Idea Research Based on Kernel Method in Fault Diagnosis

it is important to reduce keeping costs and hold up unscheduled downtimes for machinery. So knowledge of what, where and how faults occur is very important. In machine rotation and machine learning Fault diagnosis and detection are important rule. In this paper offer a method based on kernel method that using in fault occur. For this reason create kernel by wavelet packet with associate rule mining and information fusion for decision rule. This kernel has best time detection and optimization misclassification. Our proposed data fusion strategies take into account that a support vector machine with multi kernel Wavelet-Entropy by finding the optimal hyper plane with maximal margin. Keywords— Fault Diagnosis, Wavelet Entropy, Information Fusion, kernel method. I. Int roduct ion Numerous studies (both theoretical and empirical) have proved that are effective in achieving improved classification performance for various application problems. The failure of machinery reduces the production rate and increases the costs of production and maintenance [1]. Therefore, it is important to reduce noise and inspected event in machine learning, so knowledge of fault occur is very important. In pattern Recognition, kernel method is a Discriminant-based classification ( with linear discriminate analysis (LDA) whose suppose conditional probability is Gaussian distribution. In large data sets, best selection of kernel is important task. In this paper we offer a new model for fault diagnosis. This research consist of 3 steps for accrue fault diagnosis based on kernel method with best position for kernel. First step is feature extraction based on wavelet packet with associate entropy, in this step input data convert to signal model (feature map) by wavelet, and then data extract with wavelet packet tree and finally in this step select data by max entropy energy. In second step create kernel with Mercel kernel model with Morlet mother wavelet on extract data for classification. In step 3 fused data by kernel fused, in this step selecting best kernel in fusion kernel. Our proposed data fusion strategies take into account that a support vector machine with multi kernel Wavelet-Entropy by finding the optimal hyper plane with maximal margin [2]. In the distributed schemes, the individual data sources are processed separately and modeled by using the Support Vector Machine [3]. Fault diagnosis is to detect, isolate, and assess faults and failures of engine system and its major components. II. Material and Methodology In pattern recognition fault is important rule. A pattern is a set of objects, processes or events which consist of both deterministic and stochastic components [4].Recognition is identification of a pattern as a member of a category that we know or we want learns (in Classification known categories and in Clustering learning categories) [5].Therefore, pattern recognition have 2 section, in pattern section make a category or class of pattern and in section of recognition make a decision about the “category” or “class” of the pattern [figure 1]. Figure 1: fault diagnosis pattern recognition In new method for diagnosis of fault have 3 steps: I. Feature extraction using wavelet packet with associate rule mining II. Kernel method Classification with kernel wavelet III. Fault decision using Information fusion by feature level fusion(kernel fusion) I. Feature extraction based on wavelet packet transform(WPT) with associate rule mining Feature extraction is combining attributes into a new reduced set of feature.in pattern recognition and image processing feature extraction is a reduce dimension in feature space until improve classification. Wavelet transform is powerful than other transform because wavelet transform analyze signal in both time and frequency domain. Selection of suitable wavelet transform for given application is important, wavelet packet transform (WPT) was more suitable for understanding of the time-frequency characteristics. Associate rule mining is a method for detection best relation between variable in large data sets.one of the quantitative measures associated with wavelet packet transform (WPT) is Entropy. Entropy can be an associate for WPT with mathematical rule. Entropy provides valuable information for analyzing non-static signals. For express the signals characteristic many various wavelet entropy presented, these entropies based on different algorithm so they have different essential meaning in application. Wavelet energy entropy is the statistical analysis of signal energy on frequency band and presents the distributing complexity of signal energy in frequency domain. Wavelet energy entropy in this paper used to obtain energy distributing information which useful in decision rule in information fusion. International Journal of Scientific Engineering and Technology (ISSN : 2277-1581) Volume No.3 Issue No.3, pp : 310-315 1 March 2014 IJSET@2014 Page 311 II. Kernel method The second step in fault diagnosis is classification. Support vector machine (SVM) can dodge the problems of over local minimum in the classical study method, and is applied in many classification problems successfully [6]. We assume a training set of N data points k = 1,2,..., N, where is the input data, and is k-th output. The SVM constructs a decision function that is showed by: (2-1) In SVM for the function estimation the following optimization problem can be given [7]: