Intelligent judgment method for machining state of numerical control machine tool
By constructing graph network data and training fault detection models, using vibration data from multiple data acquisition points for information aggregation, the problem of low fault detection accuracy in the prior art is solved, and more accurate identification of CNC machine tools is achieved.
Patent Information
- Application Number
- CN202411061771.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing CNC machine tool fault detection methods are based only on a single vibration sensor data analysis, and cannot fully reflect the processing status of the entire machine tool, resulting in low fault detection accuracy.
By collecting vibration data and operation data of multiple data acquisition points within a preset time period, the graph network data is constructed, and the vibration data of each data acquisition point is aggregated using the graph network model, the feature sequence is extracted, and fault detection is trained Model to identify the type of failure of CNC machine tools.
It improves the accuracy of CNC machine tool fault detection and can more accurately identify bearing damage, tool wear and rail failure.
Smart Images

Figure CN119004205A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of numerical control machine tool fault detection, and more specifically, to an intelligent determination method for the machining state of a numerical control machine tool. Background Art
[0002] CNC machine tools are indispensable key equipment in modern manufacturing. They control the movements of machine tools through pre-programmed computer software, and can accurately perform cutting, drilling, milling and other operations according to the requirements of design drawings, greatly improving production efficiency and part processing accuracy. As the manufacturing industry continues to increase its requirements for product quality and production efficiency, the reliability and stability of CNC machine tools have become particularly important. Once a failure occurs, it will not only affect the processing accuracy, but may also delay production. Therefore, effective fault detection of CNC machine tools is crucial.
[0003] The existing fault detection method for CNC machine tools usually collects vibration data through a vibration sensor, extracts the time domain features of the vibration data by combining a timing analysis model, inputs the features into a classifier, and outputs the fault type of the CNC machine tool. The fault type may be bearing damage, tool wear, or guide rail failure, etc. The timing analysis model may be RNN or LSTM, etc.; or the vibration data is converted into frequency domain data through Fourier transform, the frequency domain features of the frequency domain data are first extracted through CNN, and then the time domain features of the vibration data are extracted through a timing analysis model and input into a classifier, and the fault type of the CNC machine tool is output.
[0004] However, the above scheme only analyzes the vibration data collected by a single vibration sensor, which may not reflect the processing status of the entire CNC machine tool. The vibration data of different fault points may overlap, resulting in low accuracy in fault detection of the CNC machine tool. Summary of the invention
[0005] The present invention provides a method for intelligently distinguishing the machining status of a numerically controlled machine tool, which solves the technical problems in the above-mentioned background technology.
[0006] The present invention provides a method for intelligently distinguishing the machining state of a numerically controlled machine tool, comprising the following steps:
[0007] Step S101, within a preset time period T, collecting vibration data and operation data of M data collection points of a CNC machine tool through a vibration sensor;
[0008] The vibration data includes N time points, each of which corresponds to an acceleration value;
[0009] The operation data includes N time points, each of which corresponds to a temperature value;
[0010] Step S102, preprocessing the vibration data of the M data collection points to obtain a feature sequence of the M data collection points;
[0011] The feature sequence includes N sequence units, the nth sequence unit represents the feature vector at the nth time point, and the feature vector is generated according to the vibration data and the operation data, where 1≤n≤N;
[0012] Step S103, constructing graph network data according to the feature sequences of the M data collection points;
[0013] Graph network data includes: nodes, initial features of nodes, and edges between nodes;
[0014] The node establishes data connection with the data collection point;
[0015] The initial features of a node are represented by the feature sequence of the data collection points that establish data contact with the node;
[0016] If the similarity value between the initial features of any two nodes is greater than or equal to the preset threshold, an edge is constructed between the two nodes;
[0017] Step S104, constructing and training a fault detection model;
[0018] Step S105, inputting the graph network data into the trained fault detection model, and the output value represents the fault type of the CNC machine tool;
[0019] Fault types include: bearing damage, tool wear and guideway failure.
[0020] Furthermore, T, M, N and the preset threshold are all custom parameters.
[0021] Furthermore, the vibration data is preprocessed to obtain a feature sequence, including the following steps:
[0022] Step S201, sampling the vibration data and operation data within a preset time period T at equal time intervals to obtain vibration data and operation data including N time points;
[0023] Step S202, performing noise reduction processing on the vibration data at N time points by using a Gaussian filter;
[0024] Step S203, converting the noise-reduced vibration data at N time points into frequency domain data by fast Fourier transform;
[0025] The frequency domain data includes the power spectrum density values corresponding to K frequency values, where K is a custom parameter;
[0026] Step S204: concatenate the frequency domain data of N time points with the operation data to obtain a feature vector.
[0027] Furthermore, the edges between nodes also include constructed edges between adjacent nodes.
[0028] Furthermore, the calculation formula for the similarity value between the initial features of any two nodes is as follows:
[0029]
[0030] Where 1≤u≤M, 1≤v≤M, u≠v, sim u,v represents the similarity value between the initial features of the u-th node and the v-th node, and The feature vectors at the nth time point represent the initial features of the uth node and the vth node respectively.
[0031] Further, the fault detection model includes: M first hidden layers, M extraction layers, a second hidden layer, a third hidden layer and a first classifier;
[0032] M first hidden layers share weight parameters and bias parameters
[0033] Each first hidden layer includes N hidden units;
[0034] The nth hidden unit of the mth first hidden layer inputs the nth sequence unit of the feature sequence corresponding to the initial feature of the mth node, and outputs the first update vector, where 1≤m≤M, 1≤n≤N;
[0035] The m-th extraction layer is used to extract the first update vector output by the N-th hidden unit of the m-th first hidden layer as the initial feature of the m-th node of the graph network data;
[0036] The second hidden layer inputs the graph network data and outputs a graph update matrix, where a row vector of the graph update matrix corresponds to the second update vector of a node;
[0037] The third hidden layer inputs the graph update matrix and outputs the third update vector;
[0038] The first classifier inputs the third update vector, and the classification space of the first classifier represents the fault type of the CNC machine tool.
[0039] Furthermore, the calculation formula for the nth hidden unit includes:
[0040]
[0041] in represents the first update vector of the output of the nth hidden unit, X n represents the nth sequence unit of the feature sequence input to the nth hidden unit, and They represent the forward update gate, forward reset gate, forward first intermediate state, and forward second intermediate state of the nth hidden unit, respectively. and They represent the backward update gate, backward reset gate, backward first intermediate state and backward second intermediate state of the nth hidden unit respectively. and denote the first, second, third, fourth, fifth and sixth weight parameters of the nth hidden unit, respectively. and denote the first, second, and third forward bias parameters of the nth hidden unit, respectively. and denote the backward first, second, third, fourth, fifth and sixth weight parameters of the nth hidden unit, respectively. and denote the first, second, and third backward bias parameters of the nth hidden unit, respectively. and They represent the forward second intermediate state and backward second intermediate state of the n-1th hidden unit, respectively. Assign a value of 0, The value is assigned to 0, sigmoid represents the sigmoid activation function, tanh represents the hyperbolic tangent activation function, ⊙ represents point-by-point multiplication, and concat represents the concatenation operation of vectors.
[0042] Furthermore, the calculation formula of the second hidden layer includes:
[0043]
[0044] Attion u,v =(U1h u T ) T ⊙tanh(Amount u,v U2h v T +b1);
[0045] Where 1≤u≤M, 1≤v≤M, u≠v, S represents the graph update matrix, Col u represents the set of nodes that have edges connected to the u-th node, represents the second update vector of the u-th node, h u and h v Represent the first update vector of the u-th node and the v-th node respectively, Attion u,v , U u,v and Amount u,vThey represent the attention score, weight parameter and number of edges between the uth node and the vth node, respectively. U1, U2 and b1 represent the first weight parameter, the second weight parameter and the bias parameter, respectively. M Indicates that the second update vectors of M nodes are stacked, T indicates a transpose operation, sigmoid indicates a sigmoid activation function, and tanh indicates a hyperbolic tangent activation function;
[0046] The calculation formula for the third hidden layer is as follows:
[0047] Vec=(S×U3) T ×(S×U4)+b3;
[0048] Wherein Vec represents the third update vector output by the third hidden layer, S represents the graph update matrix input by the third hidden layer, U3 and U4 represent the third weight parameter and the fourth weight parameter respectively, b3 represents the third bias parameter, and T represents the transpose operation.
[0049] Furthermore, the fault type of a simulated CNC machine tool is used as a sample label of a training sample for training a fault detection model, and graph network data is constructed according to steps S101 to S103 as sample data of a training sample for training a fault detection model.
[0050] Furthermore, the first hidden layer is pre-trained before the fault detection model is trained. During the pre-training process, the first update vector output by the Nth hidden unit of the first hidden layer is input into the second classifier and the third classifier. The classification spaces of the second classifier and the third classifier represent the acceleration values at the N-1th time point and the N+1th time point, respectively.
[0051] The beneficial effect of the present invention is that the present invention extracts features from the vibration data of each data collection point through a fault detection model, and uses a graph network to aggregate information on the vibration data of each data collection point, thereby improving the accuracy of fault detection of CNC machine tools. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 It is a flow chart of a method for intelligently distinguishing machining status of a numerical control machine tool according to the present invention;
[0053] Figure 2 It is a flow chart of the present invention for preprocessing vibration data to obtain a feature sequence;
[0054] Figure 3 It is a comparison chart of the test results of the present invention. DETAILED DESCRIPTION
[0055] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.
[0056] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0057] like Figure 1 to Figure 3 As shown, a method for intelligently distinguishing the machining state of a CNC machine tool comprises the following steps:
[0058] Step S101, within a preset time period T, collecting vibration data and operation data of M data collection points of a CNC machine tool through a vibration sensor;
[0059] The vibration data includes N time points, each of which corresponds to an acceleration value;
[0060] The operation data includes N time points, each of which corresponds to a temperature value;
[0061] Step S102, preprocessing the vibration data of the M data collection points to obtain a feature sequence of the M data collection points;
[0062] The feature sequence includes N sequence units, the nth sequence unit represents the feature vector at the nth time point, and the feature vector is generated according to the vibration data and the operation data, where 1≤n≤N;
[0063] Step S103, constructing graph network data according to the feature sequences of the M data collection points;
[0064] Graph network data includes: nodes, initial features of nodes, and edges between nodes;
[0065] The node establishes data connection with the data collection point;
[0066] The initial features of a node are represented by the feature sequence of the data collection points that establish data contact with the node;
[0067] If the similarity value between the initial features of any two nodes is greater than or equal to the preset threshold, an edge is constructed between the two nodes;
[0068] Step S104, constructing and training a fault detection model;
[0069] Step S105, inputting the graph network data into the trained fault detection model, and the output value represents the fault type of the CNC machine tool;
[0070] Fault types include: bearing damage, tool wear and guideway failure.
[0071] It should be noted that the fault types of CNC machine tools can also include: gearbox failure, transmission system failure and loose mechanical structure, etc. In addition, bearing failure, gearbox failure, transmission system failure and guide rail failure can be further refined. For example, bearing failure can include: inner ring, outer ring or rolling element wear, gearbox failure can include: gear wear and poor gear meshing, transmission system failure can include: belt or chain looseness, coupling damage, guide rail failure can include: guide rail wear and increased screw nut clearance.
[0072] In one embodiment of the present invention, T, M, N and the preset threshold are all custom parameters. Preferably, T is set to 10 seconds, M is set to 8, N is set to 20, and the preset threshold is set to 0.5.
[0073] It should be noted that when M is set to 8, that is, the number of data collection points is 8, which can be set at the following positions of the CNC machine tool: 4 vibration sensors are installed near the spindle housing or bearing seat, 2 vibration sensors are installed on the tool clamping mechanism, 1 vibration sensor is installed near the support structure or guide seat of the linear guide, and 1 vibration sensor is installed near the bearing seat or support seat of the ball screw; if the CNC machine tool does not have a linear guide and a ball screw, 6 vibration sensors are installed near the spindle housing or the bearing seat.
[0074] In one embodiment of the present invention, the operation data may further include: a voltage value and a current value of the servo motor.
[0075] In one embodiment of the present invention, Figure 2 As shown, the vibration data is preprocessed to obtain a feature sequence, including the following steps:
[0076] Step S201, sampling the vibration data and operation data within a preset time period T at equal time intervals to obtain vibration data and operation data including N time points;
[0077] For example, if T is set to 10 seconds and N is set to 20, the time interval is set to 0.5 seconds;
[0078] Step S202, performing noise reduction processing on the vibration data at N time points by using a Gaussian filter;
[0079] Step S203, converting the noise-reduced vibration data at N time points into frequency domain data by fast Fourier transform;
[0080] The frequency domain data includes power spectrum density values corresponding to K frequency values, where K is a custom parameter, and preferably, K is set to 1000;
[0081] Step S204: concatenate the frequency domain data of N time points with the operation data to obtain a feature vector.
[0082] It should be noted that the characteristic sequence is expressed as: A = {a1…a N}, where a1…a N Respectively represent the feature vectors from the 1st time point to the Nth time point;
[0083] The feature vector at the Nth time point is expressed as: in They represent the power spectral density values corresponding to the 1st frequency value to the Kth frequency value at the Nth time point, respectively. The K+1th dimension value of the feature vector at the Nth time point, that is, the temperature value.
[0084] In one embodiment of the present invention, the edges between nodes also include construction edges between adjacent nodes.
[0085] It should be noted that the similarity value between the initial features of any two nodes may be less than the preset threshold, and edges cannot be built between the nodes in the graph network data. Therefore, it is necessary to build edges between adjacent nodes to ensure the transmission and aggregation of the initial features between nodes, thereby ensuring the normal use of the fault detection model.
[0086] In one embodiment of the present invention, the calculation formula of the similarity value between the initial features of any two nodes is as follows:
[0087]
[0088] Where 1≤u≤M, 1≤v≤M, u≠v, sim u,vrepresents the similarity value between the initial features of the u-th node and the v-th node, and The feature vectors at the nth time point represent the initial features of the uth node and the vth node respectively.
[0089] In one embodiment of the present invention, the fault detection model includes: M first hidden layers, M extraction layers, a second hidden layer, a third hidden layer and a first classifier;
[0090] M first hidden layers share weight parameters and bias parameters
[0091] Each first hidden layer includes N hidden units;
[0092] The nth hidden unit of the mth first hidden layer inputs the nth sequence unit of the feature sequence corresponding to the initial feature of the mth node, and outputs the first update vector, where 1≤m≤M, 1≤n≤N;
[0093] The m-th extraction layer is used to extract the first update vector output by the N-th hidden unit of the m-th first hidden layer as the initial feature of the m-th node of the graph network data;
[0094] The second hidden layer inputs the graph network data and outputs a graph update matrix, where a row vector of the graph update matrix corresponds to a second update vector of a node; the third hidden layer inputs the graph update matrix and outputs a third update vector;
[0095] The first classifier inputs the third update vector, and the classification space of the first classifier represents the fault type of the CNC machine tool.
[0096] In one embodiment of the present invention, the calculation formula of the nth hidden unit includes:
[0097]
[0098]
[0099] in represents the first update vector of the output of the nth hidden unit, X n represents the nth sequence unit of the feature sequence input to the nth hidden unit, and They represent the forward update gate, forward reset gate, forward first intermediate state, and forward second intermediate state of the nth hidden unit, respectively. and They represent the backward update gate, backward reset gate, backward first intermediate state and backward second intermediate state of the nth hidden unit respectively. and denote the first, second, third, fourth, fifth and sixth weight parameters of the nth hidden unit, respectively. and denote the first, second, and third forward bias parameters of the nth hidden unit, respectively. and denote the backward first, second, third, fourth, fifth and sixth weight parameters of the nth hidden unit, respectively. and denote the first, second, and third backward bias parameters of the nth hidden unit, respectively. and They represent the forward second intermediate state and backward second intermediate state of the n-1th hidden unit, respectively. Assign a value of 0, The value is assigned to 0, sigmoid represents the sigmoid activation function, tanh represents the hyperbolic tangent activation function, ⊙ represents point-by-point multiplication, and concat represents the concatenation operation of vectors.
[0100] In one embodiment of the present invention, the calculation formula of the second hidden layer includes:
[0101]
[0102] Attion u,v =(U1h u T ) T ⊙tanh(Amount u,v U2h v T +b1);
[0103] Where 1≤u≤M, 1≤v≤M, u≠v, S represents the graph update matrix, Col u represents the set of nodes that have edges connected to the u-th node, represents the second update vector of the u-th node, h u and h v Represents the first update vector of the u-th node and the v-th node respectively, Attion u,v , U u,v and Amount u,v They represent the attention score, weight parameter and number of edges between the uth node and the vth node, respectively. U1, U2 and b1 represent the first weight parameter, the second weight parameter and the bias parameter, respectively. M It indicates that the second update vectors of M nodes are stacked, T indicates the transpose operation, sigmoid indicates the sigmoid activation function, and tanh indicates the hyperbolic tangent activation function.
[0104] It should be noted that, for example, the first update vector of the node is a vector of size 1×64, and the first update vector is transposed to obtain a vector of size 64×1. The first weight parameter can be designed as a matrix of size 16×64, and the first weight parameter is multiplied by the transpose of the first update vector to obtain a vector of size 16×1. Then the vector of size 16×1 is transposed to obtain a vector of size 1×16. Similarly, the second weight parameter can be designed as a matrix of size 16×64, and then multiplied by the transpose of the first update vector of the node to obtain a vector of size 16×1. Therefore, the attention score between the u-th node and the v-th node is a real number of size 1×1.
[0105] In one embodiment of the present invention, the calculation formula of the third hidden layer is as follows:
[0106] Vec=(S×U3) T ×(S×U4)+b3;
[0107] Wherein Vec represents the third update vector output by the third hidden layer, S represents the graph update matrix input by the third hidden layer, U3 and U4 represent the third weight parameter and the fourth weight parameter respectively, b3 represents the third bias parameter, and T represents the transpose operation.
[0108] It should be noted that, for example, S is a vector of size 8×64, the third weight parameter can be designed as a matrix of size 64×32, the two are multiplied to obtain a matrix of size 8×32, and after transposition, a matrix of size 32×8 is obtained. The fourth weight parameter can be designed as a vector of size 64×1, S is multiplied by the fourth weight parameter to obtain a vector of size 8×1, and the matrix of size 32×8 and the vector of size 8×1 are multiplied to obtain a vector of size 32×1.
[0109] In one embodiment of the present invention, the fault type of a simulated CNC machine tool is used as a sample label of a training sample for training a fault detection model, and graph network data is constructed according to steps S101 to S103 as sample data of a training sample for training a fault detection model.
[0110] It should be noted that the weight parameters and bias parameters in the fault detection model are all learnable parameters. In the process of fault detection model training, the difference between the predicted value output by the fault detection model at each iteration and the true value corresponding to the sample label of the training sample is specified as the loss function. The gradient of the weight parameters and bias parameters in the fault detection model is first calculated using the chain rule, and then the weight parameters and bias parameters in the fault detection model are updated through back propagation and gradient descent algorithm according to the gradient of the weight parameters and bias parameters, so that the loss function is minimized until the iteration termination condition is met. The training is completed. The training of the neural network model is a conventional technical means and will not be elaborated here.
[0111] In one embodiment of the present invention, the first hidden layer is pre-trained before the fault detection model is trained. During the pre-training process, the first update vector output by the Nth hidden unit of the first hidden layer is input into the second classifier and the third classifier. The classification spaces of the second classifier and the third classifier represent the acceleration values at the N-1th time point and the N+1th time point, respectively.
[0112] It should be noted that pre-training can improve the ability of the first hidden layer to extract features, accelerate the learning process of subsequent tasks, improve the generalization ability of the model, reduce the number of training samples to a certain extent, and improve the accuracy of fault detection by the fault detection model.
[0113] like Figure 3 As shown, the fault detection model provided by the present invention tests the fault types of 300 test data, the number of test data for each fault type is 100, and the detection accuracy of the fault detection model provided by the present invention is 100+99+99 divided by 300, which is equal to 99.33%.
[0114] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.
Claims
1. A method for intelligently distinguishing the machining status of a CNC machine tool, characterized in that: The following steps are involved: Step S101, within a preset time period T, collecting vibration data and operation data of M data collection points of a CNC machine tool through a vibration sensor; The vibration data includes N time points, each of which corresponds to an acceleration value; The operation data includes N time points, each of which corresponds to a temperature value; Step S102, preprocessing the vibration data of the M data collection points to obtain a feature sequence of the M data collection points; The feature sequence includes N sequence units, the nth sequence unit represents the feature vector at the nth time point, and the feature vector is generated according to the vibration data and the operation data, where 1≤n≤N; Step S103, constructing graph network data according to the feature sequences of the M data collection points; Graph network data includes: nodes, initial features of nodes, and edges between nodes; The node establishes data connection with the data collection point; The initial features of a node are represented by the feature sequence of the data collection points that establish data contact with the node; If the similarity value between the initial features of any two nodes is greater than or equal to the preset threshold, an edge is constructed between the two nodes; Step S104, constructing and training a fault detection model; Step S105, inputting the graph network data into the trained fault detection model, and the output value represents the fault type of the CNC machine tool; Fault types include: bearing damage, tool wear and guideway failure.
2. The method for intelligently distinguishing machining status of a CNC machine tool according to claim 1, characterized in that: T, M, N and preset thresholds are all custom parameters.
3. The method for intelligently distinguishing machining status of a CNC machine tool according to claim 1, characterized in that: Preprocessing the vibration data to obtain the feature sequence includes the following steps: Step S201, sampling the vibration data and operation data within a preset time period T at equal time intervals to obtain vibration data and operation data including N time points; Step S202, performing noise reduction processing on the vibration data at N time points by using a Gaussian filter; Step S203, converting the noise-reduced vibration data at N time points into frequency domain data by fast Fourier transform; The frequency domain data includes the power spectrum density values corresponding to K frequency values, where K is a custom parameter; Step S204: concatenate the frequency domain data of N time points with the operation data to obtain a feature vector.
4. The method for intelligently distinguishing machining status of a CNC machine tool according to claim 1, characterized in that: The edges between nodes also include the edges between adjacent nodes.
5. The method for intelligently distinguishing machining status of a CNC machine tool according to claim 1, characterized in that: The calculation formula for the similarity value between the initial features of any two nodes is as follows: Where 1≤u≤M, 1≤v≤M, u≠v, sim u,v represents the similarity value between the initial features of the u-th node and the v-th node, and The feature vectors at the nth time point represent the initial features of the uth node and the vth node respectively.
6. The method for intelligently distinguishing machining status of a CNC machine tool according to claim 1, characterized in that: The fault detection model includes: M first hidden layers, M extraction layers, a second hidden layer, a third hidden layer and a first classifier; M first hidden layers share weight parameters and bias parameters Each first hidden layer includes N hidden units; The nth hidden unit of the mth first hidden layer inputs the nth sequence unit of the feature sequence corresponding to the initial feature of the mth node, and outputs the first update vector, where 1≤m≤M, 1≤n≤N; The m-th extraction layer is used to extract the first update vector output by the N-th hidden unit of the m-th first hidden layer as the initial feature of the m-th node of the graph network data; The second hidden layer inputs the graph network data and outputs a graph update matrix, where a row vector of the graph update matrix corresponds to the second update vector of a node; The third hidden layer inputs the graph update matrix and outputs the third update vector; The first classifier inputs the third update vector, and the classification space of the first classifier represents the fault type of the CNC machine tool.
7. The method for intelligently distinguishing machining status of a CNC machine tool according to claim 6, characterized in that: The calculation formula for the nth hidden unit includes: in represents the first update vector of the output of the nth hidden unit, X n represents the nth sequence unit of the feature sequence input to the nth hidden unit, and They represent the forward update gate, forward reset gate, forward first intermediate state, and forward second intermediate state of the nth hidden unit, respectively. and They represent the backward update gate, backward reset gate, backward first intermediate state and backward second intermediate state of the nth hidden unit respectively. and denote the first, second, third, fourth, fifth and sixth weight parameters of the nth hidden unit, respectively. and denote the first, second, and third forward bias parameters of the nth hidden unit, respectively. and denote the backward first, second, third, fourth, fifth and sixth weight parameters of the nth hidden unit, respectively. and denote the first, second, and third backward bias parameters of the nth hidden unit, respectively. and They represent the forward second intermediate state and backward second intermediate state of the n-1th hidden unit, respectively. Assign a value of 0, The value is assigned to 0, sigmoid represents the sigmoid activation function, tanh represents the hyperbolic tangent activation function, ⊙ represents point-by-point multiplication, and concat represents the concatenation operation of vectors.
8. The method for intelligently distinguishing machining status of a CNC machine tool according to claim 6, characterized in that: The calculation formula for the second hidden layer includes: Attion u,v =(U1h u T ) T ⊙tanh(Amount u,v U2h v T +b1); Where 1≤u≤M, 1≤v≤M, u≠v, S represents the graph update matrix, Col u represents the set of nodes that have edges connected to the u-th node, represents the second update vector of the u-th node, h u and h v Represents the first update vector of the u-th node and the v-th node respectively, Attion u,v , U u,v and Amount u,v They represent the attention score, weight parameter and number of edges between the uth node and the vth node, respectively. U1, U2 and b1 represent the first weight parameter, the second weight parameter and the bias parameter, respectively. M Indicates that the second update vectors of M nodes are stacked, T indicates a transpose operation, sigmoid indicates a sigmoid activation function, and tanh indicates a hyperbolic tangent activation function; The calculation formula for the third hidden layer is as follows: More=(S×U3) T ×(S×U4)+b3; Wherein Vec represents the third update vector output by the third hidden layer, S represents the graph update matrix input by the third hidden layer, U3 and U4 represent the third weight parameter and the fourth weight parameter respectively, b3 represents the third bias parameter, and T represents the transpose operation.
9. The method for intelligently distinguishing machining status of a CNC machine tool according to claim 1, characterized in that: The fault type of a simulated CNC machine tool is used as a sample label of a training sample for training a fault detection model, and graph network data is constructed according to steps S101 to S103 as sample data of a training sample for training a fault detection model.
10. The method for intelligently distinguishing machining status of a CNC machine tool according to claim 6, characterized in that: The first hidden layer is pre-trained before the fault detection model is trained. During the pre-training process, the first update vector output by the Nth hidden unit of the first hidden layer is input into the second classifier and the third classifier. The classification spaces of the second classifier and the third classifier represent the acceleration values at the N-1th time point and the N+1th time point, respectively.
Citation Information
Cited By
Electricity utilization information acquisition system and method for identifying faults of electric energy meter
CN120559564A
