The Research On Technique Of Multiple Concurrent Fault Diagnosis Based-on Non-dimension Immune Detectors

It's prevalent for rotor system of rotary machine to exist concurrent fault at actual project. At present, the investigations about the composite fault diagnosis technology of rotary mainly concentrate into the single fault diagnosis (SFD), and the studies about the multiple concurrent fault diagnosis (M-CFD) technology are at the fledgling stage because of the complex mechanism of the composite fault.The immune system is a complex study of distributed information processing systems that have kinds of function such as immune protection, immune tolerance, immunologic memory, the immune surveillance, particularly have the characteristics of strong adaptability, diversity, learning, memory and identify. Various artificial intelligence methods based on immune mechanism gained the combination between functions and characteristics to solve the massive nonlinear scientific problems. The information processing mechanism of immune system has important theoretical and practical value in the application of fault diagnosis. Support Vector Machine (SVM) is a new machine learning method, which developed on the basis of a limited sample of statistical learning theory. SVM can solve practical problems such as the small sample, on-linear and high-dimensional pattern recognition, and overcome the deficiencies of the neural network learning methods such as difficult to determine network structure, slow convergence local minima, over learning, not enough learning and training needs of large amounts of data samples. SVM has good performance to generalize, and become a hot spot of study succeeding the research of neural network.The main work and innovation of this paper has done as follows:(1) The more studies are made on the mechanism of multiple concurrent faults. According as the non-dimension parameter in time field is not hypersensitive to multiple concurrent fault diagnosis, a generally applicable laws between multiple concurrent fault and some single faults what combine and form M-CF, has been explored. The correlation coefficient of the two has been defined over again, then its suitable parameter and scope are indicated, which offers the new idea when applying the non-dimension Parameter in the effective diagnosis of multiple fault.(2) Aiming at the difficulties of the diversity and the filtering redundant data of immune algorithm and using the idea of artificial immune system clone selection combining with the theory of the immune network, a Novel Immune Network Learning Algorithm (NINL) is proposed. Antibody suppression is firstly introduced for the elimination of redundant antibodies in the progress of generating beginning antibodies. In addition to the velocity is re-defended, which results into the increasing pace that antibody approach to antigen. The result of the trial, which the algorithm is applied the fault diagnosis of machine unit, indicate validity of the algorithm.(3) In this paper, according to the definition of the relationship between multiple concurrent fault and some single faults, based on the non-dimension parameter, FH-SVM algorithm is proposed aiming at multiple concurrent failures. Fuzzy clustering is introduced to the hierarchical SVMs. The algorithm of fusing with the novel immune network learning algorithm and FH-SVM to diagnosis multiple concurrent fault, test results show that the algorithm with the effectiveness of fault diagnosis.