CN112801139B - Intelligent cutter wear state identification method based on heterogeneous domain self-adaptive transfer learning - Google Patents
Intelligent cutter wear state identification method based on heterogeneous domain self-adaptive transfer learning Download PDFInfo
- Publication number
- CN112801139B CN112801139B CN202110016625.7A CN202110016625A CN112801139B CN 112801139 B CN112801139 B CN 112801139B CN 202110016625 A CN202110016625 A CN 202110016625A CN 112801139 B CN112801139 B CN 112801139B
- Authority
- CN
- China
- Prior art keywords
- fwelm
- adaptive
- domain data
- wear state
- cutter
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000013526 transfer learning Methods 0.000 title claims abstract description 23
- 239000011159 matrix material Substances 0.000 claims abstract description 43
- 238000005299 abrasion Methods 0.000 claims abstract description 27
- 230000003044 adaptive effect Effects 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000013145 classification model Methods 0.000 claims abstract description 14
- 230000009467 reduction Effects 0.000 claims abstract description 13
- 238000000605 extraction Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 23
- 238000012545 processing Methods 0.000 claims description 18
- 238000012417 linear regression Methods 0.000 claims description 13
- 238000002472 multi-metal deposition Methods 0.000 description 43
- 238000009826 distribution Methods 0.000 description 20
- 238000003754 machining Methods 0.000 description 17
- 238000005520 cutting process Methods 0.000 description 15
- 238000012544 monitoring process Methods 0.000 description 13
- 238000013507 mapping Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 7
- 238000004519 manufacturing process Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000012546 transfer Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000014759 maintenance of location Effects 0.000 description 2
- 238000000691 measurement method Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000010008 shearing Methods 0.000 description 1
- 230000003746 surface roughness Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Operations Research (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Numerical Control (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a cutter wear state intelligent identification method based on heterogeneous domain self-adaptive transfer learning, which comprises the following steps: constructing a cutter wear state identification system based on heterogeneous domain self-adaptive transfer learning; acquiring source domain data S and target domain data T according to different wear curves of a plurality of cutters, and carrying out feature extraction and feature dimension reduction on the data; constructing an MMD matrix M, and initializing parameters and the maximum iteration times; initializing FWELM random input weights and calculating an hidden layer output matrix H of the random input weights; calculating a reconstruction output weight by using DST-FWELM; calculating reconstructed cutter abrasion source domain data S 'and target domain data T'; training an adaptive FWELM classification model by using the reconstructed source domain data; updating the target pseudo tag and the condition matrix M k by using the self-adaptive FWELM classification model; and predicting the tool wear state by using a final self-adaptive FWELM classification model after the maximum iteration number is reached.
Description
Technical Field
The invention belongs to the field of cutter wear monitoring of numerical control manufacturing equipment, and particularly relates to an intelligent cutter wear state identification method based on heterogeneous domain self-adaptive transfer learning.
Background
The tool is a direct executor of machining and manufacturing, and the increased abrasion of the tool can lead to increased cutting force, increased surface roughness of a workpiece, out-of-tolerance requirement of the workpiece size and even stop of machining, so that the machining efficiency is reduced. The tool state monitoring technology can timely master the tool abrasion state, and has important and profound significance for improving the processing quality and the surface precision of the workpiece, improving the economic benefit of products, saving the processing time and the like. In order to find a better monitoring method, picking up an original signal with a close relation with the cutter wear, analyzing and acquiring characteristic information with an obvious mapping relation with the cutter wear state, and identifying the cutter state by adopting an identification model with good generalization performance.
However, due to the continuous variation of cutting force, cutting heat, cutting vibration and machining environment during machining and manufacturing, and random errors in clamping, parameter setting, machining and disassembly operations during machining and manufacturing, the whole machining process is full of dynamic uncertainties, and the uncertainties directly affect tool wear and the quality of the machining process of the workpiece.
Therefore, the identification of the wear state of the cutter in the uncertain processing environment has important significance in the aspects of ensuring the safety of a processing system, smoothly proceeding processing, reducing the production cost, improving the production efficiency and the like.
The dynamic uncertainty of the machining process causes the cutting force, the cutting heat and the cutter abrasion in the machining process to present strong randomness, thereby causing the unknowability of the parameters of the machining process, the uncertainty of the state change, the ambiguity of the information and the coupling of the multidimensional information and causing the uncertainty of the establishment of a machining process model. In addition, factors such as cutter abrasion, uneven materials, voltage and load changes and processing environment in the processing process greatly increase the modeling difficulty of the processing process. In response to the above problems, an extreme learning machine recognition model based on fuzzy sets has been developed.
The basic idea of fuzzy sets is to take the things of processing the fuzzy uncertainty of the concept as the research targets and precisely quantize the things into information which can be processed by a computer. The concept of fuzzy set is added into the extreme learning machine, so that the distribution of each sample in the feature space can be fully excavated, the respective features of the sample can be subjected to fuzzy and personalized setting, the range and capability of the extreme learning machine for processing information can be greatly widened, accurate information can be processed, fuzzy information or other uncertain information in the processing process can be processed, accurate association and mapping can be realized, and inaccurate line association and mapping, particularly fuzzy association and fuzzy mapping can be realized. Therefore, the fuzzy set is used for the extreme learning machine to effectively reduce the influence caused by the uncertainty of a processing system, and the generalization performance of the recognition model can be improved. However, it is not enough to suppress the influence of the uncertainty of the machining system alone, and the dynamic time-varying property of the machining system and the tool wear state is also an important factor affecting the tool wear state recognition model.
In the recognition problem of the cutter abrasion state, the data set has the recognition problem of non-uniformity, and the complicated segmentation characteristic of the cutter abrasion state causes that the fuzzy extreme learning machine can not well process the recognition problem of the cutter abrasion state, and the fuzzy wavelet extreme learning machine is used as a product of wavelet analysis and combination of the fuzzy extreme learning machine, and the hidden layer neuron of the fuzzy extreme learning machine is implanted with a wavelet basis function, thereby inheriting the time-frequency local characteristic and focusing characteristic of the wavelet, effectively recognizing the singularity of signals, and effectively overcoming the problems of slow convergence speed, easy sinking into local extremum, difficult determination of learning step length and the like of the traditional neural network. Therefore, the fuzzy wavelet extreme learning machine is used for identifying the wear state of the cutter, the interference problem caused by the dynamic time-varying characteristics and uncertainty of the machining system and the wear state of the cutter can be effectively solved, and the training and testing speeds of an identification model are higher, and the result is more accurate.
The invention relates to a cutter abrasion monitoring method based on a cutting force model (CN 106002488B), which utilizes a non-equally divided shearing area model to calculate the cutting force under the action of a sharp cutter generated by the formation of chips in the cutting process, and establishes a cutter monitoring model, thereby realizing real-time monitoring of cutter abrasion. However, the method needs to acquire the cutting force in the cutting process, and the uncertainty of a cutting system can lead to unstable cutting force, so that the accuracy of a monitoring result of cutter abrasion is insufficient.
The invention relates to a method and a system for detecting the broken and worn state of a cutter (CN 102765010B), which are used for detecting the broken and worn state of the cutter and the running state of a machine tool by measuring vibration signals in the cutting and grinding processes of the cutter, classifying and sorting the vibration signals and extracting characteristics, counting kurtosis indexes and peak indexes of various signals and utilizing the dynamic distribution of the indexes. However, the method directly judges the state of the cutter through the characteristics of the signals, so that the detection result is easy to be interfered by the system environment, and the accuracy is not enough.
The invention relates to a method for monitoring tool wear of a numerical control machine tool (CN 102091972B), which is characterized in that collected servo drive current signals are analyzed and processed, the signals are decomposed in a frequency domain by utilizing a wavelet packet analysis technology, so that time-frequency domain characteristics of the signals in each frequency band are obtained, and a neural network is utilized to learn and monitor the tool wear process. However, the neural network utilized by the method can only monitor the same type of cutters, but cannot monitor different types of cutters, so that the applicability is not enough.
The invention patent 'a cutter abrasion state monitoring method based on a conditional random field model' (CN 102689230B) is characterized in that acoustic emission signals in the cutting process are collected, preprocessed and relevant characteristics are extracted, the acoustic emission signals are used as training data of the conditional random field model, then a test sample is input into the built model, and the corresponding cutter abrasion state is output. However, the method needs to extract a large amount of training data, increases training time, leads to insufficient real-time performance of the model, and can not monitor different types of cutters, so that the method has great limitation in use.
Aiming at the problems, the invention aims to design the intelligent recognition method of the cutter abrasion state based on heterogeneous domain self-adaptive transfer learning, and the method utilizes a Maximum mean difference (Maximum MEAN DISCREPANCY, MMD) algorithm and a fuzzy wavelet extreme learning machine (Fuzzy wavelet extreme LEARNING MACHINE, FWELM) model, combines the domain transfer thought, and effectively solves the problems of single recognition target and insufficient accuracy of a cutter abrasion state recognition system in the current processing process.
Disclosure of Invention
Aiming at the defects of the existing cutter wear monitoring method, the invention aims to provide the intelligent cutter wear state identification method based on heterogeneous domain self-adaptive transfer learning, which has good accuracy and wide universality and can be used for monitoring the wear states of different types of cutters.
The embodiment of the disclosure provides a cutter wear state intelligent identification method based on heterogeneous domain adaptive transfer learning, which can comprise the following steps:
step one: constructing a cutter wear state identification system based on heterogeneous domain self-adaptive transfer learning;
step two: acquiring source domain data S and target domain data T according to different wear curves of a plurality of cutters, and carrying out feature extraction and feature dimension reduction on the data;
Step three: constructing an MMD matrix M, and initializing parameters and the maximum iteration times;
step four: initializing FWELM random input weights and calculating a fuzzy hidden layer output matrix H;
Step five: calculating a reconstruction output weight by using a Domain space migration-fuzzy wavelet extreme learning machine (Domain SPACE TRANSFER-Fuzzy Wavelet Extreme LEARNING MACHINE, DST-FWELM);
step six: calculating reconstructed cutter abrasion source domain data S 'and target domain data T';
step seven: training an adaptive FWELM classification model by using the reconstructed source domain data;
Step eight: updating the target pseudo tag and the condition matrix M k by using the self-adaptive FWELM classification model; and
Step nine: and after the maximum iteration times are reached, predicting the tool wear state by utilizing a final self-adaptive FWELM classification model.
According to some exemplary embodiments, in step one, the tool wear state recognition system based on heterogeneous domain adaptive transfer learning is divided into a tool state source domain data acquisition object, a tool state target domain data acquisition object, a data processing model and a DST-FWELM model. The cutter state source domain data acquisition objects are different cutters, and source domain data are acquired from the cutter state source domain data acquisition objects; the tool state target domain data acquisition object is a tool to be monitored, and a small amount of target domain data is acquired from the tool state target domain data acquisition object; the data processing model comprises feature extraction and feature dimension reduction of data, wherein feature dimension reduction refers to extracting significant features from all features by using a linear regression method, and the significant features are used as features for final training and are removed;
The ST-FWELM model mainly comprises an MMD algorithm and a FWELM model, wherein the MMD algorithm is used for measuring the distribution distance between source domain data and target domain data, reconstructing the source domain data and the target domain data through the minimization of the distribution distance, and the FWELM model is used for completing training of the reconstructed source domain data and testing of the target domain data; finally, the trained DST-FWELM model is used for monitoring the abrasion state of the target cutter.
According to some exemplary embodiments, in step two, a method of linear regression is used to select features that are significant among the features of the data and reject other features that are not significant, where the dimension reduction of the linear regression is as follows:
(1) Extracting statistical features, frequency domain features and time-frequency domain features in the source domain data and the target domain data;
(2) Taking different samples of cutter wear as independent variables, taking the corresponding characteristics of each sample as different independent variable types, taking a cutter wear value as a target value, and performing linear regression;
(3) Each variable coefficient and the significance level p value of the coefficient obtained by linear regression are used, and the smaller the p value is, the higher the significance of the coefficient is proved; and
(4) And reserving the coefficient with the p value smaller than 0.05 and the corresponding characteristic, and eliminating other characteristics.
According to some exemplary embodiments, in step three, the number of fuzzy rules L for the scaling coefficients λ, FWELM in the coefficient matrix (bi＝[bi1,…,bin]T,di＝[di1,…,din]T,i＝1,...,L),MMD of the wavelet function in the coefficient matrix (ci＝[ci1,…,cin]T,ai＝[ai1,…,ain]T,i＝1,...,L),FWELM of the membership function in FWELM is required to be initialized.
According to some exemplary embodiments, in step four, FWELM random input weights are w i＝[wi1,wi2,...,win, the initial threshold is o i e R, and the H matrix is calculated as follows:
And converting the cutter abrasion source domain data and the target domain data into the same characteristic space by utilizing the H matrix.
According to some exemplary embodiments, in step five, the MMD is calculated as follows:
MMD is an effective non-parametric distribution distance metric based on the maximum average function value difference between two distributions:
Where F is a type of mapping function, MMD 2 (F, p, q) =0 holds if and only if p=q.
For two data sets following the distributions p and q, respectivelyAnd (3) withThe MMD distance in H space for the two datasets is:
Wherein phi is a feature mapping function, and I.I H is RKHS space.
According to some exemplary embodiments, in step five, the MMD based on the tool wear data is calculated as follows:
As a non-parametric estimator of the distribution distance, MMD is used to transfer knowledge from one domain to another. In DST-FWELM, an empirical MMD term is added as a penalty to the FWELM formula. The distance measurement method of MMD is that
Representing MMDs in matrix form as
MMD2(ΦS,ΦT)＝Tr[βTHTMHβ] (5)
In the middle ofIs an MMD matrix, specifically expressed as
Preferably, the calculation method of the reconstructed output weight in the fifth step is as follows:
DST-FWELM simultaneously performs information retention and distribution distance reduction of target tool wear data, and reconstructs data of two domains to obtain DST-FWELM target function
Expanding the matrix into a matrix with uniform size, and making Rewriting minimization problem as
Obviously, the minimization problem is a convex quadratic minimization problem with an optimum equal to the fixed point, i.e
β*+HTΔHβ*-HTΔX′+λHTMHβ*＝0 (9)
Find the output weight beta * as
When L is larger than n S+nT, the total number of the components is larger than n S+nT,
When L is larger than n S+nT, the total number of the components is larger than n S+nT,
β*＝(IL+HT(Δ+λM)H)-1HTΔX′ (11)
Preferably, in the fourth step, the method for calculating the output weight by using the conditional MMD is as follows:
In the DST-FWELM model, the marginal distribution distance is considered, but no conditional distribution is involved. In practice, algorithms that merely reduce the boundary distance between domains may not perform well because the tag information of the source data is not utilized. In order to fully exploit the available data sets, source data tag knowledge should also be transferred to the target domain to improve data reconstruction.
To this end, a conditional MMD is added to the above formula, wherein the initial pseudo tag of the target sample is obtained from a FWELM classifier trained from source data or source data reconstructed using marginal MMD metrics. We express the square of the conditional MMD asIs defined as
Convert it into matrix form
In the middle ofMMD condition matrix for kth class
The conditional MMD is incorporated into DST-FWELM to give the following formula
When L is larger than n S+nT, the total number of the components is larger than n S+nT,
When L is less than or equal to n S+nT,
According to some exemplary embodiments, in step seven, the structure of the Fuzzy Wavelet (FW) is defined as:
r i: if x 1 is X 2 is/>…, X n are/>Then y m is/>In/>Depending on the parity of n, when n is even,
When n is an odd number, the number of the n,
The number of linear coefficients of each fuzzy rule isWherein/>Representing the smallest integer greater than or equal to n, the computational formula of the membership function part of FW is
Where c is the center position and a controls the slope of the intersection at x=c.
According to some exemplary embodiments, the classification model of FWELM in step seven is:
For N different training data (x i,ti), where x i＝[xi1,…,xin]T,ti＝[ti1,…,tim]T, L is the number of fuzzy rules, the FW model containing the parameter (. Beta. i,ci,ai) is
Where q iFW (i=1, the term "L") is
Based on the parity of n, the wavelet function in FWELM is
When n is an even number, the number,
When n is an odd number, the number of the n,
Obtained according to the above equation, the FW model is
It is subjected to rewritten as matrix form
HQ＝T (27)
Wherein H is a parameter matrix of the fuzzy wavelet model
Q is a linear parameter matrix
According to some exemplary embodiments, the algorithm steps of FWELM in step seven include:
For N different training data (x i,ti), where x i＝[xi1,...,xin]T,ti＝[ti1,…,tim]T, L is the number of fuzzy rules.
(1) Randomly setting parameters of membership function and wavelet function (ci＝[ci1,…,cin]T,ai＝[ai1,...,ain]T,bi＝[bi1,...,bin]T,di＝[di1,...,din]T,i＝1,...,L);
(2) Calculating an H matrix;
(3) And obtaining Q=H + T according to HQ=T, wherein H + is a generalized inverse matrix of H.
Compared with the prior art, the technical scheme provided by the embodiment of the invention has the following beneficial effects:
(1) The domain migration idea adopted by the invention has the following advantages:
1) The tool wear state recognition system can train training data with different feature numbers, and the model training universality is remarkably improved;
2) The difference between different cutter abrasion training data sets can be reduced by using the cutter abrasion state recognition system, so that the model can be suitable for the abrasion state recognition of different types of cutters;
(2) The fuzzy wavelet extreme learning machine model adopted by the invention has the following advantages:
1) The cutter abrasion state recognition system can reduce the interference caused by the environmental factors and the dynamic uncertainty of the machining process, so that the state recognition result is more accurate;
2) The complexity of the cutter abrasion state identification system is reduced, the system is easier to debug, and the training speed is improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention and do not constitute a limitation on the invention. In the drawings:
Fig. 1 is a block diagram of a tool wear state recognition system based on heterogeneous domain adaptive transfer learning according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present disclosure more apparent, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present disclosure. It will be apparent that the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments, which can be made by one of ordinary skill in the art without the need for inventive faculty, are within the scope of the present disclosure, based on the described embodiments of the present disclosure.
Unless defined otherwise, technical or scientific terms used in this disclosure should be given the ordinary meaning as understood by one of ordinary skill in the art. The terms "first," "second," and the like, as used in this disclosure, do not denote any order, quantity, or importance, but rather are used to distinguish one element from another. The word "comprising" or "comprises", and the like, means that elements or items preceding the word are included in the element or item listed after the word and equivalents thereof, but does not exclude other elements or items.
As shown in fig. 1, an embodiment of the present disclosure provides a method for intelligently identifying a tool wear state based on heterogeneous domain adaptive transfer learning, which may include the following steps.
In step one, a tool wear state recognition system based on heterogeneous domain adaptive transfer learning is constructed, as shown in fig. 1.
Specifically, the cutter abrasion state recognition system based on heterogeneous domain self-adaptive transfer learning in the first step is divided into a cutter state source domain data acquisition object, a cutter state target domain data acquisition object, a data processing model and a DST-FWELM model. Wherein the cutter state source domain data acquisition objects are different cutters, and a large amount of source domain data is acquired from the cutter state source domain data acquisition objects; the tool state target domain data acquisition object is a tool to be monitored, and a small amount of target domain data is acquired from the tool state target domain data acquisition object; the data processing model comprises feature extraction and feature dimension reduction of data, wherein feature dimension reduction refers to extracting significant features from all features by using a linear regression method, and the significant features are used as features for final training and are removed; the DST-FWELM model mainly comprises an MMD algorithm and a FWELM model, wherein the MMD algorithm is used for measuring the distribution distance between source domain data and target domain data, reconstructing the source domain data and the target domain data through the minimization of the distribution distance, and the FWELM model is used for completing training of the reconstructed source domain data and testing of the target domain data; finally, the trained DST-FWELM model is used for monitoring the abrasion state of the target cutter.
In the second step, source domain data S and target domain data T are obtained according to different wear curves of several kinds of cutters, and feature extraction and feature dimension reduction are carried out on the data.
Specifically, in the second step, a method of linear regression is used to select features which are obvious in the features of the data, and other features which are not obvious in the features are removed, wherein the step of dimension reduction of the linear regression is as follows:
(1) Extracting statistical features, frequency domain features and time-frequency domain features in the source domain data and the target domain data;
(2) Taking different samples of cutter wear as independent variables, taking the corresponding characteristics of each sample as different independent variable types, taking a cutter wear value as a target value, and performing linear regression;
(3) Each variable coefficient and the significance level p value of the coefficient obtained by linear regression are used, and the smaller the p value is, the higher the significance of the coefficient is proved;
(4) And reserving the coefficient with the p value smaller than 0.05 and the corresponding characteristic, and eliminating other characteristics.
In step three, an MMD matrix M is constructed, and parameters and maximum iteration times are initialized.
Specifically, the parameter to be initialized in the third step is the fuzzy rule number L of the scaling coefficient λ, FWELM in the coefficient matrix (bi＝[bi1,...,bin]T,di＝[di1,...,din]T,i＝1,...,L),MMD of the wavelet function in the coefficient matrix (ci＝[ci1,…,cin]T,ai＝[ai1,…,ain]T,i＝1,...,L),FWELM of the membership function in FWELM.
In step four, random input weights are initialized FWELM and their fuzzy hidden layer output matrix H is calculated.
Specifically, in the fourth step FWELM, the random input weight is w i＝[wi1,wi2,...,win, the initial threshold is o i e R, and the H matrix is calculated according to the following formula:
And converting the cutter abrasion source domain data and the target domain data into the same characteristic space by utilizing the H matrix.
In step five, the reconstructed output weights are calculated using DST-FWELM.
Specifically, the calculation method of MMD in the fifth step is as follows:
MMD is an effective non-parametric distribution distance metric based on the maximum average function value difference between two distributions:
Where F is a type of mapping function, MMD 2 (F, p, q) =0 holds if and only if p=q.
For two data sets following the distributions p and q, respectivelyAnd (3) withThe MMD distance in H space for the two datasets is:
Wherein phi is a feature mapping function, and I.I H is RKHS space.
Specifically, the MMD calculation method based on the tool wear data in the fifth step is as follows:
As a non-parametric estimator of the distribution distance, MMD is used to transfer knowledge from one domain to another. In DST-FWELM, an empirical MMD term is added as a penalty to the FWELM formula. The distance measurement method of MMD is that
Representing MMDs in matrix form as
MMD2(ΦS,ΦT)＝Tr[βTHTMHβ] (5)
In the middle ofIs an MMD matrix, specifically expressed as
Specifically, the calculation method of the reconstructed output weight in the fifth step is as follows:
DST-FWELM simultaneously performs information retention and distribution distance reduction of target tool wear data and reconstructs the data of the two fields in a unique manner to obtain a DST-FWELM objective function
Expanding the matrix into a consistent matrix, letting Rewriting minimization problem as
Obviously, the minimization problem is a convex quadratic minimization problem with an optimum equal to the fixed point, i.e
β*+HTΔHβ*-HTΔX′+λHTMHβ*＝0 (9)
Find the output weight beta * as
When L is larger than n S+nT, the total number of the components is larger than n S+nT,
When L is less than or equal to n S+nT,
β*＝(IL+HT(Δ+λM)H)-1HTΔX′ (11)
Specifically, the method for calculating the output weight by using the conditional MMD in the fifth step is as follows:
In the aforementioned DST-FWELM model, the marginal distribution distance is considered, but no conditional distribution is involved. In practice, algorithms that merely reduce the boundary distance between domains may not perform well because the tag information of the source data is not utilized. To fully exploit the available dataset, source class label knowledge should also be transferred to the target domain to improve data reconstruction.
To this end, a conditional MMD is added to the above formula, wherein the initial pseudo-markers of the target samples are obtained from a FWELM classifier trained from the raw source data or source samples reconstructed using marginal MMD metrics. We express the square of the conditional MMD asIs defined as
Convert it into matrix form
In the middle ofMMD condition matrix for kth class
The conditional MMD is incorporated into DST-FWELM to give the following formula
When L is larger than n S+nT, the total number of the components is larger than n S+nT,
When L is larger than n S+nT, the total number of the components is larger than n S+nT,
In step six, the reconstructed tool wear source field data S 'and target field data T' are calculated.
In step seven, an adaptive FWELM classification model is trained using the reconstructed source domain data.
Specifically, the structure of FW in step seven is defined as:
r i: if x 1 is X 2 is/>…, X n are/>Then y m is/>
In the middle ofThe value of (c) depends on the parity of n,
When n is an even number, the number,
When n is an odd number, the number of the n,
The number of linear coefficients of each fuzzy rule isWherein/>Representing the smallest integer greater than or equal to n, the computational formula of the membership function part of FW is
Where c is the center position and a controls the slope of the intersection at x=c.
Specifically, the classification model of FWELM in step seven is:
For N different training data (x i,ti), where x i＝[xi1,...,xin]T,ti＝[ti1,...,tim]T, L is the number of fuzzy rules, the FW model containing the parameter (. Beta. i,ci,ai) is
Where q iFW (i=1, the term "L") is
Based on the parity of n, the wavelet function in FWELM is
When n is an even number, the number,
When n is an odd number, the number of the n,
Obtained according to the above equation, the FW model is
It is subjected to rewritten as matrix form
HQ＝T (27)
Wherein H is a parameter matrix of the fuzzy wavelet model
Q is a linear parameter matrix
Specifically, the algorithm in step FWELM is as follows:
for N different training data (x i,ti), where x i＝[xi1,...,xin]T,ti＝[ti1,...,tim]T, L is the number of fuzzy rules.
(1) Randomly setting parameters of membership function and wavelet function (ci＝[ci1,…,cin]T,ai＝[ai1,...,ain]T,bi＝[bi1,...,bin]T,di＝[di1,...,din]T,i＝1,…,L);
(2) Calculating an H matrix;
(3) And obtaining Q=H + T according to HQ=T, wherein H + is a generalized inverse matrix of H.
In step eight, the target pseudo tag and condition matrix M k are updated with the adaptive FWELM classification model.
In step nine, after the maximum iteration number is reached, the tool wear state is predicted by using a final self-adaptive FWELM classification model.
While the foregoing description illustrates and describes the embodiments of the present invention, it is to be understood that the invention is not limited to the forms disclosed herein, but is not to be construed as limited to other embodiments, and is capable of use in various other combinations, modifications and environments and is capable of changes or modifications within the spirit of the invention described herein, either as a result of the foregoing teachings or as a result of the knowledge or skill of the relevant art. And that modifications and variations which do not depart from the spirit and scope of the invention are intended to be within the scope of the appended claims.
Claims (8)
1. The intelligent identification method for the cutter wear state based on heterogeneous domain self-adaptive transfer learning is characterized by comprising the following steps of:
step one: constructing a cutter wear state identification system based on heterogeneous domain self-adaptive transfer learning;
step two: acquiring source domain data S and target domain data T according to different wear curves of a plurality of cutters, and carrying out feature extraction and feature dimension reduction on the data;
Step three: constructing an MMD matrix M, and initializing parameters and the maximum iteration times;
step four: initializing FWELM random input weights and calculating an hidden layer output matrix H of the random input weights;
Step five: calculating a reconstruction output weight by using DST-FWELM;
step six: calculating reconstructed cutter abrasion source domain data S 'and target domain data T';
step seven: training an adaptive FWELM classification model by using the reconstructed source domain data;
Step eight: updating the target pseudo tag and the condition matrix M k by using the self-adaptive FWELM classification model; and
Step nine: and after the maximum iteration times are reached, predicting the tool wear state by utilizing a final self-adaptive FWELM classification model.
2. The intelligent recognition method for the cutter wear state based on heterogeneous domain adaptive transfer learning according to claim 1, wherein in the second step, the characteristic which is obvious in appearance is selected by using a linear regression method, and other characteristics which are not obvious in appearance are removed.
3. The intelligent recognition method of tool wear state based on heterogeneous domain adaptive transfer learning according to claim 1, wherein in the third step, parameters to be initialized include a membership function coefficient in FWELM, a wavelet function coefficient and a scaling coefficient in MMD.
4. The intelligent recognition method of tool wear state based on heterogeneous domain adaptive transfer learning according to claim 1, wherein in the fifth step, the reconstructed output weight is used to transform the source domain data and the target domain data into the same feature space.
5. The intelligent recognition method of tool wear state based on heterogeneous domain adaptive transfer learning according to claim 1, wherein in the seventh step, the reconstructed source domain data is directly input into the adaptive FWELM classifier, and the adaptive FWELM classifier reinitializes the input weights and thresholds inside the classifier.
6. The intelligent recognition method of the tool wear state based on heterogeneous domain adaptive transfer learning according to claim 1, wherein in the eighth step, the reconstructed target domain data is input into a trained adaptive FWELM classifier to obtain a pseudo tag corresponding to the target domain, and the condition matrix is updated by using the pseudo tag.
7. The intelligent tool wear state identification method based on heterogeneous domain adaptive transfer learning of claim 1, wherein the tool wear state identification system comprises a tool state source domain data acquisition object, a tool state target domain data acquisition object, a data processing model and a DST-FWELM model.
8. The intelligent recognition method for the cutter wear state based on heterogeneous domain adaptive transfer learning according to claim 2, wherein the feature dimension reduction comprises:
extracting statistical features, frequency domain features and time-frequency domain features in the source domain data and the target domain data;
taking different samples of cutter wear as independent variables, taking the corresponding characteristics of each sample as different independent variable types, taking a cutter wear value as a target value, and performing linear regression;
a significance level p value of each variable coefficient and coefficient obtained by linear regression; and
And reserving the coefficient with the p value smaller than 0.05 and the corresponding characteristic, and eliminating other characteristics.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110016625.7A CN112801139B (en) | 2021-01-06 | 2021-01-06 | Intelligent cutter wear state identification method based on heterogeneous domain self-adaptive transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110016625.7A CN112801139B (en) | 2021-01-06 | 2021-01-06 | Intelligent cutter wear state identification method based on heterogeneous domain self-adaptive transfer learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112801139A CN112801139A (en) | 2021-05-14 |
| CN112801139B true CN112801139B (en) | 2024-04-26 |
Family
ID=75808751
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110016625.7A Active CN112801139B (en) | 2021-01-06 | 2021-01-06 | Intelligent cutter wear state identification method based on heterogeneous domain self-adaptive transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112801139B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113752089B (en) * | 2021-10-19 | 2023-05-30 | 山东农业大学 | Cutter state monitoring method based on singular Lev index |
| CN116360342A (en) * | 2023-02-13 | 2023-06-30 | 广东工业大学 | Machine tool thermal error prediction modeling method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2015105471A (en) * | 2013-11-28 | 2015-06-08 | 国立研究開発法人産業技術総合研究所 | Pdc cutter and pdc drill bit for pit drilling |
| CN110348579A (en) * | 2019-05-28 | 2019-10-18 | 北京理工大学 | A kind of domain-adaptive migration feature method and system |
| CN111639461A (en) * | 2020-05-26 | 2020-09-08 | 浙江大学 | Tool wear state detection method aiming at industrial unbalanced data |
-
2021
- 2021-01-06 CN CN202110016625.7A patent/CN112801139B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2015105471A (en) * | 2013-11-28 | 2015-06-08 | 国立研究開発法人産業技術総合研究所 | Pdc cutter and pdc drill bit for pit drilling |
| CN110348579A (en) * | 2019-05-28 | 2019-10-18 | 北京理工大学 | A kind of domain-adaptive migration feature method and system |
| CN111639461A (en) * | 2020-05-26 | 2020-09-08 | 浙江大学 | Tool wear state detection method aiming at industrial unbalanced data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112801139A (en) | 2021-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Marei et al. | Transfer learning enabled convolutional neural networks for estimating health state of cutting tools | |
| Dun et al. | A chatter detection method in milling of thin-walled TC4 alloy workpiece based on auto-encoding and hybrid clustering | |
| Shi et al. | Tool wear prediction via multidimensional stacked sparse autoencoders with feature fusion | |
| CN110070145B (en) | LSTM hub single-product energy consumption prediction based on incremental clustering | |
| CN112801139B (en) | Intelligent cutter wear state identification method based on heterogeneous domain self-adaptive transfer learning | |
| CN111562108A (en) | Rolling bearing intelligent fault diagnosis method based on CNN and FCMC | |
| CN111633467B (en) | Cutter wear state monitoring method based on one-dimensional depth convolution automatic encoder | |
| Wang et al. | A new multitask learning method for tool wear condition and part surface quality prediction | |
| CN110561191B (en) | Numerical control machine tool cutter abrasion data processing method based on PCA and self-encoder | |
| Wang et al. | A sequence-to-sequence model with attention and monotonicity loss for tool wear monitoring and prediction | |
| CN112518425B (en) | Intelligent machining cutter wear prediction method based on multi-source sample migration reinforcement learning | |
| Niu et al. | Remaining useful life prediction of machining tools by 1D-CNN LSTM network | |
| CN110766060B (en) | Time series similarity calculation method, system and medium based on deep learning | |
| CN113798920A (en) | Cutter wear state monitoring method based on variational automatic encoder and extreme learning machine | |
| CN113601261B (en) | Monitoring method of online rapid optimization model for cutter | |
| CN111126255A (en) | Numerical control machine tool cutter wear value prediction method based on deep learning regression algorithm | |
| Shaban et al. | Cutting tool wear detection using multiclass logical analysis of data | |
| Chaowen et al. | Research on tool wear monitoring based on GRU-CNN | |
| CN115741235A (en) | Wear prediction and health management method based on five-axis machining center cutter | |
| CN113627544A (en) | Machine tool milling cutter state identification method based on multi-source heterogeneous data fusion | |
| CN110020680B (en) | PMU data classification method based on random matrix theory and fuzzy C-means clustering algorithm | |
| Zegarra et al. | Comparison of CNN and CNN-LSTM architectures for tool wear estimation | |
| CN113203953A (en) | Lithium battery residual service life prediction method based on improved extreme learning machine | |
| CN116821828A (en) | Multi-dimensional time sequence prediction method based on industrial data | |
| CN116372665A (en) | OPC UA-based aviation numerical control machine tool health monitoring method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |