CN111126255A - Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm - Google Patents
Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm Download PDFInfo
- Publication number
- CN111126255A CN111126255A CN201911335398.3A CN201911335398A CN111126255A CN 111126255 A CN111126255 A CN 111126255A CN 201911335398 A CN201911335398 A CN 201911335398A CN 111126255 A CN111126255 A CN 111126255A
- Authority
- CN
- China
- Prior art keywords
- numerical control
- control machine
- machine tool
- value
- deep learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23Q—DETAILS, COMPONENTS, OR ACCESSORIES FOR MACHINE TOOLS, e.g. ARRANGEMENTS FOR COPYING OR CONTROLLING; MACHINE TOOLS IN GENERAL CHARACTERISED BY THE CONSTRUCTION OF PARTICULAR DETAILS OR COMPONENTS; COMBINATIONS OR ASSOCIATIONS OF METAL-WORKING MACHINES, NOT DIRECTED TO A PARTICULAR RESULT
- B23Q17/00—Arrangements for observing, indicating or measuring on machine tools
- B23Q17/09—Arrangements for observing, indicating or measuring on machine tools for indicating or measuring cutting pressure or for determining cutting-tool condition, e.g. cutting ability, load on tool
- B23Q17/0952—Arrangements for observing, indicating or measuring on machine tools for indicating or measuring cutting pressure or for determining cutting-tool condition, e.g. cutting ability, load on tool during machining
- B23Q17/0957—Detection of tool breakage
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The invention belongs to the technical field of numerical control machine tool cutter state monitoring and discloses a numerical control machine tool cutter wear value prediction method based on a deep learning regression algorithm, which comprises the following steps: (1) collecting three-dimensional vibration signals when the numerical control machine tool performs cutting machining; (2) respectively inputting vibration signals into a stack type sparse automatic encoder network for training, and outputting the last coding vector obtained by each training as a feature obtained by self-adaptive extraction; (3) inputting the features into a non-linear regression function; (4) calculating the difference between the obtained predicted value and the actual value of the cutter abrasion, further finely adjusting the stack type sparse automatic encoder network, judging whether the iteration number reaches a threshold value, and if not, turning to the step (2); otherwise, finishing the training, thus obtaining a deep learning regression algorithm model and further predicting the wear value of the numerical control machine tool in real time. The invention improves the real-time performance and the prediction precision.
Description
Technical Field
The invention belongs to the technical field related to monitoring of the state of a numerical control machine tool cutter, and particularly relates to a numerical control machine tool cutter wear value prediction method based on a deep learning regression algorithm.
Background
The tool wear monitoring of the numerical control machine tool means that in the product processing process, a computer judges and predicts whether a tool is worn or not by detecting signal changes of various sensors, the tool wear monitoring process is a pattern recognition process essentially, and a complete tool wear monitoring system function comprises a research object (tool), a processing condition, a sensor network, signal processing, feature extraction, pattern recognition and the like.
The inevitable abrasion of the cutter in the machining process can directly affect the utilization rate of a machine tool and the quality of a workpiece, so that the quality of the machined workpiece is reduced by a light person, the workpiece is scrapped by a heavy person, and some mechanical parts can be damaged even under extreme conditions. Therefore, it is necessary to quickly detect the wear state of the tool in real time during machining.
After decades of development, the monitoring technology of the cutter has reached a certain level in the aspects of breadth and depth, but at present, no method which can be applied to different processing conditions and can detect the abrasion of various cutters exists, the application range of the existing various methods is limited, the requirements of automation and intellectualization are not met, and certain limitations exist in the aspect of practical application, such as the problems of insufficient monitoring real-time performance, missing report and misinformation.
Disclosure of Invention
Aiming at the defects or improvement requirements of the prior art, the invention provides a numerical control machine tool cutter wear value prediction method based on a deep learning regression algorithm, which can get rid of dependence on a large number of signal processing technologies and diagnosis experiences, realize self-adaptive extraction of sensitivity characteristics capable of representing cutter wear, and simultaneously combine the sensitivity characteristics with a nonlinear regression function, can effectively realize real-time prediction on the cutter wear value, has the advantages of strong identification real-time performance, high prediction precision, stable generalization capability and the like, and can quickly, accurately and real-time identify and predict the cutter wear value in the numerical control machine tool machining process. Specifically, the prediction method comprises the steps of firstly collecting three-way vibration signals of an acceleration sensor arranged on a spindle seat when a numerical control machine tool actually processes a die, observing the rear tool face of the tool by using a microscope after each workpiece is processed, recording the actual wear value of the tool, normalizing the vibration signals, improving an activation function in a sparse automatic encoder due to the characteristics of industrial signals, improving a loss function of a stacked sparse automatic encoder, inputting the vibration signals after normalization processing into the improved stacked sparse automatic encoder for training, extracting the last sparse vector as the characteristic of the vibration signals, inputting the characteristic into a nonlinear regression function, and comparing the obtained prediction with the actual wear value to obtain an error value, reversely fine-tuning the stacked sparse automatic encoder model based on the obtained error value, and repeatedly iterating and updating in such a way until the deep learning regression algorithm training is completed; finally, the method can realize the real-time prediction of the tool wear value, so that the invention improves the deep learning loss function, and simultaneously, the combination with the nonlinear regression algorithm can stably and accurately predict the wear value of the numerical control machine tool in real time, and the method has the advantages of strong identification real-time performance, high prediction precision, stable generalization capability and the like.
In order to achieve the above object, according to one aspect of the present invention, there is provided a method for predicting a wear value of a cutting tool of a numerical control machine based on a deep learning regression algorithm, the method mainly comprising the steps of:
(1) collecting three-way vibration signals when a numerical control machine tool performs cutting machining, wherein each machined workpiece corresponds to one group of vibration signals, and meanwhile, each machined workpiece records the actual value of the tool abrasion at the moment; then, all the obtained vibration signals are subjected to normalization processing;
(2) respectively inputting the obtained vibration signals into a stack type sparse automatic encoder network for training, and outputting the last coding vector obtained by each training as a feature obtained by self-adaptive extraction, thereby completing the feature self-adaptive extraction and obtaining a plurality of features;
(3) inputting the obtained characteristics into a nonlinear regression function to obtain a predicted value of the tool wear value, wherein the nonlinear regression function is as follows:
wherein A is a parameter, B is a matrix, C is a vector, and Y ismFeatures extracted for adaptation;
(4) calculating the difference between the obtained predicted value and the actual value of the cutter abrasion, finely adjusting the stack type sparse automatic encoder network based on the obtained difference value, simultaneously judging whether the current iteration number reaches an iteration number threshold value, and if not, turning to the step (2); otherwise, completing network training of the stack type sparse automatic encoder, obtaining a deep learning regression algorithm model, and predicting the wear value of the numerical control machine tool in real time based on the deep learning regression algorithm model and a vibration signal in the machining process of the numerical control machine tool.
Further, the threshold number of iterations is 500.
Further, the vibration signal contains processing information in the whole life cycle of the cutter; the number of vibration signals corresponds to the number of workpieces to be machined.
Further, the step (2) includes the following sub-steps:
(21) inputting each vibration signal after normalization processing into a stack type sparse automatic encoder network as an input signal, and encoding the input signal by using an encoding layer to obtain an encoding vector;
(22) inputting the coding vector into a decoder of a stack type sparse automatic encoder network for decoding, wherein the obtained output is reconstruction data;
(23) and (3) substituting the vibration signal and the reconstruction data after the normalization processing into a loss function to obtain a loss value, and repeating the step (21) and the step (22) until the loss value is smaller than a preset value, so as to finish the training of the stack type sparse automatic encoder network.
Further, the code vector is represented by hm,iExpressed, its expression is:
hm,i＝σ(υm,i)＝sf(Wυm,i+b)
wherein σ is a coding function; upsilon ism,iM is a vibration signal after the mth normalization processing and smoothing processing, M is more than or equal to 1 and less than or equal to M, M is the number of the processing workpieces, and M is the number of the processing workpieces; sfAn activation function for the coding network; w, b are neural network coding weights.
Further, v 'for data was reconstructed'm,iExpressed, the calculation formula is:
υ'm,i＝σ'(hm,i)＝sg(W'hm,i+d)
where σ' is the decoding function, sgAnd W' and d are the decoding weights of the neural network.
Further, the loss function is expressed as L' (upsilon)m,i,υ'm,i) Expressed, the calculation formula is:
where i is the vector dimension.
Further, the difference in step (4) is expressed by Error, and the calculation formula is as follows:
in the formula, VmThe actual tool wear value is obtained; a is parameter, B is matrix, C is vector, YmFeatures are extracted for adaptation.
Generally, compared with the prior art, the numerical control machine tool wear value prediction method based on the deep learning regression algorithm provided by the invention has the following beneficial effects:
1. the obtained vibration signals are respectively input into a stack type sparse automatic encoder network for training, and the last coding vector obtained by each training is used as the feature output obtained by self-adaptive extraction, so that the feature self-adaptive extraction is completed, the dependence on a large number of signal processing technologies and diagnosis experience is eliminated, the high-level features hidden in the signals corresponding to the cutter state can be extracted in an unsupervised and self-adaptive manner, the reliability and the robustness of the monitoring system are high, and the precision and the efficiency are improved.
2. The self-adaptive extraction of the sensitivity characteristic representing the tool abrasion is combined with the nonlinear regression function, the abrasion value of the tool can be effectively predicted in real time, the method has the advantages of strong identification real-time performance, high prediction precision, stable generalization capability and the like, and the abrasion value of the tool in the machining process of the numerical control machine can be quickly, accurately and in real time identified and predicted.
3. And inputting each vibration signal after normalization and smoothing treatment into a stack type sparse automatic encoder network as an input signal, and encoding the input signal by using an encoding layer to obtain an encoding vector, so that input data of a high-dimensional space is converted into an encoding vector of a low-dimensional space, namely, dimension reduction is realized.
4. Under the condition of monitoring the actual cutter wear value, the method has higher efficiency and less manual interference degree, and can accurately and quickly realize the monitoring of the cutter wear state and the prediction of the cutter wear value by improving and perfecting the self-learning, self-organizing, self-adapting, self-decision making and self-diagnosis capabilities of the intelligent monitoring system, thereby well meeting the requirements of actual production.
Drawings
FIG. 1 is a schematic flow chart of a method for predicting a wear value of a tool of a numerically-controlled machine tool based on a deep learning regression algorithm according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of vibration signal acquisition and preprocessing processes involved in the numerical control machine tool wear value prediction method based on the deep learning regression algorithm in FIG. 1;
FIG. 3 is a graph of vibration signal amplitude for cutter wear values of 0um and 1 um;
FIG. 4 is a graph of vibration signal amplitude for cutter wear values of 3um and 5 um;
FIGS. 5(a) and (b) are schematic diagrams of an improved stacked sparse autoencoder network and a single autoencoder, respectively;
in fig. 6, (a) to (d) are edge diagrams corresponding to the tool wear values of 0um, 1um, 3um, and 5um, respectively;
fig. 7 and 8 are graphs of prediction results obtained by predicting the tool life cycle data after model training is completed, respectively.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The invention provides a numerical control machine tool wear value prediction method based on a deep learning regression algorithm, which is characterized in that a feature vector of a signal is extracted by acquiring a three-dimensional vibration signal of a tool machining full life cycle and using an improved stack type sparse automatic encoder algorithm in a deep learning theory; the feature vector is then input into a non-linear regression function to obtain a specific value that can be used to characterize the actual wear value of the tool. Because the extracted depth features contain a lot of information, and the information does not have specific physical significance, the method can monitor the micro-abrasion of the cutter in the machining process. In order to verify the method, the vibration signals of the continuous machining of the cutter on the actual die production line are collected, and the result obtained after processing shows that the parameters can well represent the wear state of the cutter. In addition, the change condition of the tool wear value has high correlation with the features extracted by the deep learning regression algorithm, so that the intelligent online monitoring and prediction of the tool wear value can be realized by establishing a tool wear value real-time prediction model.
According to the method, a stack type sparse self-encoder network structure is built, input data of a high-dimensional space is converted into a coding vector of a low-dimensional space through a coding network, the coding vector of the low-dimensional space is reconstructed back to original input data through a decoding network, and bottom layer characteristics are combined to form more abstract high-level characteristics.
The numerical control machine tool wear value prediction method based on the deep learning regression algorithm mainly comprises the following steps:
firstly, preprocessing a signal to obtain a sample set.
Collecting three-way vibration signals when a numerical control machine tool performs cutting machining, wherein each machined workpiece corresponds to one group of vibration signals, and meanwhile, each machined workpiece records the actual value of the tool abrasion at the moment; subsequently, all the obtained vibration signals are normalized.
Specifically, an acceleration sensor is used for acquiring three-way vibration signals in the high-precision machining process of the numerical control machine tool, the vibration signals contain machining information of all tools in the full life cycle (namely, from the use of a new tool to the severe wear of the tool), and each tool can machine a plurality of workpieces, such as M workpieces, in the full life cycle, so that the vibration signals corresponding to the full machining cycle of the tool to be monitored are intercepted from the vibration signals, and then the vibration signals are subjected to smoothing treatment, meanwhile, the number of the intercepted vibration signals is consistent with the number of the machined workpieces, for example, the corresponding tools can machine M workpieces in the full life cycle, and then M vibration signals are provided, each signal corresponds to the machining process of the corresponding machined workpiece, and each machined workpiece also corresponds to an actual tool wear value.
After the vibration signals are subjected to smoothing processing, in order to ensure that the amplitude of the vibration signals is kept within a specific range so as to eliminate interference caused by noise in the acquisition process, normalization processing is performed on each vibration signal, each obtained vibration signal is a sample, and all vibration signals form a sample set.
And step two, pre-training a deep learning algorithm.
And respectively inputting the obtained vibration signals into a stack type sparse automatic encoder network in a deep learning network for training, and outputting the last coding vector obtained by each training as a feature obtained by self-adaptive extraction, thereby completing the feature self-adaptive extraction and obtaining a plurality of features.
Specifically, in order to input each data into a deep learning network for training, each vibration signal obtained through normalization and smoothing is respectively input into a stack-type sparse automatic encoder network for training, and the last coding vector obtained through each training is output as a feature obtained through adaptive extraction, namely, a first vibration signal is input into the stack-type sparse automatic encoder network for training to output a feature vector, a second vibration signal is input into the stack-type sparse automatic encoder network for training to output a feature, and by analogy, M features can be output in total, and each feature is composed of a vector. The method comprises the following specific steps:
and (2.1) inputting each vibration signal after normalization and smoothing as an input signal to a stack type sparse automatic encoder network, and encoding the input signal by using an encoding layer to obtain an encoding vector, so that input data of a high-dimensional space is converted into an encoding vector of a low-dimensional space, namely, dimension reduction is realized.
And (2.2) inputting the coding vector into a decoder of the stack type sparse automatic encoder network for decoding, wherein the obtained output is reconstruction data.
And (2.3) substituting the vibration signal and the reconstruction data after the normalization processing into a loss function to obtain a loss value, and repeating the step (2.1) and the step (2.2) until the loss value is smaller than a preset value or the iteration number reaches the preset value, so as to finish the training of the stack type sparse automatic encoder network. Wherein, the preset value can be set according to the actual requirement, the smaller the preset value is, the closer the reconstructed data is to the original data, or when the iteration number reaches the upper limit, the training is automatically stopped, namely, the original signal is subjected to layer-by-layer feature transformation through multiple times of coding and decoding (namely, dimension reduction step by step), the feature representation of the sample in the original space is transformed to a new feature space, the hierarchical feature representation is obtained through automatic learning, through the final output and the original input contrast error (namely the loss function value), the whole network parameter is gradually adjusted, the loss function value is minimized until a good training result is achieved, the smaller the loss function value is, the better the low-dimensional coding vector can represent the original high-dimensional vector, the loss function value is reduced along with the increase of the training times, and the training times are taken to be 500 times in the embodiment, namely the preset value of the iteration times is 500.
In this embodiment, h is used for a coded vectorm,iExpressed, its expression is:
hm,i＝σ(υm,i)＝sf(Wυm,i+b)
where σ is the encoding function, vm,iM is a vibration signal after the mth normalization processing and smoothing processing, M is more than or equal to 1 and less than or equal to M, M is the number of the processing workpieces, and M is the number of the processing workpieces; sfFor the activation function of the coding network, W, b is the neural network coding weight. Wherein the content of the first and second substances,
Xm,iis the original vibration signal.
Reconstructed data was [ v'm,iExpressed, the calculation formula is:
υ'm,i＝σ'(hm,i)＝sg(W'hm,i+d)
where σ' is the decoding function, sgAnd W' and d are the decoding weights of the neural network.
The loss function is expressed as L' (upsilon)m,i,υ'm,i) Expressed, the calculation formula is:
where i is the vector dimension.
And step three, nonlinear regression function.
Inputting the obtained characteristics into a nonlinear regression function to obtain a predicted value of the tool wear value, wherein the nonlinear regression function is as follows:
wherein A is a parameter, B is a matrix, C is a vector, and Y ismFeatures extracted for adaptation.
Specifically, once pre-training is completed, each vibration signal will obtain a corresponding feature vector, and inputting the feature vector into the non-linear regression function will obtain a specific predicted value of tool wear, which is f (Y)m) Represents:
wherein A is a parameter, B is a matrix, C is a vector, and Y ismFeatures are extracted for adaptation.
And an Error exists between the predicted value and the actual tool wear value, which is expressed by Error, and the following conditions are met:
in the formula, VmThe actual tool wear value is obtained.
Step four, reverse fine adjustment.
Calculating the difference between the obtained predicted value and the actual value of the cutter abrasion, finely adjusting the stack type sparse automatic encoder network based on the obtained difference value, simultaneously judging whether the current iteration number reaches an iteration number threshold value, and if not, turning to the step two; otherwise, completing the network training of the stack type sparse automatic encoder, obtaining a deep learning regression algorithm model, and predicting the abrasion value of the numerical control machine tool in real time based on the deep learning regression algorithm model and a vibration signal in the machining process of the numerical control machine tool so that an operator can perform timely tool compensation and tool changing operation.
Referring to fig. 1, fig. 2, fig. 3, fig. 4, fig. 5, fig. 6, fig. 7 and fig. 8, the following takes the tool wear value prediction in the mold processing production line as an example, and the method for predicting the tool wear value of the numerical control machine based on the deep learning regression algorithm provided by the embodiment of the present invention mainly includes the following steps:
(1) an acceleration sensor is arranged on a spindle seat of a numerical control machine tool, three-way vibration signal data are collected, the vibration signals in the whole life cycle machining process of each tool are respectively subjected to smoothing processing, then each vibration signal is respectively subjected to normalization processing, each vibration signal is taken as a sample and is used as the input of a deep learning regression model, each vibration signal corresponds to an actual wear value of a tool and is used as the basis of reverse fine tuning when the vibration signals are output by the model, and the vibration signal amplitude values when the tool is worn by 0um, 1um, 3um and 5um are shown in fig. 3 and 4.
(2) Inputting the smoothed and normalized vibration signal into a stack-type sparse automatic encoder network for training, as shown in fig. 5, the training steps of the stack-type sparse automatic encoder network are as follows:
given a labeled preprocessed vibration signal sample set
The stack type sparse automatic encoder network enables each vibration signal sample to be upsilon through an encoding function sigmam,iTransformed into a coded vector hm,iAnd M is the number of the processed workpieces:
hm,i＝σ(υm,i)＝sf(Wυm,i+b)
in the formula, sfAn activation function for the coding network; w, b is the neural network coding weight;
then encodes vector hm,iInverse transformation into upsilon by a decoding function sigmam,iIs reconstructed to represent upsilon'm,i：
υ'm,i＝σ'(hm,i)＝sg(W'hm,i+d)
Where σ' is the decoding function, sgAnd W' and d are the decoding weights of the neural network. Wherein the improved encoding and decoding function epsilon (X), and
by minimizing upsilonm,iAnd υ'm,iLoss function L' (upsilon)m,i,υ'm,i) The training of the whole stack type sparse automatic encoder network is completed by repeating encoding and decoding for multiple times to enable the loss value to be smaller than a preset value or reach the upper limit of iteration times, the encoding vector obtained at the last time is taken as a feature to be output, each vibration signal can obtain one feature, each feature is one vector, and M feature vectors can be obtained in total.
Wherein, L' (upsilon)m,i,υ'm,i) Represented by the following formula:
where i is the vector dimension and the reconstruction error is small enough to represent the code-passing vector hm,iCan represent the original vector vm,i；L'(υm,i,υ'm,i) The loss function is gradually decreased with the increase of the training times, and the training times can be limited according to practical limits, such as 500 times of training.
(3) Once the pre-training is completed, each vibration signal obtains a corresponding feature vector, the feature vector is input into a nonlinear regression function, and a specific prediction value is obtained, wherein the specific value is f (Y)m) Indicates that is full ofThe following conditions are satisfied:
wherein A is a parameter, B is a matrix, C is a vector, and Y ismFeatures are extracted for adaptation.
And the predicted value has an Error with the actual tool wear value, which is expressed by Error, and the following conditions are met:
in the formula, VmThe actual tool wear value is obtained.
(4) And (4) carrying out stack type sparse self-encoder network model parameter fine adjustment based on the error, and repeating the steps until the upper limit of iteration times is reached, so as to obtain a deep learning regression algorithm model, thereby realizing the real-time prediction of the cutter wear value.
And inputting a group of new vibration signals of unknown tool states in the whole life cycle into the model according to the trained model so as to predict the actual wear value of the tool, thus realizing the prediction of tool wear. In the prediction process, due to the generalization and stability of the model, the prediction error of the model is kept within 1um, so that the tool can be compensated according to the predicted wear value, if a certain vibration signal is subjected to the predicted wear value of 3um, a reaction can be given at the moment, and the tool is compensated by 3um, so that the machining precision is ensured.
Specifically, aiming at milling on a die production line in a high-precision machining process, an adopted end milling cutter is predicted by model training according to the machining process requirement of a part and the performance of the milling cutter, and a specific cutter wear value can be obtained by giving a vibration signal.
FIGS. 6(a) - (d) are edge views of the tool with wear values of 0um, 1um, 3um, 5um, and this significant difference is found; FIGS. 5(a) - (b) correspond to the amplitude of the vibration signal at the moment, respectively, and there is no obvious difference in the magnitudes; FIGS. 7 and 8 are graphs of prediction results obtained by predicting the tool life cycle data after model training is completed, respectively; according to the graph 7 and the graph 8, in the processing process, when the abrasion value of the cutter is 0um, the cutter belongs to a normal state, the abrasion degree is continuously increased along with the continuous processing of the cutter, when the abrasion value reaches 6-7 um, the cutter enters a serious abrasion state, and at the moment, the machine tool needs to give an alarm to prompt the cutter changing operation according to the condition; by utilizing a deep learning regression algorithm, vibration signals on the spindle seat are collected in real time and trained, so that the wear value condition of the tool at the moment can be predicted in real time, and a machine tool is reminded to perform tool changing operation.
In summary, after the method provided by the present invention adaptively extracts the high-level features of the tool wear state signal, the high-level features are input into the nonlinear regression function for tool wear value prediction analysis, and it is found that the method has great significance for the research of tool big data wear value prediction and monitoring diagnosis, both from the detailed explanation of the tool wear value, the accuracy of the tool wear value prediction, and the generalization and stability of the method in application.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (7)
1. A numerical control machine tool wear value prediction method based on a deep learning regression algorithm is characterized by comprising the following steps:
(1) collecting three-way vibration signals when a numerical control machine tool performs cutting machining, wherein each machined workpiece corresponds to one group of vibration signals, and meanwhile, each machined workpiece records the actual value of the tool abrasion at the moment; then, all the obtained vibration signals are subjected to normalization processing;
(2) respectively inputting the obtained vibration signals into a stack type sparse automatic encoder network for training, and outputting the last coding vector obtained by each training as a feature obtained by self-adaptive extraction, thereby completing the feature self-adaptive extraction and obtaining a plurality of features;
(3) inputting the obtained characteristics into a nonlinear regression function to obtain a predicted value of the tool wear value, wherein the nonlinear regression function is as follows:
wherein A is a parameter, B is a matrix, C is a vector, and Y ismFeatures extracted for adaptation;
(4) calculating the difference between the obtained predicted value and the actual value of the cutter abrasion, finely adjusting the stack type sparse automatic encoder network based on the obtained difference value, simultaneously judging whether the current iteration number reaches an iteration number threshold value, and if not, turning to the step (2); otherwise, completing network training of the stack type sparse automatic encoder, obtaining a deep learning regression algorithm model, and predicting the wear value of the numerical control machine tool in real time based on the deep learning regression algorithm model and a vibration signal in the machining process of the numerical control machine tool.
2. The numerical control machine tool wear value prediction method based on the deep learning regression algorithm as claimed in claim 1, wherein: the vibration signal comprises processing information in the whole life cycle of the cutter; the number of vibration signals corresponds to the number of workpieces to be machined.
3. The numerical control machine tool wear value prediction method based on the deep learning regression algorithm as claimed in claim 1, wherein: the step (2) comprises the following substeps:
(21) inputting each vibration signal after normalization processing into a stack type sparse automatic encoder network as an input signal, and encoding the input signal by using an encoding layer to obtain an encoding vector;
(22) inputting the coding vector into a decoder of a stack type sparse automatic encoder network for decoding, wherein the obtained output is reconstruction data;
(23) and (3) substituting the vibration signal and the reconstruction data after the normalization processing into a loss function to obtain a loss value, and repeating the step (21) and the step (22) until the loss value is smaller than a preset value, so as to finish the training of the stack type sparse automatic encoder network.
4. The numerical control machine tool wear value prediction method based on the deep learning regression algorithm as claimed in claim 3, wherein: h for coding vectorm,iExpressed, its expression is:
hm,i＝σ(υm,i)＝sf(Wυm,i+b)
wherein σ is a coding function; upsilon ism,iM is a vibration signal after the mth normalization processing and smoothing processing, M is more than or equal to 1 and less than or equal to M, M is the number of the processing workpieces, and M is the number of the processing workpieces; sfAn activation function for the coding network; w, b are neural network coding weights.
5. The numerical control machine tool wear value prediction method based on the deep learning regression algorithm as claimed in claim 4, wherein: reconstructed data was [ v'm,iExpressed, the calculation formula is:
υ'm,i＝σ'(hm,i)＝sg(W'hm,i+d)
where σ' is the decoding function, sgAnd W' and d are the decoding weights of the neural network.
6. The numerical control machine tool wear value prediction method based on the deep learning regression algorithm as claimed in claim 5, wherein: the loss function is expressed as L' (upsilon)m,i,υ'm,i) Expressed, the calculation formula is:
where i is the vector dimension.
7. The numerical control machine tool wear value prediction method based on the deep learning regression algorithm as claimed in any one of claims 1 to 6, wherein: the difference value in the step (4) is expressed by Error, and the calculation formula is as follows:
in the formula, VmThe actual tool wear value is obtained; a is parameter, B is matrix, C is vector, YmFeatures are extracted for adaptation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911335398.3A CN111126255A (en) | 2019-12-23 | 2019-12-23 | Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911335398.3A CN111126255A (en) | 2019-12-23 | 2019-12-23 | Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111126255A true CN111126255A (en) | 2020-05-08 |
Family
ID=70501590
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911335398.3A Pending CN111126255A (en) | 2019-12-23 | 2019-12-23 | Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111126255A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111716150A (en) * | 2020-06-30 | 2020-09-29 | 大连理工大学 | Evolution learning method for intelligently monitoring cutter state |
| CN113103068A (en) * | 2021-04-19 | 2021-07-13 | 大连理工大学 | Cutter state monitoring method based on deep migration learning |
| CN113569903A (en) * | 2021-06-09 | 2021-10-29 | 西安电子科技大学 | Method, system, equipment, medium and terminal for predicting abrasion of numerical control machine tool cutter |
| CN114675547A (en) * | 2022-05-30 | 2022-06-28 | 华中科技大学 | MIMO active vibration reduction control method and system with deep learning automatic diagnosis mechanism |
| CN116911176A (en) * | 2023-07-08 | 2023-10-20 | 哈尔滨理工大学 | Terrain trafficability prediction method based on speed and vibration state of wheeled mobile robot |
| CN116911176B (en) * | 2023-07-08 | 2024-04-30 | 哈尔滨理工大学 | Terrain trafficability prediction method based on speed and vibration state of wheeled mobile robot |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105608698A (en) * | 2015-12-25 | 2016-05-25 | 西北工业大学 | Remote image change detection method based on SAE |
| 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 |
| 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 |
| CN108921343A (en) * | 2018-06-26 | 2018-11-30 | 浙江工业大学 | Based on storehouse self-encoding encoder-support vector regression traffic flow forecasting method |
| CN109636061A (en) * | 2018-12-25 | 2019-04-16 | 深圳市南山区人民医院 | Training method, device, equipment and the storage medium of medical insurance Fraud Prediction network |
| CN110561191A (en) * | 2019-07-30 | 2019-12-13 | 西安电子科技大学 | Numerical control machine tool cutter abrasion data processing method based on PCA and self-encoder |
-
2019
- 2019-12-23 CN CN201911335398.3A patent/CN111126255A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105608698A (en) * | 2015-12-25 | 2016-05-25 | 西北工业大学 | Remote image change detection method based on SAE |
| 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 |
| 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 |
| CN108921343A (en) * | 2018-06-26 | 2018-11-30 | 浙江工业大学 | Based on storehouse self-encoding encoder-support vector regression traffic flow forecasting method |
| CN109636061A (en) * | 2018-12-25 | 2019-04-16 | 深圳市南山区人民医院 | Training method, device, equipment and the storage medium of medical insurance Fraud Prediction network |
| CN110561191A (en) * | 2019-07-30 | 2019-12-13 | 西安电子科技大学 | Numerical control machine tool cutter abrasion data processing method based on PCA and self-encoder |
Non-Patent Citations (2)
| Title |
|---|
| 高永彬等: "《Hadoop大数据分析》", 31 July 2019 * |
| 黄曦等: "基于SDAE的航空发动机燃油流量基线模型构建", 《航空发动机》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111716150A (en) * | 2020-06-30 | 2020-09-29 | 大连理工大学 | Evolution learning method for intelligently monitoring cutter state |
| CN113103068A (en) * | 2021-04-19 | 2021-07-13 | 大连理工大学 | Cutter state monitoring method based on deep migration learning |
| CN113103068B (en) * | 2021-04-19 | 2022-03-29 | 大连理工大学 | Cutter state monitoring method based on deep migration learning |
| CN113569903A (en) * | 2021-06-09 | 2021-10-29 | 西安电子科技大学 | Method, system, equipment, medium and terminal for predicting abrasion of numerical control machine tool cutter |
| CN113569903B (en) * | 2021-06-09 | 2024-04-09 | 西安电子科技大学 | Method, system, equipment, medium and terminal for predicting cutter abrasion of numerical control machine tool |
| CN114675547A (en) * | 2022-05-30 | 2022-06-28 | 华中科技大学 | MIMO active vibration reduction control method and system with deep learning automatic diagnosis mechanism |
| CN116911176A (en) * | 2023-07-08 | 2023-10-20 | 哈尔滨理工大学 | Terrain trafficability prediction method based on speed and vibration state of wheeled mobile robot |
| CN116911176B (en) * | 2023-07-08 | 2024-04-30 | 哈尔滨理工大学 | Terrain trafficability prediction method based on speed and vibration state of wheeled mobile robot |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111126255A (en) | Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm | |
| CN107877262B (en) | A kind of numerical control machine tool wear monitoring method based on deep learning | |
| CN111633467B (en) | Cutter wear state monitoring method based on one-dimensional depth convolution automatic encoder | |
| CN111716150B (en) | Evolution learning method for intelligently monitoring cutter state | |
| US20210287098A1 (en) | On-line prediction method of surface roughness of parts based on sdae-dbn algorithm | |
| CN110561191B (en) | Numerical control machine tool cutter abrasion data processing method based on PCA and self-encoder | |
| CN113664612A (en) | Numerical control machine tool milling cutter abrasion real-time monitoring method based on deep convolutional neural network | |
| CN113798920B (en) | Cutter wear state monitoring method based on variational automatic encoder and extreme learning machine | |
| CN113569903B (en) | Method, system, equipment, medium and terminal for predicting cutter abrasion of numerical control machine tool | |
| CN111136509B (en) | Cutter state monitoring method based on ELM-SDAE algorithm | |
| CN114619292B (en) | Milling cutter wear monitoring method based on fusion of wavelet denoising and attention mechanism with GRU network | |
| Ochoa et al. | New approach based on autoencoders to monitor the tool wear condition in HSM | |
| CN113414638B (en) | Variable working condition milling cutter wear state prediction method based on milling force time sequence diagram deep learning | |
| CN112518425B (en) | Intelligent machining cutter wear prediction method based on multi-source sample migration reinforcement learning | |
| Qin et al. | Tool wear identification and prediction method based on stack sparse self-coding network | |
| Oberle et al. | A use case to implement machine learning for life time prediction of manufacturing tools | |
| CN113627544B (en) | Machine tool milling cutter state identification method based on multi-source heterogeneous data fusion | |
| CN113458873B (en) | Method for predicting wear loss and residual life of cutter | |
| CN113043073A (en) | Cutter abrasion and service life prediction method and device | |
| CN111863153A (en) | Method for predicting total amount of suspended solids in wastewater based on data mining | |
| Xiao et al. | Deep learning based modeling for cutting energy consumed in CNC turning process | |
| Dong et al. | Discrete Wavelet Transformation and Genetic Algorithm-Back Propagation Neural Network Applied in Monitoring Woodworking Tool Wear Conditions in the Milling Operation Spindle Power Signals. | |
| CN114357851A (en) | Numerical control milling surface roughness prediction method based on DAE-RNN | |
| CN117341261B (en) | Intelligent control method and system for servo direct-drive screw press | |
| CN112801139B (en) | Intelligent cutter wear state identification method based on heterogeneous domain self-adaptive transfer learning |
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: 20200508 |
|
| RJ01 | Rejection of invention patent application after publication |