A magnetic inverse problem using neural networks
Yi Huang, Pengfei Yuan · 2005
This paper summaries a recent study on a magnetic inverse problem where an object with magnetic material is illuminated by an external magnetic field. A neural network (NN) is developed as an efficient alternative of solving such an inverse problem. Since training data are needed for the learning process, a forward model is derived to generate the training data. This model is suitable for numerical computation (method of moment). A singularity problem is encountered and resolved. The effects of the number and positions of the sensors on the NN results are studies. The case when noise and measurement errors are present is also investigated. It is demonstrated that the NN constructed using feed-forward back-propagation learning algorithm is efficient and accurate to locate the source elements (magnetization and impressed currents) under the illumination of an external magnetic field for the tested cases.