Research On Quantum Neural Computing Technology For Fault Diagnosis Of Rotating Machinery

Quantum neural computation, which is based on the combination of classical neural computation and quantum computation, is a foreland paradigm with very high value for theoretic study and applicable potential. On the basis of quantum computation principle, the occurrence and the characteristic of quantum neural computation are discussed in the paper. In theory, several quantum neural network models, for instance quantum neural network based on multilevel transfer function, quantum neural network based on universal quantum gate, fuzzy quantum neural network and quantum neural network based on many universes, and their application to fault diagnosis for rotating machinery are specially studied. And in practice, a network system of online monitoring and intelligent fault diagnosis for large-scale rotating machinery is developed according to the requirements of some iron and steel enterprise.The work in this thesis is supported by National Science & Technology Development Program'Network system of fault intelligent diagnosis for machinery'and relative horizontal projects. The concretely works are listed as follows:①The goals and the meaning of this research are presented. And a comprehensive survey of the characteristics of vibration faults of rotating machinery, the present research and the research methods of online monitoring and fault diagnosis, the development of neural network and fuzzy set theory and their application in the fault diagnosis are summarized.②Firstly, quantum theory and quantum computation principle are presented in a systemic way. And then the reason of the occurrence of quantum neural computation is introduced with the summary of its research status and level. In principle, corresponding concepts of quantum computation and neural computation, quantum expanding methods of neural computation and the capability of quantum neural computation are analyzed. Finally, several possible models of quantum neural network are also presented.③The prediction of machinery condition based on quantum neural networks is studied. 1) As far as the nonstationary during the long period operation of machinery is concerned, the application of quantum neural network based on multilevel transfer function to the prediction of nonstationary time series is studied. It avoids complex data pre-processing, model recognition etc in traditional neural network and time series analysis. A forecast instance indicates that the quantum neural network can forecast with higher accuracy than traditional time series analysis. 2) On the basis of regarding universal quantum gate as a transfer function, a quantum neural network based on universal quantum gate is constructed by expanding Qubit, phase-shift gate and controlled-NOT gate to complex number field. Application to single-step and multi-step ahead prediction of the vibration time series of a rotating machinery indicates that the performance of quantum neural network based on universal quantum gate is more appropriate for prediction of time series than BP neural network with much higher training speed and accuracy.④On the aspect of fault diagnosis, fuzzy quantum neural network and quantum neural network based on many universes and its application to fault diagnosis for rotating machinery are specially studied. 1) In order to decrease the fuzziness of fault class boundaries and the diagnostic uncertainty of overlapping classes, a fuzzy quantum neural network based on multilevel transfer function is suggested in the fault diagnosis of rotating machinery at the low level combination of fuzzy set theory and quantum neural network. Theoretics and practice prove that the method effectively improves the precision and reliability of breakdown diagnosis and provides an effective way of fault diagnosis for rotating machinery. 2) The idea of multilevel transfer function is introduced into fuzzy membership function, and the new membership function is called quantum membership function. Then, a diagnosis model of quantum neuro-fuzzy inference system is brought forward with a way of fuzzy c-means clustering algorithm and cluster validity functions to ascertain the number of quantum interval. Compared with Anfis, RBFand BP neural network, the model is provided with the features of fast convergence and high diagnosis accuracy. 3) According to the viewpoint of many universes in quantum theory, a fault diagnosis model of fuzzy quantum neural network for rotating machinery based on many universes is proposed. Fuzzy c-means clustering algorithm and cluster validity functions are introduced into the collapse rules of the quantum neural network based on many universes. As a result, the network can collapse to multiple universes when multiple faults emerge, i.e. the network can not only diagnose a single fault but also diagnose multiple faults effectively. And high adaptability, high anti-jamming, high expansibility, excellent convergence rate and the potential of eliminating catastrophic forgetting are the merits of the network proposed. 4) According to fault characteristics of hierarchy, correlation and uncertainty, and symptom characteristics of diversity, fuzziness and multivocality, a multiple-symptom integrated diagnosis network model based on fuzzy quantum neural network is put forward with the way of concrete implementation. On the Basis of the symptom characteristics of rotating machinery, the model syncretizes fuzzy quantum neural network, quantum neural network base on many universes and reverse reasoning strategy based on rules. And the whole diagnosis process is separated into three sub-processes of primary diagnosis, detailed diagnosis and precise diagnosis. Application results show that the model has a high value for theoretic study and high applicable significance.⑤According to the requirements in engineering application, a network system of online monitoring and intelligent fault diagnosis for large-scale rotating machinery based on a mixed structure of C/S and B/S is developed. The system integrates local online monitoring and diagnosis, remote monitoring and diagnosis, and remote diagnosis center. Concerning the characteristics of the vibration in rotating machinery, complete functions of signal analysis are provided by the system. Based on the study and summary of many references, the modules of auto diagnosis and human-computer co-diagnosis are designed. And the network system offers potent support for the equipments to run effectively, safely and economically.