Research On The Monitoring System Of Compressor Operating Status

Any abnormal state of mechanical equipment in operation will cause great loss to the entire production process.As an important equipment in petrochemical industry,the compressor running condition monitoring can not only ensure the safe and reliable operation of equipment,but also effectively avoid major accidents.According to the vibration signal produced in the process of compressor running,this thesis uses the neural network method to monitor the state of compressor.Firstly,the vibration signal of one-dimensional time series is extended to the high-dimensional space by using the phase space reconstruction technique.On the one hand,it can better show the operation characteristics of the whole compressor system.On the other hand,it can provide technical support for the determination of the number of input neurons in neural network,effectively avoid subjectivity and uncertainty caused by the traditional manual random selection process,and improve the accuracy and reliability of monitoring results.-In the traditional Grassberger-Procaccia(G-P)algorithm,scaleless range identification relies on experience too much.In this thesis,a DBSCAN(Density-Based Spatial Clustering of Applications with Noise)clustering algorithm combined with particle swarm optimization(PSO)is proposed to automatically identify scaleless range.According to the data distribution characteristics of the double logarithm curve,DBSCAN algorithm was used to cluster the points on the double logarithm curve and determine the range of the initial solution set.With the maximum correlation index and the minimum residual sum of squares as the objective,an optimization model of scaleless range identification was established.Particle swarm optimization algorithm was used to obtain the optimal solution,and automatic scaleless range identification was realized.Finally,the effectiveness of the proposed method is verified by Lorenz equation experiment,and it is applied to compressor condition monitoring.It is very important to get the predicted value of vibration signal accurately for the early warning signal of compressor abnormal state.For the time series data in actual production,BP(Back Propagation)neural network prediction only depends on the current imput.However,RNN(Recurrent Neural Network)has a strong "memory"function,and the combined effect of current signal input and past output is taken into account,so the prediction result is more accurate.On this basis,RNN prediction model is constructed based on phase space reconstruction for vibration signal prediction in this thesis.It is proved that the prediction accuracy is higher and it is more sensitive to the abnormal state in the process of compressor running.Based on the vibration signal characteristics of compressor running condition monitoring.In view of the traditional neural network can not realize the multidimensional characteristic time series model establishment problem.The multi-level features are learned by using deep belief network(DBN)stack RBM(Restricted Boltzmann Machines),and the model generalization ability is improved through reverse fine-tuning process.The predicted vibration signals are used for phase space reconstruction and the dimensional and dimensionless parameters are extracted for condition monitoring.The experimental results show that different characteristics have different angles to characterize the running state of the mechanical equipment,and the monitoring results are also different.The comprehensive consideration of multi-dimensional characteristic parameters for the condition monitoring can obtain better monitoring results.Finally,based on the research of compressor vibration signal,a condition monitoring system is designed to realize the real-time display of compressor running state.