A fast milling chatter detection method based on correlation coefficient

Abstract

The application discloses a kind of fast milling chatter detection methods based on correlation coefficient, comprising the following steps: determining the installation position of vibration sensor and rotational speed sensor based on workpiece processing, determine the data amount collected per revolution of main shaft, the data length of time window, specify chatter threshold;Periodic synchronization multi-point sampling data is obtained;Standardized processing sampling data;Extension is multidimensional data;The correlation coefficient between adjacent data in multidimensional data is calculated to obtain coefficient data, the average of coefficient data is calculated;Determine whether chatter occurs based on average and chatter threshold.This method uses the average correlation coefficient of two adjacent rows of data after extension and standardization as fault indicator, reduces the influence of signal noise on chatter indicator, and further reduces the misdiagnosis rate and missed diagnosis rate of chatter.

Description

Technical Field

[0001] This invention relates to the field of condition detection technology, and in particular to a rapid milling chatter detection method based on correlation coefficients. Background Technology

[0002] Milling is an advanced manufacturing technology characterized by high efficiency, high precision, and high surface quality. It has been widely applied in mold making, aerospace, and automotive manufacturing. During milling, inappropriate combinations of machining parameters often cause chatter. Chatter is a common abnormal vibration phenomenon in milling. Chatter not only affects workpiece quality and machining efficiency but also accelerates tool wear and can even lead to machine tool damage. Therefore, to improve milling quality and efficiency, chatter detection is necessary during the milling process to detect chatter early and prevent its harmful effects.

[0003] Currently, researchers and engineers have proposed various chatter detection techniques. Patent application number 201910656987.5 discloses a chatter detection method based on signal processing and artificial intelligence. This method first uses empirical mode decomposition (EMD) to decompose the chatter signal into a set of intrinsic mode function (IMF) signals. Then, it selects several IMF signals containing more chatter information for analysis, extracting chatter features and constructing a multidimensional chatter feature observation space. Next, it uses a manifold learning algorithm to reduce the dimensionality of the multidimensional chatter feature observation space, then fuses information from multiple chatter features, and finally uses a support vector machine (SVM) to identify the chatter. While the above chatter detection methods can reliably detect chatter, the computational cost is relatively high, and it may not be suitable for real-time chatter detection in high-speed milling. Patent application number 202210619937.1 discloses a fast and reliable method for detecting milling chatter. This method uses vibration sensors to acquire multiple sets of periodically synchronized single-point sampling data per spindle revolution, fusing chatter information collected by multiple vibration sensors. This shortens data acquisition time, increases the amount of data collected, and improves the real-time performance of chatter detection. Although this method can achieve real-time chatter detection, its chatter performance is susceptible to noise, resulting in relatively low reliability. Furthermore, the use of multiple vibration sensors not only increases the cost of chatter detection but also increases the false alarm rate due to sensor failure. Summary of the Invention

[0004] Purpose of the invention: In order to overcome the shortcomings of the prior art, this invention requests the disclosure of a rapid milling chatter detection method based on correlation coefficient. This method uses the average correlation coefficient of two adjacent rows of data after extended standardization as the fault index, which reduces the influence of signal noise on chatter index, thereby reducing the false alarm rate and false negative rate of chatter detection.

[0005] Technical solution: The rapid milling chatter detection method based on correlation coefficient disclosed in this invention includes the following steps:

[0006] S1. Determine the installation positions of the vibration sensor and speed sensor based on the workpiece being processed, determine the amount of data collected per spindle revolution, the data length of the time window, and specify the chatter threshold;

[0007] S2. Acquire periodic synchronous multi-point sampling data;

[0008] S3, standardized processing to obtain periodic synchronous multi-point sampling data;

[0009] S4. Expand the standardized periodic synchronous multi-point sampling data into multi-dimensional data;

[0010] S5. Calculate the correlation coefficient between adjacent data in the multidimensional data to obtain the coefficient data, and calculate the average value of the coefficient data;

[0011] S6. Determine whether flutter occurs based on the above average value and flutter threshold.

[0012] Furthermore, the vibration sensor in S1 is an acceleration sensor, a velocity sensor, or a displacement sensor.

[0013] Furthermore, in S1, the workpiece being processed is either a non-thin-walled workpiece or a thin-walled workpiece;

[0014] For milling of non-thin-walled workpieces, vibration and speed sensors are mounted on the high-speed spindle to acquire vibration and speed data; for milling of thin-walled workpieces, vibration sensors are mounted on the thin-walled workpiece or the high-speed spindle to acquire vibration data; speed sensors are mounted on the high-speed spindle to acquire speed data, with the spindle revolutions in meters (m) within a time window.

[0015] Furthermore, the periodically synchronized multi-point sampling data X in S2 is represented as:

[0016] X=[x(1) x(2) x(3) … x(i) … x(nm)];

[0017] In the formula: nm represents the data length of the time window.

[0018] Furthermore, in S3, the acquired periodic synchronous sampling data X is standardized using the Z-score standardization method. The processed data is represented as follows:

[0019]

[0020] In the formula, σ represents the mean of data X, and σ represents the standard deviation of data X.

[0021] Furthermore, S4 specifically involves: based on the amount of data n collected per revolution, expanding the standardized periodic synchronous multi-point sampling data X' into n-dimensional data Y, which is represented as:

[0022]

[0023] make The expanded n-dimensional data Y is represented as:

[0024]

[0025] Furthermore, S5 specifically involves: calculating the neighboring data Y in the n-dimensional data Y. i And data Y i+1 The correlation coefficient r(i) is used to obtain the data C, and the average value M of the data C is calculated.

[0026] Data C is represented as:

[0027] C=[r(1) r(2) r(3) … r(i) … r(n-1)];

[0028] Among them, data Y i And data Y i+1 The correlation coefficient r(i) is expressed as:

[0029]

[0030] In the formula, cov(Y) i ,Y i+1 (Y) is the data i And data Y i+1 covariance, σ Yi and σ Yi+1 Representing data Y respectively i And data Y i+1 Standard deviation;

[0031] The average value M of data C is expressed as:

[0032]

[0033] Furthermore, S6 specifically states: if the average value is greater than or equal to the flutter threshold, then flutter is determined to have occurred; otherwise, it is considered that flutter has not occurred, and the process returns to S2 to continue flutter detection.

[0034] Beneficial effects: Compared with the prior art, the advantages of the present invention are:

[0035] 1. The method of the present invention has low computational cost, and can use a control unit with slower computation speed, thereby reducing the cost of the control unit; and the detection time delay is short, which can quickly detect flutter.

[0036] 2. The flutter index used in the method of the present invention is not affected by noise in the signal, which reduces the false alarm rate and false negative rate of flutter and improves the reliability of flutter detection.

[0037] 3. The method of the present invention can be applied to variable speed milling systems. Attached Figure Description

[0038] Figure 1 This is a flowchart of the flutter detection method of the present invention;

[0039] Figure 2 Displacement data without flutter and periodically synchronized multi-point sampled displacement data;

[0040] Figure 3 This includes displacement data during flutter and periodically synchronized multi-point sampling displacement data.

[0041] Figure 4 The standardized, periodically synchronized multi-point sampling displacement data under flutter-free conditions;

[0042] Figure 5 This is standardized, periodically synchronized, multi-point sampled displacement data for situations with flutter.

[0043] Figure 6 This is a graph showing the changes in mean standard deviation and mean correlation coefficient with noise. Detailed Implementation

[0044] like Figure 1 The fast milling chatter detection method based on correlation coefficient shown includes the following steps:

[0045] Step 1: Based on the workpiece being machined, determine the amount of data n collected synchronously per spindle revolution, the data length nm of the time window, specify the chatter threshold T, and select the vibration sensor and speed sensor and their installation positions.

[0046] The amount of data synchronously collected per spindle revolution is determined to be n=100, the data length of each time window is nm=1000, the chatter detection threshold T is specified as 0.01, and the selected vibration sensor is a displacement sensor, but it can also be an acceleration sensor or a velocity sensor.

[0047] The installation position of the vibration sensor is selected based on the type of workpiece being machined. When the workpiece is not thin-walled, the vibration sensor and speed sensor are installed on the milling machine spindle to measure spindle vibration and speed. For milling thin-walled workpieces, a displacement sensor can be installed on the thin-walled workpiece or the high-speed spindle to obtain vibration data. Additionally, a speed sensor is installed on the spindle to obtain speed data; m can also be considered as the number of spindle revolutions within a time window. In this embodiment, the workpiece being machined is a thin-walled workpiece; therefore, the vibration sensor can be installed on the thin-walled workpiece or the spindle. In this embodiment, the displacement sensor is installed on the thin-walled workpiece to measure the displacement of the thin-walled workpiece perpendicular to the machining direction.

[0048] Step 2: Acquire periodic synchronous multi-point sampling data.

[0049] Based on the vibration sensor and speed sensor described in step 1, periodic synchronous multi-point sampling data is acquired, which can be represented as:

[0050] X=[x(1) x(2) x(3) … x(100) … x(1000)].

[0051] In this embodiment, a displacement sensor mounted on the thin-walled workpiece is used to measure the vibration displacement signal of the thin-walled workpiece perpendicular to the machining direction. A speed sensor mounted on the spindle is used to collect spindle speed information. Displacement data without chatter and periodically synchronized multi-point displacement data are shown below. Figure 2 As shown. Displacement data during flutter and periodically synchronized multi-point displacement data are as follows: Figure 3 As shown.

[0052] Step 3: Standardize the acquired periodic synchronous multi-point sampling data X.

[0053] The acquired periodically synchronized multi-point sampling data X is standardized using the Z-score standardization method. The standardized data can be represented as:

[0054]

[0055] In the formula, σ represents the mean of data X, and σ represents the standard deviation of data X.

[0056] In this embodiment, the standardized periodic synchronous multi-point sampling signal without flutter is as follows: Figure 4 As shown. The standardized, periodically synchronized multi-point sampled signal during flutter is as follows. Figure 5 As shown, through standardization, the periodically synchronized multi-point sampling data is standardized to a uniform scale. This facilitates subsequent flutter detection.

[0057] Step 4: Expand the periodic synchronous multi-point sampling data after the extended standardization process into multidimensional data Y.

[0058] Based on the data volume of 100 collected per revolution, the standardized periodic synchronous multi-point sampling data is expanded into 100-dimensional data. The expanded 100-dimensional data Y can be represented as:

[0059]

[0060] For ease of expression, let The expanded 100-dimensional data Y is represented as:

[0061]

[0062] The purpose of expanding the periodically synchronized multi-point sampling data is to facilitate subsequent correlation coefficient calculations. The expanded and normalized periodically synchronized multi-point sampling data can effectively display flutter information. Compared to periodically synchronized single-point sampling data, periodically synchronized multi-point sampling data has a larger data volume and contains more flutter information.

[0063] In this embodiment, since 100 data points are collected per revolution, the acquired periodic synchronous multi-point sampling data can be expanded into a set of 100 rows and 10 columns of data.

[0064] Step 5: Calculate the neighboring data Y in the 100-dimensional data Y. i And data Y i+1 The correlation coefficient r(i) is used to obtain the data C, and the average value M of the data C is calculated.

[0065] In step 5, the adjacent data Y in the 100-dimensional data Y are calculated. i And data Y i+1 The correlation coefficient data C can be expressed as:

[0066] C=[r(1) r(2) r(3) … r(i) … r(99)],

[0067] Among them, data Y i And data Y i+1 The correlation coefficient r(i) can be expressed as:

[0068]

[0069] In the formula, cov(Y) i ,Y i+1 (Y) is the data i And data Y i+1 covariance, σ Yi and σ Yi+1 Representing data Y respectively i And data Yi+1 The standard deviation.

[0070] The average value M of data C can be expressed as:

[0071]

[0072] Step 6: Determine if flutter has occurred.

[0073] Compare the calculated average value M with the predetermined threshold T. If M is greater than or equal to the predetermined threshold T, it can be determined that flutter has occurred. Otherwise, it is considered that flutter has not occurred, and the process returns to step 2 to continue flutter detection.

[0074] It can be concluded that for periodically synchronized multi-point sampling data without flutter, the calculated average value M is equal to 0.004, which is less than the predetermined threshold T = 0.1, and the flutter detection program returns to the second step to continue flutter detection; for periodically synchronized multi-point sampling data with flutter, the calculated average value M is equal to 0.25, which is greater than the predetermined threshold T = 0.1, and it can be determined that flutter has occurred.

[0075] It should be noted that the average correlation coefficient used in this invention patent is not affected by disturbance noise, while the standard deviation used in the invention patent with application number 202210619937.1 is greatly affected by disturbance noise. Figure 6 The effects of disturbance noise on the average correlation coefficient and average standard deviation are shown. As can be seen from the figure, the average correlation coefficient exhibits better robustness to disturbance noise than the average standard deviation.

Claims

1. A rapid milling chatter detection method based on correlation coefficient, characterized in that, Includes the following steps: S1. Determine the installation positions of the vibration sensor and speed sensor based on the workpiece being processed, determine the amount of data collected per spindle revolution, the data length of the time window, and specify the chatter threshold; S2. Based on the vibration sensor and speed sensor described in S1, acquire periodic synchronous multi-point sampling data; S3, standardized processing to obtain periodic synchronous multi-point sampling data; S4. Expand the standardized periodic synchronous multi-point sampling data into multi-dimensional data; S5. Calculate the correlation coefficient between adjacent data in the multidimensional data to obtain the coefficient data, and calculate the average value of the coefficient data; S6. Determine whether flutter occurs based on the above average value and flutter threshold; S2 periodic synchronous multi-point sampling data X Represented as: ； In the formula: nm represents the data length of the time window; In S3, the Z-score normalization method is used to normalize the acquired periodic synchronous sampling data. X After standardization, the processed data is represented as follows: ； In the formula, Represents data X The average value, Represents data X Standard deviation; S4 specifically involves: based on the amount of data n collected per revolution, synchronizing the standardized periodic multi-point sampling data. Expanded to n-dimensional data Y, the expanded n-dimensional data Y is represented as: ； make The expanded n-dimensional data Y is represented as: ； S5 specifically involves calculating the adjacent data in the n-dimensional data Y. and data The correlation coefficient r(i) is used to obtain the data C, and the average value M of the data C is calculated. Data C is represented as: ； Among them, data and data The correlation coefficient r(i) is expressed as: ， In the formula, For data and data covariance, and Representing data respectively and data Standard deviation; The average value M of data C is expressed as: 。 2. The rapid milling chatter detection method based on correlation coefficient according to claim 1, characterized in that: The vibration sensor in S1 is an acceleration sensor, velocity sensor, or displacement sensor.

3. The rapid milling chatter detection method based on correlation coefficient according to claim 1, characterized in that: In S1, the workpiece being processed can be either a non-thin-walled workpiece or a thin-walled workpiece; For milling of non-thin-walled workpieces, vibration and speed sensors are mounted on the high-speed spindle to acquire vibration and speed data; for milling of thin-walled workpieces, vibration sensors are mounted on the thin-walled workpiece or the high-speed spindle to acquire vibration data; speed sensors are mounted on the high-speed spindle to acquire speed data, with the spindle revolutions in meters (m) within a time window.

4. The rapid milling chatter detection method based on correlation coefficient according to claim 1, characterized in that, S6 specifically states: if the average value is greater than or equal to the flutter threshold, then flutter is determined to have occurred; otherwise, it is considered that flutter has not occurred, and the process returns to S2 to continue flutter detection.

Images