阅读说明：本技术 一种基于自适应分段pca的非平稳钻削过程监测方法 (Non-stationary drilling process monitoring method based on adaptive segmented PCA ) 是由 王国锋 李雨航 丛君宇 李杰峰 杨凯 于 2021-07-26 设计创作，主要内容包括：本发明公开一种基于自适应分段PCA的非平稳钻削过程监测方法,该方法通过采集机床的控制系统监控信息以及所安装的力传感器信号特征,判断钻削加工开始时间,接着根据钻头的引导部分长度和钻孔深度以及轴向进给速度,计算出每一钻削阶段的时间,以实现对钻削阶段的自适应分段,准确截取稳定钻削阶段的信号。采集不同刀具磨损值对应的轴向力信号,进行低通滤波,再进行特征选择排除无效特征,再根据加工工艺的要求定义训练集和测试集,以正常磨损的特征矩阵建立PCA监测模型,以测试数据集检验所建立的模型的准确性。(The invention discloses a non-steady drilling process monitoring method based on adaptive segmented PCA, which judges the drilling processing starting time by collecting the control system monitoring information of a machine tool and the signal characteristics of an installed force sensor, and then calculates the time of each drilling stage according to the length of a guide part of a drill bit, the drilling depth and the axial feeding speed so as to realize the adaptive segmentation of the drilling stages and accurately intercept the signals of stable drilling stages. Axial force signals corresponding to different tool wear values are collected, low-pass filtering is carried out, then feature selection is carried out to eliminate invalid features, a training set and a testing set are defined according to the requirements of a machining process, a PCA monitoring model is established according to a feature matrix of normal wear, and the accuracy of the established model is checked according to a testing data set.)

1. A non-steady drilling process monitoring method based on self-adaptive segmented PCA is characterized in that the drilling processing starting time is judged according to monitoring information of a numerical control system and signal characteristics output by an installed force sensor, and then the time of each drilling stage is calculated according to the length of a guide part of a drill bit, the drilling depth and the axial feeding speed, so that the self-adaptive segmentation of the drilling stages is realized, and signals of stable drilling stages are accurately intercepted.

2. The method for monitoring a non-stationary drilling process based on adaptive segmented PCA as claimed in claim 1, comprising the steps of:

step one, a signal acquisition platform is built, and a dynamometer is installed for acquiring axial force signals in the drilling process; reading NC codes of numerical control machining, acquiring technological parameters of drilling machining, and acquiring position information of a cutter in a numerical control system;

judging when the tool nose contacts the workpiece through a numerical control code or a dynamometer, and dividing drilling into a drilling-in stage, a stable drilling stage and a drilling-out stage according to technological parameters; defining a complete stage from the contact of the tool nose with the workpiece to the guide part of the drill bit as a drilling stage, starting to acquire force signals after the complete drilling, stopping acquiring until the tool nose is about to drill the workpiece, and acquiring force signal data in a stable drilling stage for judging and identifying tool abrasion;

performing frequency spectrum analysis on the acquired force signal data in the stable drilling stage to determine the range of effective frequency components, and then performing low-pass filtering on the force signal data in the stable drilling stage; then extracting time domain characteristics and frequency domain characteristics to form a characteristic matrix;

step four, feature selection is carried out on the feature matrix extracted in the step three stages, meaningless feature values are screened out, and the feature matrix after feature selection is obtained;

acquiring drilling axial force signals of the tool under different wear conditions, repeating the second step, the third step and the fourth step, establishing a training set by using the normal wear signals, and establishing a test set by using the normal wear signals and the abnormal wear signals;

establishing a tool wear state monitoring model based on PCA (principal component analysis) by using the training set, and checking the effectiveness of the tool wear state monitoring model by using the test set;

step seven, determining that the cutter wear state monitoring model is effective, and storing the characteristic mean value vector, the characteristic standard deviation vector, the number of the principal elements, the SPE statistic control limit and the T under the normal cutter wear condition according to the calculation result of the step six²A statistic control limit, a feature vector variance matrix and a load matrix; using the 7 parameters as parameters of a tool wear state monitoring model; and finally, nesting the tool wear state monitoring model into corresponding machining state monitoring software used in a machine tool numerical control system.

3. The method of claim 1, wherein the time domain features extracted in step three are mean, root mean square, variance, peak-to-peak, kurtosis, skewness, peak factor, pulser factor, form factor, and the frequency domain power spectrum features extracted are band power sum, band power mean, band power skewness, band power kurtosis, band power variance, band power peak, ratio of band power peak to mean, and peak frequency, and since the intercepted signal of the stable drilling stage is a stationary signal, no time-frequency feature is extracted.

Technical Field

The invention relates to the technical field of state monitoring and identification of a machine tool cutter, in particular to a non-stable drilling process monitoring method based on self-adaptive segmented PCA.

Background

At present, a tool state monitoring method is a process of filtering and denoising a machining signal acquired by a sensor, extracting characteristics and finally establishing a tool state monitoring model. Although pattern recognition algorithms such as an Artificial Neural Network (ANN), a Hidden Markov Model (HMM), a Support Vector Machine (SVM), a fuzzy clustering and the like have been developed at present and applied to the research of tool state monitoring, tool state monitoring models established for a drilling process are few because signals of the drilling process generally represent non-stationarity and greatly fluctuate, and if signals of a full process are analyzed, problems exist, firstly, the whole process contains a lot of useless noise information, and secondly, the fluctuation of the drilling signals is high, which can greatly affect the monitoring result.

Publication No. isCN105834835BThe patent introduces a tool wear online monitoring method based on multi-scale principal component analysis, which decomposes a training sample into a plurality of scales through wavelet decomposition, and performs dimension reduction processing on each scale and the whole body by using principal component analysis to construct a model; performing wavelet decomposition on the test sample according to the number of layers of the training sample, loading the data under each scale to the principal component model of the corresponding scale, and calculating the SPE and T of the data under each scale²Statistics; forming a new test sample for signals on a significant scale by using a wavelet reconstruction method, loading the new test sample on an integral principal component model, and calculating SPE and T²And (4) counting the quantity and judging whether the quantity exceeds the control limit, if so, indicating that the cutter is abnormally worn in the process, otherwise, indicating that the cutter is normally worn, thereby monitoring the wear state of the cutter.

However, the above patent does not specify the drilling process, and if the principal component analysis is performed on the signal of the whole process, the accuracy of the monitoring result is very low, because the principal component analysis algorithm is poor for monitoring data with large fluctuation degree. For drilling machining, the tool state monitoring model is characterized by being more distinct compared with other machining modes, so that when the tool state monitoring model in other machining processes is likely to fail in the drilling process, the drilling process is generally divided into three stages from the time that a tool nose contacts a workpiece to the time that a drill bit guide part completely enters the workpiece, stable drilling is carried out, the tool nose is about to drill the workpiece to the time that the drill bit completely drills the workpiece, signal fluctuation in the whole drilling process is large, the monitoring model is likely to fail, and signals in different drilling machining stages can be reasonably utilized to accurately monitor the wear state of the drill bit. Namely, a monitoring method which can adaptively segment the drilling process and well identify the non-stable drilling process is needed.

Disclosure of Invention

The invention aims to overcome the defects in the prior art and provides a non-stationary drilling process monitoring method based on adaptive segmented PCA.

The purpose of the invention is realized by the following technical scheme:

a non-steady drilling process monitoring method based on self-adaptive segmented PCA judges the drilling processing starting time according to the monitoring information of a numerical control system and the signal characteristics output by an installed force sensor, and then calculates the time of each drilling stage according to the length of a guide part of a drill bit, the drilling depth and the axial feeding speed so as to realize the self-adaptive segmentation of the drilling stage and accurately intercept and stabilize the signals of the drilling stage.

Further, a non-stationary drilling process monitoring method based on adaptive segmented PCA comprises the following steps:

step one, a signal acquisition platform is built, and a dynamometer is installed for acquiring axial force signals in the drilling process; reading NC codes of numerical control machining, acquiring technological parameters of drilling machining, and acquiring position information of a cutter in a numerical control system;

judging when the tool nose contacts the workpiece through a numerical control code or a dynamometer, and dividing drilling into a drilling-in stage, a stable drilling stage and a drilling-out stage according to technological parameters; defining a complete stage from the contact of the tool nose with the workpiece to the guide part of the drill bit as a drilling stage, starting to acquire force signals after the complete drilling, stopping acquiring until the tool nose is about to drill the workpiece, and acquiring force signal data in a stable drilling stage for judging and identifying tool abrasion;

performing frequency spectrum analysis on the acquired force signal data in the stable drilling stage to determine the range of effective frequency components, and then performing low-pass filtering on the force signal data in the stable drilling stage; then extracting time domain characteristics and frequency domain characteristics to form a characteristic matrix;

step four, feature selection is carried out on the feature matrix extracted in the step three stages, meaningless feature values are screened out, and the feature matrix after feature selection is obtained;

acquiring drilling axial force signals of the tool under different wear conditions, repeating the second step, the third step and the fourth step, establishing a training set by using the normal wear signals, and establishing a test set by using the normal wear signals and the abnormal wear signals;

establishing a tool wear state monitoring model based on PCA (principal component analysis) by using the training set, and checking the effectiveness of the tool wear state monitoring model by using the test set;

step seven, determining that the cutter wear state monitoring model is effective, and storing the characteristic mean value vector, the characteristic standard deviation vector, the number of the principal elements, the SPE statistic control limit and the T under the normal cutter wear condition according to the calculation result of the step six²A statistic control limit, a feature vector variance matrix and a load matrix; using the 7 parameters as parameters of a tool wear state monitoring model; and finally, nesting the tool wear state monitoring model into corresponding machining state monitoring software used in a machine tool numerical control system.

Further, the time domain features extracted in the third step are mean value, root mean square, mean square value, variance, peak-to-peak value, kurtosis, skewness, peak factor, pulse factor and form factor, the extracted frequency domain power spectrum features are frequency band power sum, frequency band power mean value, frequency band power skewness, frequency band power kurtosis, frequency band power variance, frequency band power peak value, ratio of frequency band power peak value to mean value and peak frequency, and the intercepted signal in the stable drilling stage is a stable signal without extracting time-frequency features.

Compared with the prior art, the technical scheme of the invention has the following beneficial effects:

1. the method can very quickly and accurately identify the abnormal wear signal of the tool in the non-steady drilling process, does not blindly analyze the signal of the whole drilling process through accurately dividing the drilling signal, only collects the signal extraction characteristics of the steady drilling stage, establishes a PCA monitoring model, realizes the monitoring of the non-steady drilling process, reduces the quantity of the collected data and improves the running speed of the tool wear state monitoring model compared with the tool wear state monitoring model which collects the signal of the whole drilling process.

2. The method selects stable drilling stage signals with more gradual changes according to the characteristic of large fluctuation of drilling axial force signals, reduces the fluctuation degree of data used for principal component analysis, and improves the accuracy and robustness of a tool wear state monitoring model.

3. The method can also reduce the dependence on the experience of machine tool operators, obtain the working state of the cutter in real time through the online monitoring of the abrasion degree of the cutter, give reference to whether to change the cutter or not, and has great significance for developing highly automated and intelligent production and processing.

Drawings

FIG. 1 is a flow chart of a method for monitoring a non-stationary drilling process based on adaptive segmented PCA in accordance with the present invention;

FIG. 2 is a diagram illustrating the labeling of several important dimensions in step S2;

FIG. 3 is a waveform diagram of an example drilling axial force signal.

Detailed Description

The invention is described in further detail below with reference to the figures and specific examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

As shown in FIG. 1, the invention provides an online monitoring method for the wear state of a drilling tool, which comprises the following steps:

step S1, a signal acquisition platform is built, an axial force sensor is installed, and an axial force signal in the drilling process is acquired; reading the NC code of numerical control machining, and acquiring the machining parameters of the machining equipment such as: spindle speed n, feed rate a, etc., parameters of the workpiece performing the drilling process, such as: the diameter d of the drill bit and the length l of a guide part of the drill bit are used for processing the thickness h of the drilled hole of the workpiece, and the position information of the drill bit during processing is collected from a numerical control system.

Step S2, the time when the drill reaches the surface of the workpiece from the initial position is calculated as t₀Z/a, where Z is the distance from the bottom of the drill to the surface of the workpiece, the time from when the workpiece contacts the surface of the workpiece to when the pilot portion completely enters the workpiece, and the calculation formula is t₁I.e. the phase in which the cutting edges are all involved in the cutting, the time calculation formula of which is t₂(h-l)/a, then (t) is counted as zero time from the start of the machine operation₀+t₁) To (t)₀+t₁+t₂) The signal of the time period is taken as an effective signal. The stage division of an example is shown in figure 3, wherein stage A is that the tool nose contacts the workpiece until the guiding part of the drill completely enters the workpiece, stage B is stable drilling stage, stage C is that the tool nose is about to drill the workpiece until the drill completely drills the workpiece, and stage D is tool retracting stage of drilling.

And step S3, acquiring signals obtained by drilling with drills with different wear degrees, performing spectrum analysis on the signal data in the stable stage defined in the step S2, calculating an amplitude spectrum by adopting DFT (discrete Fourier transform) to determine the range of 0-fc of effective frequency components, and performing low-pass filtering on the signals by taking fc as cut-off frequency. And then extracting 11 time domain features and 8 frequency domain power spectrum features to form a feature matrix containing 19 features.

And step S4, selecting the characteristics of the characteristic matrix extracted in the step S3, selecting the characteristics with high linear correlation degree and good robustness with the wear value, removing the meaningless characteristic value, and obtaining the characteristic matrix after the characteristic selection.

Step S5, the feature matrix obtained in step S4 is divided into two categories, generally according to the tool wear value requirement in actual production, where category a is the feature that is considered to be extracted from the drill bit machining axial force signal with acceptable wear degree in production, category B is the feature that is considered to be extracted from the drill bit machining axial force signal with unacceptable tool wear degree in production, and machining with a drill bit reaching this wear degree may cause abnormal machining conditions.

And S6, selecting the A-class characteristics containing the normal wear value of the tool in the step S5 to establish a training set, training a tool wear state monitoring model based on PCA, establishing a test set by using the characteristic matrixes in the A-class and the B-class in the step S5, and checking the accuracy of the established model.

Step S7, according to the calculation result of step S6, storing the characteristic mean value vector, the characteristic standard deviation vector, the main element number, the SPE statistic control limit, and the T obtained by establishing the cutter wear state monitoring model by using the training set²And the statistical control limit, the eigenvector variance matrix and the load matrix are used as main parameters of the established PCA-based drilling tool wear state monitoring model.

Step S2 specifically includes the following processing:

step 2.1, the initial height Z of the cutter is obtained according to the numerical control machining code, so that the time t of the drill reaching the surface of the workpiece from the initial position is calculated₀Z/a ', where a ' is the speed of the axial feed at this time (there may be a fast feed at the beginning of the process, a ' may be the feed speed a of the drill hole, or a fast feed speed), in which case the following calculation is: the time from the contact of the workpiece with the surface of the workpiece until the leading portion completely enters the workpiece is calculated by the formula t₁I.e. the phase in which the cutting edges are all involved in the cutting, the time calculation formula of which is t₂(h-l)/a, then (t) is counted as zero time from the start of the machine operation₀+t₁) To (t)₀+t₁+t₂) The signal of the time period is taken as an effective signal.

Step 2.2, if the fast feeding speed is not easy to determine, the contact time of the tool and the workpiece can be captured by a sensor and recorded as 0 time, and the subsequent calculation is as follows: the time from the contact of the workpiece with the surface of the workpiece until the leading portion completely enters the workpiece is calculated by the formula t₁I.e. the phase in which the cutting edges are all involved in the cutting, the time calculation formula of which is t₂(h-l)/a, then the time t is counted as zero from the start of the machine operation, and t is counted₁To t₁+t₂The signal of the time period is taken as an effective signal.

Step S3 specifically includes the following processing:

the Discrete Fourier Transform (DFT) formula in step 3.1 is as follows:

for a data sequence x of a signal_iThe data length is q, then x_iThe calculation formula of the N-point DFT is as follows:

where k represents the kth spectrum of the fourier transform.

In step 3.2, x is used_iRepresenting the data sequence of the signal and q representing the length of the data, the calculation formulas of the 11 time domain features of the signal are respectively as follows:

1. mean value: the average amplitude of a segment of signal is represented by the formula:

2. root mean square: the effective value of a section of signal is represented by the following calculation formula:

3. the mean square value is the square of the root mean square.

4. Variance: the degree of deviation of a section of signal from the average value is represented by the following formula:

5. peak-to-peak value: the difference between the maximum value and the minimum value in a section of signal is represented by the following calculation formula: x is the number of_pp＝Max(x_i)-Min(x_i)。

6. Peak value: the maximum value of the absolute value of the signal in a section of signal is represented by the following formula: x is the number of_p＝Max(abs(x_i))。

7. Kurtosis: indicating the degree of waveform steepness in a segment of the signal,the reaction processing system changes according to the following calculation formula:

8. skewness: the physical quantity representing the symmetry degree of a section of signal waveform is calculated by the following formula:

9. crest factor: the ratio of the peak value to the root mean square is calculated by the formula:

10. pulse factor: the ratio of the peak value to the average value is calculated by the formula:

11. form factor: the ratio of the root mean square to the average is calculated as:

in step 3.3, the power spectrum is calculated by a Welch method, and the obtained power spectrum density is P_kThe frequency domain power spectrum characteristic calculation formula of m (dividing the data sequence into m segments) is as follows:

1. band power sum:

2. average power of frequency band:

3. band power skewness:

4. band power kurtosis:

5. band power variance:

6. band power peak: pfp ═ Max (P)_k)。

7. Ratio of band power peak to mean:

8. peak frequency: the frequency corresponding to the peak of the power spectrum.

Step S6 specifically includes the following processing:

the process of establishing the tool wear state monitoring model based on principal component analysis in the step 6.1 is as follows:

1. the feature set in the normal wear state, that is, the training set mentioned in step S5, is selected, and is first normalized with respect to the features of different dimensions, so that the interference of the dimensions is eliminated, and data with a mean value of 0 and a standard deviation of 1 is obtained.

2. And calculating a correlation coefficient matrix of the feature matrix subjected to the standardization processing.

3. And solving the eigenvalue and the corresponding eigenvector of the correlation coefficient matrix, and arranging the eigenvalue and the corresponding eigenvector from large to small.

4. Determining the number k of principal components according to the cumulative contribution rate method, and knowing var (y)_k)＝λ_kWill beAnd defining the contribution rate of the kth principal component, wherein p represents the number of features used when a tool wear state monitoring model based on principal component analysis is established, and the contribution rate represents the proportion of information contained in the principal component to total data information. Scale typeThe number of principal components with the cumulative contribution rate of more than 90% is taken as the cumulative contribution rate of the principal componentsAnd find the corresponding load matrix P_k。

5. Calculating the square prediction error SPE at each sample point and the SPE control limit of the model, wherein the SPE statistic calculation formula at a certain sample point can be expressed as:

SPE_i＝X_i(I-P_kP_k ^T)X_i ^T

whereinσ is a diagonal matrix of eigenvalues of the first k principal elements, X_iIs normalized feature of the ith sample, I is identity matrix, P_kIs a load matrix when establishing a model.

The SPE control limit is calculated as follows:

t at a certain sample point i²The statistic calculation formula is as follows:

T_i ²＝X_iP_kO^-1P_kX_i ^T

T²the control limit is calculated by the formula:

where O is a diagonal matrix of eigenvalues of the first k principal elements, F_k,m-k,αIs the F distribution critical value under the conditions that the value checking level is alpha, the degree of freedom is k and n-1.

In general, application T²And when the SPE statistic is used for monitoring the system state, whether the processing is in a normal state or not is judged according to whether the SPE statistic and the SPE statistic exceed the control limit or not, the accuracy of the model can be verified by using a test set, and the SPE statistic and the T statistic are compared²The higher accuracy one is used as the index for monitoring the wear state of the drilling tool, and in this case, the wear state of the drilling tool is finally selectedSPE statistics.

The present invention is not limited to the above-described embodiments. The foregoing description of the specific embodiments is intended to describe and illustrate the technical solutions of the present invention, and the above specific embodiments are merely illustrative and not restrictive. Those skilled in the art can make many changes and modifications to the invention without departing from the spirit and scope of the invention as defined in the appended claims.