Research On Remaining Useful Life Prediction Of Rotating Machinery Based On Quantum Deep Neural Networks

Rotating machinery is widely used in metallurgy,coal mine,petrochemical,electric power,nuclear energy and other pillar industries.Large rotating machinery equipment represented by compressor,steam turbine,motor and many other core equipments usually continuously work under continuous operation state of heavy load and high speed,which leads to different forms of failure and affects the normal operation of the equipments.Therefore,it is of great theoretical significance and practical engineering value to use advanced equipment fault maintenance technology for early warning of rotating machinery faults,monitoring the health status of mechanical equipment operation,and ensuring the long-term safe operation of equipment.The current technical field of rotating machinery fault maintenance has been progressed from the passive maintenance mode such as post maintenance and regular maintenance to the active condition-based maintenance.Compared with the previous two maintenance modes,the condition-based maintenance can monitor the failure evolution process and predict RUL on the basis of the actual health condition of the equipment,which can provide supporting decision and early warning information for users and maintainers of equipment.Thus it can better guarantee the safety of the system and the integrity of the equipment.Therefore,the research on the RUL prediction method of rotating machinery has significant implications for ensuring safe operation,saving maintenance cost and avoiding production loss.The data-driven prediction methods are a kind of RUL estimation method which donot need a specific degradation physical model,but utilize a degradation index of the tested equipment or the historical operation data of similar equipment.It has some advantages such as simple modeling,good compatibility and wide application range.However,the existing data-driven prediction methods need to be improved in prediction accuracy and computation efficiency due to their difficulties of model training and the distortion of historical information tracing.Therefore,based on frontier quantum computation and deep learning theory,the RUL prediction methods characterized by their unification of high precision,and calculation efficiency have been studied in this thesis.The main research works of the thesis can be summarized as follows:(1)A novel RUL prediction method of rotating machinery is proposed based on quantum convolutional unit reconstructed recurrent neural network(QCURRNN).Firstly,the degradation features of rotating machinery are extracted by power spectrum entropy.And then these features are input to QCURRNN to accomplish the performance degradation trend prediction of rotating machinery.Finally,the predicted power spectrum entropies are input to the failure probability model to calculate RUL.In the proposed QCURRNN,the previous layer informations are preserved by the reduction of dimension for input data through quantum convolution unit so that the time series which are consistent with the input layer can be reconstructed in the following quantum recurrent neural network.Therefore,the higher RUL prediction accuracy of the proposed method based on QCURRNN is obtained.Besides,the weights of QCURRNN are quickly updated by the back-propagation through time training algorithm with dynamic learning rate,which can improve the convergence speed of the network,accordingly,the higher computational efficiency can be obtained for the proposed RUL prediction method.The example of RUL prediction for rolling bearing demonstrates the validity of the proposed method.(2)In order to further improve the prediction accuracy of the above method,a quantum gated circuit neural network(QGCNN)is proposed to predict the RUL of rotating machinery.Compared with QCURRNN,the quantized update gate and reset gate are constructed to form a gated structure,which can simplify the gated structure,reduce the number of learning parameters and optimize the information transmission mode in QGCNN.The important information is integrated into the hidden layer state through the memory unit and passed to the next memory unit.Therefore,on the premise of ensuring the calculation efficiency,the proposed QGCNN-based prediction method shows higher prediction accuracy in RUL prediction of rolling bearings,which indicates that the QGCNN-based prediction method can be effectively applied to the prediction RUL rotating machinery.(3)Moreover,to further improve the defects of the first two methods in network structure feedback mechanism and global optimization ability and then higher prediction accuracy and desirable efficiency can be obtained in RUL prediction.Another novel prediction model,quantum gene chain coding bidirectional neural network(QGCCBNN)is proposed to predict the RUL of rotating machinery more accurately and efficiently.In the proposed QGCCBNN,the quantum bidirectional transmission mechanism is designed to establish the pre-and post-relationships of time series for readjusting the weight parameters according to the feedback from the output layer,so that higher consistency between the input information and the overall memory of the network can be realized,thus endowing QGCCBNN with better nonlinear approximation ability.Moreover,in order to improve the global optimization ability and convergence speed,the quantum gene chain coding instead of gradient descent method is constructed to transmit and update data,in which the qubit probability amplitude real number coding is adopted and the cosine and sinusoidal qubit probability amplitudes corresponding to the minimum loss function are compared with those of the current time by the phase selection matrix for the directional parallel updating of the weight parameters.On this basis,higher prediction accuracy as well as desirable efficiency can be obtained due to the advantages of QGCCBNN.(4)On the basis of the above researches,a software system for predicting RUL of rotating machinery has been designed in Matlab 2018 software and Python 3.6.The system,which includes modules such as degradation feature extraction,degradation trend prediction,failure probability analysis and RUL calculation,can realize the operations of input data reading,method parameter setting,result display and storage in each module.Thus it has the advantages of good man-machine interaction,simple operation and stable operation and can provide engineering technical support for the RUL prediction of rotating machinery.