Magnetic Dipole Moment Estimation from Nearfield Measurements Using Stochastic Gradient Descent AI Model

With the proliferation of CubeSats, the need to determine the permanent magnetic field of a spacecraft has again become an important aspect of modern aerospace engineering. The spacecraft represents a magnetic multipole system that could be better modeled as a single dipole system. This can be problematic as robust measurement data must be obtained in the nearfield due to the steep roll-off of DC magnetic fields. Thus, making it difficult to determine the effective dipole moment of the system. A method is presented to calculate the magnetic dipole moment of a spacecraft from measured flux density using a machine learning model with stochastic gradient descent. By rephrasing the problem into a machine learning problem, we can take advantage of learning methods used to teach neural networks to generate the solution. The dipole moment is assumed to be the weight of the neural network. The model is trained on the test data until it is overfitting. The solution is then determined to be the weights of the model. For this model, stochastic gradient descent is chosen to be the optimizer. This method shows a high amount of accuracy. The model was able to solve the on-origin problem with an accuracy well above 99%. For the off-origin problem, we can directly measure the mean squared error to determine how well the solution fits the test data. The technique is guaranteed to find minima to the problem. This model can also easily be extended to determine the pole locations and magnitudes in a multipole system where the poles are not located at the origin.