Method for measuring and calculating rotating speed of fan bearing under variable-rotating-speed working condition based on meta-learning

Abstract

The invention discloses a method for measuring and calculating the rotating speed of a fan bearing under a variable-rotating-speed working condition based on meta-learning, and the method comprises the steps: collecting a vibration signal of the fan bearing, dividing the vibration signal into a plurality of sections of sample data through a sliding window, dividing a rotating speed interval into a working condition A and a working condition B, and constructing a label for the corresponding sample data; taking the data of the working condition of the rotating speed interval A section as a training set in meta learning, taking the data of the working condition of the rotating speed interval B section as a test set in meta learning, and dividing the training set into a support set S and a query set Q according to an N-way K-shot mode in meta learning; training the backbone network model through feature extraction of a support set in the training set; and performing prototype correction on the backbone network model through the test set under the working condition of the B section of the rotating speed interval to obtain a rotating speed detection model, and obtaining the relationship between the fan bearing vibration signal and the rotating speed interval through the rotating speed detection model. According to the method, the purpose of detecting the real-time rotating speed by using the fan bearing vibration signal under the variable-rotating-speed working condition is achieved.

Description

Technical Field

[0001] The present invention relates to the field of detection technology, and in particular to a method for calculating the speed of a fan bearing under variable speed conditions based on meta-learning. Background Art

[0002] In the steel smelting industry, fans can play the role of air supply, dust removal, cooling, etc., and are indispensable equipment in all aspects of steel smelting. The speed of the fan can directly reflect its working condition and health status. Therefore, the fan speed has become an important indicator in fan status monitoring and fault warning. However, since ordinary speed sensors are not suitable for direct installation on the fan, or the measurement effect is not good after installation, the speed sensor can only measure the fan speed within a short period of time. The method of directly measuring the real-time fan speed with it is not feasible in practice.

[0003] As a component that directly affects the rotation of the fan, the vibration of the bearing is related to the speed of the fan to a certain extent. In addition, sensors related to bearing vibration data are more convenient to install and data acquisition is also easier. However, the specific relationship between bearing vibration and fan speed is difficult to directly obtain through manual calculation, so machine learning can be used to find out the relationship between fan bearing vibration and its speed. However, due to different specific conditions such as load, the speed range of the fan will not remain unchanged in actual production. If traditional machine learning methods are used, it is likely that the model generalization effect will be poor after the speed range changes, resulting in a decrease in the reliability of the detection results. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for measuring the speed of a fan bearing under variable speed conditions based on meta-learning. This method overcomes the defects of the difficulty in installing a suitable speed sensor and the difficulty in data acquisition in traditional fan speed detection, establishes the relationship between the vibration signal of the fan bearing and its real-time speed, and establishes a semi-supervised category prototype correction mechanism to achieve the purpose of detecting the real-time speed using the vibration signal of the fan bearing under variable speed conditions.

[0005] In order to solve the above technical problems, the method for calculating the speed of a fan bearing under variable speed conditions based on meta-learning of the present invention comprises the following steps: Step 1: Collect the vibration signal of the fan bearing, divide the collected vibration signal into multiple sections of sample data through a sliding window, divide the speed range into section A and section B according to the actual speed, and build labels for the corresponding sample data; Step 2: Use the labeled data of the speed range A as the training set in meta-learning, and use the data of the speed range B with a small amount of labels as the test set in meta-learning, and divide the training set into the support set S and the query set Q according to the N-way K-shot mode in meta-learning; Step 3: Use the first layer of deep convolutional neural network with wide convolution kernel as the backbone network, extract the speed-related features from the input vibration signal through the support set data in the training set under the speed range A working condition, and calculate the prototype points of different speed range labels. Use the distance calculation between the query set data in the training set and different prototype points to obtain the category prediction of the query set data, compare it with the real label of the query set data, calculate the loss, and train the backbone network model through back propagation; Step 4. Through the test set under the speed range B condition, the obtained backbone network model is prototyped and corrected so that its prototype point distribution is close to the actual situation of the speed range B condition. Through the meta-learning backbone network model training and the final prototype correction, the speed detection model trained with the data under working condition A is also applicable to working condition B, and the relationship between the fan bearing vibration signal and the fan speed range is obtained through the speed detection model.

[0006] Further, in the step 2, Support Set in, represents the jth sample under category i, n represents the sample category, k represents the number of samples contained in each category in the support set, Indicates the label corresponding to the sample. The support set contains a total of n×k labeled samples; Query Set , in, represents the jth sample under category i, n represents the sample category, q represents the number of samples contained in each category in the query set, Indicates the label corresponding to the sample. The query set contains a total of n×q labeled samples. Since the method for calculating the speed of a fan bearing under variable speed conditions based on meta-learning of the present invention adopts the above-mentioned technical scheme, that is, the method collects the vibration signal of the fan bearing, divides it into multiple sections of sample data through a sliding window, divides the speed range into section A and section B, and constructs labels for the corresponding sample data; uses the data of the speed range section A as the training set in meta-learning, uses the data of the speed range B as the test set in meta-learning, and divides the training set into the support set S and the query set Q according to the N-way K-shot mode in meta-learning; trains the backbone network model by extracting features of the support set in the training set; prototypes the backbone network model by using the test set under the speed range B, and then obtains the speed detection model, and obtains the relationship between the fan bearing vibration signal and the fan speed range through the speed detection model. The method achieves the purpose of detecting the real-time speed using the fan bearing vibration signal under variable speed conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments: Figure 1 This is a flowchart of the method for calculating the fan bearing speed under variable speed conditions based on meta-learning of the present invention. DETAILED DESCRIPTION

[0008] Implementation example Figure 1 As shown, the method for calculating the fan bearing speed under variable speed conditions based on meta-learning of the present invention includes the following steps: Step 1: Collect the vibration signal of the fan bearing, divide the collected vibration signal into multiple sections of sample data through a sliding window, divide the speed range into section A and section B according to the actual speed, and build labels for the corresponding sample data; Step 2: Use the labeled data of the speed range A as the training set in meta-learning, and use the data of the speed range B with a small amount of labels as the test set in meta-learning, and divide the training set into the support set S and the query set Q according to the N-way K-shot mode in meta-learning; Step 3: Use the first layer of deep convolutional neural network with wide convolution kernel as the backbone network, extract the speed-related features from the input vibration signal through the support set data in the training set under the speed range A working condition, and calculate the prototype points of different speed range labels. Use the distance calculation between the query set data in the training set and different prototype points to obtain the category prediction of the query set data, compare it with the real label of the query set data, calculate the loss, and train the backbone network model through back propagation; Step 4. Through the test set under the speed range B condition, the obtained backbone network model is prototyped and corrected so that its prototype point distribution is close to the actual situation of the speed range B condition. Through the meta-learning backbone network model training and the final prototype correction, the speed detection model trained with the data under working condition A is also applicable to working condition B, and the relationship between the fan bearing vibration signal and the fan speed range is obtained through the speed detection model.

[0009] Preferably, in the step 2, Support Set , in, represents the jth sample under category i, n represents the sample category, k represents the number of samples contained in each category in the support set, Indicates the label corresponding to the sample. The support set contains a total of n×k labeled samples; Query Set , in, represents the jth sample under category i, n represents the sample category, q represents the number of samples contained in each category in the query set, Indicates the label corresponding to the sample. The query set contains a total of n×q labeled samples.

[0010] Meta-learning in this method is a machine learning method whose goal is to enable the model to better adapt to new tasks by learning from the experience of the task. In traditional machine learning, the model is usually trained with a large amount of labeled training data and then applied to new tasks that are similar to the training data. Meta-learning attempts to enable the model to learn faster and more effectively when faced with new tasks, so as to achieve good performance on a small number of samples.

[0011] The input data of this method is simple to obtain, low-cost, easy to deploy, and has low requirements for computing software resources during operation. It can be used for real-time detection of fan speed and can adapt to different speed conditions. In actual application, a bearing vibration sensor is installed on the fan to collect the vibration signal of the bearing, and a small amount of speed data is obtained by using a speed sensor or manual calculation to be used for model establishment and migration between speed conditions. According to the vibration signal collected under a single condition and the corresponding speed label, a model with input as vibration signal and output as speed range is constructed, and the model is saved in the computer. If the speed condition changes, the original model is prototyped by a small amount of vibration data with speed label under the new speed condition to obtain a new model that can be used under the new speed condition. The computer extracts the vibration signal at the current time point at a certain frequency, inputs it into the model, and outputs the current speed range after calculation.

[0012] The input of this method is the vibration signal of the fan bearing. There is no need to set up overly complex sensors. The output is the current speed range of the fan. According to the value of this range, further analysis can be performed, such as the current health status of the fan and the life prediction of the fan bearing. If an abnormality occurs, the staff can choose to inspect and replace the parts involved during the recent routine downtime maintenance to avoid unplanned downtime and fan failure, so as to improve the stability of the production line and increase production capacity. It is of great significance to enhance the stability of the production line, increase production capacity, and reduce inspection and maintenance costs.

Claims

1. A method for calculating the speed of a fan bearing under variable speed conditions based on meta-learning, characterized in that The steps include: Step 1: Collect the vibration signal of the fan bearing, divide the collected vibration signal into multiple sections of sample data through a sliding window, divide the speed range into section A and section B according to the actual speed, and build labels for the corresponding sample data; Step 2: Use the labeled data of the speed range A as the training set in meta-learning, and use the data of the speed range B with a small amount of labels as the test set in meta-learning, and divide the training set into the support set S and the query set Q according to the N-way K-shot mode in meta-learning; Step 3: Use the first layer of deep convolutional neural network with wide convolution kernel as the backbone network, extract the speed-related features from the input vibration signal through the support set data in the training set under the speed range A working condition, and calculate the prototype points of different speed range labels. Use the distance calculation between the query set data in the training set and different prototype points to obtain the category prediction of the query set data, compare it with the real label of the query set data, calculate the loss, and train the backbone network model through back propagation; Step 4. Through the test set under the speed range B condition, the obtained backbone network model is prototyped and corrected so that its prototype point distribution is close to the actual situation of the speed range B condition. Through the meta-learning backbone network model training and the final prototype correction, the speed detection model trained with the data under working condition A is also applicable to working condition B, and the relationship between the fan bearing vibration signal and the fan speed range is obtained through the speed detection model.

2. The method for calculating the bearing speed of a fan under variable speed conditions based on meta-learning according to claim 1 is characterized in that: In step 2, the support set, in, represents the jth sample under category i, n represents the sample category, k represents the number of samples contained in each category in the support set, Indicates the label corresponding to the sample. The support set contains a total of n×k labeled samples; Query Set , in, represents the jth sample under category i, n represents the sample category, q represents the number of samples contained in each category in the query set, Indicates the label corresponding to the sample. The query set contains a total of n×q labeled samples.

Images