CN112541233A - Rotary machine fault diagnosis method based on improved convolutional neural network - Google Patents
Rotary machine fault diagnosis method based on improved convolutional neural network Download PDFInfo
- Publication number
- CN112541233A CN112541233A CN201910890191.6A CN201910890191A CN112541233A CN 112541233 A CN112541233 A CN 112541233A CN 201910890191 A CN201910890191 A CN 201910890191A CN 112541233 A CN112541233 A CN 112541233A
- Authority
- CN
- China
- Prior art keywords
- data
- layer
- fault
- training
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a rotary machine fault diagnosis method based on an improved convolutional neural network, which comprises the following steps of: (1) acquiring one-dimensional time series fault data to obtain an original fault data set; (2) performing preprocessing operation on the acquired data of the original fault data set, wherein the preprocessing comprises the following steps: standardization, data truncation and data reconstruction; (3) dividing the preprocessed samples of each type of fault into a training set, a verification set and a test set; (4) establishing an improved convolutional neural network fault diagnosis model, wherein the model comprises an input layer, a feature extraction layer, a dimension reduction parameter reduction layer and a softmax classification output layer, and the dimension reduction parameter reduction layer comprises a 1 × 1 transition convolution layer and a global mean pooling layer; (5) training and testing the model; the diagnosis model can automatically extract and diagnose the fault data without any manual feature extraction operation, so that people can diagnose the fault of the rotating machine more conveniently and quickly.
Description
Technical Field
The invention belongs to the technical field of fault diagnosis and detection of rotary machines, and particularly relates to a fault diagnosis method of a rotary machine based on an improved convolutional neural network.
Background
With the rapid development of modern industrial technologies, rotary machinery equipment is rapidly developed towards high speed, precision, automation and integration, and the rotary machinery mainly comprises power devices such as diesel engines, steam turbines, engines, motors and the like, and also comprises rotating parts such as bearings, spindles and the like. With the diversification of working environments of rotary machines, especially when the rotary machines continuously run for a long time in complicated and variable working environments, various faults are easy to occur due to the workload, variable load, saline-alkali corrosion, high temperature and the like. If the fault cannot be diagnosed and removed effectively and timely, once the fault damage spreads under the strong coupling state, a great loss may be brought. Therefore, fault diagnosis and state online monitoring of the rotating machine are important for guaranteeing safe and stable operation of the equipment.
Before the invention, the current products or methods for fault diagnosis and state monitoring of rotating machinery on the market are rare, and more still traditional 'after repair', 'planned repair' and 'timed maintenance' modes are applied, the method is often very low in efficiency and has no intelligence, in addition, the traditional maintenance mode of regularly maintaining and regularly replacing parts according to experience and estimating the service life of the parts by experience is easy to cause waste and misjudgment, and brings potential safety hazards, so that the requirements of technical personnel on intelligent fault diagnosis and online state monitoring cannot be met.
With the continuous rise of machine learning research, fault diagnosis methods based on artificial intelligence gradually become a hotspot of research in the field of fault diagnosis. The intelligent diagnosis method generally comprises three steps of feature extraction, feature selection and fault classification, wherein the feature extraction is to perform time domain, frequency domain and time-frequency domain processing on original data signals acquired by a sensor, and extract useful signals representing fault features for subsequent fault identification. The feature selection is to remove features with low sensitivity and low utility from the extracted features, so as to reduce the number of the features, and the commonly used methods mainly comprise a Principal Component Analysis (PCA) method, an Independent Component Analysis (ICA) method and the like. The fault classification is to input the well-selected characteristics into a fault classifier to perform pattern recognition, and finally output a fault classification result through repeated iterative training, so that a BP neural network (BPNN), a Support Vector Machine (SVM) and a K nearest neighbor method (KNN) are widely used as rear-end fault classifiers and achieve a good effect. Through years of application verification, the three methods have poor feature extraction capability due to the shallow network structure, and are difficult to adapt to application in the present large data sample environment. Currently, the experts in the industry generally adopt a method of combining manual feature extraction and shallow machine learning to perform intelligent fault diagnosis. Although the above-mentioned existing intelligent fault diagnosis method has been applied and achieved certain results, three major disadvantages are still highlighted: (1) various advanced signal processing technologies are required to be mastered for feature extraction, feature selection is required to be completed by depending on experience and professional knowledge of engineers, and the method has high subjectivity and blindness; (2) the feature extraction is mainly used for solving the specific fault problem, has poor universality and is difficult to complete in the mass data sample environment; (3) the manually extracted fault features are not comprehensive, and the features reflecting the tiny faults are easy to be deleted by mistake and covered by noise. The main reason for the above defects is that most of network models used in the existing intelligent fault diagnosis method are shallow structures, and the feature extraction capability of the network models is weak.
Convolutional Neural Network (CNN) is one of the important branches of deep learning, has a strong feature extraction capability, and is mainly applied to image recognition at present. In recent years, partial scholars apply CNN to the field of fault diagnosis, although a convolutional neural network is used, the rotary mechanical fault diagnosis method disclosed by Chinese invention patent (a rotary mechanical fault feature intelligent diagnosis method based on a deep convolutional neural network structure, application number: CN 201810240234.1) still has two defects, one of which is that the method still needs to use the traditional feature extraction method (short-time Fourier transform) to carry out feature extraction pretreatment on original fault data, and the strong feature extraction capability of the convolutional neural network is not fully utilized, so that the further improvement of the fault diagnosis effect is limited; the traditional convolutional neural network comprises a 2-3 layers of full-connection layer network structure part which is usually positioned between the last pooling layer and the Softmax classification output layer, the training parameters generated by the existence of the full-connection layer occupy 80% -90% of the total parameters of the CNN, the defect is that the CNN is greatly offset by the advantage of pooling dimension reduction and parameter reduction, the structure of the full-connection layer occupies too much calculation resources, meanwhile, the CNN model is easy to train and fit, particularly the full-connection layer comprising a plurality of hidden layers, the CNN model parameters are exponentially increased along with the increase of the number of the full-connection layers, and therefore, the traditional CNN model is not beneficial to real-time rapid diagnosis of faults because the excessive parameter quantity leads to time consumption in testing when used for online diagnosis of the overlong faults.
Disclosure of Invention
Aiming at the defects or needs to be improved in the prior art, the invention provides a rotary machine fault diagnosis method based on an improved convolutional neural network, which is an improved design aiming at the structure of the traditional convolutional neural network. The end-to-end method structure enables the whole diagnosis process to be free from any manual feature extraction, and overcomes the defect that the existing fault diagnosis method excessively depends on expert prior knowledge. The invention can automatically diagnose the fault of the rotating machine and monitor the working state of the rotating machine on line in real time, so that technical personnel and equipment maintenance personnel can better master the current operating condition of the equipment, and the technical personnel can more flexibly and conveniently diagnose the fault of the rotating machine and monitor the operating state.
In order to achieve the above object, the present invention provides a method for diagnosing a fault of a rotating machine based on an improved convolutional neural network, which comprises the following steps (1) to (7).
(1) The method comprises the steps of collecting one-dimensional time series fault data, obtaining an original fault data set, arranging a vibration acceleration sensor on a rotary machine, and collecting the one-dimensional time series vibration acceleration original fault data of the rotary machine in a plurality of fault states and normal states by using the vibration acceleration sensor to form the original fault data set, wherein the data set comprises 1 normal state and s-1 fault states.
(2) Performing data preprocessing operation on all one-dimensional time series fault data in the acquired original fault data set, wherein the preprocessing comprises the following steps: standardization, data truncation and data format reconstruction; first, for the collectedThe method comprises the following steps of carrying out standardized data processing on one-dimensional time series fault data, and converting magnitude of values of all data points into 0-1, wherein the standardized method comprises the following steps:X={x i }=(x i -x min)/(x max-x min) (ii) a Secondly, equally dividing and truncating the standardized one-dimensional time sequence data to obtain h equal-length one-dimensional time sequence small data segments (assuming that the length of each small data segment is k data points, and the value of k is 100-10000), wherein each small data segment is a sample, and s original data segments of the one-dimensional time sequence can obtain s multiplied by h fault samples; then, format reconstruction is carried out on the one-dimensional time sequence data form of each small data segment with the length of k to form a two-dimensional characteristic map [ m, n](assuming that the size of each feature map is m × n = k), s × h fault sample two-dimensional feature maps can be obtained according to the method.
(3) Dividing h samples of each type of preprocessed faults into a training set, a verification set and a test set, wherein the dividing method comprises the following steps: randomly selecting 30% of samples from the h samples of each type of fault after pretreatment as a test set, and randomly taking 80% of the rest 70% of samples to be classified as a training set and 20% as a verification set.
(4) Establishing an improved convolutional neural network fault diagnosis model, inputting data samples of a training set into the CNN diagnosis model for training and learning model parameters, and repeatedly executing forward propagation and backward propagation iterative computation processes to adjust parameters.
(5) And (3) verifying the diagnosis accuracy of the CNN model in the training process in real time by using sample data of the verification set, verifying whether the diagnosis result has an overfitting phenomenon, indicating that the model parameter training is normal when the accuracy on the training set and the accuracy on the verification set both increase along with the increase of the number of training rounds and the accuracy on the training set is higher than that on the verification set, continuing to execute the step (6), stopping the model training when the accuracy on the training set and the accuracy on the verification set both increase along with the increase of the number of training rounds and the accuracy on the training set starts to be lower than that on the verification set, indicating that the model parameter training has overfitting at the moment, skipping to the step (4), revising the improved CNN model parameters, and repeating the steps until the proper CNN model hyperparameters and the verification accuracy are obtained.
(6) And when the verification accuracy reaches a set target value or training iteration times, finishing the training of the model, and simultaneously storing the optimal CNN diagnosis model parameters.
(7) And finally, inputting the sample data of the test set into the trained improved CNN fault diagnosis model to complete the final test, and obtaining the final fault diagnosis result.
The invention designs an improved convolutional neural network fault diagnosis model which comprises an input layer, a feature extraction layer, a dimension reduction parameter reduction layer and a softmax classification output layer, wherein the feature extraction layer comprises a first convolutional layer, a first pooling layer, a second convolutional layer and a second pooling layer, the dimension reduction parameter reduction layer is arranged between the feature extraction layer and the softmax classification output layer and is used for replacing a full-connection network layer of a traditional convolutional neural network, the dimension reduction parameter reduction layer comprises a transition convolutional layer with a convolution kernel of 1 multiplied by 1 and a global mean pooling layer, the 1 multiplied by 1 transition convolutional layer is used for receiving an output feature map of the second pooling layer, the output feature map of the global mean pooling layer is directly used as the input of the softmax classification output layer, and the layers are sequentially connected end to form a complete improved convolutional neural network diagnosis model.
The invention designs an input layer of the improved convolutional neural network fault diagnosis model, wherein the input layer is used for receiving original fault one-dimensional time sequence data of a rotary machine, and the input layer is characterized in that: and the input layer carries out standardization, data truncation and data format reconstruction operations on the acquired one-dimensional time series fault data, and finally transmits the reconstructed two-dimensional characteristic diagram to a first convolution layer of a characteristic extraction layer of the improved convolution neural network diagnosis model.
The invention designs that the construction method of the two-dimensional characteristic diagram is set as follows: dividing a data segment containing k data points into m parts, wherein each part contains n data points, and the arrangement sequence is as follows: and placing the 1 st n data point on the 1 st line, the 2 nd n data point on the 2 nd line, the 3 rd n data point on the 3 rd line, and sequentially sequencing, … …, wherein the mth n data point is placed on the mth line, thereby obtaining an m multiplied by n two-dimensional characteristic diagram.
Compared with the prior art, the rotary machine fault diagnosis method based on the improved convolutional neural network provided by the invention has the advantages that the 1 x 1 transition convolutional layer and the global mean pooling layer replace the full-connection layer network structure of the traditional CNN model, the training parameters of the traditional CNN full-connection layer can be greatly reduced, the calculation time of the model is reduced on the premise of ensuring the diagnosis accuracy, the diagnosis speed and efficiency of the model are effectively improved, and the rotary machine fault diagnosis method based on the improved convolutional neural network is more suitable for rapid diagnosis of faults.
Meanwhile, the invention does not need any manual feature extraction operation and does not need an operator to master various complex advanced signal processing technologies, the invention can directly input the original fault data into the improved convolutional neural network fault diagnosis model, the diagnosis model can automatically carry out data format reconstruction, feature automatic extraction and fault diagnosis on the one-dimensional time series fault data of the rotating machine, and the final diagnosis result is automatically output, the whole diagnosis process belongs to end-to-end operation, so that the invention has better operability and lower use threshold, and the fault diagnosis practitioner can more conveniently and quickly diagnose the rotating machine fault.
Drawings
FIG. 1 is a flow chart of a method of fault diagnosis for a rotating machine based on an improved convolutional neural network.
Fig. 2 is a schematic structural diagram of the improved convolutional neural network model of fig. 1.
FIG. 3 is a fault diagnosis framework of a preferred embodiment of the present invention.
FIG. 4 is a schematic diagram of the reconstruction of one-dimensional time series data according to the present invention.
FIG. 5 is a schematic diagram of a rotating machine fault data generation test stand in accordance with a preferred embodiment of the present invention.
Detailed Description
For a better understanding of the present invention, reference is made to the following description taken in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Referring to fig. 1 to 5, a method for diagnosing a fault of a rotating machine based on an improved convolutional neural network according to a preferred embodiment of the present invention includes steps one to seven.
Step one, collecting a fault vibration signal of a one-dimensional time sequence of a rotary machine. Specifically, a vibration acceleration sensor is arranged on the rotary machine, and the vibration sensor is used for collecting one-dimensional time series original fault data of the rotary machine in a normal state and s-1 fault states, so that an original fault data set with s one-dimensional time series data segments is formed.
Step two, performing data preprocessing operation on all the one-dimensional time series fault data in the acquired original fault data set, wherein the preprocessing comprises the following steps: standardization, data truncation and data format reconstruction; firstly, carrying out standardized data processing on the collected one-dimensional time series fault data, and converting the magnitude of the values of all data points into 0-1, wherein the standardized method comprises the following steps:X={x i }=(x i -x min)/(x max-x min) (ii) a Secondly, equally dividing and truncating the normalized one-dimensional time sequence data to obtain h equal one-dimensional time sequence data segments, for example, if each original data segment of the s one-dimensional time sequence original data segments comprises 100000 data points, 200 small data segments can be obtained if the original data segments are equally divided by 200 parts, each small data segment is a sample, and s × h fault samples can be obtained from the s one-dimensional time sequence original data segments; finally, performing data format reconstruction on each small data segment, according to the method shown in fig. 4, assuming that the length of each data segment is 500 data points, equally dividing the data segment containing 500 data points into 25 shares, each share containing 20 data points, and arranging in the following order: the 1 st 20 data points are placed on line 1, the 2 nd 20 data points on line 2, the 3 rd 20 data points on line 3,sorting in sequence, … …, the 25 th 20 th data point is placed in the 25 th row, thus obtaining a 25X 20 characteristic diagram, namely the reconstructed data form is marked as a two-dimensional characteristic diagram [25,20]](the size of each feature map is 25 × 20= 500), s × h fault sample two-dimensional feature maps can be obtained according to this method.
Step three, dividing h samples of each type of preprocessed faults into a training set, a verification set and a test set, wherein the dividing method comprises the following steps: randomly selecting 30% of samples from the h samples of each type of fault after pretreatment as a test set, and randomly taking 80% of the rest 70% of samples to be classified as a training set and 20% as a verification set.
Step four, establishing an improved convolutional neural network fault diagnosis model, wherein the improved convolutional neural network fault diagnosis model comprises an input layer 1, a feature extraction layer 2, a dimensionality reduction parameter layer 3 and a softmax classification output layer 4 as shown in fig. 2, the feature extraction layer 2 comprises a first convolutional layer 21, a first pooling layer 22, a second convolutional layer 23 and a second pooling layer 24, a fully-connected network layer for replacing the traditional convolutional neural network is arranged between the feature extraction layer 2 and the softmax classification output layer 4, the dimensionality reduction parameter layer 3 comprises a transition convolutional layer 31 with a convolution kernel of 1 × 1 and a global mean value pooling layer 32, the transition convolutional layer 31 with a convolution kernel of 1 × 1 is used for receiving the output feature map of the second pooling layer 24, the output feature map of the global mean value pooling layer 32 is directly used as the input of the softmax classification output layer 4, the input layer 1 is used for receiving the original fault one-dimensional time of the rotary machine, the input layer carries out standardization, data truncation and data format reconstruction operations on the acquired one-dimensional time series fault data, and finally the reconstructed two-dimensional feature map 10 is transmitted to a first convolution layer 21 of a feature extraction layer 2 of the improved convolutional neural network diagnostic model, wherein the input layer 1, the first convolution layer 21, a first pooling layer 22, a second convolution layer 23, a second pooling layer 24, a 1 × 1 transition convolution layer 31, a global mean pooling layer 32 and a softmax classification output layer 4 are sequentially connected end to form the complete improved convolutional neural network diagnostic model; inputting the data samples of the training set into the CNN diagnostic model for training and learning model parameters, and repeatedly executing the iterative computation process of forward propagation and backward propagation to adjust parameters.
And step five, carrying out real-time verification on the diagnosis accuracy rate of the CNN model in the training process by using the sample data of the verification set, verifying whether the diagnosis result has an overfitting phenomenon, indicating that the model parameter training is normal when the accuracy rates of the training set and the verification set increase along with the increase of the number of training rounds and the accuracy rate of the training set is higher than that of the verification set, continuing to execute the step (6), indicating that the model parameter training has overfitting when the accuracy rates of the training set and the verification set increase along with the increase of the number of training rounds and the accuracy rate of the training set is lower than that of the verification set, stopping the model training, skipping to the step (4), revising the improved CNN model parameters, and repeatedly executing the steps until the appropriate CNN model hyperparameter and the verification accuracy rate are obtained.
And step six, when the verification accuracy reaches a set target value or training iteration times, finishing the training of the model, and simultaneously storing the optimal CNN diagnosis model parameters.
And step seven, finally, inputting the sample data of the test set into the trained improved CNN fault diagnosis model to finish the final test, and obtaining the final fault diagnosis result.
In fig. 3, three modules are included: the bottom layer is a fault data acquisition module, the top layer is a fault diagnosis result output module, and the middle layer is an improved CNNs module.
In this embodiment, in order to further illustrate the feasibility and effectiveness of the fault diagnosis method for a rotating machine based on an improved convolutional neural network, in the fault diagnosis for the rotating machine, in this embodiment, the application example is further illustrated and verified with the bearing data of the bearing laboratory (as shown in fig. 5) of the electrical engineering laboratory of the university of kasceskun storage, usa.
As shown in fig. 5, in the experiment of this embodiment, a vibration acceleration sensor is installed at a position right above a driving end of a motor to collect vibration acceleration signals of a faulty bearing, 12 mechanical health states consisting of 11 bearing fault states and 1 normal state are classified and diagnosed in this embodiment, each health state type includes 100,000 data points, 200 parts are equally divided, each part includes 500 data points, and one-dimensional time series data of 500 data points are reconstructed into a two-dimensional characteristic diagram of [25,20] according to the data reconstruction method described in fig. 4, that is, each fault type includes 200 samples of the two-dimensional characteristic diagram of [25,20], as shown in table 1.
Table 1: failure data set of experimental bearings.
Since 30% of 200 samples of each fault type were taken as a test set (200 × 0.3= 60), 20% of the remaining 70% were taken as a verification set (200 × 0.7 × 0.2= 28), and the remaining 80% were taken as a training set (200 × 0.7 × 0.8= 112), the present experiment included 12 state types, and thus the total number of training set samples was 1344 (12 × 112 ═ 1344), the total number of verification set samples was 336 (12 × 28 ═ 336), and the total number of test set samples was 720 (12 × 60 ═ 720).
In this embodiment, the selected improved convolutional neural network diagnostic model is shown in fig. 2, and includes a first convolutional layer 21, a first pooling layer 22, a second convolutional layer 23, a second pooling layer 24, a 1 × 1 transition convolutional layer 31, a global mean pooling layer 32, and a softmax classification output layer 4, which are connected in sequence, and the detailed fault diagnostic model hyper-parameters of this experiment are shown in table 2.
Table 2: the improved CNNs model is hyper-parametric.
In this embodiment, the data processing flow of the improved CNNs fault diagnosis model designed in this experiment shown in table 2 is as follows: the feature map format of the input sample is [25,20], the first convolution layer 21 respectively performs the same convolution operation on the input feature maps [25,20] by adopting 64 convolution kernels of 2 × 2, and an output feature map of 64 channels is obtained: [25,20,64 ]; then, performing pooling operation on the output characteristic diagram [25,20 and 64] of the first convolution layer 21 by the first pooling layer 22, wherein the pooling core of the first pooling layer 22 is 2 x 2, and obtaining a characteristic diagram [12,10 and 64] after the pooling operation; then, the second convolution layer 23 performs a second convolution operation on the output feature map [12,10,64] of the first pooling layer 22, and the second convolution layer 23 respectively performs a same convolution operation on the feature maps [12,10,64] by using 32 2 × 2 convolution kernels to obtain an output feature map of 32 channels: [12,10,32 ]; then, performing pooling operation on the output characteristic diagram [12,10,32] of the second convolution layer 23 by the second pooling layer 24, wherein the pooling core of the second pooling layer 24 is 2 × 2, and obtaining a characteristic diagram [6,5,32] after the pooling operation; then, a third convolution operation is performed on the output feature maps [6,5,32] of the second pooling layer 24 by the 1 × 1 transition convolution layer 31, and the third convolution layer 31 performs a same convolution operation on the feature maps [6,5,32] by using 12 1 × 1 convolution kernels respectively to obtain 12-channel output feature maps: [6,5,12 ]; then, a global mean pooling layer 32 is arranged behind the third convolutional layer 31, the global mean pooling layer 32 performs global mean pooling calculation on the output feature maps [6,5 and 12] of the third convolutional layer 31 by adopting 12 pooling cores of 6 × 5, and all values in the feature maps of each [6 and 5] are subjected to a global mean value [1 and 12 ]; finally, the feature vectors of [1,12] are input into the Softmax classification output layer 4 to complete the final fault classification, as shown in table 2.
The output of the improved convolutional neural network adopts a 'One-hot' coding method, and is defined as follows: if the input signature corresponds to the 1 st health state, the output is [1,0,0,0,0,0,0,0,0, 0], if the input signature corresponds to the 2 nd health state, the output is [0,1,0,0,0,0,0,0,0, 0], and so on, and if the input data corresponds to the 12 th health state, the output is [0,0,0,0,0,0,0,0,0, 1 ].
Compared with the traditional CNN with the same scale, the rotary machine fault diagnosis method based on the improved convolutional neural network provided by the invention has the advantages that the parameter quantity is obviously reduced, as shown in Table 3, the total number of model parameters of the traditional CNNs-full-connection network is 92,140, and the number of model parameters provided by the invention is only 9,096.
Table 3: and (5) a CNN model training parameter comparison table.
| Name (R) | Improved CNNs | CNNs full |
| Input layer | ||
| 0 | 0 | |
| |
320 | 320 |
| |
0 | 0 |
| |
8244 | 8224 |
| |
0 | 0 |
| 1 x 1 transitional convolutional layer | 396 | Is free of |
| Global |
0 | Is free of |
| 1 st full connection layer | Is free of | 82048 |
| 2 nd full connection layer | Is free of | 1548 |
| |
0 | 0 |
| Total amount of ginseng | 9096 | 92140 |
Compared with the traditional CNN with the same scale, the rotary machine fault diagnosis method based on the improved convolutional neural network has the advantages that the accuracy is improved, and the test time and the training time of the fault are reduced, as shown in Table 4. As can be seen from comparison of table 4, the performance of the improved CNNs is significantly improved compared to the conventional fully-connected CNN. In terms of accuracy: the accuracy rate of the improved CNNs reaches 99.04 percent, while the accuracy rate of the traditional CNNs is 98.75 percent; in terms of time, the improved CNNs greatly reduce the quantity of model parameters due to the removal of a full connection part, the training time is obviously reduced, and more importantly, the testing time is only 0.198 second, which is of great significance to the application of the method to online rapid diagnosis and monitoring of faults.
Table 4: and comparing the fault diagnosis results.
| Name (R) | Accuracy of test | Training time | Time of measurement |
| Improved CNNs methods | 99.04% | 229.53 seconds | 0.198 second |
| Traditional CNNs method | 98.75% | 248.96 seconds | 0.279 second |
According to the rotary machine fault diagnosis method based on the improved convolutional neural network, the 1 x 1 transition convolutional layer and the global mean pooling layer replace a full-connection layer network structure of a traditional CNN model, the training parameters of the traditional CNN full-connection layer can be greatly reduced, the calculation time of the model is reduced on the premise of ensuring the diagnosis accuracy, the diagnosis speed and efficiency of the model are effectively improved, and the rotary machine fault diagnosis method is more suitable for rapid diagnosis of faults.
The invention does not need any manual feature extraction operation and does not need an operator to master various complex advanced signal processing technologies, the original fault data can be directly input into an improved convolutional neural network fault diagnosis model, the diagnosis model can automatically carry out data format reconstruction, feature automatic extraction and fault diagnosis on the one-dimensional time sequence fault data of the rotary machine, the final diagnosis result is automatically output, the whole diagnosis process belongs to end-to-end operation, and the invention has better operability and lower use threshold, so that fault diagnosis practitioners can more conveniently and quickly diagnose the rotary machine faults.
It should also be understood that the above description is only a preferred embodiment of the present invention, and not intended to limit the present invention, and that the technical matters in the present invention are included in the technical contents of the present invention.
Claims (4)
1. A rotary machine fault diagnosis method based on an improved convolutional neural network is characterized by comprising the following steps:
(1) acquiring one-dimensional time series fault data, acquiring an original fault data set, arranging a vibration acceleration sensor on a rotary machine, and acquiring one-dimensional time series vibration acceleration original fault data of the rotary machine in a plurality of fault states and normal states by using the vibration acceleration sensor to form an original fault data set, wherein the data set comprises 1 normal state and s-1 fault states;
(2) performing data preprocessing operation on all one-dimensional time series fault data in the acquired original fault data set, wherein the preprocessing comprises the following steps: standardization, data truncation and data format reconstruction; firstly, carrying out standardized data processing on the collected one-dimensional time series fault data, and converting the magnitude of the values of all data points into 0-1, wherein the standardized method comprises the following steps:X={x i }=(x i -x min)/(x max-x min) (ii) a Secondly, equally dividing and truncating the standardized one-dimensional time sequence data to obtain h equal-length one-dimensional time sequence small data segments (assuming that the length of each small data segment is k data points, and the value of k is 100-10000), wherein each small data segment is a sample, and s original data segments of the one-dimensional time sequence can obtain s multiplied by h fault samples; then, format reconstruction is carried out on the one-dimensional time sequence data form of each small data segment with the length of k to form a two-dimensional characteristic map [ m, n](assuming that the size of each feature map is m × n = k)According to the method, s multiplied by h fault sample two-dimensional characteristic graphs can be obtained;
(3) dividing h samples of each type of preprocessed faults into a training set, a verification set and a test set, wherein the dividing method comprises the following steps: randomly selecting 30% of samples from the h samples of each type of fault after pretreatment as a test set, and randomly taking out 80% of the rest 70% of samples to be classified as a training set and 20% of samples to be a verification set;
(4) establishing an improved convolutional neural network fault diagnosis model, inputting data samples of a training set into a CNN diagnosis model for training and learning model parameters, and repeatedly executing forward propagation and backward propagation iterative computation processes to adjust parameters;
(5) verifying the diagnosis accuracy of the CNN model in the training process in real time by using sample data of a verification set, verifying whether the diagnosis result has an overfitting phenomenon, indicating that the model parameter training is normal when the accuracy on the training set and the accuracy on the verification set both increase along with the increase of the number of training rounds and the accuracy on the training set is higher than that on the verification set, continuing to execute the step (6), stopping the model training when the accuracy on the training set and the accuracy on the verification set both increase along with the increase of the number of training rounds and the accuracy on the training set starts to be lower than that on the verification set, indicating that the model parameter training has overfitting at the moment, skipping to the step (4), revising the improved CNN model parameters, and repeating the steps until proper CNN model hyperparameters and verification accuracy are obtained;
(6) when the verification accuracy reaches a set target value or training iteration times, ending the training of the model, and simultaneously storing the optimal CNN diagnosis model parameters;
(7) and finally, inputting the sample data of the test set into the trained improved CNN fault diagnosis model to complete the final test, and obtaining the final fault diagnosis result.
2. The method according to claim 1, wherein the improved convolutional neural network fault diagnosis model comprises an input layer, a feature extraction layer, a dimensionality reduction parameter reduction layer and a softmax classification output layer, the feature extraction layer comprises a first convolution layer, a first pooling layer, a second convolution layer and a second pooling layer, a fully-connected network layer for replacing a traditional convolutional neural network is arranged between the feature extraction layer and the softmax classification output layer, the dimensionality reduction parameter reduction layer comprises a transition convolution layer with a convolution kernel of 1 x 1 and a global mean value pooling layer, the transition convolution layer with 1 x 1 is used for receiving an output feature map of the second pooling layer, an output feature map of the global mean value pooling layer is directly used as an input of the softmax classification output layer, and the first convolution layer, the second convolution layer, the first convolution layer, the second pooling layer, the third convolution layer and the third convolution layer are sequentially connected, and forming a complete improved convolutional neural network diagnosis model.
3. The improved convolutional neural network-based rotary machine fault diagnosis method according to claim 2, wherein the improved convolutional neural network fault diagnosis model comprises an input layer, the input layer is used for receiving original fault one-dimensional time series data of the rotary machine, the input layer performs the operations of standardization, data truncation and data format reconstruction on the collected one-dimensional time series fault data, and finally the reconstructed two-dimensional feature map is transmitted to the first convolutional layer of the feature extraction layer of the improved convolutional neural network diagnosis model.
4. The method for diagnosing faults of rotary machines based on the improved convolutional neural network as claimed in claim 1, 2 or 3, wherein the construction method of the two-dimensional feature map is set as follows: dividing a data segment containing k data points into m parts, wherein each part contains n data points, and the arrangement sequence is as follows: and placing the 1 st n data point on the 1 st line, the 2 nd n data point on the 2 nd line, the 3 rd n data point on the 3 rd line, and sequentially sequencing, … …, wherein the mth n data point is placed on the mth line, thereby obtaining an m multiplied by n two-dimensional characteristic diagram.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910890191.6A CN112541233A (en) | 2019-09-20 | 2019-09-20 | Rotary machine fault diagnosis method based on improved convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910890191.6A CN112541233A (en) | 2019-09-20 | 2019-09-20 | Rotary machine fault diagnosis method based on improved convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112541233A true CN112541233A (en) | 2021-03-23 |
Family
ID=75012485
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910890191.6A Withdrawn CN112541233A (en) | 2019-09-20 | 2019-09-20 | Rotary machine fault diagnosis method based on improved convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112541233A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113092044A (en) * | 2021-03-31 | 2021-07-09 | 东南大学 | Rotary machine fault diagnosis method based on weighted level visible graph |
| CN113177577A (en) * | 2021-04-01 | 2021-07-27 | 上海吞山智能科技有限公司 | Bearing fault diagnosis method based on improved convolutional neural network |
| CN113469256A (en) * | 2021-07-06 | 2021-10-01 | 吉林大学重庆研究院 | Gear part mechanical damage node prediction method |
| CN114088405A (en) * | 2021-11-10 | 2022-02-25 | 中国人民解放军陆军炮兵防空兵学院 | Engine fault diagnosis method of CNN fault diagnosis model based on spectrogram |
| CN114233581A (en) * | 2021-12-13 | 2022-03-25 | 山东神戎电子股份有限公司 | Intelligent patrol alarm system for fan engine room |
| CN114332531A (en) * | 2021-12-23 | 2022-04-12 | 南京航空航天大学 | Mechanical equipment rapid fault diagnosis method and device based on 2D-CNN and t-SNE |
| CN115062648A (en) * | 2022-06-02 | 2022-09-16 | 华南理工大学 | Fault diagnosis method, system and device for rotary machine and storage medium |
| CN116008733A (en) * | 2023-03-21 | 2023-04-25 | 成都信息工程大学 | Single-phase grounding fault diagnosis method based on integrated deep neural network |
| CN116861343A (en) * | 2023-07-10 | 2023-10-10 | 广东德尔智慧科技股份有限公司 | Bearing fault diagnosis method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108896296A (en) * | 2018-04-18 | 2018-11-27 | 北京信息科技大学 | A kind of wind turbine gearbox method for diagnosing faults based on convolutional neural networks |
-
2019
- 2019-09-20 CN CN201910890191.6A patent/CN112541233A/en not_active Withdrawn
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108896296A (en) * | 2018-04-18 | 2018-11-27 | 北京信息科技大学 | A kind of wind turbine gearbox method for diagnosing faults based on convolutional neural networks |
Non-Patent Citations (1)
| Title |
|---|
| GONG WENFENG等: ""A Novel Deep Learning Method for Intelligent Fault Diagnosis of Rotating Machinery Based on Improved CNN-SVM and Multichannel Data Fusion"", 《SENSORS》, pages 1 - 37 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113092044A (en) * | 2021-03-31 | 2021-07-09 | 东南大学 | Rotary machine fault diagnosis method based on weighted level visible graph |
| CN113177577A (en) * | 2021-04-01 | 2021-07-27 | 上海吞山智能科技有限公司 | Bearing fault diagnosis method based on improved convolutional neural network |
| CN113469256A (en) * | 2021-07-06 | 2021-10-01 | 吉林大学重庆研究院 | Gear part mechanical damage node prediction method |
| CN114088405A (en) * | 2021-11-10 | 2022-02-25 | 中国人民解放军陆军炮兵防空兵学院 | Engine fault diagnosis method of CNN fault diagnosis model based on spectrogram |
| CN114233581A (en) * | 2021-12-13 | 2022-03-25 | 山东神戎电子股份有限公司 | Intelligent patrol alarm system for fan engine room |
| CN114332531A (en) * | 2021-12-23 | 2022-04-12 | 南京航空航天大学 | Mechanical equipment rapid fault diagnosis method and device based on 2D-CNN and t-SNE |
| CN115062648A (en) * | 2022-06-02 | 2022-09-16 | 华南理工大学 | Fault diagnosis method, system and device for rotary machine and storage medium |
| CN115062648B (en) * | 2022-06-02 | 2024-06-07 | 华南理工大学 | Rotary machine fault diagnosis method, system, device and storage medium |
| CN116008733A (en) * | 2023-03-21 | 2023-04-25 | 成都信息工程大学 | Single-phase grounding fault diagnosis method based on integrated deep neural network |
| CN116861343A (en) * | 2023-07-10 | 2023-10-10 | 广东德尔智慧科技股份有限公司 | Bearing fault diagnosis method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112541233A (en) | Rotary machine fault diagnosis method based on improved convolutional neural network | |
| CN109635677B (en) | Compound fault diagnosis method and device based on multi-label classification convolutional neural network | |
| CN112067294A (en) | Rolling bearing intelligent fault diagnosis method based on deep learning | |
| CN111046945B (en) | Fault type and damage degree diagnosis method based on combined convolutional neural network | |
| CN112067916A (en) | Time series data intelligent fault diagnosis method based on deep learning | |
| CN111709448A (en) | Mechanical fault diagnosis method based on migration relation network | |
| CN113378725B (en) | Multi-scale-channel attention network-based tool fault diagnosis method, equipment and storage medium | |
| CN106017876A (en) | Wheel set bearing fault diagnosis method based on equally-weighted local feature sparse filter network | |
| CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
| CN111650453A (en) | Power equipment diagnosis method and system based on windowing characteristic Hilbert imaging | |
| CN112541511A (en) | Multi-channel time series data fault diagnosis method based on convolutional neural network | |
| CN112541510A (en) | Intelligent fault diagnosis method based on multi-channel time series data | |
| CN111190349A (en) | Method, system and medium for monitoring state and diagnosing fault of ship engine room equipment | |
| CN113505664B (en) | Fault diagnosis method for planetary gear box of wind turbine generator | |
| CN112257528A (en) | Wind power gear box fault diagnosis method based on wavelet transformation and dense connection expansion convolutional neural network | |
| CN112146882B (en) | Bearing fault diagnosis method based on transfer learning vibration signal image recognition | |
| CN114091504A (en) | Rotary machine small sample fault diagnosis method based on generation countermeasure network | |
| CN112113755A (en) | Mechanical fault intelligent diagnosis method based on deep convolution-kurtosis neural network | |
| CN111855202A (en) | Gear box fault diagnosis method and system | |
| CN113095179A (en) | Metametric learning driven helicopter planetary gearbox fault diagnosis method | |
| CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
| CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network | |
| CN114298237A (en) | Transformer fault diagnosis method, device, equipment and storage medium based on CNN | |
| CN114462480A (en) | Multi-source sensor rolling mill fault diagnosis method based on non-equilibrium data set | |
| CN113240022A (en) | Wind power gear box fault detection method of multi-scale single-classification convolutional network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20210323 |