Study On Bearing Weak Fault Diagnosis Combined With Duffing System And Bi-LSTM Network

Bearings are widely used in mechanical equipments.The running state of bearings would affect the running state and the life of the related equipment and the precision of the processed products.A serious failure of the running bearing would lead to unexpected shutdown or accidents.In order to find and diagnose bearing faults as soon as possible,the main research content of this thesis is to diagnose bearing weak faults in early stage.In this thesis,the Holmes-type Duffing system which is insensitive to noise is selected to avoid the influence of noise on the diagnosis of bearing weak fault vibration signals.By calculating the positive and negative properties of the two Lyapunov exponents of the Duffing system,the motion state of the system can be accurately judged.The internal policy force of Duffing system is adjusted and the chaotic threshold is determined by the jump of the system motion state.It is found that the initial phase angle of vibration signal sometimes causes the misjudgment of Duffing system.The signal whose initial angle has been changed by phase shifting method is used to excite the Duffing system respectively to avoid the misjudgment of Duffing system.The time scale of the Duffing system is increased by generalized time scale transformation,which is equivalent to reducing the frequency of the vibration signal.Other properties of the signal remain the same during this process.The Duffing system after generalized time transformation can be used to detect high frequency vibration signals.So far,the improvement of Duffing system is completed,which provides the necessary data for the subsequent fault diagnosis using network.The Bi-LSTM network that can achieve bidirectional propagation is selected to complete the diagnosis of bearing weak fault.The attention mechanism is proposed to optimize the Bi-LSTM network to highlight the characteristic information in bearing vibration signals and improve the effectiveness of using Bi-LSTM network to diagnose bearing faults.The Dropout technology is added to optimize the Bi-LSTM network.The principle of the Dropout is that part of the neurons in the Bi-LSTM network are lost efficacy according to the set probability,so the overfitting problem of the Bi-LSTM network can be avoided and the calculated amount of the Bi-LSTM network can be reduced by the Dropout technology.Four groups of simulated vibration signals with different frequencies are used to excite the Duffing system to make a training set and a test set.The Bi-LSTM network is trained by the training set to diagnose the test set.The weak fault diagnosis method is verified firstly.The simulation results show that the addition of attention mechanism and Dropout technology can effectively improve the accuracy of bearing fault diagnosis using Bi-LSTM networkDuffing system and Bi-LSTM network are combined to diagnose the measured signals.The Bi-LSTM network is used to diagnose the flaw detection test data from Case Western Reserve University and XJTU-SY Bearing Datasets respectively.It is found that the accuracy of the Bi-LSTM network reaches 89.38%and 87.67%respectively.The flaw detection test data from Case Western Reserve University is made into a training set,and the XJTU-SY Rolling Element Bearing Accelerated Life Test Datasets are made into a test set.The Bi-LSTM network is trained by the training set to diagnose the test set.It is found that the accuracy of the Bi-LSTM network reaches 85.87%.Through the orthogonal experiment,the selection of the optimal parameters in the Bi-LSTM network is completed.It is found that the highest accuracy of the Bi-LSTM network in diagnosis of weak bearing faults reaches 91.97%.A method of bearing weak fault diagnosis combining Duffing system and Bi-LSTM network is proposed in this paper,which provides a new idea for diagnosis of bearing faults.