Application Research On Compressed Sensing In Health Monitoring Of Rotating Machinery

As an important part of the large-scale equipment,rotating machinery plays an important role in the operating of the equipments effectively and stably,and will be affected in operation safety and effectiveness once faults or failures happened.Therefore,the research on the health management of the rotating machinery is important for the equipments security and personnel safety.With the continuous development of modern science and technology,the equipments become more and more complex.To assess the states of the equipments more accurately,more types and a great number of operational state data should be collected and used.For the devices in long-term operation,the real-time monitoring means that a long uninterrupted continuous sampling,which will produce vast amounts of state data of the equipments.On the one hand,this will bring enormous pressure to the data storage.For the equipments which need to transmit the state data to the ground control center in a long-distance such as fighter jets and helicopters and so on,the huge amounts of the data will burden the transmission heavily due to the limitation of wireless transmission rate and bandwidth.On the other hand,although the increase of the data can provide more status information of the equipments,there is often a lot of redundancy in these data.Therefore,the greater amount of data is obtained,the more redundant the data would be.This will result in a huge waste in storage resources and communication resources.Meanwhile,greater amount of the data will burden the signal processing and increase the computational cost.As a new sensing technology,compressed sensing just needs to collect small amount of compressed measurements while keeping the status information,which could alleviate the burden in mass data storage and transmission.Meanwhile,the status information is contained in the low-dimensional compressed measurements,so the compressed measurements can be used for equipment health monitoring directly.Focusing on these problems mentioned above in rotating machinery health monitoring,the applications of compressed sensing in the health monitoring of the rotating machinery are studied in this thesis,and the main works are as follows:(1)The basic theory of compressed sensing and the sparse representation of the rotating machinery vibration signal are analyzed.Three types of dictionaries on which the vibration signals can be represented sparsely are constructed and analyzed,including the comprehensive dictionaries,the fixed over-complete dictionaries and the over-complete dictionaries based on dictionary learning theory.The performances of these dictionaries in decomposing the vibration signals sparsely are also analyzed.The vibration data compression methods based on these dictionaries are studied and five indicators for evaluating the results of the vibration data compressing are proposed,with which the performances of the dictionaries suitable for the sparse representation of the vibration signals are analyzed in vibration signal compression.(2)The bearing fault detection method and diagnosing method based on the signal sparse decomposition theory are proposed.The bearing fault detection model and diagnosing model based on dictionary learning are built respectively.The methods are validated using the deep groove ball bearing vibration data in motor driven end and the effects of the sparse representation error threshold and the number of atoms involved in signal decomposition to the fault detecting and diagnosing results are analyzed.(3)The mathematical theory of the signal detecting and classification using the compressed measurements directly are analyzed and the corresponding detecting probability and classification probability are calculated.The differences between the energy distributions of the vibration signals with different states in frequency domain are analyzed and a bearing fault detection method based on the energy distributions of the vibration signals just using the compressed measurements directly is presented.The method is validated using the deep groove ball bearing vibration data in motor driven end and the effects of different parameters such as the amount of the compressed measurements to the fault detecting results are analyzed.(4)The over-complete dictionaries on which the vibration signal in different states can be sparsely represented respectively are trained using the vibration signals in the corresponded states.The bearing fault detection method and fault diagnosing method are presented just using the low-dimensional compressed measurements directly based on these over-complete dictionaries and sparse decomposition theory.The effects of different parameters to the fault detecting results and fault diagnosing results such as the amount of the measurements,the sparse representation threshold,the sparsity,the compressed observation ways and so on are analyzed,and the principles in setting these parameters are discussed.(5)The recovery method to the lost vibration data based on the compressed sensing theory is presented.The process of losing the vibration data is converted into the compressed observing process.The corresponded compressed measurements and compressed observation matrix are constructed based on the property in vibration data missing,with which the original completed signal without data missing is reconstructed and the lost data recovery can be achieved.The recovery method is validated using the simulation data and the deep groove ball bearing vibration data in motor driven end respectively,and the effects of different missing ways and different amounts of the missing data to the recovery results are analyzed.(6)A de-nosing method for vibration signals based on compressed sensing theory is proposed.For the noise problem of the rotating machinery vibration data,the compressed sensing process is carried out for the vibration signals contaminated by the noise based on the fact that the vibration signal can be represented sparsely on some dictionary while the noise can't be decomposed sparsely on the same dictionary.Then the de-noising for the vibration signals is achieved in this compressed sensing and signal reconstruction process.The de-nosing method is validated using the simulation gear data and the deep groove ball bearing vibration data in motor driven end respectively,and the effects with different types of dictionaries to the de-noising results are analyzed.In summary,the compressed sensing theory and its applications in rotating machinery health monitoring are studied in this thesis.Three types of dictionaries which can be used in sparse representation of vibration signal are constructed,and the analysis results indicate that these dictionaries could decompose the bearing vibration signals sparsely.The bearing fault detection method and fault diagnosing method based on the signal sparse decomposition theory are proposed and viladated.The mathematical theory of the signal detecting and classification using the compressed measurements directly are analyzed and the corresponding detecting probability and classification probability are calculated.The bearing fault detection method and fault diagnosing method are proposed and validated.The recovery method and the de-noising method for vibration signals are presented,and the proposed methods are validated using the simulation data and the deep groove ball bearing vibration data in motor driven end respectively.