterwards, the reliability-based robust design information can be obtained by combining the reliability optimization design theory, the reliability sensitivity technique and the robust design approach. This method will enhance the efficiency of computation and save the experimental cost. The Neural Network-Based robust design could approach any function relationship between the structure parameter and structure response accurately making full use of the non-linear mapping ability and generalization ability of the NN. Therefore, it solves the explicit expression of the limit state function which is difficult to obtain in practical engineering, and then the process of reliability-based robust design can be implemented conveniently, with very good engineering practical value.