CN111360578A - Method for identifying lubricating state of ball screw pair - Google Patents
Method for identifying lubricating state of ball screw pair Download PDFInfo
- Publication number
- CN111360578A CN111360578A CN202010194479.2A CN202010194479A CN111360578A CN 111360578 A CN111360578 A CN 111360578A CN 202010194479 A CN202010194479 A CN 202010194479A CN 111360578 A CN111360578 A CN 111360578A
- Authority
- CN
- China
- Prior art keywords
- characteristic
- lubrication
- feature
- ball screw
- screw pair
- 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.)
- Granted
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
- B23Q11/00—Accessories fitted to machine tools for keeping tools or parts of the machine in good working condition or for cooling work; Safety devices specially combined with or arranged in, or specially adapted for use in connection with, machine tools
- B23Q11/12—Arrangements for cooling or lubricating parts of the machine
- B23Q11/121—Arrangements for cooling or lubricating parts of the machine with lubricating effect for reducing friction
- B23Q11/125—Arrangements for cooling or lubricating parts of the machine with lubricating effect for reducing friction for lubricating ball screw systems
Abstract
The invention discloses a method for identifying the lubrication state of a ball screw pair, which comprises the following steps: collecting vibration signals of the ball screw pair in different lubrication states; wavelet packet transformation is carried out on the vibration signals, and the wavelet packet transformation is combined with the time domain and frequency domain characteristic values to construct a characteristic set; performing primary optimization on the feature set by using a Relief-F, screening out a feature vector with a large identification weight value for a lubrication state, and constructing a new feature set; inputting the feature set subjected to the primary optimization into a support vector machine, and training and optimizing the SVM through an optimization algorithm; and setting the identification accuracy to carry out secondary optimization, and finding out the optimal characteristic combination for identifying the lubricating state, thereby judging the lubricating state of the ball screw pair. The method for identifying the lubricating state of the ball screw pair can effectively solve the problems of difficult identification, low efficiency, low accuracy and the like of the lubricating state of the current ball screw pair.
Description
Technical Field
The invention belongs to the field of identification of a lubricating state of a ball screw pair, and particularly relates to a method for identifying a lubricating state of a ball screw pair.
Background
The ball screw pair is a key rolling functional component in the numerical control machine tool, has the characteristics of small friction, zero clearance, high transmission efficiency and the like, is widely applied to the fields of aerospace, ships, automobiles, numerical control machine tools and the like, and the performance of the ball screw pair directly influences the processing precision of the numerical control machine tool. However, in the actual processing process, the friction between the balls and the roller paths of the ball screw pair is aggravated due to poor maintenance and poor lubrication, so that the local temperature of the screw is suddenly increased, the processing precision of the numerical control machine tool is affected, and the processing quality of a product is further affected. Therefore, in the actual operation process of the numerical control machine tool, the lubricating state of the ball screw pair needs to be monitored, the lubricating state of the ball screw pair is known, and once the poor lubricating phenomenon occurs, the lubricating state of the ball screw pair can be timely improved, so that the machining precision of the ball screw pair is ensured, and the service life is prolonged.
At present, the research on the aspects of monitoring and fault diagnosis of the lubrication state of the ball screw pair is less, most research methods are single, and the combination optimization among various methods is not considered, so that the judgment on the lubrication state of the ball screw pair is inaccurate in the actual lubrication state identification process, and the development of the state monitoring direction of the ball screw pair is limited to a certain extent. Therefore, a more complete method for more accurately judging the lubrication state of the ball screw pair is urgently needed.
Disclosure of Invention
The invention aims to provide a method for identifying the lubricating state of a ball screw pair, which solves the problems of difficult identification, low efficiency, low accuracy and the like of the lubricating state of the ball screw pair at present.
The technical solution for realizing the purpose of the invention is as follows: a method for identifying a lubrication state of a ball screw pair, the method comprising the steps of:
step 1, collecting vibration signals of a ball screw pair in different lubrication states;
step 3, training an optimized SVM by using the lubricating feature set;
and 4, optimizing the lubricating feature set according to the SVM recognition accuracy to obtain the optimal feature combination of the lubricating state recognition, and then finishing the judgment of the lubricating state of the ball screw pair to be recognized by combining the optimized SVM.
Further, the step 2 of extracting vibration signal features and constructing a lubrication feature set comprises the following specific processes:
step 2-1, extracting a plurality of groups of vibration data from the vibration signals;
step 2-2, performing wavelet packet transformation on each group of vibration data;
step 2-3, solving a time domain characteristic value of each group of vibration data;
step 2-4, obtaining the frequency domain characteristic value of each group of vibration data;
step 2-5, regarding each characteristic obtained in the process as a lubricating characteristic, and constructing a lubricating characteristic set according to all results obtained in the process, wherein each sample in the set is a lubricating characteristic value corresponding to each group of vibration data;
step 2-6, optimizing the lubricating feature set, specifically: and according to the weight value occupied by each lubrication characteristic pair in the lubrication state, screening out the characteristic value with the weight value larger than a preset weight threshold value from the lubrication characteristic set, and constructing a new lubrication characteristic set.
Further, in step 2-2, performing wavelet packet transformation on each group of vibration data, the specific process includes:
(1) using a high-pass filter gk}k∈ZAnd a low pass filter hk}k∈ZPerforming full-band decomposition on the vibration data signal, wherein the formula is as follows:
in the formula, d represents wavelet coefficient of wavelet packet decomposition frequency band, k, l ∈ Z, and the two formulas respectively represent that the nth frequency band of the jth layer is decomposed into the 2n and 2n +1 frequency bands of the jth +1 layer;
(2) reconstructing the decomposed signal, wherein the reconstruction formula is as follows:
in the formula, pkAnd q iskAre respectively hkAnd gkThe dual filter of (2);
(3) calculating each node signal x after decompositionkmEnergy value of (2):
in the formula, ejkRepresenting the energy value of the frequency band of the kth node after the wavelet packet is subjected to the jth layer decomposition and reconstruction, wherein x is the amplitude of a reconstructed signal at a discrete point, and N is the signal length;
(4) calculating the energy ratio of each node frequency band:
in the formula, EjkIndicating the energy-occupied ratio of each node frequency band.
Further, the time domain characteristic values in the step 2-3 comprise dimensional parameters and dimensionless parameters;
wherein, the dimension parameters comprise:
(1) mean value
Wherein x represents the discrete data amplitude of each vibration signal, and n represents the number of discrete data;
(2) root mean square value
(3) Square root amplitude
(4) Absolute mean value
(5) Variance (variance)
(6) Peak value: taking the maximum numbers in the signals, and calculating the arithmetic mean value of the numbers as a peak value;
the dimensionless parameters include:
(1) waveform index
(2) Peak index
(3) Pulse index
(4) Margin index
(5) Kurtosis index
Further, the frequency domain feature values in step 2-4 include:
(1) frequency of center of gravity
In the formula (f)iThe instantaneous frequency of the vibration signal corresponding to time i, F (F)i) Representing the power spectrum amplitude corresponding to the instantaneous frequency;
(2) mean square frequency
(3) Frequency variance
Further, the step 2-6 optimizes the lubricating feature set, specifically optimizes the lubricating feature set by using a Relief-F, and the specific process includes:
step 2-6-1, establishing a feature weight set W, and making each feature initial weight W (i) in the feature weight set 0, i-1, 2,3, …, N1，N1Representing the number of features, setting the initial iteration number s to be 1, and setting an iteration number threshold value M;
2-6-2, randomly extracting a characteristic sample r from the lubricating characteristic set;
step 2-6-3, finding out k in the same lubrication state as the characteristic sample r from the lubrication characteristic set0A nearest neighbor feature sample Qa，a＝1,2,3,…,k0And separately find k in a lubrication state different from that of the sample r0A nearest neighbor feature sample Qa(c) And c represents the number of categories of non-homogeneous lubrication states;
and 2-6-4, updating the weight value of each feature in the feature weight set W, wherein the updating formula is as follows:
wherein p (c) represents the distribution probability of the c-th lubrication state, C (r) represents the lubrication state of the characteristic sample r, p (C (r)) represents the distribution probability of the lubrication state to which the characteristic sample r belongs, and diff (F, r, Q)a(c) ) represent feature samples r and Qa(c) Distance difference on ith feature F; wherein diff (F, r, Q)a) Discrete or continuous definition of feature data from a feature set is:
in the formula, Fr、FQaRespectively representing the feature samples r and the featuresSample QaThe value on the ith feature F;
and 2-6-5, judging whether s is smaller than M, if so, increasing s by 1, returning to execute the step 2-6-2 until the condition is met, and then screening out a characteristic value with a weight value larger than a preset weight threshold value from the lubricating characteristic set to construct a new lubricating characteristic set.
Further, the step 3 of training and optimizing the SVM by using the lubrication feature set specifically includes:
step 3-1, inputting the lubricating feature set into the SVM, performing parameter optimization on the SVM through an optimization algorithm, and obtaining an optimal parameter combination of a penalty factor c and a kernel function g of the SVM;
and 3-2, inputting the obtained optimal parameter combinations c and g into parameter setting of the SVM to obtain the optimized SVM.
Further, in the step 4, the lubricating feature set is optimized according to the SVM recognition accuracy, an optimal feature combination for the lubricating state recognition is obtained, and then the judgment of the lubricating state of the ball screw pair to be recognized is completed by combining the optimized SVM, and the specific process comprises the following steps:
step 4-1, setting a recognition accuracy threshold;
step 4-2, extracting all characteristic values corresponding to the lubrication characteristics with the maximum weight value from the lubrication characteristic set, inputting characteristic vectors formed by the characteristic values into the optimized SVM, judging whether the accuracy is smaller than an identification accuracy threshold value, if so, adding the lubrication characteristics with the maximum weight value to the optimal characteristic combination, and repeatedly executing the step aiming at the rest fields; otherwise, adding the lubrication characteristic with the maximum weight value to the optimal characteristic combination, and executing the next step;
and 4-3, acquiring vibration signals of the ball screw pair to be recognized, extracting a combination characteristic value of the vibration signals according to the optimal characteristic combination, inputting the combination characteristic value into the optimized SVM, and obtaining the lubrication state of the ball screw pair to be recognized.
Compared with the prior art, the invention has the following remarkable advantages: 1) the problems of difficult identification, low efficiency, low accuracy and the like of the lubricating state of the current ball screw pair can be effectively solved; 2) the vibration signal lubrication feature set is subjected to primary optimization by adopting a Relief-F algorithm, so that the dimensionality of the lubrication feature set is greatly reduced, the subsequent calculation time is reduced, and the identification efficiency is improved; 3) the lubricating characteristic set is secondarily optimized by setting the accuracy of the identification model, so that the optimal characteristic combination capable of representing the lubricating state can be effectively screened out, and the complexity of lubricating state identification is reduced while the accuracy of lubricating state identification is ensured; 4) the whole method is simple and convenient to operate, accurate, scientific and reasonable.
The present invention is described in further detail below with reference to the attached drawing figures.
Drawings
FIG. 1 is a flow chart of a method for identifying a lubrication state of a ball screw assembly according to an embodiment.
FIG. 2 is a schematic diagram of an exemplary 4-layer wavelet packet decomposition of a vibration signal.
FIG. 3 is a diagram illustrating the recognition effect of the test set after the secondary feature optimization of the vibration signal in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
In one embodiment, in combination with fig. 1, a method for identifying the lubrication state of a ball screw pair based on Relief-F and SVM is provided, and the method comprises the following steps:
step 1, collecting vibration signals of a ball screw pair in different lubrication states;
step 3, training an optimized SVM by using the lubricating feature set;
and 4, optimizing the lubricating feature set according to the SVM recognition accuracy to obtain the optimal feature combination of the lubricating state recognition, and then finishing the judgment of the lubricating state of the ball screw pair to be recognized by combining the optimized SVM.
Here, it should be noted that, step 3 and step 4 may be implemented by using other machine learning means such as an SVM, a neural network, and the like, and any algorithm that can achieve the purposes of training, classification, recognition, and the like is within the scope of the present invention.
Further, in one embodiment, step 1 collects vibration signals of the ball screw pair in different lubrication states, and is realized based on a ball screw pair vibration signal collection system which takes a data collection device, a computer and a sensor as a core.
Here, the sensor may employ a single-axis acceleration sensor, and may also employ a three-axis acceleration sensor, or the like.
Here, the sensor may be mounted at a position of a cross section of the nut flange, at a position of an end face of the flange, or the like, for different mounting structures of the ball screw pair.
Further, in one embodiment, the vibration signal feature is extracted in step 2 to construct a lubrication feature set, and the specific process includes:
step 2-1, extracting a plurality of groups of vibration data from the vibration signals;
step 2-2, performing wavelet packet transformation on each group of vibration data;
step 2-3, solving a time domain characteristic value of each group of vibration data;
step 2-4, obtaining the frequency domain characteristic value of each group of vibration data;
step 2-5, regarding each characteristic obtained in the process as a lubricating characteristic, and constructing a lubricating characteristic set according to all results obtained in the process, wherein each sample in the set is a lubricating characteristic value corresponding to each group of vibration data;
step 2-6, optimizing the lubricating feature set, specifically: and according to the weight value occupied by each lubrication characteristic pair in the lubrication state, screening out the characteristic value with the weight value larger than a preset weight threshold value from the lubrication characteristic set, and constructing a new lubrication characteristic set.
Further, in one embodiment, step 2-2 performs wavelet packet transformation on each group of vibration data, and the specific process includes:
(1) using a high-pass filter gk}k∈ZAnd a low pass filter hk}k∈ZPerforming full-band decomposition on the vibration data signal, wherein the formula is as follows:
in the formula, d represents wavelet coefficient of wavelet packet decomposition frequency band, k, l ∈ Z, and the two formulas respectively represent that the nth frequency band of the jth layer is decomposed into the 2n and 2n +1 frequency bands of the jth +1 layer;
(2) reconstructing the decomposed signal, wherein the reconstruction formula is as follows:
in the formula, pkAnd q iskAre respectively hkAnd gkThe dual filter of (2);
(3) calculating each node signal x after decompositionkmEnergy value of (2):
in the formula, ejkRepresenting the energy value of the frequency band of the kth node after the wavelet packet is subjected to the jth layer decomposition and reconstruction, wherein x is the amplitude of a reconstructed signal at a discrete point, and N is the signal length;
(4) calculating the energy ratio of each node frequency band:
in the formula, EjkIndicating the energy-occupied ratio of each node frequency band.
Exemplary preferably, in one of the embodiments, the wavelet packet decomposition selects a 4-level decomposition with a wavelet basis function of "db 10".
Further, in one embodiment, the time-domain feature values in step 2-3 include dimensional parameters and dimensionless parameters;
wherein, the dimension parameters comprise:
(1) mean value
Wherein x represents the discrete data amplitude of each vibration signal, and n represents the number of discrete data;
(2) root mean square value
(3) Square root amplitude
(4) Absolute mean value
(5) Variance (variance)
(6) Peak value: taking the maximum numbers in the signals, and calculating the arithmetic mean value of the numbers as a peak value;
the dimensionless parameters include:
(1) waveform index
(2) Peak index
(3) Pulse index
(4) Margin index
(5) Kurtosis index
Further, in one embodiment, the frequency domain feature values in step 2-4 include:
(1) frequency of center of gravity
In the formula (f)iThe instantaneous frequency of the vibration signal corresponding to time i, F (F)i) Representing the power spectrum amplitude corresponding to the instantaneous frequency;
(2) mean square frequency
(3) Frequency variance
Further, in one embodiment, the step 2-6 optimizes the lubricating feature set, specifically optimizes the lubricating feature set by using a Relief-F, and the specific process includes:
step 2-6-1, establishing a feature weight set W, and making each feature initial weight W (i) in the feature weight set 0, i-1, 2,3, …, N1，N1Representing the number of features, setting the initial iteration number s to be 1, and setting an iteration number threshold value M;
2-6-2, randomly extracting a characteristic sample r from the lubricating characteristic set;
step 2-6-3, finding out k in the same lubrication state as the characteristic sample r from the lubrication characteristic set0A nearest neighbor feature sample Qa，a＝1,2,3,…,k0And separately find k in a lubrication state different from that of the sample r0A nearest neighborCharacteristic sample Qa(c) And c represents the number of categories of non-homogeneous lubrication states;
and 2-6-4, updating the weight value of each feature in the feature weight set W, wherein the updating formula is as follows:
wherein p (c) represents the distribution probability of the c-th lubrication state, C (r) represents the lubrication state of the characteristic sample r, p (C (r)) represents the distribution probability of the lubrication state to which the characteristic sample r belongs, and diff (F, r, Q)a(c) ) represent feature samples r and Qa(c) Distance difference on ith feature F; wherein diff (F, r, Q)a) Discrete or continuous definition of feature data from a feature set is:
in the formula, Fr、FQaRespectively representing a feature sample r and a feature sample QaThe value on the ith feature F;
and 2-6-5, judging whether s is smaller than M, if so, increasing s by 1, returning to execute the step 2-6-2 until the condition is met, and then screening out a characteristic value with a weight value larger than a preset weight threshold value from the lubricating characteristic set to construct a new lubricating characteristic set.
Exemplarily and preferably, in one embodiment, the above-mentioned weight threshold of the Relief-F algorithm is 0.02, and the number of iterations M is set to 20.
Further, in one embodiment, the step 3 trains the optimization SVM by using the lubricating feature set, and the specific process includes:
step 3-1, inputting the lubricating feature set into the SVM, performing parameter optimization on the SVM through an optimization algorithm, and obtaining an optimal parameter combination of a penalty factor c and a kernel function g of the SVM;
and 3-2, inputting the obtained optimal parameter combinations c and g into parameter setting of the SVM to obtain the optimized SVM.
Exemplarily and preferably, in one embodiment, the optimization algorithm in the step 3-1 specifically adopts a grid search method or a genetic optimization algorithm or a particle swarm optimization algorithm.
Further, in one embodiment, the step 4 optimizes the lubrication feature set according to the SVM recognition accuracy to obtain the optimal feature combination of the lubrication state recognition, and then completes the judgment of the lubrication state of the ball screw pair to be recognized by combining the optimized SVM, and the specific process includes:
step 4-1, setting a recognition accuracy threshold;
step 4-2, extracting all characteristic values corresponding to the lubrication characteristics with the maximum weight value from the lubrication characteristic set, inputting characteristic vectors formed by the characteristic values into the optimized SVM, judging whether the accuracy is smaller than an identification accuracy threshold value, if so, adding the lubrication characteristics with the maximum weight value to the optimal characteristic combination, and repeatedly executing the step aiming at the rest fields; otherwise, adding the lubrication characteristic with the maximum weight value to the optimal characteristic combination, and executing the next step;
and 4-3, acquiring vibration signals of the ball screw pair to be recognized, extracting a combination characteristic value of the vibration signals according to the optimal characteristic combination, inputting the combination characteristic value into the optimized SVM, and obtaining the lubrication state of the ball screw pair to be recognized.
Further, in one embodiment, the step 3 and the step 4 may be performed simultaneously, and the specific process includes:
step 34-1, aiming at the lubricating feature set, forming a feature vector by all feature values corresponding to the same feature in the set;
step 34-2, selecting a characteristic vector corresponding to the lubrication characteristic with the maximum weight value from the lubrication characteristic sets to construct a new lubrication characteristic set;
step 34-3, inputting the new lubricating feature set into the SVM, performing parameter optimization on the SVM through an optimization algorithm, and obtaining the optimal parameter combination of a penalty factor c and a kernel function g of the SVM;
step 34-4, inputting the obtained optimal parameter combinations c and g into parameter setting of the SVM to obtain the optimized SVM;
step 34-5, setting a recognition accuracy threshold;
step 34-6, extracting a feature vector corresponding to the lubrication feature with the largest weight value from the new lubrication feature set, inputting the feature vector into the optimized SVM, judging whether the accuracy is smaller than an identification accuracy threshold value, if so, adding the lubrication feature with the largest weight value to the optimal feature combination, extracting the feature vector corresponding to the lubrication feature with the largest weight value from the rest feature domains (if the feature corresponding to wavelet packet transformation is extracted, the rest feature domains represent a time domain and a frequency domain), further constructing a new lubrication feature set, and then returning to the step 34-3; otherwise, adding the feature with the maximum weight value to the optimal feature combination, and executing the next step;
and step 34-7, acquiring vibration signals of the ball screw pair to be identified, extracting a combination characteristic value of the vibration signals according to the optimal characteristic combination, inputting the combination characteristic value into the optimized SVM, and obtaining the lubrication state of the ball screw pair to be identified.
As a specific example, the method of the present invention is verified, and the specific contents include:
1. a vibration signal of an X-axis lead screw after an automatic lubricating pump is closed is collected on a certain vertical machining center, and 60 groups of data of 'lubrication excellent', 'lubrication middle', 'lubrication poor' are extracted.
2. Performing wavelet packet 4-layer decomposition on the acquired vibration data as shown in fig. 2, setting a wavelet basis function to be db10, extracting energy values of all frequency bands of the last layer, and combining time domain dimensional parameter mean, root mean square value, root mean square amplitude, absolute mean, variance, peak value and dimensionless parameter waveform index, peak index, pulse index, margin index, kurtosis index, frequency domain characteristic center-of-gravity frequency, frequency variance and mean square frequency to construct a lubrication characteristic set.
3. And (3) performing primary optimization on the lubricating feature set by using a Relief-F, setting the weight value to be 0.02, screening out feature values meeting the requirements, and reconstructing the lubricating feature set as shown in table 1.
TABLE 1 Primary optimization results for feature set
4. Randomly dividing the lubricating feature set into a training sample set and a testing sample set according to the proportion of 3:1, arranging the feature vectors (all feature values corresponding to each feature form a feature vector) in each set in a descending order according to the weight value of the lubricating feature, and selecting the feature vector with the largest weight value from the training sample set to construct a new training sample set. An improved particle swarm optimization algorithm is adopted to find the optimal parameter combination of a penalty factor c and a kernel function g of the SVM, initial parameters of a self-learning factor 1.3, a social learning factor 1.2 and a population quantity 40 are set, initial inertia weights and termination inertia weights are respectively 0.1 and 0.9, and the iteration times are 200.
5. Setting a recognition accuracy threshold value to be 95%, inputting the feature vector with the maximum weight value in the new training sample set into the optimized SVM, if the accuracy is smaller than the set threshold value, adding the lubrication feature with the maximum weight value into the optimal feature combination, extracting the feature vector with the maximum weight value in the rest domains from the training sample set to construct a new training sample set, and then returning to the execution process 4 and the process 5 until the requirements are met, and finishing the secondary optimization. The secondary preferred results for the lubricating feature set are shown in table 2 below:
feature set secondary preferred results are shown below in table 2:
TABLE 2 lubricating status characteristics set Secondary optimization results
The features extracted after the secondary optimization are used as the optimal feature combination for state recognition, a new test set is constructed by the features and is input into the SVM for test verification, and the accuracy of the test result is 93.33%, as shown in FIG. 3.
According to the method for identifying the lubricating state of the ball screw pair based on the Relief-F and the SVM, the characteristic values of all domains are optimized through the Relief-F, the lubricating state is classified and predicted through the SVM to judge the lubricating state of the ball screw pair, and the problems that the existing ball screw pair is difficult to identify the lubricating state, low in efficiency, low in accuracy and the like can be effectively solved.
Claims (9)
1. A method for identifying the lubrication state of a ball screw pair is characterized by comprising the following steps:
step 1, collecting vibration signals of a ball screw pair in different lubrication states;
step 2, extracting vibration signal characteristics and constructing a lubrication characteristic set;
step 3, training an optimized SVM by using the lubricating feature set;
and 4, optimizing the lubricating feature set according to the SVM recognition accuracy to obtain the optimal feature combination of the lubricating state recognition, and then finishing the judgment of the lubricating state of the ball screw pair to be recognized by combining the optimized SVM.
2. The method for identifying the lubrication state of the ball screw pair according to claim 1, wherein the step 2 of extracting the vibration signal features and constructing the lubrication feature set comprises the following specific processes:
step 2-1, extracting a plurality of groups of vibration data from the vibration signals;
step 2-2, performing wavelet packet transformation on each group of vibration data;
step 2-3, solving a time domain characteristic value of each group of vibration data;
step 2-4, obtaining the frequency domain characteristic value of each group of vibration data;
step 2-5, regarding each characteristic obtained in the process as a lubricating characteristic, and constructing a lubricating characteristic set according to all results obtained in the process, wherein each sample in the set is a lubricating characteristic value corresponding to each group of vibration data;
step 2-6, optimizing the lubricating feature set, specifically: and according to the weight value occupied by each lubrication characteristic pair in the lubrication state, screening out the characteristic value with the weight value larger than a preset weight threshold value from the lubrication characteristic set, and constructing a new lubrication characteristic set.
3. The method for identifying the lubrication state of the ball screw pair according to claim 2, wherein the step 2-2 is to perform wavelet packet transformation on each group of vibration data, and the specific process comprises the following steps:
(1) using a high-pass filter gk}k∈ZAnd a low pass filter hk}k∈ZPerforming full-band decomposition on the vibration data signal, wherein the formula is as follows:
in the formula, d represents wavelet coefficient of wavelet packet decomposition frequency band, k, l ∈ Z, and the two formulas respectively represent that the nth frequency band of the jth layer is decomposed into the 2n and 2n +1 frequency bands of the jth +1 layer;
(2) reconstructing the decomposed signal, wherein the reconstruction formula is as follows:
in the formula, pkAnd q iskAre respectively hkAnd gkThe dual filter of (2);
(3) calculating each node signal x after decompositionkmEnergy value of (2):
in the formula, ejkRepresenting the energy value of the frequency band of the kth node after the wavelet packet is subjected to the jth layer decomposition and reconstruction, wherein x is the amplitude of a reconstructed signal at a discrete point, and N is the signal length;
(4) calculating the energy ratio of each node frequency band:
in the formula, EjkIndicating the energy-occupied ratio of each node frequency band.
4. The method for identifying the lubrication state of the ball screw pair according to claim 3, wherein the time domain characteristic values in the step 2-3 comprise dimensional parameters and dimensionless parameters;
wherein, the dimension parameters comprise:
(1) mean value
Wherein x represents the discrete data amplitude of each vibration signal, and n represents the number of discrete data;
(2) root mean square value
(3) Square root amplitude
(4) Absolute mean value
(5) Variance (variance)
(6) Peak value: taking the maximum numbers in the signals, and calculating the arithmetic mean value of the numbers as a peak value;
the dimensionless parameters include:
(1) waveform index
(2) Peak index
(3) Pulse index
(4) Margin index
(5) Kurtosis index
5. The method for identifying the lubrication state of the ball screw pair according to claim 4, wherein the frequency domain characteristic values in steps 2 to 4 include:
(1) frequency of center of gravity
In the formula (f)iThe instantaneous frequency of the vibration signal corresponding to time i, F (F)i) Representing the power spectrum amplitude corresponding to the instantaneous frequency;
(2) mean square frequency
(3) Frequency variance
6. The method for identifying the lubrication state of the ball screw pair according to claim 5, wherein the step 2-6 is implemented by optimizing a lubrication feature set, specifically optimizing the lubrication feature set by using a Relief-F, and the specific process includes:
step 2-6-1, establishing a feature weight set W, and making each feature initial weight W (i) in the feature weight set 0, i-1, 2,3, …, N1，N1Representing the number of features, setting the initial iteration number s to be 1, and setting an iteration number threshold value M;
2-6-2, randomly extracting a characteristic sample r from the lubricating characteristic set;
step 2-6-3, finding out k in the same lubrication state as the characteristic sample r from the lubrication characteristic set0A nearest neighbor feature sample Qa，a＝1,2,3,…,k0And separately find k in a lubrication state different from that of the sample r0A nearest neighbor feature sample Qa(c) And c represents the number of categories of non-homogeneous lubrication states;
and 2-6-4, updating the weight value of each feature in the feature weight set W, wherein the updating formula is as follows:
wherein p (c) represents the distribution probability of the c-th lubrication state, C (r) represents the lubrication state of the characteristic sample r, p (C (r)) represents the distribution probability of the lubrication state to which the characteristic sample r belongs, and diff (F, r, Q)a(c) ) represent feature samples r and Qa(c) Distance difference on ith feature F; wherein diff (F, r, Q)a) Discrete or continuous definition of feature data from a feature set is:
in the formula, Fr、FQaRespectively representing a feature sample r and a feature sample QaThe value on the ith feature F;
and 2-6-5, judging whether s is smaller than M, if so, increasing s by 1, returning to execute the step 2-6-2 until the condition is met, and then screening out a characteristic value with a weight value larger than a preset weight threshold value from the lubricating characteristic set to construct a new lubricating characteristic set.
7. The method for identifying the lubrication state of the ball screw pair according to claim 6, wherein the step 3 of training the optimization SVM by using the lubrication feature set comprises the following specific processes:
step 3-1, inputting the lubricating feature set into the SVM, performing parameter optimization on the SVM through an optimization algorithm, and obtaining an optimal parameter combination of a penalty factor c and a kernel function g of the SVM;
and 3-2, inputting the obtained optimal parameter combinations c and g into parameter setting of the SVM to obtain the optimized SVM.
8. The method for identifying the lubrication state of the ball screw pair according to claim 7, wherein the optimization algorithm in step 3-1 specifically adopts a grid search method or a genetic optimization algorithm or a particle swarm optimization algorithm.
9. The method for identifying the lubrication state of the ball screw pair according to claim 7 or 8, wherein in the step 4, the set of lubrication features is optimized according to the SVM recognition accuracy to obtain the optimal feature combination for the lubrication state recognition, and then the judgment on the lubrication state of the ball screw pair to be recognized is completed by combining the optimized SVM, and the specific process comprises the following steps:
step 4-1, setting a recognition accuracy threshold;
step 4-2, extracting all characteristic values corresponding to the lubrication characteristics with the maximum weight value from the lubrication characteristic set, inputting characteristic vectors formed by the characteristic values into the optimized SVM, judging whether the accuracy is smaller than an identification accuracy threshold value, if so, adding the lubrication characteristics with the maximum weight value to the optimal characteristic combination, and repeatedly executing the step aiming at the rest fields; otherwise, adding the lubrication characteristic with the maximum weight value to the optimal characteristic combination, and executing the next step;
and 4-3, acquiring vibration signals of the ball screw pair to be recognized, extracting a combination characteristic value of the vibration signals according to the optimal characteristic combination, inputting the combination characteristic value into the optimized SVM, and obtaining the lubrication state of the ball screw pair to be recognized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010194479.2A CN111360578B (en) | 2020-03-19 | 2020-03-19 | Method for identifying lubricating state of ball screw pair |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010194479.2A CN111360578B (en) | 2020-03-19 | 2020-03-19 | Method for identifying lubricating state of ball screw pair |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111360578A true CN111360578A (en) | 2020-07-03 |
| CN111360578B CN111360578B (en) | 2021-03-26 |
Family
ID=71200646
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010194479.2A Active CN111360578B (en) | 2020-03-19 | 2020-03-19 | Method for identifying lubricating state of ball screw pair |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111360578B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112763155A (en) * | 2021-02-02 | 2021-05-07 | 南京信息工程大学 | Pressure container gas leakage detection method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4761876A (en) * | 1986-04-18 | 1988-08-09 | Dynamotion Corporation | High speed precision drilling system |
| CN104162749A (en) * | 2013-05-17 | 2014-11-26 | 宝山钢铁股份有限公司 | State diagnosis method of multi-slippery bell high-speed ball thread screw rod and device for same |
| CN108844725A (en) * | 2018-04-24 | 2018-11-20 | 天津大学 | A kind of automobile engine bearing wear fault diagnosis method |
| CN109746928A (en) * | 2019-01-30 | 2019-05-14 | 南京航空航天大学 | A kind of the intelligent flexible production line and its operation method of removable dual robot milling automatic |
| CN109916624A (en) * | 2019-03-22 | 2019-06-21 | 南京理工大学 | A kind of ball screw assembly, fatigue failure diagnostic method based on Hilbert Huang |
-
2020
- 2020-03-19 CN CN202010194479.2A patent/CN111360578B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4761876A (en) * | 1986-04-18 | 1988-08-09 | Dynamotion Corporation | High speed precision drilling system |
| CN104162749A (en) * | 2013-05-17 | 2014-11-26 | 宝山钢铁股份有限公司 | State diagnosis method of multi-slippery bell high-speed ball thread screw rod and device for same |
| CN108844725A (en) * | 2018-04-24 | 2018-11-20 | 天津大学 | A kind of automobile engine bearing wear fault diagnosis method |
| CN109746928A (en) * | 2019-01-30 | 2019-05-14 | 南京航空航天大学 | A kind of the intelligent flexible production line and its operation method of removable dual robot milling automatic |
| CN109916624A (en) * | 2019-03-22 | 2019-06-21 | 南京理工大学 | A kind of ball screw assembly, fatigue failure diagnostic method based on Hilbert Huang |
Non-Patent Citations (1)
| Title |
|---|
| 王志荣等: "基于WPD与GA-BP网络的滚珠丝杠副润滑失效诊断研究", 《组合机床与自动化加工技术》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112763155A (en) * | 2021-02-02 | 2021-05-07 | 南京信息工程大学 | Pressure container gas leakage detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111360578B (en) | 2021-03-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Fu et al. | Fault diagnosis for rolling bearings based on composite multiscale fine-sorted dispersion entropy and SVM with hybrid mutation SCA-HHO algorithm optimization | |
| CN109186973B (en) | Mechanical fault diagnosis method of unsupervised deep learning network | |
| CN111539152B (en) | Rolling bearing fault self-learning method based on two-stage twin convolutional neural network | |
| CN109685366A (en) | Equipment health state evaluation method based on mutation data | |
| CN108918137B (en) | Gear box fault diagnosis device and method based on improved WPA-BP neural network | |
| CN110110768B (en) | Rolling bearing fault diagnosis method based on parallel feature learning and multiple classifiers | |
| CN110647923A (en) | Variable working condition mechanical fault intelligent diagnosis method based on self-learning under small sample | |
| CN111046945A (en) | Fault type and damage degree diagnosis method based on combined convolutional neural network | |
| CN111458148A (en) | CBAM-based convolutional neural network rolling bearing fault diagnosis method | |
| CN109781411A (en) | A kind of combination improves the Method for Bearing Fault Diagnosis of sparse filter and KELM | |
| Di et al. | Ensemble deep transfer learning driven by multisensor signals for the fault diagnosis of bevel-gear cross-operation conditions | |
| CN109726802B (en) | Machine learning prediction method for wind speed in railway and wind farm environment | |
| CN110487547A (en) | Fault Diagnosis of Roller Bearings under variable working condition based on vibrorecord and transfer learning | |
| CN113375941A (en) | Open set fault diagnosis method for high-speed motor train unit bearing | |
| CN114429152A (en) | Rolling bearing fault diagnosis method based on dynamic index antagonism self-adaption | |
| CN110737976A (en) | mechanical equipment health assessment method based on multi-dimensional information fusion | |
| CN112200263B (en) | Self-organizing federal clustering method applied to power distribution internet of things | |
| CN112329520B (en) | Truck bearing fault identification method based on generation countermeasure learning | |
| CN111360578B (en) | Method for identifying lubricating state of ball screw pair | |
| CN114462480A (en) | Multi-source sensor rolling mill fault diagnosis method based on non-equilibrium data set | |
| CN115392333A (en) | Equipment fault diagnosis method based on improved end-to-end ResNet-BilSTM dual-channel model | |
| CN115587290A (en) | Aero-engine fault diagnosis method based on variational self-coding generation countermeasure network | |
| CN113887717A (en) | Method for predicting neural network training duration based on deep learning | |
| CN112380782A (en) | Rotating equipment fault prediction method based on mixed indexes and neural network | |
| Zhang et al. | Gearbox health condition identification by neuro-fuzzy ensemble |
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 |