Intelligent Fault Diagnostic Methods For Planetary Gearboxes Under Rotating Speed Variations

Vibration based condition monitoring and fault diagnosis is widely used in automobile,wind energy,mining and etc.It can detect early faults of machine and is easy to be deplored.Planetary gearbox is the key component on the drive train of many heavy machines such as wind turbines,and fault diagnosis for planetary fearbox is an important topic.However,due to the structural complexity and non-stationary working conditions of planetary gearboxes,traditional ways of diagnosis face challenges.Besides,as the amount of industrial data grows rapidly,current manual feature extracting methods are limited.New fault diagnostic techniques are needed.This thesis based on machine learning and deep learning technologies to study the fault diagnosis of planetary gearboxes under complex and time-varying working conditions.Two main focuses of the thesis are non-stationary signal processing and intelligent fault features extraction.We mainly consider the following types of fault in a planetary gearbox: the casing cracks,several common sun gear faults,and rotor cracks.New diagnostic methods are proposed.Generally,the contributions of this research can be summarized as follows:(1).Using sweep excitation and a novel angular resampling method to detect and evaluate cracks in structures.To this end,the computed order tracking technique,which are extensively used in diagnosing rotating machines,is adapted for non-rotating machines.Besides,a window based spectral feature extracting are proposed and validated with Support Vector Machines.Experimental results prove that the proposed method is more effective than harmonic excitation based methods.(2).A new convolutional neural network is designed to take raw vibratory data as inputs for fault diagnostic purposes.As an “end-to-end” solution,fault features are automatically extracted by the neural net and a fault classification can be directly given.The method is validated with an experimental planetary gearbox test rig under large speed fluctuations.(3).Two speed fluctuation scenarios for rotating machines are considered,introducing a domain adaption problem in diagnosing.This paper proposes a novel rotating speed normalization method to mitigate the influence of speed changes,regularizing the intelligent models by normalizing their input data.