Research On LLTSA A Data Set Dimensionality Reduction Method In Rotor Fault

With the rapid development of rotating machinery to the direction of exceedingly large ones, it has been widely studied in the field of condition monitoring and fault diagnosing. Using equipment condition in implementation of vibrating signal, which generated during the process of machinery operating, to monitor and diagnose the fault is one of the main measures to ensure its stability, safety and efficient operation. The vibrating information of rotors collected by sensor, converter, the singal acquisition card, etc, which would be saved into the database, thus guaranteeing the correct fault-decision. These datas have the features of large quatity and redundancy, how to remove redundant and irrelevant datas from those datas, which is feature extraction problem in short needs to be resolved in fault diagnosis. The traditional method of linear dimension reduction is difficult to proceed the nonlinear data dimension accurately, therefore, it has become the focus of the current researches to effectively proceed the dimension of nonlinear datas. When the machinery system needs read-time condition monitoring, how to process and analyze the datas correctly is particularly critical. Concering to the above two questions, this study is based on the features of one of manifold learning algorithm’s characteristic named as local tangent space alignment(LTSA), to improve and apply it to data dimension reduction, which has the statistical characteristic. In the process of real-time condition monitoring, the writer is going to analyze and process the new datas correctly through ILTSA algorithm. Which we are improving the accuracy of rotating machinery fault-patten’s recognition, in the meantime, strengthening the correctness of on-line machinery monitoring and diagnosis. The main contents and results of this study are as follows:1) Concering that the original datas can not be directly used in the fault analysis, this research studies the time domain characteristics of main fault analysis. Further, to analyze the suitable methods for nonlinear data dimension on the basis of linear dimension.2) Concering to difficulties of the traditional linear dimension method for effective nonlinear dimension, LTSA algorithm, one of manifold learning algorithms, is proposed, however, its neighbour k value is still difficult to be selected accuratly. The paper is based on the linear block algorithm, trying to locally linearize in a reasonable way, consequently, the effect of dimension reduction will get improved.3) Concering to the problems that how to deal with new data effectively by real-time, on-time monitoring and diagnosing. This paper adopted the incremental manifold learning algorithm to reduce the new data’ dimension of which makes full use of the historical data information, as well as get the access to the new data information, therefore, it provides the reference information to the current situation and future trend of the equipment.4) Combining LabVIEW with MATLAB, fully using the advantages of both, we can successfully embed the LLTSA dimensions reduction algorithm into the hybrid programming which is based on these two kinds of software, thus highlight the effect of analysis and visual effect.The research shows a new direction to the development of machinery fault diagnosis are generation of new data of nonlinear data’s characteristic in machinery system, which is full of significance and potential in research. In the meantime, also providing a mew idea for data-mining and intelligence-diagnosing.