Machinery Fault Diagnosis Based On Bionic Pattern Recognition

Fault diagnose technology for rotating machinery has obtained highly important attention. Although a lot of diagnose methods had been widely developed and got great achievements, few people take the incomplete fault information into account. In fact, incomplete information is common in practical applications. In order to propose a new way for machinery fault diagnosis with the incomplete information, a widely interested method called as bionic pattern recognition is used in this thesis.At first, three methods of incomplete information processing are discussed, they are regression imputation, EM(Expectation Maximization) imputation, and MI(Multiple Imputation) base on the Markov Chain Monte Carlo. The influence of three methods to pattern recognition is analyzed. Secondly, the principle of bionic recognition was researched, and two programs such as hypersausages neural network and two-weight neural network are designed, and tested by several classical datasets. The results showed that in most cases, the hypersausages neural network and two-weight neural network recognition results are superior to the traditional BP(Back-Propagation) network and RBF(Radial Basis Function) network in pattern recognition.Finally, the algorithms of bionic pattern recognition are used to machinery faults diagnosis, and the results are were compared with traditional artificial neural network’s. The results showed that the algorithms proposed in this thesis are feasible and efficient. Meanwhile, considering incomplete information under the different absence rate, three imputation methods combined with the bionic pattern recognition are used in this paper. This study can be a good reference of selecting an imputation method in practical applications.