Study On On-Line Monitoring System And Method Of Fault Diagnosis Based On Improved Svdd

As the development of industrial equipment towards large-scale, complexity and automation, analyzing and designing an accurate and effective on-line monitoring system is becoming an important objection in the fault diagnosis science field. And the system should be able to on-line detect the status of the equipments, diagnose faults and update the learning ability. With the equipments running, a great amount of process data can be sampled and collected to store in the database. How to analyze those data and extract the features to improve the diagnosis capability has been becoming one of the focuses in the field of fault diagnosis.This thesis mainly focuses on some problems of on-line monitoring system and training the process data based on SVDD (Support Vector Data Description). The main contributions of this thesis are as follows:(1) To address the problem of the high dimension of the feature space of the equipment status, this thesis proposes a new dimension reduction methods based on cosine similarity. This method reduces the features which are similar to others to improve the orthogonality of the feature space based on cosine similarity. The method reduces the computation cost of SVDD and increases the prediction accuracy;(2) To address the problem of SVDD processing huge samples dataset with high time complexity, this thesis proposes a STING-SVDD (STatistical INformation Grid SVDD). Firstly, the algorithm applies STING division to sample set space; secondly, it rejects some samples based on support vectors distribution characteristics; lastly, it trains simplified sample set to get the SVDD classifier. This method reduces training scales and time without decreasing classification accuracy;(3) To discuss and study the feasibility of SVDD incremental learning algorithm, this thesis proposes a new SVDD incremental learning algorithm based on boost method. This method boosts the erroneous samples into training dataset to reduce training scales and time, and make on-line fault diagnosis possible; (4) On the basis of the above analysis and study, this thesis prepares on-line fault diagnosis software. The software realizes on-line data analysis functionality, simulates on-line incremental learning algorithm and uses the experimental data to verify the validity and utility of the software. The results demonstrate the feasibility of the software based on the simulating experimental device.