优化的粗神经网络在机械故障中的预测研究

The Prediction Research of Optimized Rough Neural Network in Machinery Faults

为了尽早发现机械故障,做到防患于未然,实施安全生产,在神经网络中引入粗糙集理论和模糊聚类方法,实现建模预测。首先用粗糙集和模糊聚类进行属性约简,去掉冗余的属性。然后根据模糊逻辑规则获取合理的网络输入层、隐含层和输出层,建立优化的粗神经网络预测模型。该模型可以有效地去除神经网络中输入层的冗余神经元,合理的确定隐含层神经元的数目,使神经网络提高了收敛性能,获得更好的非线性逼近能力。应用车床的机械振动采样数据进行仿真实验,结果说明:优化的粗神经网络预测模型,可提取有用信息、简化网络结构,减少训练时间,提高预测精度。在机械振动位移的采样数据预测实验中,取得了良好的效果,对于减少机械故障、实现安全生产、提高经济效益具有重要意义。

In order to find the mechanical faults as soon as possible, to nip them in the bud, and to ensure safety in production, the theory of rough set and fuzzy clustering method in the neural network are introduced to re- alize modeling and forecasting. Firstly, the rough set and the fuzzy clustering are used to achieve attribute reduction and to remove redundancy attributes. Secondly, according to fuzzy logic rules, obtain reasonable network input lay- ers, hidden layers and output layers to establish an optimized rough neural network model. The model can effective- ly remove redundant neurons in the input layers of neural network, reasonably ensure the numbers of neurons in the hidden layers, and improve the convergence performance of the neural network so as to obtain better nonlinear ap- proximation ability. The simulation experiment has been carried out by using the sampling data of lathe＇ s mechani- cal vibration. The results show that the optimized rough neural network prediction model can extract useful informa- tion, simplify the network structure, reduce the training time, and improve the prediction accuracy. In the experi- ment of predicting the sampling data of the mechanical vibration displacement, good results has been achieved, which has great significance to can reduce mechanical faults, ensure safety production and improve economic bene- fit.