Study On Servo Motor Gear Fault Detection Of SCR Denitration Catalyst Mixer

Selective Catalytic Reduction(SCR)technology can use denitrification catalysts to chemically react with nitrogen oxides in industrial exhaust gases,converting harmful nitrogen oxides into harmless nitrogen and water vapor,thus reducing the concentration of nitrogen oxides in the air and protecting the environment and human health,and is therefore widely used in the industrial Therefore,it is widely used in the fields of industrial denitrification of waste gas in thermal power enterprises,exhaust gas purification of diesel vehicles,etc.The mixing process for manufacturing SCR denitration catalyst is completed using an internal mixer.The servo motor is one of the key components of the internal mixer,mainly used to control the rotational speed and speed ratio of the internal mixer rotor to ensure accurate mixing and processing of materials.Prolonged operation of a servo motor may cause damage to its gear due to long-term excessive load,resulting in servo motor failure,affecting the normal operation of the production line,and even causing irreversible serious consequences.Therefore,scientific and effective fault diagnosis of servo motor gears is of great significance.Traditional fault diagnosis methods include support vector machine,decision tree,BP neural network,etc.,but the learning depth of these methods is insufficient,which leads to the complexity and poor flexibility of the fault diagnosis process.With the rapid updating of computers,deep learning develops quickly,and Convolutional Neural Network(CNN)can extract good abstract and generalization features from fault data,which is especially suitable for processing complex fault data.Based on this,this paper firstly proposes an improved convolutional neural network fault diagnosis model,which adopts multiple continuous convolutional layers to extract data features,reduces data dimensions through pooling layer,and uses global maximum pooling layer instead of Flatten layer to avoid data feature loss.Finally,the accuracy of the model on UConn gear fault data set reached 98.93% through the classification of Softmax function at the full connection layer.However,gear fault vibration signals are one-dimensional data based on time series,and CNN does not consider the long-term dependence hidden in time series data,which may lead to data loss,anomalies and other problems.Therefore,it is necessary to extract the temporal features of data.Long Short-Term Memory(LSTM)is an improved network of recurrent neural networks with long-term memory function,which can discover the potential relationship between data and time.Based on this,an improved CNN-LSTM gear fault diagnosis model is proposed in this paper.LSTM network mining data timing feature is added to CNN,and Hinge function is used as loss function.The overall accuracy of the model on the gear fault data set of the University of Connecticut can reach 99.64%,which can meet the accuracy and timeliness requirements of gear fault diagnosis.