动态递归小波神经网络在回转窑故障诊断中的应用

Application of Dynamic Recursive Wavelet Neural Network in the Fault Diagnosis of Rotary Kiln

针对回转窑作为一个复杂的非线性系统而难以提取有效故障特征的问题,构造了一种动态递归小波神经网络。采用基于梯度符号变化的变学习率以及引入动量项的算法,以回转窑窑尾温度、分解炉温度、窑尾负压、窑头负压、窑头温度、烧成带温度、窑电流、筒体表面温度8个物理量,作为神经网络的8个输入节点,将小波神经网络理论较好地应用在回转窑故障诊断中。归纳了回转窑的主要故障及现象。采用动态递归小波神经网络,从输出层反馈到输入层形成关联层,以存储上一时刻的输出信息。小波神经网络在网络训练时可以有效地利用输出信号。对回转窑故障数据进行归一化处理并作为网络的输入向量,再用小波函数代替神经网络中的激励函数,以故障序列作为网络的输出向量。试验仿真结果表明,该网络具有较好的故障识别率和时间收敛性能。

The rotary kiln,as a complicated nonlinear system,its fault features are difficult to be extracted.In order to realize thefault diagnosis of rotary kiln,a dynamic recurrent wavelet neural network is constructed.By using the variable learning rate basedon gradient symbol change and introducing the momentum algorithm, with eight of the physical quantities, the kiln endtemperature,calciner temperature,kiln tail negative pressure,kiln head negative pressure,kiln head temperature,firing zonetemperature,kiln current and shell surface temperature as the eight of input nodes of the neural network,the wavelet neuralnetwork theory is applied to the fault diagnosis of rotary kiln. The main faults and their phenomenon of the rotary kiln aresummarized.Dynamic recurrent wavelet neural network is used to form an association layer from the output layer to the input layer,which is used to store the output information of the previous time.The output signal could be effectively used in the networktraining by wavelet neural network.The fault data of the rotary kiln is normalized and used as the input vector of the network,theexcitation function of neural network is replaced by wavelet function,the fault sequence is used as the output vector of the network.The simulation results show good effects of the network in the fault identification accuracy and convergence time.