Image Based Approximate Bayesian Reverse Method And Its Application In Fiber Metal Laminate Forming

Fiber metal laminate structure(FMLs)is widely used in engineering because of its excellent lightweight performance.Batch forming of FMLs by hot forming technology has high efficiency and good economy,which is an important research direction in the field of FMLs in recent years.However,the structure of FMLs is complex.It is difficult to control the material selection,blank holder force,fiber angle,and other parameters in the hot forming process.It is easy to produce wrinkles,tears,and other defects,which seriously affect the forming quality and restrict the wide application of FMLs hot forming method.The research of traditional deep drawing process and material mainly depends on the experience parameters and test.However,for FMLs hot forming technology,on the one hand,due to the complexity of the process,there is little design experience for reference;on the other hand,due to the complexity of the forming condition,it is difficult to find a single characteristic index as the reverse objective function,so the traditional research methods can not meet the requirements.To solve this problem,an image-based approximate Bayesian inverse method is proposed in this paper.Compared with other mainstream hybrid numerical methods,this method takes the simulation cloud image as the objective function,so it can more comprehensively feedback all kinds of defects in the thermoforming process.In the process of FMLs thermoforming parameters reverse seeking,the simulation cloud images are classified by benchmarking the ideal cloud images,and the feature dimensionality reduction and extraction of the cloud images are completed by the variational self encoder based on the depth neural network.On this basis,the spatial mapping between the design parameters and the hidden variables of the variational self-encoder is constructed by the proxy model method,and the hidden variables are extended to the approximate range Finally,the optimal posterior distribution of design parameters is obtained by approximate Bayesian inverse method.Taking FMLs thermoforming as an example,the material parameters and process parameters are identified by this method.The main contents of this paper are as follows(1)Based on the hot forming system used in this paper,the sandwich structure of FMLs is designed.Combined with the prediction model of high-temperature material properties of composite materials,the parametric hot forming finite element simulation of FMLs is realized in ABAQUS.The Matlab-Abaqus co-simulation iterative model is established,which can better simulate the stress and displacement in the process of composite hot forming,and feedback the wrinkle and fiber tearing that may occur in the process of hot forming,to provide a data basis for reverse research.(2)The simulation cloud image is combined with the reverse height of parameters,and the image is processed based on open CV to obtain the input of the model.To reduce the dimension of cloud images,the cloud image features are extracted into hidden variables by a variational self-encoder.The hidden variables are used as the approximate Bayes measure to realize the data feature extraction and save the calculation cost.(3)Least squares support vector regression(LSSVR)is used to establish the mapping relationship between design parameters and hidden variables to solve the problem that particle generation needs a large number of simulation results in the process of approximate Bayesian solution.The accuracy of the surrogate model and variational self-encoder is tested by structural similarity.(4)To ensure the convergence of the model,ABC-name sampling algorithm is adopted in this paper to ensure the high accuracy of the model and improve the solution speed of the model.The results show that the method can accurately obtain the optimal parameters of FMLs thermoforming,improve the quality of FMLs thermoforming as a whole,and has good feasibility and robustness.