CN118152757A - A visual fault diagnosis method for vibrating screen based on sound sensor - Google Patents

Abstract

The present invention discloses a visual fault diagnosis method for a vibrating screen based on a sound sensor, comprising the following steps performed in sequence: step 1: the sound sensor collects a plurality of first sound signals of the vibrating screen; step 2: each of the first sound signals is converted into a grayscale image to form a grayscale image data set; step 3: a convolutional neural network is constructed to extract fault features from the grayscale image data set; step 4: a training network is constructed according to the fault features extracted in step 3, and the grayscale image data set is trained multiple times using the training network to obtain a fault diagnosis model; step 5: the fault diagnosis model is verified using a verification data set; step 6: the fault diagnosis model verified in step 5 is used to perform real-time diagnosis of the vibrating screen fault. The fault diagnosis model can accurately identify the fault problem of the vibrating screen, and effectively improves the fault diagnosis accuracy of the vibrating screen.

Description

A visual fault diagnosis method for vibrating screen based on sound sensor

Technical Field

The invention relates to the field of vibration screen fault diagnosis, and in particular to a vibration screen visual fault diagnosis method based on a sound sensor.

Background technique

As an important screening equipment, the vibrating screen is of great importance in the material sorting industry. As an important equipment for material sorting, the vibrating screen is in a harsh working environment for a long time. In addition, the long-term continuous work and the complex loads during work make the main components of the vibrating screen, such as the exciter bearing, the exciter beam, the screen frame, etc., very easy to fail. As an important part of the screening production line, once the vibrating screen has a serious failure, it will cause the entire production line to stop working, causing huge economic losses.

The development of technology has brought new directions to fault diagnosis. Fault diagnosis technology based on expert knowledge and machine learning algorithms has gradually shown advantages such as high recognition accuracy and strong adaptability, and has been widely used. However, although the intelligent fault diagnosis method based on expert knowledge and machine learning algorithms can make a relatively intuitive judgment on the fault of the vibrating screen equipment, the manual feature extraction method of the machine learning algorithm is very subjective and highly dependent on expert knowledge and experience, resulting in low fault diagnosis accuracy for the vibrating screen.

In view of this, the inventor of this case conducted in-depth research on the above-mentioned issues, which led to the emergence of this case.

Summary of the invention

The object of the present invention is to provide a visual fault diagnosis method for a vibrating screen based on a sound sensor, which can improve the fault diagnosis accuracy of the vibrating screen.

In order to achieve the above object, the present invention adopts such technical solution:

A visual fault diagnosis method for a vibrating screen based on a sound sensor comprises the following steps performed in sequence:

Step 1: The sound sensor collects a plurality of first sound signals of the vibrating screen;

Step 2: Convert each of the first sound signals into a grayscale image to form a grayscale image data set;

Step 3: Construct a convolutional neural network to extract fault features from the grayscale image data set;

Step 4: Building a training network based on the fault features extracted in step 3, and using the training network to train the grayscale image data set multiple times to obtain a fault diagnosis model;

Step 5: Verify the fault diagnosis model using a verification data set;

Step 6: Use the fault diagnosis model verified in step 5 to perform real-time diagnosis of the vibration screen fault.

Preferably, the multiple first sound signals in step 1 are: the left exciting force is 10% smaller, the left exciting force is 5% larger, the right exciting force is 10% smaller, the right exciting force is 5% larger and the exciting force is balanced.

Preferably, step 2 includes the following specific operations: dividing each of the first sound signals into a sample data according to continuous data points of a specified length, normalizing each of the sample data, and converting the sample data into the grayscale image to form the grayscale image data set.

Preferably, the convolutional neural network includes five alternating convolutional pooling layers: conv1, maxpool1, conv2, maxpool2, conv3, maxpool3, conv4, maxpool4, conv5, maxpool5; two FC layers, and ReLU activation function is used between layers for activation.

Preferably, the sound sensor re-collects a plurality of second sound signals of the vibrating screen, and produces the second sound signals into the verification data set.

Preferably, the fault diagnosis model is deployed on a Labview platform to perform real-time diagnosis of vibration screen faults.

Preferably, when training the convolutional neural network, the number of iterations is set to 100, the learning rate is set to 0.0001, and the batch size is set to 32.

By adopting the aforementioned design scheme, the beneficial effects of the present invention are as follows: the fault vibration method of the vibrating screen collects the first sound signal of the vibrating screen through a sound sensor, converts the first sound signal into a grayscale image, and then trains the convolutional neural network to obtain a fault diagnosis model, and the fault diagnosis model is trained through the first sound signal, so the fault diagnosis model can accurately identify the fault problem of the vibrating screen, and effectively improves the fault diagnosis accuracy of the vibrating screen; effectively solves the problem that the vibrating screen is difficult to inspect under harsh working conditions, and further reduces the complexity of the vibrating screen fault diagnosis; at the same time, avoids the economic losses caused by scheduled shutdown inspections, and has higher economic benefits.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a fault diagnosis method of the present invention;

FIG2 is an image example of a first sound signal of the present invention;

FIG3 is a schematic diagram of the conversion of a first sound signal into a grayscale image according to the present invention;

FIG4 is a network structure diagram of the present invention;

FIG5 is a graph showing the training loss of a convolutional neural network according to the present invention;

FIG6 is a graph showing the accuracy of the invention;

FIG. 7 is a confusion matrix diagram of the present invention.

Detailed ways

The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

A visual fault diagnosis method for a vibrating screen based on a sound sensor, as shown in the process of FIG1, specifically comprises the following steps performed in sequence:

Step 1: The sound sensor collects a plurality of first sound signals of the vibrating screen;

In this embodiment, the multiple first sound signals in step 1 are: the left exciting force is 10% smaller, the left exciting force is 5% larger, the right exciting force is 10% smaller, the right exciting force is 5% larger and the exciting force is balanced, a total of five types of first sound signals, as shown in Figure 2.

Step 2: Convert each first sound signal into a grayscale image to form a grayscale image data set; In this embodiment, each first sound signal is divided into a sample data according to continuous data points of a specified length, each sample data is normalized, and the sample data is converted into a grayscale image to form a grayscale image data set. The specific conversion method is: if you want to obtain a grayscale image of size M×M, divide each first sound signal into sample data of length M2, fill the pixel matrix of the grayscale image in sequence, and normalize the pixel matrix from 0-255 to obtain a grayscale image data set, as shown in Figure 3.

Step 3: Construct a convolutional neural network to automatically extract fault features from the grayscale image data set; the convolutional neural network constructed in this embodiment includes five alternating convolutional pooling layers, conv1, maxpool1, conv2, maxpool2, conv3, maxpool3, conv4, maxpool4, conv5, maxpool5; the first three convolutional layers use a 5×5 convolution kernel, and the last two convolutional layers use a 3×3 convolution kernel. After each convolution, the border is padded with zeros to keep the image size unchanged. A maximum pooling layer with a convolution kernel size of 2×2 is set behind each convolutional layer, where the step size of each convolutional layer is set to 2, the step size of each maximum pooling layer is set to 1, and two FC layers are used. ReLU activation function is used between layers. The network structure is shown in Figure 4.

The structural parameters of each layer of the network model are shown in Table 1:

Table 1 Structural parameters of each layer of the network model

Step 4: Build a training network based on the fault features extracted in step 3, and use the training network to train the grayscale image data set multiple times to obtain a fault diagnosis model with the best training effect; in this embodiment, when the grayscale image data set is trained multiple times, 70% of the grayscale image data set is divided into a training set for training the fault diagnosis model, and 30% is divided into a verification data set for selecting the optimal fault diagnosis model. The optimal fault diagnosis model in this embodiment refers to the fault diagnosis model with the highest prediction accuracy. When training the convolutional neural network, set the number of iterations to 100 times, the learning rate to 0.0001, the number of batches to 32, and use the Adam optimizer to update the network weights; cross entropy is used as the error function; and the Softmax classifier is used to output the classification results. The training loss curve of the convolutional neural network is shown in Figure 5, and the accuracy curve is shown in Figure 6.

Step 5: Use the validation data set to validate the fault diagnosis model; the validation data set here is a validation data set formed by recollecting multiple second sound signals of the vibrating screen by the sound sensor and converting the second sound signals into grayscale images. The validation data set is used to calculate the number of correct classifications and the number of incorrect classifications, and a confusion matrix is constructed to measure the accuracy of the fault diagnosis model and to achieve performance evaluation of the fault diagnosis model. The confusion matrix is shown in Figure 7. The validation data set finally achieved an accuracy rate of 99.8%, indicating that the fault diagnosis model can effectively identify the fault category of the vibrating screen.

Step 6: Perform real-time diagnosis of the vibration screen fault using the fault diagnosis model verified in step 5. The fault diagnosis model of this embodiment is deployed on the Labview platform to perform real-time diagnosis of the vibration screen fault.

To sum up, the fault vibration method of the vibrating screen collects the first sound signal of the vibrating screen through a sound sensor, converts the first sound signal into a grayscale image, and then trains the convolutional neural network to obtain a fault diagnosis model. The fault diagnosis model is trained through the first sound signal, so the fault diagnosis model can accurately identify the fault problem of the vibrating screen, effectively improving the fault diagnosis accuracy of the vibrating screen; effectively solving the problem that the vibrating screen is difficult to inspect under harsh working conditions, and further reducing the complexity of the vibrating screen fault diagnosis; at the same time, avoiding the economic losses caused by scheduled shutdown inspections, and having higher economic benefits.

Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims (7)

1. A visual fault diagnosis method of a vibrating screen based on a sound sensor is characterized in that: the method comprises the following steps of:

step 1: the sound sensor collects a plurality of first sound signals of the vibrating screen;

step 2: converting each first sound signal into a gray level image to form a gray level image data set;

step 3: constructing a convolutional neural network to extract fault characteristics of the gray image data set;

Step 4: building a training network according to the fault characteristics extracted in the step 3, and training the gray image dataset for multiple times by using the training network to obtain a fault diagnosis model;

step 5: validating the fault diagnosis model with a validation dataset;

step 6: and (5) performing real-time diagnosis on the fault of the vibrating screen by adopting the fault diagnosis model verified in the step (5).

2. A method for visual fault diagnosis of a vibrating screen based on a sound sensor as claimed in claim 1, wherein: the first sound signals in the step 1 are respectively that the left exciting force is smaller than 10%, the left exciting force is larger than 5%, the right exciting force is smaller than 10%, the right exciting force is larger than 5% and the exciting force is balanced.

3. A method for visual fault diagnosis of a vibrating screen based on a sound sensor as claimed in claim 1, wherein: step 2 comprises the following specific operations: dividing each first sound signal into sample data according to continuous data points with specified lengths, carrying out normalization processing on each sample data, and converting the sample data into the gray image to form the gray image data set.

4. A method for visual fault diagnosis of a vibrating screen based on a sound sensor as claimed in claim 1, wherein: the convolutional neural network includes five alternating convolutional pooling layers: conv1, maxpool, conv2, maxpool 2, conv3, maxpool, conv4, maxpool, conv5, maxpool5; two FC layers, layer-by-layer, are activated using a ReLU activation function.

5. A method for visual fault diagnosis of a vibrating screen based on a sound sensor as claimed in claim 1, wherein: a sound sensor re-collects a plurality of second sound signals of the vibrating screen and prepares the second sound signals into the validation data set.

6. A method for visual fault diagnosis of a vibrating screen based on a sound sensor as claimed in claim 1 or 5, wherein: and deploying the fault diagnosis model on a Labview platform to perform real-time diagnosis of the faults of the vibrating screen.

7. A method for visual fault diagnosis of a vibrating screen based on a sound sensor as claimed in claim 1, wherein: when the convolutional neural network is trained, the iteration times are set to 100 times, the learning rate is 0.0001, and the batch processing number is set to 32.