Method for detecting clamping chips of cutter of numerical control machine tool, electronic equipment and storage medium

Abstract

The invention relates to the field of numerical control machine tool detection.The method comprises the steps that regularization inverse operation is conducted on vibration snapshots according to a pre-calibrated frequency response function matrix between a tool nose and a sensor installation point; performing continuous wavelet transform on the tool nose time-domain signals to obtain time-frequency spectrums, and splicing the time-frequency spectrums to obtain a two-dimensional time-frequency-space atlas; segmenting the two-dimensional time-frequency-space map into a plurality of super-pixel regions, calculating an abnormal score of each super-pixel region, judging the super-pixel region of which the abnormal score exceeds a preset score threshold as an abnormal region, and taking the maximum abnormal score in all the abnormal regions as a chip blocking index; according to the method, a multi-source feature vector is constructed by utilizing the chip blocking index, and a machine tool cutter chip blocking detection result is obtained based on the multi-source feature vector, so that chip blocking features which are originally fuzzy due to transmission path distortion can be clearly displayed.

Description

Technical Field

[0001] This invention relates to the field of CNC machine tool inspection, and more particularly to a method, electronic device, and storage medium for detecting chip jamming in CNC machine tool cutting tools. Background Technology

[0002] In the precision machining process of CNC machine tools, abnormal accumulation of chips in the tool-workpiece contact area (i.e., chip jamming) is the main cause of tool breakage, deterioration of workpiece surface quality, and even machine system shutdown.

[0003] To ensure machining stability, existing technologies often install vibration sensors on the spindle or machine bed to identify chip jamming by analyzing vibration signals. However, vibration sensors are usually installed at the far end of the spindle bearing housing or the worktable, rather than directly at the tool tip. The transient impact signal generated by chip jamming must pass through multiple mechanical contact surfaces, such as the tool holder-spindle interface and the bearing-housing contact surface, before being picked up by the sensor. During this process, the high-frequency impact signal undergoes severe attenuation, reflection, and mode mixing, resulting in a complex nonlinear distortion relationship between the signal waveform received by the sensor and the actual excitation at the tool tip. This distortion effect means that traditional methods based on time-domain waveforms, frequency-domain energy, or simple statistical feature extraction often only capture the "contaminated" signal, resulting in a low signal-to-noise ratio and blurred chip jamming characteristics. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method, electronic device and storage medium for detecting chip jamming of CNC machine tool tools, which solves the technical problems existing in the prior art, such as low signal-to-noise ratio of the acquired vibration signal and blurred chip jamming characteristics due to the distortion of the signal transmission path caused by the vibration sensor being far away from the tool tip.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0008] In a first aspect, embodiments of the present invention provide a method for detecting chip jamming in CNC machine tool cutting tools, comprising: according to the angle acquisition window corresponding to the current machining process, triggering acquisition when the spindle angle falls into the angle acquisition window during each tool rotation cycle to obtain a vibration snapshot under the process; performing a regularized inverse operation on the vibration snapshot according to a pre-calibrated frequency response function matrix between the tool tip and the sensor mounting point to reconstruct the tool tip time-domain signal; performing continuous wavelet transform on the tool tip time-domain signal to obtain a time spectrum, and sequentially splicing the time spectra of multiple consecutive cycles according to the machining feed direction to obtain a two-dimensional time-frequency-space spectrum; and performing spectrum aggregation... The algorithm divides the two-dimensional time-frequency-spatial spectrum into multiple superpixel regions. It calculates the anomaly score of each superpixel region through an anomaly detection model and identifies superpixel regions with anomaly scores exceeding a preset score threshold as anomaly regions. It also takes the maximum anomaly score among all anomaly regions as the chip jamming index and uses the center feed displacement of the anomaly region corresponding to the maximum anomaly score as the position information. It constructs a multi-source feature vector using the built-in process parameters of the CNC system, the chip jamming index, and the position information, and inputs the multi-source feature vector into the classification decision model to obtain the chip jamming detection result of the machine tool.

[0009] Optionally, the angle acquisition window is preset according to the machining process type: when the current machining process is milling, the corresponding angle acquisition window is 0° to 45°; when the current machining process is drilling, the corresponding angle acquisition window is 10° to 60°; when the current machining process is turning, the corresponding angle acquisition window is 0° to 90°.

[0010] Optionally, based on the pre-calibrated frequency response function matrix between the tool tip and the sensor mounting point, a regularized inverse operation is performed on the vibration snapshot to reconstruct the tool tip time-domain signal. This includes: after obtaining the frequency response function matrix, performing a fast Fourier transform on the vibration snapshot to obtain the corresponding response spectrum; using the Tikhonov regularization method, performing an inverse operation based on the frequency response function matrix and the response spectrum to solve for the equivalent excitation source spectrum at the tool tip; wherein, the Tikhonov regularization method adaptively selects the regularization parameter through the L-curve method to achieve a balance between fitting accuracy and solution smoothness; and performing an inverse Fourier transform on the equivalent excitation source spectrum to obtain the reconstructed tool tip time-domain signal.

[0011] Optionally, the time spectra of multiple consecutive cycles are sequentially spliced ​​in the processing feed direction to obtain a two-dimensional time-frequency-space spectrum, including: updating the two-dimensional time-frequency-space spectrum in real time through a sliding window mechanism; wherein, the sliding window maintains a fixed number of periodic spectra, and each time spectrum of a new cycle is obtained, it is spliced ​​to one side of the spectrum, and the time spectrum of the oldest cycle is removed.

[0012] Optionally, the two-dimensional time-frequency-spatial map is segmented into multiple superpixel regions using a spectral clustering algorithm, including: constructing a similarity matrix based on the spatial distance and grayscale difference between pixels in the two-dimensional time-frequency-spatial map; calculating the first few eigenvectors of the corresponding Laplacian matrix based on the similarity matrix, and clustering the first few eigenvectors to obtain multiple superpixel regions.

[0013] Optionally, the anomaly detection model is an isolated forest model; the anomaly score of each superpixel region in multiple superpixel regions is calculated by the anomaly detection model, including: constructing a feature vector for each superpixel region; wherein, the feature vector of each superpixel region includes at least one of region energy, texture entropy, energy variance, region area and centroid frequency; the feature vector of each superpixel region is used as input to the isolated forest model, which consists of a preset number of isolated trees, and the corresponding anomaly score is obtained by calculating the average path length of each feature vector in all isolated trees.

[0014] Optionally, the CNC system has built-in process parameters including at least one of spindle load rate, feed axis current rate, tool life remaining, and load change rate.

[0015] Optionally, the method further includes: periodically calculating the deviation between the machine tool chip jamming detection results and the actual situation to obtain the false alarm rate and the missed alarm rate; if the false alarm rate exceeds the preset false alarm rate threshold, the preset score threshold is increased by one step, but not exceeding the preset upper limit; if the missed alarm rate exceeds the preset missed alarm rate threshold, the preset score threshold is decreased by one step, but not lower than the preset lower limit; otherwise, the preset score threshold remains unchanged.

[0016] In a second aspect, embodiments of the present invention provide an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0017] The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the method for detecting chip jamming in a CNC machine tool as described in any of the first aspects.

[0018] Thirdly, embodiments of the present invention provide a storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of the method for detecting chip jamming in a CNC machine tool as described in any of the first aspects.

[0019] (III) Beneficial Effects

[0020] The beneficial effects of this invention are:

[0021] This application provides a method, electronic device, and storage medium for detecting chip jamming in CNC machine tool tools. By using an angle acquisition window corresponding to the current machining process, acquisition is triggered when the spindle angle falls within the angle acquisition window during each tool rotation cycle, obtaining a vibration snapshot under that process. Based on a pre-calibrated frequency response function matrix between the tool tip and the sensor mounting point, a regularized inverse operation is performed on the vibration snapshot to reconstruct the tool tip time-domain signal. A continuous wavelet transform is then performed on the tool tip time-domain signal to obtain a time-frequency spectrum. The time-frequency spectra of multiple consecutive cycles are sequentially stitched together according to the machining feed direction to obtain a two-dimensional time-frequency-spatial map. A spectral clustering algorithm is used to segment the two-dimensional time-frequency-spatial map into multiple superpixel regions, and an anomaly score for each superpixel region is calculated using an anomaly detection model. The system identifies superpixel regions with abnormal scores exceeding a preset score threshold as abnormal regions. It also takes the maximum abnormal score among all abnormal regions as the chip jamming index and uses the center feed displacement of the abnormal region corresponding to this maximum abnormal score as position information. Finally, it constructs a multi-source feature vector using the CNC system's built-in process parameters, chip jamming index, and position information. This multi-source feature vector is then input into a classification decision model to obtain the machine tool chip jamming detection result. This application utilizes virtual sensing reconstruction technology based on inverse frequency response function operations to deduce the true excitation signal at the tool tip from the distorted sensor response, thereby clearly revealing the chip jamming features that were originally blurred due to transmission path distortion. This provides a high-quality signal close to the true excitation for subsequent time-frequency analysis and classification decisions, ultimately establishing a stable and reliable chip jamming discrimination standard. Attached Figure Description

[0022] Figure 1 A flowchart illustrating a method for detecting chip jamming in CNC machine tool cutting tools according to an embodiment of this application is shown.

[0023] Figure 2 This paper shows a structural block diagram of another anomaly detection model provided in an embodiment of this application;

[0024] Figure 3 This paper illustrates a multi-source information fusion decision-making flowchart provided in an embodiment of this application. Detailed Implementation

[0025] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] In existing technologies, to capture the high-frequency transient components (typically above 5kHz) generated by chip debris, continuous acquisition at a sampling rate higher than 50kHz is required. The resulting massive amounts of data place enormous pressure on the storage and computing capabilities of embedded systems. On the one hand, edge devices (such as industrial controllers and embedded industrial PCs) have limited computing power and memory, making it difficult to support real-time processing of long-term, high-frequency data streams. On the other hand, uploading the raw data to the cloud for analysis introduces communication latency and bandwidth pressure, making it extremely difficult to implement real-time, complex intelligent diagnostic algorithms at the edge.

[0027] Furthermore, chip jamming is a low-probability event with diverse forms, influenced by factors such as tool geometry, cutting parameters, and material properties. It is difficult to obtain a comprehensive fault sample library covering all working conditions to train a supervised deep learning model. Therefore, existing technologies mostly rely on fixed thresholds or manually experienced thresholds for judgment. However, in actual machining, the background vibration and noise level dynamically changes with tool wear, fine-tuning of cutting parameters, or changes in workpiece material. Such static thresholds are prone to false alarms (threshold too low) or false negatives (threshold too high), lacking adaptive adjustment capabilities.

[0028] Furthermore, existing technologies typically only provide an alarm signal indicating "chip jamming has occurred," without accurately specifying the exact machining location where the jamming occurred (such as feed displacement or spindle angle coordinates). This forces process engineers to spend a significant amount of time troubleshooting and root cause analysis after machine downtime. The lack of spatial positioning information also hinders subsequent optimization and adjustment of process parameters (such as modifying feed paths or adjusting cutting parameters).

[0029] In recent years, although unsupervised learning algorithms such as isolated forests have emerged for fault detection, these methods do not require fault samples and can be directly applied to vibration signals for anomaly identification. However, existing solutions mostly analyze the original vibration signals directly without considering the distortion effect of the signal transmission path, nor are they combined with time-frequency spectrum analysis. As a result, they lack sufficient sensitivity to detect early, weak chip jamming signals, have a high false alarm rate, and cannot meet the high reliability requirements of industrial sites.

[0030] Therefore, existing technologies have significant shortcomings in terms of signal quality, data processing, sample dependence, process traceability, and detection reliability. There is an urgent need for an online chip detection method that can overcome transmission path distortion, compress data volume, eliminate the need for fault samples, possess spatial positioning capabilities, and have a low false alarm rate.

[0031] Based on this, embodiments of this application provide a method, electronic device, and storage medium for online detection of chip jamming in CNC machine tools. By using an angle acquisition window corresponding to the current machining process, acquisition is triggered within each tool rotation cycle when the spindle angle falls into the angle acquisition window, obtaining a vibration snapshot under that process. Then, based on a pre-calibrated frequency response function matrix between the tool tip and the sensor mounting point, a regularized inverse operation is performed on the vibration snapshot to reconstruct the tool tip time-domain signal. A continuous wavelet transform is performed on the tool tip time-domain signal to obtain a time-frequency spectrum. The time-frequency spectra of multiple consecutive cycles are sequentially stitched together according to the machining feed direction to obtain a two-dimensional time-frequency-space map. Finally, a spectral clustering algorithm is used to segment the two-dimensional time-frequency-space map into multiple superpixel regions, and an anomaly detection model is used to calculate the anomalies in each superpixel region. The system calculates a constant score and identifies superpixel regions with abnormal scores exceeding a preset score threshold as abnormal regions. It also takes the maximum abnormal score among all abnormal regions as the chip jamming index and uses the center feed displacement of the abnormal region corresponding to the maximum abnormal score as position information. Finally, it constructs a multi-source feature vector using the CNC system's built-in process parameters, chip jamming index, and position information. This multi-source feature vector is then input into a classification decision model to obtain the machine tool chip jamming detection result. This application utilizes virtual sensing reconstruction technology based on the inverse operation of the frequency response function to deduce the true excitation signal at the tool tip from the distorted sensor response, thereby clearly revealing the chip jamming features that were originally blurred due to transmission path distortion. This provides a high-quality signal close to the true excitation for subsequent time-frequency analysis and classification decisions, ultimately establishing a stable and reliable chip jamming discrimination standard.

[0032] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0033] Please see Figure 1 , Figure 1 A flowchart illustrating a method for detecting chip jamming in CNC machine tool cutting tools according to an embodiment of this application is shown. It should be understood that this method can be executed by an electronic device, and the specific device of the electronic device can be configured according to actual needs; this application embodiment is not limited thereto. For example, the electronic device can be a computer or a server, etc. Specifically, the method includes:

[0034] Step S110: Based on the angle acquisition window corresponding to the current machining process, in each tool rotation cycle, when the spindle angle falls into the angle acquisition window, acquisition is triggered to obtain a vibration snapshot under this process. Here, the vibration snapshot refers to a short-time discrete vibration signal segment obtained by high-speed acquisition when the spindle angle falls into the preset angle acquisition window in each tool rotation cycle.

[0035] It should be understood that the specific angle corresponding to the angle acquisition window can be set according to actual needs, and the embodiments of this application are not limited thereto.

[0036] Optionally, this application pre-sets an angle acquisition window by analyzing the high-incidence phase intervals of chip jamming in different processing techniques:

[0037] When the machining process is milling, chip jamming is likely to occur during the tool entry into the workpiece. Therefore, the angle acquisition window can be set from 0° to 45°.

[0038] When the machining process is drilling, chips are easily generated when the cross-cutting edge contacts the workpiece, so the angle acquisition window can be set to 10° to 60°.

[0039] When the machining process is turning, chip jamming may occur throughout the entire process of constant contact between the tool and the workpiece. Therefore, the angle acquisition window can be set from 0° to 90°.

[0040] In addition, the angle acquisition window can be fine-tuned according to data such as tool geometry, cutting parameters and workpiece material to achieve process adaptation and ensure that the acquisition window covers areas with high chip jamming.

[0041] To facilitate understanding of step S110, it will be described below through specific embodiments.

[0042] Specifically, the system can read the angle signal output by the spindle encoder in real time. Furthermore, within each tool rotation cycle, it is determined whether the current angle falls within a preset window. If it does, the data acquisition card is immediately triggered to acquire multiple data points at a preset acquisition frequency to form a vibration snapshot X_n; if it does not fall within the window, the cycle is skipped, and no acquisition is performed. In this way, only one snapshot is acquired per cycle, avoiding the acquisition and transmission of invalid data. This "stroboscopic" acquisition mechanism transforms a continuous data stream into a sparse, synchronous sequence of snapshots with cutting events, laying the data foundation for subsequent processing. It should be understood that the specific frequency of the preset acquisition frequency can be set according to actual needs, and the embodiments of this application are not limited to this. For example, the preset acquisition frequency can be 65.536kHz, etc.

[0043] For example, to quantitatively assess the advantages of this application in data processing efficiency, a typical three-flute end milling machining condition is used as an example for analysis. The spindle speed is set to 3000 rpm (i.e., 50 revolutions per second), and the feed per tooth is 0.05 mm / tooth.

[0044] For traditional continuous acquisition mode: In order to capture the high-frequency transient components (above 5kHz) generated by debris, the traditional solution requires continuous acquisition of vibration signals at a high sampling rate of 65.536kHz.

[0045] Data volume per second: 65,536 sampling points / second.

[0046] Data volume per revolution: 65536 sampling points / second ÷ 50 revolutions / second = 1310.72 sampling points / revolution.

[0047] In this mode, the system not only collects effective cutting data of the tool entering / exiting the workpiece, but also has to continuously record a large number of invalid vibration signals during the tool's idle period, resulting in serious waste of computing resources and data redundancy.

[0048] For the stroboscopic acquisition mode of this application: This application presets the acquisition window angle according to the processing technology. For milling, the window is set to 0° to 45°, corresponding to an effective acquisition range of 1 / 8 revolution.

[0049] Effective window duration per cycle: (1 / 8 revolution) ÷ 50 revolutions / second = 0.0025 seconds.

[0050] Number of sampling points per cycle: The sampling rate of 65.536kHz is still used during the window period.

[0051] The number of sampling points per cycle is: 65536 × 0.0025 = 163.84 points. In practical applications, this can be rounded to 128 or 256 points depending on hardware conditions. Here, the theoretical value is used to illustrate the principle.

[0052] Data volume per second: 163.84 points / revolution × 50 revolutions / second = 8192 sampling points / second.

[0053] Data compression ratio: 65536 / 8192 = 8:1.

[0054] Therefore, this application reduces the amount of data to be processed per second from 65,536 points to 8,192 points through a stroboscopic acquisition mechanism, a reduction of 87.5%. This level of compression enables real-time data stream processing, which originally required high-performance industrial control computers, to run smoothly on embedded edge devices such as the ARM Cortex-A series, laying the hardware foundation for the subsequent edge deployment of complex algorithms such as virtual sensing reconstruction and continuous wavelet transform.

[0055] Furthermore, since the background vibration data from the non-cutting stage is actively removed, the algorithm does not need to sift through broadband noise to obtain effective information, which is equivalent to indirectly improving the signal-to-noise ratio of the input signal and is beneficial for the extraction of early and weak chip features.

[0056] In addition, the significant reduction in data volume effectively alleviates the storage pressure of long-term online monitoring, while also reducing the demand for data transmission bandwidth. This allows the system to store historical spectral data over a longer period, providing data support for post-fault tracing and process parameter optimization.

[0057] Step S120: Based on the pre-calibrated frequency response function matrix between the blade tip and the sensor mounting point, perform regularized inverse operation on the vibration snapshot to reconstruct the blade tip time domain signal.

[0058] Optionally, after obtaining the frequency response function matrix, a fast Fourier transform is performed on the vibration snapshot to obtain the corresponding response spectrum; the equivalent excitation source spectrum at the tool tip is solved by inverse operation based on the frequency response function matrix and the response spectrum using the Tikhonov regularization method; wherein, the Tikhonov regularization method adaptively selects the regularization parameter through the L-curve method to achieve a balance between fitting accuracy and solution smoothness; the equivalent excitation source spectrum is then subjected to an inverse Fourier transform to obtain the reconstructed tool tip time-domain signal.

[0059] For example, to address the signal transmission path distortion problem, this application introduces virtual sensing technology. After offline testing (such as machine tool assembly or overhaul), the frequency response function matrix H(f) between the tool tip and the physical sensor mounting point can be obtained through a hammer impact experiment. This may require multiple hammer impacts (e.g., 5 times) and averaging to ensure the coherence of the frequency response function measurement is greater than 0.9, thus guaranteeing calibration accuracy. Specifically, a known pulse excitation F_ref(f) is applied to the tool tip using a force hammer, while simultaneously acquiring the sensor position response Y_ref(f). Based on these two data points, the transfer function h(f) = Y_ref(f) / F_ref(f) is calculated to cover the frequency range of 100Hz to 10kHz. Furthermore, a frequency response function matrix H(f) is constructed based on multiple transfer function values. This matrix fully characterizes the distortion characteristics of the mechanical structure to the impact signal (including amplitude attenuation, phase shift, and mode aliasing), serving as the basis for subsequent inverse operations.

[0060] Furthermore, during online operation, a Fast Fourier Transform (FFT) can be performed on the vibration snapshot X_n to obtain the response spectrum Y. n (f). Due to modal anti-resonance and noise amplification in the transmission path, directly inverting the frequency response function matrix H(f) would lead to unstable solutions and a sharp amplification of noise. Therefore, this application uses Tikhonov regularization to solve the knife-tip equivalent excitation source spectrum F. n (f):

[0061] ;

[0062] In the formula, n represents the period; T represents the transpose operation; represents the regularization parameter, used to strike a balance between fitting accuracy and solution smoothness; I represents the identity matrix, used to introduce the regularization term; This represents a regularization term that effectively suppresses high-frequency noise amplification while preserving the key characteristics of chip impact.

[0063] Furthermore, to address variations in signal-to-noise ratio under different operating conditions, this invention employs the L-curve method to determine the optimal signal-to-noise ratio online. Specifically: by traversing the candidates Calculate the residual norm using values ​​(e.g., logarithmic intervals from 10E-8 to 10E-2). reconciliation norm The residual norm values ​​are obtained. Reconciliation Norm Subsequently, with x-axis Using the vertical axis, plot the following... The curve is changing, and this curve is usually "L"-shaped. The point of maximum curvature (inflection point) of the curve is selected. As the optimal regularization parameter. And, this optimal... The value is usually between 10E-6 and 10E-4, which can ensure that the residual is small enough and avoid excessive oscillation of the solution, so as to achieve adaptive adjustment of the signal-to-noise ratio.

[0064] In addition, in the optimal Substituting the values ​​into the calculation formula for the equivalent excitation source spectrum of the cutting edge, we obtain the equivalent excitation source spectrum F of the cutting edge. n (f), and the equivalent excitation source spectrum F of the blade tip n (f) Perform an inverse Fourier transform to obtain the reconstructed time-domain signal x'_n(t) at the tool tip, where t represents time. The above process demonstrates that after reconstruction, the energy in the 5kHz-10kHz frequency band, which was originally weakened by structural filtering, is significantly restored, the signal-to-noise ratio is improved by 8-15dB, and the peak amplitude of chip impact is increased by 2-3 times. This step is equivalent to "virtually migrating" the physical sensor to the tool tip at the software level, providing a high-quality signal close to the real excitation for subsequent feature extraction.

[0065] Step S130: Perform continuous wavelet transform on the tool tip time-domain signal to obtain the time spectrum, and then stitch together the time spectra of multiple consecutive periods according to the machining feed direction to obtain a two-dimensional time-frequency-space spectrum. Here, the machining feed direction refers to the direction of tool movement relative to the workpiece.

[0066] It should be understood that the specific process of performing continuous wavelet transform on the time-domain signal of the knife tip to obtain the time spectrum can be set according to actual needs, and the embodiments of this application are not limited thereto.

[0067] Optionally, in order to capture the broadband transient characteristics generated by the chip debris, a continuous wavelet transform is performed on the reconstructed signal x'_n(t), and a complex Morlet wavelet (such as the center frequency) can be used. It can be configured to level 6 (which can be set according to actual needs), because the complex Morlet wavelet can simultaneously extract the amplitude and phase information of the signal, and its shape is similar to the impact response waveform generated by the chip, with a center frequency of... Choosing 6 can better satisfy the allowable conditions of wavelet transform, achieving a balance between time and frequency resolution.

[0068] Furthermore, compared to the short-time Fourier transform, wavelet transform offers better frequency resolution in the low-frequency range and better time resolution in the high-frequency range, making it highly suitable for analyzing non-stationary, wide-bandwidth transient events such as chip jamming. A continuous wavelet transform of the blade tip time-domain signal yields the time-spectrum. ,in Here, f is the time shift parameter, and f is the frequency.

[0069] It should also be understood that the specific process of stitching together the time spectrum of multiple consecutive cycles in the direction of processing feed to obtain a two-dimensional time-frequency-space spectrum can be set according to actual needs, and the embodiments of this application are not limited thereto.

[0070] Optionally, the two-dimensional time-frequency-spatial spectrum is updated in real time through a sliding window mechanism; wherein, the sliding window maintains a fixed number of periodic spectra, and each time a new period's time spectrum is obtained, it is spliced ​​to one side of the spectrum, and the time spectrum of the oldest period is removed.

[0071] For example, the time-frequency spectra of N consecutive periods can be stitched together in the feed direction to form a two-dimensional spectrum P(x,y), specifically:

[0072] The horizontal axis x corresponds to the feed displacement, which can be obtained by summing the cycle number n, the feed per tooth f_z, and the number of tool teeth z: x = n × f_z × z; the vertical axis y is a logarithmic frequency scale (100Hz-10kHz); the pixel value P(x,y) is the amplitude of the wavelet coefficients normalized to 0-255. The resulting graph visually illustrates the evolution of vibration energy with processing position and frequency, making chip jamming events visible in the spatial dimension.

[0073] Furthermore, to accommodate real-time monitoring requirements, the system maintains a fixed-width spectrum window of size K, the specific value of which can be set according to actual needs. For example, K equals a spectrum window of 500 cycles (corresponding to approximately 100mm feed stroke). Each time a new cycle's spectrum is obtained, it is stitched to the right side of the spectrum, while the data from the leftmost cycle is removed and re-normalized. This sliding update mechanism ensures that the spectrum always reflects the dynamic changes of the most recent processing state, while keeping memory usage within 1024×512 pixels, meeting the real-time processing requirements at the edge.

[0074] Step S140: The two-dimensional time-frequency-spatial map is divided into multiple superpixel regions by a spectral clustering algorithm. An anomaly score is calculated for each superpixel region in the multiple superpixel regions by an anomaly detection model. Superpixel regions with anomaly scores exceeding a preset score threshold are identified as anomaly regions. The maximum anomaly score among all anomaly regions is taken as the chipping index, and the center feed displacement of the anomaly region corresponding to the maximum anomaly score is taken as the position information.

[0075] It should be understood that the specific process of segmenting a two-dimensional time-frequency-spatial map into multiple superpixel regions using a spectral clustering algorithm can be set according to actual needs, and the embodiments of this application are not limited thereto.

[0076] Optionally, a similarity matrix is ​​constructed based on the spatial distance and grayscale difference between pixels in the two-dimensional time-frequency-spatial map; the first few feature vectors of the corresponding Laplacian matrix are calculated based on the similarity matrix, and the feature vectors are clustered to obtain multiple superpixel regions.

[0077] For example, a two-dimensional time-frequency-spatial map can be viewed as a graph structure, with each pixel as a node. A similarity matrix W is constructed, where the elements W... ij This represents the similarity between pixel i and pixel j. Its value takes into account both the spatial distance and grayscale difference between the pixels. The specific calculation formula is as follows:

[0078] ;

[0079] In the formula, exp represents an exponential function with the natural constant e as its base; and This is a weighting coefficient used to balance the influence of spatial proximity and spectral similarity on similarity, and its value can be set according to actual needs, such as by optimizing it through experiments. =0.6, =0.4; This represents the Euclidean distance between pixel i and pixel j. This represents the absolute value of the difference between the gray values ​​of pixel i and pixel j. This represents the median distance between all pixel pairs. Based on this similarity matrix W, the first M eigenvectors of the Laplacian matrix are further solved (M can be 15~25, adaptively adjusted according to the map size). After normalizing the eigenvectors, K-means clustering is performed to obtain M superpixel regions. Thus, this segmentation divides the map into semantic units with similar texture and spatial adjacency, providing physically meaningful regions for subsequent feature extraction.

[0080] It should also be understood that the specific process of calculating the anomaly score of each superpixel region in multiple superpixel regions through the anomaly detection model can also be set according to actual needs, and the embodiments of this application are not limited thereto.

[0081] Optionally, an anomaly score for each superpixel region among multiple superpixel regions is calculated using an anomaly detection model, including: constructing a feature vector for each superpixel region; wherein the feature vector for each superpixel region includes at least one of region energy, texture entropy, energy variance, region area, and centroid frequency.

[0082] For example, due to the significant differences in the dimensions and orders of magnitude of different features (such as energy and area), the feature vectors need to be Z-score normalized before being input into the isolated forest to avoid the influence of dimensions on the outlier detection results. For each superpixel region, at least one of the following five parameters can be extracted to construct the corresponding feature vector:

[0083] Region energy E: It represents the sum of the gray values ​​of all pixels within the region, characterizing the vibration intensity;

[0084] Texture entropy H: It represents the Shannon entropy of the gray-level probability distribution within a region, characterizing the complexity of the texture;

[0085] Energy variance V: It represents the variance of pixel grayscale values ​​within a region, characterizing the degree of energy fluctuation;

[0086] Region area A: It represents the number of pixels contained in the region, and characterizes the size of the anomaly range;

[0087] Centroid frequency C: a weighted average of frequency coordinates with gray values ​​as weights, representing the location of the dominant frequency.

[0088] Furthermore, these parameters characterize the abnormal features of vibration signals within the region from different dimensions, forming the basis for subsequent anomaly detection.

[0089] It should also be understood that the specific model and its structure of the anomaly detection model can be set according to actual needs, and the embodiments of this application are not limited thereto.

[0090] Optionally, when the anomaly detection model is an isolated forest model, the feature vector of each superpixel region is used as the input to the isolated forest model. The isolated forest model consists of a preset number of isolated trees, and the anomaly score is obtained by calculating the average path length of each feature vector in all isolated trees. The specific number of preset trees can be set according to actual needs, and this embodiment is not limited to this.

[0091] For example, the feature vectors of all superpixel regions can be combined into a dataset F. Additionally, an isolated forest model can be constructed, which can contain a preset number of isolated trees (e.g., 100 isolated trees). Each tree randomly selects its feature dimension and split point, recursively partitioning until a height limit is reached. , where m represents the number of superpixels. For each feature vector, calculate its average path length E[h] across all isolated trees to obtain the anomaly score:

[0092] ;

[0093] In the formula, S represents the anomaly score (or outlier score), and s∈[0,1], with a larger value indicating that the region is more likely to be an anomaly; c(m) is the path length normalization constant. This algorithm does not require fault samples and can identify outliers based solely on the distribution characteristics of normal data, achieving zero-sample adaptive detection.

[0094] It should be noted that, in addition to the isolated forest model mentioned above, other models or methods can be used to achieve the same result, and the embodiments of this application are not limited to this.

[0095] Optionally, such as Figure 2 As shown, Figure 2 A structural block diagram of another anomaly detection model provided in an embodiment of this application is shown. For example... Figure 2 As shown, the model includes a local energy comparison screening layer, a temporal transient mutation verification layer, a time-frequency texture consistency test layer, and an evidence accumulation and anomaly score calculation layer connected in sequence.

[0096] In the local energy contrast screening layer, the input to this layer is the feature vectors of all superpixel regions, and the feature vector of each superpixel region contains at least the energy E of that region (i.e., the sum of the gray values ​​of the pixels within that region). This layer calculates the local energy contrast C of each superpixel region. i :

[0097] ;

[0098] In the formula, C i E represents the local energy contrast of the i-th superpixel region, measuring the prominence of this region's energy relative to the average energy of its neighboring regions; iN(i) represents the region energy of the i-th superpixel region; N(i) represents the set of indices (e.g., four-neighbor or eight-neighbor) of all superpixel regions spatially adjacent to the i-th superpixel region. E represents the number of adjacent regions, i.e., the number of elements in set N(i); j This represents the energy of the j-th adjacent superpixel region; This indicates a positive number used to prevent the denominator from being zero.

[0099] Then, this layer selects regions with local energy contrast greater than a preset contrast as candidate set R1, and appends the local energy contrast of each candidate region in candidate set R1 to its feature vector and passes it to the next layer. The specific value of the preset contrast can be set according to actual needs, and this embodiment is not limited to this. For example, the preset contrast can be 1.5.

[0100] Furthermore, for the temporal transient mutation verification layer, the input to this layer is the current periodic feature vector of each candidate region in the candidate set R1 (which contains C i and E i The layer calculates the transient rate of change T for each candidate region, along with the feature vectors (containing at least energy) of regions with the same feed displacement location in the historical cycle. i :

[0101] ;

[0102] In the formula, This indicates the regional energy level for the current cycle; It represents the median of the energy values ​​of the i-th region at the same feed displacement position in the historical cycle (such as usually looking back at the previous 3 cycles), which represents the normal energy reference at that position; This indicates a positive number used to prevent the denominator from being zero.

[0103] Then, this layer selects regions with transient change rates greater than a preset change rate as candidate set R2, and appends the transient change rate of each candidate region in candidate set R2 to its feature vector and passes it to the next layer. The specific value of the preset change rate can be set according to actual needs, and this embodiment is not limited to this. For example, the preset change rate can be 0.5.

[0104] Furthermore, for the time-frequency texture consistency verification layer, the input to this layer is the current periodic feature vector of each region in the candidate set R2, and the feature vectors of its neighboring regions. This layer calculates the frequency band energy distribution entropy for each region. S and the structural similarity index with the neighborhood i This leads to the texture difference degree D. i =sigmoid( )×(1-Si Then, the layer attaches the texture difference to the corresponding feature vector and passes it to the next layer.

[0105] Finally, for the evidence accumulation and anomaly score calculation layer, the input to this layer is the feature vector of each region in the candidate set R2 (which contains C i T i and D i Furthermore, this layer first calculates the contrast excess. and transient excess quantity Then, the anomaly score for each region is obtained through the evidence accumulation function. For regions not in the candidate set R2, their anomaly score is directly set to 0; for regions in the candidate set R2, their anomaly score is calculated using the following formula:

[0106] ;

[0107] In the formula, S i C represents the outlier score of the i-th region; max This indicates that the contrast exceeds the preset upper limit, such as C. max Equals 5; T max This indicates that the transient rate of change exceeds a preset upper limit of the quantity, such as T. max It equals 3.

[0108] Furthermore, after obtaining the abnormal scores, superpixel regions with abnormal scores exceeding a preset score threshold are identified as abnormal regions. The maximum abnormal score among all abnormal regions is taken as the chipping index, and the center feed displacement of the abnormal region corresponding to the maximum abnormal score is used as the position information. The specific value of the preset score threshold can be set according to actual needs. For example, the initial value of the preset score threshold can be set to 0.75.

[0109] For example, with an initial preset score threshold θ, all superpixel regions where S > θ can be marked as anomalous regions, and the location (represented by the feed displacement coordinates of its center pixel) and energy intensity E of each anomalous region can be recorded. Furthermore, the largest S among all anomalous regions is taken as the chipping index CI for the current period, used to quantify the severity of chipping. This step transforms the abstract unsupervised score into an intuitive chipping metric and provides spatial location information for subsequent decision-making.

[0110] Step S150: Construct a multi-source feature vector using the built-in process parameters, chip jamming index, and position information of the CNC system, and input the multi-source feature vector into the classification decision model to obtain the chip jamming detection result of the machine tool.

[0111] It should be understood that the parameters included in the built-in process parameters of the CNC system can be set according to actual needs, and the embodiments of this application are not limited thereto.

[0112] Optionally, the CNC system has built-in process parameters including at least one of spindle load rate, feed axis current rate, tool life remaining, and load change rate.

[0113] Among them, spindle load rate T_load / T_rated: the ratio of actual load to rated load, reflecting the change in cutting force;

[0114] Feed axis current rate I_feed / I_rated: The ratio of feed axis current to rated current, reflecting the feed resistance;

[0115] Remaining tool life T_rul (%): Estimated life based on usage time, used to eliminate abnormal vibrations at the end of tool life;

[0116] Load change rate dT_load / dt (% / s): Short-term rate of change of load, used to detect sudden jamming.

[0117] It should also be understood that the specific process of inputting multi-source feature vectors into the classification decision model to obtain the machine tool chip detection result can also be set according to actual needs, and the embodiments of this application are not limited thereto.

[0118] Optionally, if the classification decision model is a LightGBM classifier, such as Figure 3 As shown, the constructed multi-source feature vector can be input into a pre-trained LightGBM classifier, which outputs machine tool jamming detection results including: category 0 (normal), category 1 (minor jamming), and category 2 (severe jamming). The LightGBM classifier has 100 trees, a depth of 5, and 31 leaf nodes. The jamming severity of category 2 is greater than that of category 1.

[0119] In addition, if historical failure samples are lacking in the initial stage of the system, the following rule-based decision-making can be used as a transition:

[0120] When the chipping index CI > 0.85 and the load rate suddenly increases (> 10% / s), it is judged as severe chipping.

[0121] When 0.75 < chipping index (CI) ≤ 0.85, it is judged as slight chipping; otherwise, it is judged as normal.

[0122] In addition, the "minor jamming" and "severe jamming" cases accumulated in the initial rule-based decision-making process, after confirmation by process engineers, can be fed back into the training set as positive samples to incrementally update the LightGBM model, achieving a smooth transition from unsupervised to supervised learning and iterative performance optimization. As operational data accumulates, the rules can be gradually replaced with supervised models to achieve continuous performance optimization.

[0123] It should be noted that although the above description uses the LightGBM classifier as an example, those skilled in the art should understand that it can be replaced with other classifiers according to actual needs, and the embodiments of this application are not limited thereto.

[0124] Further, see also Figure 3 It can also perform corresponding actions based on the chip detection results of the machine tool cutting tools:

[0125] If everything is normal, the system continues processing and only records the graphs for offline trend analysis.

[0126] If it is a minor chip jam, send a prompt message to the operator (such as "Small chip jam is suspected at the xx mm mark, it is recommended to check"), but do not interrupt the process to avoid frequent downtime affecting production efficiency;

[0127] If there is severe chip jamming, an emergency stop is triggered immediately, the current fault map is saved and the abnormal location x_pos is marked, and the relevant data is stored in the fault knowledge base to provide a basis for subsequent process improvement and model iteration.

[0128] Furthermore, with tool wear, adjustments to cutting parameters, or changes in workpiece material, the normal vibration baseline may drift, and the fixed threshold θ=0.75 may no longer be optimal. Therefore, this application introduces a threshold adaptive mechanism based on false alarm / false negative feedback, specifically:

[0129] Regularly analyze the deviation between the machine tool chip jamming detection results and the actual situation to obtain the false alarm rate and the missed alarm rate;

[0130] If the false alarm rate exceeds the preset false alarm rate threshold, the preset score threshold will be increased by one step, but not exceeding the preset upper limit. The specific values ​​of the preset false alarm rate threshold, the step size, and the preset upper limit can all be set according to actual needs.

[0131] If the false negative rate exceeds the preset false negative rate threshold, the preset score threshold will be lowered by one step, but not lower than the preset lower limit; the specific values ​​of the preset score threshold and the preset lower limit can be set according to actual needs.

[0132] Otherwise, the preset score threshold will remain unchanged.

[0133] For example, the system periodically (e.g., every 24 hours) calculates the false alarm rate R_fp (alarms triggered but no alarm is triggered) and the missed alarm rate R_fn (debris jamming occurred but no alarm was triggered) over the past 24 hours. The thresholds are dynamically adjusted based on the statistical results.

[0134] If R_fp>5%, it indicates that the threshold is too low, which may easily misjudge normal vibration as abnormal. Adjust θ up by 0.05 (not exceeding the upper limit of 0.90).

[0135] If R_fn>3%, it indicates that the threshold is too high and real card debris is easily missed. Adjust θ down by 0.05 (not lower than the lower limit of 0.65).

[0136] Otherwise, the current threshold will remain unchanged.

[0137] Therefore, through this closed-loop adjustment, the detection threshold can adapt to changes in operating conditions, always maintain optimal detection performance, and ensure long-term stability and reliability.

[0138] In summary, by employing the above technical solution, this application has the following advantages:

[0139] First, this application employs a stroboscopic transient snapshot acquisition based on spindle phase synchronization, triggering high-speed acquisition only within the angle window where chipping is most frequent, significantly reducing the data rate. Compared to the massive redundant data resulting from continuous high-frequency sampling in existing technologies, this application enables real-time data streams that previously required high-performance industrial control computers to run smoothly on embedded edge devices, providing hardware feasibility for the edge deployment of subsequent complex algorithms and fundamentally resolving the contradiction between high-frequency sampling and real-time processing.

[0140] Secondly, this application reconstructs the true excitation signal at the tool tip from the distorted sensor response through regularization and adaptive parameter selection. Experiments show that the reconstructed signal exhibits significant high-frequency energy recovery, a markedly improved signal-to-noise ratio, and a substantial increase in the peak amplitude of chip impact, thus fundamentally solving the signal transmission path distortion problem and providing a high-quality signal close to the true excitation for subsequent feature extraction.

[0141] Furthermore, this application constructs a two-dimensional time-frequency-spatial evolution map, stitching together continuous periodic time-frequency spectra along the feed direction, thus visualizing the evolution of vibration energy with processing position and frequency. Through a sliding window update mechanism, high-precision spatial positioning of chip jamming events is achieved, providing process engineers with direct evidence for fault tracing and significantly reducing downtime for troubleshooting.

[0142] Furthermore, this application employs spectral clustering to segment the time-frequency-spatial map into physically meaningful superpixel regions, extracting region energy, texture entropy, energy variance, area, and centroid frequency to construct feature vectors. Then, it utilizes an isolated forest algorithm for unsupervised outlier detection. This algorithm requires no fault samples and can be started and run solely based on the distribution characteristics of normal data, truly achieving zero-shot learning capability. Experimental optimization shows that the outlier threshold achieves a balance between false positive and false negative rates, and the detection sensitivity is significantly better than the scheme that directly applies an isolated forest to the original signal.

[0143] Furthermore, this application constructs a multi-source information fusion decision-making mechanism, combining vibration characteristics (chip jamming index, location information) with built-in process parameters of the CNC system (such as spindle load rate, feed axis current rate, tool life remaining, load change rate) into a multi-source feature vector, which is then input into a classification decision model for three-state decision-making (normal, slight chip jamming, severe chip jamming). The multi-source information corroborates each other, effectively suppressing occasional interference from a single signal source, significantly reducing the false alarm rate, and significantly improving system robustness. Initially, when there are no faulty samples, rule-based decision-making can be used, gradually transitioning to a supervised model as data accumulates, demonstrating good engineering adaptability.

[0144] Furthermore, this application introduces a threshold adaptive mechanism based on false alarm and false negative feedback. It periodically calculates the false alarm and false negative rates of historical alarms and dynamically adjusts the outlier threshold of the unsupervised detection model. Through this closed-loop feedback, the detection threshold can adapt to changes in operating conditions (such as tool wear, cutting parameter adjustments, and workpiece material changes), maintaining optimal detection performance at all times. This significantly reduces manual maintenance costs and ensures the long-term stability and reliability of the system.

[0145] Furthermore, this application organically integrates multiple steps, including stroboscopic acquisition, virtual reconstruction, map construction, unsupervised anomaly extraction, multi-source fusion decision-making, and threshold adaptive updating, to form a complete closed-loop solution for online chip detection that can be deployed in industrial settings. The data flow between each step is clear, and the functional boundaries are well-defined, ensuring both the algorithm's advanced nature and the feasibility of engineering implementation, while also possessing good scalability and maintainability. Compared to the single-step improvements in existing technologies, this application systematically addresses core pain points such as signal quality, data processing, sample dependence, process traceability, and detection reliability, significantly improving the performance and engineering practicality of online chip detection for CNC machine tools.

[0146] It should be understood that the above-described method for detecting chip jamming in CNC machine tool tools is merely exemplary. Those skilled in the art can make various modifications based on the above system, and such modified solutions also fall within the protection scope of this application.

[0147] This application also provides an electronic device, which includes: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory stores a computer program, and when the program is executed by the processor, the processor performs the steps of the above-described method for detecting chip jamming in CNC machine tool tools.

[0148] This application also provides a storage medium storing a computer program executable by an electronic device. When the program is run on the electronic device, it causes the electronic device to perform the steps of the above-described method for detecting chip jamming in CNC machine tool tools.

[0149] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0150] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0151] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0152] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0153] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0154] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

Claims

1. A method for detecting chip jamming in CNC machine tool cutting tools, characterized in that, include: Based on the angle acquisition window corresponding to the current machining process, acquisition is triggered when the spindle angle falls into the angle acquisition window during each tool rotation cycle, thus obtaining a vibration snapshot under this process. Based on the pre-calibrated frequency response function matrix between the blade tip and the sensor mounting point, the vibration snapshot is subjected to a regularized inverse operation to reconstruct the blade tip time domain signal; The time-domain signal of the cutting edge is subjected to continuous wavelet transform to obtain the time spectrum, and the time spectra of multiple consecutive periods are spliced ​​together in the machining feed direction to obtain a two-dimensional time-frequency-space spectrum; The two-dimensional time-frequency-spatial spectrum is divided into multiple superpixel regions by a spectral clustering algorithm, and an anomaly score of each superpixel region is calculated by an anomaly detection model. The superpixel regions whose anomaly scores exceed a preset score threshold are identified as anomaly regions. The maximum anomaly score among all anomaly regions is taken as the chipping index, and the center feed displacement of the anomaly region corresponding to the maximum anomaly score is taken as the position information. A multi-source feature vector is constructed using the built-in process parameters of the CNC system, the chip jamming index, and the position information. The multi-source feature vector is then input into a classification decision model to obtain the chip jamming detection result of the machine tool.

2. The method according to claim 1, characterized in that, The angle acquisition window is preset according to the processing technology type as follows: When the current machining process is milling, the corresponding angle acquisition window is 0° to 45°. When the current machining process is drilling, the corresponding angle acquisition window is 10° to 60°. When the current machining process is turning, the corresponding angle acquisition window is 0° to 90°.

3. The method according to claim 1, characterized in that, The step of performing a regularized inverse operation on the vibration snapshot based on the pre-calibrated frequency response function matrix between the blade tip and the sensor mounting point to reconstruct the blade tip time-domain signal includes: Having obtained the frequency response function matrix, a fast Fourier transform is performed on the vibration snapshot to obtain the corresponding response spectrum; The equivalent excitation source spectrum at the knife tip is solved by inverse operation based on the frequency response function matrix and the response spectrum using the Tikhonov regularization method. The Tikhonov regularization method adaptively selects the regularization parameter through the L-curve method to achieve a balance between fitting accuracy and solution smoothness. The equivalent excitation source spectrum is subjected to inverse Fourier transform to obtain the reconstructed knife-tip time-domain signal.

4. The method according to claim 1, characterized in that, The step of sequentially stitching together the time-frequency spectra of multiple consecutive periods according to the processing feed direction to obtain a two-dimensional time-frequency-space spectrum includes: The two-dimensional time-frequency-spatial spectrum is updated in real time using a sliding window mechanism. The sliding window maintains a fixed number of periodic spectra. Each time a new period's time spectrum is obtained, it is spliced ​​to one side of the spectrum, and the time spectrum of the oldest period is removed.

5. The method according to claim 1, characterized in that, The step of segmenting the two-dimensional time-frequency-spatial map into multiple superpixel regions using a spectral clustering algorithm includes: A similarity matrix is ​​constructed based on the spatial distance and grayscale difference between pixels in the two-dimensional time-frequency-spatial map; The first few feature vectors of the corresponding Laplacian matrix are calculated based on the similarity matrix, and the first few feature vectors are clustered to obtain the multiple superpixel regions.

6. The method according to claim 1, characterized in that, The anomaly detection model is an isolated forest model; the calculation of the anomaly score for each superpixel region in the plurality of superpixel regions using the anomaly detection model includes: Construct a feature vector for each superpixel region; wherein the feature vector for each superpixel region includes at least one of region energy, texture entropy, energy variance, region area, and centroid frequency; The feature vector of each superpixel region is used as the input to the isolated forest model, which consists of a preset number of isolated trees. The anomaly score is obtained by calculating the average path length of each feature vector in all the isolated trees.

7. The method according to claim 1, characterized in that, The built-in process parameters of the CNC system include at least one of spindle load rate, feed axis current rate, tool life remaining, and load change rate.

8. The method according to claim 7, characterized in that, The method further includes: The deviation between the chip jamming detection results of the machine tool tools and the actual situation is analyzed periodically to obtain the false alarm rate and the false alarm rate; If the false alarm rate exceeds the preset false alarm rate threshold, the preset score threshold is increased by one step, but not exceeding the preset upper limit; If the false negative rate exceeds the preset false negative rate threshold, the preset score threshold is lowered by one step, but not lower than the preset lower limit; Otherwise, the preset score threshold remains unchanged.

9. An electronic device, characterized in that, include: The processor, communication interface, memory, and communication bus are connected, with the processor, communication interface, and memory communicating with each other via the communication bus. The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the method for detecting chip jamming in CNC machine tool tools according to any one of claims 1 to 8.

10. A storage medium, characterized in that, It stores a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of the method for detecting chip jamming in a CNC machine tool according to any one of claims 1 to 8.

Images