Research On Rolling Bearing Fault Pattern Identification Method Based On Ensemble Learning

In order to meet the development needs of intelligent manufacturing,rolling bearing fault diagnosis technology must move in the direction of intelligence.With the expansion of the scale of rotating machinery and equipment fault monitoring,a huge amount of industrial data resources are generated,which pushes the field of rolling bearing fault diagnosis into the era of "big data".In the face of huge data resources,it is necessary to use reasonable and efficient data mining technology to explore the hidden knowledge information,which is of strategic significance for the rapid development of intelligent manufacturing technology of rotating machinery.The key step of rolling bearing fault diagnosis lies in the mode identification of operation status.This thesis is based on the ensemble learning method and focuses on the rolling bearing fault mode identification method.The specific research contents and the research progress achieved are as follows.2(1)Rolling bearing fault data sets often have nonlinear and non-smooth characteristics,and the traditional rotating forest algorithm has a low accuracy rate for fault identification,and a rolling bearing fault id entification method is proposed to improve the rotating forest algorithm for this problem.The method uses CART decision tree as the base classifier and uses kernel principal component analysis to increase the variability among the base classifiers by feat ure transformation of the feature data set,which makes the integrated pattern recognition more effective.The improved rotating forest algorithm is experimentally verified to improve the accuracy of rolling bearing fault identification.(2)The clustering of rotating machinery equipment leads to an increase in the dimensionality of the collected fault data set,and the traditional extreme learning machines cannot obtain the best classification performance when dealing with complex data.An ensemble extreme learning machines fault identification method based on feature selection is proposed to address this problem.The method selects a high-efficiency extreme learning machines as the base classifier and uses multi-scale fuzzy approximation entropy to construct a high-dimensional feature set.The subset of features obtained by feature simplification under different neighborhood radii is used to train the base classifier,resulting in greater variability among base classifiers and shorter training time for individual classifiers.The integration of feature selection into an extreme learning machines can theoretically improve the pattern recognition accuracy,and the classification advantages of this method on high-dimensional datasets are further validated by experiments.(3)In the era of mechanical big data,how to effectively accumulate and utilize data has become the necessary path for the development of intelligent decision-making technology.A rolling bearing fault diagnosis system based on C# and SQL Server is developed for this problem.The system realizes the collection,diagnosis,storage and management of rolling bearing operation status data through hardware facilities and software interface.In the system,an improved rotating forest and an ensemble extreme learning machines pattern recognition method are embedded to achieve intelligent decision making.The feasibility and practicality of this system is verified by a set of double span rotor experimental bench.Ensemble learning has a broad development prospect in the field of rolling bearing fault pattern recognition,which can improve the accuracy and enhance the generalization of the pattern recognition algorithm,provide high-quality algorithm guarantee for the development of fault dia gnosis system,and then can better use the accumulated industrial data resources of the fault diagnosis system to carry out intelligent decision-making technology.