On Information Fusion And Bayesian Networks Integrated Fault Diagnosis Theoretical Method And Experiments

With the development of industrial system for the direction of high-speed, high power and high reliability, there are lots of complicated and coupling relationship as well as uncertain elements and information between and among the components. When the fault took place in the system, it can be expressed many kinds of symptoms which is often caused by a variety of faults. Therefore, it's difficult to make an accurate diagnosis by means of a single theory and single information source. Moreover, the theory of fault diagnosis need urgently further refinement and can not meet the practical requirements because of complexity and diversity of industrial systems. The purpose of this paper is to resolve the above problems so that fault diagnosis technology can be better applied to actual industrial systems.Two theory methods that are combined based on information fusion technology and Bayesian network are proposed in this paper on the basis of previous research results. On the one hand, multi-sensor information fusion method can increase the completeness of the fault information to overcome the disadvantages of the single sensor; On the other hand, Bayesian network is one of most effective means which not only can decrease the ambiguity of fault information but also improve the speed of fault diagnosis. At last, the experimental study of hydraulic pump is verified to be valid, which both enriches the theory system of fault diagnosis and strengthens the practical value.The main works in this dissertation are showed as follows:(1)It designs multi-source signal collection methods according to the structural characteristics and working conditions, focusing on the collection process of vibration and noise signals. In addition, another signal processing method is presented which uses band-pass filtering denoising and envelop demodulation. Firstly, wavelet decomposition and reconstruction algorithm process filter denoising by means of wavelet packet. Then, envelope demodulation algorithm based on Hilbert transform is applied for fault signals after they are filtered. Finally, it utilizes the analysis of envelope spectrum of the vibration signal and noise signal to express changeable features of specific frequency components of fault mode.(2)All various of feature parameters of signal amplitude domain are introduced ,especially for pieces of the domain dimensionless parameters indicators. The vibration signal on the cover of pump, for instance, it compares the sensitivity to every fault for non-dimensional indicators and discuss the extract method based on wavelet packet decomposition band energy from the angle of time-frequency domain. Moreover, feature vectors can be extracted from the point of the amplitude domain and time-frequency domain in order to increase the completeness of fault characteristic information, which is very guiding significant for hydraulic pump fault diagnosis in the future.(3)The attribute reduction algorithm based on Rough Set is studied due to redundant phenomena of vector of fault. On the one hand, heuristic attribute reduction algorithm based on the main meta-mode is proposed in order to obtain the best attribute reduction with the idea of methods integration. On the other hand, multi-variable decision tree constructed method is presented resolving the problem of high-dimensions of data.(4)Considering more dimension of eigenvector and relativity between the characteristics, the dimension reduction of decoupling method based on principal component analysis also is researched. Furthermore, fault detection methods of PCA on loose shoe of axial piston pump are brought forward so as to achieve real-time fault diagnosis of hydraulic pump. Ultimately, experimental verification provides theory instruction a reference for multi-fault mode diagnosis.(5)The basic concepts, fusion structure and fusion method of multi-sensor information fusion technology are expatiated in this paper. And, it focuses on Bayesian information fusion algorithm based on parameter estimation. At the same time, for some construction method of concepts and classifier of Bayesian are discussed in detail. Then, aiming at incompleteness and ambiguity of fault characteristics and using the broad definition of multi-sensor information fusion , a diagnosis method of single sensor and multi-characteristics fusion based on vibration diagnosis technology is put forward and to the efficacy of the method can be proved and the assumption of multi-sensor diagnosis is presented furtherly by experiment of fault diagnosis that collects vibration signals at lid on the top of end.(6)The fault diagnosis mode of Bayesian network based on multi-sensor information fusion can be established in line with incompleteness and ambiguity of the characteristics of hydraulic pump. Besides, the program of experimental system based on virtual instrument is put forward in this paper, which selects vibration, noise signal and outlet pressure signal as a monitoring signal at three perpendicular directions. Next, a series of fault samples of artificially simulating piston pump faults, such as: loose-shoe, off-shoe, abrasion of sliding boots, abrasion of the swash plate and center spring failure are collected and in addition, these monitor signals are pre-processed by pre-processing program and then use diagnose mode established by multi-sensor information fusion to test. Then, the effectiveness of this method is verified. Finally, compared with two methods that based on rough set model and principal component analysis model mentioned in this paper, it is proved that this innovative method is more applicable to fault diagnosis of hydraulic pumps...