NC Machine Tools Fault Diagnosis Based on Kernel PCA and k -Nearest Neighbor Using Vibration Signals

Abstract

This paper focuses on the fault diagnosis for NC machine tools and puts forward a fault diagnosis method based on kernel principal component analysis (KPCA) and k -nearest neighbor ( k NN). A data-dependent KPCA based on covariance matrix of sample data is designed to overcome the subjectivity in parameter selection of kernel function and is used to transform original high-dimensional data into low-dimensional manifold feature space with the intrinsic dimensionality. The k NN method is modified to adapt the fault diagnosis of tools that can determine thresholds of multifault classes and is applied to detect potential faults. An experimental analysis in NC milling machine tools is developed; the testing result shows that the proposed method is outperforming compared to the other two methods in tool fault diagnosis.

Academic Editor:Changjun Zheng

1, College of Mechanical Engineering of Wenzhou University, Wenzhou 325035, China

Received 13 January 2015; Accepted 15 February 2015; 25 October 2015

This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction

Numerical control (NC) machines are some of the most common classes of machining equipment and play important roles in industry relying on their high automation and stability. However, NC machines generally are subject to failures by inner flimsy tools in practice, which may decrease machinery service performance such as the manufacturing quality and operation safety. Therefore, the demand to fault diagnosis for NC machine tools accurately and reliably has attracted considerable interests. However, it is a challenge to develop and adopt effective signal processing techniques that can discover the crucial damage information from responsive signals.

Tool's damage, such as wear and gap, will occur on the tool's face contacting with the chip or on the flank due to the friction between the tool and the workpiece material. A commonly used effective method is analyzing signals gathered by sensors to reveal fault features with the help of effective signal processing techniques. Traditional signal processing techniques in the fault diagnosis include time-domain analysis [1], frequency-domain analysis [2], time-frequency analysis [3], and wavelet analysis [4]. Wavelet analysis method can be viewed as an extension of the conventional spectral technique by an adjustable window size and adds a scale variable through a wavelet basis function with the inner product transform. So far, wavelet analysis has been widely used in machinery fault diagnosis, such as rotor [5], bearing [6], and gear [7].

However, how to choose a suitable wavelet basis function to match the machine signal structure remains an open issue despite many current alternative types. There is still a lack of generally accepted effective methods to solve the issue [8]. Although some authors provided multiwavelet transform for fault diagnosis [9, 10], these methods also determined the wavelet basis functions previously. In this paper, for overcoming deficiencies of above methods, a fault diagnosis method based on kernel principal component analysis (KPCA) and [figure omitted; refer to PDF] -nearest neighbor ( [figure omitted; refer to PDF] NN) is put forward. KPCA is used to obtain independent local relevance in parameters. Firstly, the original data set transformed high-dimensional data into low-dimensional manifold feature space with the intrinsic dimensionality by KPCA, and the kernel function of KPCA is constructed with covariance matrix of sample data to avoid the subjectivity of choosing kernel function. The [figure omitted; refer to PDF] NN technique is applied to low-dimensional manifold feature space to detect potential faults.

The rest of this paper is organized as follows. Section 2 discusses the selection of impressible statistical parameters in time and frequency domains. Section 3 reviews the standard KPCA method and provides a data-dependent kernel PCA method. Section 4 describes the [figure omitted; refer to PDF] NN rule and improves it to adapt the situation of multifault classes. Section 5 presents the novel fault diagnosis method of KPCA- [figure omitted; refer to PDF] NN. In Section 6, the proposed method is applied to experimental study, and the fault diagnosis results are compared with other approaches. The conclusions of this work are summarized in Section 8.

2. Statistical Feature Parameters in Time-Frequency Domain with Tool Vibration Signal

In the manufacturing process, tool vibration is caused by a variety of factors that are main reasons for tool damages, such as cutting force, built up edge, material inhomogeneity, and ambient conditions. Hence, using the vibration signals to identify tool's damage is reasonable [11, 12]. Both amplitude and distribution of the vibration signals change along with tools' state from sharp to worn. Previous studies [11-13] have shown that some time and frequency domain features can indicate tool's damage. Although these mentioned features are indicative of tool's fault from different aspects, they have different importance. In addition, considering that the dimensions of these parameters are different, this paper employs dimensionless feature parameters to analysis. According to previous literatures [14] and our experimental studies, eight impressible dimensionless statistical feature parameters in time and frequency domains are chosen. Brief mathematical descriptions of these features are shown in Table 1.

Table 1: Dimensionless statistical feature parameters.

Domain	Parameter	Formula	Remark
Time	Waveform	[figure omitted; refer to PDF]	[figure omitted; refer to PDF] ,
Peak	[figure omitted; refer to PDF]	[figure omitted; refer to PDF] ,
Kurtosis	[figure omitted; refer to PDF]	[figure omitted; refer to PDF]
Where [figure omitted; refer to PDF] is a signal series for [figure omitted; refer to PDF] = 1, 2, ..., [figure omitted; refer to PDF] , [figure omitted; refer to PDF] is the number of data points
Frequency	Stabilization ratio	[figure omitted; refer to PDF]	[figure omitted; refer to PDF] ,
Wave-height ratio	[figure omitted; refer to PDF]
Power spectrum standard deviation	[figure omitted; refer to PDF]	[figure omitted; refer to PDF] ,
Frequency high-low ratio	[figure omitted; refer to PDF]	[figure omitted; refer to PDF] ,
Degree of irregular	[figure omitted; refer to PDF]	[figure omitted; refer to PDF]
Where [figure omitted; refer to PDF] is the frequency signal with [figure omitted; refer to PDF] by Fast Fourier Transform (FFT) and [figure omitted; refer to PDF] is the power spectrum of [figure omitted; refer to PDF]

3. Data-Dependent Kernel Principal Component Analysis (KPCA)

3.1. Kernel Principal Component Analysis

Principal component analysis (PCA) has played a significant role in dimensionality reduction, noise removal, and feature extraction from the original data set as a preprocessing step in data mining. It is readily implemented and it is effective when dealing with common cases encountered frequently. However, for complicated cases in industrial practice, especially nonlinearity, PCA is unsuccessful as it is linear by nature. A modified PCA technique, kernel principal component analysis (KPCA) [15] has been proposed to tackle the nonlinear problems in recent years. As a nonlinear form of PCA, KPCA can efficiently compute principal components in a high-dimensional feature space by the use of integral operators and nonlinear kernel functions [16].

Let [figure omitted; refer to PDF] , [figure omitted; refer to PDF] be the [figure omitted; refer to PDF] training samples for KPCA learning. The basic idea of KPCA is first to define a nonlinear map [figure omitted; refer to PDF] : [figure omitted; refer to PDF] , [figure omitted; refer to PDF] which represents a high-dimensional feature space [figure omitted; refer to PDF] and then to apply PCA to the data in space [figure omitted; refer to PDF] . However, it is difficult to do so directly because the dimension of the feature space [figure omitted; refer to PDF] can be arbitrarily large or even infinite. In implementation, the implicit feature vector in [figure omitted; refer to PDF] does not need to be computed explicitly, while it is just done by computing the inner product of two vectors in [figure omitted; refer to PDF] with a kernel function.

The mapping of [figure omitted; refer to PDF] is simply noted as [figure omitted; refer to PDF] . Principal components may be extracted by solving the eigenvalue equation of the sample covariance [figure omitted; refer to PDF] : [figure omitted; refer to PDF] . By defining the kernel matrix [figure omitted; refer to PDF] of dimension [figure omitted; refer to PDF] such that [figure omitted; refer to PDF] Here, [figure omitted; refer to PDF] is the calculation of the inner product of two vectors in [figure omitted; refer to PDF] with a kernel function. The eigenvalue problem may be put in the form [figure omitted; refer to PDF] , where [figure omitted; refer to PDF] identifies the eigenvector [figure omitted; refer to PDF] after normalization. Centralize [figure omitted; refer to PDF] by [figure omitted; refer to PDF] Here, matrix [figure omitted; refer to PDF] .

The eigenvectors identified in the feature space [figure omitted; refer to PDF] can be considered as kernel principal components (KPCs) which characterize the dynamical system. Note that, since the number of eigenvectors (i.e., nonlinear PCs) is the same as the number of samples, it is higher than the number of (linear) PCs given by PCA. The KPCA method is termed "nonlinear" since the feature mapping in the space [figure omitted; refer to PDF] is achieved by a nonlinear function.

3.2. Data-Dependent Kernel Function

For a given data set, the key step is to select a corresponding optimal kernel function in order to obtain the optimal nonlinear feature transformation. There are several kernel functions can be chosen, such as Gaussian, Quadratic, Radial basis function, and polynomial. It is worth noting that the above kernel functions give different results if inappropriate parameters are chosen. However the selection of parameters in these kernel functions is arbitrary and does not have theoretical rules for reference. Since the geometric structure of data in the kernel feature space [figure omitted; refer to PDF] is determined by the kernel function, it is desirable to choose the target kernel function to be data-dependent.

In this paper, a data-dependent kernel based on the Gaussian kernel is designed. The Gaussian kernel function is [figure omitted; refer to PDF] where [figure omitted; refer to PDF] is a measure of the data spread, and it implicitly assumes the same variance for each entry of the sample vector (i.e., the data is isotropic), which cannot be satisfied in many practice situations.

We consider the generalized Gaussian kernel: [figure omitted; refer to PDF] where [figure omitted; refer to PDF] is a transformation matrix. In cases where certain input transformations are known to leave function values unchanged, the use of [figure omitted; refer to PDF] can also allow such invariance to be incorporated into the kernel function.

A choice for [figure omitted; refer to PDF] is the [figure omitted; refer to PDF] covariance matrix: [figure omitted; refer to PDF] where [figure omitted; refer to PDF] is the data vector and [figure omitted; refer to PDF] is the corresponding mean vector. By assuming that the entries of the sample vector are independent, it can be simplified by dropping the off-diagonal elements as [figure omitted; refer to PDF] Here, [figure omitted; refer to PDF] extracts the diagonal elements of covariance matrix [figure omitted; refer to PDF] , which are then used to construct a diagonal matrix by [figure omitted; refer to PDF] , and the obtained kernel by this matrix [figure omitted; refer to PDF] is called independent DDK [17].

While (3) implicitly assumes all sample vectors have the same variance, this is not necessary for the generalized Gaussian kernel. And if all sample vectors have the same variance [figure omitted; refer to PDF] , then [figure omitted; refer to PDF] and the generalized Gaussian kernel (4) will reduce to the conventional Gaussian kernel (3). Using the kernel function (4) with covariance matrix of sample data, the local geometric structure of data can remain in the kernel feature space [figure omitted; refer to PDF] .

3.3. Algorithm of Data-Dependent KPCA

The description of feature extraction process of NC tools with data-dependent KPCA is as follows.

(1) Gain [figure omitted; refer to PDF] groups vibration time-domain signals data of NC machine, and transform every group data to frequency-domain signals by FFT.

(2) Calculate eight feature parameters [figure omitted; refer to PDF] of each group, respectively, to compose observed data set [figure omitted; refer to PDF] , define the average and standard deviation of column vectors as [figure omitted; refer to PDF] , [figure omitted; refer to PDF] , and normalize data set [figure omitted; refer to PDF] .

(3) Reduce dimension of [figure omitted; refer to PDF] by KPCA described in Section 3. In the operation of KPCA, take (4) as the kernel function and the covariance matrix of [figure omitted; refer to PDF] as the matrix [figure omitted; refer to PDF] to transform [figure omitted; refer to PDF] into new data [figure omitted; refer to PDF] .

(4) Compute the [figure omitted; refer to PDF] kernel matrix [figure omitted; refer to PDF] and centralize [figure omitted; refer to PDF] to [figure omitted; refer to PDF] using (2). Solve the eigenvalue problem of [figure omitted; refer to PDF] , obtain the eigenvalues [figure omitted; refer to PDF] and eigenvectors [figure omitted; refer to PDF] of [figure omitted; refer to PDF] . Set [figure omitted; refer to PDF] as the eigenvalue list sort by decreasing order, choose the number of kernel principal components [figure omitted; refer to PDF] with the principle of the minimum value that the cumulative contribution rate is above 75% [18, 19].

(5) Calculate the load matrix [figure omitted; refer to PDF] : [figure omitted; refer to PDF] , where [figure omitted; refer to PDF] denotes the diagonal matrix by the first [figure omitted; refer to PDF] eigenvalues in [figure omitted; refer to PDF] , and [figure omitted; refer to PDF] is the eigenvectors matrix corresponding to the first [figure omitted; refer to PDF] eigenvalues.

4. Multiclasses Fault Diagnosis [figure omitted; refer to PDF] -Nearest Neighbor

The [figure omitted; refer to PDF] NN method is the simplest, yet most useful approach to general pattern classification, and has been widely applied in pattern classification. Its principle is based on an intuitive concept that data points of the same class should be closer in the feature space. Specifically, for a given unlabeled sample, [figure omitted; refer to PDF] NN finds the [figure omitted; refer to PDF] -nearest labeled samples in the training data set and assigns to the class that appears most frequently within the [figure omitted; refer to PDF] -subset (i.e., [figure omitted; refer to PDF] -nearest neighbors). [figure omitted; refer to PDF] NN's error rate has been proven to be asymptotically at most twice that of the Bayesian error rate [20]. [figure omitted; refer to PDF] NN has been used for image clustering [21] and cardiac arrhythmia diagnosis [22], and so forth, but few published researches can be found in machine fault detection. In this paper, the [figure omitted; refer to PDF] NN classifier is used to evaluate the reduced feature vectors by KPCA for avoiding relatively local relevance between parameters and reducing computation complexity.

He and Wang provided a fault detection method using the [figure omitted; refer to PDF] -nearest neighbor method for semiconductor manufacturing processes, named FD- [figure omitted; refer to PDF] NN [23]. It is based on the idea that the trajectory of a normal sample is similar to the trajectories of normal samples in the training data, and the trajectory of a fault sample must exhibit some deviation from the trajectories of normal training samples. However, FD- [figure omitted; refer to PDF] NN just determines two classes (normal and abnormal) which cannot be applied directly in NC machine tools, because they have several fault classes (normal, wear, gap, etc.) that we want to determine. In this paper, a multiclasses fault diagnosis [figure omitted; refer to PDF] NN method (MFD- [figure omitted; refer to PDF] NN) is proposed based on FD- [figure omitted; refer to PDF] NN.

The proposed MFD- [figure omitted; refer to PDF] NN method consists of three steps for model training.

(1) For the [figure omitted; refer to PDF] th fault class [figure omitted; refer to PDF] in training data, finding [figure omitted; refer to PDF] nearest neighbors for each sample [figure omitted; refer to PDF] in the same class, [figure omitted; refer to PDF] denotes the [figure omitted; refer to PDF] th sample in the [figure omitted; refer to PDF] th fault class.

(2) Calculate the [figure omitted; refer to PDF] NN squared distance for each sample: [figure omitted; refer to PDF] , where [figure omitted; refer to PDF] denotes squared Euclidean distance from sample [figure omitted; refer to PDF] to its [figure omitted; refer to PDF] th nearest neighbor in the [figure omitted; refer to PDF] th fault class.

(3) Determine the corresponding control limit [figure omitted; refer to PDF] of the [figure omitted; refer to PDF] th fault class [figure omitted; refer to PDF] . The threshold [figure omitted; refer to PDF] with a significance level [figure omitted; refer to PDF] can be determined by an assumption that [figure omitted; refer to PDF] following a noncentral [figure omitted; refer to PDF] distribution, which can be estimated using the Matlab function chilimit in the PLS Toolbox.

The choice of [figure omitted; refer to PDF] is usually uncritical [23]. In general, the value of [figure omitted; refer to PDF] is greater and the effect of noise is fewer, but the boundary between normal and faults is less distinct. The value of [figure omitted; refer to PDF] can be set to be a modest number, such as 5, 7, and 10, based on the sample size.

5. Fault Diagnosis Approach for NC Machine Tools Based on KPCA and MFD- [figure omitted; refer to PDF] NN

The fault diagnosis approach for NC machine tools in this paper includes two phases. The first phase focuses on model training, reducing dimension of feature parameters by data-dependent KPCA, and determining thresholds of different fault classes using MFD- [figure omitted; refer to PDF] NN method. The second phase is fault detection for unknown working tools. The specific procedure is constructed through the following main steps as shown in Figure 1.

Figure 1: Sketch of fault diagnosis process for NC machine tools based on KPCA and MFD- [figure omitted; refer to PDF] NN.

[figure omitted; refer to PDF]

5.1. Model Training

Step 1.

Gain vibration time-domain signals of NC machine in several fault classes [figure omitted; refer to PDF] , such as all tools normal, single tool wear, and single tool gap. Then divide signals of each fault class into [figure omitted; refer to PDF] groups which include continuous [figure omitted; refer to PDF] signals data, respectively; transform every group data to frequency-domain signals by FFT.

Step 2.

For the normal class (all tools are normal), calculate the average [figure omitted; refer to PDF] and standard deviation [figure omitted; refer to PDF] of feature vectors, and extract nonlinear features using the data-dependent KPCA algorithm provided in Section 3, obtain the kernel principal components [figure omitted; refer to PDF] , and load matrix [figure omitted; refer to PDF] .

Step 3.

Standardize [figure omitted; refer to PDF] by the average [figure omitted; refer to PDF] and the standard deviation [figure omitted; refer to PDF] : [figure omitted; refer to PDF] , and multiply [figure omitted; refer to PDF] to the load matrix [figure omitted; refer to PDF] to obtain the time-frequency principal components [figure omitted; refer to PDF] of the [figure omitted; refer to PDF] fault: [figure omitted; refer to PDF] .

Step 4.

For every fault class [figure omitted; refer to PDF] , determine the corresponding control limit [figure omitted; refer to PDF] using the MFD- [figure omitted; refer to PDF] NN method proposed in Section 4.

5.2. Fault Detection

Step 5.

Collect continuous [figure omitted; refer to PDF] vibration time-domain signals data [figure omitted; refer to PDF] of NC machine under operating condition periodically while the state of tools is unknown and to identify, transform signals data of [figure omitted; refer to PDF] to frequency-domain signals by FFT.

Step 6.

Calculate eight feature parameters [figure omitted; refer to PDF] of signals, and standardize [figure omitted; refer to PDF] by the average [figure omitted; refer to PDF] and the standard deviation [figure omitted; refer to PDF] : [figure omitted; refer to PDF] . Then multiply to the load matrix [figure omitted; refer to PDF] to obtain the time-frequency principal components [figure omitted; refer to PDF] of the unknown tools: [figure omitted; refer to PDF] .

Step 7.

Identify [figure omitted; refer to PDF] 's [figure omitted; refer to PDF] nearest neighbors for each fault class [figure omitted; refer to PDF] in the training data set.

Step 8.

Calculate [figure omitted; refer to PDF] 's [figure omitted; refer to PDF] NN squared distance [figure omitted; refer to PDF] .

Step 9.

Calculate [figure omitted; refer to PDF] , and compare [figure omitted; refer to PDF] with the number 1, if [figure omitted; refer to PDF] , [figure omitted; refer to PDF] can be considered as a sample of the [figure omitted; refer to PDF] th fault class. Otherwise, it is not belong to the [figure omitted; refer to PDF] th fault class.

Step 10.

Three results could possibly be obtained by Step 9. One result is that [figure omitted; refer to PDF] belongs to a single class; it can be determined that [figure omitted; refer to PDF] belongs to the corresponding class distinctly. The second result is that [figure omitted; refer to PDF] does not belong to any known class, in other words, [figure omitted; refer to PDF] for all [figure omitted; refer to PDF] . In this case, they have to stop working and check tools for a possible unknown damage, handle it, and update the diagnosis model. And the last result is that [figure omitted; refer to PDF] belongs to several classes. This situation could happen despite very small probability. In this case, calculate the confidence level [figure omitted; refer to PDF] of [figure omitted; refer to PDF] using the Matlab function chilimit with scaling value and degrees of freedom calculated in Step 4. [figure omitted; refer to PDF] belongs to the class [figure omitted; refer to PDF] : [figure omitted; refer to PDF] .

6. Experiment Design

The FV-2215 horizontal Longmen machining center (see Figure 2) is used for the experimental test, in which characteristics are illustrated in Table 2. The cutting tools used in this test are carbide tools #RPMW1003# in a face milling cutter #EMR 5R160-40-8T# with four teeth, its shape and different tool state are shown in Figure 3. The experimental setup for the in-process detection of tool fault is illustrated in Figure 4. The tool vibration signals are monitored by a uniaxial accelerometer (PCB352C22) which was placed at the tool rest in the direction perpendicular to tool face, and sent to a data acquisition instrument (ECON Avant (MI-7016)) and a portable computer.

Table 2: Characteristics of the machining center.

Type	Horizontal
Model	FV-2215
Maximum speed of rotation	4000 (rpm)
Positioning accuracy	0.025 (mm)
Maximum feed rate	5000 (mm/min)

Figure 2: The FV-2215 horizontal Longmen machining center.

[figure omitted; refer to PDF]

Figure 3: Samples of milling cutter shape and different tools.

(a) Face milling cutter

[figure omitted; refer to PDF]

(b) Normal tool sample

[figure omitted; refer to PDF]

(c) Wear tool sample

[figure omitted; refer to PDF]

(d) Gap tool sample

[figure omitted; refer to PDF]

Figure 4: Experiment illustration.

[figure omitted; refer to PDF]

Tool states in this experiment are alternatively three situations: normal, single tool wear, and single tool gap, and tool gap is cut with approximately 4 mm depth using electro-discharge machining. Aluminum alloy samples #7075-T351# of size (150 mm, 100 mm, 500 mm) are fixed on the machining center. In this experiment, cutting speed and depth are 500 rpm and 2 mm respectively.

7. Result and Discussion

In this experiment, three tool fault classes were measured including [figure omitted; refer to PDF] all tools normal ( [figure omitted; refer to PDF] ), [figure omitted; refer to PDF] single tool wear ( [figure omitted; refer to PDF] ), and [figure omitted; refer to PDF] single tool gap ( [figure omitted; refer to PDF] ). It is assumed that all investigated tools have the same damage mechanism. The vibration signals were measured and the sampling frequency was 48,000 Hz, which was determined by the frequency characteristics of tool vibration. A time-domain waveform and its frequency-domain waveform of vibration signals sample of classes ( [figure omitted; refer to PDF] ) and ( [figure omitted; refer to PDF] ) in 10 s are shown in Figure 5. It can be observed from the figure that the time-domain data appear periodically and the energy of frequency-domain data is dispersed to several band ranges. It is difficult to identify tool's fault based on time-domain or frequency-domain waveform simply.

Figure 5: Time-frequency domain waveform sample of normal and single tool wear classes.

(a) Time-domain waveform with all tools normal

[figure omitted; refer to PDF]

(b) Frequency-domain waveform with all tools normal

[figure omitted; refer to PDF]

(c) Time-domain waveform with single tool wear

[figure omitted; refer to PDF]

(d) Frequency-domain waveform with single tool wear

[figure omitted; refer to PDF]

Vibration signals in three tool fault classes are measured 50 times which changes various tools partly in the same state every time. These vibration signals are divided into two parts: 40 times as training set and 10 times testing set. The training set is used to train model in Matlab R2012, which includes calculation of time-frequency statistical parameters (shown in Table 1) and dimensionality reduction by KPCA, and determine the control limit of three fault classes by the MFD- [figure omitted; refer to PDF] NN method. First in all, 40 group vibration signals in class ( [figure omitted; refer to PDF] ) (all tools are normal) are employed to reduce dimension by data-dependent KPCA. The cumulative eigen contribution rate with the first three eigenvalues is 94.3%, which is above the down limit [figure omitted; refer to PDF] [figure omitted; refer to PDF] . Hence the number of kernel principal components [figure omitted; refer to PDF] , and kernel principal components [figure omitted; refer to PDF] and the load matrix [figure omitted; refer to PDF] are obtained in KPCA. Then, the other two classes' vibration signals are standardized by the average and the standard deviation of the normal class signals, and transformed into kernel principal components [figure omitted; refer to PDF] and [figure omitted; refer to PDF] which are multiplied by the load matrix [figure omitted; refer to PDF] .

For every fault class [figure omitted; refer to PDF] , determine the corresponding control limit [figure omitted; refer to PDF] using the MFD- [figure omitted; refer to PDF] NN method proposed in which the value of [figure omitted; refer to PDF] is set to be 5. Take [figure omitted; refer to PDF] as an example, find [figure omitted; refer to PDF] nearest neighbors for each sample [figure omitted; refer to PDF] in the [figure omitted; refer to PDF] class, and calculate the [figure omitted; refer to PDF] NN squared distance for each sample [figure omitted; refer to PDF] : [figure omitted; refer to PDF] where [figure omitted; refer to PDF] and [figure omitted; refer to PDF] denote the vector of the [figure omitted; refer to PDF] th sample and its [figure omitted; refer to PDF] th nearest neighbor.

The threshold [figure omitted; refer to PDF] ( [figure omitted; refer to PDF] ) is calculated using the chilimit function in the Matlab PLS Toolbox.

The values of threshold [figure omitted; refer to PDF] of three fault classes are shown in Table 3.

Table 3: The values of threshold [figure omitted; refer to PDF] of three fault classes.

Fault class	Threshold value
All tools normal ( [figure omitted; refer to PDF] )	16.35
Single tool wear ( [figure omitted; refer to PDF] )	20.19
Single tool gap ( [figure omitted; refer to PDF] )	19.75

For validating the effectiveness of this proposed approach, the remaining 10 times vibration signals in each fault class are used as testing set. Firstly, calculate the testing signals' time-frequency statistical parameters (shown in Table 1), and then these statistical parameters are standardized by the average and the standard deviation of the normal class signals in the training phase and transformed into kernel principal components multiplied by the load matrix [figure omitted; refer to PDF] . Sequentially, find [figure omitted; refer to PDF] nearest neighbors for each testing data [figure omitted; refer to PDF] by calculating its distance with training data, and calculate its [figure omitted; refer to PDF] NN squared distance with the [figure omitted; refer to PDF] th fault class [figure omitted; refer to PDF] : [figure omitted; refer to PDF] where [figure omitted; refer to PDF] and [figure omitted; refer to PDF] denote the vector of the [figure omitted; refer to PDF] th testing data and its [figure omitted; refer to PDF] th nearest neighbor.

Finally, the value of [figure omitted; refer to PDF] is compared with 1, the result is shown in Figure 6. Figure 6(a) is the testing result with [figure omitted; refer to PDF] (all tools normal) training set, all points of [figure omitted; refer to PDF] testing set located below the threshold and the testing data of other two classes are above the threshold. Figure 6(b) is the testing result with [figure omitted; refer to PDF] (single tool wear) training set; the second data of [figure omitted; refer to PDF] testing set is above the threshold which should be below it, so this point is not determined to [figure omitted; refer to PDF] . Figure 6(c) is the testing result with [figure omitted; refer to PDF] (single tool gap) training set, the second data of [figure omitted; refer to PDF] testing set is below the threshold which should be above it, so this point is determined to [figure omitted; refer to PDF] . In brief, there is one fault detection error by the proposed method.

Figure 6: Test results of three classes with proposed method.

(a) Testing result with [figure omitted; refer to PDF] (all tools normal) training set

[figure omitted; refer to PDF]

(b) Testing result with [figure omitted; refer to PDF] (single tool wear) training set

[figure omitted; refer to PDF]

(c) Testing result with [figure omitted; refer to PDF] (single tool gap) training set

[figure omitted; refer to PDF]

PCA-MFD- [figure omitted; refer to PDF] NN method and standard KPCA-MFD- [figure omitted; refer to PDF] NN method with the same vibration signals are used for comparison, the standard KPCA take the conventional Gaussian kernel (3) as the kernel function. Despite the cumulative eigen contribution rate with the first three values by PCA and standard KPCA being 77.19% and 88.35% which are above the down limit CPV, they are both less than that value of the proposed method. The testing results with PCA-based method are shown in Figure 7. It can be observed that ten points are judged erroneously in Figure 7, one point of [figure omitted; refer to PDF] class and one point of [figure omitted; refer to PDF] class in training set are excluded from their class, and in the aspect of testing set, two points of [figure omitted; refer to PDF] class are determined to [figure omitted; refer to PDF] class, three points of [figure omitted; refer to PDF] class and one point of [figure omitted; refer to PDF] class are determined to [figure omitted; refer to PDF] class, and two points of [figure omitted; refer to PDF] class are determined to [figure omitted; refer to PDF] class.

Figure 7: Test results of three classes with PCA-MFD- [figure omitted; refer to PDF] NN method.

(a) Testing result with [figure omitted; refer to PDF] (all tools normal) training set

[figure omitted; refer to PDF]

(b) Testing result with [figure omitted; refer to PDF] (single tool wear) training set

[figure omitted; refer to PDF]

(c) Testing result with [figure omitted; refer to PDF] (single tool gap) training set

[figure omitted; refer to PDF]

The testing results with standard KPCA-based method are shown in Figure 8. It can be observed that five points are judged erroneously in Figure 8, one point of [figure omitted; refer to PDF] class in training set is excluded from its class, and in the aspect of testing set, two points of [figure omitted; refer to PDF] class are determined to [figure omitted; refer to PDF] class, two points of [figure omitted; refer to PDF] class are determined to [figure omitted; refer to PDF] class. It is clear that the proposed method is outperforming than the PCA-MFD- [figure omitted; refer to PDF] NN and standard KPCA-MFD- [figure omitted; refer to PDF] NN methods in tool fault diagnosis.

Figure 8: Test results of three classes with standard KPCA-MFD- [figure omitted; refer to PDF] NN method.

(a) Testing result with [figure omitted; refer to PDF] (all tools normal) training set

[figure omitted; refer to PDF]

(b) Testing result with [figure omitted; refer to PDF] (single tool wear) training set

[figure omitted; refer to PDF]

(c) Testing result with [figure omitted; refer to PDF] (single tool gap) training set

[figure omitted; refer to PDF]

8. Conclusion

The present study proposes a fault detection approach to NC machine tools, which is based on data-dependent kernel principal component analysis (DKPCA) and multiclasses fault diagnosis [figure omitted; refer to PDF] -nearest neighbor (MFD- [figure omitted; refer to PDF] NN). Firstly, the original data set are transformed high dimensional data into low dimensional manifold feature space with the intrinsic dimensionality by DKPCA, in which the kernel function is constructed with covariance matrix of sample data. The [figure omitted; refer to PDF] NN technique is applied to low dimensional manifold feature space to detect potential faults. Future work can be centered on the expansion of fault classes of NC machine' tool and the efficiency and calculation speed of the proposed method.

Acknowledgments

This work was supported by the youth project of the National Nature Science Foundation of China (Grant nos. 51405346 and 71101112) and the key science and technology innovation team project of Wenzhou City in China (Grant no. C20120002-02).

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

References

[1] J. Lv, J. Wang, M. Wang, Y. Wu, "Research on tool wear condition monitoring based on combination of SOM and HMM," China Mechanical Engineering , vol. 21, no. 13, pp. 1531-1535, 2010.

[2] I. Zaghbani, M. Lamraoui, V. Songmene, M. Thomas, M. El Badaoui, "Robotic high speed machining of aluminum alloys," Advanced Materials Research , vol. 188, pp. 584-589, 2011.

[3] M. Jachan, G. Matz, F. Hlawatsch, "Time-frequency ARMA models and parameter estimators for underspread nonstationary random processes," IEEE Transactions on Signal Processing , vol. 55, no. 9, pp. 4366-4381, 2007.

[4] R. Shao, W. Hu, J. Li, "Multi-fault feature extraction and diagnosis of gear transmission system using time-frequency analysis and wavelet threshold de-noising based on EMD," Shock and Vibration , vol. 20, no. 4, pp. 763-780, 2013.

[5] P. K. Kankar, S. C. Sharma, S. P. Harsha, "Rolling element bearing fault diagnosis using autocorrelation and continuous wavelet transform," Journal of Vibration and Control , vol. 17, no. 14, pp. 2081-2094, 2011.

[6] M. Singh, R. Kumar, "Thrust bearing groove race defect measurement by wavelet decomposition of pre-processed vibration signal," Measurement , vol. 46, no. 9, pp. 3508-3515, 2013.

[7] D. P. Jena, S. N. Panigrahi, R. Kumar, "Gear fault identification and localization using analytic wavelet transform of vibration signal," Measurement , vol. 46, no. 3, pp. 1115-1124, 2013.

[8] Z. Feng, M. Liang, F. Chu, "Recent advances in time-frequency analysis methods for machinery fault diagnosis: a review with application examples," Mechanical Systems and Signal Processing , vol. 38, no. 1, pp. 165-205, 2013.

[9] J. L. Chen, Y. Y. Zi, Z. J. He, X. D. Wang, "Construction of adaptive redundant multiwavelet packet and its application to compound faults detection of rotating machinery," Science China Technological Sciences , vol. 55, no. 8, pp. 2083-2090, 2012.

[10] X. Jiawei, M. Toshiro, W. Yanxue, J. Zhansi, "A hybrid of interval wavelets and wavelet finite element model for damage detection in structures," Computer Modeling in Engineering & Sciences , vol. 81, no. 3-4, pp. 269-294, 2011.

[11] V. H. Nguyen, C. Rutten, J.-C. Golinval, "Fault diagnosis in industrial systems based on blind source separation techniques using one single vibration sensor," Shock and Vibration , vol. 19, no. 5, pp. 795-801, 2012.

[12] A. Parus, B. Powalka, K. Marchelek, S. Domek, M. Hoffmann, "Active vibration control in milling flexible workpieces," Journal of Vibration and Control , vol. 19, no. 7, pp. 1103-1120, 2013.

[13] B. Chen, X. Chen, B. Li, Z. He, H. Cao, G. Cai, "Reliability estimation for cutting tools based on logistic regression model using vibration signals," Mechanical Systems and Signal Processing , vol. 25, no. 7, pp. 2526-2537, 2011.

[14] X. Houzheng, H. Min, "Research of numerical control machine tools wear monitoring method based on vibration testing," Instrument Technique and Sensor , no. 2, pp. 73-75, 2013.

[15] R. Sun, F. Tsung, L. Qu, "Evolving kernel principal component analysis for fault diagnosis," Computers and Industrial Engineering , vol. 53, no. 2, pp. 361-371, 2007.

[16] V. H. Nguyen, J.-C. Golinval, "Fault detection based on Kernel Principal Component Analysis," Engineering Structures , vol. 32, no. 11, pp. 3683-3691, 2010.

[17] J. Wang, J. T. Kwok, H. C. Shen, L. Quan, "Data-dependent kernels for high-dimensional data classification," in Proceedings of the International Joint Conference on Neural Networks (IJCNN '05), pp. 102-107, August 2005.

[18] H. Abdi, L. J. Williams, "Principal component analysis," Wiley Interdisciplinary Reviews: Computational Statistics , vol. 2, no. 4, pp. 433-459, 2010.

[19] Z. Jiulong, D. Xiaonan, Z. Zhiyu, "Probabilistic kernel PCA with application," Computer Engineering and Applications , vol. 47, no. 4, pp. 165-167, 2011.

[20] F. Angiulli, G. Folino, "Distributed nearest neighbor-based condensation of very large data sets," IEEE Transactions on Knowledge and Data Engineering , vol. 19, no. 12, pp. 1593-1606, 2007.

[21] T. N. Tran, R. Wehrens, L. M. C. Buydens, "KNN-kernel density-based clustering for high-dimensional multivariate data," Computational Statistics and Data Analysis , vol. 51, no. 2, pp. 513-525, 2006.

[22] W. M. Zuo, W. G. Lu, K. Q. Wang, H. Zhang, "Diagnosis of cardiac arrhythmia using kernel difference weighted KNN classifier," in Proceedings of the Computers in Cardiology (CAR '08), vol. 35, pp. 253-256, September 2008.

[23] Q. P. He, J. Wang, "Fault detection using the k-nearest neighbor rule for semiconductor manufacturing processes," IEEE Transactions on Semiconductor Manufacturing , vol. 20, no. 4, pp. 345-354, 2007.