Development Of Fault Diagnosis System For Rotating Machinery Rotor In Petrochemica Plant

The growth of the petrochemical industry has promoted the demand for petrochemical equipment.Key equipment in industries,such as centrifugal compressor,is developing towards clustering and intelligence.Once these cluster equipment failures occur,they may cause financial losses and serious casualties.The equipment fault diagnosis systems not only relies heavily on the experience and expert knowledge of technicians,but also has time-consuming judgment process and poor versatility,which can no longer meet the intelligent development of equipment and catch up with the trend of big data era.So a rotor fault diagnosis system is developed in this thesis for intelligent diagnosis,it utilizes the CNN for rotor trouble judgement,which is able to pick up the signal traits of its own accord.Firstly,the vibration principle of rotor is explored and the several common features of traits are analyzed.Secondly,wavelet packet denoising is used to denoise the input signal.The results show that the two indexes of wavelet packet denoising method are better than wavelet denoising,so the former has better de-noising effect.Then,A CNN model suitable for rotor fault diagnosis is proposed.In order to extract signal features more effectively,this model has the special structure of the first and second large convolution kernel and multi-layer small convolution kernel.Lookahead and RAdam are adopted to improve the training process of the model,and restrain the risk of over fitting.Wavelet packet denoising technology is used to reduce the noise of input signal to remove useless information.The normal data,unbalanced data and impact grinding data collected at three speeds are made into fault data set in proportion,which is used for training and verification of network model.In the test,the accuracy of rotor fault diagnosis reaches 100%,and the recognition rate of the model with mixed data set training is 100% under variable speed.Finally,combined with the characteristics of existing rotor test bench and data acquisition card,the trouble judgement system is exploited.The function of each module is devised by using C# in windows environment.Adopting C# and python joint programming in the fault diagnosis module,the CNN model is integrated into the development module to realize the diagnosis function.The main functions of the developed system are tested by using the simulated fault data from the rotor test bench.The results show that each function meets the requirements,and the fault diagnosis function of the system can accurately identify the rotor faults.