Fast Topology Optimal Design Of Electromagnetic Inverse Problem Based On Deep Learning

In order to achieve sustainable economic development and make the best use of various resources,when various structures is designed,designers are required to achieve the design objectives through the limited material resources,and then make full use of the properties of the materials themselves.Therefore,in the design of electromagnetic equipment,it is necessary to make a scientific and reasonable allocation of materials in the structure.Traditional structural design methods usually involve designing the initial structure of a product through the designer’s actual experience or the analogy,and then the static and dynamic analysis are carried out on it.If it is not suitable,the design method is modified to re-analyse it until it meets the design requirements.At present,there are two major difficulties in the optimal design of the structure of electromagnetic equipment.On the one hand,it is difficult to set up a design space included all the design variables.On the other hand,the solution of the inverse problem of electromagnetic field mainly applies to many repeated numerical calculations of electromagnetic fields,such as the modeling and simulation of finite element analysis,which is a process that consumes a lot of computational time.Therefore,in order to solve the above difficulties,the main work of this thesis is as follows:Firstly,in this thesis,the topology optimal design is applied at the beginning of the structural design,where the geometry and materials are represented in bitmaps and the entire design area is described as a set of pixels by means of a topological approach.The appearance of the material boundaries and flux barriers are freely modified in the design area,resulting in a new device structure shape.Secondly,a surrogate model is built to evaluate the topology optimisation results in reducing the number of finite element analyses and hence the running time.Deep convolution neural network subclasses have achieved great success in the field of image recognition.In this thesis,the deep convolution neural network is used to address the complexity of physical problems and the overfitting behavior of neural network to obtain the results that satisfy the optimization objectives.The deep convolutional neural network is trained by using the topological information map as the input to the convolutional neural network.Unlike parametric optimisation,topological optimisation does not require the introduction of design parameters to extract the important features of the image.Finally,through the topology optimization analysis of two types of electromagnetic equipment,this thesis verifies the feasibility of topology optimization design based on deep convolution neural network to solve the inverse problem of electromagnetic fields.In this thesis,deep convolution neural network model independent of specific parameters is used to solve different types of problems in electromagnetic fields,which provides a framework for the structural optimization of electromagnetic equipment.