CN109753923A - Method, system, device and computer-readable storage medium for monitoring tool wear amount - Google Patents
Method, system, device and computer-readable storage medium for monitoring tool wear amount Download PDFInfo
- Publication number
- CN109753923A CN109753923A CN201811643774.0A CN201811643774A CN109753923A CN 109753923 A CN109753923 A CN 109753923A CN 201811643774 A CN201811643774 A CN 201811643774A CN 109753923 A CN109753923 A CN 109753923A
- Authority
- CN
- China
- Prior art keywords
- tdcnn
- lstm network
- tool
- network model
- 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
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
本发明提供了一种刀具磨损量的监测方法，包括：采集监测刀具的输入数据；对所述输入数据进行预处理得到输入样本，并获取所述输入样本对应的真实刀具磨损值，生成训练样本和测试样本；构建TDCNN‑LSTM网络模型；根据所述训练样本和所述测试样本，训练并调整所述TDCNN‑LSTM网络模型的所述网络参数，以得到改进TDCNN‑LSTM网络模型；以及获取待测数据，将所述待测数据经预处理后输入改进TDCNN‑LSTM网络模型，以得到相应的目标刀具磨损值。采用TDCNN和LSTM网络的结合，可以充分提取不同尺度的信息，提高了刀具磨损量预测的精度及其提高了工作效率，此外，本发明还提供了一种刀具磨损量的监测系统、设备及计算机可读存储介质。
The present invention provides a method for monitoring tool wear, comprising: collecting input data for monitoring tools; preprocessing the input data to obtain input samples, and obtaining the real tool wear values corresponding to the input samples to generate training samples and test samples; construct a TDCNN-LSTM network model; according to the training samples and the test samples, train and adjust the network parameters of the TDCNN-LSTM network model to obtain an improved TDCNN-LSTM network model; The measured data is input into the improved TDCNN-LSTM network model after preprocessing, so as to obtain the corresponding target tool wear value. The combination of TDCNN and LSTM network can fully extract information of different scales, improve the precision of tool wear amount prediction and improve work efficiency, in addition, the invention also provides a tool wear amount monitoring system, equipment and computer Readable storage medium.
Description
Technical field
The present invention relates to the technical field of mechanical fault diagnosis, in particular to a kind of monitoring method of tool abrasion is
System, equipment and computer readable storage medium.
Background technique
Being constantly progressive and develop with manufacturing industry technology, manufacturing field wants work pieces process efficiency and processing quality
Ask also higher and higher, wherein the degeneration of cutting performance is the principal element for causing machine failure to shut down, and influences processing matter
The direct factor of amount and processing efficiency.Currently, in actual processing, when unknown cutting-tool wear state, to ensure to add
The precision of work workpiece generally regularly replaces cutter by the way of conservative estimation, the inaccuracy of this Cutter wear situation
There are the following problems for the mode of judgement: it is replaced when if (1) tool abrasion being lower than blunt standard, will cause cutter waste,
Increase processing cost；(2) if cutter has occurred and that abrasion or breakage without replacing cutter in time, not only will affect workpiece
Surface quality and dimensional accuracy, cause waste of material, or even can also damage lathe；(3) because failure leads to the time shut down
It is longer, it will seriously affect processing efficiency etc..
The key for solving problem above is the real-time monitoring of realization cutter health status, passes through Cutter wear amount
Real-time monitoring can identify the abrasion loss of cutter in real time and accurately, then by the wear threshold of setting, accurately capture
Optimal tool replacement time, both can ensure that as a result, in time replacement cutter, with guarantee workpieces processing surface quality and dimensional accuracy,
It reduces the waste of workpiece raw material and reduces downtime, and be avoided that cutter working life caused by frequently replacement cutter
Waste and the increase of the cost of charp tool.
However, the signal processing of some technical literatures is using traditional in existing cutter real time monitoring
Signal processing and machine Learning Theory are technical support, and data handling procedure is cumbersome, and the manual features being directed to mention
Step is taken to need to rely on the domain knowledge and expertise of profession, time-consuming and accuracy is not high for the signal processing.Depth
Learn the disruptive technology as machine learning field, the deep layer nonlinear characteristic of study initial data that can be adaptive, from
And disadvantages mentioned above can be overcome, therefore, other technical literatures are then based on convolutional neural networks Cutter wear amount and are supervised
It surveys, corresponding energy frequency spectrum figure will be converted to by wavelet packet without the original vibration signal of denoising, then using convolution mind
Tool wear degree is obtained through network analysis.But its network model used is only single convolutional neural networks model, energy
The data characteristics enough extracted is limited, and the precision of monitoring is not high.
Summary of the invention
In view of this, the present invention is intended to provide a kind of tool abrasion that can be improved working efficiency, promote monitoring accuracy
Monitoring method, system, equipment and computer readable storage medium, with solve monitoring efficiency existing in the prior art it is low and
The not high technical problem of monitoring accuracy.
For this purpose, first aspect present invention provides a kind of monitoring method of tool abrasion comprising: acquisition monitoring knife
The input data of tool；The input data is pre-processed to obtain input sample, and it is corresponding true to obtain the input sample
Real tool wear value generates training sample and test sample；Construct TDCNN-LSTM network model comprising: building TDCNN
Simultaneously its network parameter is arranged in network, and the TDCNN network is used to extract the local spatial feature of the input sample；And structure
It builds LSTM network and its network parameter is set, the LSTM network is used to extract the changing character of the input sample；
According to the training sample and the test sample, training and the network ginseng for adjusting the TDCNN-LSTM network model
Number, to obtain improving TDCNN-LSTM network model；And testing data is obtained, the testing data is defeated after pretreatment
Enter to improve TDCNN-LSTM network model, to obtain corresponding target tool attrition value.
In the first aspect of the invention, the implied feature in data is automatically extracted by TDCNN-LSTM network model,
And corresponding tool wear value is exported, feature set is extracted and selected without artificial, solves the feature selecting needs of conventional method
The problem of relying on a large amount of signal processing knowledge and expertise, improves work efficiency；Also, it is mentioned using TDCNN network model
The local spatial feature and LSTM network model of evidence of fetching extract the temporal aspect of data, pass through the knot of TDCNN and LSTM network
It closes, can sufficiently extract the information of different scale, can be excavated than single CNN network model and single LSTM network model
More implied features out, therefore the characteristic of significantly more efficient tool abrasion sensitivity can be extracted, to improve knife
Has the precision of Abrasion prediction.
In addition, in the monitoring method of the tool abrasion involved in the first aspect of the present invention, optionally, described
Before building TDCNN-LSTM network model, further includes: the input sample is divided into N number of subsequence along time dimension.?
In this case, after input data being split, can it is significantly more efficient realize local spatial feature extraction, improve
The reliability that local spatial feature is extracted.
In addition, in the monitoring method of the tool abrasion involved in the first aspect of the present invention, it is optionally, described
TDCNN network includes N number of local spatial feature extractor, and the local spatial feature extractor includes more than two convolution knots
Structure, wherein each convolutional coding structure includes a convolutional layer, one batch of standardization layer, an active coating and an average pond
Change layer.In this case, each local spatial feature extractor can carry out convolutional calculation etc. to its corresponding subsequence
Operation, and by the collateral action of multiple local spatial feature extractors, can calculating in effective lifting feature extraction process
Speed, to promote the monitoring efficiency of tool abrasion.
In addition, in the monitoring method of the tool abrasion involved in the first aspect of the present invention, optionally, the office
The extraction of portion's space characteristics includes: that N number of local spatial feature extractor is independently acted on its corresponding son
The subsequence is converted to fisrt feature figure by calculating, wherein the meter of N number of local spatial feature extractor by sequence
Calculation process is parallel.In such a case, it is possible to the significantly more efficient extraction for realizing local spatial feature, and by multiple
The collateral action of local shape factor device, can calculating speed effectively in lifting feature extraction process, to promote cutter mill
The monitoring efficiency of damage amount.
In addition, in the monitoring method of the tool abrasion involved in the first aspect of the present invention, it is optionally, described defeated
Entering data includes: vibration data, is the vibration signal of monitoring cutter, by being installed on electric machine main shaft and close to the monitoring
The vibrating sensor of cutter measures；And current data, it is the current signal of monitoring cutter, by being installed on motor driver
The current transformer of delivery outlet measure.As a result, by the combination of a variety of data, the accuracy of prediction result can be improved, and
Without manually extracting and selecting feature set, the feature selecting for solving conventional method need to rely on a large amount of signal processing knowledge and
The problem of expertise.
In addition, in the monitoring method of the tool abrasion involved in the first aspect of the present invention, it is optionally, described true
Real tool wear value is measured by being measured microscopically tool.Thus, it is possible to which more intuitive obtain the big of real tool attrition value
It is small, to more easily realize it compared with the judgement of prediction tool wear value.
In addition, in the monitoring method of the tool abrasion involved in the first aspect of the present invention, it is optionally, described pre-
Processing is to be standardized using Z-score method to the input data.In such a case, it is possible to by input number
According to standardization, the validity of input sample is promoted, to improve the operating efficiency of TDCNN-LSTM network model and steady
It is qualitative.
In addition, in the monitoring method of the tool abrasion involved in the first aspect of the present invention, it is optionally, described
LSTM network is two layers or more, and is multiple-input and multiple-output structure.It, can be with as a result, by the setting of multilayer LSTM network
It realizes and multistep extraction is carried out to the changing character of input data, to improve the precision and effectively of changing character extraction
Property.
In addition, in the monitoring method of the tool abrasion involved in the first aspect of the present invention, optionally, when described
The extraction of sequence variation characteristic includes: to convert second feature figure for the fisrt feature figure using the LSTM network；It will be described
Second feature figure is input to Flatten layers, and the second feature figure is converted to one-dimensional characteristic；The one-dimensional characteristic is defeated
Enter to full articulamentum, the one-dimensional characteristic is converted into output feature；And the output feature is input to output layer,
Tool wear value is predicted with output.In this case, enter data into Flatten layers and full articulamentum and output layer it
Before, using LSTM network to data carry out changing character extraction, because LSTM Web vector graphic input gate, forget door and
The more complicated structure such as out gate controls the function of network implementations long-term memory, has than common full articulamentum etc.
More excellent long short-term memory circulatory function, the changing character for the data that can be extracted more have specific aim, and
And the use of LSTM network, it is thus also avoided that enter data directly into the explosion of gradient caused by full articulamentum etc. and disappear with gradient
The problems such as, improve the validity of changing character extraction.
In addition, in the monitoring method of the tool abrasion involved in the first aspect of the present invention, optionally, to described
The training and adjustment of the network parameter of TDCNN-LSTM network model include: using the training sample to described
TDCNN-LSTM network model is trained；The test sample is inputted in trained TDCNN-LSTM network model, is obtained
To prediction tool wear value；According to the prediction tool wear value real tool mill corresponding with the test sample
Damage value, judges whether the precision of prediction of the TDCNN-LSTM network model is greater than or equal to threshold value；If it is not, then described in adjustment
It is trained again after the network parameter of TDCNN-LSTM network model, the TDCNN-LSTM net after training
The precision of prediction of network model is greater than or equal to the threshold value, to obtain the improvement TDCNN-LSTM network model.At this
In the case of kind, by the repetition training and adjustment to TDCNN-LSTM network model, TDCNN-LSTM network can be continuously improved
The precision of prediction of model promotes it so that it is guaranteed that improving the numerical reliability of TDCNN-LSTM network model during the test
The precision of measurement.
In addition, in the monitoring method of the tool abrasion involved in the first aspect of the present invention, it is optionally, described pre-
Survey precision judge index can for the real tool attrition value and it is described prediction tool wear value mean absolute error and
Root-mean-square error.In this case, judge index uses the combination of mean absolute error and root-mean-square error, can be into one
Step promotes the validity and reliability of judging result, to promote the precision of prediction of TDCNN-LSTM network model.
In addition, second aspect of the present invention provides a kind of monitoring system of tool abrasion comprising: data acquisition module
Block, for acquiring the input data of monitoring cutter；Preprocessing module, it is defeated for being pre-processed to obtain to the input data
Enter sample, and obtain the corresponding real tool attrition value of the input sample, generates training sample and test sample；Model structure
Block is modeled, for constructing TDCNN-LSTM network model comprising: simultaneously its network parameter is arranged in building TDCNN network, described
TDCNN network is used to extract the local spatial feature of the input sample；And it constructs LSTM network and its network ginseng is set
Number, the LSTM network are used to extract the changing character of the input sample；Training and adjustment module, for according to institute
Training sample and the test sample, training and the network parameter for adjusting the TDCNN-LSTM network model are stated, with
To improvement TDCNN-LSTM network model；And test module is located the testing data for obtaining testing data in advance
Input improves TDCNN-LSTM network model after reason, to obtain corresponding target tool attrition value.
In addition, optionally, further including in the monitoring system of the tool abrasion involved in the second aspect of the present invention
The data segmentation module being connected with the model construction module, for the input sample to be divided into N number of son along time dimension
Sequence.
In addition, third aspect present invention provides a kind of monitoring device of tool abrasion comprising: memory is used for
Store computer program；And processor, as above any tool wear is realized when for executing the computer program
The step of monitoring method of amount.
In addition, fourth aspect present invention provides a kind of computer readable storage medium, optionally, the computer can
It reads to be stored with computer program on storage medium, be realized when the computer program is executed by processor as above any described
The step of monitoring method of tool abrasion.
Compared with the existing technology, the monitoring method, system, equipment of tool abrasion of the present invention and computer can
Storage medium is read to have the advantage that
The implied feature in data is automatically extracted by TDCNN-LSTM network model, and exports corresponding tool wear
Value, without artificial extraction and selection feature set, the feature selecting for solving conventional method needs to rely on a large amount of signal processing knowledge
And the problem of expertise, it improves work efficiency；Also, the local spatial feature of data is extracted using TDCNN network model
And the temporal aspect of LSTM network model extraction data can sufficiently extract difference by the combination of TDCNN and LSTM network
The information of scale can excavate more implied features than single CNN network model and single LSTM network model, because
This can extract the characteristic of significantly more efficient tool abrasion sensitivity, to improve the precision of tool abrasion prediction.
Detailed description of the invention
The attached drawing for constituting a part of the invention is used to provide further understanding of the present invention, and of the invention is schematic
Examples and descriptions thereof are used to explain the present invention, does not constitute improper limitations of the present invention.In the accompanying drawings:
Fig. 1 is the flow chart of the monitoring method of tool abrasion involved in embodiments of the present invention；
Fig. 2 is in the monitoring method of tool abrasion involved in embodiments of the present invention to TDCNN-LSTM network
The training of the network parameter of model and the flow chart adjusted；
Fig. 3 is the structural schematic diagram that TDCNN-LSTM network model is improved involved in embodiments of the present invention；
Fig. 4 is the structural schematic diagram of the monitoring system of tool abrasion involved in embodiments of the present invention；
Fig. 5 is the structural schematic diagram of the monitoring device of tool abrasion involved in embodiments of the present invention.
Specific embodiment
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the present invention can
To be combined with each other.Term " first ", " second " in description and claims of this specification and above-mentioned attached drawing, "
Three " and " the 4th " etc. are not use to describe a particular order for distinguishing different objects.
The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiment.
Fig. 1 is the flow chart of the monitoring method of tool abrasion involved in embodiments of the present invention；Fig. 2 is this hair
To the instruction of the network parameter of TDCNN-LSTM network model in the monitoring method of tool abrasion involved in bright embodiment
The flow chart practiced and adjusted；Fig. 3 is that the structure of improvement TDCNN-LSTM network model involved in embodiments of the present invention is shown
It is intended to.
Hereinafter, referring to figs. 1 to Fig. 3, describing the monitoring method of tool abrasion involved in present embodiment in detail.
As shown in Figure 1, tool wear quantity monitoring method involved in present embodiment may include steps of: acquisition
Monitor the input data (step S101) of cutter；Input data is pre-processed to obtain input sample, and obtains input sample
Corresponding real tool attrition value generates training sample and test sample (step S102)；By input sample along time dimension point
It is segmented into N number of subsequence (step S103)；Construct TDCNN-LSTM network model comprising: simultaneously it is arranged in building TDCNN network
Network parameter, TDCNN network are used to extract the local spatial feature of input sample, and, it constructs LSTM network and it is set
Network parameter, LSTM network are used to extract the changing character (step S104) of input sample；According to training sample and test
Sample, training and the network parameter for adjusting TDCNN-LSTM network model, to obtain improving TDCNN-LSTM network model (step
Rapid S105)；And testing data is obtained, the testing data is inputted after pretreatment and improves TDCNN-LSTM network model,
To obtain corresponding target tool attrition value (step S106).
In the present embodiment, the implied feature in data is automatically extracted by TDCNN-LSTM network model, and exported
Feature set is extracted and selected to corresponding tool wear value without artificial, and the feature selecting for solving conventional method needs to rely on big
The problem of measuring signal processing knowledge and expertise, improves work efficiency；Also, data are extracted using TDCNN network model
Local spatial feature and LSTM network model extract data temporal aspect can by the combination of TDCNN and LSTM network
Sufficiently to extract the information of different scale, can be excavated more than single CNN network model and single LSTM network model
Implied feature, therefore the characteristic of significantly more efficient tool abrasion sensitivity can be extracted, to improve tool wear
Measure the precision of prediction.
In the present embodiment, in step s101, the input data of acquisition monitoring cutter.Wherein, the kind of input data
Class is not particularly limited.In some instances, input data can be vibration data and current data, such as monitor cutter
Vibration signal and current signal.In other examples, the quantity of vibration data and current data can be respectively one with
On, such as can be respectively one, three or five etc..As a result, by the combination of a variety of or multiple data, can be improved pre-
Survey the accuracy of result.
In addition, in the present embodiment, the acquisition mode of vibration data and current data is not particularly limited.Some
In example, vibration data can be measured on electric machine main shaft and close to the vibrating sensor of monitoring cutter by being installed on, current data
It can be measured by the current transformer for being installed on the delivery outlet of motor driver.In other examples, vibration data can be by
It is installed on electric machine main shaft and is measured close to the three-way vibration sensor of monitoring cutter, current data can be by being installed on motor
Three current transformers of three delivery outlets of driver measure.Pass through the combination of a variety of or multiple data, Ke Yiti as a result,
The accuracy of high prediction result.
In the present embodiment, in step s 102, input data is pre-processed to obtain input sample, and is obtained
The corresponding real tool attrition value of input sample generates training sample and test sample.Wherein, pretreated form is without spy
It does not limit.In some instances, pretreatment can be to be standardized using Z-score method to input data.In this feelings
Under condition, the validity of input sample can be promoted by the standardization to input data, to improve TDCNN-LSTM
The operating efficiency and stability of network model.
In addition, in the present embodiment, the acquisition mode of real tool attrition value is not particularly limited.In some examples
In, real tool attrition value can be measured by being measured microscopically tool.Thus, it is possible to which more intuitive obtain real tool
The size of attrition value, to more easily realize it compared with the judgement of prediction tool wear value.
As described above, in step S101 and S102, the step of preferred embodiment, is as follows: firstly, three-dimensional is shaken
Dynamic sensor is mounted on electric machine main shaft and close to the place of monitoring cutter, and three current transformers are mounted on motor driven
Three delivery outlets of device；Then, logical per at regular intervals, acquire certain length 6 in the process of monitoring cutter
Road sensor signal is as input data X={ x1,x2,…,xL, wherein L is the length of input sample, the input at per moment
xiIt is the vector comprising 6 elements.Finally, measuring work with abrasion loss of the tool that is measured microscopically to monitoring cutter
To export Y, a tape label sample data (X, Y) is formed.So circulation, the full Life Cycle of available multiple monitoring cutters
Phase tape label sample data.
It is then possible to be standardized respectively using Z-score method to 6 channel signals, the method for specific standards is such as
Under:
In formula, xjIt is the time series of j-th of sensor passage, μjAnd σjIt is xjAverage and standard deviation,It is Z-
Time series data after score normalization.
In the present embodiment, in step s 103, input sample is divided into N number of subsequence along time dimension.At this
Kind in the case of, after input data is split, can it is significantly more efficient realize local spatial feature extraction, improve office
The reliability that portion's space characteristics extract.
As described above, in step s 103, preferred embodiment can be with are as follows: by the input sample X in step S102
={ x1,x2,…,xLAlong time dimension it is divided into N number of subsequence, input becomes X={ PT1,PT2,…,PTN, whereinAnd PTi∈RM×l。
In the present embodiment, in step S104, building TDCNN-LSTM network model may include: building TDCNN
(Chinese name: Annual distribution formula convolutional neural networks, full name in English: Time Distributed Convolutional
Neural Network) network and its network parameter is set, TDCNN network is used to extract the local spatial feature of input sample,
And building LSTM (Chinese name: long short-term memory Recognition with Recurrent Neural Network, full name in English: Long Short-Term
Memory network and its network parameter) is set, LSTM network is used to extract the changing character of input sample.Pass through as a result,
The combination of TDCNN and LSTM network can sufficiently extract the information of different scale, can be than single CNN network model and single
LSTM network model excavate more implied features, therefore it is sensitive to extract significantly more efficient tool abrasion
Characteristic, to improve the precision of tool abrasion prediction.
In addition, in the present embodiment, TDCNN network may include N number of local spatial feature extractor, local space
Feature extractor may include more than two convolutional coding structures, wherein each convolutional coding structure may include a convolutional layer, one
Criticize standardization layer, an active coating and an average pond layer.In this case, each local spatial feature is extracted
Device can carry out the operation such as convolutional calculation to its corresponding subsequence, and by multiple local spatial feature extractors and
Row effect, can calculating speed effectively in lifting feature extraction process, to promote the monitoring efficiency of tool abrasion.
In addition, in the present embodiment, the extraction of local spatial feature may include: to extract N number of local spatial feature
Device independently acts on its corresponding subsequence, subsequence is converted to fisrt feature figure by calculating, wherein N number of part
The calculating process of space characteristics extractor is parallel.In such a case, it is possible to significantly more efficient realization local spatial feature
Extraction can calculating speed effectively in lifting feature extraction process and by the collateral action of multiple local shape factor devices
Degree, to promote the monitoring efficiency of tool abrasion.
In addition, in the present embodiment, the extraction of changing character may include: special by first using LSTM network
Sign figure is converted into second feature figure；Second feature figure is input to Flatten layers, second feature figure is converted into one-dimensional spy
Sign；One-dimensional characteristic is input to full articulamentum, one-dimensional characteristic is converted into output feature；And output feature is input to
Output layer predicts tool wear value with output.In this case, Flatten layers and full articulamentum and defeated are being entered data into
Out before layer, using LSTM network data are carried out with the extraction of changing character, because of LSTM Web vector graphic input gate, is lost
The more complicated structure such as door and out gate is forgotten to control the function of network implementations long-term memory, than common full articulamentum
More have Deng the changing character with more excellent long short-term memory circulatory function, the data that can be extracted and is directed to
Property, also, the use of LSTM network, it is thus also avoided that enter data directly into the explosion of gradient caused by full articulamentum etc. and ladder
The problems such as degree disappears improves the validity of changing character extraction.
As described above, in step S104, the preferred method and step for constructing TDCNN-LSTM network model can be with
Are as follows:
(1) building TDCNN network is used to extract the local spatial feature of input data, by N number of local shape factor device point
Other independent action is in its corresponding subsequence P obtained in step s 103Ti, the calculating process of N number of local shape factor device is
Parallel.Local shape factor device includes 2 convolutional coding structures, and each convolutional coding structure includes 1 convolutional layer, and 1 is criticized at standardization
Manage layer, 1 active coating and 1 average pond layer；
(2) feature that TDCNN network extracts is input in two layers of LSTM Recognition with Recurrent Neural Network, for extracting all sub- sequences
Changing character between column.
Wherein, detailed process is as follows for step (1):
1. by subsequenceIt is input in convolutional layer C1, is (k with size1, 1) volume
Product verification PTiConvolution algorithm is carried out, convolution step-length is (0.5k1, 1), convolution kernel number is c, and obtained characteristic pattern is criticized
Then standardization carries out activation operation with activation primitive ReLU, having a size of (2,1), finally obtaining size is (l in pond1,
M, c) characteristic pattern.
2. by step 1. obtained in characteristic pattern be input in convolutional layer C2, with size be (k2, M) convolution kernel to spy
Sign figure carries out convolution algorithm, and convolution step-length is (1,1), and convolution kernel number is c, and obtained characteristic pattern is carried out at batch standardization
Then reason carries out activation operation with activation primitive ReLU, having a size of (2,1), finally obtaining size is (l in pond2, 1, c) spy
Sign figure.
3. being directed to each subsequence PTi, execute step 1. with step 2..It is by the characteristic pattern dimension transformation of each subsequence
(l2×c).By N number of parallel local shape factor device, converting size for original input data is (N, l2× c) feature
Figure.
Detailed process is as follows for step (2):
(I) feature that step (1) obtains is input in the long memory network LSTM1 in short-term of first layer.LSTM1 is using more
Multi output structure is inputted, N number of time step is shared, wherein a time step of the local feature of each subsequence as LSTM1
Input, the quantity of the output neuron of each time step of LSTM1 are 10, and carry out criticizing standardization to obtained characteristic pattern,
Obtain the characteristic pattern that size is (N, 10)；
(II) feature obtained in step (I) is input in the long memory network LSTM2 in short-term of the second layer.LSTM2 is same
For multiple-input and multiple-output structure, N number of time step is shared, the quantity of the output neuron of each time step of LSTM2 is 5, and to
To characteristic pattern carry out batch standardization, obtain the characteristic pattern that size is (N, 5)；
(III) characteristic pattern that step (II) exports is input in Flatten layers, characteristic pattern is converted into one-dimensional characteristic,
Size is N × 5；
(IV) one-dimensional characteristic for exporting step (III), is input in full articulamentum, the neuronal quantity of full articulamentum is
20, activation primitive is ReLU (Chinese name: amendment linear unit, full name in English: Rectified Linear Units)；
(V) feature for exporting step (IV), is input in output layer, and output layer is full articulamentum, output neuron
Quantity be 1, activation primitive ReLU.
In the present embodiment, in step s105, according to training sample and test sample, training simultaneously adjusts TDCNN-
The network parameter of LSTM network model, to obtain improving TDCNN-LSTM network model.In some instances, as shown in Fig. 2,
Training and adjustment to the network parameter of TDCNN-LSTM network model may include: using training sample to TDCNN-LSTM
Network model is trained (step S201)；Test sample is inputted in trained TDCNN-LSTM network model, is obtained pre-
It surveys tool wear value (step S202)；According to prediction tool wear value and real tool attrition value corresponding in test sample, sentence
Whether the precision of prediction of disconnected TDCNN-LSTM network model is greater than or equal to threshold value (step S203)；If it is not, then adjusting
It is trained again after the network parameter (step S204) of TDCNN-LSTM network model, the TDCNN-LSTM after training
The precision of prediction of network model be greater than or equal to the threshold value, with obtain improve TDCNN-LSTM network model (step S205,
The structure for improving TDCNN-LSTM network model is as shown in Figure 3).In this case, by TDCNN-LSTM network model
Repetition training and adjustment, the precision of prediction of TDCNN-LSTM network model can be continuously improved, so that it is guaranteed that improve
The numerical reliability of TDCNN-LSTM network model during the test promotes the precision of its measurement.
In addition, in the present embodiment, in step S203, the judge index of precision of prediction is not particularly limited.One
In a little examples, the judge index of precision of prediction can be missed for the average absolute of real tool attrition value and prediction tool wear value
Poor (MAE) and root-mean-square error (RMSE).In other examples, the calculation formula of MAE and RMSE be can be such that
In formula, y_test is real tool attrition value, and y_pre is the prediction tool wear value in test sample, and n is to survey
The quantity of sample sheet.
In this case, judge index uses the combination of mean absolute error and root-mean-square error, can further mention
The validity and reliability for rising judging result, to promote the precision of prediction of TDCNN-LSTM network model.
In addition, in the present embodiment, in step s 304, the adjustment to the parameter of TDCNN-LSTM network model can
To include: the adjustment of size, the adjustment of quantity pond size of convolution kernel, the adjustment of LSTM output dimension, Quan Lian of convolution kernel
Connect the adjustment of the quantity of layer neuron, the adjustment of batch size (batch size), the adjustment of training total degree and model optimizer
Adjustment etc..Thus, it is possible to many-sided parameter for improving TDCNN-LSTM network model, improve its measurement accuracy and can
By property.
In the present embodiment, in step s 106, testing data is obtained, testing data is inputted after pretreatment and is changed
Into TDCNN-LSTM network model, to obtain corresponding target tool attrition value.Wherein, the type of testing data is not special
Limitation.In some instances, testing data can be vibration data and current data.In other examples, vibration data and
The quantity of current data can be respectively more than one, such as can be respectively one, three or five etc..As a result, by more
The combination of kind or multiple data, can be improved the accuracy of prediction result.
In addition, in the present embodiment, in step s 106, the acquisition mode of vibration data and current data is without spy
It does not limit.In some instances, vibration data can be by being installed on electric machine main shaft and close to the vibrating sensor of monitoring cutter
It measures, current data can be measured by the current transformer for being installed on the delivery outlet of motor driver.In other examples,
Vibration data can be measured on electric machine main shaft and close to the three-way vibration sensor of monitoring cutter by being installed on, and current data can
It is measured with three current transformers by be installed on motor driver three delivery outlets.Pass through a variety of or multiple data as a result,
Combination, the accuracy of prediction result can be improved.
As described above, in step s 106, the step of preferred embodiment is as follows: firstly, three-way vibration is sensed
Device is mounted on electric machine main shaft and close to the place of monitoring cutter, and three current transformers are mounted on the three of motor driver
A delivery outlet；Then, in the process of monitoring cutter, per the 6 channels sensing at regular intervals, acquiring certain length
Device signal is as input data X={ x1,x2,…,xL, wherein L is the length of input sample, the input x at per momentiIt is one
A includes the vector of 6 elements.
In addition, in the present embodiment, in step s 106, not limiting especially the pretreated form of testing data
System.In some instances, pretreatment can be to be standardized using Z-score method to input data.In such case
Under, the validity of input sample can be promoted by the standardization to input data, to improve the cutter mill of prediction
The precision of damage value.
As described above, in step s 106, preferred pretreated method can be logical to 6 using Z-score method
Road signal is standardized respectively, and the method for specific standards is as follows:
In formula, xjIt is the time series of j-th of sensor passage, μjAnd σjIt is xjAverage and standard deviation,It is Z-
Time series data after score normalization.
Fig. 4 is the structural schematic diagram of the monitoring system of tool abrasion involved in embodiments of the present invention；
Fig. 5 is the structural schematic diagram of the monitoring device of tool abrasion involved in embodiments of the present invention.Join below
Fig. 4 and Fig. 5 are examined, describes the monitoring system and equipment of tool abrasion involved in present embodiment in detail.
In the present embodiment, as shown in figure 4, the monitoring system of tool abrasion may include: data acquisition module
401, for acquiring the input data of monitoring cutter；Preprocessing module 402, it is defeated for being pre-processed to obtain to input data
Enter sample, and obtain the corresponding real tool attrition value of input sample, generates training sample and test sample；Model construction mould
Block 403, for constructing TDCNN-LSTM network model comprising: simultaneously its network parameter, TDCNN is arranged in building TDCNN network
Network is used to extract the local spatial feature of input sample；And it constructs LSTM network and its network parameter, LSTM net is set
Network is used to extract the changing character of input sample；Training and adjustment module 404, for according to training sample and test specimens
This, training and the network parameter for adjusting TDCNN-LSTM network model, to obtain improving TDCNN-LSTM network model；And
Testing data is inputted after pretreatment for obtaining testing data and improves TDCNN-LSTM network model by test module 405,
To obtain corresponding target tool attrition value.
In addition, can also include and model structure in the monitoring system of tool abrasion involved in the present embodiment
The connected data segmentation module (not shown) of block 403 is modeled, for input sample to be divided into N number of subsequence along time dimension.
In the present embodiment, it as shown in figure 5, the monitoring device of tool abrasion may include: memory 501, is used for
Store computer program；And processor 502, realized when for executing computer program as above it is any it is described based on
The step of tool wear quantity monitoring method of TDCNN-LSTM.
In addition, in the present embodiment, the monitoring device of tool abrasion can also include: to be connected with processor 502
The input interface (not shown) of computer program, parameter or instruction for obtaining external importing etc., is connected with processor 502
And the display unit (not shown) of the data sent for video-stream processor 502, what is be connected with processor 502 is used for and outside
The building blocks such as the network port (not shown) that each terminal device is communicatively coupled.Thus, it is possible to promote the reality of monitoring device
With the reliability of property and monitoring.
Present embodiment discloses a kind of computer readable storage medium, in the present embodiment, computer-readable storage medium
Computer program is can store in matter, computer program can realize as above any knife when being executed by processor
Has the step of monitoring method of abrasion loss.It will be appreciated by those skilled in the art that all or part side in above embodiment
Method step can control relevant executable instruction by computer program and complete, which can store
In computer readable storage medium, storage medium includes read-only memory (Read-Only Memory, ROM), random storage
Device (Random Access Memory, RAM), programmable read only memory (Programmable Read-Only Memory,
PROM), programmable read only memory (Programmable Read-Only Memory, PROM), erasable programmable are read-only
Memory (Erasable Programmable Read-Only Memory, EPROM), disposable programmable read-only memory
(One-Time Programmable Read-Only Memory, OTPROM), the electronics formula of erasing can make carbon copies read-only memory
(Electrically-Erasable Programmable Read-Only Memory, EEPROM), CD-ROM (Compact
Disc Read-Only Memory, CD-ROM) or other disc memories, magnetic disk storage, magnetic tape storage, or can use
In carrying or computer-readable any other medium, such as USB flash disk, mobile hard disk of storing data etc..
It should be noted that being stated that a series of for simple description for each method example above-mentioned
Combination of actions, but those skilled in the art should understand that, the present invention is not limited by the sequence of acts described, because
According to the application, some steps may be performed in other sequences or simultaneously.
In addition, method and step according to the present invention can according to actual needs the adjustment of carry out sequence, merge or/and delete
Subtract, the modular unit in system according to the present invention can also be merged according to actual needs, divides or/and be deleted.
The foregoing is merely better embodiments of the invention, are not intended to limit the invention, all of the invention
Spirit and principle within, any modification, equivalent replacement, improvement and so on, should be included in protection scope of the present invention it
It is interior.
Claims (10)
1. a kind of monitoring method of tool abrasion characterized by comprising
The input data of acquisition monitoring cutter；
The input data is pre-processed to obtain input sample, and obtains the corresponding real tool abrasion of the input sample
Value generates training sample and test sample；
Construct TDCNN-LSTM network model comprising: simultaneously its network parameter, the TDCNN network is arranged in building TDCNN network
For extracting the local spatial feature of the input sample；And it constructs LSTM network and its network parameter, the LSTM is set
Network is used to extract the changing character of the input sample；
According to the training sample and the test sample, training and the network for adjusting the TDCNN-LSTM network model
Parameter, to obtain improving TDCNN-LSTM network model；And
Testing data is obtained, the testing data is inputted after pretreatment and improves TDCNN-LSTM network model, to obtain phase
The target tool attrition value answered.
2. the monitoring method of tool abrasion according to claim 1, which is characterized in that the TDCNN network includes N number of
Local spatial feature extractor, the local spatial feature extractor include more than two convolutional coding structures, wherein each convolution knot
Structure includes a convolutional layer, one batch of standardization layer, an active coating and an average pond layer.
3. the monitoring method of tool abrasion according to claim 1, which is characterized in that the LSTM network be two layers with
On, and be multiple-input and multiple-output structure.
4. the monitoring method of tool abrasion according to claim 1, which is characterized in that the input data includes:
Vibration data is the vibration signal of monitoring cutter, by being installed on electric machine main shaft and close to the vibration of the monitoring cutter
Dynamic sensor measures；And
Current data is surveyed for the current signal of monitoring cutter by being installed on the current transformer of delivery outlet of motor driver
?.
5. the monitoring method of tool abrasion according to claim 1, which is characterized in that the real tool attrition value is logical
The tool of being measured microscopically is crossed to measure.
6. the monitoring method of tool abrasion according to claim 1, which is characterized in that the pretreatment is using Z-
Score method is standardized the input data.
7. the monitoring method of tool abrasion according to claim 1, which is characterized in that the TDCNN-LSTM network
The training of the network parameter of model simultaneously adjusts and includes:
The TDCNN-LSTM network model is trained using the training sample；
The test sample is inputted in trained TDCNN-LSTM network model, prediction tool wear value is obtained；
According to the prediction tool wear value real tool attrition value corresponding with the test sample, described in judgement
Whether the precision of prediction of TDCNN-LSTM network model is greater than or equal to threshold value；
If it is not, being trained again after then adjusting the network parameter of the TDCNN-LSTM network model, after training
The TDCNN-LSTM network model the precision of prediction be greater than or equal to the threshold value, to obtain the improvement TDCNN-
LSTM network model.
8. a kind of monitoring system of tool abrasion characterized by comprising
Data acquisition module, for acquiring the input data of monitoring cutter；
Preprocessing module obtains input sample for being pre-processed to the input data, and obtains the input sample pair
The real tool attrition value answered generates training sample and test sample；
Model construction module, for constructing TDCNN-LSTM network model comprising: simultaneously its network is arranged in building TDCNN network
Parameter, the TDCNN network are used to extract the local spatial feature of the input sample；And it constructs LSTM network and is arranged
Its network parameter, the LSTM network are used to extract the changing character of the input sample；
Training and adjustment module, for according to the training sample and the test sample, training simultaneously to adjust the TDCNN-
The network parameter of LSTM network model, to obtain improving TDCNN-LSTM network model；And
The testing data is inputted after pretreatment for obtaining testing data and improves TDCNN-LSTM network by test module
Model, to obtain corresponding target tool attrition value.
9. a kind of monitoring device of tool abrasion characterized by comprising
Memory, for storing computer program；And
Processor is realized when for executing the computer program such as the tool abrasion as described in any in claim 1 to 7
The step of monitoring method.
10. a kind of computer readable storage medium, which is characterized in that be stored with computer on the computer readable storage medium
Program realizes the prison of the tool abrasion as described in any in claim 1 to 7 when the computer program is executed by processor
The step of survey method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811643774.0A CN109753923A (en) | 2018-12-29 | 2018-12-29 | Method, system, device and computer-readable storage medium for monitoring tool wear amount |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811643774.0A CN109753923A (en) | 2018-12-29 | 2018-12-29 | Method, system, device and computer-readable storage medium for monitoring tool wear amount |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109753923A true CN109753923A (en) | 2019-05-14 |
Family
ID=66404943
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811643774.0A Pending CN109753923A (en) | 2018-12-29 | 2018-12-29 | Method, system, device and computer-readable storage medium for monitoring tool wear amount |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109753923A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110421408A (en) * | 2019-07-02 | 2019-11-08 | 绍兴安迪自动化设备有限公司 | A kind of monitoring method of cutting-tool wear state |
| CN110554657A (en) * | 2019-10-16 | 2019-12-10 | 河北工业大学 | Health diagnosis system and diagnosis method for operation state of numerical control machine tool |
| CN110726505A (en) * | 2019-10-23 | 2020-01-24 | 上海电力高压实业有限公司 | Transmission tower bolt fastening detection method and system based on long-time and short-time memory network |
| CN111300146A (en) * | 2019-11-29 | 2020-06-19 | 上海交通大学 | Numerical control machine tool cutter abrasion loss online prediction method based on spindle current and vibration signal |
| CN111366123A (en) * | 2020-03-06 | 2020-07-03 | 大连理工大学 | A prediction method of part surface roughness and tool wear based on multi-task learning |
| CN111390648A (en) * | 2020-03-31 | 2020-07-10 | 江麓机电集团有限公司 | Turning tool abrasion judging method based on antagonistic neural network |
| CN111633467A (en) * | 2020-05-15 | 2020-09-08 | 大连理工大学 | A tool wear state monitoring method based on one-dimensional depthwise convolutional autoencoder |
| CN111814728A (en) * | 2020-07-22 | 2020-10-23 | 同济大学 | Recognition method and storage medium for tool wear state of CNC machine tools |
| CN112518425A (en) * | 2020-12-10 | 2021-03-19 | 南京航空航天大学 | Intelligent machining cutter wear prediction method based on multi-source sample migration reinforcement learning |
| CN113317791A (en) * | 2021-05-28 | 2021-08-31 | 温州康宁医院股份有限公司 | Method and device for determining severity of depression based on audio frequency of testee |
| CN113496312A (en) * | 2021-07-05 | 2021-10-12 | 山东大学 | Cutter residual life prediction method and equipment based on multi-dimensional feature extraction fusion and long-term and short-term memory network and storage medium |
| CN114048958A (en) * | 2021-10-19 | 2022-02-15 | 上海交通大学 | Tool wear prediction method and system based on deep convolutional residual shrinkage network |
| CN114186617A (en) * | 2021-11-23 | 2022-03-15 | 浙江大学 | A mechanical fault diagnosis method based on distributed deep learning |
| CN114417697A (en) * | 2021-12-07 | 2022-04-29 | 山东大学 | Neural network-based TBM hob abrasion real-time prediction method and system |
| CN114676716A (en) * | 2020-12-09 | 2022-06-28 | 电科云(北京)科技有限公司 | Tool remaining life prediction method, device and medium |
| CN114833636A (en) * | 2022-04-12 | 2022-08-02 | 安徽大学 | Cutter wear monitoring method based on multi-feature space convolution neural network |
| CN115476199A (en) * | 2021-05-31 | 2022-12-16 | 沈阳中科数控技术股份有限公司 | Numerical control machine tool cutter wear prediction method based on time convolution network |
| CN118969230A (en) * | 2024-10-14 | 2024-11-15 | 浙江微盾环保科技有限公司 | Tool monitoring method and system based on non-incineration friction heat medical waste treatment equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107877262A (en) * | 2017-11-13 | 2018-04-06 | 华中科技大学 | A kind of numerical control machine tool wear monitoring 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 |
| US20180210780A1 (en) * | 2011-06-03 | 2018-07-26 | Beet, Llc | Method for generating a machine heartbeat |
| CN108536938A (en) * | 2018-03-29 | 2018-09-14 | 上海交通大学 | A kind of machine tool life prediction system and prediction technique |
| US20180330493A1 (en) * | 2017-04-28 | 2018-11-15 | Mentor Graphics Corporation | Wafer Map Pattern Detection Based On Supervised Machine Learning |
-
2018
- 2018-12-29 CN CN201811643774.0A patent/CN109753923A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180210780A1 (en) * | 2011-06-03 | 2018-07-26 | Beet, Llc | Method for generating a machine heartbeat |
| US20180330493A1 (en) * | 2017-04-28 | 2018-11-15 | Mentor Graphics Corporation | Wafer Map Pattern Detection Based On Supervised Machine Learning |
| CN107877262A (en) * | 2017-11-13 | 2018-04-06 | 华中科技大学 | A kind of numerical control machine tool wear monitoring 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 |
| CN108536938A (en) * | 2018-03-29 | 2018-09-14 | 上海交通大学 | A kind of machine tool life prediction system and prediction technique |
Non-Patent Citations (3)
| Title |
|---|
| HUIHUI QIAO ET AL.: "《A Time-Distributed Spatiotemporal Feature Learning Method for Machine Health Monitoring with Multi-Sensor Time Series》", 《SENSORS》 * |
| RUI ZHAO ET AL.: "《Learning to Monitor Machine Health with Convolutional Bi-Directional LSTM Networks》", 《SENSORS》 * |
| 张存吉 等: "《基于深度学习的刀具磨损监测方法》", 《计算机集成制造系统》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110421408B (en) * | 2019-07-02 | 2021-07-02 | 绍兴安迪自动化设备有限公司 | Method for monitoring wear state of cutter |
| CN110421408A (en) * | 2019-07-02 | 2019-11-08 | 绍兴安迪自动化设备有限公司 | A kind of monitoring method of cutting-tool wear state |
| CN110554657A (en) * | 2019-10-16 | 2019-12-10 | 河北工业大学 | Health diagnosis system and diagnosis method for operation state of numerical control machine tool |
| CN110726505A (en) * | 2019-10-23 | 2020-01-24 | 上海电力高压实业有限公司 | Transmission tower bolt fastening detection method and system based on long-time and short-time memory network |
| CN110726505B (en) * | 2019-10-23 | 2021-10-15 | 上海电力高压实业有限公司 | Transmission tower bolt fastening detection method and system based on long-time and short-time memory network |
| CN111300146A (en) * | 2019-11-29 | 2020-06-19 | 上海交通大学 | Numerical control machine tool cutter abrasion loss online prediction method based on spindle current and vibration signal |
| CN111366123A (en) * | 2020-03-06 | 2020-07-03 | 大连理工大学 | A prediction method of part surface roughness and tool wear based on multi-task learning |
| CN111390648A (en) * | 2020-03-31 | 2020-07-10 | 江麓机电集团有限公司 | Turning tool abrasion judging method based on antagonistic neural network |
| CN111390648B (en) * | 2020-03-31 | 2021-12-28 | 江麓机电集团有限公司 | Turning tool abrasion judging method based on antagonistic neural network |
| CN111633467A (en) * | 2020-05-15 | 2020-09-08 | 大连理工大学 | A tool wear state monitoring method based on one-dimensional depthwise convolutional autoencoder |
| CN111814728A (en) * | 2020-07-22 | 2020-10-23 | 同济大学 | Recognition method and storage medium for tool wear state of CNC machine tools |
| CN114676716A (en) * | 2020-12-09 | 2022-06-28 | 电科云(北京)科技有限公司 | Tool remaining life prediction method, device and medium |
| CN112518425A (en) * | 2020-12-10 | 2021-03-19 | 南京航空航天大学 | Intelligent machining cutter wear prediction method based on multi-source sample migration reinforcement learning |
| CN113317791A (en) * | 2021-05-28 | 2021-08-31 | 温州康宁医院股份有限公司 | Method and device for determining severity of depression based on audio frequency of testee |
| CN113317791B (en) * | 2021-05-28 | 2023-03-14 | 温州康宁医院股份有限公司 | Method and device for determining severity of depression based on audio frequency of testee |
| CN115476199A (en) * | 2021-05-31 | 2022-12-16 | 沈阳中科数控技术股份有限公司 | Numerical control machine tool cutter wear prediction method based on time convolution network |
| CN113496312A (en) * | 2021-07-05 | 2021-10-12 | 山东大学 | Cutter residual life prediction method and equipment based on multi-dimensional feature extraction fusion and long-term and short-term memory network and storage medium |
| CN114048958A (en) * | 2021-10-19 | 2022-02-15 | 上海交通大学 | Tool wear prediction method and system based on deep convolutional residual shrinkage network |
| CN114186617A (en) * | 2021-11-23 | 2022-03-15 | 浙江大学 | A mechanical fault diagnosis method based on distributed deep learning |
| CN114417697A (en) * | 2021-12-07 | 2022-04-29 | 山东大学 | Neural network-based TBM hob abrasion real-time prediction method and system |
| CN114833636A (en) * | 2022-04-12 | 2022-08-02 | 安徽大学 | Cutter wear monitoring method based on multi-feature space convolution neural network |
| CN114833636B (en) * | 2022-04-12 | 2023-02-28 | 安徽大学 | Cutter wear monitoring method based on multi-feature space convolution neural network |
| CN118969230A (en) * | 2024-10-14 | 2024-11-15 | 浙江微盾环保科技有限公司 | Tool monitoring method and system based on non-incineration friction heat medical waste treatment equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109753923A (en) | Method, system, device and computer-readable storage medium for monitoring tool wear amount | |
| CN112508105B (en) | Fault detection and retrieval method for oil extraction machine | |
| WO2021004154A1 (en) | Method for predicting remaining life of numerical control machine tool | |
| CN112034789A (en) | Health assessment method, system and assessment terminal for key parts and complete machine of numerical control machine tool | |
| CN112735094A (en) | Geological disaster prediction method and device based on machine learning and electronic equipment | |
| CN116070527B (en) | Prediction Method of Remaining Life of Milling Tool Based on Degradation Model | |
| CN113762182A (en) | Milling cutter wear state prediction method and device based on depth network self-adaptation | |
| CN118655488B (en) | Power supply failure detection system and method | |
| CN119067225B (en) | Industrial control anomaly interpretation method and system based on difference of generated counterfactual samples | |
| KR102778651B1 (en) | System and method for measuring the thickness of a multilayer thin film structure | |
| CN118782172A (en) | A water environment detection method and device | |
| CN106874251A (en) | A kind of measuring method of rock mass RBI and RQD indexs | |
| CN117471922B (en) | Intelligent control method and system for oil casing electric punching equipment | |
| CN117892192A (en) | Data-driven-based flow prediction method for onboard hydraulic system | |
| CN114581697B (en) | A method for monitoring the fluid supply capacity of oil wells based on dynamometer diagram | |
| Wang et al. | Rapid pattern recognition of electric submersible pump ammeter card based on artificial neural network | |
| CN119989082B (en) | Hoisting machinery fault diagnosis method based on multi-element self-adaptive singular spectrum analysis | |
| CN119004275B (en) | A detection method for data line production | |
| CN119845365B (en) | Cable channel fault location method and system based on distributed optical fiber sensing | |
| CN117391146B (en) | Experiment model construction method and system based on neural network | |
| CN119559564B (en) | Power transmission line icing analysis method and system based on image recognition | |
| CN119413077B (en) | An audio control panel installation device and a fixing performance detection method thereof | |
| CN119598301B (en) | A tunnel disease monitoring data enhancement method and system based on CTGAN | |
| CN119168022B (en) | AI-based intelligent data analysis method, system and storage medium | |
| CN119200009A (en) | An intelligent prediction method for tunnel groundwater conditions |
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: 20190514 |
|
| RJ01 | Rejection of invention patent application after publication |