Fault Diagnosis And Life Prediction Of Gearbox Key Components Based On Deep Learning

Gearboxes,a key component of mechanical equipment,provide variable-speed and variable-load functions for system.In the process of operation,different speed,different load and complex working conditions all greatly affect the operating state of the gearbox.Gears and bearings are vulnerable to damage.Because of their complex working state and internal structure,their vibration signals are also very complex.The traditional signal processing methods are difficult to separate the effective information from the fault signals.This paper presents a method for fault diagnosis and life prediction of components in gearbox,which based on the framework of deep learning and combining with traditional fault diagnosis methods.(1)This article proposes a one-dimensional convolutional neural network intelligent diagnosis method with improved Soft Max function.Local mean decomposition decomposes the signal into different components.The components are input into the matrix sample entropy based on the Euclidean distance,and the components that best reflects the characteristics of fault are selected.Finally,those components are used to train the network which can identify the fault of the gearbox.Experiments show this method can successfully identify single faults,combined faults and gearbox faults under different loads.It can effectively solve the problem of difficult signal decomposition of gearbox composite faults.(2)The intelligent fault diagnosis method based on deep learning has achieved high accuracy in the diagnosis of complex gear faults.Eeach model-training requires a large amount of fault data,which makes this method have certain limitations.Therefore,a deep migration model based on the adaptive network of Wasserstein distance is proposed for fault diagnosis.The essence of this method is to train the adaptive network to get the minimum distance between the distribution of source domain and target domain to migrate the fault features.The experimental results show that the method has a high accuracy in bearing fault diagnosis under different working conditions.(3)Signal processing methods have many shortcomings about feature expression in the the fault location and prediction.Due to the small number of lifetime data and inconsistent distribution of data among different bearings,the prediction results are not satisfactory.A new prediction method based on transfer learning includes two stages.In the first stage,the original vibration signal is transformed into Hilbert Yellow and the depth features are extracted.Divide the signals of each bearing into a steady state and a rapidly degraded state.The least squares regression algorithm is used to obtain the RUL prediction model of degraded state.In the second stage,migration feature analysis is used to modulate the characteristics of bearing and the faulty bearing in order.The migration feature is used to predict the life.Experiments prove that the presented method has greatly improved the prediction precision and robustness.In this paper,an intelligent fault diagnosis and life prediction method are proposed,which combine deep learning with traditional signal processing methods.First,a deep learning model is established for fault diagnosis.Then the deep network is migrated for fault diagnosis.Finally,deep features are used for life prediction.The method is effective after model training and experimental verification.The research results are of great significance to the fault diagnosis and life prediction of gears and bearings.