CN111325112A - Cutter wear state monitoring method based on depth gate control circulation unit neural network - Google Patents
Cutter wear state monitoring method based on depth gate control circulation unit neural network Download PDFInfo
- Publication number
- CN111325112A CN111325112A CN202010077631.9A CN202010077631A CN111325112A CN 111325112 A CN111325112 A CN 111325112A CN 202010077631 A CN202010077631 A CN 202010077631A CN 111325112 A CN111325112 A CN 111325112A
- Authority
- CN
- China
- Prior art keywords
- time
- time sequence
- neural network
- network
- signal
- 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.)
- Granted
Links
Images
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Feedback Control In General (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a cutter wear state monitoring method based on a deep gate control circulation unit neural network, which comprises the following steps: the method comprises the steps that a vibration signal generated in the cutter machining process is collected in real time by a sensor, is input into a one-dimensional convolutional neural network for single time step time sequence signal local feature extraction after being denoised by a wavelet threshold, is input into an improved depth gate control cyclic unit neural network CABGRAs for time sequence signal time sequence feature extraction, an Attention mechanism is introduced to calculate network weights and reasonably distribute the network weights, and finally, signal feature information with different weights is put into a Softmax classifier to classify the cutter wear state, so that the complexity and the limitation caused by manual feature extraction are avoided; meanwhile, the problem of correlation before and after the single convolution neural network ignores the time sequence signal is effectively solved, and the accuracy of the model is improved by introducing an Attention mechanism. Therefore, the method has the characteristics of improving the real-time performance and accuracy of monitoring the wear state of the cutter.
Description
Technical Field
The invention belongs to the field of manufacturing process monitoring, and particularly relates to a cutter wear state monitoring method based on a deep gate control circulation unit neural network.
Background
In the machining process, cutting machining is the most important machining mode for part forming, and the wear state of a cutter directly influences the machining precision, the surface quality and the production efficiency of parts, so that the cutter condition monitoring (TCM) technology has very important significance for ensuring the machining quality and realizing continuous automatic machining. The tool state monitoring method mainly adopts an indirect measurement method at present, and can acquire signals in real time through a sensor in the cutting process of the tool, and monitor the tool abrasion loss by adopting a Machine Learning (ML) model after data processing and feature extraction.
In the prior art, zhansi et al propose a device remaining life prediction model based on transfer learning and long-short term memory network (LSTM), which is trained on different but related device remaining life prediction data sets in advance, and then network structure and training parameters are fine-tuned for a target data set. The experimental result shows that the prediction accuracy of the model can be improved by the transfer learning method on the premise of only possessing a small number of samples. The method for diagnosing the fault of the planetary gearbox based on deep learning diversity feature extraction and information fusion is provided for the golden chess and the like, a plurality of Stack Denoising Automatic Encoders (SDAE) are optimized by a multi-objective optimization algorithm, the diversity fault features are extracted at the same time, and then a multi-response linear regression model is adopted to integrate the diversity fault features to realize the information fusion. Experimental results show that the method for extracting the diversity characteristics and fusing the information can effectively improve the accuracy and the stability of fault diagnosis and has strong generalization capability. Zhangjie et al propose to convert the vibration signal of the cutter in the machining process into an energy spectrogram through Wavelet Packet Transform (WPT), then input the energy spectrogram into a convolutional neural network to automatically extract characteristics and accurately classify the characteristics, and experimental results show that the accuracy of the deep convolutional neural network is superior to that of a traditional neural network model. The Cao university theory and the like propose that a deep neural network DenseNet is constructed in a dense connection mode, tiny features hidden in a cutter processing signal are extracted from an original time sequence signal in a self-adaptive mode, and an experimental result shows that the deepening of the number of network layers is beneficial to excavating high-dimensional features hidden in the processing signal and improves the precision of a cutter wear monitoring model. The above methods all adopt a deep learning mode to extract features in a self-adaptive manner, but the used convolutional neural network is excessively dependent on high-dimensional feature extraction, gradient diffusion easily occurs when the number of convolutional layers is too large, the global situation cannot be grasped when the number of convolutional layers is too small, and the important feature of generating the correlation between time sequence signal samples during cutter processing is not considered.
Disclosure of Invention
The invention aims to overcome the defects and provides a cutter wear state monitoring method based on a depth gate control circulation unit neural network, which can improve the real-time performance and accuracy of cutter wear state monitoring.
The invention discloses a cutter wear state monitoring method based on a deep gate control circulation unit neural network, which comprises the following steps of:
the method comprises the following steps: the method comprises the following steps of collecting triaxial vibration signals generated during cutter milling by using an acceleration sensor, carrying out denoising processing on the collected original signals by using a wavelet threshold denoising method, and cutting the vibration signals generated by cutter feeding each time into a plurality of short sequence time sequence signals with the length of 2000, wherein the method comprises the following specific steps: and intercepting continuous 100000 points in each sampling signal, and dividing the intercepted points into 50 samples by taking 2000 as the number of the samples, wherein the 50 samples correspond to the same wear state label.
Step two: local feature extraction of single time step time sequence signals: processing the short sequence time sequence signal generated in the cutter processing process by adopting a one-dimensional Convolutional neural network, wherein the Convolutional neural network part comprises 2 layers of Convolutional layers (CONV) and 1 Layer of Pooling layers (POOL), the Convolutional layers carry out neighborhood filtering on the time sequence signal of each dimension in a one-dimensional convolution operation mode to generate feature mapping, and each feature map can be regarded as convolution operation of different filters on the time sequence signal of the current time step; the calculation formula of the one-dimensional convolutional layer operation is as follows:
wherein:j-th feature map representing the l-th layer, f representing the activation function, M representing the number of input feature maps,the ith feature map representing the l-1 layer,a trainable convolution kernel is represented that is capable of being trained,representing a bias parameter; the activation function adopts a Relu activation function;
the pooling layer adopts maximum pooling to take maximum value for the feature points in the neighborhood, and the formula is as follows:
wherein:the value of the t-th neuron in the i-th feature vector of the l-th layer, t ∈ [ (j-1) w +1, jw](ii) a w is the width of the pooling region; pi l+1(j) Indicating the corresponding value for layer l +1 neurons.
Step three: in order to dig the time sequence change rule of relatively longer intervals in a time sequence, an improved gate control circulation unit is adopted to extract the time sequence characteristics of the time sequence signals and learn the dependency relationship of the time sequence characteristics among the time sequence signals;
the improved gate control cycle unit forms a CABGRUs network by constructing two bidirectional BiGRU networks which are overlapped together, and simultaneously introduces an Attention mechanism in the CABGRUs network and increases an Attention layer, so that the model not only obtains the capability of simultaneously extracting time sequence signal characteristics from the forward direction and the reverse direction, but also obtains the capability of selectively learning key information in the signal characteristics;
each bidirectional BiGRU network in the improved depth-gated cyclic unit neural networks CABGRUs comprises 256 neurons, each of the forward and reverse BiGRU networks is composed of 128 neurons, each of the BiGRU neurons comprises an update gate and a reset gate, and z is used respectivelytAnd rtTo indicate the manner in which, among others,representing candidate hidden states at time step t, htRepresenting hidden states at time step t, xtRepresenting the input vector at time step t. Updating the door ztFor controlling how much state information, z, the current state is updatedtThe closer to 1, the more information representing the current state is utilized for the previous time instance. Reset gate rtFor controlling which state information is removed from the previous state, rtThe closer to 0, the smaller the proportion of the output state from the previous time. The formula is as follows:
rt=σ(xtWxr+ht-1Whr+br)
zt=σ(xtWxz+ht-1Whz+bz)
wherein: wxrAnd WhrWeight vector, W, representing reset gatexzAnd WhzWeight vector, W, representing the update gatexhAnd WhhWeight vector representing candidate hidden states, br、bz、bhRepresenting the bias vector, □ represents the Hadamard product, i.e. the dot product of the matrix, σ (·) represents the Sigmod function, and tanh represents the hyperbolic tangent activation function.
Input timingHigh dimensional characteristics of the signal, outputting the hidden state via a forward BiGRU networkOutputting the hidden state to the BiGRU networkCABGRUs network outputting hidden state P at time step tt。
The formula is as follows:
step four: introducing an Attention mechanism to calculate the importance distribution of continuous time step time sequence signal characteristics, wherein the introduced Attention mechanism performs weighted summation with each time step output vector of a BiGRU layer by distributing different initialization probability weights, and finally obtains a numerical value by calculating a Sigmod function; the equation for the Attention mechanism is as follows:
ut=tanh(WsPt+bs)
ν=∑αtPt
wherein, PtRepresenting the output eigenvector, u, of the BiGRU layer at time step ttRepresents PtHidden layer representation, u, obtained by a neural network layersContext vector representing random initialization, αtRepresents utThe importance weight obtained by the Softmax function normalization, v, represents the feature vector of the final text information, i.e., usRandomly generating in the training process, and finally mapping the Attention layer output value v through a Softmax function to obtain a real-time classification result of the tool wear state.
Step five: training of a network model: dropout technology is introduced to prevent the model from being over-fitted during the training process; and (3) performing wear classification on the time sequence signal characteristics obtained in the steps by adopting Softmax as an activation function and adopting Categorical _ cross control as a loss function of the network model to obtain a classification result.
The input data is a time sequence signal, the characteristic extraction and expression of the time sequence signal are realized by convolution layers (C1 and C1), a Dropout layer, a pooling layer (P1), a Flatten layer, BiGRU (B1 and B1) layers, an Attention layer A1 and a full-link layer (F1 and F1), wherein the time sequence signal with the size of (2000,3) is input to a deep learning neural network as input data, convolution layer C1 is convolved with the convolution check time sequence signal of 3 1 1, the convolution kernel step size is 1, a characteristic diagram (20,98,128) is generated, the time sequence signal is input to convolution layer C1 by convolution layer C1, convolution layer C1 is convolved with the convolution check time sequence signal of 3 1, the convolution kernel is 1, a characteristic diagram (20,96,128) is generated, the convolution kernel is input to the pooling layer P1 by Dropout layer C1, the Dropout layer is 0.5, the maximum value is generated, the convolution kernel is input to BiGRU 1, the BiorP 72, the BiorP 1, the Biopten layer B1 is input to the Biopten layer 1, the Biopten layer B1, the Bioptel 1, the Biopten layer B1, the Biopten layer 1, the Bioptel 1, the Biopten layer 1 is input to generate a1, the Biopten layer 1, the Bioptel 1, the Biopten layer 1.
Compared with the prior art, the method has obvious beneficial effects, and the scheme shows that the existing gated cyclic unit (GRU) neural network is improved, the CABGRUs network is formed by constructing two bidirectional BiGRU networks which are overlapped together, and meanwhile, an Attention mechanism is introduced into the CABGRUs network, and an Attention layer is added, so that the model not only has the capability of simultaneously extracting the time sequence signal characteristics from the forward direction and the reverse direction, but also has the capability of selectively learning key information in the signal characteristics. In the CABG networks, each bidirectional GRU network comprises 256 neurons, each forward GRU network and each reverse GRU network are composed of 128 neurons, and each GRU neuron comprises an update gate and a reset gate. The tool wear state real-time monitoring model can better learn the dependency relationship of time series characteristics among time sequence signals, and the accuracy of model classification is improved. In addition, an Attention mechanism is introduced, wherein the Attention mechanism is a brain signal processing mechanism similar to that specific to human vision, different initialization probability weights are distributed to carry out weighted summation with output vectors of each time step of a BiGRU layer, and then the weighted summation is carried out to be brought into a Sigmod function, and final calculation is carried out to obtain a numerical value. The method and the device realize selective filtering of partial key information from a large number of signal characteristics and focusing, the focusing process is embodied in calculation of weight coefficients, different weights are distributed to different key information, the proportion of the key information is strengthened in a weight improving mode, and loss of the key information of long-sequence time sequence signals is reduced.
In a word, the vibration signals generated in the cutter processing process are collected in real time by using a sensor, are input into a one-dimensional convolutional neural network for single time step local characteristic extraction after being denoised by a wavelet threshold, then are input into improved depth gate control cycle unit neural networks CABGRAUs for time sequence signal time sequence characteristic extraction, an Attention mechanism is introduced to calculate network weights and reasonably distribute the network weights, and finally, signal characteristic information with different weights is put into a Softmax classifier to classify the cutter wear state, so that the complexity and the limitation caused by manual characteristic extraction are avoided; meanwhile, the problem of correlation before and after the single convolution neural network ignores the time sequence signal is effectively solved, the problems of gradient dispersion and gradient explosion of the circulating neural network are avoided, and the accuracy of the model is improved by introducing an Attention mechanism. Therefore, the method has the characteristics of improving the real-time performance and accuracy of monitoring the wear state of the cutter.
The advantageous effects of the present invention will be further described below by way of specific embodiments.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a schematic diagram of a CABGRUs network structure according to the present invention;
FIG. 3 is a CNN model training and validation diagram in an embodiment;
FIG. 4 is a BiGRU model training and validation diagram in an embodiment;
FIG. 5 is a CBLSTMs model training and validation graph in an example embodiment;
FIG. 6 is a diagram of CABGRUs model training and validation in an example embodiment.
Detailed Description
The following detailed description will be made with reference to the accompanying drawings and preferred embodiments of a method for monitoring a wear state of a tool based on a neural network of a depth-gated cyclic unit according to the present invention.
Referring to fig. 1, the method for monitoring the wear state of a tool based on a neural network of a depth gated cyclic unit of the present invention includes the following steps:
firstly, acquiring vibration signals generated by numerical control machining equipment in the process of machining a workpiece in real time by using an acceleration sensor, and taking the signals subjected to wavelet threshold denoising as input signals of a tool wear state real-time monitoring model, wherein the input signals comprise αx、αy、αzAnd the original vibration signal is a tensor composed of 2000 sampling points (2000,3) obtained by continuously sampling and cutting the vibration signals in the x, y and z directions into the vibration signals, and the tensor is taken as input data of the model and is brought into the model.
Step two: local feature extraction is carried out on a single time step time sequence signal: inputting the time sequence signal with the size of (2000,3) as input data into a one-dimensional Convolutional Neural Network (CNN) for neighborhood filtering, and calculating by using a sliding window to finally obtain the high-dimensional characteristic of the single time step time sequence signal;
directly processing time sequence signals generated in the machining process of a cutter by adopting a one-dimensional Convolutional neural network, wherein the Convolutional neural network part comprises 2 Convolutional layers (CONV) and 1 Pooling Layer (POOL), and the Convolutional layers are subjected to one-dimensional convolution operationEach feature map may be viewed as a convolution operation of a different filter on the current time step timing signal. When the input timing signal is xtWhen the filter is wtFeature map y of convolutional layertCan be expressed as:
in the convolutional layer, each neuron of the l layer is connected with a neuron in a local window of the l-1 layer to form a local connection network. The calculation formula of the one-dimensional convolutional layer is as follows:
wherein:j-th feature map representing the l-th layer, f representing the activation function, M representing the number of input feature maps,the ith feature map representing the l-1 layer,a trainable convolution kernel is represented that is capable of being trained,representing the bias parameter. In consideration of the convergence speed and the over-fitting problem, the nonlinear activation function of the invention selects a corrected Linear unit (Relu) with a higher convergence speed to improve the sparsity of the network, reduce the interdependence relationship of parameters and relieve the over-fitting phenomenon. The formula for the Relu activation function is:
The convolutional layer is followed by a Pooling layer for local maxima or local means, i.e. maximum Pooling (MaxPooling) and Mean Pooling (Mean Pooling). The pooling layer has a function similar to feature selection, and can reduce feature dimension, accelerate network training speed, reduce parameter quantity and improve feature robustness while ensuring that the features have anti-deformation capability. The invention selects maximum value pooling to take maximum value for the feature points in the neighborhood, and the formula is as follows:
wherein:the value of the t-th neuron in the i-th feature vector of the l-th layer, t ∈ [ (j-1) w +1, jw](ii) a w is the width of the pooling region; pi l+1(j) Indicating the corresponding value for layer l +1 neurons.
The method has the advantages that the characteristics of the original data are extracted through the one-dimensional convolution neural network, 3-dimensional characteristics of the time sequence signals are well expressed into high-dimensional characteristics, and the time sequence characteristics of the subsequent network can be conveniently extracted.
Step three: and (3) performing feature extraction on the time sequence of the time sequence signal: adopting an improved gate control cycle unit to process high-dimensional characteristics generated by continuous time step time sequence signals, and gradually synthesizing vector characteristic representation of input signals;
the improved gated circulation unit forms a CABGRUs network by constructing two bidirectional gated circulation unit BiGRU networks which are overlapped together, and simultaneously introduces an Attention mechanism in the CABGRUs network to increase an Attention layer, so that the model not only can obtain the capability of simultaneously extracting time sequence signal characteristics from the forward direction and the reverse direction, but also can obtain the capability of selectively learning key information in the signal characteristics;
the original signals generated in the machining process of the cutter have a time sequence relation, the RNN can encode time sequence signal time sequences, and a relatively long-interval time sequence change rule in the time sequences is mined. In order to enable the cutter wear state real-time monitoring model to better learn the dependency relationship of time series characteristics among time sequence signals, the accuracy of model classification is improved. The invention improves the existing gated cycle units (GRUs), and constructs a CABGRUs network by jointly superposing two bidirectional gated cycle unit BiGRU networks, and simultaneously introduces an Attention mechanism into the CABGRUs network to increase an Attention layer, so that the model not only obtains the capability of simultaneously extracting time sequence signal characteristics from the forward direction and the reverse direction, but also obtains the capability of selectively learning key information in the signal characteristics.
Each bidirectional BiGRU network in the deep-gated cyclic unit neural networks CABGRUs constructed by the invention comprises 256 neurons, the forward and reverse BiGRU networks are respectively composed of 128 neurons, each BiGRU neuron comprises an update gate and a reset gate, and z is respectively usedtAnd rtTo indicate.
Wherein the content of the first and second substances,representing candidate hidden states at time step t, htRepresenting hidden states at time step t, xtRepresenting the input vector at time step t. Updating the door ztFor controlling how much state information, z, the current state is updatedtThe closer to 1, the more information representing the current state is utilized for the previous time instance. Reset gate rtFor controlling which state information is removed from the previous state, rtThe closer to 0, the smaller the proportion of the output state from the previous time. The formula is as follows:
rt=σ(xtWxr+ht-1Whr+br)
zt=σ(xtWxz+ht-1Whz+bz)
wherein: wxrAnd WhrWeight vector, W, representing reset gatexzAnd WhzWeight vector, W, representing the update gatexhAnd WhhWeight vector representing candidate hidden states, br、bz、bhRepresenting the bias vector, □ represents the Hadamard product, i.e. the dot product of the matrix, σ (·) represents the Sigmod function, and tanh represents the hyperbolic tangent activation function.
Inputting the high dimensional characteristics of the timing signal, outputting the hidden state via the forward BiGRU networkOutputting the hidden state to the BiGRU networkCABGRUs network outputting hidden state P at time step tt。
The formula is as follows:
step four: introduction of the Attention mechanism: calculating the importance distribution of the time sequence signal characteristics of continuous time steps by using an Attention mechanism, and generating a time sequence signal characteristic model containing Attention probability distribution; the introduced Attenttion mechanism carries out weighted summation with output vectors of each time step of the BiGRU layer by distributing different initialization probability weights, and finally, a numerical value is obtained by calculation of a Sigmod function.
The Attenttion mechanism introduced by the invention performs weighted summation with each time step output vector of the BiGRU layer by distributing different initialization probability weights, and finally obtains a numerical value through the calculation of a Sigmod function. The method and the device realize selective filtering of partial key information from a large number of signal characteristics and focusing, the focusing process is embodied in calculation of weight coefficients, different weights are distributed to different key information, the proportion of the key information is strengthened in a weight improving mode, and loss of the key information of long-sequence time sequence signals is reduced. The equation for the Attention mechanism is as follows:
ut=tanh(WsPt+bs)
ν=∑αtPt
wherein, PtRepresenting the output eigenvector, u, of the BiGRU layer at time step ttRepresents PtHidden layer representation, u, obtained by a neural network layersContext vector representing random initialization, αtRepresents utThe importance weight obtained by the Softmax function normalization, v, represents the feature vector of the final text information. u. ofsRandomly generating in the training process, and finally mapping the Attention layer output value v through a Softmax function to obtain a real-time classification result of the tool wear state.
Step five: training of a network model: dropout technology is introduced to prevent the model from being over-fitted during the training process; and (3) performing wear classification on the time sequence signal characteristics obtained in the steps by adopting Softmax as an activation function and adopting Categorical _ cross control as a loss function of the network model to obtain a classification result.
A Dropout technology is introduced into the cutter wear state real-time monitoring model to prevent the model from being over-fitted in the training process. And (3) carrying out wear classification on the obtained time sequence signal characteristics by using Softmax as an activation function and using Categorical _ cross control as a loss function of the network model.
The formula is as follows:
y is a vector with one dimension being the size of the number of categories, the value of each dimension is between [0,1], the sum of all the dimensions is 1, the value represents the probability that the tool wear state belongs to a certain category, and M is the possible number of categories. During the training of the model, the entire model was trained by the category _ cross Loss. The cross entropy error calculation formula is as follows:
…
wherein: m represents the number of classifications, n represents the number of samples,i-th value, y, in the true category label vector representing the wear state of the toolimRepresenting the ith value of the output vector y of the Softmax classifier. For the cross entropy error obtained, the average is finally taken as a modelA loss function. The Adam method, which is essentially RMSprop with momentum terms, dynamically adjusts the learning rate of each parameter using first and second moment estimates of the gradient, is used to minimize the objective function when training the model. Adam has the advantages that after offset correction, the learning rate of each iteration has a certain range, so that the parameter change is relatively stable.
Such as the schematic diagram of the network structure of the cabarus shown in fig. 2. The input data based on the CABGRUs model neural network comprises time sequence signals, and the feature extraction and expression of the time sequence signals are realized through convolution layers (C1 and C2), a Dropout layer, a pooling layer (P1), a Flatten layer, BiGRU layers (B1 and B2), an Attention layer A1 (Attention mechanism) and full-connection layers (F1 and F2).
Inputting a time sequence signal with the size of (2000,3) as input data into a deep learning neural network, convolving layer C1 with a convolution kernel of 3 × to convolve the time sequence signal, the convolution kernel step size is 1, generating a (20,98,128) feature map, inputting the (20,98,128) feature map from convolving layer C1 to convolving layer C2, convolving layer C2 with a convolution kernel step size of 3 × to convolve the time sequence signal, the convolution kernel step size is 1, generating a (20,96,128) feature map, inputting the (20,96,128) feature map from convolving layer C2 to pooling layer P1 via Dropout, and 0.5, generating a (20,48,128) feature map by maximum pooling layer P1 to inputting the (20,6144) feature map into BiGRU layer B1, generating a (20,256) feature map, inputting the BiGRU layer B1 into BiGRU layer L1, generating a (BiGRU layer B2) feature map, inputting the BiGRU layer L1 feature map via BiGRU layer B599, generating a full-linking layer A867, connecting the Bigrunt layer B863 to generate a knife switch insert 3653, and connecting the BiGRU insert 3653, generating a full-state map, and connecting 3628, and connecting the wear-state map, and connecting the BigrU insert 3653.
The examples are as follows:
1 design of the experiment
(1) Condition monitoring
In the experiment, a high-precision numerical control vertical milling machine (model: VM600) is used for milling a workpiece, cooling liquid is not added in the milling process, the milled workpiece is die steel (S136), a milling cutter is an ultrafine particle tungsten steel hard alloy four-edge milling cutter, and the surface of a cutting edge is covered with a TiAIN pattern layer. Table 1 shows the milling experimental cutting parameters.
TABLE 1 milling test cutting parameters
In the experiment, three acceleration sensors (model: INV9822) are magnetically adsorbed on a machine tool clamp in the directions of x, y and z and are used for acquiring original vibration signals generated in the machining process of the cutter in real time; the real-time signal is processed by a high-precision digital acquisition instrument (model: INV3018CT) of Beijing eastern vibration and noise institute and transmitted to a computer. The sampling frequency of the signal is 20KHz, 200mm is milled along the x direction in each feed, the milling stroke is recorded as one milling stroke, 330 strokes are milled in each cutter, and after each milling stroke is finished, the wear value of the rear cutter face of each milling edge of the milling cutter is measured by adopting a pre-calibrated high-precision digital microscope.
(2) Data analysis
The deep learning hardware platform of the experiment adopts a high-performance server: an Intel Xeon E5-2650 processor, a main frequency 2.3GHz 256GB memory and a GPU (graphics processing Unit) which is an NVIDIA GeForce TITAN X graphics processor. The software platform uses an Ubuntu16.04.4 operating system, and the deep learning framework selects Keras as the front end and TensorFlow as the rear end for data analysis.
In the experiment, 4 milling cutters (C1, C2, C3 and C4) are used for completing milling operation, milling is carried out 1320 times to obtain 1320 original signal samples, 3 data of the milling cutters (C1, C2 and C3) are used for training and verifying a model, 1 data of the milling cutters (C4) are used for testing the model, 80% of 990 samples are randomly selected to serve as a training set, and 20% of the 990 samples serve as a verifying set. A sufficient number of samples are needed in the deep learning training process to improve the learning quality of the neural network. The data expansion can increase experimental data on the basis of the original magnitude data, and the robustness is improved. According to the signal sampling principle, each sample is continuously sampled and cut into a plurality of short sequence time sequence signals with equal length, and the short sequence time sequence signals are used for inputting the model after data normalization.
Each sample comprises a three-dimensional signal and wear values of four flank surfaces, and in order to prevent the mutual interference of the wear values of different edges, the maximum value of the four edges is selected as the wear value of the milling tool. The wear state of the cutter is divided into: initial wear, normal wear, and rapid wear. The method defines the wear state of the cutter according to the actual wear curve of each milling cutter, is used for determining the wear degree of the cutter, divides the wear degree of the cutter into 3 types of label data, and converts the label data in a one-hot coding mode, so that the wear state of the cutter can be classified finally.
2 deep learning contrast experiment results
TABLE 2 model specific training parameters
In the experiment, an original signal generated in the milling cutter machining process is sampled, cut and input into a CABGRUs neural network model, the model adaptively extracts high-dimensional characteristics implicit in a time sequence signal, the error distance between the actual output value and the true value of the model is calculated, the Loss is reduced by adopting an Adam algorithm, the network weight is continuously updated, and the actual output value of the model approaches to the true value more. The invention uses CNN, BiGRU and CBLSTMs deep learning neural networks to compare with the CABGRUs model provided by the invention, and 4 models set the same training parameters in the training process. Table 2 is a model specific training parameter table.
The method comprises the steps of obtaining different loss function values and accuracy rates after deep learning neural network training and verification, and respectively obtaining the loss function values and the accuracy rates of a verification set by a training set and the verification set output by CNN models, BiGRU models, CBLSTMs models and CABGRUs models by using a graph shown in figure 3, a graph shown in figure 4, a graph shown in figure 5 and a graph shown in figure 6, wherein the x axis represents the iteration times of a milling cutter data set, and the double y axes represent the loss function values and the model verification accuracy rates respectively. The loss function value of the network model training set is reduced along with the increase of the iteration times and finally tends to be stable, the loss function value of the verification set is in periodic fluctuation, the oscillation amplitude of the loss function of the CNN and BiGRU network models is large, the CBLSTMs and CABGRUs network models are relatively stable, the overall trend of the loss function is reduced continuously and converged finally, the gradient explosion or dispersion phenomenon does not occur, and the network convergence speed is high. Accuracy rates of verification sets of the CNN and the BiGRU network models are 89.75% and 88.02% respectively, and prediction accuracy is low, which indicates that although a single deep learning network can predict the wear state of the tool, the single deep learning network cannot capture deeper features hidden in the tool vibration signal due to the limitation of the network model capability. The CABG networks provided by the method are superior to CNN and BiGRU network models, because the network structure is relatively deep, the method is favorable for excavating deeper features, firstly, the CNN network can effectively extract hidden local features in time sequence signals, meanwhile, the length of the time sequence signal features is compressed, the dependency relationship of time sequence features among the time sequence signals can be conveniently learned by a subsequent network, and the model prediction capability is improved. Compared with the depth CBLSTMs network model, the CABGRUs network model provided by the method obtains higher prediction precision. CBLSTMs constructs a double-layer BilSTM network, and accesses past and future information by utilizing a bidirectional LSTM network, namely, time sequence signal characteristics can be simultaneously extracted from the forward direction and the reverse direction, richer information characteristics are mined, after 22 iterations, the accuracy of a verification set is basically stabilized at more than 96%, and after 50 iterations, the accuracy is 96.75%. CABGRUs improves the internal structure of neurons on the basis of CBLSTMs, introduces an Attention mechanism, realizes that part of key information is selectively filtered from a large amount of information and is focused, reduces the loss of key information characteristics of long-sequence texts, and after 20 iterations, the accuracy of a verification set is basically stabilized at more than 96%, and after 50 iterations, the accuracy is 98.02%, the loss function value reaches 0.0595, and the network stability is higher. Table 3, loss function and accuracy for model validation.
TABLE 3 loss function and accuracy of model validation
The data of a milling cutter (C4) is selected to be used for a test set of the network model, the total number of test samples is 330, wherein the number of initial wear samples is 23, the number of normal wear samples is 232, and the number of rapid wear samples is 75, and the samples are randomly brought into the trained CABGUs network model. According to the test result, the CABG network model provided by the invention has strong generalization capability, and although the test time is not as long as part of comparison models, the algorithm finds a good balance point between time and precision. Table 4 shows the test time for a single sample and the accuracy of the model test set.
TABLE 4 test time for individual samples and accuracy of model test set
3 deep learning versus machine learning
TABLE 5 accuracy of machine learning and deep learning predictions
In order to further verify the feasibility of the algorithm provided by the invention, the experimental data acquired by the method is used in a tool wear state monitoring model of a BP neural network (BPNN), a Support Vector Machine (SVM), a Hidden Markov Model (HMM) and a Fuzzy Neural Network (FNN), wavelet threshold denoising processing is carried out on an original signal acquired by an acceleration sensor, and after feature extraction and feature screening, the features with less noise interference and large tool wear correlation degree are obtained. The feature extraction includes time domain features, frequency domain features, and time-frequency domain features. And reflecting the correlation degree between the characteristics and the abrasion loss by adopting a Pearson correlation coefficient method, selecting the characteristics with the correlation coefficient larger than 0.9 as an extraction object, realizing the dimension reduction of the characteristics, and taking the extracted characteristics as the input of a machine learning model. And comparing the test result with the test result of the CABGRUs network model provided by the invention. As can be seen from the table, the accuracy of the conventional machine is very different, because the instability of the manually extracted features and the construction of the model affect the prediction result. The prediction accuracy of deep learning is obviously higher than that of machine learning BPNN, SVM and HMM, however, the prediction accuracy of machine learning FNN reaches 94.24%, because the FNN learns the rules of the fuzzy system by using a neural network, the design parameters of the fuzzy system are automatically designed and adjusted according to input and output learning samples, and the self-learning and self-adaptive functions of the fuzzy system are realized. Compared with other algorithm models, the method has the advantage that the performance is greatly improved. The test sample speed of the CABGRUs model can reach 8ms, and the requirement of monitoring the wear state of the cutter in real time during industrial actual production is met. Table 5 shows the accuracy of the predictions for machine learning and deep learning.
In a word, the CNN and RNN fused machine learning method is applied to a cutter wear state real-time monitoring task, network parameters and a structure are modified according to the characteristics of high-frequency vibration signal noise and sample redundancy, and the CNN and RNN fused machine learning method is used for monitoring the cutter wear degree in real time. Performing wavelet threshold denoising on a time sequence signal acquired by an acceleration sensor in a preprocessing stage, dividing a redundant signal generated by cutter feeding each time into a plurality of training samples in a data expansion mode, and adding experimental data on the basis of original magnitude data to filter noise and improve the robustness of an algorithm; the wear state of the cutter is defined according to the actual wear curve, the wear degree of the cutter is determined, and the accuracy of data label classification is improved; the method adopts the one-dimensional convolutional neural network to extract local features, and excavates rich high-dimensional features from the de-noised signal, thereby better representing the cutter wear state information hidden in the original signal and shortening the training time of a network model; the idea of the Attention mechanism is innovatively introduced into the improved CABGRUs network model, so that the identification precision and the generalization performance of the real-time monitoring of the network model are effectively improved. The feasibility of the method is verified by using a real-time tool wear state monitoring system. A signal acquisition unit and an upper computer analysis unit are set up in an experiment, and a depth learning framework is adopted to predict the wear state of the cutter in real time. The experimental results show that: the prediction accuracy of the CABGRUs network model is 97.58%, the CABGRUs network model is superior to the traditional machine learning algorithm, and meanwhile, the CABGRUs network model can adapt to hardware systems in most production environments and can meet industrial requirements on identification accuracy and identification speed.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, and any simple modification, equivalent change and modification made to the above embodiment according to the technical spirit of the present invention are within the scope of the present invention without departing from the technical spirit of the present invention.
Claims (3)
1. A cutter wear state monitoring method based on a depth gate control circulation unit neural network comprises the following steps:
the method comprises the following steps: collecting vibration signals generated by a cutter by using a sensor, carrying out denoising treatment on the collected original vibration signals by using a wavelet threshold denoising method, and cutting the vibration signals generated by cutter feeding each time into short sequence time sequence signals with the length of 2000 points;
step two: local feature extraction of single time step time sequence signals: processing the short sequence time sequence signal by adopting a one-dimensional convolutional neural network, wherein the convolutional neural network part comprises 2 convolutional layers CONV and 1 pooling layer POOL, the convolutional layers perform neighborhood filtering on the time sequence signal of each dimension in a one-dimensional convolutional operation mode to generate feature mapping, each feature map can be regarded as convolution operation of different filters on the current time step time sequence signal, namely, the signal self-adaptive feature extraction is performed through the one-dimensional convolutional neural network, the input parameters of a subsequent network are reduced, the calculation speed is improved, meanwhile, the feature map is reduced to a certain extent in the vector dimension, the features of a vibration signal are highlighted, and the subsequent neural network can perform time sequence feature extraction conveniently;
step three: time series signal time series feature extraction: in order to dig the time sequence change rule of relatively longer intervals in a time sequence, extracting time sequence signal time sequence characteristics by adopting an improved depth gate control cycle unit neural network CABGRUS and learning the dependency relationship of the time sequence characteristics among the time sequence signals;
the improved depth gating circulating unit neural network CABGRUs is formed by constructing two depth bidirectional gating circulating unit BiGRU networks which are overlapped together, and meanwhile, an Attention mechanism is introduced into the CABGRUs network to increase an Attention layer, so that the model not only obtains the capability of simultaneously extracting time sequence signal characteristics from the forward direction and the reverse direction, but also obtains the capability of selectively learning key information in the signal characteristics;
each bidirectional BiGRU network in the improved depth-gated cyclic unit neural networks CABGRUs comprises 256 neurons, each of the forward and reverse BiGRU networks consists of 128 neurons, each of the BiGRU neurons comprises an update gate and a reset gate, and the two neurons are used respectivelyAndto indicate the manner in which, among others,is shown at time stepThe candidate hidden state of the time-varying,is shown at time stepIn the hidden state of the time,is shown at time stepInput vector of time, update gateFor controlling how much state information the current state is updated,the closer to 1, the more information representing the current state is utilized for the previous time, the reset gateFor controlling which state information is removed from the previous state,the closer to 0, the smaller the proportion of the output state from the previous time, the formula is as follows:
wherein:anda weight vector representing the reset gate,anda weight vector representing the updated gate,anda weight vector representing a candidate hidden state,a vector of the offset is represented, and,representing the Hadamard product, i.e. the dot product of the matrix,represents a Sigmod function, and the tanh function represents a hyperbolic tangent activation function;
step four: an Attention mechanism is introduced to calculate an importance distribution numerical value of continuous time step time sequence signal characteristics, the introduced Attention mechanism carries out weighted summation by distributing different initialization probability weights and each time step output vector of a BiGRU layer of a depth bidirectional gating circulating unit, and finally a numerical value is obtained by calculating a Sigmod function, so that the proportion of key information is enhanced in a weight increasing mode, and the loss of the key information of a long sequence time sequence signal is reduced;
step five: training of a network model: dropout technology is introduced to prevent the model from being over-fitted during the training process; and (3) performing wear classification on the time sequence signal characteristics obtained in the steps by adopting Softmax as an activation function and adopting category _ cross control as a loss function of the network model to obtain a classification result, and confirming the wear state of the tool at the current moment.
2. The tool wear state monitoring method based on the deep gated cyclic unit neural network as claimed in claim 1, wherein: the method is characterized in that a vibration signal generated by cutter feeding each time is cut into a short sequence time sequence signal with the length of 2000, and the method specifically comprises the following steps: and intercepting continuous 100000 points in each sampling signal, and dividing the intercepted points into 50 samples by taking 2000 as the number of the samples, wherein the 50 samples correspond to the same wear state label.
3. The tool wear state monitoring method based on the deep gated cyclic unit neural network as claimed in claim 1, wherein: the equation for the Attention mechanism in step four is as follows:
wherein the content of the first and second substances,represents the output feature vector of the BiGRU layer at the time step,to representA hidden layer representation obtained by the neural network layer,a context vector representing a random initialization is shown,to representThe importance weights obtained by the Softmax function normalization,feature vectors representing the final text information, i.e.Randomly generating in the training process, and finally outputting the Attention layer output value through a Softmax functionAnd mapping to obtain a real-time classification result of the wear state of the cutter.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010077631.9A CN111325112B (en) | 2020-01-31 | 2020-01-31 | Cutter wear state monitoring method based on depth gate control circulation unit neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010077631.9A CN111325112B (en) | 2020-01-31 | 2020-01-31 | Cutter wear state monitoring method based on depth gate control circulation unit neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111325112A true CN111325112A (en) | 2020-06-23 |
CN111325112B CN111325112B (en) | 2023-04-07 |
Family
ID=71167049
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010077631.9A Active CN111325112B (en) | 2020-01-31 | 2020-01-31 | Cutter wear state monitoring method based on depth gate control circulation unit neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111325112B (en) |
Cited By (36)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111814728A (en) * | 2020-07-22 | 2020-10-23 | 同济大学 | Method for recognizing wear state of cutting tool of numerical control machine tool and storage medium |
CN111879397A (en) * | 2020-09-01 | 2020-11-03 | 国网河北省电力有限公司检修分公司 | Fault diagnosis method for energy storage mechanism of high-voltage circuit breaker |
CN112001527A (en) * | 2020-07-29 | 2020-11-27 | 中国计量大学 | Industrial production process target data prediction method of multi-feature fusion deep neural network |
CN112033463A (en) * | 2020-09-02 | 2020-12-04 | 哈尔滨工程大学 | Nuclear power equipment state evaluation and prediction integrated method and system |
CN112070208A (en) * | 2020-08-05 | 2020-12-11 | 同济大学 | Tool wear prediction method based on encoder-decoder stage attention mechanism |
CN112115550A (en) * | 2020-09-13 | 2020-12-22 | 西北工业大学 | Aircraft maneuvering trajectory prediction method based on Mogrifier-BiGRU |
CN112149355A (en) * | 2020-09-27 | 2020-12-29 | 浙江科技学院 | Soft measurement method based on semi-supervised dynamic feedback stack noise reduction self-encoder model |
CN112200032A (en) * | 2020-09-28 | 2021-01-08 | 辽宁石油化工大学 | Attention mechanism-based high-voltage circuit breaker mechanical state online monitoring method |
CN112329625A (en) * | 2020-11-05 | 2021-02-05 | 中国科学技术大学 | Cutter wear state real-time identification method and model based on deep learning |
CN112629854A (en) * | 2020-11-25 | 2021-04-09 | 西安交通大学 | Bearing fault classification method based on neural network attention mechanism |
CN112668507A (en) * | 2020-12-31 | 2021-04-16 | 南京信息工程大学 | Sea clutter prediction method and system based on hybrid neural network and attention mechanism |
CN112712063A (en) * | 2021-01-18 | 2021-04-27 | 贵州大学 | Tool wear value monitoring method, electronic device and storage medium |
CN112785092A (en) * | 2021-03-09 | 2021-05-11 | 中铁电气化局集团有限公司 | Turnout residual life prediction method based on self-adaptive deep layer feature extraction |
CN112798956A (en) * | 2020-12-31 | 2021-05-14 | 江苏国科智能电气有限公司 | Wind turbine generator fault diagnosis method based on multi-resolution time sequence cyclic neural network |
CN113051689A (en) * | 2021-04-25 | 2021-06-29 | 石家庄铁道大学 | Bearing residual service life prediction method based on convolution gating circulation network |
CN113177584A (en) * | 2021-04-19 | 2021-07-27 | 合肥工业大学 | Zero sample learning-based composite fault diagnosis method |
CN113204640A (en) * | 2021-04-02 | 2021-08-03 | 南京邮电大学 | Text classification method based on attention mechanism |
CN113286309A (en) * | 2021-05-18 | 2021-08-20 | 合肥工业大学 | Heterogeneous communication method and system based on CSI |
CN113305645A (en) * | 2021-06-22 | 2021-08-27 | 重庆邮电大学工业互联网研究院 | Numerical control machine tool cutter residual life prediction method based on mixed neural model |
CN113359577A (en) * | 2021-07-02 | 2021-09-07 | 中国科学院空间应用工程与技术中心 | Embedded state monitoring and fault diagnosis system and method for ultrasonic motor |
CN113414638A (en) * | 2021-06-04 | 2021-09-21 | 西北工业大学 | Variable working condition milling cutter wear state prediction method based on milling force time sequence diagram deep learning |
CN113537472A (en) * | 2021-07-26 | 2021-10-22 | 北京计算机技术及应用研究所 | Bidirectional recurrent neural network with low computation and storage consumption |
CN113657454A (en) * | 2021-07-23 | 2021-11-16 | 杭州安脉盛智能技术有限公司 | Autoregressive BiGRU-based nuclear power rotating machine state monitoring method |
CN113664612A (en) * | 2021-08-24 | 2021-11-19 | 沈阳工业大学 | Numerical control machine tool milling cutter abrasion real-time monitoring method based on deep convolutional neural network |
CN113779101A (en) * | 2021-11-10 | 2021-12-10 | 北京航空航天大学 | Time sequence set recommendation system and method based on deep neural network |
CN113780153A (en) * | 2021-09-07 | 2021-12-10 | 北京理工大学 | Cutter wear monitoring and predicting method |
CN113822139A (en) * | 2021-07-27 | 2021-12-21 | 河北工业大学 | Equipment fault diagnosis method based on improved 1DCNN-BilSTM |
CN113971489A (en) * | 2021-10-25 | 2022-01-25 | 哈尔滨工业大学 | Method and system for predicting remaining service life based on hybrid neural network |
CN114915496A (en) * | 2022-07-11 | 2022-08-16 | 广州番禺职业技术学院 | Network intrusion detection method and device based on time weight and deep neural network |
CN114925807A (en) * | 2022-04-11 | 2022-08-19 | 大连理工大学 | Wear data marking method for centrifugal compressor impeller machining tool |
CN115299962A (en) * | 2022-08-12 | 2022-11-08 | 山东大学 | Anesthesia depth monitoring method based on bidirectional gate control circulation unit and attention mechanism |
CN115446663A (en) * | 2022-10-14 | 2022-12-09 | 常州先进制造技术研究所 | Cutter wear state monitoring method based on deep learning network of physical guidance and application |
CN115582733A (en) * | 2022-10-28 | 2023-01-10 | 大连工业大学 | Milling cutter wear monitoring method based on residual error structure and convolutional neural network |
CN116308304A (en) * | 2023-05-24 | 2023-06-23 | 山东建筑大学 | New energy intelligent operation and maintenance method and system based on meta learning concept drift detection |
CN117272138A (en) * | 2023-09-15 | 2023-12-22 | 东华理工大学 | Geomagnetic data denoising method and system based on reference channel data constraint and deep learning |
CN117633528A (en) * | 2023-11-21 | 2024-03-01 | 元始智能科技(南通)有限公司 | Manufacturing workshop energy consumption prediction technology based on small sample data restoration and enhancement |
Citations (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5579232A (en) * | 1993-03-29 | 1996-11-26 | General Electric Company | System and method including neural net for tool break detection |
US20050265124A1 (en) * | 2004-04-22 | 2005-12-01 | Smith Gregory C | Method for detecting acoustic emission using a microwave doppler radar detector |
GB0809516D0 (en) * | 2006-02-06 | 2008-07-02 | Smith International | Method of real-time drilling simulation |
US8781982B1 (en) * | 2011-09-23 | 2014-07-15 | Lockheed Martin Corporation | System and method for estimating remaining useful life |
CN103962888A (en) * | 2014-05-12 | 2014-08-06 | 西北工业大学 | Tool abrasion monitoring method based on wavelet denoising and Hilbert-Huang transformation |
WO2017192821A1 (en) * | 2016-05-06 | 2017-11-09 | Massachusetts Institute Of Technology | Method and apparatus for efficient use of cnc machine shaping tool including cessation of use no later than the onset of tool deterioration by monitoring audible sound during shaping |
CN107584334A (en) * | 2017-08-25 | 2018-01-16 | 南京航空航天大学 | A kind of complex structural member numerical control machining cutter status real time monitor method based on deep learning |
CN108319962A (en) * | 2018-01-29 | 2018-07-24 | 安徽大学 | A kind of Tool Wear Monitoring method based on convolutional neural networks |
CN108520125A (en) * | 2018-03-29 | 2018-09-11 | 上海理工大学 | A kind of method and system of prediction cutting-tool wear state |
CN108747590A (en) * | 2018-06-28 | 2018-11-06 | 哈尔滨理工大学 | A kind of tool wear measurement method based on rumble spectrum and neural network |
CN108960077A (en) * | 2018-06-12 | 2018-12-07 | 南京航空航天大学 | A kind of intelligent failure diagnosis method based on Recognition with Recurrent Neural Network |
CN109145319A (en) * | 2017-06-16 | 2019-01-04 | 哈尔滨理工大学 | Key equipment cutting tool method for predicting residual useful life based on deep neural network |
CN109571141A (en) * | 2018-11-01 | 2019-04-05 | 北京理工大学 | A kind of Monitoring Tool Wear States in Turning based on machine learning |
CN109822399A (en) * | 2019-04-08 | 2019-05-31 | 浙江大学 | Cutting tool for CNC machine state of wear prediction technique based on parallel deep neural network |
CN110000610A (en) * | 2019-04-17 | 2019-07-12 | 哈尔滨理工大学 | A kind of Tool Wear Monitoring method based on multi-sensor information fusion and depth confidence network |
CN110153802A (en) * | 2019-07-04 | 2019-08-23 | 西南交通大学 | A kind of cutting-tool wear state discrimination method based on convolutional neural networks and long Memory Neural Networks conjunctive model in short-term |
CN110188637A (en) * | 2019-05-17 | 2019-08-30 | 西安电子科技大学 | A kind of Activity recognition technical method based on deep learning |
CN110509109A (en) * | 2019-07-16 | 2019-11-29 | 西安交通大学 | Tool Wear Monitoring method based on multiple dimensioned depth convolution loop neural network |
-
2020
- 2020-01-31 CN CN202010077631.9A patent/CN111325112B/en active Active
Patent Citations (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5579232A (en) * | 1993-03-29 | 1996-11-26 | General Electric Company | System and method including neural net for tool break detection |
US20050265124A1 (en) * | 2004-04-22 | 2005-12-01 | Smith Gregory C | Method for detecting acoustic emission using a microwave doppler radar detector |
GB0809516D0 (en) * | 2006-02-06 | 2008-07-02 | Smith International | Method of real-time drilling simulation |
US8781982B1 (en) * | 2011-09-23 | 2014-07-15 | Lockheed Martin Corporation | System and method for estimating remaining useful life |
CN103962888A (en) * | 2014-05-12 | 2014-08-06 | 西北工业大学 | Tool abrasion monitoring method based on wavelet denoising and Hilbert-Huang transformation |
WO2017192821A1 (en) * | 2016-05-06 | 2017-11-09 | Massachusetts Institute Of Technology | Method and apparatus for efficient use of cnc machine shaping tool including cessation of use no later than the onset of tool deterioration by monitoring audible sound during shaping |
CN109145319A (en) * | 2017-06-16 | 2019-01-04 | 哈尔滨理工大学 | Key equipment cutting tool method for predicting residual useful life based on deep neural network |
CN107584334A (en) * | 2017-08-25 | 2018-01-16 | 南京航空航天大学 | A kind of complex structural member numerical control machining cutter status real time monitor method based on deep learning |
CN108319962A (en) * | 2018-01-29 | 2018-07-24 | 安徽大学 | A kind of Tool Wear Monitoring method based on convolutional neural networks |
CN108520125A (en) * | 2018-03-29 | 2018-09-11 | 上海理工大学 | A kind of method and system of prediction cutting-tool wear state |
CN108960077A (en) * | 2018-06-12 | 2018-12-07 | 南京航空航天大学 | A kind of intelligent failure diagnosis method based on Recognition with Recurrent Neural Network |
CN108747590A (en) * | 2018-06-28 | 2018-11-06 | 哈尔滨理工大学 | A kind of tool wear measurement method based on rumble spectrum and neural network |
CN109571141A (en) * | 2018-11-01 | 2019-04-05 | 北京理工大学 | A kind of Monitoring Tool Wear States in Turning based on machine learning |
CN109822399A (en) * | 2019-04-08 | 2019-05-31 | 浙江大学 | Cutting tool for CNC machine state of wear prediction technique based on parallel deep neural network |
CN110000610A (en) * | 2019-04-17 | 2019-07-12 | 哈尔滨理工大学 | A kind of Tool Wear Monitoring method based on multi-sensor information fusion and depth confidence network |
CN110188637A (en) * | 2019-05-17 | 2019-08-30 | 西安电子科技大学 | A kind of Activity recognition technical method based on deep learning |
CN110153802A (en) * | 2019-07-04 | 2019-08-23 | 西南交通大学 | A kind of cutting-tool wear state discrimination method based on convolutional neural networks and long Memory Neural Networks conjunctive model in short-term |
CN110509109A (en) * | 2019-07-16 | 2019-11-29 | 西安交通大学 | Tool Wear Monitoring method based on multiple dimensioned depth convolution loop neural network |
Non-Patent Citations (1)
Title |
---|
唐利平等: "刀具磨损状态特征参数提取与识别方法研究" * |
Cited By (55)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111814728A (en) * | 2020-07-22 | 2020-10-23 | 同济大学 | Method for recognizing wear state of cutting tool of numerical control machine tool and storage medium |
CN112001527B (en) * | 2020-07-29 | 2024-01-30 | 中国计量大学 | Industrial production process target data prediction method of multi-feature fusion depth neural network |
CN112001527A (en) * | 2020-07-29 | 2020-11-27 | 中国计量大学 | Industrial production process target data prediction method of multi-feature fusion deep neural network |
CN112070208A (en) * | 2020-08-05 | 2020-12-11 | 同济大学 | Tool wear prediction method based on encoder-decoder stage attention mechanism |
CN112070208B (en) * | 2020-08-05 | 2022-08-30 | 同济大学 | Tool wear prediction method based on encoder-decoder stage attention mechanism |
CN111879397A (en) * | 2020-09-01 | 2020-11-03 | 国网河北省电力有限公司检修分公司 | Fault diagnosis method for energy storage mechanism of high-voltage circuit breaker |
CN112033463A (en) * | 2020-09-02 | 2020-12-04 | 哈尔滨工程大学 | Nuclear power equipment state evaluation and prediction integrated method and system |
CN112033463B (en) * | 2020-09-02 | 2022-09-06 | 哈尔滨工程大学 | Nuclear power equipment state evaluation and prediction integrated method and system |
CN112115550A (en) * | 2020-09-13 | 2020-12-22 | 西北工业大学 | Aircraft maneuvering trajectory prediction method based on Mogrifier-BiGRU |
CN112115550B (en) * | 2020-09-13 | 2022-04-19 | 西北工业大学 | Aircraft maneuvering trajectory prediction method based on Mogrifier-BiGRU |
CN112149355A (en) * | 2020-09-27 | 2020-12-29 | 浙江科技学院 | Soft measurement method based on semi-supervised dynamic feedback stack noise reduction self-encoder model |
CN112149355B (en) * | 2020-09-27 | 2023-08-22 | 浙江科技学院 | Soft measurement method based on semi-supervised dynamic feedback stack noise reduction self-encoder model |
CN112200032A (en) * | 2020-09-28 | 2021-01-08 | 辽宁石油化工大学 | Attention mechanism-based high-voltage circuit breaker mechanical state online monitoring method |
CN112200032B (en) * | 2020-09-28 | 2023-05-30 | 辽宁石油化工大学 | Online monitoring method for mechanical state of high-voltage circuit breaker based on attention mechanism |
CN112329625A (en) * | 2020-11-05 | 2021-02-05 | 中国科学技术大学 | Cutter wear state real-time identification method and model based on deep learning |
CN112629854A (en) * | 2020-11-25 | 2021-04-09 | 西安交通大学 | Bearing fault classification method based on neural network attention mechanism |
CN112668507A (en) * | 2020-12-31 | 2021-04-16 | 南京信息工程大学 | Sea clutter prediction method and system based on hybrid neural network and attention mechanism |
CN112798956A (en) * | 2020-12-31 | 2021-05-14 | 江苏国科智能电气有限公司 | Wind turbine generator fault diagnosis method based on multi-resolution time sequence cyclic neural network |
CN112712063B (en) * | 2021-01-18 | 2022-04-26 | 贵州大学 | Tool wear value monitoring method, electronic device and storage medium |
CN112712063A (en) * | 2021-01-18 | 2021-04-27 | 贵州大学 | Tool wear value monitoring method, electronic device and storage medium |
CN112785092B (en) * | 2021-03-09 | 2024-05-07 | 中铁电气化局集团有限公司 | Switch residual life prediction method based on self-adaptive deep feature extraction |
CN112785092A (en) * | 2021-03-09 | 2021-05-11 | 中铁电气化局集团有限公司 | Turnout residual life prediction method based on self-adaptive deep layer feature extraction |
CN113204640A (en) * | 2021-04-02 | 2021-08-03 | 南京邮电大学 | Text classification method based on attention mechanism |
CN113204640B (en) * | 2021-04-02 | 2023-05-30 | 南京邮电大学 | Text classification method based on attention mechanism |
CN113177584A (en) * | 2021-04-19 | 2021-07-27 | 合肥工业大学 | Zero sample learning-based composite fault diagnosis method |
CN113177584B (en) * | 2021-04-19 | 2022-10-28 | 合肥工业大学 | Compound fault diagnosis method based on zero sample learning |
CN113051689A (en) * | 2021-04-25 | 2021-06-29 | 石家庄铁道大学 | Bearing residual service life prediction method based on convolution gating circulation network |
CN113286309B (en) * | 2021-05-18 | 2023-02-07 | 合肥工业大学 | Heterogeneous communication method and system based on CSI |
CN113286309A (en) * | 2021-05-18 | 2021-08-20 | 合肥工业大学 | Heterogeneous communication method and system based on CSI |
CN113414638A (en) * | 2021-06-04 | 2021-09-21 | 西北工业大学 | Variable working condition milling cutter wear state prediction method based on milling force time sequence diagram deep learning |
CN113305645A (en) * | 2021-06-22 | 2021-08-27 | 重庆邮电大学工业互联网研究院 | Numerical control machine tool cutter residual life prediction method based on mixed neural model |
CN113359577B (en) * | 2021-07-02 | 2023-08-11 | 中国科学院空间应用工程与技术中心 | Ultrasonic motor embedded state monitoring and fault diagnosis system and method |
CN113359577A (en) * | 2021-07-02 | 2021-09-07 | 中国科学院空间应用工程与技术中心 | Embedded state monitoring and fault diagnosis system and method for ultrasonic motor |
CN113657454B (en) * | 2021-07-23 | 2024-02-23 | 杭州安脉盛智能技术有限公司 | Nuclear power rotating machinery state monitoring method based on autoregressive BiGRU |
CN113657454A (en) * | 2021-07-23 | 2021-11-16 | 杭州安脉盛智能技术有限公司 | Autoregressive BiGRU-based nuclear power rotating machine state monitoring method |
CN113537472B (en) * | 2021-07-26 | 2024-04-09 | 北京计算机技术及应用研究所 | Construction method of bidirectional recurrent neural network with low calculation and storage consumption |
CN113537472A (en) * | 2021-07-26 | 2021-10-22 | 北京计算机技术及应用研究所 | Bidirectional recurrent neural network with low computation and storage consumption |
CN113822139B (en) * | 2021-07-27 | 2023-08-25 | 河北工业大学 | Equipment fault diagnosis method based on improved 1DCNN-BiLSTM |
CN113822139A (en) * | 2021-07-27 | 2021-12-21 | 河北工业大学 | Equipment fault diagnosis method based on improved 1DCNN-BilSTM |
CN113664612A (en) * | 2021-08-24 | 2021-11-19 | 沈阳工业大学 | Numerical control machine tool milling cutter abrasion real-time monitoring method based on deep convolutional neural network |
CN113780153A (en) * | 2021-09-07 | 2021-12-10 | 北京理工大学 | Cutter wear monitoring and predicting method |
CN113971489A (en) * | 2021-10-25 | 2022-01-25 | 哈尔滨工业大学 | Method and system for predicting remaining service life based on hybrid neural network |
CN113779101A (en) * | 2021-11-10 | 2021-12-10 | 北京航空航天大学 | Time sequence set recommendation system and method based on deep neural network |
CN113779101B (en) * | 2021-11-10 | 2022-03-18 | 北京航空航天大学 | Time sequence set recommendation system and method based on deep neural network |
CN114925807A (en) * | 2022-04-11 | 2022-08-19 | 大连理工大学 | Wear data marking method for centrifugal compressor impeller machining tool |
CN114925807B (en) * | 2022-04-11 | 2024-06-11 | 大连理工大学 | Method for marking abrasion data of impeller machining cutter of centrifugal compressor |
CN114915496A (en) * | 2022-07-11 | 2022-08-16 | 广州番禺职业技术学院 | Network intrusion detection method and device based on time weight and deep neural network |
CN115299962A (en) * | 2022-08-12 | 2022-11-08 | 山东大学 | Anesthesia depth monitoring method based on bidirectional gate control circulation unit and attention mechanism |
CN115446663A (en) * | 2022-10-14 | 2022-12-09 | 常州先进制造技术研究所 | Cutter wear state monitoring method based on deep learning network of physical guidance and application |
CN115446663B (en) * | 2022-10-14 | 2023-10-20 | 常州先进制造技术研究所 | Tool wear state monitoring method based on physical guidance deep learning network and application |
CN115582733A (en) * | 2022-10-28 | 2023-01-10 | 大连工业大学 | Milling cutter wear monitoring method based on residual error structure and convolutional neural network |
CN116308304A (en) * | 2023-05-24 | 2023-06-23 | 山东建筑大学 | New energy intelligent operation and maintenance method and system based on meta learning concept drift detection |
CN116308304B (en) * | 2023-05-24 | 2023-08-25 | 山东建筑大学 | New energy intelligent operation and maintenance method and system based on meta learning concept drift detection |
CN117272138A (en) * | 2023-09-15 | 2023-12-22 | 东华理工大学 | Geomagnetic data denoising method and system based on reference channel data constraint and deep learning |
CN117633528A (en) * | 2023-11-21 | 2024-03-01 | 元始智能科技(南通)有限公司 | Manufacturing workshop energy consumption prediction technology based on small sample data restoration and enhancement |
Also Published As
Publication number | Publication date |
---|---|
CN111325112B (en) | 2023-04-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111325112B (en) | Cutter wear state monitoring method based on depth gate control circulation unit neural network | |
CN112435363B (en) | Cutter wear state real-time monitoring method | |
CN111177911B (en) | Part surface roughness online prediction method based on SDAE-DBN algorithm | |
CN112070209B (en) | Stable controllable image generation model training method based on W distance | |
Liang et al. | Rolling bearing fault diagnosis based on one-dimensional dilated convolution network with residual connection | |
CN111626355A (en) | Unet + + convolutional neural network-based seismic data first arrival pickup method | |
CN109389171B (en) | Medical image classification method based on multi-granularity convolution noise reduction automatic encoder technology | |
CN109934158B (en) | Video emotion recognition method based on local enhanced motion history map and recursive convolutional neural network | |
CN113664612A (en) | Numerical control machine tool milling cutter abrasion real-time monitoring method based on deep convolutional neural network | |
CN113923104B (en) | Network fault diagnosis method, equipment and storage medium based on wavelet neural network | |
CN113822139B (en) | Equipment fault diagnosis method based on improved 1DCNN-BiLSTM | |
CN113554621B (en) | Cutter wear state identification system and method based on wavelet scale map and deep migration learning | |
CN115100574A (en) | Action identification method and system based on fusion graph convolution network and Transformer network | |
CN112464004A (en) | Multi-view depth generation image clustering method | |
CN111145145B (en) | Image surface defect detection method based on MobileNet | |
CN108596044B (en) | Pedestrian detection method based on deep convolutional neural network | |
CN113705715B (en) | Time sequence classification method based on LSTM and multi-scale FCN | |
Ma et al. | Automatic recognition of machining features based on point cloud data using convolution neural networks | |
Wu et al. | Quality estimation method for gear hobbing based on attention and adversarial transfer learning | |
CN117009916A (en) | Actuator fault diagnosis method based on multi-sensor information fusion and transfer learning | |
Zhang et al. | A deep learning-based multi-sensor data fusion method for degradation monitoring of ball screws | |
Essiet et al. | Big data analysis for gas sensor using convolutional neural network and ensemble of evolutionary algorithms | |
CN117590173A (en) | Cable partial discharge pattern recognition method based on convolutional neural network | |
CN117496157A (en) | Relatively small part point cloud semantic segmentation method based on MS module | |
CN115357862B (en) | Positioning method in long and narrow space |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |