Weiwei Qian^{1} , Shunming Li^{2} , Jinrui Wang^{3} , Zenghui An^{4} , Xingxing Jiang^{5}
^{1, 2, 3, 4}College of Energy and Power Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing, 210016, P. R. China
^{5}School of Urban Rail Transportation, Soochow University, Suzhou, 215137, P. R. China
^{1}Corresponding author
Journal of Vibroengineering, Vol. 20, Issue 8, 2018, p. 28392854.
https://doi.org/10.21595/jve.2018.19339
Received 1 November 2017; received in revised form 14 May 2018; accepted 24 May 2018; published 31 December 2018
Copyright © 2018 Weiwei Qian, et al. 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.
Traditional intelligent fault diagnosis methods take advantage of diagnostic expertise but are laborintensive and timeconsuming. Among various unsupervised feature extraction methods, sparse filtering computes fast and has less hyperparameters. However, the standard sparse filtering has poor generalization ability and the extracted features are not so discriminative by only constraining the sparsity of the feature matrix. Therefore, an improved sparse filtering with L1 regularization (L1SF) is proposed to improve the generalization ability by improving the sparsity of the weight matrix, which can extract more discriminative features. Based on Fourier transformation (FFT), L1SF, softmax regression, a new threestage intelligent fault diagnosis method of rotating machinery is developed. It first transforms timedomain samples into frequencydomain samples by FFT, then extracts features in L1regularized sparse filtering and finally identifies the health condition in softmax regression. Meanwhile, we propose employing different activation functions in the optimization of L1SF and feedforward for considering their different requirements of the nonsaturating and antinoise properties. Furthermore, the effectiveness of the proposed method is verified by a bearing dataset and a gearbox dataset respectively. Through comparisons with the standard sparse filtering and L2regularized sparse filtering, the superiority of the proposed method is verified. Finally, an interpretation of the weight matrix is given and two useful sparse properties of weight matrix are defined, which explain the effectiveness of L1SF.
Keywords: fault diagnosis, automatic feature extraction, sparse filtering, L1 regularization, activation function.
With the rapid development of technology, industrial Internet of Things (IoT) and data driven techniques [1] have been revolutionizing manufacturing. As machines are becoming more efficient and automatic than ever before, it is more significant to have a safe and sound operation [2]. Meanwhile, rotating speed oscillation usually occurs during machine operation which can accelerate the damage on machines [3]. Therefore, condition monitoring systems are adopted to collect the massive amount of data from monitored machines [4, 5]. At the same time, with substantial development of computing systems and sensors, machinery fault diagnosis has fully embraced the extensive revolution in modern manufacturing.
Traditionally, there are mainly three sequential steps in the framework of fault diagnosis systems: 1) signal acquisition; 2) feature extraction and selection and 3) fault classification [2, 6, 7]. In the first step, vibration signals are widely used as the data source since essential information about machine health condition is embedded in them. However, feature extraction is always needed because the raw signals contain various kinds of noise. In the second step, representative features are extracted for fault classification. The more discriminative features are extracted from vibration signals, the more accurate fault diagnosis result would be. Signal processing techniques have been proposed and widely used in feature extraction and selection such as timedomain analysis [8], frequency transform [9], high resolution timefrequency analysis [10], wavelet transform [11], and envelope demodulation algorithms [12]. Besides, as entropy is effective in detection the dynamic characteristics of time series, some methods use entropy in feature extraction and selection. Modiﬁed multiscale symbolic dynamic entropy (MMSDE) [13] and modified hierarchical permutation entropy (MHPE) [14] were proposed to extract features for gearbox fault diagnosis. Zhao et al. [15] combined the EEMD and multiscale fuzzy entropy to extract more discriminative features. However, the further analysis of features extracted by these algorithms requires manpower and prior knowledge. At the same time, although these techniques can effectively extract primary features from signals, these features may prevail among all signals of various health conditions. Therefore, dimensionreduction strategies such as nonlinear PCA [16], independent component analysis (ICA) [17] are required to select sensitive features. It means that these extracted features should be sensitive to different health conditions but be insensitive to different operating conditions such as large speed oscillation. Intelligent feature learning methods can also be utilized in feature extraction along with feature selection. Among various intelligent feature learning methods, unsupervised ones have been successfully applied to extract useful features in many image, video and audio tasks.
Well developed as unsupervised feature learning algorithms are, there are still two main drawbacks [18]. Firstly, unsupervised learning algorithms generally have plenty of parameters, which are difficult to tune and crucial to the performance of the network. The comparison of tunable parameters in various unsupervised learning algorithms is shown in Table 1 [18, 19]. It can be seen that there are massive parameters in sparse restricted Boltzmann machines (RBMs) [20], sparse autoencoder [21], sparse coding [22]. Secondly, computational complexity is also a major drawback. Independent component analysis (ICA) has just one hyperparameter needing tuning [23] and provides commendable results in object recognition experiments, but it scales poorly to large datasets and highdimension inputs. Sparse coding has an expensive inference process, which requires a costly optimizing iteration. Stacked autoencoder (SAE) [24] has to do a greedy layerwise training process which includes decoding and encoding networks, and the parameters of decoding network are useless in fine tuning which will increase the computational complexity. In the final step, the learned features are utilized to train the classifiers such as support vector machine (SVM) and softmax regression [25], which are always used to identify the health condition. In Ref. [26], the improved Particle Swarm Optimization (PSO) algorithm was proposed to optimize the parameters of least squares support vector machines (LSSVM) in order to construct an optimal classifier.
Table 1. Tunable parameters in various algorithms
Algorithm

Tunable parameters

Sparse filtering

Feature dimension

ICA

Feature dimension

Sparse coding

Feature dimension, minibatch size, sparse penalty

Sparse autoencoders

Feature dimension, target activation, sparse penalty, weight decay

Sparse RBMs

Features, target activation, sparse penalty, weight decay, momentum learning rate

As sparsity is one of the desirable properties of a good feature representation [27], Ngiam et al. [19] proposed an unsupervised algorithm called sparse filtering to solve the problems pointed out above, which is a twolayer neural network. Only the dimension of features learned from sparse filtering needs tuning and its main idea is the sparsity of the feature matrix [18]. Meanwhile, it was validated on image recognition and phone classification tasks, which obtained the stateoftheart performance. For its simplicity and good performance, Lei et al. [2] introduced sparse filtering into rotating machine fault diagnosis and constructed a twostage intelligent fault diagnosis method, which got better performance than other networks. Sparse filtering works by imposing sparse constraints on the distribution sparsity of the feature matrix but the overefitting risk of it is quite high, namely having poor generalization ability. Yang et al. [28] utilized L2 norm to regularize the weight matrix of sparse filtering. However, its effectiveness in preventing overfitting is limited and it takes no consideration of the sparsity of the weight matrix. Meanwhile, the activation functions used in weight matrix optimization of sparse filtering [2] are softabsolute function, which computes fast and can partly avoid the gradient vanishing problem. Nevertheless, when it is used in feedforward, its robustness to noise is weak. L1 norm can also be utilized to reduce the overfitting risk. However, when introducing L1 regularization into the original network, the optimization of the new network is difficult sometimes. Recently, Deng et al. [29, 30] introduced new algorithms into solving complex optimization problems, which can be beneficial to the network with many terms to be optimized. With the help of them, L1 norm can be utilized better.
Inspired by Ref. [2830], aiming at reducing the overfitting risk, an improved sparse filtering network with L1 regularization is proposed, which can improve the performance of the network by exploring the sparsity of the weight matrix. Meanwhile, for considering the different requirements of nonsaturating and antinoise properties, we propose using different activation functions in feature extraction network training and feedforward, which is inspired by the training process of deep learning that different activation functions can be adopted in greedy layerwise pretraining and further fine tuning.
The main contributions of this paper are described as follows:
(1) A threestage intelligent fault diagnosis method of rotating machinery is developed. In the first stage, timedomain samples are automatically preprocessed into frequencydomain samples by FFT. In the second stage, an improved feature extraction network called L1regularized sparse filtering is proposed to extract and select discriminative features. In the last stage, features are fed into softmax regression to identify the health condition. Experiments conducted on a bearing dataset and a gearbox dataset verify the effectiveness of the proposed method.
(2) L1regularized sparse filtering is proposed to constrain the weight matrix sparsity and improve the generalization ability of standard sparse filtering. We also construct the L2regularized sparse filtering and do experiments on it. The comparison with standard sparse filtering demonstrates its effectiveness. The comparison with L1regularized sparse filtering shows it is inferior when using frequencydomain samples.
(3) An interpretation of the weight matrix learned by sparse filtering is given, which is the supplement to the interpretation given in Ref. [2]. Based on it, two sparse properties of the weight matrix are defined, which are validated significant in obtaining discriminative features. They also explain why L1regularized sparse filtering performs better when frequencydomain samples are adopted.
This paper is organized as follows. Section 2 describes sparse filtering and L1 regularization briefly. Section 3 details the proposed L1regularized sparse filtering and the using of the new activation function. In Section 4, diagnosis cases of a bearing dataset and a gearbox dataset are studied separately using the proposed method and other models. Section 5 discusses the properties of the weight matrix learned by the improved sparse filtering. Finally, the conclusion is drawn in Section 6.
Generally, unsupervised feature learning algorithms can be divided into two main categories [18]: explicitly modeling the input distribution or not. ICA, sparse coding, SAE and sparse RBMs explicitly model the input distribution and reconstruct mechanical dynamic signal structure by minimizing the reconstruction error. Although it’s desirable to learn a good approximation of the input distribution, algorithms such as sparse filtering show that this is not so important if the aim is to get discriminative features. The fundamental idea of sparse filtering focuses on constraining the sparsity of output features instead of explicitly modeling the input distribution. Sparse filtering tries to learn sparse feature distribution that satisfies three principles:
(1) Population sparsity, which refers that all the features of one sample should be activated sparsely at a time;
(2) Lifetime sparsity, which refers that the same kind of features belonging to all samples should be activated sparsely at a time;
(3) High dispersal, which refers that all kinds of features should have similar statistics.
The architecture of the sparse filtering network is shown in Fig. 1. Firstly, the collected raw vibration data is alternately separated into $K$ samples. Then they are transformed from original timedomain samples into Fourier spectra $\{{\mathbf{x}}^{i}{\}}_{i=1}^{K}$ by fast Fourier transformation (FFT), where ${\mathbf{x}}^{i}\in {R}^{N\times 1}$ is $i$th sample containing $N$ Fourier coefficients, which are the amplitudes of the spectral lines. Next, the training dataset is used to train sparse filtering to get an optimized weight matrix $\mathrm{W}\in {R}^{N\times {N}^{out}}\text{,}$ where $L$ is the dimension of output feature vectors. Finally, the frequencydomain sample ${\mathbf{x}}^{i}\in {R}^{N\times 1}$ is mapped into a feature vector ${\text{f}}^{i}\in {R}^{{N}^{out}\times 1}$ in feedforward. The mapping can be realized through $\mathbf{W}$ and activation function 2.
Fig. 1. Architecture of the sparse filtering network
In the optimization, the standard sparse filtering uses a softabsolute function as activation function 1 [19], so the features can be computed as follows:
where ${f}_{j}^{i}$ refers to the $j$th feature of the $i$th sample and $\epsilon $ is set to 10^{8}. The feature matrix learned from training dataset is $\mathbf{f}\in {R}^{{N}^{out}\times K}$, where $K$ is the number of training samples and ${N}^{out}$ is the dimension of output feature vectors. Three steps should be applied to realize the sparse principles mentioned above, which are shown as follows.
Firstly, by means of dividing each kind of feature using its L2 norm [2] across all samples, each kind of feature ${\mathbf{f}}_{j}$ is normalized to be equally active:
Then, the new feature matrix is normalized by columns, so that features of each sample ${\stackrel{~}{\mathbf{f}}}^{i}$ lie on the unit L2 ball:
At last, the normalized features are optimized for sparsity by L1 norm. The objective function of sparse filtering can be described as follows:
In this paper, we take a novel view on sparse filtering: considering it as a greedy layerwise training process [31], which means that the activation function used in finetuning process could be different from the one used in weight matrix initialization and that is analogous to the training process in SAE. Therefore, to get a more noiserobust feature representation, we adopt Eq. (5) as activation function 2:
instead of the widely used softabsolute function, as shown in Fig. 1.
Overfitting can be simply described as an undesirable phenomenon that the training accuracy is higher than testing accuracy. The network with higher overfitting risk means it has lower generalization ability. It always occurs when the network model is too complex or when the training dataset is too small. However, it is desirable to use a smaller dataset in training and get a higher diagnosis accuracy in testing. To reduce the overfitting risk of the sparse filtering, we use strategies like L2 and L1 regularization. L2 norm penalty, which can constrain the magnitudes of parameters is usually used in overfitting problem as a regularization approach [32] and is generally called L2 regularization. It works by penalizing the bigger parameters in the weight matrix to make them smaller. In addition, it should be noticed that L2 regularization cannot make weight matrix sparse theoretically [33].
Different from L2 norm, L1 norm can impose a sparse constraint on the parameters. Therefore, L1 norm is used to realize the sparsity of the feature matrix in sparse filtering, as shown in the subsection above. However, it is noticed that the sparsity of the weight matrix is not constrained. As weight matrix is closely linked with the output features, heuristically we propose utilizing L1 norm to directly constrain the sparsity of the weight matrix and then further constrain the sparsity of the feature matrix. It is called L1 regularization and can make the sparse filtering perform better. In this paper, we will also compare the performance of L1 regularization and L2 regularization. Specifically, L1 regularization and L2 regularization are added to the objective function of the standard sparse filtering in Eq. (4) respectively, and the new objective functions are defined separately in Eq. (6) and Eq. (7) as follows:
where ${\lambda}_{1}$ and ${\lambda}_{2}$ are the regularization parameters, ${\lambda}_{1}$ controls the relative importance of feature sparse distribution and weight matrix sparse distribution; ${\lambda}_{2}$ controls the relative importance of feature sparse distribution and weight decay; $N$ is the input dimension of samples and ${N}^{out}$ is dimension of learned feature vector; ${w}_{ij}$ is the $i$th row and the $j$th column element of the weight matrix $\mathbf{W}$. The gradient of the regularization term can be derived correspondingly and combined into the original gradient.
The unsupervised feature learning process of sparse filtering is described above and it is always followed by a classifier. Softmax regression [31] is often applied at the last stage for multiclass classification and performs well. Therefore, we implement softmax regression as the classifier in this paper.
Suppose that the output feature dataset $\{{\mathbf{f}}^{i}{\}}_{i=1}^{K}$ has label set $\{{y}^{i}{\}}_{i=1}^{K}$, where ${y}^{i}\in \{\mathrm{1,2},\dots ,R\}$. For each ${\mathbf{f}}^{i}$, the output layer attempts to estimate the probability $p({y}^{i}=r{\mathbf{f}}^{i})$ for each label of $r=$ 1, 2, ..., $R$. Thus, the output layer is optimized by minimizing the objective function:
where $1\{\cdot \}$ represents the indicator function, which outputs 1 if the condition is true, and 0 otherwise; ${\mathbf{W}}_{2}$ is the weight matrix of the output layer and ${\mathbf{W}}_{2}^{r}$ is the $r$th row of ${\mathbf{W}}_{2}$. The second term is the weight decay term, and ${\mathbf{W}}_{2}^{k,j}$ is the $k$th row and the $j$th column element of ${\mathbf{W}}_{2}$; ${\lambda}_{3}$ is the regular parameter.
This section details how the proposed method automatically extracts features from raw data and then briefly describes how the diagnosis results are obtained. The framework of our method is illustrated in Fig. 2 and the process can be described as follows.
(1) Data sampling and preprocessing. The timedomain raw signals are sampled into dataset $\{{\mathbf{z}}^{i}{\}}_{i=1}^{M}$, where ${\mathbf{z}}^{i}\in {R}^{2N\times 1}$ is the $i$th sample containing $2N$ data points. Then each sample ${\mathbf{z}}^{i}$ is transformed into frequencydomain sample ${\mathbf{t}}^{i}$ by FFT, where ${\mathbf{t}}^{i}\in {R}^{N\times 1}$ is the $i$th sample containing $N$ Fourier coefficients. These samples are combined into dataset ${\left\{{\mathbf{t}}^{i}\right\}}_{i=1}^{M}$.
(2) Training dataset selection and whitening. $K$ samples are randomly selected from ${\left\{{\mathbf{t}}^{i}\right\}}_{i=1}^{M}$ composing the training dataset ${\left\{{\mathbf{x}}^{i}\right\}}_{i=1}^{K}$ and the label set of the training samples is ${\left\{{l}^{i}\right\}}_{i=1}^{K}$, where ${l}^{i}$ is the label of sample ${\mathbf{x}}^{i}$. The rest samples compose the testing dataset ${\left\{{\mathbf{s}}^{i}\right\}}_{i=1}^{MK}$. A whitening approach [2] is utilized to preprocess the training samples and obtain the whitened dataset ${\mathbf{S}}_{w}\in {R}^{N\times K}$, where $K$ is the number of samples and $N$ is the dimension of samples.
(3) Sparse filtering training. Firstly, the whitened dataset ${\mathbf{S}}_{w}$ is fed into the L1regularized sparse filtering model. Then, the weight matrix ${\mathbf{W}}_{1}\in {R}^{N\times {N}^{out}}$ is obtained by minimizing the new objective function as Eq. (6), where ${N}^{out}$ is the output dimension. It should be noted that Eq. (1) is used as activation function 1.
(4) Output features calculation. A feedforward process is carried out on the training dataset ${\left\{{\mathbf{x}}^{i}\right\}}_{i=1}^{K}$ with the optimized weight matrix ${\mathbf{W}}_{1}$ and activation function 2. The new function described in Eq. (5) is adopted as activation function 2. Each sample ${\mathbf{x}}^{i}$ is mapped into feature vector ${\mathbf{f}}^{i}\in {R}^{{N}^{out}\times 1}$, so the feature matrix of training samples is ${\left\{{\mathbf{f}}^{i}\right\}}_{i=1}^{K}$.
(5) Softmax regression training. The feature matrix ${\left\{{\mathbf{f}}^{i}\right\}}_{i=1}^{K}$ together with the labels of the training samples ${\left\{{l}^{i}\right\}}_{i=1}^{K}$ is fed into the softmax regression to train the network. The trained softmax regression will be used to identify the health conditions of feature vectors.
(6) Diagnose the testing samples. The trained network is employed to identify health conditions of testing samples${\left\{{\mathbf{s}}^{i}\right\}}_{i=1}^{MK}$, as shown in Fig. 2.
Fig. 2. Flowchart of the proposed method
In this section, a bearing dataset and a gearbox dataset are employed to validate the effectiveness of our method. In order to further verify the superiority of our method, six different networks are constructed, as shown in Table 2, where SF denotes sparse filtering; L1SF denotes Lregularized sparse filtering; L2SF denotes L2regularized sparse filtering. Meanwhile, 20 trials are carried out for each experiment to reduce the effect of randomness. The parameters are all tuned to the best, and the regular parameter of softmax regression is tuned to 1E5 [2]. The computation platform is a PC with an Intel I5 CPU and 8G RAM.
Table 2. Description of networks
Method

Main model

Activation function 1

Activation function 2

SFABS

SF

Eq. (1)

Eq. (1)

SFLOG

SF

Eq. (1)

Eq. (5)

L1SFLOG

L1SF

Eq. (1)

Eq. (5)

L2SFLOG

L2SF

Eq. (1)

Eq. (5)

L1SFABS

L1SF

Eq. (1)

Eq. (1)

L2SFABS

L2SF

Eq. (1)

Eq. (1)

The motor bearing experimental dataset provided by Case Western Reserve University [34] is employed in this section. The vibration signals were collected from the drive end of the motor through the triaxial accelerator. The dataset contains four different health conditions: 1) normal condition (NC); 2) outer race fault (OF); 3) inner race fault (IF) and 4) roller fault (RF). Vibration signals of three different damage severity levels (0.07, 0.14 and 0.21 inch) were separately collected for health condition OF, IF and RF. In addition, the signals were all collected under three different loads (1, 2 and 3 hp) and the sampling frequency was 48 kHz. The total dataset is composed of ten health conditions under three loads, and we treat the same health condition under different loads as the same class. 300 samples are obtained from each health condition under one load, so the dataset is composed of 9000 samples and each sample includes 1600 timedomain data points.
In this subsection, the dataset will be processed by the proposed method. Firstly, FFT is implemented on each timedomain sample and 1600 Fourier coefficients are obtained. As the coefficients are symmetric, we utilize the first 800 coefficients in each sample, so each sample contains 800 Fourier coefficients. As there is no obvious periodic property in each frequencydomain sample, the coefficients of each sample are treated as a whole. Therefore, there is no need to segment it in the way mentioned in Ref. [2] and the dataset ${\left\{{\mathbf{x}}^{j}\right\}}_{j=1}^{9000}$ is used to train and test the network, where ${\mathbf{x}}^{j}\in {R}^{800\times 1}$.
First, we investigate the selection of the parameters of sparse filtering. We randomly select 10 % of samples to train the proposed method and the rest of the samples are used for testing. The averaged diagnosis results of L1SFLOG with different output dimensions are shown in Fig. 3. It presents that the testing accuracy is higher than 99.2 % when ${N}^{out}$ is bigger than 600. It is evident that the testing accuracy goes higher and the corresponding standard deviation drops continuously with the increase of ${N}^{out}$, but the averaged training time keeps increasing. It indicates that we should make a tradeoff between the diagnosis accuracy and the time spent while selecting ${N}^{out}$. It can be seen that the increase of accuracy is unobvious when ${N}^{out}$ is bigger than 700, so we choose 700 as ${N}^{out}$. The testing accuracy reaches the highest and the standard deviation reaches the lowest when the regular parameter of L1SFLOG is around 100, so it is set to 100, as displayed in Fig. 4. The output dimension and regular parameter of L2SFLOG are tuned to 400 and 10 respectively. The output dimensions of SFLOG and SFABS are tuned to 700 and 100 separately.
Fig. 3. Output dimension selection
The diagnosis results using various numbers of samples for training are presented in Fig. 5. It presents in Fig. 5(a) that the testing accuracy rises moderately with the increase of training sample number because the testing accuracy is higher than 96 % with just 1 % samples for training. Fig. 5(a) also shows that L1SFLOG can diagnose ten health conditions with averaged testing accuracy up to 98 % when only 2 % samples are used for training, and the testing accuracy rises to 99 % when the percentage of training samples increases to 5 %. At the same time, it can be observed from Fig. 5 that the test accuracy is up to 96 % when the training samples for L2SFLOG, SFLOG, and SFABS increase to 10 %, 10 % and 20 % respectively. It indicates that L1SFLOG can achieve excellent performance even while lacking training samples and L1SFLOG performs better than L2SFLOG.
Fig. 4. The regular parameter of L1SFLOG
Fig. 5. Diagnosis results of: a) L1SFLOG, b) L2SFLOG, c) SFLOG, d) SFABS with different percentage of training samples
a)
b)
c)
d)
It also indicates in Fig. 5 that L1SFLOG can prevent the overfitting dramatically since the gaps between training accuracy and testing accuracy are always smaller than 3.7 %. By comparison, it shows that L1SFLOG and L2SFLOG both have higher testing accuracies and smaller gaps between training accuracy and testing accuracy than SFLOG under the same training sample number. Meanwhile, it also presents that SFLOG outperforms SFABS, which benefits directly from the using of the new activation function. Furthermore, when using same numbers of samples in training, L1SFLOG always holds the lowest standard deviation, which indicates L1SFLOG is more stable than the other three networks. In the following experiments, we randomly select 15 % samples from the whole dataset for training.
To further present the efficiency of the proposed method, 20 trials averaged accuracies and their standard deviations of all the six networks are shown in Table 3. It shows that L1SFLOG holds the highest accuracy and lowest standard deviation, which means L1SFLOG is more stable than the others. Meanwhile, it is worth mentioning that SFLOG performs better than SFABS. Comparing the results of SFLOG and SFABS, L2SFLOG and L2SFABS, L1SFLOG and L1SFABS, it presents that the networks which adopt the new activation in feedforward can achieve higher testing accuracies and are much more stable.
The 20 trials averaged confusion matrix of L1SFLOG shows that the health conditions IR7 and OR7 will be misclassified as each other most probably, as displayed in Table 4. It can be assumed that the health conditions misclassified as each other, mainly because signals of them are similar to each other. The normal health condition is classified perfectly mainly because it has no fault and is most distinct from other health conditions.
Table 3. The setting and performance of 6 networks
Method

Output dimension

Training accuracy (%)

Testing accuracy (%)

Training time (s)

SFABS

100

98.26±2.94

95.57±2.21

6.41±0.34

SFLOG

700

100

97.37±0.27

23.30±0.58

L1SFLOG

700

100

99.39±0.23

29.84±0.39

L2SFLOG

400

99.73±0.27

97.61±0.27

16.68±0.38

L1SFABS

700

99.50±0.68

98.08±0.89

25.30±1.06

L2SFABS

700

99.49±0.24

97.23±0.42

23.30±0.74

Note: The format of the result is: mean value ± standard deviation

Table 4. Confusion matrix of 20 trials averaged diagnosis results
Predict true

NOR

BALL7

BALL14

BALL21

IR7

IR14

IR21

OR7

OR14

OR21

NOR

765

0

0

0

0

0

0

0

0

0

BALL7

0

765

0

0

0

0

0

0

0

0

BALL14

0

0

761.65

0

0

0.15

3.2

0

0

0

BALL21

0

0

0

764.85

0

0

0

0

0

0.15

IR7

0.25

1.85

0.15

0.7

748.9

3.45

0

1.05

6.15

2.5

IR14

0

0

0

0

1.1

760.3

0.3

0

3.25

0.05

IR21

0

0

4.25

0.25

0.1

1.05

759.3

0

0.05

0

OR7

0

0

0

0

1.4

0

0

763.2

0.15

0.25

OR14

0

0

0.7

0.1

13.9

1.3

0.1

1.05

747.8

0.05

OR21

0

0.85

0.5

1.15

0.05

0.4

0

0

0

762.05

The vibration signals of a fourspeed motorcycle gearbox [35] used in this section were collected from the test rig as shown in Fig. 6. The test rig was also equipped with a triaxial accelerometer, a data acquisition system, a load mechanism, an electrical motor, a tachometer and four shock absorbers. The vibration signals of four conditions were collected: 1) normal condition (NC); 2) mediumworn (MW); 3) brokentooth (BT) and 4) slightworn (SW), as shown in Fig. 6 and each health condition has one load. The sampling frequency was 16384 Hz. 80 samples are sampled for each health condition from raw data, where each timedomain sample contains 800 data points.
Fig. 6. a) test rig, b) schematic of the gearbox, c) accelerometer location, d) broken teeth defect, e) worn teeth defect, f) worn model defect
a)
b)
c)
d)
e)
f)
The preprocessing of each sample is the same as the bearing case, 400 Fourier coefficients are obtained from each sample. 10 % samples are randomly selected for training and the rest for testing. The output dimensions ${N}^{out}$ for experiments of various networks are all tuned to 400 and the regular parameters ${\lambda}_{1}$, ${\lambda}_{2}$ are all tuned to 100.
Fig. 7. Diagnosis results of 20 trials
20 trials are carried out for each experiment, and the comparison of diagnosis results among 4 networks is depicted in Fig. 7. It shows that the averaged accuracies are mostly over 99.5 % for L1SFLOG and 99.0 % for L2SFLOG, and both of them also hold quite low accuracy fluctuations. By comparison, it indicates that the new activation function can extract better features than the softabsolute function, and L1 regularization is effective in improving the performance of standard sparse filtering.
Table 5. Training and testing accuracies of 6 networks
Method

Training accuracy (%)

Testing accuracy (%)

SFABS

100

88.35±7.85

SFLOG

100

94.51±6.19

L1SFLOG

100

99.13±0.50

L2SFLOG

100

98.80±0.56

L1SFABS

100

94.82±4.24

L2SFABS

100

98.05±0.61

Note: The format of the result is: mean value ± standard deviation

To quantitatively show the efficiency of L1SFLOG and L2SFLOG in preventing overfitting, the averaged accuracies and standard deviations of each network are shown in Table 5. It shows that the improved networks all perform better than SFABS, and L1SFLOG holds the highest classification accuracy and the lowest standard deviation once again. Meanwhile, it also presents that the difference between averaged training accuracy and testing accuracy of L1SFLOG is the smallest, namely 0.87 %. Furthermore, it is worth mentioning that the standard deviation is 0.5 % for L1SFLOG and 0.56 % for L2SFLOG, which means L1SFLOG and L2SFLOG are stable, and L1SFLOG is superior to L2SFLOG.
Experiments on these networks show that proposed L1SFLOG can obtain much higher testing accuracy, stability and generalization ability than the others, and networks with the new activation function perform better than the ones without. It also demonstrates that although inferior to L1regularized sparse filtering, the L2regularized sparse filtering is also effective in improving some aspects of performance.
We focus on the process of feature calculation and investigate why L1SFLOG holds the best performance. As sparse filtering has no bias term, if we use $F\left(x\right)=x$ as the activation function, then we can describe the feature calculation process as follows:
Fig. 8. Columns and rows of weight matrices learned from the bearing dataset: a) L1SFLOG, b) L2SFLOG, c) SFLOG
a)
b)
c)
We regard each column of $\mathbf{W}$ as a filter [2] and the features are extracted by the inner product of weight matrix columns and sample ${\mathbf{x}}^{i}$. When samples are fixed, we want the filters to extract discriminative features and do not want these features to be extracted repeatedly, which means that the sparsity of the weight matrix should be ensured in a certain degree. Heuristically, as sparse filtering achieves the sparsity of the feature matrix by implementing L1 norm in the objective function, we utilize L1 regularization to constrain the sparsity of the weight matrix. Similarly, when the samples are frequency spectra, we propose defining the population sparsity and lifetime sparsity of weight matrix as follows:
(1) Population sparsity, which refers that the columns of the weight matrix have very narrow passing bandwidths. It should be noticed that different from common filters, each column of the weight matrix can have several main lobes, which means that it can fuse several weakly discriminative features by composing them into a prominent one.
(2) Lifetime sparsity, which refers that the rows of the weight matrix should be sparsely activated, which means the same frequency band will not be filtered repeatedly. It prevents the discriminative features from being extracted by many times and reduces the information redundancy in feature representation. As a result, more discriminative features are mined.
To further present the effectiveness of sparsity in the weight matrix, typical rows and columns of weight matrices learned from L1SFLOG, L2SFLOG, SFLOG are presented in Fig. 8 and Fig. 9. Firstly, it can be seen that all the columns of weight matrix learned by each network serve as bandwidth filters for frequencydomain samples and the ones learned by L1SFLOG have the best frequencydomain filtering property. By comparison, it can be concluded that L1SFLOG holds this property because its weight matrix is constrained by L1 regularization. Secondly, it also presents that the columns and rows of weight matrices learned by L1SFLOG are all highly sparse, which means that the weight matrix has the population sparsity and lifetime sparsity defined above. Therefore, it is validated that L1 regularization is effective in constraining the sparsity of the weight matrix in sparse filtering. Furthermore, it can also be found that the columns of weight matrix learned by L2SFLOG are smoother than SFLOG.
Fig. 9. Columns and rows of weight matrices learned from the gearbox dataset: a) L1SFLOG, b) L2SFLOG, c) SFLOG
a)
b)
c)
A new feature extraction method called L1regularized sparse filtering is proposed to extract more discriminative features and improve generalization ability of standard sparse filtering through constraining the sparsity of the weight matrix. Based on FFT, L1regularized sparse filtering and softmax regression, a novel threestage fault diagnosis method is developed. Experiments on a bearing dataset and a gearbox dataset demonstrate that L1SFLOG is more effective in preventing overfitting than L2SFLOG, SFLOG and SFABS. Meanwhile, the proposed L1SFLOG can obtain much higher diagnosis accuracy with less training samples and be more stable.
We propose adopting different activation functions in feedforward and weight matrix optimization for considering their different requirements of antinoise and nonsaturating properties. Experiments show that this operation improves the testing accuracy and the diagnosis network stability a lot. Furthermore, we give an interpretation of the weight matrix and define two sparse properties of weight matrix. The significance of the two properties is validated by experiments. It shows that the weight matrix learned by L1SFLOG has the sparse properties described above, which accounts for its superiority.
The research was supported by National Natural Science Foundation of China (51675262) and also supported by the Project of National Key Research and Development Plan of China “New EnergySaving Environmental Protection Agricultural Engine Development” (2016YFD0700800) and the Advance Research Field Fund Project of China (6140210020102).