CN116070527B - Milling cutter residual life prediction method based on degradation model - Google Patents

Milling cutter residual life prediction method based on degradation model Download PDF

Info

Publication number
CN116070527B
CN116070527B CN202310206684.XA CN202310206684A CN116070527B CN 116070527 B CN116070527 B CN 116070527B CN 202310206684 A CN202310206684 A CN 202310206684A CN 116070527 B CN116070527 B CN 116070527B
Authority
CN
China
Prior art keywords
feature
time
frequency domain
representing
degradation model
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
Application number
CN202310206684.XA
Other languages
Chinese (zh)
Other versions
CN116070527A (en
Inventor
赵正彩
李尧
朱夏林
张创
傅玉灿
徐九华
吴尚霖
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Aeronautics and Astronautics
Nanjing Chenguang Group Co Ltd
Original Assignee
Nanjing University of Aeronautics and Astronautics
Nanjing Chenguang Group Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing University of Aeronautics and Astronautics, Nanjing Chenguang Group Co Ltd filed Critical Nanjing University of Aeronautics and Astronautics
Priority to CN202310206684.XA priority Critical patent/CN116070527B/en
Publication of CN116070527A publication Critical patent/CN116070527A/en
Application granted granted Critical
Publication of CN116070527B publication Critical patent/CN116070527B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/04—Ageing analysis or optimisation against ageing
    • 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)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a milling cutter residual life prediction method based on a degradation model, which mainly comprises the following steps: firstly, data preprocessing is carried out by combining vibration, load current, main shaft power signals and PLC auxiliary information, characteristics with the front sequence are selected to participate in main component fusion and health factors are constructed to serve as observation data, and finally, an online update degradation model is designed to learn complex nonlinear relation between multi-source input characteristics and the residual service life of a cutter, so that iterative estimation of the residual service life of the cutter is realized. The invention takes the health factors of fusion analysis as the indexes for evaluating the residual life of the cutter, and can evaluate the residual life of the cutter more scientifically, accurately and reliably.

Description

Milling cutter residual life prediction method based on degradation model
Technical Field
The invention belongs to the field of operation reliability and life prediction of mechanical products, and particularly relates to a milling cutter residual life prediction method based on a degradation model, which can be used for guiding life tracking of a cutter in a machining process.
Background
The real-time state of the cutter in the cutting process directly influences the processing quality, the processing precision and the processing efficiency of the part as an executor of the cutting processing, and even causes serious obstacle to the whole mechanical processing system, thereby causing huge economic loss. The main failure modes of milling machine tools are abrasion and fracture, and the degradation process of the milling machine tools cannot be directly observed in practice and can only be indirectly expressed by using characteristic indexes shown in the degradation process.
At present, the prediction methods about the residual life of the cutter are mainly divided into three types, one type is a method based on an empirical model, such as a Taylor formula and an expansion equation thereof; one is a method based on numerical simulation analysis, such as a finite element analysis model; the other type is a machine learning-based method, which is the most widely used method at present and mainly comprises a shallow machine learning method and a deep learning method.
With the development and utilization of high-precision monitoring sensors and intelligent algorithms, the online and real-time tracking of the characterization of the degradation process of the tool is realized, so that the degradation state can be estimated by utilizing the online monitoring signals.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a milling cutter residual life prediction method based on a degradation model, and aims to track cutter abrasion in the machining process and realize the prediction of cutter residual life.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
the milling cutter residual life prediction method based on the degradation model is characterized by comprising the following steps of:
step 1: acquiring sample data comprising machining process signal data and tool wear values;
step 2: extracting signal characteristics of a time domain, a frequency domain and a time-frequency domain from the signal data in the processing process to generate massive characteristics Chi Juzhen comprising the time domain characteristics, the frequency domain characteristics and the time-frequency domain characteristics;
step 3: three rounds of signal screening are carried out on the mass feature pool matrixes according to the feature time sequence, wherein the three rounds of signal screening are monotonicity sequencing, correlation optimization and redundancy elimination respectively;
step 4: splicing the screened characteristics to form a multidimensional characteristic matrix, analyzing the multidimensional characteristic matrix by adopting a principal component analysis method, and taking principal components as health factors for driving the prediction of the residual life;
step 5: constructing a degradation model based on the health factors, taking the cutter abrasion value as a label, reading the label and the health factors as a sample data set, and training the degradation model;
step 6: and predicting the residual life of the cutter according to the trained degradation model.
In order to optimize the technical scheme, the specific measures adopted further comprise:
further, in the step 1, the machining process signal data includes a vibration signal, a load current, a spindle power signal, and PLC information.
Further, the step 2 specifically includes the following steps:
in the time domain, 13 characteristics of gap coefficient, crest coefficient, pulse coefficient, kurtosis, average value, peak value, root mean square, distortion rate, signal to noise ratio, shape factor, deflection, standard deviation and total harmonic distortion are calculated on the processing signal data to form a time domain characteristic pool
Figure SMS_1
，/>
Figure SMS_2
，/>
Figure SMS_3
Represent the firstlThe number of time-domain features,lrepresenting the number of time domain features;
in the frequency domain, 5 characteristics of peak amplitude, peak frequency, natural frequency, damping coefficient and band power are calculated on the processing signal data to form a frequency domain characteristic pool
Figure SMS_4
，/>
Figure SMS_5
，/>
Figure SMS_6
Represent the firstmThe frequency domain characteristics of the frequency domain,mrepresenting frequency domain featuresA number of;
on the time-frequency domain, the time-frequency domain signal characteristics are obtained based on the signal data in the wavelet packet decomposition processing process, and a time-frequency domain characteristic pool is formed
Figure SMS_7
，/>
Figure SMS_8
，/>
Figure SMS_9
Represent the firstnThe characteristics of the time-frequency domain,nrepresenting the number of time-frequency domain features;
integrating the time domain feature pool, the frequency domain feature pool and the time-frequency domain feature pool to form massive features Chi Juzhen
Figure SMS_10
，
Figure SMS_11
。
Further, in the step 3, monotonically sorting is performed according to the following formula:
Figure SMS_13
wherein->
Figure SMS_17
For feature sets ordered by monotonicity, +.>
Figure SMS_20
For the ranking algorithm based on the spearman's rank correlation coefficient judgment, < >>
Figure SMS_14
Representation->
Figure SMS_15
To->
Figure SMS_19
Is provided with a set of features of (a),Mrepresenting the length of the time series>
Figure SMS_22
Matrix for massive feature pool>
Figure SMS_12
Middle (f)jPersonal characteristics (I)>
Figure SMS_16
Is the firstjTime series,/->
Figure SMS_18
For time series rank->
Figure SMS_21
For covariance calculation.
Further, in the step 3, the correlation is preferably performed according to the following formula:
Figure SMS_23
selecting signal characteristics consistent with the change rule of the cutter abrasion value; in (1) the->
Figure SMS_24
Feature set of ranking algorithm is preferred for correlation between two classes of variables>
Figure SMS_25
For a linear correlation calculation algorithm based on the pearson correlation coefficient evaluation, ++>
Figure SMS_26
For correlation calculation, ++>
Figure SMS_27
For the calculation of the variance of the values,
Figure SMS_28
for mass characteristics Chi Juzhen->
Figure SMS_29
Is a tool wear value vector.
Further, in the step 3, the inter-feature autocorrelation operation is performed by the following formula, and redundancy elimination is performed:
Figure SMS_30
the method comprises the steps of carrying out a first treatment on the surface of the In (1) the->
Figure SMS_31
For feature sets that operate by auto-correlation between feature sets,
Figure SMS_32
for the feature autocorrelation calculation, +.>
Figure SMS_33
For correlation calculation, ++>
Figure SMS_34
For variance calculation, < >>
Figure SMS_35
Is a mass feature Chi Juzhen.
Further, the step 4 specifically includes the following steps:
splicing the features which are sorted to be front after screening to form a multidimensional feature matrix
Figure SMS_36
：
Figure SMS_37
，
In the method, in the process of the invention,
Figure SMS_38
、/>
Figure SMS_39
and->
Figure SMS_40
Respectively representing a time domain feature vector, a frequency domain feature vector and a time-frequency domain feature vector which are screened by three rounds of signals, < ->
Figure SMS_41
For the expanded multidimensional feature matrix, +.>
Figure SMS_42
，
Figure SMS_43
Represent the firstMLine 1NThe characteristics of the columns are such that,Mthe length of the time series is indicated,Nrepresenting the number of feature vectors;
according to the principal component analysis method, the multidimensional feature matrix is obtained
Figure SMS_46
Conversion into a scoring matrix>
Figure SMS_49
，
Figure SMS_52
Wherein->
Figure SMS_45
Representing a transformation matrixATranspose of->
Figure SMS_50
The transformation matrix is composed of eigenvectors->
Figure SMS_53
Composition (S)/(S)>
Figure SMS_55
Representation->
Figure SMS_44
Middle (f)MThe elements of the time series,
Figure SMS_48
，/>
Figure SMS_51
indicating transpose,/->
Figure SMS_54
The corresponding maximum characteristic value is +.>
Figure SMS_47
；
Score matrix
Figure SMS_56
The first column of (2) is used as the main component, and the main component is used as the health factor for residual life prediction
Figure SMS_57
Then (1)jHealth factor of the individual time series->
Figure SMS_58
The method comprises the following steps:
Figure SMS_59
，
in the method, in the process of the invention,
Figure SMS_60
representing the initial characteristic value ∈ ->
Figure SMS_61
AndSthe mean and variance of the principal components, < ->
Figure SMS_62
Representing the first of the score matricesiCharacteristic vector numberjCharacteristic values corresponding to the respective time sequences, +.>
Figure SMS_63
Representing the transpose.
Further, in the step 5, the following degradation model is constructed:
Figure SMS_64
，
in the method, in the process of the invention,
Figure SMS_67
is by training->
Figure SMS_68
The resulting health factor as a function of time, < >>
Figure SMS_72
The intercept term is represented as such,
Figure SMS_66
，/>
Figure SMS_69
is a random variable conforming to a lognormal distribution, +.>
Figure SMS_70
Is a random variable conforming to a gaussian distribution, +.>
Figure SMS_71
Representing white gaussian noise->
Figure SMS_65
Representing the noise variance;
taking logarithm of the degradation model and then reducing to obtain the following components:
Figure SMS_73
，
in the method, in the process of the invention,
Figure SMS_74
is->
Figure SMS_75
Variants; at each time step, based on +.>
Figure SMS_76
Will be->
Figure SMS_77
And->
Figure SMS_78
The distribution of (c) is updated as a posterior.
Further, in the step 5, the training process of the degradation model is as follows:
the cutter abrasion value of each cutting edge in the sample data is averaged and then used as a label to obtain a plurality of label matrixes
Figure SMS_79
；
Reading tag matrix
Figure SMS_80
And health factor->
Figure SMS_81
The health factors are standardized to obtain data conforming to standard normal distribution, each tag matrix is matched with the corresponding health factors to serve as a sample data set, the sample data set is divided into a training set and a testing set, a degradation model is trained, and dynamic update is performed>
Figure SMS_82
And->
Figure SMS_83
。
Further, 80% of the sample dataset was used as the training set and 20% was used as the test set.
The beneficial effects of the invention are as follows: according to the milling cutter residual life prediction method based on the degradation model, the actual machining condition of the milling cutter is fully considered, the vibration, load current, main shaft power signals and PLC auxiliary information are combined for data preprocessing, the time domain, the frequency domain and the time-frequency domain are respectively extracted from signal characteristics, and the health factors subjected to fusion analysis are used as indexes for evaluating the cutter residual life, so that the cutter residual life can be predicted more scientifically, accurately and reliably.
Drawings
FIG. 1 is a flow chart of a method for predicting remaining life of a milling tool based on a degradation model.
FIG. 2a is a schematic diagram before feature reduction for a multi-dimensional feature of 54 items; fig. 2b is a schematic diagram of a feature reduction for a multi-dimensional feature of 54 items.
FIG. 3 is a schematic diagram of feature ordering results.
Fig. 4 is a schematic diagram of the duty cycle of each component in the multi-dimensional feature in principal component analysis.
Fig. 5 is a schematic diagram of the result of the implementation of the milling tool remaining life prediction method based on the degradation model.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings of the embodiments of the present application, and it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
The milling cutter residual life prediction method based on the degradation model aims at tracking cutter abrasion in the machining process and realizing the tracking of the cutter residual life. As shown in fig. 1, the implementation process of the method is as follows:
step one, collecting multi-sensor cutter monitoring data, including machining process signal data and cutter abrasion value sample data; each machining process signal data comprises a workbench X-direction vibration signal, a Y-direction vibration signal, a Z-direction vibration signal, a main shaft S3 load current signal and a main shaft S3 main shaft power signal from different acquisition channels
Figure SMS_84
，/>
Figure SMS_85
Is the firstkSensing signals in the individual sampling channels, +.>
Figure SMS_86
Is the collection of acquired signals.
Step two, carrying out data preprocessing on the combined vibration signal, the load current signal, the main shaft power signal and the PLC auxiliary information (such as signal generation time and the like); signal characteristic extraction is carried out by utilizing time domain, frequency domain and time-frequency domain to generate massive characteristics Chi Juzhen
Figure SMS_87
. In the time domain, the gap system is calculated for each signal13 features of number, crest factor, pulse factor, kurtosis, average value, peak value, root mean square, distortion ratio, signal to noise ratio, shape factor, skew, standard deviation and total harmonic distortion, forming a time domain feature pool: />
Figure SMS_88
，/>
Figure SMS_89
Represent the firstlThe number of time-domain features,lrepresenting the number of time domain features; on the frequency domain, 5 characteristics of peak amplitude, peak frequency, natural frequency, damping coefficient and band power are calculated to form a frequency domain characteristic pool: />
Figure SMS_90
，/>
Figure SMS_91
Represent the firstmThe frequency domain characteristics of the frequency domain,mrepresenting the number of frequency domain features; on the time-frequency domain, 8 sub-features are extracted based on the time-frequency domain signal features decomposed by the wavelet packet, and a time-frequency domain feature pool is formed:
Figure SMS_92
，/>
Figure SMS_93
represent the firstnThe characteristics of the time-frequency domain,nrepresenting the number of time-frequency domain features;
integrating the time domain, the frequency domain and the time-frequency domain feature pool to form massive features Chi Juzhen for life prediction
Figure SMS_94
。
Step three, in order to extract the effective signal characteristics, the analysis characteristic linear correlation realizes characteristic reduction, and the specific process is as follows:
calculating the linear correlation coefficient of the time sequence of the extracted features of the multichannel signals and the corresponding time vector, and performing three-cycle signal screening on the feature sequence, wherein the three-cycle signal screening is monotonicity sequencing, correlation optimization and redundancy elimination respectively;
(1) Monotonicity sequencing:
Figure SMS_95
；
in the method, in the process of the invention,
Figure SMS_96
for feature sets ordered by monotonicity, +.>
Figure SMS_100
For the ranking algorithm based on the spearman's rank correlation coefficient judgment, < >>
Figure SMS_103
Representation->
Figure SMS_98
To->
Figure SMS_101
Is provided with a set of features of (a),Mrepresenting the length of the time series>
Figure SMS_104
Matrix for massive feature pool>
Figure SMS_105
Middle (f)jPersonal characteristics (I)>
Figure SMS_97
Is the firstjTime series,/->
Figure SMS_99
For time series rank->
Figure SMS_102
For covariance calculation.
(2) The correlation is preferably:
Figure SMS_106
；
selecting signal characteristics consistent with the change rule of the cutter abrasion value; in the method, in the process of the invention,
Figure SMS_107
feature set of ranking algorithm is preferred for correlation between two classes of variables>
Figure SMS_108
For a linear correlation calculation algorithm based on the pearson correlation coefficient evaluation, ++>
Figure SMS_109
For correlation calculation, ++>
Figure SMS_110
For variance calculation, < >>
Figure SMS_111
Is a tool wear value vector.
(3) Redundancy elimination:
Figure SMS_112
the method comprises the steps of carrying out a first treatment on the surface of the And eliminating repeated feature vectors by calculating the auto-correlation operation among the features. In (1) the->
Figure SMS_113
For feature set by autocorrelation operation between feature sets, +.>
Figure SMS_114
Is a feature autocorrelation calculation.
Feature screening path:
Figure SMS_115
，/>
Figure SMS_116
、/>
Figure SMS_117
and->
Figure SMS_118
Respectively representing the feature matrix after monotonicity sequencing, relevance optimization and redundancy elimination calculation operation. Fig. 2a and 2b are illustrations of feature reduction before and after feature reduction for a multi-dimensional feature of 54 items, respectivelyIt is intended that fig. 3 is a feature ordering result.
Splicing the features ranked at the front to form a multi-dimensional feature matrix
Figure SMS_119
：
Figure SMS_120
，
In the method, in the process of the invention,
Figure SMS_121
、/>
Figure SMS_122
and->
Figure SMS_123
Respectively representing a time domain feature vector, a frequency domain feature vector and a time-frequency domain feature vector which are screened by three rounds of signals, < ->
Figure SMS_124
For the expanded multidimensional feature matrix, +.>
Figure SMS_125
，/>
Figure SMS_126
Represent the firstMLine 1NThe characteristics of the columns are such that,Mthe length of the time series is indicated,Nrepresenting the number of feature vectors.
According to the principal component analysis method, the multidimensional feature matrix is obtained
Figure SMS_130
Conversion into a scoring matrix>
Figure SMS_135
，
Figure SMS_139
Wherein->
Figure SMS_129
Representing a transformation matrixATranspose of->
Figure SMS_131
The transformation matrix is composed of eigenvectors->
Figure SMS_134
Composition (S)/(S)>
Figure SMS_138
Representation->
Figure SMS_127
Middle (f)MThe elements of the time series,
Figure SMS_132
，/>
Figure SMS_136
indicating transpose,/->
Figure SMS_140
The corresponding maximum characteristic value is +.>
Figure SMS_128
. Score matrix->
Figure SMS_133
The projection vector of (a) is the main component for defining health factors, and the scoring matrix is +.>
Figure SMS_137
The first column of (2) is used as the main component, and the main component is used as the health factor matrix for driving the residual life recognition algorithm>
Figure SMS_141
. Fig. 4 is a graph showing the ratio of each component in the multi-dimensional feature in the principal component analysis.
Therefore, the firstjHealth factor of individual time series
Figure SMS_142
The method comprises the following steps:
Figure SMS_143
，
in the method, in the process of the invention,
Figure SMS_144
representing the initial characteristic value ∈ ->
Figure SMS_145
AndSthe mean and variance of the principal components, < ->
Figure SMS_146
Representing the first of the score matricesiCharacteristic vector numberjCharacteristic values corresponding to the respective time sequences, +.>
Figure SMS_147
Representing the transpose.
Step five, constructing an identification algorithm for updating parameters of the degradation model on line, wherein the index degradation model is specifically expressed as follows:
Figure SMS_148
，
in the method, in the process of the invention,
Figure SMS_151
is by training->
Figure SMS_153
The resulting health factor as a function of time, < >>
Figure SMS_155
Is an intercept term that is considered to be a constant. For easier updating of the parameters of the exponential model +.>
Figure SMS_150
Is ignored from the above equation and then the natural logarithm is taken to obtain the equation of the formula, wherein +.>
Figure SMS_152
，/>
Figure SMS_156
Is a random variable conforming to a lognormal distribution, +.>
Figure SMS_158
Is a random variable conforming to a gaussian distribution, +.>
Figure SMS_149
Is subject to->
Figure SMS_154
Distributed white gaussian noise +.>
Figure SMS_157
Representing the noise variance.
Figure SMS_159
Is->
Figure SMS_160
Variant, taking the log via an exponential degradation model, the result is reduced:
Figure SMS_161
，
at each time step (i.e. number of processes)jBased on
Figure SMS_162
Will be +.>
Figure SMS_163
And->
Figure SMS_164
The distribution of (c) is updated as a posterior.
Then, the average value of the cutter abrasion loss of the cutter on each cutting edge is used as a label to obtain a plurality of label matrixes
Figure SMS_165
；
Reading tag matrix
Figure SMS_166
And health factor->
Figure SMS_167
After the health factor data is standardized, data which accords with standard normal distribution, namely, the mean value is 0 and the variance is 1, are obtained; each tag matrix is matched to a corresponding health factor as a sample dataset.
80% of the sample data set is used as a training set, 20% is used as a test set, an algorithm network model is trained, and online dynamic update is performed
Figure SMS_168
、/>
Figure SMS_169
。
And step six, predicting the residual life of the cutter through the trained degradation model, wherein the implementation result is shown in fig. 5.
The beneficial effects of the invention are as follows: according to the milling cutter residual life prediction method based on the degradation model, the health factors subjected to fusion analysis are used as indexes for evaluating the cutter residual life, so that the cutter residual life can be evaluated more scientifically, accurately and reliably.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to the invention without departing from the principles thereof are intended to be within the scope of the invention as set forth in the following claims.

Claims (4)

1. The milling cutter residual life prediction method based on the degradation model is characterized by comprising the following steps of:
step 1: acquiring sample data comprising machining process signal data and tool wear values;
step 2: extracting signal characteristics of a time domain, a frequency domain and a time-frequency domain from the signal data in the processing process to generate massive characteristics Chi Juzhen comprising the time domain characteristics, the frequency domain characteristics and the time-frequency domain characteristics;
step 3: three rounds of signal screening are carried out on the mass feature pool matrixes according to the feature time sequence, wherein the three rounds of signal screening are monotonicity sequencing, correlation optimization and redundancy elimination respectively;
in the step 3, monotonically ordering is performed according to the following formula:
Figure FDA0004227681810000011
wherein S is 1 For feature sets ordered by monotonicity, S mon (. Cndot.) is a ranking algorithm based on the Szelman-level correlation coefficient evaluation, F {1：M} Represents F 1 To F M Is set according to the characteristic of the sequence, M represents the time sequence length, F j Is the j-th feature, t in the mass features Chi Juzhen F j For the j-th time sequence, rank (·) is the time sequence rank, corr (·) is the covariance calculation;
the correlation is preferably performed according to the formula:
Figure FDA0004227681810000012
selecting signal characteristics consistent with the change rule of the cutter abrasion value; wherein S is 2 For preference of feature sets of the ranking algorithm by correlation between two classes of variables, S corr (. Cndot.) is linear correlation calculation algorithm based on Pearson correlation coefficient judgment, cov (-) is correlation calculation, var (-) is variance calculation, F is mass feature Chi Juzhen, V B Is a cutter abrasion value vector;
performing inter-feature autocorrelation operation by the following formula, and performing redundancy elimination:
Figure FDA0004227681810000013
wherein S is 3 S is a feature set calculated by auto-correlation between feature sets self (. Cndot.) is the feature autocorrelation calculation, cov (-) is the correlation calculation, var (-) is the variance calculation, and F is the mass feature Chi Juzhen;
step 4: splicing the screened characteristics to form a multidimensional characteristic matrix, analyzing the multidimensional characteristic matrix by adopting a principal component analysis method, and taking principal components as health factors for driving the prediction of the residual life; the step 4 specifically comprises the following steps:
splicing the features which are sorted to be front after screening to form a multidimensional feature matrix F opt ：
Figure FDA0004227681810000021
Wherein T is opt 、f opt And E is opt Respectively representing time domain feature vectors, frequency domain feature vectors and time-frequency domain feature vectors which are screened by three rounds of signals,
Figure FDA0004227681810000022
for the expanded multidimensional feature matrix, +.>
Figure FDA0004227681810000023
Figure FDA0004227681810000024
Representing the features of the Mth row and the Nth column, M representing the time sequence length, and N representing the number of feature vectors;
according to the principal component analysis method, a multidimensional feature matrix F is obtained opt Conversion to a scoring matrix index, index=a T F opt In the formula, A T Represents the transpose of the transformation matrix a, a= [ a ] 1 ，A 2 ，…，A N ]The transformation matrix is composed of eigenvectors
Figure FDA0004227681810000025
Composition (S)/(S)>
Figure FDA0004227681810000026
Representation A i In M time series, i.e. [1, N ]]，[·] T Represent transpose, A i The corresponding maximum eigenvalue is lambda i ；
Score matrixThe first column of index is the main component, and the main component is used as the health factor S for residual life prediction HI Then the health factor of the j-th time series
Figure FDA0004227681810000027
The method comprises the following steps:
Figure FDA0004227681810000028
wherein lambda is oi Representing the initial eigenvalues, μ and S are the mean and variance of the principal components respectively,
Figure FDA0004227681810000029
representing eigenvalues corresponding to the jth time series of the ith eigenvector in the scoring matrix, (·) T Representing a transpose;
step 5: constructing a degradation model based on the health factors, taking the cutter abrasion value as a label, reading the label and the health factors as a sample data set, and training the degradation model;
in the step 5, the following degradation model is constructed:
Figure FDA00042276818100000210
wherein S (j) is by training
Figure FDA00042276818100000212
The resulting health factor, phi, as a function of time, represents the intercept term,
Figure FDA00042276818100000211
θ (j) is a random variable conforming to a lognormal distribution, β (j) is a random variable conforming to a gaussian distribution, ε represents Gaussian white noise, σ 2 Representing the noise variance;
taking logarithm of the degradation model and then reducing to obtain the following components:
Figure FDA0004227681810000031
wherein L (j) is a variant S (j); updating the distribution of θ (j) and β (j) to a posterior based on the results of L (j) at each time step;
the process of training the degradation model is as follows:
the tool wear value of each blade in the sample data is averaged and then used as a label to obtain a plurality of label matrixes V B ；
Reading tag matrix V B And health factor S HI Carrying out standardized processing on the health factors to obtain data conforming to standard normal distribution, matching each tag matrix with the corresponding health factors to be used as a sample data set, dividing the sample data set into a training set and a testing set, training a degradation model, and dynamically updating theta (j) and beta (j);
step 6: and predicting the residual life of the cutter according to the trained degradation model.
2. The degradation model-based milling tool remaining life prediction method according to claim 1, wherein: in the step 1, the machining process signal data includes a vibration signal, a load current, a spindle power signal and PLC information.
3. The degradation model-based milling tool remaining life prediction method according to claim 1, wherein: the step 2 specifically comprises the following steps:
in the time domain, 13 characteristics of gap coefficient, crest coefficient, pulse coefficient, kurtosis, average value, peak value, root mean square, distortion rate, signal to noise ratio, shape factor, deflection, standard deviation and total harmonic distortion are calculated on the processing signal data to form a time domain characteristic pool T, T= [ T ] (1) ，T (2) ，…，T (l) ]，T (l) Representing a first time domain feature, l representing the number of time domain features;
in the frequency domain, processing is signaledThe number data calculates 5 characteristics of peak amplitude, peak frequency, natural frequency, damping coefficient and band power to form a frequency domain characteristic pool f, f= [ f ] (1) ，f (2) ，…，f (m) ]，f (m) Representing the mth frequency domain feature, m representing the number of frequency domain features;
on a time-frequency domain, obtaining time-frequency domain signal characteristics based on wavelet packet decomposition processing process signal data to form a time-frequency domain characteristic pool E, E= [ E ] (1) ，E (2) ，…，E (n) ]，E (n) Representing the nth time-frequency domain feature, n representing the number of the time-frequency domain features;
integrating the time domain feature pool, the frequency domain feature pool and the time-frequency domain feature pool to form massive features Chi Juzhen F, F= { T (1) ，T (2) ，…，T (l) ，f (l+1) ，…，f (l+m) ，E (l+m+1) ，…，E (l+m+n) }。
4. The degradation model-based milling tool remaining life prediction method according to claim 1, wherein: 80% of the sample dataset was used as training set and 20% as test set.
CN202310206684.XA 2023-03-07 2023-03-07 Milling cutter residual life prediction method based on degradation model Active CN116070527B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310206684.XA CN116070527B (en) 2023-03-07 2023-03-07 Milling cutter residual life prediction method based on degradation model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310206684.XA CN116070527B (en) 2023-03-07 2023-03-07 Milling cutter residual life prediction method based on degradation model

Publications (2)

Publication Number Publication Date
CN116070527A CN116070527A (en) 2023-05-05
CN116070527B true CN116070527B (en) 2023-06-16

Family

ID=86176956

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310206684.XA Active CN116070527B (en) 2023-03-07 2023-03-07 Milling cutter residual life prediction method based on degradation model

Country Status (1)

Country Link
CN (1) CN116070527B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116822383A (en) * 2023-08-31 2023-09-29 成都态坦测试科技有限公司 Equipment life prediction model construction method and device, readable storage medium and equipment
CN117033912B (en) * 2023-10-07 2024-02-13 成都态坦测试科技有限公司 Equipment fault prediction method and device, readable storage medium and electronic equipment
CN118013808B (en) * 2024-04-09 2024-07-09 中国海洋大学 Method for identifying looseness health state of bolt of offshore wind power generation structure

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108959778B (en) * 2018-07-06 2020-09-15 南京航空航天大学 Method for predicting residual life of aircraft engine based on consistency of degradation modes
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
CN113934982B (en) * 2021-10-18 2024-04-19 河北工业大学 Method for predicting mechanical life of breaker operating mechanism based on vibration-electric signal fusion
CN114819315A (en) * 2022-04-17 2022-07-29 北京化工大学 Bearing degradation trend prediction method based on multi-parameter fusion health factor and time convolution neural network

Also Published As

Publication number Publication date
CN116070527A (en) 2023-05-05

Similar Documents

Publication Publication Date Title
CN116070527B (en) Milling cutter residual life prediction method based on degradation model
CN109158953B (en) A kind of cutting-tool wear state on-line monitoring method and system
CN112257530B (en) Rolling bearing fault diagnosis method based on blind signal separation and support vector machine
CN110509109B (en) Cutter wear monitoring method based on multi-scale depth convolution cyclic neural network
CN111687689A (en) Cutter wear state prediction method and device based on LSTM and CNN
CN110303380A (en) A kind of cutting tool for CNC machine method for predicting residual useful life
CN107368809A (en) A kind of bearing fault sorting technique based on rarefaction representation and dictionary learning
CN113378725B (en) Multi-scale-channel attention network-based tool fault diagnosis method, equipment and storage medium
CN114619292B (en) Milling cutter wear monitoring method based on fusion of wavelet denoising and attention mechanism with GRU network
CN108776017B (en) Method for predicting residual life of rolling bearing by improving CHSMM
CN113496312A (en) Cutter residual life prediction method and equipment based on multi-dimensional feature extraction fusion and long-term and short-term memory network and storage medium
CN112207631B (en) Method for generating tool detection model, method, system, device and medium for detecting tool detection model
CN114429152A (en) Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption
CN109434562A (en) Milling cutter state of wear recognition methods based on partition clustering
CN114749996A (en) Tool residual life prediction method based on deep learning and time sequence regression model
CN108393744B (en) Multi-sensing monitoring method for cutter state
Kumar et al. Performance evaluation for tool wear prediction based on Bi-directional, Encoder–Decoder and Hybrid Long Short-Term Memory models
CN114354184A (en) Deep learning-based method and device for establishing health early warning model of main shaft of large-scale rotating equipment
CN114871850B (en) Tool wear state assessment method based on vibration signals and BP neural network
CN113504768B (en) High-precision product digital twin computability method for assembly quality prediction
Wang et al. Self-adaptive fusion of local-temporal features for tool condition monitoring: A human experience free model
CN110108474A (en) A kind of rotating machinery operation stability on-line monitoring and appraisal procedure and system
CN107341503B (en) Identification method for multi-source energy efficiency state in cutting process
CN115392292A (en) Cutter abrasion on-line monitoring method based on attention circulation neural network
CN112966449A (en) Manufacturing process quality defect root cause diagnosis method based on variable selection and probabilistic neural network

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