CN114193229B - Variable working condition cutter wear prediction method based on causal inference - Google Patents
Variable working condition cutter wear prediction method based on causal inference Download PDFInfo
- Publication number
- CN114193229B CN114193229B CN202111682699.0A CN202111682699A CN114193229B CN 114193229 B CN114193229 B CN 114193229B CN 202111682699 A CN202111682699 A CN 202111682699A CN 114193229 B CN114193229 B CN 114193229B
- Authority
- CN
- China
- Prior art keywords
- causal
- working condition
- signal
- abrasion
- change
- 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
- 230000001364 causal effect Effects 0.000 title claims abstract description 112
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000005299 abrasion Methods 0.000 claims abstract description 89
- 230000008859 change Effects 0.000 claims abstract description 85
- 238000012544 monitoring process Methods 0.000 claims abstract description 21
- 238000004458 analytical method Methods 0.000 claims abstract description 15
- 238000004364 calculation method Methods 0.000 claims abstract description 14
- 238000001228 spectrum Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 9
- 238000000540 analysis of variance Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 7
- 230000007613 environmental effect Effects 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 230000003321 amplification Effects 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 238000001914 filtration Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 claims 1
- 238000003754 machining Methods 0.000 abstract description 8
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 230000001808 coupling effect Effects 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 5
- 238000010219 correlation analysis Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
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
- B23Q17/0904—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 before or after machining
Abstract
The invention relates to a causal inference-based variable working condition cutter wear prediction method, which aims at the difficult problem that accurate prediction of cutter wear is difficult due to the coupling action relation between a machining working condition and a monitoring signal and cutter wear in the numerical control machining process, and provides a causal inference-based method for extracting monitoring signal characteristics. Determining the influence coefficient of the abrasion and the working condition change on the signal change through the relation analysis and the causal effect calculation among the working condition change, the signal change and the abrasion change, and fully extracting partial information related to the abrasion from the signal characteristics according to the influence coefficient so as to realize the accurate prediction of the abrasion quantity of the cutter; the method realizes accurate prediction of the wear of the variable working condition cutter, greatly reduces the influence caused by the working condition change, and is beneficial to accurate modeling of a cutter wear prediction model.
Description
Technical Field
The invention relates to the field of cutter abrasion loss prediction in numerical control machining, in particular to a variable working condition cutter abrasion loss prediction method under big manufacturing data, and specifically relates to a variable working condition cutter abrasion loss prediction method based on causal inference.
Background
In the numerical control machining process, the cutter is affected by multiple factors, the abrasion mode is complex, only the machining working condition, the monitoring signal and the cutter abrasion exist in practical observables, the influence of the machining working condition on the cutter abrasion and the monitoring signal is coupled, and the prediction model between the monitoring signal and the cutter abrasion is directly built according to the data under the variable working condition. The working condition is used as a quantifiable tool abrasion influence factor, the uncertainty of the working condition is quantified, the causal relation among working condition change, signal change and abrasion change can be cleared, and the internal characteristics of the signal and abrasion data change during the working condition change are searched, so that the decisive characteristics of the tool in the degradation process are fully excavated, the redundancy and interference information of monitoring data under the variable working condition are effectively removed, the input quality of a model is improved, the sample size requirement is reduced, the training complexity is reduced, and the tool abrasion prediction precision under the variable working condition is improved.
At present, the variable working condition signal characteristic extraction method comprises a working condition decoupling method based on a stability coefficient and a working condition decoupling method based on correlation analysis. The working condition decoupling method based on the stability coefficient only directly ignores the influence of the working condition on the cutter wear in the modeling process, analyzes and selects the stability coefficient irrelevant to the working condition, and a cutting force model established based on the stability coefficient contains very little relevant cutter wear amount information, can only conduct stage prediction, and is difficult to realize accurate prediction of cutter wear amount. The working condition decoupling method based on the correlation analysis is to analyze the correlation among various monitoring signal characteristics, tool wear and processing working conditions, but the characteristic selection process only selects part of monitoring signals according to the strength of the correlation, the influence of the working conditions can not be completely shielded when the information loss exists, and particularly when the working conditions are changed greatly, the information loss is more serious, so that the prediction accuracy of the tool wear amount is greatly reduced. In practice, there is a complex coupling relationship between the machining conditions, the monitoring signal and the tool wear. The working condition belongs to the confounding factor on the signal characteristic and tool wear modeling path, and an error association relation is generated between the affected signal and wear.
Disclosure of Invention
The invention aims at solving the problem that the existing method adopts the relationship of stability coefficient and correlation analysis decoupling working condition change, signal change and abrasion change to have information loss and error correlation so as to influence the cutter abrasion prediction precision, and discloses a causally inferred variable working condition cutter abrasion prediction method.
The technical scheme of the invention is as follows:
a cutter abrasion prediction method based on causal inference mainly comprises the following steps: firstly, collecting and processing monitoring signals of a vibration sensor, a current sensor and a power sensor on a part, extracting features, and collecting and preprocessing labels of cutter abrasion loss; secondly, signal characteristic optimization is carried out based on causal inference, influence coefficients of abrasion and working condition changes on the signal changes are determined through causal relation analysis and causal effect calculation on the working condition changes, the signal changes and the abrasion changes, and partial information related to the abrasion is fully extracted from monitoring signal characteristics according to the influence coefficients; and finally, realizing accurate prediction of the cutter abrasion loss based on the data driving model. The main steps of signal feature optimization based on causal inference include: causal network establishment, causal effect calculation and signal characteristic update.
The invention collects three types of common monitoring of vibration signals, main shaft current and main shaft power, which are strongly related to the abrasion change of the cutterAnd then, carrying out Gaussian filtering on the signal to reduce noise, compensating the missing value by adopting median random sample interpolation, and carrying out further interpolation correction by adopting an absolute difference median method. Measuring the maximum wear width VB of the relief surface by means of an industrial microscope max And (5) representing the abrasion condition of the cutter and carrying out pretreatment operations such as abnormal data rejection, data amplification and the like.
The invention adopts time domain, frequency domain and time-frequency domain analysis methods to respectively extract the characteristics of three signals of vibration, main shaft current and main shaft power, and the three signals respectively obtain peak-peak value PP, mean value Mean, median Med, variance Var and skewness value C s Peak, root mean square RM, peak factor C f 8 time domain features; mean of power spectrum ps Mean square error MSE of power spectrum ps Power spectrum bias value C sps Peak power spectrum ps 8 frequency domain features such as average frequency MeanF, median frequency MedF, power bandwidth BPW, average power MBP and the like; the node energy after 12 eigenvalues and 8 wavelet packets are decomposed, and 28 time-frequency domain features such as 8 decomposition coefficient vectors, etc. are 132 features in total.
The causal network is established aiming at the causal network, causal relationships among working condition changes, signal changes and abrasion changes are represented by adopting a causal graph, verification of the causal network is carried out through analysis of variance, and an accurate causal network is established for subsequent influence coefficient analysis and calculation. The main steps of establishing a causal network of working condition-signal-abrasion comprise:
(1) Selecting a variable to be analyzed;
(2) A causal relationship representation based on a structural causal model;
(3) Causal relationship verification based on analysis of variance.
The invention selects the variables to be analyzed, namely, determines all exogenous and endogenous variables in the causal network. Exogenous variables refer to system exogenous variables that are not of interest in the causal network establishment process, including noise, and other unknown environmental factors. Because exogenous variables are not main influencing factors of the system, and the characteristics of randomness, unobservability and the like exist, the invention reasonably omits simplification during specific analysis. Endogenous variables refer to intra-system variables of interest in the causal network establishment process, including operating condition change C, wear change W, and signal change S. And then, according to the structural causal model and the processing experience knowledge, the variable relation in the numerical control processing cutter abrasion system is preliminarily represented by a functional father-son relation, and can be subsequently converted into conditional probability for analysis based on a product rule.
Under the condition that exogenous variables are not considered, the signal variables are unfolded into four types, and the functional relation can be further simplified into a causal structure diagram formed by working conditions, abrasion and four types of signal characteristics. The first signal characteristic is the signal characteristic S which is only influenced by the working condition C and is not influenced by the abrasion W I The method comprises the steps of carrying out a first treatment on the surface of the The second type of signal features are signal features S influenced by the working condition C and the abrasion W at the same time II The method comprises the steps of carrying out a first treatment on the surface of the The third kind of signal features are the signal features S which are only affected by the abrasion W and are not affected by the working condition C III The method comprises the steps of carrying out a first treatment on the surface of the The fourth signal characteristic is a signal characteristic S which is not affected by the abrasion W and the working condition C IV 。
The causal relationship verification based on analysis of variance mainly comprises three parts:
firstly, respectively providing assumptions for two factors of working conditions and abrasion, wherein the original/alternative assumption of the working condition C is that the working condition change has no/obvious influence on a certain signal characteristic, and the original/alternative assumption of the abrasion W is that the abrasion change has no/obvious influence on a certain signal characteristic;
second, construct and calculate the correlation statistic for hypothesis testing, the sum of squares SS of signal features S T Is the sum of squares error of all sample observations and the average of the total samples, and the sum of squares error SS generated by the condition C C Sum of squares error SS due to wear W W Error term and SS generated by an interaction C×W Sum of squares of random errors SS E The four items are formed, and then mean square errors are calculated respectively to construct test F statistics;
thirdly, judging the causal relation between the working condition and the abrasion and the signals, and setting two degrees of freedom values f of the working condition C and the abrasion W at a significance level sig C ，f W Through the critical value F found in the F table sig And respectively calculating corresponding P values. If P < sig, i.e. F > F sig The original assumption is refused, and the working condition/abrasion factors are considered to have obvious influence on the signals; otherwise, the original assumption is received, and the working condition/abrasion factors are considered to have no significant influence on the signals.
The invention relates to a causal inference-based variable working condition cutter wear prediction method, which adopts intervention operation to change data distribution aiming at causal effect calculation, calculates influence coefficients on causal paths among working condition change, wear change and signal change, and mainly comprises the following steps:
(1) Calculating the causal effect of the wear change on the signal characteristic change;
(2) And calculating the causal effect of the working condition change on the signal characteristic change.
The invention calculates the causal effect of the wear change on the signal characteristic change, namely, calculates the increment of the signal change under the condition of the wear change relative to the signal change under the condition of the unchanged wear by adopting the average causal effect (Average Causal Effect, ACE), namely, the influence coefficient of the wear on the signal characteristic change:
ACE＝P(S＝1|do(W＝1))-P(S＝1|do(W＝0))
the causal effect of the working condition change on the signal characteristic value change is calculated, namely the increment of the signal change under the condition of the working condition change relative to the signal change under the condition of the unchanged working condition is the influence coefficient of the working condition on the signal characteristic change:
ACE＝P(S＝1|do(C＝1))-P(S＝1|do(C＝0))
according to the causal inference-based variable working condition tool wear prediction method, partial signal characteristic values are weighted and updated according to the influence coefficient obtained through calculation aiming at signal characteristic updating, and the mixed interference of working condition changes on the corresponding relation between the signal characteristics and tool wear is weakened. Operating mode C in which the tool wear is relatively slow 0 The following features are reference features S 0 Calculating the characteristic change of the signal, and facing to the new working condition C k (k=1, 2,3, …) according to the influence coefficient a k Re-weighting each dimension of the feature under the new working condition, and furtherNovel operating mode characteristics S k The updated signal characteristics are guaranteed to only contain partial information related to tool wear, and the specific formula is as follows:
calculating signal characteristic change: delta=s k -S 0
Updating signal characteristics: s'. k ＝S 0 +a k Δ
According to the causal inference-based variable working condition tool wear prediction method, the full-connection neural network model is adopted to predict the tool wear amount, the input of the model is the updated signal characteristic, the output is the tool wear amount, sufficient input and output data pairs are provided for the model during training, only the updated signal characteristic is provided as the model input during testing, and the model outputs the predicted tool wear amount value.
The beneficial effects of the invention are as follows:
(1) The prediction method can reduce the mixed influence of the working condition change and improve the cutter abrasion prediction precision under the variable working condition;
(2) Based on causal inference, quantifying the influence among working conditions, signals and abrasion, the method is beneficial to further analysis of the cutter abrasion intrinsic mechanism;
(3) The method is not only suitable for predicting the cutter wear, but also suitable for fault detection in other machining processes.
Drawings
FIG. 1 is a flow chart of a causal inference-based tool wear prediction method according to the present invention, in which S I 、S II 、S III 、S IV The method respectively represents four types of signal characteristics which are influenced by the working condition C only and are not influenced by the abrasion W, signal characteristics which are influenced by the working condition C and are influenced by the abrasion W simultaneously, signal characteristics which are influenced by the working condition C only and are not influenced by the abrasion W, and signal characteristics which are influenced by the working condition C neither are influenced by the abrasion W nor are influenced by the working condition C, wherein C represents working condition change, W represents tool abrasion change, and a k The influence coefficients generated from the arrow start object to the arrow end object are represented, delta represents signal characteristic change, mean represents Mean, med represents median and MBP represents average power;
FIG. 2 (a) is a schematic diagram of the condition-signal-wear causal relationship, U 1 、U 2 、U 3 External variables such as noise and unknown environmental factors are represented;
FIG. 2 (b) is a schematic diagram of causal relationships after signal variable expansion;
FIG. 3 (a) is an initial causal network containing four classes of features;
FIG. 3 (b) is a final causal network containing two types of features;
FIG. 4 is a diagram showing signal characteristic update under variable operating conditions according to the present invention, wherein S k ' representing the pre-update Signal feature set, S k Representing the updated set of signal features.
Detailed Description
The invention is further illustrated by the following figures and examples.
As shown in fig. 1-4.
A cutter abrasion prediction method based on causal inference is shown in a flow chart of fig. 1, and comprises the following specific steps:
step 1, collecting and processing monitoring signals of a vibration sensor, a current sensor and a power sensor on a part, extracting characteristics, and collecting and preprocessing labels of cutter abrasion loss;
step 2, extracting signal characteristics, namely extracting time domain, frequency domain and time-frequency domain characteristics of the monitoring signals mainly in a statistical mode;
step 3, signal characteristic optimization based on causal inference mainly comprises three parts: establishing a causal network, calculating causal effect and updating signal characteristics; wherein, the causal network establishment is shown in fig. 2 and 3, and the signal characteristic update is shown in fig. 4;
and 4, predicting the cutter abrasion loss, wherein a data driving model is adopted to establish a complex corresponding relation between the monitoring signal characteristics and the cutter abrasion loss, so that the cutter abrasion loss is predicted.
The details are as follows:
the tool wear prediction method based on causal inference of the invention collects three common monitoring signals of vibration signal, main shaft current and main shaft power which are strongly related to tool wear change, then adopts Gaussian filter to reduce noise, and adopts median random sample interpolation pairThe missing value is compensated, and further interpolation correction is carried out by adopting an absolute difference median method. Measuring the maximum wear width VB of the relief surface by means of an industrial microscope max And (5) representing the abrasion condition of the cutter and carrying out pretreatment operations such as abnormal data rejection, data amplification and the like.
The invention adopts time domain, frequency domain and time-frequency domain analysis methods to respectively extract the characteristics of three signals of vibration, main shaft current and main shaft power, and the three signals respectively obtain peak-peak value PP, mean value Mean, median Med, variance Var and skewness value C s Peak, root mean square RM, peak factor C f 8 time domain features; mean of power spectrum ps Mean square error MSE of power spectrum ps Power spectrum bias value C sps Peak power spectrum ps 8 frequency domain features such as average frequency MeanF, median frequency MedF, power bandwidth BPW, average power MBP and the like; the node energy after 12 eigenvalues and 8 wavelet packets are decomposed, and 28 time-frequency domain features such as 8 decomposition coefficient vectors, etc. are 132 features in total.
The causal network is established aiming at the causal network, causal relationships among working condition changes, signal changes and abrasion changes are represented by adopting a causal graph, verification of the causal network is carried out through analysis of variance, and an accurate causal network is established for subsequent influence coefficient analysis and calculation. The main steps of establishing a causal network of working condition-signal-abrasion comprise:
(1) Selecting a variable to be analyzed;
(2) A causal relationship representation based on a structural causal model;
(3) Causal relationship verification based on analysis of variance.
The invention selects the variables to be analyzed, namely, determines all exogenous and endogenous variables in the causal network. Exogenous variables refer to system exogenous variables that are not of interest in the causal network establishment process, including noise, and other unknown environmental factors. Because exogenous variables are not main influencing factors of the system, and the characteristics of randomness, unobservability and the like exist, the invention reasonably omits simplification during specific analysis. Endogenous variables refer to intra-system variables of interest in the causal network establishment process, including operating condition change C, wear change W, and signal change S. And then, according to the structural causal model and the processing experience knowledge, the variable relation in the numerical control processing cutter abrasion system is preliminarily represented by a functional father-son relation, and can be subsequently converted into conditional probability for analysis based on a product rule.
Under the condition that exogenous variables are not considered, the signal variables are unfolded into four types, and the functional relation can be further simplified into a causal structure diagram formed by working conditions, abrasion and four types of signal characteristics. The first signal characteristic is the signal characteristic S which is only influenced by the working condition C and is not influenced by the abrasion W I The method comprises the steps of carrying out a first treatment on the surface of the The second type of signal features are signal features S influenced by the working condition C and the abrasion W at the same time II The method comprises the steps of carrying out a first treatment on the surface of the The third kind of signal features are the signal features S which are only affected by the abrasion W and are not affected by the working condition C III The method comprises the steps of carrying out a first treatment on the surface of the The fourth signal characteristic is a signal characteristic S which is not affected by the abrasion W and the working condition C IV 。
The causal relationship verification based on analysis of variance mainly comprises three parts:
firstly, respectively providing assumptions for two factors of working conditions and abrasion, wherein the original/alternative assumption of the working condition C is that the working condition change has no/obvious influence on a certain signal characteristic, and the original/alternative assumption of the abrasion W is that the abrasion change has no/obvious influence on a certain signal characteristic;
second, construct and calculate the correlation statistic for hypothesis testing, the sum of squares SS of signal features S T Is the sum of squares error of all sample observations and the average of the total samples, and the sum of squares error SS generated by the condition C C Sum of squares error SS due to wear W W Error term and SS generated by an interaction C×W Sum of squares of random errors SS E The four items are formed, and then mean square errors are calculated respectively to construct test F statistics;
thirdly, judging the causal relation between the working condition and the abrasion and the signals, and setting two degrees of freedom values f of the working condition C and the abrasion W at a significance level sig C ，f W Through the critical value F found in the F table sig And respectively calculating corresponding P values. If P < sig, i.e. F > F sig Reject the original hypothesisThe operating conditions/wear factors are considered to have a significant impact on the signal; otherwise, the original assumption is received, and the working condition/abrasion factors are considered to have no significant influence on the signals.
The invention relates to a causal inference-based variable working condition cutter wear prediction method, which adopts intervention operation to change data distribution aiming at causal effect calculation, calculates influence coefficients on causal paths among working condition change, wear change and signal change, and mainly comprises the following steps:
(1) Calculating the causal effect of the wear change on the signal characteristic change;
(2) And calculating the causal effect of the working condition change on the signal characteristic change.
The invention calculates the causal effect of the wear change on the signal characteristic change, namely, calculates the increment of the signal change under the condition of the wear change relative to the signal change under the condition of the unchanged wear by adopting the average causal effect (Average Causal Effect, ACE), namely, the influence coefficient of the wear on the signal characteristic change:
ACE＝P(S＝1|do(W＝1))-P(S＝1|do(W＝0))
the causal effect of the working condition change on the signal characteristic value change is calculated, namely the increment of the signal change under the condition of the working condition change relative to the signal change under the condition of the unchanged working condition is the influence coefficient of the working condition on the signal characteristic change:
ACE＝P(S＝1|do(C＝1))-P(S＝1|do(C＝0))
according to the causal inference-based variable working condition tool wear prediction method, partial signal characteristic values are weighted and updated according to the influence coefficient obtained through calculation aiming at signal characteristic updating, and the mixed interference of working condition changes on the corresponding relation between the signal characteristics and tool wear is weakened. Operating mode C in which the tool wear is relatively slow 0 The following features are reference features S 0 Calculating the characteristic change of the signal, and facing to the new working condition C k (k=1, 2,3, …) according to the influence coefficient a k Re-weighting each dimension of the feature under the new working condition, and updating the feature S of the new working condition k Ensuring that the updated signal characteristics only include a correlation with tool wearThe specific formula is as follows:
calculating signal characteristic change: delta=s k -S 0
Updating signal characteristics: s'. k ＝S 0 +a k Δ
According to the causal inference-based variable working condition tool wear prediction method, the full-connection neural network model is adopted to predict the tool wear amount, the input of the model is the updated signal characteristic, the output is the tool wear amount, sufficient input and output data pairs are provided for the model during training, only the updated signal characteristic is provided as the model input during testing, and the model outputs the predicted tool wear amount value.
The invention is not related in part to the same as or can be practiced with the prior art.
Claims (7)
1. A causal inference-based variable working condition cutter wear prediction method is characterized by comprising the following steps of:
step 1, collecting and processing monitoring signals of a vibration sensor, a current sensor and a power sensor on a part, extracting characteristics, and collecting and preprocessing labels on the abrasion loss of a cutter;
step 2, signal characteristic optimization is carried out based on causal inference, influence coefficients of abrasion and working condition changes on signal changes are determined through causal relation analysis and causal effect calculation on working condition changes, signal changes and abrasion changes, and partial information related to the abrasion is fully extracted from monitoring signal characteristics according to the influence coefficients;
the method specifically comprises three steps of causal network establishment, causal effect calculation and signal characteristic updating;
the step of establishing a condition-signal-wear causal network includes:
(1) Selecting a variable to be analyzed; namely, determining all exogenous and endogenous variables in a causal network; exogenous variables refer to system exogenous variables that are not of interest in the causal network establishment process, including noise, and other unknown environmental factors; endogenous variables refer to intra-system variables of interest in the causal network establishment process, including operating condition changesChemical treatmentCWear changeWSum signal variationSThen, according to the structural causal model and the processing experience knowledge, the variable relation in the numerical control processing cutter abrasion system is primarily represented by a functional father-son relation, and then the variable relation can be converted into conditional probability for analysis based on a product rule;
(2) A causal relationship representation based on a structural causal model; under the condition of not considering exogenous variables, the signal variables are unfolded into four classes, and the first class of signal features are only under working conditionsCIs not affected by abrasionWAffected signal characteristicsS I The method comprises the steps of carrying out a first treatment on the surface of the The second kind of signal features are the working conditionCAnd wear and tearWAffected signal characteristicsS II The method comprises the steps of carrying out a first treatment on the surface of the The third type of signal features being only subject to wearWIs not affected by working conditionsCAffected signal characteristicsS III The method comprises the steps of carrying out a first treatment on the surface of the The fourth type of signal features are not subject to wearWIs not affected by working conditionsCAffected signal characteristicsS IV ；
(3) Causal relationship verification based on analysis of variance; mainly comprises three parts:
firstly, respectively providing assumptions for two factors of working condition and abrasion, and working conditionCThe original/alternative assumption of (1) is that the working condition change has no/obvious influence on a certain signal characteristic, and the abrasionWIs that wear changes have no/significant effect on a certain signal characteristic;
second, construct and calculate correlation statistics for hypothesis testing, signal characteristicsSSum of the total squares of (2)SS T Is the sum of squares of errors of all sample observed values and the average value of the total samples, and is determined by working conditionsCThe sum of squares of the errors producedSS C Wear and tearWThe sum of squares of the errors producedSS W Error term generated by interaction andSS C×W sum of squares of random errorsSS E The four items are formed, and then mean square errors are calculated respectively to construct test F statistics;
thirdly, judging the causal relation between the working condition and the abrasion and the signal, and setting the significance level sig and the working conditionCAnd wear and tearWTwo degree of freedom values of (2)f C ，f W By means ofCritical value F found in the F table sig Respectively calculating corresponding P values, if P is less than sig, namely F>F sig The original assumption is refused, and the working condition/abrasion factors are considered to have obvious influence on the signals; otherwise, receiving the original assumption, and considering that the working condition/abrasion factors have no obvious influence on the signals; and 3, realizing accurate prediction of the cutter abrasion loss based on a data driving model, specifically adopting a fully-connected neural network model to predict the cutter abrasion loss, wherein the input of the model is updated signal characteristics, the output is the cutter abrasion loss, sufficient input and output data pairs are provided for the model during training, only the updated signal characteristics are provided as model input during testing, and the model outputs the predicted cutter abrasion loss value.
2. The causal inference based variable condition tool wear prediction method of claim 1, wherein: in the step 1, monitoring signals of a vibration sensor, a current sensor and a power sensor on a part are collected and processed, and specifically: and collecting three sensor monitoring signals, namely a vibration signal, a main shaft current and a main shaft power, which are strongly related to the abrasion change of the cutter, then adopting Gaussian filtering to reduce noise of the signals, adopting median random sample interpolation to compensate the missing value, and adopting an absolute difference median method to carry out further interpolation correction.
3. The causal inference based variable condition tool wear prediction method of claim 1, wherein: in the step 1, the cutter abrasion loss is collected and the label is preprocessed, specifically: measuring the maximum wear width VB of the relief surface by means of an industrial microscope max， VB max And (5) representing the abrasion condition of the cutter and carrying out abnormal data rejection and data amplification pretreatment operation.
4. The causal inference based variable condition tool wear prediction method of claim 2, wherein: the monitoring signals of the vibration sensor, the current sensor and the power sensor are subjected to feature extraction, and the method specifically comprises the following steps: the vibration, the main shaft current and the main shaft power are respectively carried out by adopting time domain, frequency domain and time-frequency domain analysis methodsExtracting features of three kinds of signals to obtain peak-to-peak valuePPAverage value ofMeanMedian valueMedVariance ofVarDeviation valuePeak valuePeakRoot mean squareRMPeak factor->8 time domain features; mean value of power spectrumMean ps Mean square error of power spectrumMSE ps Power spectrum bias value->Peak of power spectrumPeak ps Average frequencyMeanFMedian frequencyMedFBandwidth of powerBPWAverage powerMBP8 frequency domain features; the method comprises the steps of 12 inherent mode function values, node energy after 8 wavelet packets are decomposed, and 28 time-frequency domain characteristics of 8 decomposition coefficient vectors, wherein 132 characteristics are obtained.
5. The causal inference based variable condition tool wear prediction method of claim 1, wherein: aiming at the causal network establishment, a causal graph is adopted to represent causal relations among working condition changes, signal changes and wear changes, the causal network is verified through analysis of variance, and an accurate causal network is established for subsequent influence coefficient analysis and calculation.
6. The causal inference based variable condition tool wear prediction method of claim 1, wherein: for causal effect calculation, the data distribution is changed by adopting intervention operation, and the influence coefficients on causal paths among working condition change, abrasion change and signal change are calculated, wherein the steps comprise:
(1) Calculating the causal effect of the wear change on the signal characteristic change;
(2) And calculating the causal effect of the working condition change on the signal characteristic change.
7. The causal inference based variable condition tool wear prediction method of claim 6, wherein: aiming at signal characteristic updating, partial signal characteristic values are weighted and updated according to the magnitude of the influence coefficient obtained by calculation, and the mixed interference of the working condition change on the corresponding relation between the signal characteristic and the cutter abrasion is weakened; by a relatively slow tool wear conditionC 0 The following features are reference featuresS 0 Calculating the characteristic change of the signal, and facing to new working conditionsC k When k=1, 2,3, …, n, k represents the working condition number, n is the maximum value of the working condition number k; according to the influence coefficienta k Re-weighting each dimension of the feature under the new working condition, and updating the feature of the new working conditionS k It is ensured that the updated signal characteristics only contain part of the information related to tool wear.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111682699.0A CN114193229B (en) | 2021-12-30 | 2021-12-30 | Variable working condition cutter wear prediction method based on causal inference |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111682699.0A CN114193229B (en) | 2021-12-30 | 2021-12-30 | Variable working condition cutter wear prediction method based on causal inference |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114193229A CN114193229A (en) | 2022-03-18 |
| CN114193229B true CN114193229B (en) | 2023-12-12 |
Family
ID=80658093
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111682699.0A Active CN114193229B (en) | 2021-12-30 | 2021-12-30 | Variable working condition cutter wear prediction method based on causal inference |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114193229B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116205384B (en) * | 2023-05-06 | 2023-07-04 | 北京航空航天大学 | Urban data prediction method based on generation cause and effect interpretation model |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2018086712A (en) * | 2016-11-30 | 2018-06-07 | 株式会社日立製作所 | Tool wear prediction device and method therefor |
| CN109333160A (en) * | 2018-10-09 | 2019-02-15 | 西北工业大学 | The on-line monitoring method of high temperature alloy drilling process bit wear form and state of wear |
| CN109571141A (en) * | 2018-11-01 | 2019-04-05 | 北京理工大学 | A kind of Monitoring Tool Wear States in Turning based on machine learning |
| CN110728667A (en) * | 2019-10-08 | 2020-01-24 | 南京航空航天大学 | Automatic and accurate cutter wear loss measuring method based on gray level image probability |
| CN111085898A (en) * | 2019-12-30 | 2020-05-01 | 南京航空航天大学 | Working condition self-adaptive high-speed milling process cutter monitoring method and system |
| CN112192319A (en) * | 2020-09-28 | 2021-01-08 | 上海交通大学 | Tool wear monitoring method and system of unsupervised model |
| EP3815645A1 (en) * | 2019-11-01 | 2021-05-05 | DENTSPLY SIRONA Inc. | Dental machining system for predicting the wear condition of a dental tool |
| CN113664612A (en) * | 2021-08-24 | 2021-11-19 | 沈阳工业大学 | Numerical control machine tool milling cutter abrasion real-time monitoring method based on deep convolutional neural network |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080161959A1 (en) * | 2006-12-01 | 2008-07-03 | Jerard Robert B | Method to measure tool wear from process model parameters |
| JP6423769B2 (en) * | 2015-08-31 | 2018-11-14 | ファナック株式会社 | Machining system with machining accuracy maintenance function |
-
2021
- 2021-12-30 CN CN202111682699.0A patent/CN114193229B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2018086712A (en) * | 2016-11-30 | 2018-06-07 | 株式会社日立製作所 | Tool wear prediction device and method therefor |
| CN109333160A (en) * | 2018-10-09 | 2019-02-15 | 西北工业大学 | The on-line monitoring method of high temperature alloy drilling process bit wear form and state of wear |
| CN109571141A (en) * | 2018-11-01 | 2019-04-05 | 北京理工大学 | A kind of Monitoring Tool Wear States in Turning based on machine learning |
| CN110728667A (en) * | 2019-10-08 | 2020-01-24 | 南京航空航天大学 | Automatic and accurate cutter wear loss measuring method based on gray level image probability |
| EP3815645A1 (en) * | 2019-11-01 | 2021-05-05 | DENTSPLY SIRONA Inc. | Dental machining system for predicting the wear condition of a dental tool |
| CN111085898A (en) * | 2019-12-30 | 2020-05-01 | 南京航空航天大学 | Working condition self-adaptive high-speed milling process cutter monitoring method and system |
| CN112192319A (en) * | 2020-09-28 | 2021-01-08 | 上海交通大学 | Tool wear monitoring method and system of unsupervised model |
| CN113664612A (en) * | 2021-08-24 | 2021-11-19 | 沈阳工业大学 | Numerical control machine tool milling cutter abrasion real-time monitoring method based on deep convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114193229A (en) | 2022-03-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7890200B2 (en) | Process-related systems and methods | |
| CN108960329B (en) | Chemical process fault detection method containing missing data | |
| CN114193229B (en) | Variable working condition cutter wear prediction method based on causal inference | |
| CN108535635B (en) | EEMD and HMM based analog circuit intermittent fault diagnosis method | |
| CN109085805B (en) | Industrial process fault detection method based on multi-sampling-rate factor analysis model | |
| CN111638707B (en) | Intermittent process fault monitoring method based on SOM clustering and MPCA | |
| CN112000081B (en) | Fault monitoring method and system based on multi-block information extraction and Mahalanobis distance | |
| CN113569990A (en) | Performance equipment fault diagnosis model construction method oriented to strong noise interference environment | |
| Jiang et al. | Recursive filtering for nonlinear systems subject to measurement outliers | |
| CN117193222A (en) | Intelligent quality control system based on industrial Internet of things and big data and control method thereof | |
| CN114757087A (en) | Tool wear prediction method based on dynamic principal component analysis and LSTM | |
| CN115922553A (en) | Method for online monitoring polishing processing state of silicon carbide wafer | |
| CN114266197A (en) | Method for diagnosing equipment fault of hydropower station | |
| CN112817955B (en) | Regression model-based data cleaning method | |
| CN113269327A (en) | Flow anomaly prediction method based on machine learning | |
| CN110855519A (en) | Network flow prediction method | |
| CN116361191A (en) | Software compatibility processing method based on artificial intelligence | |
| CN116658433A (en) | Intelligent detection method and system for particle-resistant magnetic pump | |
| CN116628869A (en) | Analysis method for fault propagation mechanism of numerical control machine tool based on transfer entropy theory | |
| CN114418420A (en) | Competitive risk survival analysis method based on causal inference | |
| Baruah et al. | An autonomous diagnostics and prognostics framework for condition-based maintenance | |
| Nakkabi et al. | A combined analytical and knowledge based method for fault detection and isolation | |
| CN114637793B (en) | Equipment fault frequent region positioning method based on big data analysis | |
| CN112104340B (en) | HMM model and Kalman filtering technology-based switching value input module BIT false alarm reduction method | |
| CN110647922A (en) | Layered non-Gaussian process monitoring method based on public and special feature extraction |
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 |