Cumulative Error Elimination for PMLSM Mover Displacement Measurement Based on BP Neural Network Model and SVD

Abstract

Linear motor position measurement faces serious cumulative error problem under long stroke and high-frequency response, which limits the mover position feedback accuracy. This work proposes a cumulative error elimination method for long-stroke displacement measurement based on BP neural network and singular value decomposition (SVD) filtering. Firstly, based on machine vision technology, an image position measurement model of linear motor is established, followed by theoretical analysis of cumulative errors under long stroke and high-frequency response. Secondly, a BP neural network model considering the cumulative error is constructed to obtain the mover displacement of linear motor with long stroke. To reduce the influence of random initialization of neural network model parameters on the fluctuation range of prediction accuracy, the relationship between maximum prediction absolute error and target accuracy was established to ensure the training time and improve the stability of model prediction accuracy. Subsequently, the Hankle matrix is constructed to filter the prediction results by SVD, which can further reduce the amplitude of cumulative error fluctuation. Finally, a linear motor displacement measurement platform is built. The experimental results demonstrate that compared to other methods, the proposed method can effectively reduce the cumulative error in linear motor displacement measurement, exhibiting high robustness and real-time performance.