Early fault detection method for automatic tool changing system of CNC machine tools based on adaptive frequency band deconvolution

Abstract

The present invention discloses an early fault detection method for an automatic tool changing system of a CNC machine tool based on adaptive frequency band deconvolution, which belongs to the technical field of mechanical fault diagnosis of CNC machine tools. The method first constructs a basic matrix and a basic vector using the vibration monitoring signal of the automatic tool changing system, and obtains a weight vector through sparse filtering; then, the basic matrix and the weight vector are subjected to signal adaptive frequency band deconvolution; after deconvolution, the obtained results are subjected to frequency sorting, separation and reconstruction to eliminate interference components in the frequency; finally, the reconstructed signal is subjected to envelope spectrum analysis to extract fault characteristics. The present invention can capture and enhance early fault characteristics from the original vibration monitoring data, has strong robustness to interference, and can quickly and effectively describe the early fault behavior of the servo tool holder power head in the automatic tool changing system; in summary, the present invention has high efficiency and practicality, and is suitable for monitoring and diagnosing equipment faults in actual engineering.

Description

Technical Field

[0001] The invention relates to the technical field of mechanical fault diagnosis of numerical control machine tools, and in particular to an early fault detection method of an automatic tool changing system of a numerical control machine tool based on adaptive frequency band deconvolution. Background Art

[0002] The automatic tool changing system is a key component of CNC machine tools, which has a direct impact on the machining accuracy and production efficiency of CNC machine tools. However, the continuous high-intensity operation of CNC machine tools may cause faults such as bearing wear and gear meshing problems. If these faults are not discovered in time, they will cause machine tool shutdown, increase costs, and may affect the quality of processed products. Traditional fault detection methods mainly rely on regular manual maintenance and inspection, but the efficiency is low, and it is difficult to capture fault signals in the early stage of the fault, which may cause the fault to deteriorate and increase the difficulty and cost of repair. With the advancement of technology and algorithms, fault detection based on vibration signal analysis has become more and more popular. This method can monitor the equipment status and diagnose faults in real time by collecting and analyzing the vibration signals in the operation of the automatic tool changing system. However, since the vibration signal contains multiple frequencies and has large noise interference, it is difficult for traditional methods to accurately extract fault features.

[0003] Deconvolution technology is a signal processing technology that can separate specific target signals from complex vibration signals and reduce noise interference through deconvolution filters. If it is applied to the early fault detection of automatic tool change systems for CNC machine tools, it can more accurately extract fault-related features and provide support for the rapid discovery and subsequent fault diagnosis of automatic tool change systems for machine tools. Summary of the invention

[0004] The purpose of the present invention is to solve the problems faced in the technical field of mechanical fault diagnosis of CNC machine tools mentioned above, and to provide an early fault detection method for automatic tool changing systems based on adaptive frequency band deconvolution, which is mainly aimed at early fault detection of automatic tool changing systems of CNC machine tools to improve the real-time and accuracy of detection, thereby reducing costs and ensuring production safety.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An early fault detection method for an automatic tool changing system based on adaptive frequency band deconvolution comprises the following steps:

[0007] 1) Generate the basic matrix and basic vector using the collected vibration monitoring signal of the automatic tool changing system;

[0008] 2) Perform sparse filtering on the basis vector to obtain the weight vector;

[0009] 3) Substitute the basis matrix and weight vector into the deconvolution formula:

[0010] ;

[0011] represents the fundamental matrix, for The transpose of represents the basis vector, weight vector;

[0012] Performing non-iterative adaptive frequency band deconvolution calculations;

[0013] 4) After deconvolution, the signal is spectrally sorted, separated and reconstructed to filter out interference components with relatively low frequency content;

[0014] 5) Perform envelope spectrum analysis on the reconstructed signal to extract fault features.

[0015] The expressions for generating the basic matrix and basic vector described in step 1) are:

[0016] ;

[0017] ;

[0018] ;

[0019] Among them, in formula (2) represents the original vibration monitoring signal, in formula (3) represents the basic matrix, in formula (4) represents the basis vector, Corresponding to the filter length during deconvolution calculation;

[0020] The sparse filtering process described in step 2 is:

[0021] ;

[0022] ;

[0023] Formula (5) means mapping the data to the interval [-1, 1] and then using the exponential Perform power operation on it to achieve the purpose of sparsification;

[0024] Formula (6) represents the threshold filtering of the result of formula (5), where is the threshold value; the final result is:

[0025] ;

[0026] ;

[0027] The result obtained by formula (6) is then subjected to one-dimensional digital filtering. The weight vector is generated by sparseness, threshold filtering and one-dimensional digital filtering, the purpose of which is to remove the influence of low-amplitude components in the signal, thereby highlighting the characteristic frequency components; formula (7) represents the rational transfer function in the Z transform domain during one-dimensional digital filtering; the final weight vector is .

[0028] The deconvolution formula (1) described in step 3) is derived as follows:

[0029] Assume that the filter to be solved is The input of the system is , the output is , then formula (8) defines the following convolution operation:

[0030] ;

[0031] Maximizing Multi D-Norm is used as the solution objective. Multi D-Norm is expressed as formula (9)

[0032] ;

[0033] The optimization problem is transformed into a derivation problem as shown in formula (10).

[0034] ;

[0035] After simplification, we can get formula (11):

[0036] ;

[0037] because is a scalar and can be regarded as a constant term coefficient, so formula (11) can be transformed into formula (1)

[0038] In formula (1), the vector With vector The proportional coefficient only affects the overall size of the vector and does not affect the magnitude trend of the vector. Therefore, the influence of this proportional coefficient can be ignored; therefore, the deconvolution filter to be solved is:

[0039] .

[0040] Step 4) The separated expression:

[0041] ;

[0042] ;

[0043] Formula (12) represents the Hilbert transform, and formula (13) represents the discrete Fourier transform, where yes Discrete representation of .

[0044] The reconstructed expression:

[0045] ;

[0046] ;

[0047] Formula (14) represents threshold filtering, Z represents separation reconstruction threshold; Formula (15) represents inverse discrete Fourier transform; the final result is the output of the present invention, and the envelope spectrum analysis is performed on the result to complete fault detection.

[0048] The present invention provides an early fault detection method for an automatic tool change system of a numerical control machine tool based on adaptive frequency band deconvolution, aiming to solve the technical difficulties in the field of mechanical fault detection. The method first constructs a basic matrix and a basic vector using the vibration monitoring signal of the automatic tool change system, and obtains a weight vector through sparse filtering. Then, a non-iterative deconvolution calculation method is used to perform signal adaptive frequency band deconvolution in combination with the weight vector and the basic matrix. After deconvolution, the obtained results are spectrally sorted, separated and reconstructed to eliminate interference components with low frequency content. Finally, the reconstructed signal is subjected to envelope spectrum analysis to extract fault features. The present invention can capture and enhance early fault features from the original vibration monitoring data, and has strong robustness to the presence of interference. The method can quickly and effectively describe the early fault behavior of the servo tool holder power head in the automatic tool change system. The implementation process of the present invention has high efficiency and practicality, is suitable for monitoring and diagnosing equipment faults in engineering practice, and has broad application and promotion prospects.

[0049] Compared with the prior art, the method for early fault detection of an automatic tool changing system based on adaptive frequency band deconvolution provided by the present invention has the following beneficial effects and advantages:

[0050] 1. The present invention adopts a non-iterative adaptive frequency band deconvolution calculation method, which can significantly improve the calculation efficiency and reduce the calculation time compared with the traditional iterative method, and is suitable for real-time or near real-time fault detection needs.

[0051] 2. The present invention performs sparse filtering on the basic vectors and uses the obtained weight vectors to participate in deconvolution calculations to effectively reduce noise and redundant information in the data, improve the purity of the signal and the significance of the fault characteristics, and make fault detection more accurate and reliable.

[0052] 3. The present invention performs spectrum sorting, separation and reconstruction on the signal after deconvolution, which can specifically screen out interference components with low frequency content, further improve the extraction effect of fault features, and contribute to the accurate identification of early faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of the deconvolution solution calculation purpose of an early fault detection method of an automatic tool changing system of a CNC machine tool based on adaptive frequency band deconvolution of the present invention;

[0054] Figure 2 It is a flow chart of the steps of an early fault detection method of an automatic tool changing system of a CNC machine tool based on adaptive frequency band deconvolution of the present invention;

[0055] Figure 3 This is a physical picture of a servo tool holder power head test bench used for verification in the automatic tool changing system of a CNC machine tool.

[0056] Figure 4 This is the direct envelope spectrum analysis of the signals of two comparison cases, (a) is case 1, (b) is case 2;

[0057] Figure 5 The results of two comparison case signals processed by the fast Kurtogram method are shown in Figure 1. (a) is the result of case 1, and (b) is the result of case 2.

[0058] Figure 6 The following are result diagrams of two comparative case signals after being processed by the early fault detection method proposed in the present invention, (a) is the result of case 1, and (b) is the result of case 2. DETAILED DESCRIPTION

[0059] In order to make the technical scheme and purpose of the present invention more clearly understood, the present invention is described in more detail below in conjunction with the drawings and specific embodiments of the present invention, and the technical effects of the present invention are demonstrated through the specific implementation cases listed below. It should be noted that the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments.

[0060] See attached Figure 2 , an early fault detection method for an automatic tool changing system based on adaptive frequency band deconvolution, comprising the following steps:

[0061] 1) Generate the basic matrix and basic vector using the collected vibration monitoring signal of the automatic tool changing system;

[0062] 2) Perform sparse filtering on the basis vector to obtain the weight vector;

[0063] 3) Substitute the basic matrix and weight vector into the deconvolution formula proposed in the present invention:

[0064]

[0065] Performing non-iterative adaptive frequency band deconvolution calculations;

[0066] 4) After deconvolution, the signal is spectrally sorted, separated and reconstructed to filter out interference components with relatively low frequency content;

[0067] 5) Perform envelope spectrum analysis on the reconstructed signal to extract fault features.

[0068] The specific expression of generating the basic matrix and basic vector by using the vibration monitoring signal of the automatic tool changing system is:

[0069] ;

[0070] ;

[0071] ;

[0072] Among them, in formula (2) represents the original vibration monitoring signal; in formula (3) represents the basic matrix; in formula (4) represents the basis vector; Corresponds to the filter length used in deconvolution calculations.

[0073] The weight vector is obtained by performing sparse filtering on the basic vector The process and specific expressions are shown in the following formulas (5) to (7):

[0074] ;

[0075] ;

[0076] ;

[0077] Among them, formula (5) means mapping the data to the interval [-1, 1] and then using the index Perform power operation on it to achieve the purpose of sparsification;

[0078] Formula (6) represents the threshold filtering of the result of formula (5), and the final result is ;

[0079] Formula (7) represents the rational transfer function in the Z transform domain during one-dimensional digital filtering. The one-dimensional digital filtering refers to performing a one-dimensional digital filtering on the result obtained by formula (6). Through sparseness, threshold filtering and one-dimensional digital filtering, the influence of low-amplitude components in the signal is removed, thereby highlighting the characteristic frequency components, and the final weight vector is generated as follows: .

[0080] The deconvolution formula (1) proposed in the present invention is deduced as follows:

[0081] Assume that the filter to be solved is The input of the system is , the output is Then formula (8) defines the following convolution operation:

[0082]

[0083] Maximizing Multi D-Norm is used as the solution objective. Multi D-Norm is expressed as formula (9)

[0084]

[0085] The optimization problem is transformed into a derivation problem as shown in formula (10) and formula (10).

[0086]

[0087] After simplification, we can get formula (10):

[0088]

[0089] because is a scalar and can be regarded as a constant term coefficient, so formula (10) can be transformed into formula (11).

[0090]

[0091] Among them, the vector in formula (1) With the vector in formula (11) The proportional coefficient only affects the overall size of the vector and does not affect the amplitude trend of the vector. Therefore, the influence of this proportional coefficient can be ignored. Therefore, the deconvolution filter to be solved is:

[0092] ;

[0093] Furthermore, the above basic matrix can be and the weight vector Substitute the proposed deconvolution formula (1) to perform non-iterative adaptive frequency band deconvolution calculation. The deconvolution calculation is solved for the purpose of the solution. See the attached Figure 1 , its core goal is to solve a filter that can effectively deconvolve and approximately restore the fault pulse component in the initial signal.

[0094] The spectrum of the deconvolved signal is sorted, separated and reconstructed. The separation process and specific expressions are shown in the following formulas (12) to (13):

[0095] ;

[0096] ;

[0097] Wherein, formula (12) represents the Hilbert transform; formula (13) represents the discrete Fourier transform, where yes Discrete representation of ;

[0098] The specific expressions of the reconstruction are shown in formula (14) and formula (15):

[0099] ;

[0100] ;

[0101] Formula (14) represents threshold filtering, represents the separation reconstruction threshold; formula (15) represents the inverse discrete Fourier transform; the final result is used as the output of the present invention, and then the envelope spectrum analysis is performed on the result to complete the fault detection.

[0102] In order to verify the performance of the present invention, the early fault data of the servo tool holder power head in the automatic tool changing system of a CNC machine tool is used to verify the practicality and performance of the early fault detection method of the automatic tool changing system based on adaptive frequency band deconvolution of the present invention through experiments. The specific process and results are as follows:

[0103] See attached Figure 3 , which is the servo tool turret power head test bench in the CNC machine tool automatic tool change system used in the experiment; during the data collection process, the power motor speed is 1400 RPM, the transmission ratio is 1:1, the number of teeth of the active and driven bevel gears in the turret is 25; the tool cutting speed is 1400RPM, the torque is 10NM; the sampling frequency is 10000Hz, and the intercepted data length is 10000 data points. In the experiment, two sections of early fault data of the servo tool turret power head were selected as two comparison cases, and the fault type was slight wear of the transmission gear.

[0104] See attached Figure 4, the results obtained by directly performing envelope analysis on the data of the two comparison cases. Since the cutting speed of the tool is 1400RPM, the basic rotation frequency is 23.33Hz; since the number of teeth of the active and driven bevel gears inside the turret is 25, the basic meshing frequency is 583.33Hz. See Appendix Figure 4 (a), its basic rotation frequency and meshing frequency can be found; see Appendix Figure 4 (b), only the meshing frequency can be found; obviously, Figure 4 There is no obvious frequency feature available for diagnostic assessment of the health status of the system.

[0105] In order to highlight the practicality of the present invention, the classic Fast Kurtogram method and the method proposed in the present invention are used to compare the effects:

[0106] 1) Participate in the Figure 5 , shows the process and results of early fault detection using fast Kurtogram: Due to the high meshing frequency of the gears, the decomposition parameter “level” is set to 2 to prevent the frequency band from becoming narrower after fast Kurtogram decomposition, thus affecting subsequent analysis.

[0107] In the attached Figure 5 In (a), there is only the fundamental rotation frequency of the motor, and there is no peak of the gear meshing frequency in the red ellipse area; similarly, in the adjacent Figure 5 In (b), only the fundamental rotation frequency of the motor and its multiples exist, and the gear meshing frequency peak does not appear in the red ellipse area. At this point, it can be considered that the fast Kurtogram method is not effective.

[0108] 2) Participate in the Figure 6 The results of early fault detection using the method proposed by the present invention are shown in the attached figure: Figure 6 (a) and attached Figure 6 (b) are all concise and clear, without interference components. In addition to the basic rotation frequency and meshing frequency, there are alternating peaks with a sideband length of 23.33Hz near the basic rotation frequency, meshing frequency and its multiple frequency. This value corresponds to the basic rotation frequency of the gear.

[0109] Therefore, it can be considered that the gear in the turret of the servo turret power head system is damaged. The results show that compared with the fast Kurtogram, the present invention has significant advantages in early fault detection.

[0110] In summary, the present invention aims at the problem that it is difficult to extract early fault features of automatic tool changing systems of CNC machine tools in engineering practice, and proposes an early fault detection method for automatic tool changing systems of CNC machine tools based on adaptive deconvolution, which is used to extract early fault features of monitored components from vibration monitoring signals, thereby improving the accuracy of fault diagnosis.

[0111] The method proposed in the present invention firstly constructs a basic matrix and a basic vector using the vibration monitoring signal of the automatic tool changing system; and obtains a weight vector through sparse filtering processing; then, a non-iterative adaptive frequency band deconvolution calculation method is used to perform signal deconvolution; after deconvolution, the obtained results are spectrally sorted, separated and reconstructed to eliminate interference components with low frequency content; finally, the reconstructed signal is subjected to envelope spectrum analysis to extract fault characteristics.

[0112] It can be seen from the experimental results that the present invention can effectively diagnose the early fault behavior of the automatic tool changing system; the implementation process of the present invention has high efficiency and practicality, is suitable for monitoring and diagnosing equipment failures in engineering practice, and has broad application and promotion prospects.

Claims

1. An early fault detection method for an automatic tool changing system of a CNC machine tool based on adaptive frequency band deconvolution comprises the following steps: 1) Generate the basic matrix and basic vector using the collected vibration monitoring signal of the automatic tool changing system; 2) Perform sparse filtering on the basis vector to obtain the weight vector; 3) Substitute the basis matrix and weight vector into the deconvolution formula: ； represents the fundamental matrix, for The transpose of represents the basis vector, weight vector; Performing non-iterative adaptive frequency band deconvolution calculations; 4) After deconvolution, the signal is spectrally sorted, separated and reconstructed to filter out interference components with relatively low frequency content; 5) Perform envelope spectrum analysis on the reconstructed signal to extract fault features; The expressions for generating the basic matrix and basic vector described in step 1) are: ； ； ； Among them, in formula (2) represents the original vibration monitoring signal, in formula (3) represents the basic matrix, in formula (4) represents the basis vector, Corresponding to the filter length during deconvolution calculation; The sparse filtering process described in step 2 is: ； ； Formula (5) means mapping the data to the interval [-1, 1] and then using the exponential Perform power operation on it to achieve the purpose of sparsification; Formula (6) represents the threshold filtering of the result of formula (5), where is the threshold value; the final result is: ； ； The result obtained by formula (6) is then subjected to one-dimensional digital filtering. The weight vector is generated by sparseness, threshold filtering and one-dimensional digital filtering. The purpose is to remove the influence of low-amplitude components in the signal, thereby highlighting the characteristic frequency components. Formula (7) represents the rational transfer function in the Z transform domain during one-dimensional digital filtering. The final weight vector is .

2. The method for early fault detection of automatic tool changing system of CNC machine tools based on adaptive frequency band deconvolution according to claim 1 is characterized in that: The deconvolution formula (1) described in step 3) is derived as follows: Assume that the filter to be solved is The input of the system is , the output is , then formula (8) defines the following convolution operation: ； Maximizing Multi D-Norm is used as the solution objective; Multi D-Norm is expressed as formula (9): ； The optimization problem is transformed into a derivation problem as shown in formula (10); ； After simplification, we can get formula (11): ； because is a scalar and can be regarded as a constant term coefficient, so formula (11) can be transformed into formula (1); In formula (1), the vector With vector Proportional, and this proportionality coefficient only affects the overall size of the vector, and does not affect the amplitude trend of the vector; therefore, the impact of this proportionality coefficient can be ignored; so the deconvolution filter to be solved is: 。 3. The method for early fault detection of automatic tool changing system of CNC machine tool based on adaptive frequency band deconvolution according to claim 1 or 2, characterized in that: Step 4) The separated expression: ； ； Formula (12) represents the Hilbert transform, and formula (13) represents the discrete Fourier transform, where yes Discrete representation of ; The reconstructed expression: ； ； Formula (14) represents threshold filtering, Z represents separation reconstruction threshold; Formula (15) represents inverse discrete Fourier transform; the final result is the output of the present invention, and the envelope spectrum analysis is performed on the result to complete fault detection.

Images