CN110378045A - A kind of pre- maintaining method of guide precision based on deep learning - Google Patents

Abstract

The embodiment of the present invention discloses a method for pre-maintenance of guide rail accuracy based on deep learning, which includes the following steps: Step 100, collecting data signals under the normal operating state of the guide rail through sensors, and preprocessing the data signals; step 200, using depth The learned deep belief network pre-trains the preprocessed data signal and establishes a calculation model; step 300, the deep learning algorithm performs online real-time judgment on the state of the guide rail operation, and completes the curve diagram of the guide rail operation state; step 400, according to the guide rail operation Status graphs for predictive maintenance of guideway wear. The present invention uses a deep learning algorithm, realizes autonomous learning, extracts features independently, performs fitting through a neural network, and grasps the wear and tear during operation in real time. It has the advantages of good self-adaptation and high robustness, and can adapt, Self-learning and self-diagnosis have high application value.

Description

A Predictive Maintenance Method of Guideway Accuracy Based on Deep Learning

technical field

The embodiment of the present invention relates to the technical field of machine tools, and in particular to a method for pre-maintenance of guide rail accuracy based on deep learning.

Background technique

With the rapid development of my country's manufacturing industry, machine tools play a huge role as the main equipment in the manufacturing field. Guide rails, as an important part of machine tools, have a great impact on the accuracy of machined parts. The existing technology mainly focuses on imitating the surface of guide rails. Laser scanning images, etc. are used to identify the wear on the surface of the guide rail. Although it achieves a certain effect, the environment of the machine tool during the cutting process changes in real time. In the face of guide rail miscellaneous wear, oxidative wear, poor lubrication and other phenomena, timely maintenance and timely repair can effectively avoid guide rail precision. Severe wear, reduced life, etc. will affect the accuracy.

Contents of the invention

For this reason, the embodiment of the present invention provides a guide rail precision pre-maintenance method based on deep learning, which can not only improve the production quality, but also avoid the problem of defective products due to the decrease in the accuracy of the guide rail due to wear and tear, and ensure the quality of enterprise production. It solves the problem in the prior art that the accuracy of the guide rail is not affected by real-time monitoring and wear and the like.

In order to achieve the above object, embodiments of the present invention provide the following technical solutions:

A method for pre-maintenance of guide rail accuracy based on deep learning, comprising the following steps:

Step 100, collecting the data signal of the guide rail under the normal operation state through the sensor, and preprocessing the data signal;

Step 200, pre-training the preprocessed data signal with a deep belief network of deep learning, and establishing a calculation model;

Step 300, the deep learning algorithm performs online real-time judgment on the running state of the guide rail, and completes the graph of the running state of the guide rail;

Step 400 , performing pre-maintenance on wear and tear of the guide rail according to the graph of the running state of the guide rail.

As a preferred solution of the present invention, in step 100, the specific step of preprocessing the data signal is: unify the vibration signal into a specified size or length.

As a preferred solution of the present invention, in step 200, the specific steps of pre-training the data signal are:

Step 201, send the preprocessed data signal segment to the restricted Boltzmann machine in the deep belief network, and train the first RBM to reach a steady state;

Step 202, outputting the result obtained by the previous RBM as the input of the next RBM visual layer until a steady state;

Step 203, repeat step 202 until the last RBM training is completed;

Step 204, adjust the number of layers of each layer with the backpropagation algorithm, so that the entire network can find the optimal parameters, and establish a calculation model through training.

As a preferred solution of the present invention, in step 300, the specific steps for online real-time judgment of the operating state of the guide rail are:

Step 301, receiving the data signal collected by the sensor under the normal operation state of the guide rail;

Step 302, let the test signal pass through the first several layers of the calculation model, and use the results as the extracted features of the feature extractor;

Step 303, extracting the eigenvectors of the guide rail state by principal component analysis, and calculating the similarity between the eigenvectors;

Step 304 , draw a deterioration curve of the operating state of the guide rail according to the calculated similarity, and complete the graph of the operating state of the guide rail.

As a preferred solution of the present invention, in step 303, the calculation method for the similarity between feature vectors is:

Among them, d(v _modulus , v _test ) is the similarity between feature vectors, v _{modulus n} is the feature vector of the calculation model, v _{test n} is the feature vector extracted by principal component analysis, and i and n represent the number.

As a preferred solution of the present invention, in step 400, a threshold is set on the graph according to the change trend on the graph of the operating state of the guide rail, and when the detection data signal exceeds the threshold, the wear and tear of the guide rail is pre-maintained.

As a preferred solution of the present invention, the setting of the threshold is determined according to the deterioration trend or wear depth.

As a preferred solution of the present invention, the specific steps for determining the threshold according to the wear depth are:

Set the straightness A of the guide rail to be expressed by the wear depth h _m ;

According to the trend of the degradation curve of the guide rail, the accuracy attenuation change △h of the guide rail is calculated, and then the accumulative calculation of the accuracy attenuation change of the guide rail is carried out. The calculation formula is:

Among them, △h _ij represents the j-th wear of the i-th sampling point, and the _base h is the benchmark of the initial wear.

Embodiments of the present invention have the following advantages:

The invention uses a deep learning algorithm, realizes autonomous learning, extracts features independently, performs fitting through a neural network, and grasps the wear and tear during operation in real time, has the advantages of good self-adaptation and high robustness, and can self-adapt Self-learning and self-diagnosis have high application value.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the accompanying drawings that are required in the description of the embodiments or the prior art. Apparently, the drawings in the following description are only exemplary, and those skilled in the art can also obtain other implementation drawings according to the provided drawings without creative work.

The structures, proportions, sizes, etc. shown in this manual are only used to cooperate with the content disclosed in the manual, so that people familiar with this technology can understand and read, and are not used to limit the conditions for the implementation of the present invention, so there is no technical In the substantive meaning above, any modification of structure, change of proportional relationship or adjustment of size shall still fall within the scope of the technical content disclosed in the present invention without affecting the functions and objectives of the present invention. within the range that can be covered.

Fig. 1 is a flowchart block diagram in the embodiment of the present invention;

FIG. 2 is a schematic diagram of a pre-training process in an embodiment of the present invention;

Fig. 3 is a schematic diagram of the wear depth structure according to the embodiment of the present invention.

Detailed ways

The implementation mode of the present invention is illustrated by specific specific examples below, and those who are familiar with this technology can easily understand other advantages and effects of the present invention from the contents disclosed in this description. Obviously, the described embodiments are a part of the present invention. , but not all examples. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

As shown in Figure 1, the present invention provides a method for pre-maintenance of guide rail accuracy based on deep learning, including the following steps:

Step 100, collecting the data signal of the guide rail under the normal operation state through the sensor, and preprocessing the data signal;

Step 200, pre-training the preprocessed data signal with a deep belief network of deep learning, and establishing a calculation model;

Step 300, the deep learning algorithm performs online real-time judgment on the running state of the guide rail, and completes the graph of the running state of the guide rail;

Step 400 , performing pre-maintenance on wear and tear of the guide rail according to the graph of the running state of the guide rail.

On the one hand, in the present invention, the data signal of the guide rail is mainly collected by the sensor, and the data signal is preprocessed, which facilitates the calculation of the data after the preprocessing. The present invention uses a part of the collected data itself as the training target to form a feature vector, and then calculates the similarity between the feature vectors, that is, there is no need to think about setting and intervention, and the collected data can be directly trained and processed in the model to generate the feature vector As the target of calculation, so as to achieve the purpose of self-adaptation and self-learning.

On the other hand, in the present invention, by establishing a calculation model, the running state of the guide rail is judged in real time online, and the graph of the running state of the guide rail is completed. From the graph, its changing area in the next period of time can be known, and in this area Set the threshold for warning. When the threshold is exceeded, the warning will be made and corresponding decisions will be made to achieve the purpose of self-diagnosis.

The present invention realizes automatic extraction of features according to the characteristics of the collected data through the above maintenance method, and grasps the status of the guide rail in real time through neural network fitting, can adapt to different guide rails independently, and can automatically convert the collected parameters without manual intervention Import into the neural network for calculation and monitoring.

In the present invention, since the vibration signal contains a large amount of state information of the guide rail during operation. Therefore, the sensor can be used to collect the vibration signal during the normal operation of the guide rail, and mark it as normal, and the principle of other signals is similar.

In order to facilitate calculation and data processing, in step 100, the data signal is first preprocessed, and the specific step is to unify the vibration signal into a specified size or length. In this embodiment, for example, the signal data length of each small segment is taken as T.

A standard restricted Boltzmann machine consists of a hidden layer and a visible layer, as shown in Figure 2, in which c ₁ , c ₂ , c ₃ , ..., c _n are all visible layers, in step 200, The specific steps of pre-training the data signal are as follows:

Step 201, send the preprocessed data signal segment with a length of T into the restricted Boltzmann machine (RBM) in the deep belief network, and train the first RBM to reach a stable state;

Step 202, outputting the result obtained by the previous RBM as the input of the next RBM visual layer until a steady state;

Step 203, repeat step 202 until the last RBM training is completed;

Step 204, use the backpropagation algorithm to fine-tune the number of layers in each layer, so that the entire network can find the optimal parameters {W, a, b}, and establish a calculation model through training.

The present invention is characterized in that the training of the neural network is directly performed on the preprocessed data. In the training process, the source and accuracy of the data are not considered, and the original data is directly trained, and the training results are directly applied to the following A training cycle until all the training is completed to establish its calculation model. The advantage of this method is that it has a strong ability to adapt to the data independently, and it can directly process the original data and make corresponding decisions without human intervention and guidance during the calculation process.

In the above steps, set specific parameters to further illustrate the above training process:

Set the RBM to consist of the visible layer V and the hidden layer H, where a _n is the bias of the nth unit of the visible layer, b _n is the bias of the nth unit of the hidden layer, W is the visible layer V and the hidden layer H A vector of link weights.

In the first step, when the training sample V ₁ =(v ₁₁ ,v ₁₂ ,v ₁₂ ,…,v _1(n-1 ),v _1n ) is input to the visible layer V, the output of the hidden layer H is obtained through Gibas sampling H ₁ =(h ₁₁ ,h ₁₂ ,h ₁₂ ,...,h _1(n-1) ,h _1n ), where h _1n =Gibas(f _s (a _n +∑ _i w _in v _in )), f _s It is an s-type activation function, and the value generated by f _s needs to be between (0, 1). If it is greater than 1, take 1, and if it is less than 0, take 0.

The second step is to calculate {W, a, b} in the RBM, as shown in Figure 2, in the first RBM, namely:

W=μW+ε(h′ ₁ v ₁ -h′ ₂ v ₂ )

a=μa+∑(v ₁ -v ₂ )'

b=μb+Σ(h ₁ -h ₂ )',

Among them, μ is the learning rate introduced in order to overcome the training falling into a local minimum, generally between 0.5 and 0.9, ε is the learning rate, and the value represents the step size of each adjustment, generally between 0.005 and 0.200.

The third step is to use the mean square error to judge the training accuracy of RBM, namely:

Among them, v ₁ is the input of the visual layer, and v ₂ is the reconstructed output of the visual layer.

The output obtained from the previous RBM (that is, the first one):

H ₁ =(h ₁₁ ,h ₁₂ ,h ₁₂ ,...,h _1(n-1) ,h _1n ),

As the input of the visual layer of the next (that is, the second) RBM, as shown in Figure 2: in the second RBM, H ₁ is equivalent to the V of the first RBM, and the output H ₂ is equivalent to the middle of the first RBM H ₁ , and calculate the parameters {W, a, b} in the second RBM.

In the two layers of associative memory shown in Figure 1, the top layer is used for layer number fitting, and the label neurons are used for supervised learning to facilitate fine-tuning of the entire network.

After the model is trained, the running status of the guide rail can be judged online. In step 300, the specific steps of real-time judgment are:

Step 301, receiving the data signal collected by the sensor under the normal operation state of the guide rail;

Step 302, let the test signal pass through the first several layers of the calculation model, and use the results as the extracted features of the feature extractor;

Step 303, extracting the eigenvectors of the guide rail state by principal component analysis, and calculating the similarity between the eigenvectors;

Step 304 , draw a deterioration curve of the operating state of the guide rail according to the calculated similarity, and complete the graph of the operating state of the guide rail.

Wherein, in step 303, the similarity calculation method between feature vectors is:

Among them, d(v _modulus , v _test ) is the similarity between feature vectors, v _{modulus n} is the feature vector of the calculation model, v _{test n} is the feature vector extracted by principal component analysis, and i and n represent the number.

The change of the straightness of the guide rail has a great influence on the size and surface accuracy of the processed workpiece. According to the deterioration curve, the running state change trend of the guide rail can be known. The accuracy of the guide rail can be guaranteed by setting a threshold on the graph, or the threshold can be set in combination with the wear depth , and make corresponding decisions when the threshold is exceeded.

Specifically, in step 400, a threshold is set on the graph according to the change trend of the guide rail operation status graph, and a corresponding decision is made when the detected data signal exceeds the threshold, wherein the setting of the threshold is determined according to the deterioration trend or wear depth.

As shown in Figure 3, the specific steps for determining the threshold according to the wear depth are:

The straightness A of the set guide rail can be expressed by the wear depth h _m ;

According to the trend of the deterioration curve of the guide rail, the accuracy attenuation change △h of the guide rail is calculated, and the cumulative calculation is carried out through each wear △h _ij . The calculation formula is:

Among them, A is the straightness of the final guide rail, △h _ij represents the jth wear of the i-th sampling point, and the _base h is the benchmark of the initial wear. Determine the final threshold A _max according to the above formula. When A<A _max , it means that the accuracy of the guide rail is within the safe range. The effective accuracy life of the guide rail at this time can be obtained through AA _max , and corresponding decisions can be made according to the remaining accuracy or threshold.

In the above, the wear depth is a fixed value, which can be determined directly, but in the present invention, how to determine its threshold according to the deterioration trend is the key. That is to say, the degree of degradation at this time is determined according to the actual wear of different sampling points. It is a process of dynamic change. Combined with the self-learning method in the present invention, it defines the specific maximum threshold in the process of dynamic change. , and make corresponding decisions based on the difference between the real-time measured straightness and the threshold.

The existing technical solutions related to the guide rail accuracy maintenance model mainly focus on the use of statistical laws, based on the Archard model, and the use of theoretical models of dynamic characteristics. Compared with them, the present invention uses deep learning algorithms to achieve autonomous learning , extract features independently, fit through neural network, and grasp the wear and tear during operation in real time. It has good self-adaptation, high robustness, self-adaptation, self-learning and self-diagnosis, and has high application value.

Although the present invention has been described in detail with general descriptions and specific examples above, it is obvious to those skilled in the art that some modifications or improvements can be made on the basis of the present invention. Therefore, the modifications or improvements made on the basis of not departing from the spirit of the present invention all belong to the protection scope of the present invention.

Claims (8)

1. a guide rail precision pre-maintenance method based on deep learning, is characterized in that, comprises the steps: Step 100, collecting the data signal of the guide rail under the normal operation state through the sensor, and preprocessing the data signal; Step 200, pre-training the preprocessed data signal with a deep belief network of deep learning, and establishing a calculation model; Step 300, the deep learning algorithm performs online real-time judgment on the running state of the guide rail, and completes the graph of the running state of the guide rail; Step 400 , performing pre-maintenance on wear and tear of the guide rail according to the graph of the running state of the guide rail. 2. A method for pre-maintenance of guide rail accuracy based on deep learning according to claim 1, characterized in that, in step 100, the specific step of preprocessing the data signal is: unify the vibration signal into a specified size or length . 3. A method for pre-maintenance of guide rail accuracy based on deep learning according to claim 1, wherein in step 200, the specific steps of pre-training the data signal are: Step 201, send the preprocessed data signal segment to the restricted Boltzmann machine in the deep belief network, and train the first RBM to reach a steady state; Step 202, outputting the result obtained by the previous RBM as the input of the next RBM visual layer until a steady state; Step 203, repeat step 202 until the last RBM training is completed; Step 204, adjust the number of layers of each layer with the backpropagation algorithm, so that the entire network can find the optimal parameters, and establish a calculation model through training. 4. A method for pre-maintenance of guide rail accuracy based on deep learning according to claim 1, characterized in that, in step 300, the specific steps for online real-time judgment of the operating state of the guide rail are: Step 301, receiving the data signal collected by the sensor under the normal operation state of the guide rail; Step 302, let the test signal pass through the first several layers of the calculation model, and use the results as the extracted features of the feature extractor; Step 303, extracting the eigenvectors of the guide rail state by principal component analysis, and calculating the similarity between the eigenvectors; Step 304 , draw a deterioration curve of the operating state of the guide rail according to the calculated similarity, and complete the graph of the operating state of the guide rail. 5. A method for pre-maintenance of guide rail accuracy based on deep learning according to claim 4, characterized in that, in step 303, the similarity calculation method between feature vectors is: Among them, d(v _modulus , v _test ) is the similarity between feature vectors, v _{modulus n} is the feature vector of the calculation model, v _{test n} is the feature vector extracted by principal component analysis, and i and n represent the number. 6. A method for pre-maintenance of guide rail accuracy based on deep learning according to claim 1, characterized in that in step 400, a threshold is set on the graph according to the change trend on the graph of the operating state of the guide rail, when detection Early preventive maintenance of guide rail wear when the data signal exceeds the threshold. 7. A method for predictive maintenance of guide rail accuracy based on deep learning according to claim 6, wherein the setting of the threshold is determined according to the deterioration trend or wear depth. 8. A method for pre-maintenance of guide rail accuracy based on deep learning according to claim 7, wherein the specific steps of determining the threshold according to the wear depth are: Set the straightness A of the guide rail to be expressed by the wear depth h _m ; According to the trend of the degradation curve of the guide rail, the accuracy attenuation change △h of the guide rail is calculated, and then the accumulative calculation of the accuracy attenuation change of the guide rail is carried out. The calculation formula is: Among them, △h _ij represents the j-th wear of the i-th sampling point, and the _base h is the benchmark of the initial wear.