# Nonlinear Compensation of the Linear Variable Differential Transducer Using an Advanced Snake Optimization Integrated with Tangential Functional Link Artificial Neural Network

## TL;DR

This paper introduces a new optimization method to improve the measurement range of a vibration sensor by reducing nonlinear output.

## Contribution

A novel ASO-TFLANN model is proposed to enhance the linear range of LVDT by overcoming overfitting and local optimization issues.

## Key findings

- The ASO-TFLANN model effectively solves overfitting and local optimal problems in training.
- The proposed method expands the linear range of LVDT and improves measurement accuracy.
- Offline and online tests confirm the effectiveness of the ASO-TFLANN model in reducing ϵfs.

## Abstract

The linear variable differential transformer is a key component for measuring vibration noise and active vibration isolation. The nonlinear output associated with increased differential displacement in LVDT constrains the measurement range. To extend the measurement range, this paper proposes an advanced Snake Optimization–Tangential Functional Link Artificial Neural Network (ASO-TFLANN) model to extend the linear range of LVDT. First, the Latin hypercube sampling method and the Levy flight method are introduced into the snake optimization (SO) algorithm, which enhances the global search ability and diversity preservation ability of the SO algorithm and effectively solves the common overfitting and local optimal problems in the training process of the gradient descent method. Second, a voltage–displacement test bench is constructed, collecting the input and output data of the LVDT under four different main excitation conditions. Then, the collected input and output data are fed into the ASO-TFLANN model to determine the optimal weight vectors of the tangential functional link Artificial Neural Network (TFLANN). Finally, by comparing with the simulation experiments of several algorithms, it is proven that the ASO proposed in this paper effectively solves the common overfitting and local optimization problems in the training process of the gradient descent method. On this basis, through offline simulation comparison experiments and online tests, it is proven that the method effectively reduces ϵfs while expanding the linear range of LVDT and significantly improves the measurement range, which provides a reliable basis for improving measurement range and accuracy.