Discharge pulse monitoring method and system for high-frequency power supply of wire cutting equipment

Abstract

The present application belongs to the technical field of monitoring, and particularly relates to a discharge pulse monitoring method and system for a high-frequency power supply of a linear cutting device. The method comprises the following steps: mixing a to-be-processed signal into a guide noise generated based on a historical frequency energy distribution, and obtaining a component through variational mode decomposition; calculating the kurtosis value of a Teager-Kaiser energy operator sequence of each component, screening a target signal, and determining a normalization coefficient based on the kurtosis value; extracting a global envelope of a Hilbert transform of the target signal and a local envelope of a local mean decomposition, and generating a composite envelope by using the normalization coefficient to weight a root mean square; quantizing the amplitude value sequence of the composite envelope in gray scale, constructing a gray level co-occurrence matrix, extracting a feature vector, and inputting the feature vector into a classification model to output a discharge state. That is, the scheme of the present application can adaptively remove noise and highlight pulse distortion details, and significantly improve the discharge state recognition accuracy under variable working conditions.

Description

Technical Field

[0001] This invention belongs to the field of monitoring technology, and in particular relates to a method and system for monitoring the discharge pulses of a high-frequency power supply in a wire cutting device. Background Technology

[0002] In wire EDM, a high-frequency pulse power supply releases electrical energy through the tiny gap between the electrode wire and the workpiece, using the instantaneous high temperature generated by the spark discharge to melt and erode the metal material. The discharge pulse state of the wire EDM power supply directly affects the workpiece surface finish, material removal efficiency, and electrode wire lifespan. During actual processing, due to the alternating influence of various factors such as chip removal conditions, inter-electrode flow field distribution, and material properties, the inter-electrode state may frequently exhibit various modes, including normal discharge, open circuit, short circuit, and abnormal arcing. If abnormal discharge states are not monitored and fed back in a timely and accurate manner, it can easily lead to workpiece surface burns, decreased processing efficiency, and even frequent wire breakage, severely impacting production continuity. Therefore, high-precision real-time monitoring and status identification of the wire EDM power supply discharge pulse signal is a crucial step in achieving stable control of the processing process and improving the intelligence level of CNC equipment.

[0003] The inter-electrode discharge environment in wire EDM is complex and rapidly changing. The acquired power discharge pulse signals are inherently highly non-stationary, nonlinear, and strongly random high-frequency signals. In actual industrial settings, factors such as stray capacitance of the discharge cable, high-frequency switching radiation from the power inverter, and electromechanical coupling interference from the machine tool combine to often result in strong broadband background noise mixed in the acquired pulse signals. Existing condition monitoring methods often struggle to fully preserve the transient characteristics of the pulses while suppressing strong interference when processing such complex signals. Furthermore, feature extraction often fails to consider both the global energy distribution trend of the signal over time and the fine local features of instantaneous changes, leading to simplistic feature representation and insufficient anti-interference capabilities. These incomplete feature information and insufficient noise reduction capabilities make it difficult to reliably guarantee the accuracy of identifying various discharge pulse states and the stability of the system under harsh processing conditions, thus hindering further improvements in the processing performance of high-end wire EDM equipment. Summary of the Invention

[0004] To improve the accuracy of discharge state identification and operational stability under varying processing conditions, this invention provides a method and system for monitoring the discharge pulses of a high-frequency power supply in a wire EDM machine.

[0005] According to one aspect of the present invention, a method for monitoring discharge pulses of a high-frequency power supply in a wire cutting device is provided, comprising the following steps:

[0006] The wire EDM power pulse signal is acquired as the signal to be processed. Based on the frequency energy distribution of the historical wire EDM power pulse signal, the corresponding frequency band of the guiding noise is generated. The guiding noise and the signal to be processed are superimposed to form a mixed signal. Variational mode decomposition is performed on the mixed signal to obtain multiple intrinsic mode function components. The kurtosis values ​​of the Teager-Kaiser energy operator sequences for each intrinsic mode function component are calculated respectively. Intrinsic mode function components with kurtosis values ​​greater than a preset threshold are selected and reconstructed into target signals. The kurtosis values ​​of the target signals are calculated and used as weighting factors. Normalization coefficients are determined based on the weighting factors. The envelope of the target signal is extracted in parallel. The global envelope of the target signal is calculated by Hilbert transform as the first envelope. The envelope of the first product function component of the target signal is obtained by local mean decomposition as the local envelope and the local envelope is used as the second envelope. The first envelope and the second envelope are weighted by root mean square operation using normalization coefficients to generate a composite envelope. The amplitude sequence of the composite envelope is quantized by grayscale to construct the corresponding grayscale co-occurrence matrix. Feature vectors containing contrast features, second-order angular moment features, and inverse difference moment features are extracted from the grayscale co-occurrence matrix. The extracted feature vectors are input into a pre-trained classification model, and the classification model outputs the discharge state of the wire cutting power supply.

[0007] Optionally, the generation of guiding noise for the corresponding frequency band based on the frequency energy distribution of the wire cutting power pulse history signal includes: The power spectral density of the historical power pulse signal of wire cutting is obtained, the frequency band with the largest energy proportion is extracted as the characteristic frequency band, and the upper and lower limit frequencies of the characteristic frequency band are obtained. A random noise sequence is generated, and the random noise sequence is filtered using a bandpass filter with the upper and lower cutoff frequencies as the cutoff frequencies; The inverse modulation coefficient is calculated based on the equivalent frequency band amplitude of the characteristic frequency band in the historical signal and the historical reference amplitude. The amplitude of the filtered signal is scaled according to the root mean square value of the signal to be processed, the preset signal-to-noise ratio, and the inverse modulation coefficient to obtain the guiding noise.

[0008] Optionally, the step of calculating the kurtosis values ​​of the Teager-Kaiser energy operator sequence for each intrinsic mode function component includes: For any intrinsic mode function component of the time series, the Teager-Kaiser energy operator sequence corresponding to the time series is calculated by subtracting the product of the amplitude of the previous sampling point and the amplitude of the next sampling point from the square of the amplitude of the current sampling point. Calculate the square of the fourth central moment and variance of the Teager-Kaiser energy operator sequence, divide the fourth central moment by the square of the variance and then subtract the constant 3 to obtain the kurtosis value of the intrinsic mode function component.

[0009] Optionally, the step of calculating the kurtosis value of the target signal and using the kurtosis value as a weighting factor, and determining the normalization coefficient based on the weighting factor, includes: The kurtosis value of the target signal is obtained by subtracting 3 from the ratio of the fourth central moment of the target signal sequence to the square of the variance, and is used as a weighting factor. The normalized coefficient is obtained by taking the negative value of the weighting factor, calculating the exponent with the natural constant as the base, adding 1, and then taking the reciprocal. The normalized coefficient is used as the weight of the second envelope, and the difference between 1 and the normalized coefficient is used as the weight of the first envelope.

[0010] Optionally, the parallel extraction of the envelope of the target signal, wherein the global envelope of the target signal is calculated by Hilbert transform as the first envelope, the envelope of the first product function component of the target signal is obtained by local mean decomposition as the local envelope, and the local envelope is used as the second envelope, includes: The imaginary part of the analytic signal is obtained by performing a Hilbert transform on the target signal, and the amplitude sequence is calculated as the first envelope by combining it with the target signal as the real part. Extract all local extreme points of the target signal, and calculate the local mean function and local envelope estimation function of adjacent extreme points using the moving average method; The target signal is subtracted from the local mean function and divided by the local envelope estimation function to obtain the frequency-modulated signal. The frequency-modulated signal is then iteratively filtered until the envelope converges to 1 to obtain the pure frequency-modulated signal. The first product function component is obtained by multiplying the pure frequency-modulated signal with all the local envelope estimation functions generated during the iteration process, and the product of all the local envelope estimation functions is used as the second envelope.

[0011] Optionally, the step of generating a composite envelope by performing a weighted root mean square operation on the first and second envelopes using normalized coefficients includes: The difference between 1 and the normalized coefficient is used as the weighting coefficient of the first envelope. The square of the amplitude of the first envelope is multiplied by the weighting coefficient of the first envelope to obtain the first product. Multiply the square of the second envelope amplitude by the normalization coefficient to obtain the second product; The first product and the second product are added together and then the square root is taken to obtain the composite envelope amplitude of each sampling point. The composite envelope amplitudes of all sampling points are combined into a time series to generate the composite envelope.

[0012] Optionally, the gray-scale quantization of the amplitude sequence of the composite envelope is performed to construct a corresponding gray-level co-occurrence matrix, and feature vectors including contrast features, second-order angular moment features, and inverse difference moment features are extracted from the gray-level co-occurrence matrix, including: The amplitude range of the composite envelope is divided into a preset number of intervals, and the amplitude of each sampling point is mapped to the corresponding interval index to achieve gray level quantization. The preset number is the set number of gray levels. Adjacent sampling points at a set distance in the quantized sequence are taken as pixel pairs. The frequency of each gray level combination is counted and normalized to construct a gray-level co-occurrence matrix. The contrast feature, angular second moment feature, and inverse difference moment feature are calculated based on the matrix elements of the gray-level co-occurrence matrix and combined to form a one-dimensional feature vector.

[0013] According to another aspect of the present invention, a discharge pulse monitoring system for a high-frequency power supply of a wire cutting equipment is provided, comprising the following modules: The generation module is used to acquire the wire EDM power pulse signal as the signal to be processed, generate the corresponding frequency band guiding noise based on the frequency energy distribution of the historical wire EDM power pulse signal, superimpose the guiding noise and the signal to be processed into a mixed signal, perform variational mode decomposition processing on the mixed signal to obtain multiple intrinsic mode function components; The calculation module is used to calculate the kurtosis value of the Teager-Kaiser energy operator sequence of each intrinsic mode function component, filter out intrinsic mode function components with kurtosis values ​​greater than a preset threshold and reconstruct the components into a target signal, calculate the kurtosis value of the target signal and use the kurtosis value as a weighting factor, and determine the normalization coefficient based on the weighting factor. An extraction module is used to extract the envelope of the target signal in parallel. The global envelope of the target signal is calculated by Hilbert transform as the first envelope. The envelope of the first product function component of the target signal is obtained by local mean decomposition as the local envelope and the local envelope is used as the second envelope. The first envelope and the second envelope are weighted by root mean square operation using normalization coefficients to generate a composite envelope. The output module is used to perform gray-scale quantization on the amplitude sequence of the composite envelope, construct the corresponding gray-scale co-occurrence matrix, and extract feature vectors containing contrast features, second-order angular moment features, and inverse difference moment features from the gray-scale co-occurrence matrix. The extracted feature vectors are then input into a pre-trained classification model, which outputs the discharge state of the wire cutting power supply.

[0014] Preferably, the generation of guiding noise for the corresponding frequency band based on the frequency energy distribution of the wire cutting power pulse history signal includes: The power spectral density of the historical power pulse signal of wire cutting is obtained, the frequency band with the largest energy proportion is extracted as the characteristic frequency band, and the upper and lower limit frequencies of the characteristic frequency band are obtained. A random noise sequence is generated, and the random noise sequence is filtered using a bandpass filter with the upper and lower cutoff frequencies as the cutoff frequencies; The inverse modulation coefficient is calculated based on the equivalent frequency band amplitude of the characteristic frequency band in the historical signal and the historical reference amplitude. The amplitude of the filtered signal is scaled according to the root mean square value of the signal to be processed, the preset signal-to-noise ratio, and the inverse modulation coefficient to obtain the guiding noise.

[0015] Preferably, the step of calculating the kurtosis value of the Teager-Kaiser energy operator sequence for each intrinsic mode function component includes: For any intrinsic mode function component of the time series, the Teager-Kaiser energy operator sequence corresponding to the time series is calculated by subtracting the product of the amplitude of the previous sampling point and the amplitude of the next sampling point from the square of the amplitude of the current sampling point. Calculate the square of the fourth central moment and variance of the Teager-Kaiser energy operator sequence, divide the fourth central moment by the square of the variance and then subtract the constant 3 to obtain the kurtosis value of the intrinsic mode function component.

[0016] Preferably, the step of calculating the kurtosis value of the target signal and using the kurtosis value as a weighting factor, and determining the normalization coefficient based on the weighting factor, includes: The kurtosis value of the target signal is obtained by subtracting 3 from the ratio of the fourth central moment of the target signal sequence to the square of the variance, and is used as a weighting factor. The normalized coefficient is obtained by taking the negative value of the weighting factor, calculating the exponent with the natural constant as the base, adding 1, and then taking the reciprocal. The normalized coefficient is used as the weight of the second envelope, and the difference between 1 and the normalized coefficient is used as the weight of the first envelope.

[0017] Preferably, the parallel extraction of the envelope of the target signal, wherein the global envelope of the target signal calculated by Hilbert transform is used as the first envelope, the envelope of the first product function component of the target signal obtained by local mean decomposition is used as the local envelope, and the local envelope is used as the second envelope, includes: The imaginary part of the analytic signal is obtained by performing a Hilbert transform on the target signal, and the amplitude sequence is calculated as the first envelope by combining it with the target signal as the real part. Extract all local extreme points of the target signal, and calculate the local mean function and local envelope estimation function of adjacent extreme points using the moving average method; The target signal is subtracted from the local mean function and divided by the local envelope estimation function to obtain the frequency-modulated signal. The frequency-modulated signal is then iteratively filtered until the envelope converges to 1 to obtain the pure frequency-modulated signal. The first product function component is obtained by multiplying the pure frequency-modulated signal with all the local envelope estimation functions generated during the iteration process, and the product of all the local envelope estimation functions is used as the second envelope.

[0018] Preferably, the step of generating a composite envelope by performing a weighted root mean square operation on the first and second envelopes using normalized coefficients includes: The difference between 1 and the normalized coefficient is used as the weighting coefficient of the first envelope. The square of the amplitude of the first envelope is multiplied by the weighting coefficient of the first envelope to obtain the first product. Multiply the square of the second envelope amplitude by the normalization coefficient to obtain the second product; The first product and the second product are added together and then the square root is taken to obtain the composite envelope amplitude of each sampling point. The composite envelope amplitudes of all sampling points are combined into a time series to generate the composite envelope.

[0019] Preferably, the step of performing gray-scale quantization on the amplitude sequence of the composite envelope to construct a corresponding gray-level co-occurrence matrix, and extracting feature vectors from the gray-level co-occurrence matrix including contrast features, second-order angular moment features, and inverse difference moment features, includes: The amplitude range of the composite envelope is divided into a preset number of intervals, and the amplitude of each sampling point is mapped to the corresponding interval index to achieve gray level quantization. The preset number is the set number of gray levels. Adjacent sampling points at a set distance in the quantized sequence are taken as pixel pairs. The frequency of each gray level combination is counted and normalized to construct a gray-level co-occurrence matrix. The contrast feature, angular second moment feature, and inverse difference moment feature are calculated based on the matrix elements of the gray-level co-occurrence matrix and combined to form a one-dimensional feature vector.

[0020] The beneficial effects of this invention are as follows: This invention constructs guiding noise based on the frequency distribution of historical signals and combines it with variational mode decomposition to achieve fine decomposition and background noise removal of the electrical signal to be processed. The modal components are screened and the target signal is reconstructed using the kurtosis value of the Teager-Kaiser energy operator, highlighting the transient impact characteristics in the discharge pulse; simultaneously, weighting factors and normalization coefficients are determined based on the kurtosis value of the reconstructed signal itself, enabling the subsequent feature fusion process to adaptively match the signal state. By extracting the global and local envelopes in parallel and using the normalization coefficients for weighted root mean square fusion, both the macroscopic energy change trend of the pulse signal and the ability to express local abrupt changes are preserved. Furthermore, the composite envelope is converted into a gray-level co-occurrence matrix, and texture feature vectors such as contrast, second-order angular moment, and inverse difference moment are extracted. Combined with a classification model, discharge state identification is completed, thereby transforming complex time-series signal analysis into a stable feature pattern discrimination process, improving the accuracy and reliability of wire EDM power supply discharge state monitoring. Attached Figure Description

[0021] Figure 1 This is a flowchart of a discharge pulse monitoring method for a high-frequency power supply of a wire cutting device according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the intrinsic modal component sequence in an embodiment of the present invention; Figure 3 This is a schematic diagram of the composite envelope in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0023] In the first embodiment, the present invention proposes a method for monitoring the discharge pulse of a high-frequency power supply in a wire cutting device, such as... Figure 1 As shown, it includes the following steps: S1. Obtain the wire EDM power pulse signal as the signal to be processed. Generate guiding noise in the corresponding frequency band based on the frequency energy distribution of the historical wire EDM power pulse signal. Superimpose the guiding noise and the signal to be processed into a mixed signal. Perform variational mode decomposition on the mixed signal to obtain multiple intrinsic mode function components.

[0024] The voltage drop signal between the electrodes of the wire EDM power supply is acquired in real time using a high-frequency data acquisition card as the signal to be processed. The acquired analog voltage data is read and converted into a digital sequence. A Fast Fourier Transform (FFT) is performed on the historical pulse data of the wire EDM power supply to calculate the power spectral density in the frequency domain to evaluate the energy distribution of different frequency bands. A bandpass filter is constructed based on the energy distribution characteristics, and an inverse modulation coefficient is determined based on the energy amplitude of the main frequency band of the historical signal. This ensures that the larger the historical energy amplitude of the main frequency band, the smaller the injected amplitude of the guiding noise, and vice versa. This generates a band-limited Gaussian random noise sequence inversely proportional to the energy amplitude of the main frequency band of the historical signal, serving as the guiding noise. The generated guiding noise is superimposed onto the signal sequence to be processed according to a set signal-to-noise ratio to construct a mixed signal. The mixed signal is processed using a variational mode decomposition (VMD) algorithm, with a quadratic penalty factor of 2000 and a decomposition mode number of 5. The mixed signal is decomposed in the frequency domain, and the center frequency and bandwidth of each mode are iteratively updated to obtain five band-limited intrinsic mode function components.

[0025] In an optional embodiment, the generation of guiding noise for the corresponding frequency band based on the frequency energy distribution of the wire-cut power pulse history signal includes: The power spectral density of the historical power pulse signal of wire cutting is obtained, the frequency band with the largest energy proportion is extracted as the characteristic frequency band, and the upper and lower limit frequencies of the characteristic frequency band are obtained. A random noise sequence is generated, and the random noise sequence is filtered using a bandpass filter with the upper and lower cutoff frequencies as the cutoff frequencies; The inverse modulation coefficient is calculated based on the equivalent frequency band amplitude of the characteristic frequency band in the historical signal and the historical reference amplitude. The amplitude of the filtered signal is scaled according to the root mean square value of the signal to be processed, the preset signal-to-noise ratio, and the inverse modulation coefficient to obtain the guiding noise.

[0026] When acquiring the characteristic frequency band, the Welch method is used to perform a Fourier transform on the historical signal sequence of wire-cut power pulses of length N=4096 to calculate the power spectral density. The Hamming window length is set to L=1024, and the overlap rate is 50%. The Welch method, also known as the Welch power spectral estimation method, is a commonly used method for estimating the power spectral density of signals. The integrated energy of different frequency bands is calculated, for example, each 100kHz is considered a frequency band interval. The frequency band with the largest proportion of energy integral to the total frequency domain energy is selected as the characteristic frequency band. The characteristic frequency band can be a single frequency band interval or a frequency range formed by merging multiple adjacent frequency band intervals with continuously concentrated energy. Assume the sampling frequency is... =10MHz. If calculations show that the energy share of the 1.5MHz to 2.0MHz band reaches 45%, then this band is determined as the characteristic frequency band, and the corresponding lower limit frequency is obtained. =1.5MHz and upper limit frequency =2.0MHz.

[0027] A Gaussian white noise sequence following a standard normal distribution N(0,1) is generated using a pseudo-random number generator as the random noise sequence. The length of this random noise sequence is consistent with the length of the signal to be processed, for example, N=4096 sampling points. A 6th-order Butterworth bandpass filter is constructed, with a lower cutoff frequency of 1.5MHz and an upper cutoff frequency of 2.0MHz as the -3dB cutoff frequency. Forward and backward zero-phase filtering is applied to the generated random noise sequence to eliminate phase distortion, obtaining the filtered noise within a specific frequency band. The energy amplitude of this characteristic frequency band in the historical signal is calculated. And set historical reference amplitude. And to prevent tiny positive numbers with a denominator of zero. ,according to Calculate the inverse modulation coefficient ;in, Used to characterize the inverse modulation relationship between the energy amplitude of historical major frequency bands and the intensity of guided noise injection. The equivalent frequency band amplitude is obtained by taking the square root of the integral value of the power spectral density of the characteristic frequency band. This refers to the reference equivalent amplitude obtained using the same calculation method within the historical reference period. Preferably, it can be... Limited to a preset range To avoid modulation that is too strong or too weak.

[0028] Calculate the root mean square (RMS) value of the current signal sequence to be processed, and set a preset signal-to-noise ratio (SNR) range of 5dB to 15dB to guide variational mode decomposition (VMD) without masking the true features. For example, take a preset SNR of... =10dB, first according to Calculate the effective value coefficient of the basic noise, and then apply the formula. The effective value coefficient of the target noise after inverse modulation with historical frequency band energy is calculated. Based on this, the amplitude of the filtered noise is linearly scaled to make the scaled root mean square value of the noise equal to the calculated value. This yields the guiding noise used for superposition. Therefore, a preset signal-to-noise ratio is used to define the basic proportion of the noise relative to the current signal to be processed, and an inverse modulation coefficient is used to reflect the adaptive correction of the guiding noise intensity by the energy amplitude of the main frequency bands of the historical signal.

[0029] S2, calculate the kurtosis value of the Teager-Kaiser energy operator sequence for each intrinsic mode function component, filter out intrinsic mode function components with kurtosis values ​​greater than a preset threshold and reconstruct the components into a target signal, calculate the kurtosis value of the target signal and use the kurtosis value as a weighting factor, and determine the normalization coefficient based on the weighting factor.

[0030] For each output intrinsic mode function (IMF) component sequence, each sampling point in the sequence is traversed. The Teager-Kaiser energy operator sequence for each IMF component is obtained by subtracting the product of the amplitudes of the previous and next sampling points from the square of the amplitude at the current sampling point. The kurtosis value of each energy operator sequence is calculated to evaluate the impact characteristic intensity of the corresponding mode component. A preset threshold of 3 for kurtosis values ​​is set. IMF components with kurtosis values ​​greater than the preset threshold of 3 are selected using logical judgment statements. The selected IMF components are then summed point-by-point along the corresponding dimension of the sequence to reconstruct the target signal sequence. If the kurtosis values ​​of all IMF components do not exceed the preset threshold of 3, at least one IMF component with the largest kurtosis value is selected as the minimum reconstructed component to generate the target signal sequence, ensuring effective input for subsequent kurtosis calculation and envelope extraction steps. An illustration of the IMF component sequence is shown below. Figure 2 As shown, the kurtosis value of the reconstructed target signal sequence is calculated again, and this kurtosis value is directly used as the weighting factor. The normalization coefficient is calculated using the exponential normalization algorithm. The negative power of the weighting factor is calculated using the exp function in the mathematical function library. The result is then incremented by 1 and the reciprocal is taken. The value between 0 and 1 is determined as the normalization coefficient.

[0031] In an optional embodiment, calculating the kurtosis values ​​of the Teager-Kaiser energy operator sequence for each intrinsic mode function component includes: For any intrinsic mode function component of the time series, the Teager-Kaiser energy operator sequence corresponding to the time series is calculated by subtracting the product of the amplitude of the previous sampling point and the amplitude of the next sampling point from the square of the amplitude of the current sampling point. Calculate the square of the fourth central moment and variance of the Teager-Kaiser energy operator sequence, divide the fourth central moment by the square of the variance and then subtract the constant 3 to obtain the kurtosis value of the intrinsic mode function component.

[0032] For any intrinsic mode function (IMF) component time series of length N obtained by variational mode decomposition (VMD) Where n represents the discrete sampling point index, n=2,3,...,N-1. The energy operator value for each point in the sequence is calculated using the Teager-Kaiser energy operator (TKEO) nonlinear tracking algorithm. For the current nth sampling point in the time series, the amplitude is extracted. and the amplitude of the adjacent previous sampling point and the amplitude of the next sampling point Through mathematical expressions Perform algebraic operations. For example, if the amplitudes of three consecutive sampling points in an IMF sequence are 0.5V, 1.2V, and 0.8V respectively, then the TKEO value at the current point of 1.2V is 1.04. Iterate through all points from n=2 to N-1 sequentially; the first and last points can be zero-padding or mirror extension to generate a TKEO energy operator sequence of the same length as the original sequence. .

[0033] After obtaining the energy operator sequence Then, statistical calculations are performed on the overall probability distribution characteristics to obtain the kurtosis value. The average value of the energy operator sequence is then calculated. Next, the fourth central moments of each element in the sequence as it deviates from the mean are calculated. and second-order central moments To obtain the normalized super-kurtosis, the obtained fourth-order central moments are... Divide by the square of the variance Then subtract a constant 3, which is used to offset the theoretical kurtosis value of the standard normal distribution. For example, when calculating a sequence of energy operators... =650, variance If the value is 10, the kurtosis value is 3.5. If this kurtosis value is greater than the preset filtering threshold... If the kurtosis value is greater than 3, then the IMF component is determined to contain discharge transient impact characteristics, and the IMF component is retained for subsequent target signal reconstruction. For IMF components with a kurtosis of 3, the IMF component with the largest kurtosis value is retained as the safety net component to avoid the target signal being empty.

[0034] In an optional embodiment, calculating the kurtosis value of the target signal and using the kurtosis value as a weighting factor, and determining the normalization coefficient based on the weighting factor, includes: The kurtosis value of the target signal is obtained by subtracting 3 from the ratio of the fourth central moment of the target signal sequence to the square of the variance, and is used as a weighting factor. The normalized coefficient is obtained by taking the negative value of the weighting factor, calculating the exponent with the natural constant as the base, adding 1, and then taking the reciprocal. The normalized coefficient is used as the weight of the second envelope, and the difference between 1 and the normalized coefficient is used as the weight of the first envelope.

[0035] Let the target signal sequence obtained after filtering and reconstruction be... The sequence length is also N=4096. Calculate the mean of this target signal sequence. Find the fourth-order central moments that represent the flatness of the signal distribution. and the variance representing the degree of energy dispersion According to the definition of kurtosis, Divide by The square of the value and the offset constant 3 are used to obtain the kurtosis value of the target signal. Assign this kurtosis value to the weighting factor. ,Right now This factor represents the relative abundance of the high-frequency transient impact component of the line-cutting signal in the target signal. The sharper and sparser the high-frequency pulses in the target signal, the higher the calculated abundance. The larger the value; for example, when there is a tendency for arc discharge, the higher the calculated value. The value can reach 2.5. If the target signal is relatively stable and the excess kurtosis is negative, then the negative value is directly entered into the subsequent Sigmoid mapping, so that the weight of the second envelope is reduced accordingly, and the absolute value is no longer taken separately.

[0036] To weight factors The probability space is mapped to 0 to 1 for subsequent weighting, employing a nonlinear transformation mechanism based on the Sigmoid function. Specifically, the weighting factors are... Inverse As a natural constant e, it is approximately equal to an exponent of 2.71828. Add 1 to the index value and take the reciprocal to calculate the normalization coefficient. For example, when When = 2.5, the normalized coefficient ≈0.924. This calculation mechanism ensures that as the kurtosis value increases, It will smoothly and monotonically approximate 1.

[0037] The normalization coefficients obtained above For example, 0.924 is configured as the weight coefficient of the second envelope; calculate 1 minus The difference, 0.076, is used as the weighting coefficient of the first envelope. When there is a sudden change in the discharge state, the model automatically assigns a larger weight to the second envelope to detect pulse distortion features.

[0038] S3, extract the envelope of the target signal in parallel, wherein the global envelope of the target signal is calculated by Hilbert transform as the first envelope, the envelope of the first product function component of the target signal is obtained by local mean decomposition as the local envelope, and the local envelope is used as the second envelope. The first envelope and the second envelope are weighted by root mean square operation using normalization coefficients to generate a composite envelope.

[0039] The target signal sequence is constructed analytically, and the absolute value sequence of the analytical signal is obtained to obtain the instantaneous amplitude sequence of the target signal as the global envelope, i.e., the first envelope. The LMD function is used in parallel to perform local mean decomposition on the target signal. Through continuous sliding smoothing, the local mean function and local envelope estimation function are calculated, iteratively separating the pure frequency-modulated signal and the envelope signal. The various levels of local envelope estimation functions generated during the generation of the first product function component are extracted, and these local envelope estimation functions are multiplied point-by-point according to the sampling points to obtain the envelope sequence of the first product function component. This envelope sequence is used as the local envelope, i.e., the second envelope. The previously determined normalization coefficients are obtained and used as the square weights of the second envelope. 1 minus the normalization coefficients is used as the square weights of the first envelope. Multiplication and addition operators are used to add the result of multiplying the square of the first envelope amplitude by 1 minus the normalization coefficients to the result of multiplying the square of the second envelope amplitude by the normalization coefficients. The square root of each sampling point is calculated for the combined sequence, generating a composite envelope containing both global and local features.

[0040] In an optional embodiment, the parallel extraction of the envelope of the target signal, wherein the global envelope of the target signal is calculated by Hilbert transform as the first envelope, the envelope of the first product function component of the target signal is obtained by local mean decomposition as the local envelope, and the local envelope is used as the second envelope, includes: The imaginary part of the analytic signal is obtained by performing a Hilbert transform on the target signal, and the amplitude sequence is calculated as the first envelope by combining it with the target signal as the real part. Extract all local extreme points of the target signal, and calculate the local mean function and local envelope estimation function of adjacent extreme points using the moving average method; The target signal is subtracted from the local mean function and divided by the local envelope estimation function to obtain the frequency-modulated signal. The frequency-modulated signal is then iteratively filtered until the envelope converges to 1 to obtain the pure frequency-modulated signal. The first product function component is obtained by multiplying the pure frequency-modulated signal with all the local envelope estimation functions generated during the iteration process, and the product of all the local envelope estimation functions is used as the second envelope.

[0041] When calculating the first envelope, for the discrete target signal Performing a Fast Fourier Transform (FFT) sets the negative frequency components of the spectrum to zero and multiplies the amplitude of the positive frequency components by 2. Then performing an Inverse Fast Fourier Transform (IFFT) yields a Hilbert Transform sequence with orthogonal imaginary parts. The original target signal As the real part, it forms a complex analytic signal together with the imaginary part. Calculate the complex modulus for each sampling point of the analytic signal sequence, i.e. The resulting time series represents the first envelope of the macroscopic transient energy of the target signal. For example, if the original signal at a certain point is 3V and the imaginary part after the Hilbert transform is 4V, then the amplitude of the first envelope at that point is 5V.

[0042] When calculating the second envelope, the Local Mean Decomposition (LMD) algorithm is applied to process the target signal. The target signal is traversed through differentiation and differencing. Extract the sequences and time indices of local maxima and minima. Calculate the midpoint of the line connecting adjacent extreme points as the initial mean point, and half the absolute value of the difference between extreme points as the initial envelope point. Use a smoothing window of width W (e.g., one-third of the distance between adjacent extreme points) for the discrete midpoint and envelope point, employing a moving average method or cubic spline interpolation with W=5 sampling points, to generate a smooth local mean function. and local envelope estimation function From the original signal Subtract To obtain a zero-mean signal Then divide the zero-mean signal by the envelope function. Obtain the first frequency modulation signal .

[0043] The step of extracting the frequency modulation signal is iteratively filtered, that is, the step of extracting the frequency modulation signal is iteratively filtered. Treating this as a new input signal, repeat the above process of extreme value extraction and moving average envelope calculation. In each iteration, denoted as the k-th iteration, calculate the new frequency-modulated signal and check whether the local envelope function of the new frequency-modulated signal approximates 1 over the entire sequence. Assume the convergence condition is that the absolute error is less than or equal to a threshold. ,in Set as Once this tolerance is met, the iteration stops, and the signal at this point is confirmed as a pure frequency-modulated signal without amplitude modulation components. The obtained pure frequency-modulated signal is multiplied by all the local envelope estimation functions generated in the k iterations to recover the first product function component, PF1. The pointwise product sequence of the k local envelope estimation functions, since it is extracted from the first component containing high-frequency discharge impulse components, is used as the second envelope for identifying local abrupt change characteristics.

[0044] In an optional embodiment, the step of generating a composite envelope by performing a weighted root mean square operation on the first and second envelopes using normalization coefficients includes: The difference between 1 and the normalized coefficient is used as the weighting coefficient of the first envelope. The square of the amplitude of the first envelope is multiplied by the weighting coefficient of the first envelope to obtain the first product. Multiply the square of the second envelope amplitude by the normalization coefficient to obtain the second product; The first product and the second product are added together and then the square root is taken to obtain the composite envelope amplitude of each sampling point. The composite envelope amplitudes of all sampling points are combined into a time series to generate the composite envelope.

[0045] Let the first envelope time series be The second envelope time series is , and the normalization coefficients determined in the preceding steps are When performing fusion calculations point by point, for the current sampling point with index n, the amplitude of the first envelope at that point is obtained and squared. , representing the transient energy value at that point, will Multiplied by weight constant This yields the first product term, which focuses on macroscopic features. For example, suppose the amplitude of the first envelope at a certain sampling point is 5V, and the normalization coefficient has been calculated. =0.9, then the first product .

[0046] For sampling points with the same index n, the second envelope amplitude is extracted and squared. and will Multiply by the normalization factor The second product term, which focuses on the characteristics of local mutations, is calculated. Continuing with the numerical example above, if the second envelope amplitude at this point is 8V, then the second product... After completing the two product calculations, the first product... With the second product By performing algebraic addition, the weighted composite energy value is obtained.

[0047] After summing, the square root of the combined energy value is taken as a whole, i.e., the formula... This is taken as the composite envelope amplitude of that sampling point. In the previous example, the sum is 60.1. The square root yields a composite envelope amplitude of approximately 7.75V. By iterating through all sampling points, the scalar amplitudes obtained from the weighted root mean square operation at each point are sequentially concatenated and combined according to the original time sequence, constructing a characteristic one-dimensional time series that considers both the macroscopic discharge trend and microscopic pulse abrupt changes. This one-dimensional time series is the generated composite envelope, reducing energy loss due to bandwidth limitations in signal envelope calculation. A schematic diagram of the composite envelope is shown below. Figure 3 As shown.

[0048] S4. The amplitude sequence of the composite envelope is quantized by gray to construct the corresponding gray-level co-occurrence matrix. Feature vectors containing contrast features, second-order angular moment features, and inverse difference moment features are extracted from the gray-level co-occurrence matrix. The extracted feature vectors are input into the pre-trained classification model, and the classification model outputs the discharge state of the wire cutting power supply.

[0049] The minimum and maximum values ​​of the composite envelope amplitude sequence are calculated separately. Using the extreme value normalization formula, each value in the composite envelope sequence is subtracted from the minimum value and then divided by the difference between the maximum and minimum values. The normalization result is then mapped to discrete gray levels between 0 and G-1 according to a preset gray level number G, resulting in a one-dimensional gray quantization sequence. Sampling point pairs with a set distance d in the one-dimensional gray quantization sequence are used as co-occurrence statistics objects. The frequency of occurrence of each gray level combination is counted and normalized to construct a gray-level co-occurrence matrix. Based on the element probability distribution of the gray-level co-occurrence matrix, contrast features, ASM angular second moment features, and inverse difference moment features are calculated respectively, and the three feature values ​​are concatenated in sequence to form a one-dimensional feature vector. A classification model is constructed using the Support Vector Machine (SVC) classification algorithm. This model has been pre-trained under supervised supervision using historical discharge data samples and the `fit` function. The `predict` function of the classification model is called, and the extracted one-dimensional feature vector is passed in as input parameters. The classification model calculates and outputs one of the following discharge states of the wire-cutting power supply: normal spark discharge, short-circuit discharge, open-circuit discharge, or abnormal arcing discharge. The training and prediction phases use the same grayscale levels, set distances, feature extraction order, and feature normalization rules. The historical discharge data samples are divided into training, validation, and test sets, with the test set not participating in the classification model training.

[0050] In an optional embodiment, the gray-scale quantization of the amplitude sequence of the composite envelope to construct a corresponding gray-level co-occurrence matrix, and the extraction of feature vectors from the gray-level co-occurrence matrix including contrast features, second-order angular moment features, and inverse difference moment features, include: The amplitude range of the composite envelope is divided into a preset number of intervals, and the amplitude of each sampling point is mapped to the corresponding interval index to achieve gray level quantization. The preset number is the set number of gray levels. Adjacent sampling points at a set distance in the quantized sequence are taken as pixel pairs. The frequency of each gray level combination is counted and normalized to construct a gray-level co-occurrence matrix. The contrast feature, angular second moment feature, and inverse difference moment feature are calculated based on the matrix elements of the gray-level co-occurrence matrix and combined to form a one-dimensional feature vector.

[0051] When performing gray quantization on a one-dimensional composite envelope time series, search for the minimum amplitude in the series. and maximum amplitude Determine the global amplitude span range. Set the preset grayscale level G to 16, or 32, or other powers of 2, and proportionally divide the amplitude span into 16 continuous sub-intervals. The width step of each sub-interval is [missing value]. For each sampling point in the sequence, the index of the interval containing the composite envelope amplitude corresponding to the current sampling point is calculated by rounding down: If it happens to be the maximum value, it is directly classified as level 15, transforming the floating-point amplitude sequence into a one-dimensional discrete grayscale sequence with values ​​ranging from 0 to 15. For example, if the range is 0V to 16V, =1V, then a sampling point with an amplitude of 3.4V is represented as gray level 3.

[0052] In the gray-level co-occurrence matrix construction stage, the quantized gray-level sequence is treated as a one-dimensional temporal gray-level structure. A specific bias step size is constructed as the set distance d. To detect the periodic characteristics of high-frequency discharge, the set distance is set to d = 1 to 5 sampling periods, such as d = 1. The temporal representation sequence is traversed, and point pairs with a distance of d are taken as a pair of "pixels". A 16×16 zero matrix is ​​initialized. If the gray values ​​of the point pair are i and j, i, j∈[0,15], then the count cell in the i-th row and j-th column of the matrix is ​​incremented by one. After the statistics are completed, the count values ​​of all elements in the matrix are divided by the total number of point pairs, so that the sum of the probabilities of all elements in the entire matrix is ​​1, thus obtaining the probability distribution of the included elements. The normalized gray-level co-occurrence matrix.

[0053] Based on the probability distribution of elements in the constructed gray-level co-occurrence matrix Calculate three texture statistical feature parameters. The contrast feature is calculated using the formula... The value is calculated to represent the severity of the abrupt change in the amplitude of the composite envelope; a larger value indicates a stronger local spark impact during wire cutting. The second moment characteristic of the angle is determined using the formula... Calculations are performed to represent the uniformity and coarseness of the envelope distribution; the inverse difference moment characteristic is expressed by the formula... The local smoothness that can represent the waveform of the discharge pulse signal is obtained. For example, if Con=45.2, ASM=0.08, and IDM=0.35 are calculated, then the above three real numbers are arranged and combined in sequence to form a feature vector [45.2,0.08,0.35]. The feature vector is output for the back-end classification model to determine the real-time discharge state of the wire cutting power supply.

[0054] The test set, consisting of 5000 discharge pulse samples, was compiled from power pulse signals collected during actual wire EDM processing, covering various operating conditions including normal spark discharge, short-circuit discharge, open-circuit discharge, and abnormal arcing discharge. The signal sampling frequency was set to 10MHz, and all data were processed on the same hardware platform. The control experiment was divided into three groups: the first group used only the first envelope calculated by Hilbert transform to extract features and perform state classification; the second group used only the second envelope obtained by local mean decomposition to extract features and perform state classification; and the third group used a composite envelope generated based on kurtosis weighted root mean square operation to extract features and perform state classification. All three groups used the same support vector machine as the backend classification model, and all samples underwent gray quantization and feature vector extraction, including contrast, second angular moment, and inverse difference moment.

[0055] Based on test set verification, the first group, relying on a single first envelope, achieved an accuracy of 82.5% in discharge state identification with a false alarm rate of 11.4%. The second group, using only the second envelope, achieved an accuracy of 86.3% with a false alarm rate of 8.7%. The third group, employing the composite envelope of this invention, achieved an accuracy of 95.8% while reducing the false alarm rate to 2.1%. In terms of the calculated feature vector values, the third group showed significantly higher numerical sensitivity and inter-class discrimination when arc discharge abruptly occurred, with the mean contrast feature values ​​being 45.2% and 22.6% higher than the first and second groups, respectively.

[0056] The composite envelope scheme improved the state recognition accuracy by 13.3 percentage points and 9.5 percentage points compared to the first and second groups, respectively. The single first envelope, focusing on macroscopic energy trends, easily misses high-frequency local transient impacts, limiting classification accuracy. While the single second envelope detects local microscopic abrupt changes, it lacks representation of the overall discharge energy profile. By using kurtosis-based normalization coefficients as weights for weighted fusion, the scheme intelligently allocates proportions according to the sharpness of the discharge signal, preserving both macroscopic energy change patterns and highlighting local pulse distortion details. This eliminates the energy omission problem caused by the limitation of a single computational bandwidth and enhances the model's real-time perception capability of complex operating conditions.

[0057] In a second embodiment, the present invention also proposes a discharge pulse monitoring system for a high-frequency power supply of a wire cutting equipment, comprising the following modules: The generation module is used to acquire the wire EDM power pulse signal as the signal to be processed, generate the corresponding frequency band guiding noise based on the frequency energy distribution of the historical wire EDM power pulse signal, superimpose the guiding noise and the signal to be processed into a mixed signal, perform variational mode decomposition processing on the mixed signal to obtain multiple intrinsic mode function components; The calculation module is used to calculate the kurtosis value of the Teager-Kaiser energy operator sequence of each intrinsic mode function component, filter out intrinsic mode function components with kurtosis values ​​greater than a preset threshold and reconstruct the components into a target signal, calculate the kurtosis value of the target signal and use the kurtosis value as a weighting factor, and determine the normalization coefficient based on the weighting factor. An extraction module is used to extract the envelope of the target signal in parallel. The global envelope of the target signal is calculated by Hilbert transform as the first envelope. The envelope of the first product function component of the target signal is obtained by local mean decomposition as the local envelope and the local envelope is used as the second envelope. The first envelope and the second envelope are weighted by root mean square operation using normalization coefficients to generate a composite envelope. The output module is used to perform gray-scale quantization on the amplitude sequence of the composite envelope, construct the corresponding gray-scale co-occurrence matrix, and extract feature vectors containing contrast features, second-order angular moment features, and inverse difference moment features from the gray-scale co-occurrence matrix. The extracted feature vectors are then input into a pre-trained classification model, which outputs the discharge state of the wire cutting power supply.

[0058] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the present invention.

Claims

1. A method for monitoring discharge pulses in the high-frequency power supply of a wire cutting machine, characterized in that, Includes the following steps: The wire EDM power pulse signal is acquired as the signal to be processed. Based on the frequency energy distribution of the historical wire EDM power pulse signal, the corresponding frequency band of the guiding noise is generated. The guiding noise and the signal to be processed are superimposed to form a mixed signal. Variational mode decomposition is performed on the mixed signal to obtain multiple intrinsic mode function components. The kurtosis values ​​of the Teager-Kaiser energy operator sequences for each intrinsic mode function component are calculated respectively. Intrinsic mode function components with kurtosis values ​​greater than a preset threshold are selected and reconstructed into target signals. The kurtosis values ​​of the target signals are calculated and used as weighting factors. Normalization coefficients are determined based on the weighting factors. The envelope of the target signal is extracted in parallel. The global envelope of the target signal is calculated by Hilbert transform as the first envelope. The envelope of the first product function component of the target signal is obtained by local mean decomposition as the local envelope and the local envelope is used as the second envelope. The first envelope and the second envelope are weighted by root mean square operation using normalization coefficients to generate a composite envelope. The amplitude sequence of the composite envelope is quantized by grayscale to construct the corresponding grayscale co-occurrence matrix. Feature vectors containing contrast features, second-order angular moment features, and inverse difference moment features are extracted from the grayscale co-occurrence matrix. The extracted feature vectors are input into a pre-trained classification model, and the classification model outputs the discharge state of the wire cutting power supply.

2. The method for monitoring the discharge pulse of a high-frequency power supply in a wire cutting device according to claim 1, characterized in that, The generation of guiding noise for the corresponding frequency band based on the frequency energy distribution of the wire cutting power pulse history signal includes: The power spectral density of the historical power pulse signal of wire cutting is obtained, the frequency band with the largest energy proportion is extracted as the characteristic frequency band, and the upper and lower limit frequencies of the characteristic frequency band are obtained. A random noise sequence is generated, and the random noise sequence is filtered using a bandpass filter with the upper and lower cutoff frequencies as the cutoff frequencies; The inverse modulation coefficient is calculated based on the equivalent frequency band amplitude of the characteristic frequency band in the historical signal and the historical reference amplitude. The amplitude of the filtered signal is scaled according to the root mean square value of the signal to be processed, the preset signal-to-noise ratio, and the inverse modulation coefficient to obtain the guiding noise.

3. The method for monitoring the discharge pulse of a high-frequency power supply in a wire cutting device according to claim 1, characterized in that, The calculation of the kurtosis values ​​of the Teager-Kaiser energy operator sequence for each intrinsic mode function component includes: For any intrinsic mode function component of the time series, the Teager-Kaiser energy operator sequence corresponding to the time series is calculated by subtracting the product of the amplitude of the previous sampling point and the amplitude of the next sampling point from the square of the amplitude of the current sampling point. Calculate the square of the fourth central moment and variance of the Teager-Kaiser energy operator sequence, divide the fourth central moment by the square of the variance and then subtract the constant 3 to obtain the kurtosis value of the intrinsic mode function component.

4. The method for monitoring the discharge pulse of a high-frequency power supply in a wire cutting device according to claim 3, characterized in that, The calculation of the kurtosis value of the target signal and the use of the kurtosis value as a weighting factor, and the determination of the normalization coefficient based on the weighting factor, include: The kurtosis value of the target signal is obtained by subtracting 3 from the ratio of the fourth central moment of the target signal sequence to the square of the variance, and is used as a weighting factor. The normalized coefficient is obtained by taking the negative value of the weighting factor, calculating the exponent with the natural constant as the base, adding 1, and then taking the reciprocal. The normalized coefficient is used as the weight of the second envelope, and the difference between 1 and the normalized coefficient is used as the weight of the first envelope.

5. A method for monitoring discharge pulses of a high-frequency power supply in a wire cutting device according to claim 1 or 2, characterized in that, The parallel extraction of the envelope of the target signal, wherein the global envelope of the target signal is calculated by Hilbert transform as the first envelope, and the envelope of the first product function component of the target signal is obtained by local mean decomposition as the local envelope, and the local envelope is used as the second envelope, includes: The imaginary part of the analytic signal is obtained by performing a Hilbert transform on the target signal, and the amplitude sequence is calculated as the first envelope by combining it with the target signal as the real part. Extract all local extreme points of the target signal, and calculate the local mean function and local envelope estimation function of adjacent extreme points using the moving average method; The target signal is subtracted from the local mean function and divided by the local envelope estimation function to obtain the frequency-modulated signal. The frequency-modulated signal is then iteratively filtered until the envelope converges to 1 to obtain the pure frequency-modulated signal. The first product function component is obtained by multiplying the pure frequency-modulated signal with all the local envelope estimation functions generated during the iteration process, and the product of all the local envelope estimation functions is used as the second envelope.

6. The method for monitoring the discharge pulse of a high-frequency power supply in a wire cutting device according to claim 1, characterized in that, The step of generating a composite envelope by performing a weighted root mean square operation on the first and second envelopes using normalization coefficients includes: The difference between 1 and the normalized coefficient is used as the weighting coefficient of the first envelope. The square of the amplitude of the first envelope is multiplied by the weighting coefficient of the first envelope to obtain the first product. Multiply the square of the second envelope amplitude by the normalization coefficient to obtain the second product; The first product and the second product are added together and then the square root is taken to obtain the composite envelope amplitude of each sampling point. The composite envelope amplitudes of all sampling points are combined into a time series to generate the composite envelope.

7. The method for monitoring the discharge pulse of a high-frequency power supply in a wire cutting device according to claim 1, characterized in that, The process involves gray-quantizing the amplitude sequence of the composite envelope to construct a corresponding gray-level co-occurrence matrix, and extracting feature vectors from the gray-level co-occurrence matrix that include contrast features, second-order angular moment features, and inverse difference moment features, including: The amplitude range of the composite envelope is divided into a preset number of intervals, and the amplitude of each sampling point is mapped to the corresponding interval index to achieve gray level quantization. The preset number is the set number of gray levels. Adjacent sampling points at a set distance in the quantized sequence are taken as pixel pairs. The frequency of each gray level combination is counted and normalized to construct a gray-level co-occurrence matrix. The contrast feature, angular second moment feature, and inverse difference moment feature are calculated based on the matrix elements of the gray-level co-occurrence matrix and combined to form a one-dimensional feature vector.

8. A discharge pulse monitoring system for a high-frequency power supply of a wire cutting equipment, characterized in that, Includes the following modules: The generation module is used to acquire the wire EDM power pulse signal as the signal to be processed, generate the corresponding frequency band guiding noise based on the frequency energy distribution of the historical wire EDM power pulse signal, superimpose the guiding noise and the signal to be processed into a mixed signal, perform variational mode decomposition processing on the mixed signal to obtain multiple intrinsic mode function components; The calculation module is used to calculate the kurtosis value of the Teager-Kaiser energy operator sequence of each intrinsic mode function component, filter out intrinsic mode function components with kurtosis values ​​greater than a preset threshold and reconstruct the components into a target signal, calculate the kurtosis value of the target signal and use the kurtosis value as a weighting factor, and determine the normalization coefficient based on the weighting factor. An extraction module is used to extract the envelope of the target signal in parallel. The global envelope of the target signal is calculated by Hilbert transform as the first envelope. The envelope of the first product function component of the target signal is obtained by local mean decomposition as the local envelope and the local envelope is used as the second envelope. The first envelope and the second envelope are weighted by root mean square operation using normalization coefficients to generate a composite envelope. The output module is used to perform gray-scale quantization on the amplitude sequence of the composite envelope, construct the corresponding gray-scale co-occurrence matrix, and extract feature vectors containing contrast features, second-order angular moment features, and inverse difference moment features from the gray-scale co-occurrence matrix. The extracted feature vectors are then input into a pre-trained classification model, which outputs the discharge state of the wire cutting power supply.

9. A discharge pulse monitoring system for a high-frequency power supply of a wire cutting equipment according to claim 8, characterized in that, The generation of guiding noise for the corresponding frequency band based on the frequency energy distribution of the wire cutting power pulse history signal includes: The power spectral density of the historical power pulse signal of wire cutting is obtained, the frequency band with the largest energy proportion is extracted as the characteristic frequency band, and the upper and lower limit frequencies of the characteristic frequency band are obtained. A random noise sequence is generated, and the random noise sequence is filtered using a bandpass filter with the upper and lower cutoff frequencies as the cutoff frequencies; The inverse modulation coefficient is calculated based on the equivalent frequency band amplitude of the characteristic frequency band in the historical signal and the historical reference amplitude. The amplitude of the filtered signal is scaled according to the root mean square value of the signal to be processed, the preset signal-to-noise ratio, and the inverse modulation coefficient to obtain the guiding noise.

10. A discharge pulse monitoring system for a high-frequency power supply of a wire cutting equipment according to claim 8, characterized in that, The calculation of the kurtosis values ​​of the Teager-Kaiser energy operator sequence for each intrinsic mode function component includes: For any intrinsic mode function component of the time series, the Teager-Kaiser energy operator sequence corresponding to the time series is calculated by subtracting the product of the amplitude of the previous sampling point and the amplitude of the next sampling point from the square of the amplitude of the current sampling point. Calculate the square of the fourth central moment and variance of the Teager-Kaiser energy operator sequence, divide the fourth central moment by the square of the variance and then subtract the constant 3 to obtain the kurtosis value of the intrinsic mode function component.

Images