Belt conveyor hidden damage identification method based on frequency domain migration of acoustic emission signals

Abstract

The present invention relates to the technical field of nondestructive detection of conveyor belts, and discloses a method for identifying hidden damage of conveyor belts based on frequency domain migration of acoustic emission signals, including: real-time acquisition of acoustic emission signals during conveyor belt operation, time-frequency analysis to construct a time-frequency energy matrix, migration of frequency band division rules of historical working conditions to current signals to form adaptive feature sensitive frequency bands through a domain adaptation algorithm in transfer learning, elimination of frequency band components overlapping with noise by comparing a noise spectrum fingerprint library, and inputting the remaining frequency bands into a lightweight convolutional network to output recognition results. Since the frequency band of acoustic emission signals of hidden damage to conveyor belts will dynamically shift with the evolution of damage and changes in working conditions, this method can perceive the signal energy distribution in real time, and dynamically adjust and focus on the frequency band that best characterizes the damage characteristics, thereby avoiding the problem of incomplete feature extraction and misjudgment caused by frequency band solidification.

Description

Technical Field

[0001] The present invention relates to a method for identifying hidden damage of a conveyor belt based on the frequency-domain migration of acoustic emission signals, belonging to the technical field of non-destructive testing of conveyor belts. Background Art

[0002] As a widely used material conveying device in modern industrial production, the safe and stable operation of a conveyor belt is crucial for ensuring production efficiency. However, due to the influence of long-term operation, material impact, and environmental factors, hidden damages such as microcracks and fiber fractures often occur inside the conveyor belt. These early damages are difficult to directly observe with the naked eye. If not discovered and processed in time, they are extremely likely to develop into serious faults, resulting in production interruption and safety accidents. Therefore, early and accurate identification of the hidden damage of the conveyor belt has important practical significance.

[0003] Currently, using acoustic emission technology to detect damage to conveyor belts has become an important means. Acoustic emission refers to the elastic waves released when materials deform or fracture under stress. By analyzing the characteristics of acoustic emission signals, the damage state of materials can be judged. In the existing technology, the acoustic emission signal identification method for hidden damage of conveyor belts usually focuses on frequency-domain analysis. Its basic principle is to perform time-frequency transformation on the collected acoustic emission signals, such as using methods like wavelet packet decomposition to extract energy or amplitude characteristics within a specific frequency band, and judge whether there is damage based on these characteristics. However, the problems existing in these methods in practical applications are as follows:

[0004] 1. Most of the existing acoustic emission damage identification methods rely on pre-set fixed frequency bands for analysis. However, the materials of conveyor belts are not completely uniform, and the operating environment is complex and changeable. The effective frequency band of the acoustic emission signals generated by hidden damage will dynamically shift with the evolution of the damage and the interference of environmental noise. This leads to the situation that if only the fixed frequency band is analyzed, it may not be able to comprehensively capture the characteristic information of the damage, and even cause misjudgment. 2. The operating environment of the conveyor belt is usually relatively harsh, with various noise interferences such as mechanical vibrations and material impacts. The frequency bands of these noise signals often highly overlap with the frequency bands of hidden damage signals, making it difficult to effectively separate the two using traditional filtering methods. The common practice in the industry is to set thresholds or rely on manual experience to screen signals, but this method has obvious subjectivity and is prone to missing weak early damage signals, resulting in a high false negative rate. 3. The existing damage identification models are usually trained based on single working conditions or laboratory data. However, during the actual operation of the conveyor belt, the working conditions such as load, speed, and environmental temperature will change significantly. These changes will affect the characteristics of acoustic emission signals, resulting in the difficulty of the model trained based on single working conditions to adapt to the changing on-site working conditions, insufficient migration ability of frequency-domain characteristics, and poor generalization performance of the model.

[0005] Therefore, how to overcome the limitations of the fixed frequency-domain features in the prior art, effectively separate noise and damage signals, and improve the adaptability of the model under different working conditions has become the technical problem to be solved by the present invention. Summary of the Invention

[0006] The present invention provides a method for identifying hidden damage of a conveyor belt based on frequency-domain migration of acoustic emission signals, and its main purpose is to solve the problems of incomplete extraction of frequency-domain features, difficult effective separation of noise interference, and poor cross-condition adaptability.

[0007] To achieve the above object, a method for identifying hidden damage of a conveyor belt based on frequency-domain migration of acoustic emission signals provided by the present invention includes the following steps:

[0008] Step 1: Continuously collect acoustic emission signals generated during the operation of the conveyor belt in real time, and perform continuous wavelet time-frequency analysis to construct a time-frequency energy matrix reflecting the distribution of signal energy in the time domain and the frequency domain;

[0009] Step 2: By analyzing the energy density gradient in the time-frequency energy matrix, dynamically identify the frequency band regions where the signal energy changes significantly, and use these frequency band regions as candidate regions for potential hidden damage feature frequency bands to achieve a preliminary tracking of the frequency-domain energy flow of the acoustic emission signals;

[0010] Step 3: Use the frequency band energy entropy and the energy correlation between adjacent frequency bands to perform self-organizing migration on the candidate regions for potential hidden damage feature frequency bands, specifically including dynamically merging adjacent frequency bands with similar energy entropy values, and performing secondary splitting according to the energy correlation between frequency bands to form sensitive frequency bands that can adaptively characterize the hidden damage features under the current working conditions, and achieve dynamic focusing of frequency-domain features;

[0011] Step 4: Construct and dynamically maintain a noise spectrum fingerprint library, which extracts the spectrum features of background noise signals collected under different working conditions and updates them using a dynamic self-evolution mechanism. The dynamic self-evolution mechanism includes, during real-time operation, when it is detected that the similarity between the current environmental noise spectrum and all records in the fingerprint library is lower than a preset threshold, automatically triggering the incremental acquisition of the noise spectrum, screening out stable noise feature segments through the verification of the stability of the frequency band energy distribution, and performing spatio-temporal correlation constraints in combination with the operating state parameters of the conveyor belt to eliminate transient interference noises irrelevant to the current working condition, and finally updating using a lightweight fingerprint fusion strategy;

[0012] Step 5: Compare the similarity between the adaptive feature-sensitive frequency band and the noise spectrum in the noise spectrum fingerprint library to decouple the noise signal and the damage signal. Through spectrum similarity comparison, remove the frequency domain components that highly overlap with the noise from the dynamically divided frequency bands to obtain the remaining frequency bands that only contain potential damage information;

[0013] Step 6, normalize the energy of the remaining frequency bands to eliminate the influence of signal amplitude differences under different working conditions;

[0014] Step 7, input the remaining frequency bands after energy normalization into a pre-trained lightweight convolutional neural network. The lightweight convolutional neural network adapts the frequency-domain features learned under historical working conditions to the current working condition through the domain adaptation algorithm in transfer learning, and outputs the recognition result of whether there is hidden damage on the conveyor belt.

[0015] Preferably, in the self-organizing migration step, the calculation of the frequency band energy entropy is used to evaluate the concentration degree of the energy distribution within the frequency band. Frequency bands with similar energy entropy values have similar damage characteristics in terms of physical meaning.

[0016] Preferably, the spatio-temporal correlation constraint includes performing correlation analysis on the timestamps of the newly added noise segments and the historical noise data, as well as the operating state parameters of the conveyor belt, to distinguish between persistent environmental noise and occasional interference.

[0017] Preferably, the training data of the lightweight convolutional neural network includes artificially simulated hidden damage signals under multiple working conditions, and realizes cross-environment generalization through transfer learning. The domain adaptation algorithm is used to align the feature distribution of the source domain (historical working conditions) to the target domain (current working conditions).

[0018] Preferably, the initial construction of the noise spectrum fingerprint library adopts a spatio-temporal homologous verification mechanism, including synchronously recording the operating state of the conveyor belt, environmental temperature and humidity, and equipment position coordinates, constructing a spatio-temporal correlation matrix, then dividing the collected noise spectra into several subsets according to the spatio-temporal correlation matrix, and only retaining the segments with the standard deviation of the spectral energy distribution within each subset less than a preset threshold, and through cross-modal cross-verification, using the low-frequency features of the vibration sensor to eliminate abnormal noise patterns.

[0019] Preferably, the energy density gradient detection is realized by calculating the difference between the energy values of adjacent frequency bands and judging whether the difference exceeds a preset gradient threshold to identify the energy mutation frequency bands.

[0020] Preferably, the frequency band energy entropy The calculation formula is:

[0021] ,

[0022] where represents the proportion of the energy density of the th frequency point within the frequency band to the total energy density of the frequency band.

[0023] Preferably, the operating state parameters of the conveyor belt include but are not limited to the load size and running speed of the conveyor belt.

[0024] Preferably, the preset threshold is an empirical value range determined based on the historical operation data of different types of conveyor belts and the analysis of the characteristics of AE signals of hidden damages.

[0025] Preferably, the recognition result includes the probability value of whether there is a hidden damage on the conveyor belt. When the probability value is greater than or equal to the set damage probability threshold, it is determined that there is a hidden damage on the conveyor belt, and an alarm signal can be output.

[0026] Compared with the problems described in the background art, the beneficial effects of the present invention are as follows:

[0027] 1. By introducing the dynamic tracking and self-organizing migration mechanism of the frequency-domain energy flow, the limitation of relying on fixed frequency band analysis in the prior art is overcome; since the frequency band of the AE signal of the hidden damage of the conveyor belt will dynamically shift with the damage evolution and working conditions changes, this method can perceive the signal energy distribution in real time, and then dynamically adjust and focus on the frequency band that can best characterize the damage characteristics, thus avoiding the problems of incomplete feature extraction and misjudgment caused by fixed frequency bands, and significantly improving the accuracy and integrity of the recognition of early hidden damages of the conveyor belt.

[0028] 2. Aiming at the problem that the noise interference is serious and the frequency band highly overlaps with the damage signal in the operating environment of the conveyor belt, the method of the present invention constructs a noise spectrum fingerprint library based on dynamic self-evolution and spatio-temporal correlation constraints; this mechanism can effectively decouple the noise components from the dynamically divided feature-sensitive frequency bands, avoiding the loss of useful signals that may be caused by traditional filtering methods, thus significantly enhancing the signal-to-noise ratio of the damage characteristics in a complex industrial background, reducing the subjectivity and missed detection rate of manual experience screening, and realizing a more objective and reliable extraction of damage signals. Description of the Drawings

[0029] Figure 1 It is a flow chart of the method for identifying hidden damages of conveyor belts based on the frequency-domain migration of AE signals of the present invention;

[0030] Figure 2 It is a schematic diagram of the dynamic migration mechanism of the method for identifying hidden damages of conveyor belts based on the frequency-domain migration of AE signals of the present invention;

[0031] Figure 3 It is a flow chart of the self-organizing migration to form a sensitive frequency band of the present invention.

[0032] The realization of the object, functional characteristics and advantages of the present invention will be further described in conjunction with the embodiments and with reference to the drawings. Detailed Embodiments

[0033] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0034] An embodiment of this application provides a method for identifying hidden damage to a conveyor belt based on the frequency-domain migration of acoustic emission signals, including the following steps:

[0035] Step 1: Continuously collect the acoustic emission signals generated during the operation of the conveyor belt in real time, and perform continuous wavelet time-frequency analysis to construct a time-frequency energy matrix reflecting the distribution of signal energy in the time domain and frequency domain;

[0036] Step 2: By analyzing the energy density gradient in the time-frequency energy matrix, dynamically identify the frequency band regions where the signal energy changes significantly, and use these frequency band regions as candidate regions for potential hidden damage feature frequency bands to achieve a preliminary tracking of the frequency-domain energy flow of the acoustic emission signals;

[0037] Step 3: Use the frequency band energy entropy and the energy correlation between adjacent frequency bands to perform self-organizing migration on the candidate regions for potential hidden damage feature frequency bands, specifically including dynamically merging adjacent frequency bands with similar energy entropy values, and performing secondary splitting according to the energy correlation between frequency bands to form sensitive frequency bands that can adaptively characterize the hidden damage features under the current working conditions, realizing the dynamic focusing of frequency-domain features;

[0038] Step 4: Construct and dynamically maintain a noise spectrum fingerprint library, which extracts the spectrum features of background noise signals collected under different working conditions and updates them using a dynamic self-evolution mechanism. The dynamic self-evolution mechanism includes, during real-time operation, when it is detected that the similarity between the current environmental noise spectrum and all records in the fingerprint library is lower than a preset threshold, automatically triggering the incremental acquisition of the noise spectrum, screening out stable noise feature segments through the verification of the stability of the frequency band energy distribution, and performing spatio-temporal correlation constraints in combination with the conveyor belt operation state parameters to eliminate transient interference noises irrelevant to the current working conditions, and finally updating using a lightweight fingerprint fusion strategy;

[0039] Step 5: Compare the similarity between the adaptive feature sensitive frequency band and the noise spectrum in the noise spectrum fingerprint library to decouple the noise signal and the damage signal. Through spectrum similarity comparison, remove the frequency domain components that highly overlap with the noise from the dynamically divided frequency bands to obtain the remaining frequency bands that only contain potential damage information;

[0040] Step 6: Normalize the energy of the remaining frequency bands to eliminate the influence of signal amplitude differences under different working conditions;

[0041] Step 7: Input the remaining frequency bands after energy normalization into a pre-trained lightweight convolutional neural network. The lightweight convolutional neural network adapts the frequency-domain features learned under historical working conditions to the current working conditions through the domain adaptation algorithm in transfer learning, and outputs the recognition result of whether there is hidden damage to the conveyor belt.

[0042] Preferably, in the self-organizing migration step, the calculation of the frequency band energy entropy is used to evaluate the concentration degree of the energy distribution within the frequency band. Frequency bands with similar energy entropy values have similar damage characteristics in terms of physical meaning.

[0043] Preferably, the spatio-temporal correlation constraint includes performing correlation analysis on the timestamps of the newly added noise segments and the historical noise data, as well as the conveyor belt operation state parameters, to distinguish between persistent environmental noise and occasional interference.

[0044] Preferably, the training data of the lightweight convolutional neural network includes artificially simulated hidden damage signals under multiple working conditions, and cross-environment generalization is achieved through transfer learning. The domain adaptation algorithm is used to align the feature distribution of the source domain (historical working condition) to the target domain (current working condition).

[0045] Preferably, the initial construction of the noise spectrum fingerprint library adopts a spatio-temporal homologous verification mechanism, including synchronously recording the conveyor belt operation state, environmental temperature and humidity, and equipment position coordinates, constructing a spatio-temporal correlation matrix, then dividing the collected noise spectra into several subsets according to the spatio-temporal correlation matrix, and only retaining the segments with the standard deviation of the spectral energy distribution within each subset less than a preset threshold, and through cross-modal cross-verification, using the low-frequency characteristics of the vibration sensor to eliminate abnormal noise patterns.

[0046] Preferably, the energy density gradient detection is achieved by calculating the difference between the energy values of adjacent frequency bands and determining whether the difference exceeds a preset gradient threshold to identify the frequency bands with energy mutations.

[0047] Preferably, the frequency band energy entropy The calculation formula is:

[0048] ,

[0049] where represents the proportion of the energy density of the th frequency point within the frequency band to the total energy density of the frequency band.

[0050] Preferably, the operation state parameters of the conveyor belt include but are not limited to the load size and running speed of the conveyor belt.

[0051] Preferably, the preset threshold is an empirical value range determined based on the analysis of the historical operation data of different types of conveyor belts and the characteristics of hidden damage acoustic emission signals.

[0052] Preferably, the recognition result includes the probability value of whether there is hidden damage on the conveyor belt. When the probability value is greater than or equal to the set damage probability threshold, it is determined that there is hidden damage on the conveyor belt, and an alarm signal can be output.

[0053] Embodiment 1: In step 2, the time-frequency energy matrix constructed in step 1 is analyzed to dynamically identify the frequency band regions where the signal energy changes; for each moment in the time-frequency energy matrix, calculate the energy difference between adjacent frequency bands, that is, the energy density gradient. Set a gradient threshold , when the energy difference between adjacent frequency bands is greater than this threshold, it is considered that the energy in this frequency band region has changed significantly, and these frequency band regions with significant energy changes are marked as potential candidate regions for hidden damage feature frequency bands. The setting of the gradient threshold can be based on the statistical analysis of historical sound emission signals of non-damaged and damaged conveyor belts, and select an empirical value range that can effectively distinguish energy mutations; for example, the distribution of energy density gradients under normal working conditions and typical damage conditions can be statistically analyzed, and the gradient value range with the highest discrimination is set as .

[0054] In step 3, self-organizing migration is performed on the potential candidate regions for hidden damage feature frequency bands identified in step 2 to form a sensitive frequency band that can adaptively characterize the hidden damage features under the current working conditions; first, calculate the energy entropy of each frequency band, and its calculation formula is shown as:

[0055] ,

[0056] where represents the proportion of the energy density of the -th frequency point in the frequency band to the total energy density of this frequency band; the energy entropy is used to evaluate the concentration degree of the energy distribution within the frequency band. Adjacent frequency bands with similar energy entropy values have similar damage features in physical meaning. Therefore, adjacent frequency bands with an energy entropy difference less than the preset entropy value threshold can be dynamically merged; the setting of the entropy value threshold can be determined according to the analysis of the energy distribution characteristics of the sound emission signal frequency bands in different damage states. Secondly, in order to more finely focus on the damage features, the energy correlation between adjacent frequency bands also needs to be considered, and calculate the energy correlation coefficient between adjacent frequency bands; if the energy correlation coefficient between adjacent frequency bands is less than the preset correlation threshold , it indicates that the energy change trends of these two adjacent frequency bands are quite different and may contain different damage feature information, and secondary splitting is required. The setting of the correlation threshold can be based on the analysis of the correlation between the frequency bands of sound emission signals of different damage types; by dynamically merging adjacent frequency bands with similar energy entropy values and performing secondary splitting according to the energy correlation between frequency bands, a sensitive frequency band that can adaptively characterize the hidden damage features under the current working conditions is finally formed.

[0057] In step 4, a noise spectrum fingerprint library is constructed and dynamically maintained. This fingerprint library collects background noise signals under different working conditions, extracts their spectral features, and updates them using a dynamic self-evolution mechanism. During real-time operation, when the similarity between the currently detected environmental noise spectrum and all the records in the fingerprint library is lower than a preset similarity threshold , an incremental acquisition of the noise spectrum is automatically triggered. The setting of the similarity threshold can be determined based on the analysis of the spectral features of different environmental noise; the newly acquired noise segments are screened to obtain stable noise feature segments through the verification of the stability of the frequency band energy distribution, and spatio-temporal correlation constraints are combined with the conveyor belt operation state parameters to eliminate transient interference noises irrelevant to the current working condition. Finally, a lightweight fingerprint fusion strategy (e.g., weighted averaging of similar noise spectra) is used to update the fingerprint library.

[0058] In step 5, the similarity between the adaptive feature-sensitive frequency band formed in step 3 and the noise spectra in the noise spectrum fingerprint library is compared. The similarity between the spectrum of the adaptive feature-sensitive frequency band and each noise spectrum in the fingerprint library can be calculated (e.g., by calculating the cosine similarity); for the frequency domain components with a similarity higher than the preset noise similarity threshold , they are considered to belong to the noise signal and are removed from the adaptive feature-sensitive frequency band, so as to obtain the remaining frequency band containing only potential damage information. The setting of the noise similarity threshold can be determined based on the discrimination analysis of the spectral features of the damage signal and the noise signal.

[0059] Embodiment 2: This embodiment combines Figures 1 to 3 to illustrate the method for identifying hidden damage of a conveyor belt based on the frequency domain migration of acoustic emission signals.

[0060] As Figure 1 shown, first, the acoustic emission signals generated during the operation of the conveyor belt are collected in real time by sensors. The collected acoustic emission signals are then sent to the time-frequency analysis module for processing to generate a time-frequency energy matrix. Thereafter, the migration engine receives this time-frequency energy matrix and sends a signal requesting noise features to the noise library; the noise library responds to this request and returns fingerprint data to the migration engine. The migration engine extracts pure frequency band features using these data and transmits these features to the recognition network; the recognition network analyzes the received pure frequency band features and outputs a damage probability, which is used by the migration engine to generate a warning instruction. Finally, the warning instruction is transmitted to the time-frequency analysis module to form a closed-loop monitoring and feedback mechanism.

[0061] As Figure 2As shown, the sensor first sends an acoustic emission signal to the time-frequency analysis module, which generates a time-frequency energy matrix and transfers it to the migration engine. The migration engine then requests noise features from the noise library, and the noise library returns fingerprint data. Based on this data, the migration engine extracts pure frequency band features and sends them to the recognition network. The recognition network analyzes them and outputs the damage probability. The recognition network transmits the damage probability to the alarm, which triggers an early warning and finally sends the warning information to the monitoring center to achieve a more comprehensive monitoring and alarm function.

[0062] As Figure 3 shown, Figure 3 Details the process of self-organizing migration to form sensitive frequency bands. The whole process starts with two parallel steps: adjacent frequency band merging and candidate frequency band input. After adjacent frequency band merging, correlation analysis is carried out, and a judgment is made according to the analysis result: if the correlation is low, the frequency band is split again; otherwise, the merging is maintained. At the same time, after the candidate frequency band is input, energy entropy calculation is carried out, and it is judged whether the entropy values are similar: if the entropy values are similar, the process returns to the adjacent frequency band merging step; otherwise, it remains independent. The final link of the process is the output of sensitive frequency bands, which can adaptively characterize the hidden damage characteristics under the current working conditions.

[0063] Example 3: In step 2, analyze the energy density gradient in the time-frequency energy matrix to dynamically identify the frequency band region where the signal energy changes significantly. For the time-frequency energy matrix obtained through continuous wavelet time-frequency analysis, calculate the energy difference between adjacent frequency bands at each time slice; for example, if at a certain moment , the frequency band has an energy of , then its energy density gradient can be obtained by calculating . Preset a gradient threshold , which is an empirical value obtained by statistical analysis of a large number of acoustic emission signals of historical conveyor belts in different health states; when the calculated energy density gradient is greater than , then mark this frequency band as the frequency band region where the energy changes significantly and include it in the candidate area of potential hidden damage characteristic frequency bands.

[0064] In step 3, perform self-organizing migration on the candidate area of potential hidden damage characteristic frequency bands to form sensitive frequency bands that can adaptively characterize the hidden damage characteristics under the current working conditions. First, calculate the energy entropy of each candidate frequency band , and the calculation formula is:

[0065] ,

[0066] where represents the frequency band the proportion of the energy density at the n-th frequency point in the total energy density of this frequency band, and set an entropy value difference threshold . For two adjacent candidate frequency bands and , if , then merge these two frequency bands; the setting of the entropy value difference threshold is determined based on the analysis of the energy distribution characteristics of the acoustic emission signal frequency bands in different damage states; then, considering the energy correlation between adjacent frequency bands, calculate the energy correlation coefficient and of adjacent frequency bands in the time domain. If is less than the preset correlation threshold , then split the frequency bands and into independent sensitive frequency bands; the setting of the correlation threshold is determined based on the analysis of the correlation between the acoustic emission signal frequency bands of different damage types. Through the above dynamic merging and splitting processes, sensitive frequency bands that can adapt to the hidden damage characteristics under the current working conditions are formed.

[0067] In step 4, construct and dynamically maintain a noise spectrum fingerprint library, which is obtained by collecting background noise signals during the operation of the conveyor belt under different working conditions and extracting their spectral characteristics; when the similarity between the currently monitored ambient noise spectrum and all existing records in the fingerprint library is lower than the preset similarity threshold , the system will automatically trigger new noise spectrum collection, and the similarity can be achieved by calculating the cosine similarity of two spectral vectors. Verify the stability of the frequency band energy distribution of the newly collected noise spectrum segment, screen out the frequency bands with stable energy distribution as noise characteristic segments. At the same time, combine operating state parameters such as the load size and running speed of the conveyor belt to perform spatio-temporal correlation constraints, and eliminate transient interference noises that do not conform to the current working conditions. Finally, use the weighted average method to fuse and update the similar noise spectra in the fingerprint library.

[0068] In step 5, compare the similarity between the adaptive feature sensitive frequency band obtained in step 3 and the noise spectra in the noise spectrum fingerprint library, calculate the cosine similarity between the spectrum of the adaptive feature sensitive frequency band and each noise spectrum in the fingerprint library. For the frequency domain components with a similarity greater than the preset noise similarity threshold , it is considered that this frequency domain component is very likely to originate from noise and is removed from the adaptive feature sensitive frequency band to obtain the remaining frequency bands that only contain potential damage information.

[0069] In step 7, the remaining frequency bands after energy normalization are input into a pre-trained lightweight convolutional neural network. Through the domain adaptation algorithm in transfer learning, the network transfers the frequency-domain feature knowledge learned under various historical working conditions to the current working condition, so as to accurately identify whether there is hidden damage on the conveyor belt and output the corresponding recognition result, such as the probability value of the existence of hidden damage on the conveyor belt.

[0070] Embodiment 4: In this embodiment, frequency-domain migration is a process of dynamically adjusting and focusing on the sensitive frequency bands representing damage characteristics by using the frequency-band division experience under historical working conditions and combining the characteristics of the current signal itself. This process may include, for example, the following two core elements, such as experience reference: extracting the rules or patterns of frequency-band division from historical working condition data as the initial migration basis; dynamic self-adaptation: on this basis, according to the characteristics such as the energy distribution, energy entropy, and frequency-band correlation of the current signal, perform self-organizing migration, that is, dynamically merge, split, and adjust the frequency bands, and finally form sensitive frequency bands adapted to the current working condition.

[0071] The update of the fingerprint database is a complex process involving multiple links, and its core goal is to ensure the accuracy, comprehensiveness, and real-time nature of the fingerprint database. Specifically, the update mechanism may include the following steps, for example, incremental acquisition trigger: when the similarity between the currently monitored ambient noise spectrum and all records in the fingerprint database is lower than the preset threshold, trigger the acquisition of a new noise spectrum; stability verification: verify the stability of the frequency-band energy distribution of the newly acquired noise segment, and screen out stable noise feature segments to reduce the influence of random noise; spatio-temporal correlation constraint: combine the operating state parameters such as the load size, running speed, ambient temperature and humidity of the conveyor belt to perform spatio-temporal correlation constraint, and eliminate transient interference noise irrelevant to the current working condition; fingerprint fusion: fuse similar noise spectra in the fingerprint database to reduce redundancy and improve the retrieval efficiency. The specific method of fusion can be weighted average, and the weights can be determined according to factors such as the occurrence frequency and duration of the noise spectrum, which all belong to the extended implementation methods known to those of ordinary skill in the art.

[0072] and in the frequency-band energy entropy In the calculation formula, energy entropy is used to evaluate the concentration degree of energy distribution within a frequency band. The lower the energy entropy, the more concentrated the energy distribution; the higher the energy entropy, the more dispersed the energy distribution. In the process of frequency-band merging, since adjacent frequency bands with similar energy entropy values have similar energy distribution characteristics and thus contain similar damage characteristics, these frequency bands can be merged to simplify subsequent processing and improve the efficiency of feature extraction. The specific merging strategy is as follows: calculate the energy entropy of each candidate frequency band , set an entropy difference threshold ; for two adjacent candidate frequency bands and , if , then these two frequency bands are merged. Here, the entropy value difference threshold is crucial, which determines the looseness of the merger. is smaller, the merger is more strict, and only the frequency bands with very close energy entropy will be merged, which can retain more detailed information but may also increase the complexity of subsequent processing. is larger, the merger is more loose, and more frequency bands will be merged, which can simplify subsequent processing but may also lose some detailed information. The specific value needs to be determined according to the actual application scenario and prior knowledge of the damage characteristics, and the best value can be selected through experimental analysis and parameter optimization. And the energy density gradient threshold : Used to identify the frequency band region where the energy changes significantly, and its value can be based on the statistical analysis of historical undamaged and damaged conveyor belt acoustic emission signals. For example, the probability distribution of the energy density gradient under normal conditions and typical damage conditions can be calculated, and the gradient value range that can most effectively distinguish these two distributions is selected as . The entropy value threshold is used for frequency band merging, and its value can be based on the analysis of the frequency band energy distribution characteristics of acoustic emission signals in different damage states. For example, the statistical characteristics of the frequency band energy entropy under different damage degrees can be calculated, and the entropy value difference range that can effectively distinguish different damage degrees is selected as ; The correlation threshold : This threshold is used for frequency band splitting, and its value can be based on the analysis of the correlation between frequency bands of acoustic emission signals of different damage types. For example, the statistical characteristics of the energy correlation coefficient of adjacent frequency bands under different damage types can be calculated, and the correlation coefficient range that can effectively distinguish different damage types is selected as ; The noise similarity threshold : Used for noise separation, and its value can be based on the analysis of the discrimination degree between the spectral characteristics of damage signals and noise signals. For example, the similarity distribution of the spectra of damage signals and noise signals can be calculated, and the similarity threshold that can most effectively distinguish these two signals is selected as , all belong to the extended implementation methods known to those of ordinary skill in the art.

[0073] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for identifying hidden damage of conveyor belts based on frequency domain migration of acoustic emission signals, characterized in that: The following steps are involved: Step 1: collect the acoustic emission signal generated by the conveyor belt in real time, perform continuous wavelet time-frequency analysis, and construct a time-frequency energy matrix reflecting the distribution of signal energy in the time domain and frequency domain; Step 2, by analyzing the energy density gradient in the time-frequency energy matrix, dynamically identifying the frequency band regions where the signal energy changes significantly, and taking these frequency band regions as potential candidate frequency bands of hidden damage characteristics, to achieve preliminary tracking of the frequency domain energy flow of the acoustic emission signal; Step 3, using the frequency band energy entropy and the energy correlation between adjacent frequency bands, self-organizing migration is performed on the candidate frequency bands of the potential hidden damage characteristics, specifically including dynamically merging adjacent frequency bands with similar energy entropy values, and performing secondary splitting according to the energy correlation between frequency bands, so as to form a sensitive frequency band that can adaptively characterize the hidden damage characteristics under the current working conditions; Step 4: construct and dynamically maintain a noise spectrum fingerprint library. The noise spectrum fingerprint library collects background noise signals under different working conditions, extracts its spectrum features and updates it using a dynamic self-evolution mechanism. The dynamic self-evolution mechanism includes automatically triggering incremental acquisition of the noise spectrum in real-time operation when it is detected that the similarity between the current environmental noise spectrum and all records in the fingerprint library is lower than a preset threshold. Stable noise feature fragments are screened out through the stability verification of frequency band energy distribution, and the spatiotemporal correlation constraints are combined with the conveyor belt operation state parameters to eliminate transient interference noise irrelevant to the current working condition. Finally, a lightweight fingerprint fusion strategy is used for updating. Step 5, performing a similarity comparison between the adaptive feature sensitive frequency band and the noise spectrum in the noise spectrum fingerprint library to decouple the noise signal and the damage signal, and removing the frequency domain components that highly overlap with the noise from the dynamically divided frequency bands through spectrum similarity comparison to obtain the remaining frequency bands that only contain potential damage information; Step 6, normalizing the energy of the remaining frequency bands to eliminate the influence of signal amplitude differences under different working conditions; Step 7, input the remaining frequency band after energy normalization processing into a pre-trained lightweight convolutional neural network, and the lightweight convolutional neural network adapts the frequency domain features learned under historical working conditions to the current working conditions through the domain adaptation algorithm in transfer learning, and outputs the identification result of whether the conveyor belt has hidden damage.

2. The method for identifying hidden damage of a conveyor belt based on frequency domain migration of acoustic emission signals according to claim 1 is characterized in that: In the self-organization migration step, the calculation of the frequency band energy entropy is used to evaluate the concentration of energy distribution in the frequency band. Frequency bands with similar energy entropy values ​​have similar damage characteristics in a physical sense.

3. The method for identifying hidden damage of a conveyor belt based on frequency domain migration of acoustic emission signals according to claim 1 is characterized in that: The spatiotemporal correlation constraints include correlating the newly added noise segments with the timestamps of historical noise data and the conveyor belt operation status parameters to distinguish between continuous environmental noise and occasional interference.

4. The method for identifying hidden damage of a conveyor belt based on frequency domain migration of acoustic emission signals according to claim 1 is characterized in that: The training data of the lightweight convolutional neural network contains artificially simulated hidden damage signals under multiple working conditions, and cross-environment generalization is achieved through transfer learning. The domain adaptation algorithm is used to align the feature distribution of the source domain to the target domain.

5. The method for identifying hidden damage of a conveyor belt based on frequency domain migration of acoustic emission signals according to claim 1 is characterized in that: The initial construction of the noise spectrum fingerprint library adopts a spatiotemporal homology verification mechanism, including synchronously recording the operating status of the conveyor belt, ambient temperature and humidity, and equipment location coordinates, constructing a spatiotemporal correlation matrix, and then dividing the collected noise spectrum into several subsets according to the spatiotemporal correlation matrix, and only retaining the fragments in each subset whose standard deviation of the spectrum energy distribution is less than the preset threshold. Through cross-modal cross-validation, the low-frequency characteristics of the vibration sensor are used to eliminate abnormal noise patterns.

6. The method for identifying hidden damage of a conveyor belt based on frequency domain migration of acoustic emission signals according to claim 1 is characterized in that: Energy density gradient detection recognizes frequency bands with energy mutations by calculating the difference in energy values ​​between adjacent frequency bands and determining whether the difference exceeds a preset gradient threshold.

7. The method for identifying hidden damage of a conveyor belt based on frequency domain migration of acoustic emission signals according to claim 1 is characterized in that: Frequency band energy entropy The calculation formula is: ， in, Indicates frequency band Neidi The ratio of the energy density of a frequency point to the total energy density of the frequency band.

8. The method for identifying hidden damage of a conveyor belt based on frequency domain migration of acoustic emission signals according to claim 3 is characterized in that: The operating state parameters of the conveyor belt include but are not limited to the load size and operating speed of the conveyor belt.

9. The method for identifying hidden damage of a conveyor belt based on frequency domain migration of acoustic emission signals according to claim 5 is characterized in that: The preset threshold is an empirical value range determined based on historical operating data of different types of conveyor belts and characteristic analysis of acoustic emission signals of hidden damage.

10. The method for identifying hidden damage of a conveyor belt based on frequency domain migration of acoustic emission signals according to claim 1, characterized in that: The identification result includes a probability value of whether the conveyor belt has hidden damage. When the probability value is greater than or equal to a set damage probability threshold, it is determined that the conveyor belt has hidden damage and an alarm signal can be output.

Images