焊接机器人可靠性数据模拟生成与分布拟合

Simulation generation and distribution fitting of the reliability data of welding robot

针对焊接机器人可靠性实验周期长、成本高、实验数据量少且采集困难的问题,提出在不改变原有数据分布规律的前提下,建立人工神经网络模型,运用径向基函数神经网络对原始数据进行模拟训练和扩充,在一定的误差范围内确定焊接机器人可靠性数据服从对数正态分布。最后通过对比分析拟合误差得出:径向基函数神经网络模型可以很好地对可靠性数据进行训练和扩充,拟合较反向传播神经网络更加精确,对分布模型的参数估计也更加准确。

Aiming at the long reliability experimental period of the welding robot, high cost, small amount of experimental data and difficult data acquisition, it proposes a method under the premise of without changing the original data distribution, establishes ANN (Artificial neural network) model and the radial basis function (RBF) neural network to train and extend the original data, confirms the welding robot reliability data as log-normal distribution model under a certain error range. Comparing the fitting errors, it can train and extend the reliability data well in RBF neural network, the fitting is more accurate than BP neural network. This method is more accurate for estimating characteristic parameters of the reliability distribution model.