Acoustic sensing based on-line monitoring method and system for cable extrusion process state

Abstract

The present application belongs to the technical field of extrusion process monitoring, and particularly relates to a cable extrusion process state online monitoring method and system based on acoustic sensing, to solve the technical problem of inaccurate cable extrusion process state monitoring in the prior art. The monitoring method comprises the following steps: S1, when the transient impact factor is greater than a preset threshold, a non-symmetrical window function with the main lobe in front and the side lobe behind is used to window the signal frame; S2, fast Fourier transform is performed on each windowed signal frame to obtain its frequency spectrum; a filter bank is used to filter the frequency spectrum of each signal frame and calculate the log energy of each filter channel; S3, the cepstrum coefficient group with a ratio greater than a selection threshold is selected to form a process state feature vector; the process state feature vector is input into a state classifier to obtain the current cable extrusion process state. The present application eliminates redundancy and noise interference, thereby improving the accuracy of process state monitoring.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of extrusion process monitoring, and particularly relates to an online monitoring method and system for cable extrusion process state based on acoustic sensing. BACKGROUND

[0002] In the cable extrusion process, the stability of the process state plays a decisive role in the uniformity and compactness of the insulation layer. Real-time, accurate online monitoring of the cable extrusion process state helps to ensure the quality of cable products, improve production efficiency, and avoid equipment damage. The sound signals generated by the cable extrusion equipment during operation usually contain two parts: one is the periodic steady-state signal generated by the screw rotation and other components, and the other is the transient impact signal caused by uneven material melting, impurity blockage or mechanical component collision and other abnormal conditions.

[0003] Mel-frequency cepstral coefficient (MFCC) is a common method in sound analysis, which can better obtain the envelope information of the signal and improve the accuracy of state recognition to a certain extent. However, when there is a transient impact in the signal, the symmetrical window function used by MFCC will cause energy leakage and ambiguity of the impact characteristics in the time-frequency spectrum, which is not conducive to the identification of early weak faults.

[0004] In addition, the Mel filter bank of MFCC is designed based on the hearing characteristics of the human ear, with fixed center frequency and bandwidth, which cannot be matched with the inherent physical characteristics of the extruder screw speed and other equipment, and it is also difficult to adaptively adjust according to the signal characteristics, resulting in the loss of key fault feature information. In terms of feature selection, MFCC usually directly uses all the cepstral coefficients or fixed low-order coefficients, which not only increases the computational burden, but also may introduce redundant or noise information that is not sensitive to the working condition, thereby affecting the recognition accuracy of the state classifier. SUMMARY

[0005] The application provides an online monitoring method and system for cable extrusion process state based on acoustic sensing to solve the technical problem of inaccurate monitoring of cable extrusion process state in the prior art.

[0006] In a first aspect, the application provides an online monitoring method for cable extrusion process state based on acoustic sensing, comprising the following steps:

[0007] S1, obtaining a sound signal in the operation of a cable extrusion equipment to obtain an original signal sequence; dividing the original signal sequence into multiple signal frames, and taking the product of the kurtosis value and the skewness value of each signal frame as a transient impact factor; when the transient impact factor is greater than a preset threshold, using a non-symmetrical window function with the main lobe in front and the side lobe behind to window the signal frame; otherwise, using a Hamming window to window the signal frame;

[0008] S2. Perform a Fast Fourier Transform on each windowed signal frame to obtain its spectrum; determine the total number of filters based on the transient impulse factor of the corresponding signal frame, and construct a filter bank, wherein the center frequency of each filter is an integer multiple of the fundamental frequency of the screw speed of the cable extrusion equipment, and the bandwidth is adjusted according to the energy entropy and transient impulse factor of the corresponding signal frame, based on the center frequency; use the filter bank to filter the spectrum of each signal frame and calculate the logarithmic energy of each filter channel;

[0009] S3. Perform discrete cosine transform on the logarithmic energy of each filter channel to obtain cepstral coefficients; based on pre-stored cepstral coefficient samples under normal and various abnormal conditions, calculate the ratio of inter-class divergence to intra-class divergence of each order of cepstral coefficients; correct the benchmark threshold according to the transient impact factor to obtain the selection threshold; select cepstral coefficients with a ratio greater than the selection threshold to form a process state feature vector; input the process state feature vector into the state classifier to obtain the current cable extrusion process state.

[0010] Furthermore, the asymmetric window function is: ,in, Let be the value of the asymmetric window function at the nth sampling point; n is the sampling point index within the signal frame, n = 0, 1, ..., N-1; N is the length of the signal frame. Preset parameters to control the peak position of the window function, and .

[0011] Furthermore, the first is calculated using the following formula. The center frequency of each filter : ;in, The fundamental frequency of the screw speed in the cable extrusion equipment; Let m be the filter number, where m = 1, 2, ..., M, and M is the total number of filters. As a preset base, and .

[0012] Furthermore, the first is calculated using the following formula. The bandwidth of each filter : ;in, For the first The center frequency of each filter, where C is a preset bandwidth scaling constant. The energy entropy of the current signal frame. This is the transient impact factor of the current signal frame.

[0013] Furthermore, the selection threshold is calculated using the following formula. : ;in, This is a preset baseline threshold. The transient impact factor of the current signal frame; The preset impact response coefficient is used to adjust the transient impact factor. For selection threshold The intensity of the influence, with a value range of .

[0014] Secondly, the present invention provides an online monitoring system for the status of cable extrusion process based on acoustic sensing, comprising:

[0015] The windowing module acquires the sound signal during the operation of the cable extrusion equipment to obtain the original signal sequence; it divides the original signal sequence into multiple signal frames and uses the product of the kurtosis and skewness values ​​of each signal frame as the transient impact factor; when the transient impact factor is greater than a preset threshold, an asymmetric window function with the main lobe in front and the side lobes behind is used to window the signal frame; otherwise, a Hamming window is used for windowing.

[0016] The filter construction module performs a Fast Fourier Transform on each windowed signal frame to obtain its spectrum; determines the total number of filters based on the transient impact factor of the corresponding signal frame, and constructs a filter bank, wherein the center frequency of each filter is an integer multiple of the fundamental frequency of the screw speed of the cable extrusion equipment, and the bandwidth is adjusted according to the energy entropy and transient impact factor of the corresponding signal frame, based on the center frequency; the filter bank is used to filter the spectrum of each signal frame and calculate the logarithmic energy of each filter channel;

[0017] The status monitoring module performs discrete cosine transform on the logarithmic energy of each filter channel to obtain cepstral coefficients; based on pre-stored cepstral coefficient samples under normal and various abnormal conditions, it calculates the ratio of inter-class divergence to intra-class divergence of each order of cepstral coefficients; it corrects the benchmark threshold according to the transient impact factor to obtain the selection threshold; it selects cepstral coefficients with a ratio greater than the selection threshold to form a process status feature vector; and it inputs the process status feature vector into the status classifier to obtain the current cable extrusion process status.

[0018] Furthermore, the asymmetric window function is: ,in, Let be the value of the asymmetric window function at the nth sampling point; n is the sampling point index within the signal frame, n = 0, 1, ..., N-1; N is the length of the signal frame. Preset parameters to control the peak position of the window function, and .

[0019] Furthermore, the first is calculated using the following formula. The center frequency of each filter : ;in, The fundamental frequency of the screw speed in the cable extrusion equipment; Let m be the filter number, where m = 1, 2, ..., M, and M is the total number of filters. As a preset base, and .

[0020] Furthermore, the first is calculated using the following formula. The bandwidth of each filter : ;in, For the first The center frequency of each filter, where C is a preset bandwidth scaling constant. The energy entropy of the current signal frame. This is the transient impact factor of the current signal frame.

[0021] Furthermore, the selection threshold is calculated using the following formula. : ;in, This is a preset baseline threshold. The transient impact factor of the current signal frame; The preset impact response coefficient is used to adjust the transient impact factor. For selection threshold The intensity of the influence, with a value range of .

[0022] The beneficial effects are as follows: For transient impacts in signals, the use of asymmetric window functions enables more accurate identification of impact characteristics, suppresses energy leakage and feature ambiguity, and improves sensitivity to early faults. The constructed filter bank's center frequency is correlated with the core parameter of the equipment's screw speed, and the filter bandwidth setting is also combined with the signal frame's own energy entropy and transient impact factor, ensuring that feature extraction matches the monitored object and facilitating the extraction of key information. Only the cepstral coefficients with the highest discriminative power for operating conditions are selected to form the process state vector, eliminating redundancy and noise interference, thereby improving the accuracy of process state monitoring. Attached Figure Description

[0023] Figure 1 This is a flowchart of a method for online monitoring of the status of cable extrusion process based on acoustic sensing;

[0024] Figure 2 Time-domain waveform of the window function;

[0025] Figure 3 This is the discrete spectrum of the signal;

[0026] Figure 4 This is the frequency response diagram of the triangular filter bank. Detailed Implementation

[0027] An embodiment of the online monitoring method for cable extrusion process status based on acoustic sensing provided by the present invention:

[0028] like Figure 1 As shown, the online monitoring method for cable extrusion process status based on acoustic sensing includes the following steps:

[0029] S1. Acquire the sound signal during the operation of the cable extrusion equipment to obtain the original signal sequence; divide the original signal sequence into multiple signal frames, and use the product of the kurtosis value and skewness value of each signal frame as the transient impact factor; when the transient impact factor is greater than the preset threshold, use an asymmetric window function with the main lobe in front and the side lobes behind to window the signal frame; otherwise, use a Hamming window to window.

[0030] During the cable extrusion process, under normal operating conditions, the cable extrusion equipment runs smoothly, emitting a soft sound with a clear periodic pattern. However, when the cable extrusion equipment experiences mechanical failure or the raw materials are abnormal, it is often accompanied by sudden, piercing abnormal sounds. Traditionally, relying on manual monitoring is not only inefficient but also makes continuous and objective monitoring difficult. Therefore, sound sensors are installed at appropriate locations near the cable extrusion equipment to collect sound signals during its operation in real time. The continuous analog signals are then converted from analog to digital to obtain a discrete digital time series, which serves as the original signal sequence.

[0031] In one embodiment, the original signal sequence is divided into multiple signal frames, for example, each signal frame contains 2048 sampling points, with an inter-frame overlap rate of 50%. For each signal frame, its skewness and kurtosis values ​​are calculated, and their product is defined as the transient impulse factor of that signal frame. The transient impulse factor can characterize whether there are short-duration, high-energy impulse components in the signal. Specifically, the skewness value is the ratio of the third central moment of the signal frame to the cube of its standard deviation, and the kurtosis value is the ratio of the fourth central moment of the signal frame to the fourth power of its standard deviation.

[0032] Subsequently, the transient impact factor is compared with a preset threshold. If the transient impact factor is greater than the preset threshold, it indicates that the current signal frame contains a significant transient impact. In this case, an asymmetric window function with the main lobe in front and the side lobes behind is used to window the signal frame to better preserve the time-frequency characteristics at the start of the impact. If the transient impact factor is not greater than the preset threshold, the signal is considered to be in a stable state, and a Hamming window is used for windowing.

[0033] The preset threshold is a set value. When setting the preset threshold, statistical analysis can be performed based on the sound data under historical normal operating conditions: collect sound signals of multiple cable extrusion equipment under stable operating conditions, divide them into frames as described above, and calculate the transient impact factor of each signal frame to obtain its distribution range; take the upper quantile (such as the 95th quantile) or the mean plus several times the standard deviation as the initial preset threshold, and make fine adjustments in combination with the on-site debugging results to ensure that the asymmetric window is rarely triggered under normal operating conditions, and can respond effectively when there is obvious impact anomaly.

[0034] In an optional embodiment, the asymmetric window function is: ,in, Let be the value of the asymmetric window function at the nth sampling point; n is the sampling point index within the signal frame, n=1, 2, ..., N-1; N is the length of the signal frame. Preset parameters to control the peak position of the window function, and .

[0035] The transient impact signal generated in the initial stage of a mechanical failure is characterized by its suddenness and short duration, with its main energy typically concentrated in the initial portion of the signal. Traditional symmetrical window functions (such as Hamming or Heining windows) place the main lobe at the center of the window, applying equal attenuation to both the beginning and end of the signal, thus weakening the crucial initial information of the impact event. In contrast, the asymmetrical window function used in this invention shifts the main lobe, i.e., the peak position, forward, giving higher weight to the early portion of the signal frame, thereby amplifying and preserving the initial characteristics of the fault impact while suppressing subsequent noise and damped oscillations.

[0036] For example, assume the signal frame length N is 2048 sampling points, and the preset parameters... With a value of 0.25, the peak of the window function, calculated using the above formula, occurs near the 512th sampling point, with the first quarter of the window receiving the highest weight. This significantly enhances the initial stage of the signal, facilitating the extraction of early impact components. In contrast, if a symmetrical window function (such as a Hamming window) is used, its peak is located at the 1024th sampling point, smoothing out these crucial early impact information and leading to feature loss, such as... Figure 2 As shown. Furthermore, by further adjusting the preset parameters... For example, reducing it to 0.15 can bring the peak position forward to about the 307th sampling point, thereby achieving a stronger response to earlier transient events.

[0037] Preset parameters The window function's main lobe (i.e., the region of maximum energy) is determined in the time domain: when The smaller the lobe, the closer the main lobe is to the beginning of the signal frame; conversely, The larger the lobe, the more it shifts towards the center. Therefore, choosing an appropriate lobe is crucial. The value is crucial for effectively capturing the initial characteristics of transient impact signals.

[0038] In practical applications, preset parameters The selection of [value] should be optimized based on the typical fault characteristics and signal characteristics of the cable extrusion equipment. Specifically, its value can be determined through the following steps:

[0039] Collect typical fault signal samples: Under the condition of a known mechanical fault in the cable extrusion equipment, collect the fault signals during its operation and divide them into signal frames (frame length) that are the same as those under normal operating conditions. (Consistent in terms of overlap rate, etc.) to ensure consistency in analysis;

[0040] Analyze the starting time distribution of the impact signal: Perform time-domain analysis on the above fault signal frames to identify the starting time of the transient impact event and count its occurrence position in the signal frame. For example, most impacts occur between the first 300 and 600 sampling points.

[0041] Setting the target main lobe location: Based on the statistical results of the impact onset time, set the target interval that the main lobe of the desired window function should cover. For example, if most impacts occur near the 300th sampling point, then the peak value of the desired window function should be located in this region;

[0042] Reverse push Value: Using the formula peak position ≈ The corresponding can be deduced by reverse deduction. Value. For example, when =2048, the expected peak occurs at point 512, then You can select =0.25 as the initial value;

[0043] Verification and optimization: Apply the initial values ​​to the actual monitoring system, and combine them with subsequent spectral analysis methods such as Short Time Fourier Transform (STFT) or Wavelet Transform to observe whether they can effectively enhance the impact characteristics and suppress noise interference; if the early impact is still weakened, the values ​​can be appropriately reduced. This allows the main lobe to move further forward, enhancing its ability to respond to earlier impacts.

[0044] S2. Perform a Fast Fourier Transform on each windowed signal frame to obtain its spectrum; determine the total number of filters based on the transient impulse factor of the corresponding signal frame, and construct a filter bank, wherein the center frequency of each filter is an integer multiple of the fundamental frequency of the screw speed of the cable extrusion equipment, and the bandwidth is adjusted according to the energy entropy and transient impulse factor of the corresponding signal frame, based on the center frequency; use the filter bank to filter the spectrum of each signal frame and calculate the logarithmic energy of each filter channel.

[0045] In an optional embodiment, the first step is calculated using the following formula. The center frequency of each filter : ;in, The fundamental frequency of the screw speed in the cable extrusion equipment; Let m be the filter number, where m = 1, 2, ..., M, and M is the total number of filters. As a preset base, and .

[0046] The sound signals generated by mechanical equipment malfunctions have complex frequency components, including not only integer multiples of the fundamental frequency harmonics but also fault-related modulation sidebands and nonlinear frequency components often appearing in the high-frequency range. Setting the filter center frequency using a nonlinear incremental method allows the filter bank to be densely distributed in the low-frequency range and sparsely distributed in the high-frequency range, thus better reflecting the energy distribution patterns of most faults. Specifically, the fundamental frequency and its main harmonics are concentrated in the low-frequency region, while modulation frequencies or resonance components caused by damage to the bearing outer ring, etc., are distributed over a wider high-frequency range.

[0047] For example, assuming the screw speed of a cable extrusion device is 600 rpm, then its fundamental frequency is... The frequency is 10Hz. The total number of filters M is set to 10, with a preset base. The value is 1.4. Therefore, the center frequency of the first filter is... The first filter is set to 10Hz and is used to monitor the fundamental frequency. The center frequency of the second filter is... The center frequency of the third filter is 14Hz. The center frequency of the first filter is 19.6Hz; the center frequencies of subsequent filters are 27.44Hz, 38.42Hz, etc. Figure 3 This design can effectively identify non-integer multiple modulation frequency components distributed around the fundamental frequency caused by damage to the bearing outer ring, as well as resonant frequency components in the high-frequency band. Without increasing the total number of filters, it can achieve both detailed analysis of the fundamental frequency and major harmonics in the low-frequency range and coverage of broadband information in the high-frequency range.

[0048] Preset base The selection steps are as follows:

[0049] Acquire typical signals: Under normal equipment operation and typical fault conditions, acquire sound signals and analyze their spectrum to determine the frequency bands where fault characteristic frequencies are mainly distributed.

[0050] Define the coverage area: Based on the analysis results, determine the highest frequency that the filter bank needs to cover. And the total number of filters M.

[0051] Calculate candidate values: using the formula Preliminary estimate .generally, The value should be between 1.1 and 1.5. If high resolution in the low-frequency band is required (e.g., dense harmonics), choose a smaller value (e.g., 1.1–1.2); if high-frequency coverage or redundancy needs to be considered, choose a larger value (e.g., 1.3–1.5).

[0052] Verification and Selection: Using different Construct a filter bank and test its feature extraction effect on normal and fault signals. Select values ​​that can clearly distinguish the operating conditions and have obvious key frequency responses as the final parameters.

[0053] In an optional embodiment, the first step is calculated using the following formula. The bandwidth of each filter : ;in, Let be the center frequency of the m-th filter, and C be a preset bandwidth scaling constant. The energy entropy of the current signal frame. This is the transient impact factor of the current signal frame.

[0054] Energy entropy This reflects the uncertainty of the signal in the frequency domain. When the cable extrusion equipment operates smoothly, the signal component is simple, and the energy is concentrated, the energy entropy is low; however, when the signal contains multiple noises or multiple fault characteristics, the energy distribution tends to be dispersed, and the energy entropy increases accordingly. Transient impact factor This factor measures whether there is a significant transient impact component in the signal. When a strong impact event occurs (such as crack propagation or foreign object impact), the value of this factor increases significantly. Combining these two factors to adjust the filter bandwidth enables adaptive optimization under different operating conditions: when the signal is complex but without obvious impact, the bandwidth is appropriately widened to improve the coverage of weak features; when a strong transient impact exists, the bandwidth is automatically narrowed to improve frequency resolution, thereby more accurately locating the impact energy and enhancing the signal-to-noise ratio.

[0055] For example, consider a filter with a center frequency of 200Hz, and assume the bandwidth scaling constant C is 0.2. During normal operation, the signal is stable, and the energy entropy... The transient impact factor is 0.3. The value is 0.5. At this point, the calculated bandwidth is approximately 34.7Hz, which is relatively wide and beneficial for capturing weak fault information under background noise. When an early crack appears in the equipment and generates a weak impact, the energy entropy... It rose to 0.4, while the transient impact factor... Increased to 5. At this point, the new bandwidth is 9.3Hz, such as... Figure 4The bandwidth is significantly narrowed, suppressing background noise interference and improving the detection sensitivity for early, minor faults. This adaptive mechanism achieves optimal feature extraction under different operating conditions, balancing robustness and sensitivity.

[0056] The bandwidth ratio constant C is a key preset parameter used to control the fundamental proportional relationship between the filter bandwidth and the center frequency. Its value directly affects the frequency resolution and coverage capability of the filter bank. The selection of C should be optimized in conjunction with the equipment operating characteristics, signal spectrum structure, and monitoring objectives. Specifically, it can be determined according to the following steps:

[0057] Collect sound signal samples under typical operating conditions: Under normal and stable operation of the cable extrusion equipment, collect multiple sets of sound signals to ensure coverage of different production batches, material types, and speed ranges. Additionally, signals under known fault conditions (such as screw wear, bearing damage, foreign matter contamination, etc.) can be collected for comparison.

[0058] Spectral analysis was performed to determine the main energy distribution range: the acquired normal operating condition signal was windowed, FFT transformed, and averaged to obtain its average amplitude spectrum. The fundamental frequency was observed. The energy concentration region of the main harmonics is recorded, along with the 3dB bandwidth of the main harmonics and the minimum frequency interval between harmonics. The 3dB bandwidth represents the frequency width when the energy drops to half of the peak value.

[0059] Defining the candidate range for C: Since the filter bandwidth needs to be proportional to the center frequency, and it is generally desirable for low-frequency filters to cover the main energy region of a single harmonic, C can be estimated based on the relative bandwidth of a typical harmonic. For example, if the harmonic 3dB bandwidth at 100Hz is approximately 20Hz, then C is initially estimated to be 0.2. Generally, for rotating machinery, the empirical value of C ranges from 0.1 to 0.3.

[0060] S3. Perform discrete cosine transform on the logarithmic energy of each filter channel to obtain cepstral coefficients; based on pre-stored cepstral coefficient samples under normal and various abnormal conditions, calculate the ratio of inter-class divergence to intra-class divergence of each order of cepstral coefficients; correct the benchmark threshold according to the transient impact factor to obtain the selection threshold; select cepstral coefficients with a ratio greater than the selection threshold to form a process state feature vector; input the process state feature vector into the state classifier to obtain the current cable extrusion process state.

[0061] A discrete cosine transform is performed on a one-dimensional vector composed of N logarithmic energy values ​​to obtain N cepstral coefficients. Based on pre-collected cepstral coefficient samples under three operating conditions—normal, material blockage, and bearing wear—the inter-class divergence and intra-class divergence of each cepstral coefficient under different operating conditions are calculated, and the ratio of inter-class divergence to intra-class divergence is further obtained.

[0062] In one embodiment, the baseline threshold is set to 1.2, and the threshold is selected based on the transient impact factor. From the formula Calculation: Iterate through the cepstral coefficients of all orders. If the ratio of inter-class divergence to intra-class divergence of a certain order cepstral coefficient is greater than the currently calculated selection threshold, retain that order cepstral coefficient; otherwise, discard it. Arrange all retained cepstral coefficients in order of order to form a process state feature vector. Finally, input the extracted process state feature vector into a support vector machine classifier that has been pre-trained using a large number of known operating condition samples. By calculating the category to which the newly input process state feature vector belongs, output a label representing the current cable extrusion process state, such as "normal operation".

[0063] In an optional embodiment, the selection threshold is calculated using the following formula. : ;in, This is a preset baseline threshold. The transient impact factor of the current signal frame; The preset impact response coefficient is used to adjust the transient impact factor. For selection threshold The intensity of the influence, with a value range of .

[0064] Transient impact factor The transient impact factor reflects whether there are sudden, non-Gaussian impulse components in the signal. When the equipment is operating smoothly and there are no obvious impulses in the signal, the transient impact factor is... The value is very small, so choose a threshold. Approaching the baseline threshold Maintain standard testing practices. When the cable extrusion equipment malfunctions and generates strong impact vibrations, The value will increase significantly, at which point the selection threshold will be lowered accordingly, improving the detection sensitivity for weak impact events. For example, setting a baseline threshold... The impact response coefficient is 3. =0.5. Under normal operating conditions, if The value is 0.2, so the threshold is selected. The value is approximately 2.73. At this point, the selection threshold decreases slightly, while the detection sensitivity remains relatively stable. If a component experiences an impact... If it rises to 8.0, then With a threshold of 0.6, the selection threshold is significantly reduced, thereby effectively enhancing the ability to identify early and subtle fault characteristics.

[0065] Among them, the impact response coefficient The selection steps are as follows:

[0066] Determine the adjustment target: Controlling transient impact factor For selection threshold The greater the value of the influence intensity, the more important it is to select a threshold when the impact occurs. The more significant the decrease, the more significant the increase in sensitivity.

[0067] Define the candidate range: based on experience, The values ​​are usually selected within the range (0, 1), and common candidate values ​​include 0.1, 0.3, 0.5, 0.7, and 1.0.

[0068] Simulation or field verification: On a dataset of known normal and fault conditions, test different... The fault detection rate and false alarm rate of the system under the given value.

[0069] Balancing sensitivity and stability: Selecting a sensor that can effectively detect early, minor impacts without significantly increasing false alarms. Value. For example, if If the detection rate is high and false alarms are controllable when the value is 0.5, then it is determined as the final parameter.

[0070] An embodiment of the online monitoring system for cable extrusion process status based on acoustic sensing provided by the present invention:

[0071] An online monitoring system for the cable extrusion process based on acoustic sensing includes:

[0072] The windowing module acquires the sound signal during the operation of the cable extrusion equipment to obtain the original signal sequence; it divides the original signal sequence into multiple signal frames and uses the product of the kurtosis and skewness values ​​of each signal frame as the transient impact factor; when the transient impact factor is greater than a preset threshold, an asymmetric window function with the main lobe in front and the side lobes behind is used to window the signal frame; otherwise, a Hamming window is used for windowing.

[0073] The filter construction module performs a Fast Fourier Transform on each windowed signal frame to obtain its spectrum; determines the total number of filters based on the transient impact factor of the corresponding signal frame, and constructs a filter bank, wherein the center frequency of each filter is an integer multiple of the fundamental frequency of the screw speed of the cable extrusion equipment, and the bandwidth is adjusted according to the energy entropy and transient impact factor of the corresponding signal frame, based on the center frequency; the filter bank is used to filter the spectrum of each signal frame and calculate the logarithmic energy of each filter channel;

[0074] The status monitoring module performs discrete cosine transform on the logarithmic energy of each filter channel to obtain cepstral coefficients; based on pre-stored cepstral coefficient samples under normal and various abnormal conditions, it calculates the ratio of inter-class divergence to intra-class divergence of each order of cepstral coefficients; it corrects the benchmark threshold according to the transient impact factor to obtain the selection threshold; it selects cepstral coefficients with a ratio greater than the selection threshold to form a process status feature vector; and it inputs the process status feature vector into the status classifier to obtain the current cable extrusion process status.

[0075] In an optional embodiment, the asymmetric window function is: ,in, Let be the value of the asymmetric window function at the nth sampling point; n is the sampling point index within the signal frame, n=1, 2, ..., N-1; N is the length of the signal frame. Preset parameters to control the peak position of the window function, and .

[0076] In an optional embodiment, the first step is calculated using the following formula. The center frequency of each filter : ;in, The fundamental frequency of the screw speed in the cable extrusion equipment; Let m be the filter number, where m = 1, 2, ..., M, and M is the total number of filters. As a preset base, and .

[0077] In an optional embodiment, the first step is calculated using the following formula. The bandwidth of each filter : ;in, Let be the center frequency of the m-th filter, and C be a preset bandwidth scaling constant. The energy entropy of the current signal frame. This is the transient impact factor of the current signal frame.

[0078] In an optional embodiment, the selection threshold is calculated using the following formula. : ;in, This is a preset baseline threshold. The transient impact factor of the current signal frame; The preset impact response coefficient is used to adjust the transient impact factor. For selection threshold The intensity of the influence, with a value range of .

[0079] In addition, in the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

Claims

1. A method for online monitoring of cable extrusion process status based on acoustic sensing, characterized in that, Includes the following steps: S1. Acquire the sound signal during the operation of the cable extrusion equipment to obtain the original signal sequence; divide the original signal sequence into multiple signal frames, and use the product of the kurtosis value and skewness value of each signal frame as the transient impact factor; when the transient impact factor is greater than the preset threshold, use an asymmetric window function with the main lobe in front and the side lobes behind to window the signal frame; otherwise, use a Hamming window to window. The asymmetric window function is , is the value of the asymmetric window function at the nth sampling point; n is the sampling point index within the signal frame, n=1, 2, ..., N-1; N is the length of the signal frame; Preset parameters to control the peak position of the window function, and ; S2. Perform a Fast Fourier Transform on each windowed signal frame to obtain its spectrum; determine the total number of filters based on the transient impulse factor of the corresponding signal frame, and construct a filter bank, wherein the center frequency of each filter is an integer multiple of the fundamental frequency of the screw speed of the cable extrusion equipment, and the bandwidth is adjusted according to the energy entropy and transient impulse factor of the corresponding signal frame, based on the center frequency; use the filter bank to filter the spectrum of each signal frame and calculate the logarithmic energy of each filter channel; S3. Perform discrete cosine transform on the logarithmic energy of each filter channel to obtain cepstral coefficients; based on pre-stored cepstral coefficient samples under normal and various abnormal conditions, calculate the ratio of inter-class divergence to intra-class divergence of each order of cepstral coefficients. The selection threshold is obtained by correcting the baseline threshold based on the transient impact factor; The cepstral coefficients with ratios greater than the selection threshold are selected to form the process state feature vector; the process state feature vector is input into the state classifier to obtain the current cable extrusion process state.

2. The method for online monitoring of cable extrusion process status based on acoustic sensing according to claim 1, characterized in that, The number is calculated using the following formula. The center frequency of each filter : ;in, The fundamental frequency of the screw speed in the cable extrusion equipment; Let m be the filter number, where m = 1, 2, ..., M, and M is the total number of filters. As a preset base, and .

3. The method for online monitoring of cable extrusion process status based on acoustic sensing according to claim 2, characterized in that, The number is calculated using the following formula. The bandwidth of each filter : ;in, For the first The center frequency of each filter, where C is a preset bandwidth scaling constant. The energy entropy of the current signal frame. This is the transient impact factor of the current signal frame.

4. The method for online monitoring of cable extrusion process status based on acoustic sensing according to claim 3, characterized in that, The selection threshold is calculated using the following formula. : ;in, This is a preset baseline threshold. The transient impact factor of the current signal frame; The preset impact response coefficient is used to adjust the transient impact factor. For selection threshold The intensity of the influence, with a value range of .

5. An online monitoring system for the cable extrusion process status based on acoustic sensing, characterized in that, include: The windowing module acquires the sound signal during the operation of the cable extrusion equipment to obtain the original signal sequence; it divides the original signal sequence into multiple signal frames and uses the product of the kurtosis and skewness values ​​of each signal frame as the transient impact factor; when the transient impact factor is greater than a preset threshold, an asymmetric window function with the main lobe in front and the side lobes behind is used to window the signal frame; otherwise, a Hamming window is used for windowing. The asymmetric window function is , is the value of the asymmetric window function at the nth sampling point; n is the sampling point index within the signal frame, n=1, 2, ..., N-1; N is the length of the signal frame; Preset parameters to control the peak position of the window function, and ; The filter construction module performs a Fast Fourier Transform on each windowed signal frame to obtain its spectrum; determines the total number of filters based on the transient impact factor of the corresponding signal frame, and constructs a filter bank, wherein the center frequency of each filter is an integer multiple of the fundamental frequency of the screw speed of the cable extrusion equipment, and the bandwidth is adjusted according to the energy entropy and transient impact factor of the corresponding signal frame, based on the center frequency; the filter bank is used to filter the spectrum of each signal frame and calculate the logarithmic energy of each filter channel; The status monitoring module performs discrete cosine transform on the logarithmic energy of each filter channel to obtain cepstral coefficients; based on pre-stored cepstral coefficient samples under normal operating conditions and various abnormal operating conditions, it calculates the ratio of inter-class divergence to intra-class divergence of each order of cepstral coefficients. The selection threshold is obtained by correcting the baseline threshold based on the transient impact factor; The cepstral coefficients with ratios greater than the selection threshold are selected to form the process state feature vector; the process state feature vector is input into the state classifier to obtain the current cable extrusion process state.

6. The online monitoring system for cable extrusion process status based on acoustic sensing according to claim 5, characterized in that, The number is calculated using the following formula. The center frequency of each filter : ;in, The fundamental frequency of the screw speed in the cable extrusion equipment; Let m be the filter number, where m = 1, 2, ..., M, and M is the total number of filters. As a preset base, and .

7. The online monitoring system for cable extrusion process status based on acoustic sensing according to claim 6, characterized in that, The number is calculated using the following formula. The bandwidth of each filter : ;in, For the first The center frequency of each filter, where C is a preset bandwidth scaling constant. The energy entropy of the current signal frame. This is the transient impact factor of the current signal frame.

8. The online monitoring system for cable extrusion process status based on acoustic sensing according to claim 7, characterized in that, The selection threshold is calculated using the following formula. : ;in, This is a preset baseline threshold. The transient impact factor of the current signal frame; The preset impact response coefficient is used to adjust the transient impact factor. For selection threshold The intensity of the influence, with a value range of .

Images