CN111832428A - A data enhancement method applied to fault diagnosis of broken strip in cold rolling mill - Google Patents

Abstract

The invention provides a data enhancement method applied to fault diagnosis of broken strips in a cold rolling mill, belonging to the technical field of iron and steel metallurgy and fault diagnosis. The method includes: collecting time-series signals of multiple features related to fault diagnosis of broken strips in cold rolling, processing the collected time-series signals of multiple features, and generating a two-dimensional fault image set; dividing the fault image set into are training data and test data; use the training data and its corresponding labels to train the auxiliary classification generative adversarial network to obtain a generative model, wherein the trained generative model is used to generate fault images required for fault diagnosis of broken belts. The invention can improve the quality of the generated fault images while improving the training speed of the generation model, so as to facilitate the directional generation of the fault images required for the fault diagnosis of the broken belt, thereby solving the problem of insufficient fault data in the fault diagnosis of the broken belt.

Description

A data enhancement method applied to fault diagnosis of broken strip in cold rolling mill

technical field

The invention relates to the technical fields of iron and steel metallurgy and fault diagnosis, in particular to a data enhancement method applied to fault diagnosis of broken strips in a cold rolling mill.

Background technique

Modern strip cold rolling is a high-quality, high-efficiency fully automated production line with flexible production to order. Strip breakage is one of the most common faults in cold rolling production lines. Once the belt breakage failure occurs, it will cause damage to the equipment. Affecting the rolling production efficiency, it will cause a fire due to the strands, posing a great threat to personal safety. The fault diagnosis of cold rolled strip can effectively prevent accidents, restrain the decline of product quality, and maximize the potential of process operation, which has important scientific significance.

The data-driven fault diagnosis method is a common method in the field of fault diagnosis, and the data quality has a huge impact on the accuracy of the method. In the fault diagnosis of broken belt, there are many factors that affect the broken belt, which leads to high data dimension, it is difficult to extract the main features during diagnosis, and the model training speed is slow. In addition, the normal operating status data of the rolling stock in cold rolling is better obtained, but relatively normal operation, the frequency of faults is not high, resulting in a lack of fault data, which has become an important factor restricting the research on data-driven belt breakage fault diagnosis. .

SUMMARY OF THE INVENTION

The embodiment of the present invention provides a data enhancement method applied to the fault diagnosis of a broken belt in a cold rolling mill, which can improve the quality of the generated fault images while improving the training speed of the generation model, so as to facilitate the directional generation of the faults required for the diagnosis of the broken belt fault. image, so as to solve the problem of insufficient fault data in broken belt fault diagnosis. The technical solution is as follows:

In one aspect, there is provided a data enhancement method applied to fault diagnosis of broken strip in a cold rolling mill, the method being applied to electronic equipment, the method comprising:

Collect time series signals of multiple features related to fault diagnosis of broken strip in cold rolling, process the collected time series signals of multiple features, and generate a two-dimensional fault image set;

dividing the fault image set into training data and test data;

The auxiliary classification generative adversarial network is trained by using the training data and its corresponding labels to obtain a generative model, wherein the trained generative model is used to generate a fault image required for fault diagnosis of a broken belt.

Further, the collection of time-series signals of multiple features related to fault diagnosis of strip breakage in cold rolling, processing the collected time-series signals of multiple features, and generating a two-dimensional fault image set includes:

Collect time series signals of multiple features related to fault diagnosis of broken strip in cold rolling;

The dimensionality reduction of the collected time series signals of multiple features is performed through the stack auto-encoding network to obtain a one-dimensional time series signal;

The one-dimensional time series signal is converted into a two-dimensional grayscale image through signal-image conversion to form a two-dimensional fault image set.

Further, the structural connection mode of the stacked self-encoding network is: input layer→full connection layer→full connection layer→full connection layer→full connection layer.

Further, the training of the auxiliary classification generative adversarial network by using the training data and its corresponding label to obtain a generative model further includes:

Flip, rotate and add noise to the divided training data to obtain the training data of the auxiliary classification generative adversarial network;

The obtained training data of the auxiliary classification generative adversarial network and its corresponding labels are input into the auxiliary classification generative adversarial network for training, and a generative model is obtained.

Further, the auxiliary classification generative adversarial network includes: a generator and a discriminator; wherein,

The generator is used to generate a fault image, and the fault image is a two-dimensional grayscale image;

The discriminator is used for judging the difference between the fault image generated by the generator and the fault image input to the auxiliary classification-generating confrontation network, and providing feedback to the generator.

Further, after using the training data and its corresponding labels to train the auxiliary classification generative adversarial network to obtain a generative model, the method further includes:

Using the generative model to generate fault images required for fault diagnosis of broken belts;

The generated fault images and the training data obtained by the original division are jointly input into the two-dimensional convolutional neural network for training, and the fault diagnosis model for the broken belt is obtained;

Among them, the trained broken belt fault diagnosis model is used for broken belt fault diagnosis, and the type of broken belt fault is output.

Further, the structural connection mode of the two-dimensional convolutional neural network is: two-dimensional convolutional layer→max pooling layer→two-dimensional convolutional layer→max pooling layer→two-dimensional convolutional layer→max pooling layer→ Two-dimensional convolutional layer → max pooling layer → fully connected layer → fully connected layer → SoftMax layer, where SoftMax represents the normalized exponential function.

Further, after the generated fault image and the training data obtained by the original division are jointly input into the two-dimensional convolutional neural network for training to obtain the fault diagnosis model, the method further includes:

Use the divided test data to test the trained fault diagnosis model for broken belts.

In one aspect, an electronic device is provided, the electronic device includes a processor and a memory, the memory stores at least one instruction, the at least one instruction is loaded and executed by the processor to implement the above-mentioned application to cold rolling A data augmentation method for fault diagnosis of broken strips in rolling mills.

In one aspect, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the above-mentioned data enhancement applied to fault diagnosis of broken strips in a cold rolling mill method.

The beneficial effects brought by the technical solutions provided by the embodiments of the present invention include at least:

In the embodiment of the present invention, the time series signals of multiple features related to the fault diagnosis of strip breakage in cold rolling are collected, and the collected time series signals of the multiple features are processed to generate a two-dimensional fault image set; It is divided into training data and test data; using the training data and its corresponding labels to train the auxiliary classification generative adversarial network, while improving the training speed of the generative model, the quality of the generated fault images can be improved, so as to facilitate the directional generation of fault images. The fault images required by the belt fault diagnosis can be solved to solve the problem of insufficient fault data in the broken belt fault diagnosis.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

FIG. 1 is a schematic flowchart of a data enhancement method applied to fault diagnosis of a broken strip in a cold rolling mill provided by an embodiment of the present invention;

FIG. 2 is a detailed flowchart of a data enhancement method applied to fault diagnosis of a broken strip in a cold rolling mill provided by an embodiment of the present invention;

3 is a schematic diagram of a comparison of loss function values before and after data enhancement provided by an embodiment of the present invention;

4 is a schematic diagram of a comparison of confusion matrices before and after data enhancement provided by an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

As shown in FIG. 1 , an embodiment of the present invention provides a data enhancement method applied to fault diagnosis of a broken strip in a cold rolling mill. The method can be implemented by an electronic device, and the electronic device can be a terminal or a server. The method includes:

S101 , collecting time-series signals of multiple features related to fault diagnosis of broken strips in cold rolling, and processing the collected time-series signals of multiple features to generate a two-dimensional fault image set;

S102, dividing the fault image set into training data and test data;

S103, using the training data and its corresponding label to train auxiliary classification generative adversarial networks (ACGANs) to obtain a generative model, wherein the trained generative model is used for glitch image.

The data enhancement method applied to the fault diagnosis of strip breakage in a cold rolling mill described in this embodiment collects time-series signals of multiple features related to fault diagnosis of strip-breaks in cold rolling, and processes the collected time-series signals of multiple features. Generate a two-dimensional fault image set; divide the fault image set into training data and test data; use the training data and its corresponding label to train the auxiliary classification generation confrontation network, while improving the training speed of the generation model, The quality of the generated fault images can be improved, so as to facilitate the directional generation of fault images required for fault diagnosis of broken belts, thereby solving the problem of insufficient fault data in fault diagnosis of broken belts.

In this embodiment, the label is the category of the broken belt fault, which specifically refers to the broken belt fault of different racks.

In this embodiment, the collection of time-series signals of multiple features related to fault diagnosis of strip breakage in cold rolling, processing the collected time-series signals of multiple features, and generating a two-dimensional fault image set includes:

A1, collect time series signals of multiple features related to fault diagnosis of broken strip in cold rolling;

In this embodiment, as shown in FIG. 2 , N (for example, 24) sensors are used to collect time series signals of 24 features related to fault diagnosis of broken strips in cold rolling to obtain a high-dimensional time series signal data matrix.

A2, reducing the dimension of the collected time series signals of multiple features through a stacked auto-encoders network (SAE) to obtain a one-dimensional time series signal;

In this embodiment, the obtained high-dimensional time series signal data matrix is put into a stacked self-encoding network (SAE) for training, and the high-dimensional time series signal data matrix is reduced to a one-dimensional time series signal to achieve the effect of information fusion.

In this embodiment, the structural connection mode of the stacked self-encoding network (SAE) is: input layer L0 → fully connected layer L1 → fully connected layer L2 → fully connected layer L3 → fully connected layer L4; wherein, the input layer L0 The number of output neurons is 30; the number of input neurons in the fully connected layer L1 is 30, and the number of output neurons is 10; the number of input neurons in the fully connected layer L2 is 10, and the number of output neurons is 1; The number of input neurons in the fully connected layer L3 is 1 and the number of output neurons is 10; the number of input neurons in the fully connected layer L4 is 10 and the number of output neurons is 30.

A3: Generate a two-dimensional grayscale image from the one-dimensional time series signal through signal-image conversion to form a two-dimensional fault image set.

In this embodiment, the specific method of signal-image conversion is shown in formula (1):

Among them, P(m,n) represents the gray value of the mth row and nth column in the generated two-dimensional grayscale image; N represents the size of the generated image is N×N; the form L(i) represents the The grayscale value of the i-th data point; Max(L) and Min(L) represent the maximum and minimum values in L respectively, and L represents the value of the one-dimensional time series signal after a single sampling, and its length is N ² ; The function of the rounding function round(x) is to round the data to ensure that the converted data is an integer between 0 and 255.

In this embodiment, the one-dimensional time-domain signal can be signal normalized, converted into gray value, rounded, signal intercepted according to the size of the image, and matrix transformed by formula (1) to obtain the one-dimensional time-series signal. Two-dimensional grayscale image, thus forming a two-dimensional fault image set.

In this embodiment, 80% of the two-dimensional fault image set may be used as training data, and the remaining 20% may be used as test data.

In this embodiment, the training of the auxiliary classification generative adversarial network by using the training data and its corresponding labels to obtain a generative model further includes:

B1: Flip, rotate and add noise to the training data obtained by division to obtain training data for the auxiliary classification generating adversarial network;

In this embodiment, flipping includes horizontal flipping and/or vertical flipping of the image.

In this embodiment, the rotation includes rotating the image by 90° to the left or right.

In this embodiment, the noise adding process refers to adding random Gaussian noise to the image, and adding Gaussian noise specifically refers to adding the image matrix directly to the numbers randomly sampled from the Gaussian distribution.

In this embodiment, each image in the training data is subjected to the above-mentioned flipping, rotating and noise processing, and training data six times the amount of the original training data can be obtained, so as to obtain the auxiliary classification generative adversarial network with high diversity. training data.

B2, the obtained training data of the auxiliary classification generative adversarial network and its corresponding labels are input into the auxiliary classification generative adversarial network (ACGANs) for training, and a generative model is obtained.

In this embodiment, the Auxiliary Classification Generative Adversarial Networks (ACGANs) can process two-dimensional images. Therefore, the ACGANs can also be called 2D-ACGANs, as shown in FIG. 2 .

In this embodiment, the auxiliary classification generative adversarial network (ACGANs) includes: a generator and a discriminator, wherein the generator is used to generate a fault image, and the discriminator is used to determine the fault image generated by the generator and the input to the auxiliary classification generator. The difference in the adversarial network's fault image, which is a two-dimensional grayscale image, is provided and feedback is provided to the generator. The generator includes 4 fractional stride 2D convolutional layers and 4 batch normalization layers. Except the last output layer uses the Tanh function as the activation function, the ReLU function is used as the activation function. The discriminator includes 4 two-dimensional convolutional layers and 4 batch normalization layers. The activation function uses the LeakyReLU function in addition to the Sigmoid function used in the output layer.

In this embodiment, the two-dimensional convolution layer has fewer parameters and better feature extraction capability for time series signals. Meanwhile, the ACGANs described in this embodiment also take into account the label information of fault data, so there is no need to train multiple models, which makes the training process simple In this way, while improving the training speed of the generation model, the quality of the generated fault images can be improved, so as to facilitate the directional generation of fault images required for fault diagnosis of broken belts, thereby solving the problem of insufficient fault data in fault diagnosis of broken belts.

In this embodiment, after using the training data and its corresponding labels to train the auxiliary classification generative adversarial network to obtain a generative model, the method further includes:

Using the generative model to generate fault images required for fault diagnosis of broken belts;

The generated fault images and the training data obtained by the original division are jointly input into a two-dimensional (2D) convolutional neural network (CNN) for training to obtain a fault diagnosis model for broken belts;

Among them, the trained broken belt fault diagnosis model is used for broken belt fault diagnosis, and the type of broken belt fault is output.

In this embodiment, the structure of the two-dimensional convolutional neural network includes a total of 11 layers, and the specific structural connection method is: two-dimensional convolutional layer L1 (5×5×32) → maximum pooling layer L2 (2×2 ) → two-dimensional convolutional layer L3 (3×3×64) → maximum pooling layer L4 (2×2) → two-dimensional convolutional layer L5 (3×3×128) → maximum pooling layer L6 (2×2 )→two-dimensional convolutional layer L7 (3×3×256)→max pooling layer L8 (2×2)→full connection layer L9 (2560-768)→full connection layer L10 (768-10)→SoftMax (normalized Uniform exponential function) layer L11; wherein, the three parameters in the brackets of the two-dimensional convolution layer represent the length of the convolution kernel, the width of the convolution layer and the number of convolution kernels respectively; the parameters in the parentheses of the maximum pooling layer represent The size of its window; the parameters in parentheses of the fully connected layer represent the number of input parameters and the number of output parameters, respectively.

In this embodiment, after the generated fault image and the training data obtained by the original division are jointly input into a two-dimensional convolutional neural network for training, and a fault diagnosis model is obtained, the method further includes:

Use the divided test data to test the trained fault diagnosis model for broken belts.

In order to verify the effectiveness of the data enhancement method applied to the fault diagnosis of cold rolling mill broken strip provided by the embodiment of the present invention, the relevant characteristic parameters of the cold rolling mill and the broken strip fault of the rolling mill in a steel plant are collected, and the specific composition is shown in Table 1. , through which the fault diagnosis experiment is carried out. In the experiment, the fault data and normal data of 4 rolling mills were collected in total, with a total of 24 parameters. The sampling frequency was 12kHz, and 4096 points were taken as a group, and a total of 1000 groups of data were collected.

Table 1 Characteristic parameters related to fault diagnosis of broken strip in cold rolling

1 1 rack drive side servo valve current 2 1 rack operation side servo valve current 3 Rolling force deviation (absolute value) 4 2-stand rolling force deviation 5 2 rack drive side servo valve current 6 2 rack operation side servo valve current 7 3 rack drive side servo valve current 8 3 rack operation side servo valve current 9 3 Rolling force deviation 10 4 rack drive side servo valve current 11 4 rack operation side servo valve current 12 4 Rolling force deviation 13 1 The actual tension value of the frame 14 1 Tension deviation value of the frame 15 2 The actual tension value of the frame 16 2 The tension deviation value of the frame 17 3 The actual tension value of the frame 18 3 The tension deviation value of the frame 19 4 The actual tension value of the frame 20 4 The tension deviation value of the frame twenty one Motor current for 1 rack twenty two Motor current for 2 racks twenty three Motor current for 3 racks twenty four Motor current for 4 racks

Figure 3(a) shows the change trend of the loss function of the fault diagnosis model for fault diagnosis without data enhancement. It can be seen that the training loss function finally converges to 0 approximately, while the test loss function has been oscillating around 0.5, indicating that the model is due to training at this time. There is an over-fitting phenomenon due to insufficient data; Fig. 3(b) is the change trend of the loss function of the fault diagnosis model for the broken belt after the data enhancement method provided by the embodiment of the present invention, and the loss function value is both on the training data and the test data. Approximately converges to 0, indicating that the overfitting phenomenon basically disappears after data enhancement.

Figure 4(a) and (b) are schematic diagrams of the comparison of confusion matrices for fault diagnosis before and after data augmentation, respectively, where the data on the symmetry axis represents the proportion of correctly identified health states of this type in all test data. The rest of the row is the fraction of other health states that were misidentified. It can be clearly seen that after data augmentation, the recognition accuracy of 3 racks has increased from 92.5% to 99%, an increase of 6.5%, and the overall average accuracy has also increased from 95% to 99.5%.

5 is a schematic structural diagram of an electronic device 600 according to an embodiment of the present invention. The electronic device 600 may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPU) 601 and one or more memories 602, wherein at least one instruction is stored in the memory 602, and the at least one instruction is loaded and executed by the processor 601 to realize the above-mentioned data applied to the fault diagnosis of strip breakage in a cold rolling mill Enhancement method.

In an exemplary embodiment, a computer-readable storage medium, such as a memory including instructions, is also provided, and the instructions can be executed by a processor in a terminal to implement the above-mentioned data enhancement method applied to fault diagnosis of a broken strip in a cold rolling mill. For example, the computer-readable storage medium may be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, and the like.

Those of ordinary skill in the art can understand that all or part of the steps of implementing the above embodiments can be completed by hardware, or can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, etc.

The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within the range.

Claims (8)

1. A data enhancement method applied to strip breakage fault diagnosis of a cold rolling mill is characterized by comprising the following steps:

collecting time sequence signals of a plurality of characteristics related to belt breakage fault diagnosis in cold rolling, and processing the collected time sequence signals of the plurality of characteristics to generate a two-dimensional fault image set;

dividing the fault image set into training data and test data;

and training the assistant classification generation countermeasure network by using the training data and the labels corresponding to the training data to obtain a generation model, wherein the trained generation model is used for generating fault images required by belt breakage fault diagnosis.

2. The data enhancement method applied to the strip breakage fault diagnosis of the cold rolling mill according to claim 1, wherein the step of collecting time sequence signals of a plurality of characteristics related to the strip breakage fault diagnosis in the cold rolling, and the step of processing the collected time sequence signals of the plurality of characteristics to generate a two-dimensional fault image set comprises the following steps:

collecting time sequence signals of a plurality of characteristics related to belt breakage fault diagnosis in cold rolling;

reducing the dimension of the collected time sequence signals with a plurality of characteristics through a stack self-coding network to obtain one-dimensional time sequence signals;

and generating a two-dimensional gray scale map from the one-dimensional time sequence signal through signal-image conversion to form a two-dimensional fault image set.

3. The data enhancement method applied to the strip breakage fault diagnosis of the cold rolling mill as claimed in claim 2, wherein the structural connection mode of the stack self-coding network is as follows: input layer → fully connected layer.

4. The data enhancement method applied to the strip breakage fault diagnosis of the cold rolling mill as claimed in claim 1, wherein the training of the auxiliary classification generation countermeasure network by using the training data and the corresponding labels thereof to obtain the generation model further comprises:

turning, rotating and denoising the training data obtained by dividing to obtain training data for assisting classification to generate an anti-network;

and inputting the training data of the obtained assistant classification generation countermeasure network and the corresponding labels thereof into the assistant classification generation countermeasure network for training to obtain a generation model.

5. The data enhancement method applied to the strip breakage fault diagnosis of the cold rolling mill as claimed in claim 1, wherein the auxiliary classification generation countermeasure network comprises: a generator and a discriminator; wherein,

the generator is used for generating a fault image, and the fault image is a two-dimensional gray scale image;

and the discriminator is used for judging the difference between the fault image generated by the generator and the fault image input to the auxiliary classification generation countermeasure network and providing feedback for the generator.

6. The data enhancement method applied to the strip breakage fault diagnosis of the cold rolling mill as claimed in claim 1, wherein after training the auxiliary classification generation countermeasure network by using the training data and the corresponding labels thereof to obtain a generation model, the method further comprises:

generating a fault image required by belt breakage fault diagnosis by using the generated model;

inputting the generated fault image and training data obtained by original division into a two-dimensional convolutional neural network together for training to obtain a belt breakage fault diagnosis model;

the trained belt breakage fault diagnosis model is used for carrying out belt breakage fault diagnosis and outputting a belt breakage fault type.

7. The data enhancement method applied to the fault diagnosis of the broken strip of the cold rolling mill according to claim 1, wherein the structural connection mode of the two-dimensional convolution neural network is as follows: two-dimensional convolution layer → maximum pooling layer → fully-connected layer → SoftMax layer, where SoftMax represents a normalized exponential function.

8. The data enhancement method applied to the strip breakage fault diagnosis of the cold rolling mill according to claim 1, wherein after the generated fault image and the training data obtained by the original division are input into a two-dimensional convolutional neural network together for training to obtain a fault diagnosis model, the method further comprises the following steps:

and testing the trained belt breakage fault diagnosis model by using the test data obtained by dividing.

Images