CN114536104B - Dynamic prediction method for tool life - Google Patents
Dynamic prediction method for tool life Download PDFInfo
- Publication number
- CN114536104B CN114536104B CN202210299596.4A CN202210299596A CN114536104B CN 114536104 B CN114536104 B CN 114536104B CN 202210299596 A CN202210299596 A CN 202210299596A CN 114536104 B CN114536104 B CN 114536104B
- Authority
- CN
- China
- Prior art keywords
- data
- cutter
- tool
- factors
- life prediction
- 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
Images
Classifications
-
- 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
- B23Q17/0995—Tool life management
-
- 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
- G06F17/15—Correlation function computation including computation of convolution operations
-
- 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)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Mechanical Engineering (AREA)
- Machine Tool Sensing Apparatuses (AREA)
Abstract
The invention relates to the field of tool life prediction, and discloses a tool life dynamic prediction method, which comprises the following steps: s1, determining characteristics influencing the service life of a cutter, collecting related information data to obtain historical data, and carrying out standardization processing on the historical data; s2, performing correlation analysis on the historical data, and deleting the characteristics of the correlation in a critical range; s3, performing principal component analysis on the features, and performing dimensionality reduction and simplification on historical data to obtain modeling data; s4, training the modeling data by using a gradient lifting regression tree, and establishing a tool life prediction model; s5, acquiring real-time data according to the characteristics of the modeling data, carrying out standardization processing on the real-time data, inputting the real-time data into a tool life prediction model, and outputting to obtain the tool life; according to the dynamic cutter life prediction method, information data influencing the cutter life are optimized, a perfect cutter life prediction model is established, and the accuracy of cutter life prediction is further improved.
Description
Technical Field
The invention relates to the field of tool life prediction, in particular to a dynamic tool life prediction method.
Background
The cutter is the important component of machine tool machining, in the course of working, the cutter constantly wears out, when wearing and tearing reach certain degree, continue processing and can lead to the part to appear the quality problem, the prediction of cutter life-span is the difficult problem that puzzles machine tool machining for a long time, can cause the cutter extravagant when cutter in-service use time is less than the cutter life-span, part manufacturing cost increases, can bring huge quality risk when cutter in-service use time is greater than the cutter life-span, part milling damage, the scheduling problem of being out of tolerance probably appears, cause the part to scrap.
The tool life in the industry at the present stage is still determined by experience, when the same tool is subjected to rough machining, the rotating speed is high, the large feed is realized, the cutting depth is large, the cutting width is large, the tool is abraded quickly, and when the fine machining is performed, the rotating speed is low, the feed is low, the cutting depth is small, the cutting width is small, the tool is abraded slowly, in order to ensure the machining quality of parts and the machining efficiency of batch production, the conservative tool life is generally adopted for judgment, namely, the tool is scrapped and replaced when the cutting time of the tool reaches the theoretical life of the tool, so that the tool is scrapped and replaced when the tool is damaged slightly, and a great amount of waste of the tool is caused.
The patent CN109465676B discloses a tool life prediction method, wherein only current signals are collected, factors including temperature, process parameters, equipment, part materials and the like are not considered, and the established model is difficult to accurately predict the tool life under various working conditions; patent CN108427841A discloses a method for predicting the life of a tool in real time, which divides the machining process into working subintervals according to different machining conditions, only considers signals acquired by a machine tool machining sensor, does not consider the wear conditions of different edges of the tool due to different part characteristics, and takes an end mill as an example: when the cutting tool is used for cutting, only the bottom corner is used, and after the bottom corner is damaged, the side edge of the cutting tool is intact, and the operation of milling the appearance of the part can still be performed.
Disclosure of Invention
The invention aims to: aiming at the problems that in the cutter service life prediction method in the prior art, the established cutter service life prediction model is not perfect enough and the cutter service life is difficult to predict accurately, the cutter service life dynamic prediction method is provided, the more perfect cutter service life prediction model is established by optimizing information data influencing the cutter service life, the effective dynamic prediction of the cutter service life can be carried out according to the acquired real-time data, and the accuracy of the cutter service life prediction is further improved.
In order to achieve the purpose, the invention adopts the technical scheme that:
a dynamic tool life prediction method is characterized by comprising the following steps:
s1: determining characteristics influencing the service life of the cutter, collecting related information data to obtain historical data, and carrying out standardization processing on the historical data;
s2: carrying out correlation analysis on the historical data, and deleting the characteristics of the correlation in a critical range;
s3: performing Principal Component Analysis (PCA) on the characteristics, and performing dimensionality reduction and simplification on historical data to obtain modeling data;
s4: training modeling data by using a Gradient Boosting Regression Tree (GBRT) and establishing a tool life prediction model;
s5: and acquiring real-time data according to the characteristics of the modeling data, standardizing the real-time data, inputting the standardized real-time data into a tool life prediction model, and outputting to obtain the tool life.
The historical data is used for establishing a cutter life model, the real-time data is used for dynamically predicting the cutter life, and the standardization processing of the steps S1 and S5 is to scale the data in proportion to enable the data to fall into a small specific interval, so that indexes of different units or orders of magnitude can be compared and weighted conveniently; the critical range of the step S2 is used for judging the correlation degree among the characteristics, can be set according to specific working conditions, avoids the deviation of the tool life prediction model to sampling highly correlated characteristics through the data processing of the step S2, and ensures the generalization performance of the tool life prediction model; because one input datum usually contains dozens of characteristics, the influence of each characteristic on the service life of the cutter is difficult to judge according to experience, the method is suitable for unsupervised learning and completely free of parameter limitation, principal component analysis does not need to consider set parameters in the calculation process or intervene in calculation according to any empirical model, and after dimension reduction is carried out on the datum through the principal component analysis, the influence of each characteristic on the service life of the cutter can be effectively judged through the step S3, so that the historical data can be reasonably simplified, the modeling datum is the datum obtained after dimension reduction of the historical datum and omission of some characteristics, and the obtained modeling datum is convenient for establishing a cutter service life prediction model; finally, training the modeling data through the step S4, so that the accuracy of establishing a tool life prediction model is improved; therefore, through the mutual cooperation of the steps S1 to S4, the information data influencing the service life of the cutter is optimized, and meanwhile, the establishment of a cutter service life prediction model is optimized.
According to the dynamic prediction method for the service life of the cutter, disclosed by the invention, a more complete cutter service life prediction model is established by optimizing the information data influencing the service life of the cutter, the service life of the cutter can be effectively and dynamically predicted according to the acquired real-time data, and the accuracy of the cutter service life prediction is further improved.
Preferably, the characteristics in step S1 are divided into part factors, equipment factors, process factors, cutter factors and environmental factors, the part factors, the equipment factors and the process factors are collected based on a production line management system, the cutter factors are collected based on a cutter management system, and the environmental factors are collected based on a sensor.
Through dividing into four main factors with the characteristic that influences the cutter life-span, select the collection mode that corresponds according to different factors, can effectively improve data acquisition's efficiency, avoided the signal source singleness of gathering, the incomplete problem of information of gathering.
Preferably, the information data related to the process factors are collected in a divided manner according to an NC (Numerical Control) program.
In numerical control machining, one surface of a machined part can be provided with a plurality of cutters, usually, one NC program corresponds to one cutter, and the NC program comprises a plurality of subprograms.
Preferably, the characteristics of the part factors comprise the material and the clamping mode of the part; the characteristics of the equipment factors comprise a main shaft type, an equipment type and an equipment model; the characteristics of the process factors comprise part characteristics, material removal amount, cutting depth, cutting width, rotating speed and feeding amount; the characteristics of the cutter factors comprise cutter category, cutting edge material, cutter diameter, base angle radius, cutter tooth number, cutter structure, inner cooling, cutter handle type and cutter abrasion loss; characteristics of environmental factors include temperature and vibration.
The selection range corresponding to the characteristics in the step S1 is determined, the information quantity of the acquired data is improved, and the accuracy of the tool life model for predicting the tool under various working conditions is further improved.
Preferably, the part material includes: aluminum alloy and titanium alloy, the clamping mode includes: screw compress tightly, clamp plate compresses tightly, counter-draws, vacuum adsorption, magnetic force adsorb, and the part characteristic includes: ribs, flanges, webs, corners, holes and special-shaped surfaces; the types of tools include: end milling cutters, slotting cutters, conical cutters, chamfering cutters, drill bits, reamers, boring cutters, and the like; the cutting edge material comprises: cemented carbide, diamond, high speed steel, ceramics, etc.; the cutter structure includes: integral knife, welding knife, and indexable knife.
The characteristic classification influencing the service life of the cutter is further refined, and the content of the acquired information data is improved, so that the establishment of a cutter service life prediction model is further perfected.
Preferably, when the historical data is collected, the tool wear amount is the tool wear amount δ k of the current processing,
δk＝k cur -k last ，
k cur is the current tool wear, k last The tool wear amount is the tool wear amount when the last machining is completed;
when real-time data is collected, the abrasion loss of the cutter is the residual abrasion loss delta k',
δk’＝k max -k cur
k cur is the current tool wear, k max Is the maximum cutter wear;
the tool wear amount includes: and selecting the corresponding tool abrasion loss to participate in data processing according to the use working condition of the tool.
The tool wear of the same tool is different under different working conditions, so each acquisitionWhen historical data are acquired, the tool wear amount of the current processing is obtained through calculation, the tool wear amount of the current processing is taken as characteristic data to be collected, and when the tool is a new tool, the tool wear amount k of the last processing is obtained last ＝0；
When the real-time data is collected to predict the service life of the cutter, the wear loss of the residual cutter needs to be obtained through calculation, the wear loss of the residual cutter is taken as characteristic data to be collected, and when the cutter is a new cutter, the current wear loss k of the cutter is obtained cur ＝0；
According to the service life test of the end mill in the national standard GB/T16460-2016, the judgment of the service life of the cutter is based on a preset numerical value of certain form of abrasion of the cutter, as shown in figure 1, the wear bandwidth value VB of a common flank surface, the wear KT of a front flank surface and the wear of a base angle are taken as judgment bases, the relationship between the wear loss of the cutter and the cutting time under different working conditions is different, but the general trends are basically the same, as shown in figure 2, the wear loss of the cutter is distinguished according to different cutting edges of the cutter, the wear loss of the corresponding cutting edge is selected according to different working conditions for prediction, and the accuracy and the rationality of the service life prediction of the cutter can be effectively improved.
Preferably, the method for collecting the tool wear amount is as follows: and (4) acquiring images of the tool after the characteristics of each part are processed in the processing area of the machine tool, and identifying and processing the images to obtain the data of the tool abrasion loss.
The data of the abrasion loss of the cutter is acquired by a method of image acquisition in a processing area, the abrasion loss of the cutter after part characteristics are processed every time can be accurately acquired, and the normal use of the cutter cannot be influenced.
Preferably, in step S1 and step S5, the normalization process includes removing a dimension, filling the blank value, and removing an abnormal value, wherein the removing dimension is transformed by the following formula:
wherein, x' is data after dimension removal, mu is a data mean value, and sigma is a data standard deviation;
filling the acquired data which does not belong to normal distribution with the median of the data when filling the blank value;
in step S2, the correlation is calculated using the pearson correlation coefficient:
wherein
And sigma X The standard fraction, the sample average value and the sample standard deviation of the Xi samples are respectively shown, r is the Pearson correlation coefficient, and the critical range is the range of the Pearson correlation coefficient between 0.9 and 1.
Blank value means that corresponding data of some characteristics in the collected data are missing, if the missing data do not belong to normal distribution, such as process parameters, tool abrasion loss, tool parameters and the like, the median of indexes capable of better representing the trend of the data center is selected from historical data for filling, and the effectiveness of filling data is improved;
because the units of different variables are different, after the dimension is eliminated, the unit limit of the data is removed, and the data is converted into a dimensionless pure numerical value, so that indexes of different units or orders of magnitude can be compared and weighted.
Because the phenomenon that two variables have common variation trend exists between a plurality of pairs of variables in the input data, namely, correlation exists between a plurality of characteristics, such as characteristics of tool abrasion amount and material removal amount, tool abrasion amount and cutting depth, tool abrasion amount and rotating speed and the like, which can mutually influence each other;
the features with normal distribution relation are extracted, the correlation among the features can be intuitively judged by adopting the Pearson correlation coefficient, the high-correlation features with the Pearson correlation coefficient between 0.9 and 1 are deleted, the model is prevented from being biased to sample the high-correlation features, and the generalization performance of the tool life prediction model is ensured.
Preferably, in step S3, the principal component analysis is to calculate the principal component contribution rate and the cumulative totalCalculating the contribution rate; the calculation formula of the P-th principal component is as follows: f p ＝a 1i *Z x1 +a 2i *Z x2 +a 3i *Z x3 +…+a pi *Z xp Principal component Z i The contribution ratio of (c):
cumulative contribution rate:
taking a characteristic value lambda of which the accumulated contribution rate reaches 85 to 95 percent 1 ，λ 2 ，λ 3 ，…λ m The corresponding first, second, 8230th main component (m is less than or equal to P);
wherein a is 1i ，a 2i ，...，a pi (i = 1.. Multidot., p) is a feature vector corresponding to a feature value of the covariance matrix Σ of X, Z x1 ，Z x2 ，...，Z xp Is data subjected to a standardization process, F p Is the pth principal component.
The dimension of the historical data can be reduced through principal component analysis, and simultaneously, the characteristics with larger influence are arranged in front, F p The larger the information is, the more important the former part can be selected as required in actual operation, and the later digit is omitted, namely P main components with the cumulative contribution rate of 85% -95% are selected, so that the historical data can be simplified, the effect of compressing the historical data is achieved, and the required modeling data is obtained after the dimension of the historical data is reduced and partial features are omitted.
Preferably, the gradient boost regression tree includes a regression tree and a gradient boost, and the model of the regression tree is:
m is the division of the data set into M units, c m Is the output value of the Mth cell, I (x ∈ R) m ) Is an indicator function, if x ∈ R m I =1, otherwise I =0.
The training process is carried out byThe iteratively obtained learner is f t 1 (x), the loss function is L (y, f) t -1 (x)), and then iterating to find the weak learner h of the regression tree model t (x) The GBRT consists of a plurality of decision trees, and the conclusions of all the decision trees are accumulated to obtain a result.
As the predicted tool life is a continuous value, GBRT can be selected to carry out effective training, the regression tree is to carry out node segmentation and tree building through information gain, and the building of a tool life prediction model can be optimized through the training of the GBRT.
Preferably, the modeling data is divided into training data and testing data, the training data is used for training the gradient boosting regression tree, and after step S4, before step S5, the following steps are further included:
a. testing the tool life prediction model by using the test data;
b. judging the accuracy of the tool life prediction model according to the fitting goodness result, if the accuracy does not meet the requirement, performing parameter optimization, and repeating the step a and the step b until the accuracy meets the requirement;
the parameter optimization is to optimize the gradient lifting regression tree frame parameters, the goodness of fit result is a result of testing by using test data, and the established tool life prediction model can be further optimized through testing and optimization of the test data, so that the accuracy of the tool life prediction model is improved.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
1. according to the dynamic prediction method for the service life of the cutter, the information data influencing the service life of the cutter are optimized, a more complete cutter service life prediction model is established, the service life of the cutter can be effectively and dynamically predicted according to the collected real-time data, and the accuracy of the service life prediction of the cutter is further improved;
2. the characteristics influencing the service life of the cutter are divided into four main factors, and the corresponding acquisition mode is selected according to different factors, so that the data acquisition efficiency can be effectively improved, the problems of single acquired signal source and incomplete acquired information are solved, the characteristics of the four main factors are comprehensive, and the established cutter service life prediction model can be suitable for various working conditions;
3. after historical data acquired by dividing the NC program are substituted into training or testing, the accuracy of the tool life prediction model can be improved;
4. the cutter abrasion loss is distinguished according to different cutting edges of the cutter, and the abrasion loss of the corresponding cutting edge is selected according to different working conditions for prediction, so that the accuracy and the rationality of the cutter service life prediction can be effectively improved;
5. by sequentially carrying out correlation analysis, principal component analysis and training of a gradient lifting regression tree on historical data, a tool life prediction model is optimized;
6. the dynamic cutter life prediction method reduces cutter waste, reduces the processing cost of parts, improves the processing quality and efficiency of the parts, provides a basis for evaluating the cutting capacity of a tool, provides data support for a cutter purchasing plan and improves the management level of enterprises.
Drawings
FIG. 1 is a schematic view of end mill rake face wear;
FIG. 2 is a graph of tool wear versus tool life;
FIG. 3 is a flow chart of a method for dynamically predicting tool life according to an embodiment;
reference numerals: 1-front tool face, 2-back tool face and 3-base angle.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings.
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and embodiments, it being understood that the specific embodiments described herein are only for the purpose of explaining the present invention and are not intended to limit the present invention.
Examples
As shown in fig. 3, the method for dynamically predicting the tool life according to the present invention includes the following steps:
s1: determining characteristics influencing the service life of the cutter, collecting related information data to obtain historical data, and carrying out standardization processing on the historical data;
s2: carrying out correlation analysis on the historical data, and deleting the characteristics of the correlation in a critical range;
s3: performing principal component analysis on the characteristics, and performing dimensionality reduction and simplification on historical data to obtain modeling data;
s4: training the modeling data by using a gradient lifting regression tree, and establishing a tool life prediction model;
s5: and acquiring real-time data according to the characteristics of the modeling data, standardizing the real-time data, inputting the standardized real-time data into a tool life prediction model, and outputting to obtain the tool life.
In the embodiment, the service life of the cutter for machining the aeronautical structural part is predicted,
step S1 is carried out, characteristics influencing the service life of the cutter are determined, and the characteristics are divided into part factors, equipment factors, process factors, cutter factors and environmental factors;
the characteristics of part factor include part material, clamping mode to aviation structure spare is the example, and the part material mainly includes: aluminum alloy and titanium alloy, the clamping mode mainly includes: screw pressing, pressing plate pressing, reverse drawing, vacuum adsorption and magnetic adsorption, and different clamping modes can cause different part stresses;
the characteristics of the equipment factors comprise a main shaft type, an equipment type and an equipment model, wherein the main shaft type mainly comprises a mechanical main shaft, an electric main shaft and the like;
the characteristics of the process factors comprise part characteristics, material removal amount, cutting depth, cutting width, rotating speed and feeding amount, wherein the part characteristics comprise: ribs, rims, webs, corners, holes, irregular surfaces, and the like;
the characteristics of cutter factor include cutter category, blade material, cutter diameter, base angle radius, cutter number of teeth, cutter structure, whether interior cold, handle of a knife type, cutter producer, cutter wearing and tearing volume to aviation structure for the example, wherein, the cutter type includes: end mills, slotting mills, conical cutters, chamfering cutters, drill bits, reamers, boring cutters, and the like; the cutting edge material comprises: cemented carbide, diamond, high speed steel, ceramics, etc.; the cutter structure includes: a solid knife, a welding knife, an indexable knife, etc.; the types of the tool shank include: a cylindrical straight handle, an integrated knife, a spiral locking groove and the like;
characteristics of environmental factors include temperature and vibration.
Under the condition of reducing the influence on the machining efficiency of the parts as much as possible, acquiring information data corresponding to characteristics possibly influencing the service life of the cutter, taking the unit as an example, and acquiring part factors, equipment factors and process factors based on a production line management and control system;
wherein, the process factors are divided and collected according to characteristics according to an NC program; in numerical control machining, it is common to machine one side of a part as a continuous process, such as: c0001.MPF — c000n. MPF (MPF is a suffix of NC program), which are called by a main program continuously, and the surface is processed after all the sub programs are finished, wherein a plurality of knives are possible to process one surface, and one knife is usually changed to be an NC program during processing, but the process parameters in the same sub program may be different, for example: the rotation speed and feed can be changed when different characteristics are milled;
in order to control variables as much as possible and make the relation between the variables as simple as possible so as to find the law therein, when historical data are collected in an experiment, the historical data are divided into a subprogram according to a cutter, a part characteristic and a process parameter, wherein the process parameter is a cutting depth, a rotating speed, a feeding amount and other processing parameters, the historical data are collected by dividing into a single piece according to a single subprogram in an NC program, and only one or more characteristics of each piece of historical data can be changed, for example, the task of the surface is to mill a web: and C0001.MPF is to rapidly mill a slot cavity, has high rotating speed, large feeding and large material removal amount, and C0002.MPF mills 4 rotating angles with the same cutter, reduces the feeding and has small material removal amount, so that two pieces of data are acquired.
In addition, the collection of the material removal amount can obtain the size of a raw material (blank) through an MES system or a part manufacturing outline, obtain the material removal amount of the current processing through process simulation, and obtain the material removal amount which is not processed for the first time through accumulative subtraction.
The tool factors are collected based on a tool management system, wherein the tool wear loss can be collected by installing a high-power image collecting camera around a numerical control machine tool machining area to collect the top end image of the tool after machining of each feature, a main shaft rotates for one circle to take a picture, and the tool wear loss value is further identified and collected by using an image identification algorithm;
when historical data is collected, the tool wear amount is obtained by collecting data according to the tool wear amount delta k of the current processing, and the tool wear amount delta k of the current processing is calculated according to the following formula:
δk＝k cur -k last ，
k cur is the current tool wear, k last The tool wear amount is the tool wear amount when the last machining is completed;
if the cutter is a new cutter initially, the wear k of the cutter is measured when the previous machining is finished last =0, the tool wear amount of the current machining is the current tool wear amount k when historical data are collected cur (ii) a For example, when a new cutter executes task 1, the current wear amount and the current task wear amount are both 0.1; when task 2 is executed, the abrasion loss delta k = k of the task at this time cur -k last 0.180-0.100=0.08; when executing task 3, the abrasion loss delta k = k of the task at this time cur -k last =0.220-0.180=0.04, and so on, as shown in table 1;
TABLE 1 calculation of wear of this task for cutter
When the real-time data is collected for predicting the service life of the cutter, the cutter abrasion loss is collected by the residual cutter abrasion loss, and the residual cutter abrasion loss delta k' is calculated by the following formula:
δk’＝k max -k cur
k cur is the current tool wear, k max Is the maximum cutter wear;
maximum wear k of the tool max The tool wear amount k can be obtained through historical data of the tool, and if the tool is initially a new tool, the current tool wear amount k is obtained cur =0, and the residual cutter wear delta k' is the maximum cutter wear k max ；
The machining part characteristics and the tool abrasion position have a relationship, so the tool abrasion loss is divided into: selecting corresponding tool wear amount according to the use working condition of the tool to participate in data processing, wherein a drill bit, a boring tool, a conical tool and a plunge milling tool in the tool generally use a base angle to process part materials and use the base angle wear amount as a characteristic to evaluate by taking an aviation structural part as an example; reamers, reamers and chamfers are generally machined by using side blades, and the wear amount of a rear cutter face is used as a characteristic for evaluation; when the end mill is used for processing a web plate or performing row cutting operation and the cutting depth is smaller than a base angle, the base angle is generally used for processing, when the end mill is used for processing ribs, rims and corners, a side edge is generally used for processing, and base angle abrasion loss and flank face abrasion loss are selected as characteristics for different characteristics for evaluation;
the environmental factors are collected based on a sensor, for example, a vibration sensor of a machine tool additionally provided with a temperature sensor collects corresponding data;
and finally, taking data collected in experiments or previous processing as historical data.
Further, the historical data is subjected to standardization: filling the missing of data corresponding to some characteristics in the collected historical data, if the missing data does not belong to normal distribution, such as process parameters, tool abrasion amount, tool parameters and the like, selecting an index median which can better represent the trend of a data center in the historical data for filling, further eliminating dimensions of the historical data, and performing data transformation as follows:
wherein x' is data after dimension removal corresponding to each feature, mu is a mean value of data acquired corresponding to each feature, and sigma is a standard deviation of the data acquired corresponding to each feature;
and further checking abnormal values of the data, and deleting the abnormal values with too large or too small data.
Step S2, carrying out correlation analysis on the historical data, and deleting the characteristics of the correlation in a critical range; and calculating the correlation by using a Pearson correlation coefficient, wherein if the covariance is a positive value, the two variables are positively correlated, and if the covariance is 0, the two variables are not correlated, and the calculation formula is as follows:
r can be represented by (X) i ,Y i ) And estimating the standard fraction mean value of the sample points to obtain an expression equivalent to the formula:
wherein
And sigma X The correlation coefficient is determined to be within a range of 0.9 to 1, and the features having a correlation between features of 0.9 to 1 are deleted.
Step S3, performing principal component analysis on the characteristics, and performing dimensionality reduction and simplification on historical data to obtain modeling data; in step S1, P features are determined in total, and principal component analysis is performed on the P features, where the calculation formula of the pth principal component is: f p ＝a 1i *Z x1 +a 2i *Z x2 +a 3i *Z x3 +…+a pi *Z xp Calculating the contribution rate of the principal component and the accumulated contribution rate; principal component Z i The contribution ratio of (c):
cumulative contribution rate:
taking a characteristic value lambda of which the accumulated contribution rate reaches 85-95% 1 ，λ 2 ，λ 3 ，…λ m The corresponding first, second, \ 8230, the m (m is less than or equal to P) major components;
wherein a is 1i ，a 2i ，...，a pi (i = 1.. Multidot., P) is a feature vector corresponding to a feature value of the covariance matrix Σ of X, Z x1 ，Z x2 ，...，Z xp Is data subjected to standardization, F p Is the pth principal component.
S4, training the modeling data by using GBRT, and establishing a tool life prediction model; GBRT includes a regression tree and gradient boosting, the model of the regression tree is:
m is the division of the data set into M units, c m Is the output value of the Mth cell, I (x ∈ R) m ) Is an indicator function, if x ∈ R m I =1, otherwise I =0;
suppose that the learner from the previous iteration is f t 1 (x), the loss function is L (y, f) t 1 (x)), the objective of the iteration is to find a weak learner h of the regression tree model t (x) And the GBRT consists of a plurality of decision trees, and the conclusions of all the decision trees are accumulated to obtain a result, so that a tool life prediction model is established.
At this time, before the next step is performed, the modeling data may be divided into training data and test data, the training data is used for GBRT training, and the tool life prediction model is tested using the test data;
judging the accuracy of the tool life prediction model according to the fitting goodness result, and if the accuracy does not meet the requirement, performing parameter optimization;
and repeating the testing and parameter adjusting until the accuracy of the tool life prediction model meets the requirement.
And S5, acquiring real-time data according to the characteristics of the modeling data, performing the same standardization processing on the real-time data as historical data, inputting the tool life prediction model, and outputting to obtain the tool life.
In addition, in practical application, when the same machine tool adopts the same process scheme and the same cutter to the same batch of parts and the same environment is used for processing, the main factor influencing the service life of the cutter is the cutter abrasion loss, a high-power optical microscope needs to be additionally arranged on the cutter abrasion loss, and the abrasion loss is collected by utilizing an image recognition technology. The changes of other part factors, equipment factors, process factors, cutter factors and environmental factors are ignored, so that before each machining, only the cutter abrasion loss is collected, the calculation of the maximum abrasion loss is used as the input of a model, and the dynamic prediction result of the cutter service life can be output by substituting into a formula.
In the distributed processing process, the parts are usually put into production in various small batches, and large-scale batch production cannot be realized. Besides the tool wear amount, part factors, process factors and tool factors can be obtained in information systems such as manufacturing execution systems MES, enterprise resource planning ERP and the like, environmental factors can be collected through sensors and are simple, a trained model is brought, and a dynamic prediction result of the tool service life can also be obtained. The method has the advantages of strong practicability, high accuracy and popularization and application values.
The above description is intended to be illustrative of the preferred embodiment of the present invention and should not be taken as limiting the invention, but rather, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.
Claims (6)
1. A dynamic tool life prediction method is characterized by comprising the following steps:
s1: determining characteristics influencing the service life of the cutter, collecting related information data to obtain historical data, and performing standardization processing on the historical data;
s2: carrying out correlation analysis on the historical data, and deleting the characteristics of the correlation in a critical range;
s3: performing principal component analysis on the characteristics, and performing dimensionality reduction and simplification on historical data to obtain modeling data;
s4: training the modeling data by using a gradient lifting regression tree, and establishing a tool life prediction model;
s5: acquiring real-time data according to the characteristics of the modeling data, carrying out standardization processing on the real-time data, inputting the real-time data into a tool life prediction model, and outputting to obtain the tool life;
the characteristics in the step S1 are divided into part factors, equipment factors, process factors, cutter factors and environment factors, wherein the part factors, the equipment factors and the process factors are collected based on a production line management system, the cutter factors are collected based on a cutter management system, and the environment factors are collected based on a sensor;
the part factors are characterized by comprising the material and the clamping mode of the part; the characteristics of the equipment factors comprise a main shaft type, an equipment type and an equipment model; the characteristics of the process factors comprise part characteristics, material removal amount, cutting depth, cutting width, rotating speed and feeding amount; the characteristics of the cutter factors comprise cutter category, cutting edge material, cutter diameter, base angle radius, cutter tooth number, cutter structure, inner cooling, cutter handle type and cutter abrasion loss; characteristics of environmental factors include temperature and vibration;
in step S1 and step S5, the normalization process includes removing a dimension, filling a blank value, and removing an abnormal value, wherein the removing dimension is transformed by the following formula:
wherein x' is the data after dimension removal,
is the mean value of the data and is,
is the standard deviation of the data;
filling the acquired data which do not belong to normal distribution with the median of the data when filling the blank value;
in step S2, the correlation is calculated using the pearson correlation coefficient:
wherein
And
are respectively a pair
The standard fraction, the average value and the standard deviation of the sample, wherein r is a Pearson correlation coefficient, and the critical range is the range of the Pearson correlation coefficient from 0.9 to 1;
in step S3, principal component analysis is to calculate the principal component contribution rate and the accumulated contribution rate; the calculation formula of the P-th principal component is as follows:
principal component of
The contribution ratio of (c):
cumulative contribution rate:
taking the characteristic values lambda 1, lambda 2, lambda 3, \ 8230 # corresponding to lambda m with the cumulative contribution rate of 85-95%The m (m is less than or equal to P) main components;
wherein
(i =1,2, \ 8230; p) is a covariance matrix of X
The feature vector corresponding to the feature value of (a),
is the data that has been subjected to the normalization process,
is the pth principal component.
2. The dynamic tool life prediction method as claimed in claim 1, wherein the information data related to the process factors are collected according to the NC program in a divided manner.
3. The method of claim 1, wherein the tool wear amount is the tool wear amount of the current machining when the historical data is collected
，
when real-time data is collected, the tool wear loss is the residual tool wear loss
，
the tool wear amount includes: and selecting the corresponding tool abrasion loss to participate in data processing according to the use working condition of the tool.
4. The dynamic tool life prediction method of claim 1, wherein the method for collecting the tool wear amount comprises the following steps: and (4) carrying out image acquisition on the tool after the characteristics of each part are processed in the processing area of the machine tool, and identifying and processing the image to obtain the data of the tool abrasion loss.
5. The dynamic tool life prediction method of claim 1, wherein the gradient boosting regression tree comprises a regression tree and a gradient boosting, and the model of the regression tree is:
m is the division of the data set into M cells,
is the output value of the mth cell,
is an indication function, if
Then I =1, otherwise I =0.
6. The dynamic tool life prediction method of claim 1, wherein the modeling data is divided into training data and testing data, the training data is used for training a gradient boosting regression tree, and after step S4 and before step S5, the method further comprises the following steps:
a. testing the tool life prediction model by using the test data;
b. and (4) judging the accuracy of the tool life prediction model according to the fitting goodness result, if the accuracy does not meet the requirement, performing parameter optimization, and repeating the step a and the step b until the accuracy meets the requirement.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210299596.4A CN114536104B (en) | 2022-03-25 | 2022-03-25 | Dynamic prediction method for tool life |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210299596.4A CN114536104B (en) | 2022-03-25 | 2022-03-25 | Dynamic prediction method for tool life |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114536104A CN114536104A (en) | 2022-05-27 |
| CN114536104B true CN114536104B (en) | 2022-12-13 |
Family
ID=81665436
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210299596.4A Active CN114536104B (en) | 2022-03-25 | 2022-03-25 | Dynamic prediction method for tool life |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114536104B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115291564B (en) * | 2022-10-08 | 2023-01-10 | 成都飞机工业（集团）有限责任公司 | Numerical control machining cutter service life evaluation method based on cutting volume |
| CN116911469B (en) * | 2023-09-12 | 2024-01-12 | 成都飞机工业（集团）有限责任公司 | Numerical control machine tool machining time prediction method |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4583415B2 (en) * | 2007-02-15 | 2010-11-17 | 株式会社神戸製鋼所 | Tool wear prediction method, tool wear prediction program, and tool wear prediction system |
| CN103813355B (en) * | 2014-02-21 | 2018-07-27 | 厦门大学 | The recognition methods of synchronous abnormal point is cooperateed in a kind of distributed network |
| CN106021796B (en) * | 2016-06-03 | 2018-12-21 | 上海工具厂有限公司 | A kind of chromium steel blade profile processing method for predicting residual useful life of rose cutter |
| CN108536938A (en) * | 2018-03-29 | 2018-09-14 | 上海交通大学 | A kind of machine tool life prediction system and prediction technique |
| CN109978379B (en) * | 2019-03-28 | 2021-08-24 | 北京百度网讯科技有限公司 | Time series data abnormity detection method and device, computer equipment and storage medium |
| CN111890124A (en) * | 2019-05-05 | 2020-11-06 | 深圳市玄羽科技有限公司 | On-line cutter monitoring system and method |
| CN110263474A (en) * | 2019-06-27 | 2019-09-20 | 重庆理工大学 | A kind of cutter life real-time predicting method of numerically-controlled machine tool |
| CN111300146B (en) * | 2019-11-29 | 2021-04-02 | 上海交通大学 | Numerical control machine tool cutter abrasion loss online prediction method based on spindle current and vibration signal |
| CN113223701A (en) * | 2021-05-16 | 2021-08-06 | 河海大学 | Sudden heart disease prediction method based on Transformer-MHP model |
-
2022
- 2022-03-25 CN CN202210299596.4A patent/CN114536104B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN114536104A (en) | 2022-05-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114536104B (en) | Dynamic prediction method for tool life | |
| CN108490880B (en) | Method for monitoring wear state of cutting tool of numerical control machine tool in real time | |
| CN109514349B (en) | Tool wear state monitoring method based on vibration signal and Stacking integrated model | |
| CN110059442B (en) | Turning tool changing method based on part surface roughness and power information | |
| CN113741377A (en) | Machining process intelligent monitoring system and method based on cutting characteristic selection | |
| CN111975453B (en) | Numerical simulation driven machining process cutter state monitoring method | |
| Karpuschewski et al. | Determination of specific cutting force components and exponents when applying high feed rates | |
| CN113305644A (en) | Cutter state monitoring and early warning method and system based on part measurement data | |
| Schorr et al. | Quality prediction of reamed bores based on process data and machine learning algorithm: A contribution to a more sustainable manufacturing | |
| CN113189937A (en) | Integrated management method, system and application of tools of automatic production line of aviation parts | |
| CN114840932A (en) | Method for improving TC4 titanium alloy surface roughness prediction precision through multi-factor coupling | |
| CN116690313B (en) | Failure monitoring method for machining cutter of web plate of aircraft structural member | |
| CN117464420A (en) | Digital twin control cutter self-adaptive matching system suitable for numerical control machine tool | |
| CN109396956B (en) | Intelligent monitoring method for hob state of numerical control gear hobbing machine | |
| Anwar et al. | Optimization of surface roughness for Al-Alloy 7075-T in milling process | |
| Duo et al. | Surface roughness assessment on hole drilled through the identification and clustering of relevant external and internal signal statistical features | |
| CN112372371B (en) | Method for evaluating abrasion state of numerical control machine tool cutter | |
| Huihui et al. | Feature transfer-based approach for tool wear monitoring of face milling | |
| Roszkowski et al. | Study on the impact of cutting tool wear on machine tool energy consumption | |
| CN115213735B (en) | System and method for monitoring cutter state in milling process | |
| Akkus | Experimental and statistical investigations of surface roughness, vibration, and energy consumption values of titanium alloy during machining using response surface method and grey relational analysis | |
| CN115540759B (en) | Detection method and detection system for modifying metal based on image recognition technology | |
| CN113656903B (en) | Evaluation and control method for cutting residual stress of diamond abrasive particles | |
| Sarker et al. | Parametric study of a CNC turning process using discriminant analysis | |
| CN113601264B (en) | Cutter rear cutter face abrasion state determination method based on variable feed trial cutting |
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 |