CN112329625A - Cutter wear state real-time identification method and model based on deep learning - Google Patents
Cutter wear state real-time identification method and model based on deep learning Download PDFInfo
- Publication number
- CN112329625A CN112329625A CN202011222866.9A CN202011222866A CN112329625A CN 112329625 A CN112329625 A CN 112329625A CN 202011222866 A CN202011222866 A CN 202011222866A CN 112329625 A CN112329625 A CN 112329625A
- Authority
- CN
- China
- Prior art keywords
- layer
- output
- time
- wear state
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Abstract
The invention discloses a method and a model for identifying the wear state of a cutter in real time based on deep learning, wherein the method comprises the following steps: a neural network model formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-short time memory network layer and an output layer is used as an identification network model; and taking the obtained original power signal generated by the machine tool in the working process as the input of the trained recognition network model, and carrying out real-time recognition on the tool wear state of the machine tool to obtain the real-time recognition result of the tool wear state of the machine tool. The method can meet the requirement of monitoring the wear state of the cutter in real time, and can achieve the effect of accurately monitoring the wear state of the cutter in real time by processing the power data generated in the operation process of the cutter on line.
Description
Technical Field
The invention relates to the field of machine tool cutter wear state identification, in particular to a cutter wear state real-time identification method and a cutter wear state real-time identification model based on deep learning.
Background
At present, the wear state of the tool can be identified by signal acquisition, and the existing identification method mainly adopts the following methods for signal acquisition, wherein the method is off-line acquisition and comprises an optical image method, a contact method and the like. Off-line acquisition can only be monitored after the tool is removed in a shutdown state, which not only results in increased downtime, but also affects machining accuracy. The other type is online acquisition, which acquires signals such as power and cutting force of a machine tool on line and then identifies the state of tool wear through the online acquired signals.
The tool wear state identification in the current online acquisition mode mainly extracts characteristics of acquired signals through signal analysis, and then processes the extracted characteristics based on a simulation model or a classifier so as to evaluate the wear state of the tool. However, in the current identification method, a series of signal processing is required for feature extraction, which causes problems of poor real-time performance, low accuracy and the like, and this process limits the application of the tool wear identification method in online monitoring. In addition, extracting the identifying features relies on a large amount of training data or expert knowledge, which also increases the difficulty of feature selection and presents a challenge to accurately identify the wear state of the tool.
In the method for monitoring the state by using other kinds of online collected signals, deep learning related technologies are used, and the methods can automatically extract the characteristics, but the signals adopted in the methods have certain difficulties in the collection process. Such as: when the acceleration signal is used for condition monitoring, an accelerometer is needed, so that the method has the defects of difficult sensor installation, expensive acquisition equipment, easy signal influence caused by noise and the like, and therefore, the method is difficult to be widely applied.
Disclosure of Invention
Based on the problems existing in the prior art, the invention aims to provide a tool wear state real-time identification method and a tool wear state real-time identification model based on deep learning, which can solve the problems of long time effect, poor real-time performance, expensive acquisition equipment, easy influence of noise on signals, difficult sensor installation and the like of the existing method for identifying the tool wear state by acquiring signals on line.
The purpose of the invention is realized by the following technical scheme:
the embodiment of the invention provides a cutter wear state real-time identification method based on deep learning, wherein a neural network model formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-short time memory network layer and an output layer is used as an identification network model;
and taking the obtained original power signal generated by the machine tool in the working process as the input of the trained recognition network model, and carrying out real-time recognition on the tool wear state of the machine tool to obtain the real-time recognition result of the tool wear state of the machine tool.
The embodiment of the invention also provides a tool wear state real-time identification model based on deep learning, which is a neural network model formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-short time memory network layer and an output layer.
According to the technical scheme provided by the invention, the method and the model for identifying the wear state of the cutter in real time based on deep learning have the advantages that:
the method comprises the steps of adopting a neural network model formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-time and short-time memory network and an output layer as an identification network model, taking an obtained original power signal generated by a machine tool in the working process as input, and utilizing the trained identification network model to carry out real-time identification on the wear state of a cutter of the machine tool so as to obtain a real-time identification result of the wear state of the cutter of the machine tool. The deep learning neural network model of the method integrates the advantages of the convolutional neural network and the long-time and short-time memory network, and the input signals adopt power signals from the cutting process of the cutter, and the power signals comprise global features and local features. Therefore, local correlation of the time series signals is acquired through a convolutional network before global features are identified to integrate local features. Then, the identification method can process the current input based on the previous state information by virtue of the long-term storage of the LSTM network, thereby achieving the effect of processing the global features; compared with the traditional TCM which processes the power signal by using a signal analysis method, the identification method can automatically extract and classify the characteristics of the original signal. Meanwhile, due to the introduction of the LSTM network, the identification method has a good classification effect when processing mass data, can meet the requirement of real-time monitoring of the wear state of the cutter, and can achieve the effect of accurately monitoring the wear state of the cutter in real time by processing power data generated in the operation process of the cutter on line.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
Fig. 1 is a schematic diagram illustrating a structure of an identification network model in a method for identifying a wear state of a tool in real time according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a tool power signal for a cutting test in an identification method according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of instantaneous power signals of an 8mm cutter in different wear states in the identification method provided by the embodiment of the invention; wherein, (a) is a schematic diagram of an instantaneous power signal of an 8mm cutter in a wear state at 00min, (b) is a schematic diagram of an instantaneous power signal of an 8mm cutter in a wear state at 40min, (c) is a schematic diagram of an instantaneous power signal of an 8mm cutter in a wear state at 60min, (d) is a schematic diagram of an instantaneous power signal of an 8mm cutter in a wear state at 80min, and (e) is a schematic diagram of an instantaneous power signal of an 8mm cutter in a wear state at 100 min;
FIG. 4 is a schematic diagram illustrating a classification result of a wear state of a tool with a diameter of 8mm in the identification method according to the embodiment of the present invention;
FIG. 5 is a schematic diagram illustrating a result of classifying a wear state of a tool with a diameter of 10mm in the identification method according to an embodiment of the present invention;
the part names corresponding to the marks in the figure are as follows: 1-identifying an input layer of a network model; 2-identifying a first convolutional layer of the network model; 3-identifying a first pooling layer of the network model; 4-identifying a second convolution layer of the network model; 5-identifying a second pooling layer of the network model; 6-identifying a long-time memory network layer of the network model; 7-identifying an output layer of the network model; a-the cutting start position in the power signal; b-first cut position in power signal; c-tool arrival to workpiece position in power signal; d-tool off-workpiece position in the power signal; e-cutter fast forward position in the power signal; f-last cut position in power signal; g-the end of cut position in the power signal.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the specific contents of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, belong to the scope of protection of the present invention. Details which are not described in detail in the embodiments of the invention belong to the prior art which is known to the person skilled in the art.
As shown in fig. 1, an embodiment of the present invention provides a method for identifying a tool wear state in real time based on deep learning, in which a neural network model formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-short term memory network layer, and an output layer is used as an identification network model;
and taking the obtained original power signal generated by the machine tool in the working process as the input of the trained recognition network model, and carrying out real-time recognition on the tool wear state of the machine tool to obtain the real-time recognition result of the tool wear state of the machine tool.
In the above method, in the identification network model, the first convolutional layer and the second convolutional layer perform convolution processing on the original power signal to obtain a feature map retaining useful information, where the convolution processing is expressed by a formula (1):
in the formula (1), f represents an activation function;representing a convolution kernel;representing the ith feature map in the ith layer;the deviation is indicated.
In the above method, the first pooling layer and the second pooling layer respectively perform pooling processing on the output of the previous convolutional layer by a maximum pooling method, and a maximum statistical value of each pooling unit is selected to represent the characteristics, wherein the pooling processing is represented by formula (2):
in the formula (2), the reaction mixture is,represents the T-th neuron in the i channel in the l layer, and T represents the pooling step.
In the method, the long-time memory network optimizes the characteristics of the output of the previous convolutional layer.
In the method, the long-short term memory network layer is formed by connecting a plurality of long-short term memory units in series, the long-short term memory units have the same structure, and each long-short term memory unit is formed by a forgetting gate, an input gate and an output gate; wherein the content of the first and second substances,
the forgetting gate determines the features that remain from the previous moment to the current moment, expressed by equation (3):
ft=σ(Wf·[ht-1,xt]+bf) (3);
in the above formula (3), ftIndicating a forgetting gate; σ represents a logic function whose output is 0 to 1; wfA weight representing a forgotten door; h ist-1Represents a previous time value; x is the number oftRepresenting the current input, which is the output of the previous moment; bfA deviation indicating a forgotten door;
the input gate determines the current network input xtIs the current unit state CtThe retained characteristics are expressed by the formulas (4), (5):
it=σ(Wi·ht-1,xt]+bi) (4);
Ct=ft·Ct-1+it·tanh(Wc·[ht-1,xt]+bc) (5);
in the above formulas (4) and (5), itRepresenting an input gate; σ represents a logic function whose output is 0 to 1; wiRepresents the weight of the input gate; h ist-1Represents a previous time value; x is the number oftRepresenting the current input, which is the output of the previous moment; biIndicating the deviation of the input gate; ctRepresenting the current unit state; f. oftIndicating a forgetting gate; tanh represents an activation function; bcRepresenting the deviation of the current state unit;
the output door controls the current unit state CtTo the current value htWhich features are output, expressed by equations (6), (7):
Ot=σ(Wo·[ht-1,xt]+bo) (6);
ht=Ot·tanh(Ct) (7);
in the above formulae (6) and (7), OtIs an output gate; σ represents a logic function whose output is 0 to 1; woIs the weight of the output gate; boIs the offset of the output gate; h istRepresents the current value; tanh represents the activation function.
In the method, the input of the Softmax function of the output layer is the output of the long-time and short-time memory network, the Softmax function can convert the probability distribution of the input neurons 0-1 to classify different wear states, and the Softmax function is represented by formula (8):
in the above formula (8), aiIs the output of the ith neuron.
In the method, the first convolution layer adopts a wide convolution kernel; the second convolutional layer employs a small convolutional kernel.
In the above method, the training of the recognition network model is performed in the following manner to obtain the trained recognition network model, including:
and 3, adjusting the hyper-parameter value of the recognition network model according to the training result of the step 2 until the training result of the recognition network model is converged.
Referring to fig. 1, an embodiment of the present invention further provides a tool wear state real-time identification model based on deep learning, which is a neural network model formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-short term memory network layer, and an output layer.
In the identification model, the first convolution layer adopts a wide convolution kernel; the second convolutional layer employs a small convolutional kernel.
The embodiments of the present invention are described in further detail below.
In the method for identifying the wear state of the cutter in real time based on deep learning, the neural network model for identification is formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-short time memory network and an output layer as shown in figure 1;
in the recognition network model, the roles and the processes of the layers are as follows:
(1) an input layer:
the input layer inputs the original power signal generated by the tool during the cutting process, and the original power signal is represented as a one-dimensional time sequence (as shown in fig. 2).
(2) First convolutional layer and third convolutional layer (convolutional network structure (CNN)) (common features):
as shown in fig. 1, the first and third layers of the method are convolutional layers, and the convolution processing of the two convolutional layers shifts the convolution kernel along the time axis for time series data (i.e., the original power signal) to perform convolution operations. Thereafter, the activation function is used to process the information by adding a bias to obtain a feature map that retains useful information, which can be expressed as:
in the formula (1), f represents an activation function;representing a convolution kernel;representing the ith feature map in the ith layer;the deviation is indicated.
As shown in fig. 1, after each convolutional layer, there is a pooling layer, i.e. a second pooling layer and a third pooling layer, and the main function of each pooling layer is to reduce the number of parameters, wherein a down-sampling technique is used to reduce the dimension of feature mapping, in the identification network model of the present invention, a maximum pooling method is used, which selects the maximum statistic of the area to represent its features, and the pooling process is represented as:
in the formula (2), the reaction mixture is,represents the T-th neuron in the i channel in the l layer, and T represents the pooling step.
As can be seen from fig. 1, in the identification network model of the present invention, the first convolution layer uses a wide convolution kernel, which can improve the network reception field and reduce the high-frequency noise; furthermore, the use of wide convolution kernels may reduce the number of convolution layers, thereby reducing the complexity of network structure and parameter computation; besides the first convolutional layer, small convolutional kernels are used on other convolutional layers, and local features can be optimized and the network depth can be deepened due to the small convolutional kernels, so that the network performance is improved.
(3) Long and short term memory networks (LSTM) (distinguishing features):
as shown in fig. 1, after each convolution layer operation, a layer of long and short term memory network (LSTM) is introduced for further feature optimization, wherein the long and short term memory network (LSTM) is formed by connecting a plurality of long and short term memory units in series, the long and short term memory units have the same structure, and each long and short term memory unit is formed by a forgetting gate, an input gate and an output gate; wherein the content of the first and second substances,
the forgetting gate determines the characteristics reserved from the previous moment to the current moment, plays a role of forgetting useless information, and is represented by the following formula: f. oft=σ(Wf·[ht-1,xt]+bf) (3);
In the above formula (3), ftIndicating a forgetting gate; σ represents a logic function whose output is 0 to 1; wfA weight representing a forgotten door; h ist-1Represents a previous time value; x is the number oftRepresenting the current input, which is the output of the previous moment; bfA deviation indicating a forgotten door;
input Gate determines Current network input xtAnd is the current unit state CtThe features that remain. Can be expressed as:
it=σ(Wi·[ht-1,xt]+bi) (4);
Ct=ft·Ct-1+it·tanh(Wc·[ht-1,xt]+bc) (5);
in the above formulas (4) and (5), itRepresenting an input gate; σ represents a logic function whose output is 0 to 1; wiRepresents the weight of the input gate; h ist-1Represents a previous time value; x is the number oftRepresenting the current input, which is the output of the previous moment; biIndicating the deviation of the input gate; ctRepresenting the current unit state; f. oftIndicating a forgetting gate; tanh represents an activation function; bcIndicating the deviation of the input gate;
which features the output gate control unit state Ct outputs to the current value ht is expressed by the following formula:
Ot=σ(Wo·[ht-1,xt]+bo) (6);
ht=Ot·tanh(Ct) (7);
in the above formulae (6) and (7), OtIs an output gate; σ represents a logic function whose output is 0 to 1; woIs the weight of the output gate; boIs the offset of the output gate; h istRepresents the current value; tanh represents the activation function.
The output of the last pooling layer is input into the LSTM layer, so that the defect of incomplete feature extraction of the CNN in a time sequence can be overcome, and the generalization capability of the model can be improved.
(4) An output layer:
finally, the output of the LSTM layer in the output layer is considered as the input of the Softmax function that can transform the probability distribution of the input neurons (0-1) in order to classify the different wear states, said Softmax function being expressed as:
in the above formula (8), aiIs the output of the ith neuron.
In the identification method, a neural network model with a specific structure formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-short time memory network and an output layer is used as the identification network model, the advantages of the convolutional neural network and the advantages of the long-short time memory network are combined, and the input signal adopts power signals from the cutting process of the cutter, and the power signals contain global characteristics and local characteristics. Therefore, local correlation of the time series signals is acquired through a convolutional network before global features are identified to integrate local features. The recognition method then enables processing of the current input based on the previous state information by means of long-term storage of the LSTM, thereby achieving the effect of processing global features. Compared with the traditional TCM which processes power signals by using a signal analysis method, the identification method can automatically extract and classify the characteristics of the original signals. Meanwhile, due to the introduction of the LSTM structure, the identification method has a good classification effect when mass data are processed, and the requirement of real-time monitoring of the wear state of the cutter can be met.
Examples
The cutter wear state real-time identification model based on deep learning is a neural network model formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-time and short-time memory network and an output layer. The mode of identifying the wear state of the cutter in real time by the identification network model is as follows, and test data come from power data generated by the vertical milling machine in the cutting process. The test data contained two different diameter cutters (8mm and 10 mm); FIG. 3 shows the instantaneous power signal corresponding to a certain wear condition of the 8mm tool, wherein only the power signal during the cutting process of the tool is selected as the input of the identification method; fig. 4 shows the selected instantaneous power signals of the wear states, which include five different wear states of 0min, 40min, 60min, 80min and 100min, which correspond to fig. 3(a) to fig. 3(e), respectively.
In the present embodiment, 1024 power points (i.e. one point value in the power signal) are taken as one sample, and 10000 samples are selected in total, wherein 70% is used for training and 30% is used for testing, and each state has 1400 training samples and 600 testing samples in five states. In order to verify the identification network model of the invention, classification precision a is introducedr:
In the above equation (9), MC represents the number of correctly predicted samples, and MT represents the total number of samples.
In this embodiment, the data corresponding to the 8mm cutter is tested, and the test result is shown in fig. 4, where the highest precision can reach 98.6%. In order to further verify the identification method of the present invention, the power data of the 10mm cutter was used for testing, and the result is shown in fig. 5, which shows that the highest classification accuracy of the identification method of the present invention can reach 100%. The average test time of the identification method is shown in table 1, and the result shows that the identification method can meet the requirement of monitoring the wear state of the cutter in real time.
TABLE 1 average test time per sample
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. A cutter wear state real-time identification method based on deep learning is characterized in that a neural network model formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-short time memory network layer and an output layer is used as an identification network model;
and taking the obtained original power signal generated by the machine tool in the working process as the input of the trained recognition network model, and carrying out real-time recognition on the tool wear state of the machine tool to obtain the real-time recognition result of the tool wear state of the machine tool.
2. The method for identifying the tool wear state based on the deep learning in real time as claimed in claim 1, wherein in the identification network model, a first convolution layer and a second convolution layer perform convolution processing on an original power signal to obtain a feature map retaining useful information, and the convolution processing is expressed by formula (1):
3. The method for identifying the tool wear state based on deep learning in real time as claimed in claim 2, wherein the first pooling layer and the second pooling layer respectively pool the output of the previous convolutional layer by a maximum pooling method, and the maximum statistic value of each pooling unit is selected to represent the characteristics, and the pooling process is represented by formula (2):
4. The method for real-time tool wear state recognition based on deep learning of claim 3, wherein the long-time and short-time memory network performs feature optimization on the output of previous convolutional layers.
5. The tool wear state real-time identification method based on deep learning of claim 4 is characterized in that the long and short time memory network layer is composed of a plurality of long and short time memory units which are connected in series, the long and short time memory units are identical in structure, and each long and short time memory unit is composed of a forgetting gate, an input gate and an output gate; wherein the content of the first and second substances,
the forgetting gate determines the features that remain from the previous moment to the current moment, expressed by equation (3):
ft=σ(Wf·[ht-1,xt]+bf) (3);
in the above formula (3), ftIndicating a forgetting gate; σ represents a logic function whose output is 0 to 1; wfA weight representing a forgetting gate; h ist-1Represents a previous time value; x is the number oftRepresenting the current input, which is the output of the previous moment; bfA deviation indicating a forgotten door;
the input gate determines the currentNetwork input xtIs the current unit state CtThe retained characteristics are expressed by the formulas (4), (5):
it=σ(Wi·[ht-1,xt]+bi) (4);
Ct=ft·Ct-1+it·tanh(Wc·[ht-1,xt]+bc) (5);
in the above formulas (4) and (5), itRepresenting an input gate; σ represents a logic function whose output is 0 to 1; wiRepresenting the weight of the input gate; h ist-1Represents a previous time value; x is the number oftRepresenting the current input, which is the output of the previous moment; biIndicating the deviation of the input gate; ctRepresenting the current unit state; f. oftIndicating a forgetting gate; tanh represents an activation function; bcIndicates the deviation of the current state cell;
the output door controls the current unit state CtTo the current value htWhich features are output, expressed by equations (6), (7):
Ot=σ(Wo·[ht-1,xt]+bo) (6);
ht=Ot·tanh(Ct) (7);
in the above formulae (6) and (7), OtIs an output gate; σ represents a logic function whose output is 0 to 1; woIs the weight of the output gate; boIs the offset of the output gate; h istRepresents the current value; tanh represents the activation function.
6. The tool wear state real-time identification method based on deep learning according to any one of claims 1 to 5, characterized in that the input of the Softmax function of the output layer is the output of the long-time memory network, the Softmax function can convert the probability distribution of input neurons 0-1 to classify different wear states, and the Softmax function is expressed by formula (8):
in the above formula (8), aiIs the output of the ith neuron.
7. The tool wear state real-time identification method based on deep learning of any one of claims 1 to 5, characterized in that the first convolution layer adopts a wide convolution kernel; the second convolutional layer employs a small convolutional kernel.
8. The method for identifying the tool wear state based on the deep learning according to any one of claims 11 to 5, wherein the training of the recognition network model is performed in the following manner, and the obtaining of the trained recognition network model comprises:
step 1, acquiring an original power signal of a machine tool in a working process as a training sample;
step 2, setting hyper-parameters of the recognition network model to be trained, and inputting the training sample into the recognition network model for training;
and 3, adjusting the hyper-parameter value of the recognition network model according to the training result of the step 2 until the training result of the recognition network model is converged.
9. A cutter wear state real-time recognition model based on deep learning is characterized by being a neural network model formed by sequentially connecting an input layer, a first convolution layer, a first pooling layer, a second convolution layer, a second pooling layer, a long-time and short-time memory network layer and an output layer.
10. The deep learning based tool wear state real-time identification model of claim 9, wherein the first convolution layer employs a wide convolution kernel; the second convolutional layer employs a small convolutional kernel.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011222866.9A CN112329625A (en) | 2020-11-05 | 2020-11-05 | Cutter wear state real-time identification method and model based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011222866.9A CN112329625A (en) | 2020-11-05 | 2020-11-05 | Cutter wear state real-time identification method and model based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112329625A true CN112329625A (en) | 2021-02-05 |
Family
ID=74315785
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011222866.9A Pending CN112329625A (en) | 2020-11-05 | 2020-11-05 | Cutter wear state real-time identification method and model based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112329625A (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200175335A1 (en) * | 2018-11-30 | 2020-06-04 | Microsoft Technology Licensing, Llc | Machine Learning Model With Depth Processing Units |
CN111325112A (en) * | 2020-01-31 | 2020-06-23 | 贵州大学 | Cutter wear state monitoring method based on depth gate control circulation unit neural network |
CN111553186A (en) * | 2020-03-05 | 2020-08-18 | 中国电子科技集团公司第二十九研究所 | Electromagnetic signal identification method based on depth long-time and short-time memory network |
-
2020
- 2020-11-05 CN CN202011222866.9A patent/CN112329625A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200175335A1 (en) * | 2018-11-30 | 2020-06-04 | Microsoft Technology Licensing, Llc | Machine Learning Model With Depth Processing Units |
CN111325112A (en) * | 2020-01-31 | 2020-06-23 | 贵州大学 | Cutter wear state monitoring method based on depth gate control circulation unit neural network |
CN111553186A (en) * | 2020-03-05 | 2020-08-18 | 中国电子科技集团公司第二十九研究所 | Electromagnetic signal identification method based on depth long-time and short-time memory network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110222701B (en) | Automatic bridge disease identification method | |
CN113378725B (en) | Multi-scale-channel attention network-based tool fault diagnosis method, equipment and storage medium | |
CN110110768B (en) | Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers | |
CN111761409A (en) | Multi-sensor numerical control machine tool cutter wear monitoring method based on deep learning | |
CN110561191B (en) | Numerical control machine tool cutter abrasion data processing method based on PCA and self-encoder | |
CN113554621B (en) | Cutter wear state identification system and method based on wavelet scale map and deep migration learning | |
CN114619292B (en) | Milling cutter wear monitoring method based on fusion of wavelet denoising and attention mechanism with GRU network | |
CN107451760B (en) | Rolling bearing fault diagnosis method based on time window slip limited Boltzmann machine | |
Mechefske et al. | Fault detection and diagnosis in low speed rolling element bearings Part II: The use of nearest neighbour classification | |
CN112904810B (en) | Process industry nonlinear process monitoring method based on effective feature selection | |
CN109977213B (en) | Optimal answer selection method for intelligent question-answering system | |
CN115953666B (en) | Substation site progress identification method based on improved Mask-RCNN | |
CN108393744B (en) | Multi-sensing monitoring method for cutter state | |
CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network | |
CN114049305A (en) | Distribution line pin defect detection method based on improved ALI and fast-RCNN | |
CN114754973A (en) | Wind tunnel force measurement test data intelligent diagnosis and analysis method based on machine learning | |
CN116070527A (en) | Milling cutter residual life prediction method based on degradation model | |
CN113627544A (en) | Machine tool milling cutter state identification method based on multi-source heterogeneous data fusion | |
CN114239810A (en) | Milling cutter wear prediction method based on improved PCANet model | |
CN112329625A (en) | Cutter wear state real-time identification method and model based on deep learning | |
CN117001423A (en) | Tool state online monitoring method based on evolutionary learning | |
CN113523904A (en) | Cutter wear detection method | |
CN115169504B (en) | Equipment abnormity identification method in coal gas fine desulfurization process | |
CN114714145B (en) | Graham angle field enhanced contrast learning monitoring method for cutter wear state | |
CN112884027B (en) | Cutting process real-time state monitoring method and device based on pattern recognition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210205 |
|
RJ01 | Rejection of invention patent application after publication |