| 摘要:电阻应变式称重传感器存在严重的非线性误差，直接影响称重结果的准确度。本文首先阐述了称重传感器的非线性误差机理与误差补偿原理，提出了一种基于导数约束的称重传感器非线性误差补偿方法。该方法根据称重传感器输入-输出特性曲线的单调递增性，构造神经网络补偿模型训练的约束条件，完成神经网络优化设计，弥补了因训练样本不足导致的网络泛化误差大的缺陷，同时讨论了惩罚因子对网络性能的影响。实验表明，采用这种基于导数约束神经网络补偿方法（DCNN方法）的称重传感器的非线性误差远小于补偿前的误差；同时当训练样本不足时，DCNN方法比传统训练方法（仅利用数据样本训练神经网络，DINN）具有更好的泛化能力，称重准确度更高。 |
| 关键词: 称重传感器，非线性误差补偿，神经网络，导数约束 |
|
| Method for Compensation of Load Cell’s Nonlinear Error Based on Derivatives Constraints |
|
|
| Abstract:The nonlinear error of the resistance strain gauge load cell has heavy nonlinear error, which will lead to the low accuracy of weighing results. In this paper, the mechanism of the load cell’s nonlinear error is introduced and a method for compensation on the load cell’s nonlinear error based on derivatives constraints neural network(DCNN) is proposed. In this method, the monotonically increasing characteristic of load cell’s input-output function is used to construct the constraint conditions of training and optimizing the error compensation model with neural network, which can decrease the model’s generalization error because of the lack of its training samples. On the other hand, the model’s performance affected by the punishing factor is discussed. The experimental results show that the nonlinear error of load cell with this proposed method is far less than that without compensation, and the DCNN’s generalization ability is more advantageous than the DINN(i.e. training neural network by only using data samples and not any constraint condition), and the weighing results of load cell with DCNN are more accurate. |
| Key words: load cell, nonlinear error compensation, neural network, derivatives constraints |