Drilling tool breakage detection method based on isolated forest

Abstract

The drilling tool breakage detection method based on the isolated forest comprises the following steps that A, vibration signals in three directions in the machining process are collected; step B, obtaining a feature matrix Am * n; c, normalizing the feature matrix in the step B to obtain an abnormal sequence sm; step D, obtaining an abnormal smooth sequence ssmoth; e, processing signals after working condition segmentation are obtained; and F, extracting effective value and kurtosis index features and constructing feature vectors through the processing signal data in the step E and the processing signal after working condition segmentation, and realizing health monitoring of the cutter by delimiting a corresponding threshold value. According to the method, the cutter health state information is obtained by analyzing the vibration signals and the like in the cutter drilling process and extracting the effective machining time period of the cutter through the working condition segmentation method, so that the characteristics related to the cutter health state are enhanced, the model monitoring precision is improved, the calculation amount is reduced, and the monitoring efficiency is improved. On the basis, the method has a good application prospect.

Description

Technical Field

[0001] The invention relates to the technical field of detection methods, and in particular to a drilling tool breakage detection method based on isolation forests. Background Art

[0002] Drilling tools (hereinafter referred to as tools) are important tools for machining machine holes, and their health status directly determines the speed and accuracy of machining. Therefore, timely and accurate grasp of the operating status of the tool, especially identifying whether the tool is broken, is of great significance to improving the service life of the tool and ensuring the safe and stable operation of the machine tool. In the prior art, there are also some methods for monitoring the health status of the tool, which mainly include the following. Method 1: Using the signal of the electric spindle inside the machine tool, the mapping relationship between the processing parameters and the power signal is established through support vector regression and adaptive fuzzy neural network, and the calculated standard power is compared with the actual power to judge in real time whether the tool processing is in a healthy state. Method 2: Based on the vibration characteristics of the fixed tool tap, the particle swarm algorithm is used to optimize the key parameters of the support vector machine to monitor and identify the health status of the tool. Method 3: The genetic optimization algorithm, BP neural network and integrated learning are combined to construct a real-time tool monitoring model to monitor the tool accuracy status during the drilling process.

[0003] Although the above tool monitoring methods have played a certain role in the health status of the tool, these methods do not take into account that in the actual processing process, the drilling process includes multiple non-steady fast movements and frequent acceleration and deceleration processes. If these signals are directly used for tool health monitoring, too much information that is weakly related or even irrelevant to the tool health status will be introduced, which will easily lead to the failure of subsequent AI-based recognition and whether the tool is in a healthy working state. The model will fail, and then the tool health monitoring will not meet the actual production needs. Therefore, how to pre-process the monitoring signal and extract the effective processing time period through working condition segmentation is a difficult problem that needs to be overcome urgently. Summary of the invention

[0004] In order to overcome the disadvantages of the prior art that it is unable to effectively monitor and identify whether the tool is in a healthy state, which will have an adverse effect on the actual processing of the workpiece by the tool, the present invention provides a method of obtaining tool health status information by analyzing the vibration signal (effective processing period) of the tool drilling process under the joint action of relevant processes, and extracting the effective processing period of the tool by using the working condition segmentation method. This method is not only conducive to enhancing the features related to the tool health status and improving the model monitoring accuracy, but also conducive to reducing the amount of calculation and improving the monitoring efficiency. Therefore, it provides a drilling tool broken tool detection method based on isolation forest, which provides favorable technical support for the model based on AI recognition and whether the tool is in a healthy working state, and effectively obtaining the tool health status.

[0005] The technical solution adopted by the present invention to solve the technical problem is:

[0006] The method for detecting broken tool of drilling tool based on isolation forest is characterized by comprising the following steps: step 1: using an acceleration sensor installed near the spindle to collect vibration signals in three directions during the processing; step 2: using a sliding window to segment the signal of each channel from the starting time, with a certain overlap rate between the front and rear windows, extracting correlation factor features and statistical indicators such as mean, root mean square value, variance, peak-to-peak value, waveform index, peak index, pulse index, margin index, skewness, etc. from the segmented signal segments to obtain a feature matrix A. m×n , where m is the number of segmented signal segments and n is the number of extracted feature indicators; Step 3: After normalizing the feature matrix, the anomaly score of each signal segment is calculated using the isolation forest outlier anomaly detection method to obtain the anomaly sequence s m ; Step 4: The abnormal sequence fluctuates greatly. In order to ensure the continuity of the original signal processing in step 5, the Savitzky-Golay filtering method is used to smooth the abnormal sequence s to obtain the abnormal smooth sequence ssmooth; Step 5: Set a threshold for the abnormal smooth sequence. If a point in the abnormal smooth sequence exceeds the threshold, the signal segment corresponding to the point is a signal segment that is not concerned. It is mapped back to the original signal and set to zero, thereby successfully extracting the effective processing time period in the drilling process and obtaining the processing signal after the working condition segmentation; Step 6: Using the processing signal obtained in step 5, extract the index features such as effective value and kurtosis, construct a feature vector, and realize detection by setting the corresponding threshold.

[0007] Furthermore, the specific method of step 2 is as follows: the correlation factor is defined as the maximum value of the cross-correlation between two adjacent signal segments, Corr = R xy (τ) max , Among them, x is a signal segment, y is the previous signal segment adjacent to the signal segment, and this feature reflects the correlation between the two signal segments in chronological order. The correlation factor is conducive to extracting the relatively stable segment in the original signal, thereby effectively locating the effective processing time period in the drilling process; in order to improve the robustness and accuracy of the method, other statistical indicators such as mean, root mean square value, variance, peak-to-peak value, waveform index, peak index, pulse index, margin index, skewness, etc. are added to perform multivariate feature comprehensive calculations.

[0008] Furthermore, the specific method of step three is as follows: randomly select a certain number of samples from the feature matrix A as a sample subset and put them into the root node of the isolation tree, randomly specify a feature, randomly generate a cutting point in the current node sample subset, and form a hyperplane to divide the current node data space into two subspaces; put the data in the specified feature that is smaller than the cutting point in the left child node of the current node, and vice versa in the right child node, and continuously randomly select features for segmentation until the child node reaches a limited length, and then an isolation tree is constructed; repeat the above steps to complete the construction of a certain number of isolation trees to form an isolation forest; for each sample in the feature matrix, its abnormality score is defined as: Where h(x) is the path length of the sample on an isolated tree, c(m) is the average path length of the isolated tree in the m sample data set: c(m) = 2H(m-1)-(2(m-1) / m), where H(i) is the harmonic number, H(i) = ln(i)+0.5772156649; calculate the anomaly score of each sample to obtain the anomaly sequence s m .

[0009] Furthermore, the specific method of step 4 is as follows: the Savitzky-Golay filtering method is used to perform weighted filtering on the abnormal sequence s in a window, and the weight is obtained by least square fitting of a given high-order polynomial. The following is to fit the abnormal sequence data within a window length of 2n+1 using a k-order polynomial: s (2n+1)×1 =X (2n+1)×k ·A k×1 +ε (2n+1)×1 , where X is the polynomial variable matrix, A is the polynomial coefficient matrix, and E is the residual matrix. A is obtained by the least squares method: A = (X T ·X) -1 ·X T ·s, then the abnormal smoothing sequence after filtering is: smooth = X·A = X·(X T ·X) -1 ·X T ·s.

[0010] The beneficial effects of the present invention are as follows: under the joint action of relevant data and steps, the present invention utilizes multivariate features and unsupervised outlier anomaly detection methods, fully utilizes the information contained in the signal, can accurately locate the effective processing time period under strong noise and multiple interferences in tool processing, improves the signal-to-noise ratio, and obtains information carriers that are strongly related to the health status of the tool by extracting the effective processing time period in the drilling process, thereby enhancing the robustness of tool status monitoring, improving the accuracy and efficiency of tool status monitoring, and providing favorable technical support for the model based on AI recognition and whether the tool is in a healthy working state, and effectively obtaining the health status of the tool. Based on the above, the present invention has good application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a structural diagram of the experimental platform of the present invention.

[0012] Figure 2 It is a flow chart of the method of the present invention.

[0013] Figure 3 It is a schematic diagram of a broken knife fault of the present invention.

[0014] Figure 4 It is the original vibration signal graph collected by the present invention.

[0015] Figure 5 It is a diagram of the amplitude variation of the Y-direction correlation factor of the present invention.

[0016] Figure 6 It is an abnormally smooth sequence diagram in the Y direction of the present invention.

[0017] Figure 7 It is the time domain waveform and spectrum diagram of the Y direction signal before segmentation of the present invention.

[0018] Figure 8 It is the time domain waveform and spectrum diagram after the Y direction signal is segmented in the present invention.

[0019] Fig. 9 It is a comparison diagram of the signal characteristics (RMS value) results of the present invention. DETAILED DESCRIPTION

[0020] The following is an example of the drilling process of a tap on a drilling machine. The machine tool structure and sensor layout are as follows: Figure 1 The drilling tool breakage detection method based on isolation forest is as follows: Figure 2 As shown, the specific steps are as follows: Step 1, use the acceleration sensor installed near the spindle to collect vibration signals {X, Y, Z} in three directions during normal processing and after tool breakage. The sampling frequency is 6250Hz and the speed is 4000rpm. Among them, the schematic diagram of tool breakage fault is as follows Figure 3 As shown in the figure, the collected vibration signal is as follows Figure 4 shown.

[0021] Step 2: Use a rectangular sliding window of size 400 to segment the vibration signals in three directions from the starting time, with a 50% overlap between the front and rear windows. Extract the correlation factor features and statistical indicators such as mean, root mean square value, variance, peak-to-peak value, waveform index, peak index, pulse index, margin index, skewness, etc. from the segmented signal segments to obtain the feature matrix A. m×n , where m is the number of segmented signal segments and n is the number of extracted feature indicators. The correlation factor is defined as the maximum value of the cross-correlation between two adjacent signal segments:

[0022] Corr=R xy (τ) max (1)

[0023]

[0024] Where Corr is the correlation factor, x is a signal segment, y is the previous signal segment adjacent to the signal segment, and R xy is the cross-correlation of signal segments x and y. Figure 5 The figure shows the change in the amplitude of the correlation factor in the Y direction during a drilling process. The amplitude of the correlation factor in the effective processing period is different from that in other processing periods. This feature reflects the degree of correlation between the two signal segments in chronological order. The correlation factor is conducive to extracting the relatively stable segment in the original signal, thereby effectively locating the effective processing period in the drilling process.

[0025] Step 3: After normalizing the feature matrix, randomly select 256 samples from feature matrix A as the sample subset and put them into the root node of the isolation tree. Randomly specify a feature, randomly generate a cutting point in the current node sample subset, and form a hyperplane to divide the current node data space into two subspaces. Put the data in the specified feature that is smaller than the cutting point in the left child node of the current node, and vice versa in the right child node. Continue to randomly select features for segmentation until the child node reaches the specified length, and then the construction of an isolation tree is completed. Repeat the above steps to complete the construction of 100 isolation trees to form an isolation forest. For each sample in the feature matrix, its anomaly score is defined as:

[0026]

[0027] Where score is the anomaly score, h(x) is the path length of the sample on an isolated tree, and c(m) is the average path length of the isolated tree in the m sample data set: c(m) = 2H(m-1)-(2(m-1) / m) (4)

[0028] Where H(i) is the harmonic number, H(i) = ln(i) + 0.5772156649. Calculate the anomaly score of each sample to get the anomaly sequence s m The abnormal sequence reflects the degree of outlier of each signal segment. The greater the degree of outlier, the greater the possibility that the signal segment is a non-effective processing period.

[0029] Step 4: Use the Savitzky-Golay filtering method to perform weighted filtering on the abnormal sequence s in a window of size 10. The weight is obtained by least square fitting of a given high-order polynomial. Formula (5) is to fit the abnormal sequence data within a window length of 2n+1 using a k-order polynomial: s(2n+1)×1 =X (2n+1)×k ·A k×1 +ε (2n+1)×1 (5)

[0030] Where X is the polynomial variable matrix, A is the polynomial coefficient matrix, and E is the residual matrix. A is obtained by the least squares method:

[0031] A=(X T ·X) -1 ·X T ·s (6)

[0032] Then the abnormal smoothing sequence after filtering is:

[0033] smooth=X·A=X·(X T ·X) -1 ·X T ·s (7)

[0034] Smoothing the abnormal sequence according to the above formula to obtain an abnormal smooth sequence is conducive to the smooth extraction of the effective processing time period.

[0035] Step 5: Figure 6 As shown in , a 3σ threshold is set for the abnormal smooth sequence. If a point in the abnormal smooth sequence exceeds the threshold, the signal segment corresponding to the point is a signal segment that is not of interest and is mapped back to the original signal and set to zero. Figure 7 The figure shows the time domain waveform and spectrum of the Y direction signal of a drilling process before segmentation. Figure 8 As shown in the figure, after the effective processing period is extracted by the working condition segmentation method, only the signal when the tool is processing steadily is found in the time domain, and the information in the frequency domain is clear, which is convenient for the subsequent extraction of fault features. Through step five, the effective processing period in the drilling process is successfully extracted, and the processing signal after working condition segmentation is obtained.

[0036] Step 6: Extract the root mean square value, kurtosis and other features of the signal before and after the working condition segmentation, and use the extreme value of the normal sample fault characteristic index as the threshold for fault diagnosis to perform broken blade detection. When the fault characteristic index exceeds the threshold, it is considered that a broken blade fault has occurred. The test results before and after the working condition segmentation are as follows: Fig. 9 As shown, the discrimination is determined by the ratio of the distance between classes to the distance within classes. It can be seen that after the working condition segmentation, the feature discrimination is improved by 57%, and the fault characteristics are amplified, which is beneficial to the tool state monitoring. In summary, the method proposed in the present invention suppresses noise interference, improves the accuracy and efficiency of monitoring, and enhances the robustness of tool state monitoring.

[0037] Under the joint action of relevant data and steps, the present invention utilizes multivariate features and unsupervised outlier anomaly detection methods, fully utilizes the information contained in the signal, and can accurately locate the effective processing time period under strong noise and multiple interferences in tool processing, thereby improving the signal-to-noise ratio. By extracting the effective processing time period in the drilling process, an information carrier strongly related to the health status of the tool is obtained, the robustness of tool status monitoring is enhanced, and the accuracy and efficiency of tool status monitoring are improved, which provides favorable technical support for the model based on AI recognition and whether the tool is in a healthy working state, and effectively deriving the health status of the tool.

[0038] It will be apparent to those skilled in the art that the present invention is limited to the details of the exemplary embodiments described above and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations that fall within the meaning and scope of the equivalent elements of the claims be included in the present invention.

[0039] In addition, it should be understood that although the present specification is described according to the implementation mode, the implementation mode does not only include an independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in the embodiments can also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A drilling tool breakage detection method based on isolation forest, characterized in that: The method comprises the following steps: step A: using an acceleration sensor installed near a tool spindle to collect vibration signals in three directions during machining; step B: using a sliding window to segment the signals collected in step A from the start time of each channel, with a certain overlap rate between the front and rear windows, extracting correlation factor features and mean, root mean square value, variance, peak-to-peak value, waveform index, peak index, pulse index, margin index, and skewness statistical index from the segmented signal segments to obtain a feature matrix A. m×n , where m is the number of segmented signal segments and n is the number of extracted feature indicators; Step C: After normalizing the feature matrix in step B, the anomaly score of each signal segment is calculated using the isolation forest outlier anomaly detection method to obtain the anomaly sequence s m ; Step D: The abnormal sequence fluctuates greatly. In order to ensure the continuity of the original signal processing in step five, the Savitzky-Golay filtering method is used to smooth the abnormal sequence s to obtain the abnormal smooth sequence ssmooth; Step E: Set a threshold for the abnormal smooth sequence. If a point in the abnormal smooth sequence exceeds the threshold, the signal segment corresponding to the point is a signal segment that is not concerned. It is mapped back to the original signal and set to zero, thereby successfully extracting the effective processing time period in the drilling process and obtaining the processing signal after the working condition segmentation; Step F: Using the processing signal obtained in step five, extract the index features such as effective value and kurtosis, construct a feature vector, and realize detection by setting the corresponding threshold.

2. The drilling tool breakage detection method based on isolation forest according to claim 1 is characterized in that: In step B, the correlation factor is defined as the maximum value of the cross-correlation between two adjacent signal segments, and the formula is as follows: Corr = R xy (τ) max , 3. The drilling tool breakage detection method based on isolation forest according to claim 1 is characterized in that: In step C, the establishment of the isolation forest includes the following steps: a: randomly select a certain number of samples from the feature matrix A in step B as the sample subset and put them into the root node of the isolation tree, randomly specify a feature, randomly generate a cutting point in the current node sample subset, and form a hyperplane to divide the current node data space into two subspaces; b: put the data in the specified feature that is smaller than the cutting point in the left child node of the current node, and vice versa in the right child node; c: continue to randomly select features for segmentation until the child node reaches a limited length, and then complete the construction of an isolation tree; d: repeat the above steps to complete the construction of a certain number of isolation trees to form an isolation forest.

4. The drilling tool breakage detection method based on isolation forest according to claim 1, characterized in that: In step C, for each sample in the feature matrix, its anomaly score is defined as follows: In the formula, h(x) is the path length of the sample on an isolated tree, c(m) is the average path length of the isolated tree in the m sample data set, and is further expressed as the following formula: c(m) = 2H(m-1)-(2(m-1) / m), where H(i) is the harmonic number, H(i) = ln(i)+0.5772156649.

5. The drilling tool breakage detection method based on isolation forest according to claim 1, characterized in that: In step D, the Savitzky-Golay filtering method is specifically used to perform weighted filtering on the abnormal sequence s within a window. The weight is obtained by least square fitting of a given high-order polynomial. The fitting formula for the abnormal sequence data within a window length of 2n+1 using a k-order polynomial is as follows: s (2n+1)×1 =X (2n+1)×k ·A k×1 +ε (2n+1)×1 , where X is the polynomial variable matrix, A is the polynomial coefficient matrix, and E is the residual matrix. The formula for A is obtained by the least squares method as follows: A = (X T ·X) -1 ·X T ·s, then the abnormal smoothing sequence after filtering is as follows: smooth = X·A = X·(X T ·X) -1 ·X T ·s.

Images