The Influence of Different Distance Sampling Step Sizes on the Accuracy of Neural Network-Based Magnetic Distance Measurement

Achieving high accuracy in neural network-based magnetic distance measurement depends critically on the spatial sampling step sizes of the training data and the number of deployed sensors. This paper systematically investigates how these two distinct factors influence the prediction accuracy of deep neural networks (DNNs). Utilizing Ansys finite element simulations of a permanent magnet translating over a 140 mm range, it is found that finer sampling steps drastically improve DNN precision. Specifically, decreasing the sampling step from 0.5 mm to 0.005 mm for a single-sensor system reduces the mean absolute error (MAE) from 0.9405 mm to 0.0503 mm. Furthermore, expanding the setup to a three-sensor array under identical conditions further suppresses the simulated MAE from 0.2407 mm to 0.0016 mm. Empirical experiments subsequently validated these theoretical findings. Notably, the physical three-sensor setup achieved a minimum MAE of 1.1881 mm at the finest step size, which represents a marked improvement over the 1.6065 mm error observed in the single-sensor system. These results confirm that both a sufficient sample size and multiple sensors enhance distance measurement accuracy.