Study On Feature Optimization Methods And Their Applications To Rolling Element Bearings’ Fault Diagnosis

The booming of manufacturing technology has stimulated people’s demand of safer and more reliable mechanical productions to a lager extend. The thriving of the demand also leads to the increase of modern mechanical equipment’s complexity, precision and integration. In the latest decade, the development of artificial intelligence technology has carved its influence on mechanical equipment and leads to the application of artificial intelligence technique to the field of machinery fault diagnosis, which field has emerged its importance to the reliable of machines. The nature of the mechanical fault diagnosis is recognizing the patterns with a variety of signal analysis and processing techniques, thus it is the application of pattern recognition technology in mechanical fault diagnosis. With the increasing number of complexity machinery, industrial robots and other artificial intelligence individuals, traditional mechanical fault diagnosis methods, which based on time series analysis signal, fail to meet the demand of real-time diagnosis and monitoring machinery in large data environments. Thus the feature-based approach has attracted more and more engineers and researchers. In addition, with the increasing of data and data types, the traditional methods are clearly unable to meet the requirements of advanced equipment’s conditional monitor and fault diagnosis. In the other sides, modern pattern recognition technology is the product of artificial intelligence and is good for big and multi-dimensional data. Therefore, optimizing the feature-based methods used in machinery fault diagnosis will have a broad development and application space. Feature optimization-based approaches not only get rid of the unexplainable problems in machinery signal with traditional signal analysis methods, but also provide a wide space for data fusion and pattern recognition. As the basis for pattern recognition, the structure of features have a significant impact on the pattern recognition methods, therefore, study the optimization methods for feature is significant for this field. Taking these into considerations, the main contents and conclusions dissertation are as follows:(1) From the direction of machinery’s development as well as the trend of artificial intelligence, linked to the importance of the mechanical fault diagnosis technology in modern industrial society, this dissertation describes the background and significance of the proposed topics. Based on the rolling element bearing, which is one of the key components in machines, this dissertation researches the application of feature optimizing methods to the field of mechanical fault diagnosis and introduces the feature optimizing methods by tow aspects, namely feature selection and feature extraction.(2) introduce a wrapper method, namely cosine similarity measure support vector machines(CSMSVM), to eliminate irrelevant or redundant features during classifier construction by introducing the cosine distance into support vector machines(SVM). Traditionally, feature selection approaches typically extract features and learn SVM parameters independently or in the attribute space, which might result in a loss of information related to classification process or lead to the increase of classification error when introduce the kernel SVM. The proposed CSMSVM framework, however, jointly performs feature selection, SVM parameter learning and remove low relevance features by optimizing the shape of an anisotropic RBF kernel in feature space. Moreover, the Bayesian interpretation of the novel methodology reveals its Bayesian character, which builds the proposed method on solid theory foundation, and the iteration algorithm, which is proposed to optimize the feature weight, has achieved to maximize the maximum a posterior(MAP).(3) presents a feature extraction method for fault diagnosis of rolling element bearing. In the proposed algorithm, the wavelet based image fusion technique is utilized,the images for fusion come from the Hankel matrix and the matrix is constructed with the power spectral density(PSD) of the vibration signal. The enhanced feature set is obtained based on the wavelet-based fusion technique. Specifically, we have made two main contributions: 1) we introduce the image fusion concept into mechanical fault diagnosis by using the PSD of the vibration signal to create the image; 2) the relationship of the failure mechanism and the feature enhancement method has been revealed, which explains why the image fusion method has the ability to resist the background noise.(4) Correctness and practicality of the proposed methods have been demonstrated by two experiments. With data from bearing test rig, the Bayesian optimization based feature selection method and wavelet based image fusion methods have proved their power in fault diagnosis for rolling element bearing. In addition, the anti-noise properties of wavelet image fusion method has been proved by adding white noise to the original signal and investigating its fault diagnosis results, verifies the proposed feature extraction method is efficiency even with strong background noise.