Weak Signal Parameter Identification Based On Periodic Motion And Deep Learning

Rolling bearing is the most common mechanical accessory and widely used in all kinds of equipment.And bearing failure is one of the main causes of mechanical equipment failure.The rolling bearing often works in the complex and high-intensity background noise which appears as weak fault and is not easy to be found in the early stage of bearing fault.Therefore,how to extract weak fault signal from vibration signal data under strong background noise is of key significance to prevent serious equipment failure shutdown or major accidents.With the development of nonlinear theory,a nonlinear detection method based on detecting weak signal with nonlinear system is proposed.Due to the influence of noise,the qualitative analysis method is difficult to be used to determine the transition critical point,which affects the identification accuracy of weak signal parameters of early fault.The quantitative analysis method has complex algorithm and large amount of calculation,which is difficult to be applied to the engineering practice of early fault identification.In this thesis,the qualitative analysis method and quantitative analysis method are combined.Based on the principle that the noise has little influence on the phase diagram of periodic motion system,the phase diagram similarity is introduced,the conditions for maintaining the periodic motion are given by using the subharmonic Melnikov function,and the persistence of the phase diagram similarity of periodic motion under the background of strong noise is discussed.On this basis,the phase diagram similarity method is proposed to identify the parameters of weak harmonic signal.Phase trajectory method is used to identify weak periodic impact signal and efficient and accurate phase trajectory library similarity method.The specific contents of this thesis are as follows:Based on the similarity of the phase diagram of the weak harmonic background,the phase diagram of the weak harmonic background is proposed.The key of this method is to construct an appropriate low-frequency weak signal,give the parameter conditions for maintaining the periodic motion by using the subharmonic Melnikov function,reconstruct the constructed low-frequency weak signal and the periodic identification system to obtain the determined reference periodic system,and input the low-frequency weak signal to be measured into the periodic identification system to obtain the system to be identified.The phase diagram similarity MSE is used to determine that the phase diagram of the reference period system and the system to be identified has the highest similarity under the same signal parameters,and the low-frequency weak signal parameters are identified.The simulation results of numerical simulation show that the SNR of the lowest detection signal-to-noise ratio is-80.71 d B.Furthermore,the weak high-frequency signal is difficult to be detected because of its complex and strong noise background.Hence,the characteristics of variable scale periodic system is used to convert large frequency signal into a small frequency signal.And based on the persistence of phase diagram similarity of variable scale periodic system under strong noise background,the variable scale periodic system is combined with phase diagram similarity method to construct the identification process of parameters for identifying high-frequency weak harmonic signals.The numerical simulation results and the early fault diagnosis results of the actual bearing show that the minimum signal-to-noise ratio SNR is-62.38 d B and the accuracy is 97.3%.For weak periodic shock signal,a phase trajectory method is proposed.By studying the influence of the driving force on the periodic motion phase trajectory of Duffing Holmes oscillator system under the action of impact signal,it is concluded that the phase diagram of periodic system under small amplitude harmonic excitation is more sensitive to the excitation of impact signal.Based on the persistence of phase trajectory of periodic motion system under strong noise background,MSE is used to determine the highest similarity of system phase trajectory under the same parameters and identify the parameters of impact signal.The experimental results of numerical simulation verify that the phase trajectory method can be used to identify the parameters of weak impact signal.According to the experimental data of a paroxysmal pulse vibration bearing,the absolute error is6.35 rad/s and the accuracy is 99.0%.In view of the complexity and large amount of calculation of the existing nonlinear detection methods,which is difficult to meet the requirements of "fast" and "accurate" for fault detection methods in engineering practical applications,a fault detection method based on phase diagram similarity method and phase trajectory method is proposed combined with deep learning technology.According to the processing process of deep learning technology for big data,this method establishes the phase trajectory database and its label library,which combines the persistence of the phase diagram similarity of periodic motion under the background of strong noise with the phase trajectory database and its label library.Based on the similarity between the phase trajectory database of the reference periodic system and the phase trajectory of the system to be identified under the same signal parameters through the phase diagram similarity,the weak signal parameters can be identified.A weak signal detection software platform based on phase trajectory library similarity method is built,and the reliability of the software platform based on phase trajectory library similarity method is verified according to the experimental data of a bearing outer ring fault.The results of the absolute error of 30 rad/s and the relative error of 3.0% prove that this method can be used to obtain high detection efficiency.After the phase trajectory library is built,the actual detection only needs about 40 s,which meets the requirements of "fast" and "accurate" in engineering practical application.