Abnormal voiceprint detection system and method based on pump characteristic frequency spectrum

Abstract

The invention relates to the technical field of voiceprint detection, and discloses an abnormal voiceprint detection system and method based on a pump characteristic spectrum, and the system comprises a collection module which sets a sampling rate, generates a voiceprint index for a pump according to an analysis result, and determines an abnormal signal; the processing module recognizes the voiceprint indexes, obtains the characteristic frequency spectrum of the pump machine based on the processing result, determines the abnormal position of the pump machine, and determines the abnormal type of the pump machine according to the abnormal voiceprint characteristics; and the verification module compares the abnormal voiceprint characteristics with historical voiceprint data, determines the similarity of abnormal voiceprints, re-determines the abnormal type of the pump machine by adopting a clustering algorithm when the abnormal type is judged to be inaccurate, and determines the abnormal prediction type of the pump machine based on a convolutional neural network model when the abnormal position is located on the wall surface of the pump machine. Verifying whether the exception type is accurate according to the exception prediction type; and the early warning module determines a graded alarm according to the abnormal position and the verified abnormal type. The method has the characteristics of good detection and analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of voiceprint detection, and in particular to an abnormal voiceprint detection system and method based on a pump characteristic spectrum. Background Art

[0002] With the development of science and technology, there are certain challenges for pipeline transportation. Pipeline transportation is achieved through pumps. The normal operation of the pump is an important factor in ensuring the safe transportation of the pipeline. In recent years, the phenomenon of oil pipeline shutdown due to abnormal shutdown of the pump has occurred frequently, which has brought negative impact on the safe operation of the pipeline.

[0003] Traditional pump abnormality detection mainly relies on manual inspections and regular maintenance, which is not only time-consuming and labor-intensive, but also affected by human experience. In the process of using voiceprints to detect the abnormal state of the pump, the abnormal voiceprint features cannot be effectively extracted, resulting in the inability to process and comprehensively analyze the abnormal voiceprint features, and the inability to effectively judge the abnormal state of the pump, thus affecting the timeliness and accuracy of the handling of pump abnormalities.

[0004] Therefore, how to provide an abnormal voiceprint detection system and method based on the pump characteristic spectrum is a technical problem that technicians in this field need to solve urgently. Summary of the invention

[0005] In view of this, the present invention proposes an abnormal voiceprint detection system and method based on a pump characteristic spectrum, aiming to solve the problem that abnormal voiceprint features cannot be effectively extracted, resulting in the inability to process and comprehensively analyze the abnormal voiceprint features, and the inability to effectively judge the abnormal state of the pump, thereby affecting the timeliness and accuracy of the abnormal processing of the pump.

[0006] In one aspect, the present invention proposes an abnormal voiceprint detection system based on a pump characteristic spectrum, comprising: A collection module, used to set a sampling rate, obtain sound signals of multiple collection points of the pump based on the sampling rate, analyze the sound signals, generate a soundprint index for the pump according to the analysis result and determine an abnormal signal, wherein the soundprint index includes a first index, a second index and a third index; a processing module, for identifying the voiceprint indicator, and when the second indicator or the third indicator is identified, processing the abnormal signal, obtaining a characteristic spectrum of the pump based on the processing result and determining the abnormal position of the pump, extracting the abnormal voiceprint feature of the characteristic spectrum, and determining the abnormal type of the pump according to the abnormal voiceprint feature; a verification module, for verifying the abnormality type according to the abnormal position; when the abnormal position is located inside the pump, comparing the abnormal voiceprint feature with the historical voiceprint data to determine the abnormal voiceprint similarity; judging whether the abnormality type is accurate according to the abnormal voiceprint similarity; when judging that the abnormality type is inaccurate, re-determining the abnormality type of the pump by using a clustering algorithm; when the abnormal position is located on the wall of the pump, determining the abnormality prediction type of the pump based on a convolutional neural network model; verifying whether the abnormality type is accurate according to the abnormality prediction type; and re-determining the abnormality type of the pump based on the abnormality prediction type; The early warning module is used to determine a graded alarm according to the abnormal location and the verified abnormal type.

[0007] Furthermore, when the sound signal is analyzed, a soundprint index is generated for the pump according to the analysis result and an abnormal signal is determined, including: The acquisition module acquires the standard sound waveform corresponding to the sound signals of all acquisition points, and determines the sound waveform corresponding to the sound signals of all acquisition points; When the sound waveforms are all equal to the standard sound waveform, generating the first indicator for the pump; When the sound waveforms are not equal to the standard sound waveform, the second index is generated for the pump, and the sound signal of any collection point is determined as an abnormal signal; When there are one or more sound waveforms that are not equal to the standard sound waveform, generating the third indicator for the pump, and determining whether the sound waveform that is not equal to the standard sound waveform is unique; When it is determined that the sound waveform that is not equal to the standard sound waveform is unique, any sound signal corresponding to the unique sound waveform is determined as an abnormal signal; When it is determined that the sound waveform that is not equal to the standard sound waveform is not unique, any sound signal corresponding to the maximum peak value of the sound waveform is determined as an abnormal signal.

[0008] Further, when the second indicator or the third indicator is identified, the abnormal signal is processed, and based on the processing result, the characteristic spectrum of the pump is obtained and the abnormal position of the pump is determined, including: When the processing module identifies that the second indicator or the third indicator exists, retaining valid information of the abnormal signal based on the first signal algorithm and removing noise of the abnormal signal; Based on the result of noise removal, the abnormal position of the pump is determined according to the time delay difference of the abnormal signal, the abnormal signal is decomposed by a sine wave function based on the second signal algorithm, and the frequency data of the abnormal signal and the amplitude data of the abnormal signal are extracted, the abnormal signal is converted from the time domain to the frequency domain, and a characteristic spectrum of the pump is established according to the frequency data and the amplitude data, wherein the horizontal axis of the characteristic spectrum is represented by frequency, and the vertical axis of the characteristic spectrum is represented by amplitude.

[0009] Furthermore, when extracting the abnormal voiceprint features of the characteristic spectrum and determining the abnormal type of the pump according to the abnormal voiceprint features, it includes: The processing module pre-sets a first preset frequency and a second preset frequency, and extracts an amplitude peak value of the characteristic spectrum; When the amplitude peak value appears at the first preset frequency, the processing module determines the abnormal type of the pump as oxidation; When the amplitude peak value appears at the second preset frequency, the processing module determines the abnormal type of the pump machine as expansion; When the amplitude peaks appear at both the first preset frequency and the second preset frequency, the processing module determines the abnormality type of the pump as fracture.

[0010] Furthermore, when comparing the abnormal voiceprint features with the historical voiceprint data to determine the abnormal voiceprint similarity, it includes: The verification module determines, based on the historical voiceprint data, a historical voiceprint feature of the same abnormal type as the abnormal voiceprint feature, wherein the historical voiceprint feature includes a historical amplitude peak value, a historical first preset frequency, a historical second preset frequency, and a historical amplitude peak number; The abnormal voiceprint similarity is obtained by the following formula: ; in, Indicates the abnormal voiceprint similarity, , , and represents the weight coefficient, and , represents the peak amplitude, Indicates the historical amplitude peak value, Indicates the first preset frequency, Indicates the first preset frequency in history, Indicates the second preset frequency, Indicates the second preset frequency in history, Indicates the number of amplitude peaks, Indicates the number of historical amplitude peaks.

[0011] Further, when judging whether the abnormal type is accurate according to the abnormal voiceprint similarity, it includes: The verification module pre-sets a similarity threshold, and compares the abnormal voiceprint similarity with the similarity threshold, and determines whether the abnormal type is accurate based on the comparison result; When the abnormal voiceprint similarity is greater than or equal to the similarity threshold, the abnormal type is determined to be accurate; When the abnormal voiceprint similarity is less than the similarity threshold, it is determined that the abnormal type is inaccurate.

[0012] Furthermore, when it is determined that the abnormality type is inaccurate, a clustering algorithm is used to re-determine the abnormality type of the pump, including: According to the historical voiceprint data, representative data of each abnormal type is obtained, and the representative data and the abnormal voiceprint features are combined to establish an aggregated data set. The number of clusters k is set to 3, and the feature vector of each data in the aggregated data set is extracted and important features are determined. The parameters of the Gaussian distribution are initialized, and the probability that each data in the aggregated data set belongs to each Gaussian distribution is calculated to obtain the responsibility value, and the cluster with the largest responsibility value is taken as the abnormal type of the data.

[0013] Further, when the abnormal position is located on the wall of the pump, the abnormal prediction type of the pump is determined based on the convolutional neural network model, and whether the abnormal type is accurate is verified according to the abnormal prediction type, and the abnormal type of the pump is re-determined based on the abnormal prediction type, including: The verification module sets the historical voiceprint data as a model training set, uses grid search to find the establishment parameters of the convolution layer, pooling layer, fully connected layer and output layer of the convolutional neural network model, obtains the convolutional neural network model based on cross-validation, and uses the model training set to fit the convolutional neural network model; When the abnormal position is located on the wall of the pump, the abnormal soundprint feature is input into the convolutional neural network model to obtain the abnormal prediction type of the pump, and the abnormal type is compared with the abnormal prediction type, and whether the abnormal type is accurate is determined according to the comparison result; When the abnormality type is equal to the abnormality prediction type, it is determined that the abnormality type is accurate; When the abnormality type is not equal to the abnormality prediction type, it is determined that the abnormality type is inaccurate, and the abnormality type is deleted, and the abnormality prediction type is used as the abnormality type of the pump.

[0014] Further, when determining the graded alarm according to the abnormal location and the verified abnormal type, it includes: The early warning module issues a first-level alarm according to the abnormal position being located on the wall of the pump and the abnormal type of the pump being oxidation or expansion; According to the abnormal position being located inside the pump machine and the abnormal type of the pump machine being oxidation or expansion, a secondary alarm is issued; According to the abnormal position being located inside or on the wall of the pump machine, and the abnormal type of the pump machine being fracture, a third-level alarm is issued; The urgency levels of the first level alarm, the second level alarm, and the third level alarm are increased in sequence.

[0015] Compared with the prior art, the beneficial effect of the present invention is that by setting the sampling rate, the sound signal of the pump can be fully acquired from multiple collection points, accurate soundprint indicators can be generated, and abnormal signals can be accurately identified. The abnormal position of the pump body is obtained based on the abnormal signal, which ensures the accuracy and comprehensiveness of the detection. By analyzing the characteristic spectrum and extracting the abnormal soundprint features, the reliability of the abnormal soundprint feature analysis and comprehensive processing is improved, and then the abnormal type of the pump is preliminarily judged, the accuracy of the judgment is improved, and the dependence on human experience is avoided. Based on the comparison with the historical soundprint data, the accuracy of the abnormal type can be verified according to the abnormal position. When the abnormal position of the pump is located inside, the accuracy of the abnormal type is judged by comparing with the historical soundprint data. When the abnormal position of the pump is located on the wall, the convolutional neural network model is used to obtain the abnormal prediction type and compare it with the abnormal type, which further improves the accuracy of the abnormal type and ensures the reliability of the detection. The graded alarm is automatically generated based on the abnormal position and the verified abnormal type. It ensures that personnel can respond in a timely and effective manner, reduces the downtime of the pump, and improves the safety management of the pump.

[0016] On the other hand, the present application also provides an abnormal voiceprint detection method based on a pump characteristic spectrum, which is used to apply the above-mentioned abnormal voiceprint detection system based on a pump characteristic spectrum, including: Setting a sampling rate, acquiring sound signals of multiple collection points of a pump machine based on the sampling rate, analyzing the sound signals, generating a soundprint index for the pump machine according to the analysis result and determining an abnormal signal, wherein the soundprint index includes a first index, a second index and a third index; Identify the voiceprint indicator, and when the second indicator or the third indicator is identified, process the abnormal signal, obtain the characteristic spectrum of the pump based on the processing result and determine the abnormal position of the pump, extract the abnormal voiceprint feature of the characteristic spectrum, and determine the abnormal type of the pump according to the abnormal voiceprint feature; The abnormality type is verified according to the abnormal position. When the abnormal position is located inside the pump, the abnormal voiceprint feature is compared with the historical voiceprint data to determine the abnormal voiceprint similarity. Whether the abnormality type is accurate is determined according to the abnormal voiceprint similarity. When the abnormality type is determined to be inaccurate, the abnormality type of the pump is re-determined using a clustering algorithm. When the abnormal position is located on the wall of the pump, the abnormality prediction type of the pump is determined based on a convolutional neural network model. Whether the abnormality type is accurate is verified according to the abnormality prediction type, and the abnormality type of the pump is re-determined based on the abnormality prediction type. A graded alarm is determined according to the abnormal location and the verified abnormal type.

[0017] It can be understood that the above-mentioned abnormal voiceprint detection system and method based on pump characteristic spectrum have the same beneficial effects, which will not be described in detail here. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings: Figure 1 A functional block diagram of an abnormal voiceprint detection system based on a pump characteristic spectrum provided by an embodiment of the present invention; Figure 2 A flowchart of an abnormal voiceprint detection method based on a pump characteristic spectrum provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0020] In some embodiments of the present application, see Figure 1 As shown, an abnormal voiceprint detection system based on a pump characteristic spectrum includes: The acquisition module is used to set the sampling rate, obtain the sound signals of multiple acquisition points of the pump based on the sampling rate, analyze the sound signals, generate a soundprint index for the pump according to the analysis result and determine the abnormal signal, the soundprint index includes a first index, a second index and a third index; a processing module, for identifying the voiceprint indicator, and when the second indicator or the third indicator is identified, processing the abnormal signal, obtaining a characteristic spectrum of the pump based on the processing result and determining the abnormal position of the pump, extracting the abnormal voiceprint feature of the characteristic spectrum, and determining the abnormal type of the pump according to the abnormal voiceprint feature; A verification module is used to verify the abnormality type according to the abnormal position. When the abnormal position is located inside the pump, the abnormal voiceprint feature is compared with the historical voiceprint data to determine the abnormal voiceprint similarity. Whether the abnormality type is accurate is determined based on the abnormal voiceprint similarity. When the abnormality type is determined to be inaccurate, the clustering algorithm is used to re-determine the abnormality type of the pump. When the abnormal position is located on the wall of the pump, the abnormality prediction type of the pump is determined based on the convolutional neural network model. Whether the abnormality type is accurate is verified based on the abnormality prediction type, and the abnormality type of the pump is re-determined based on the abnormality prediction type. The early warning module is used to determine the graded alarm according to the abnormal location and the verified abnormal type.

[0021] Specifically, the acquisition module is responsible for setting the sampling rate and obtaining the sound signals of multiple acquisition points of the pump. In order to ensure the accuracy and comprehensiveness of signal acquisition, the acquisition module usually sets a higher sampling rate to ensure that it can capture various types of small changes in sound signals generated by the pump during operation. The setting of the sampling rate is crucial for subsequent analysis. A too low sampling rate will not be able to capture sufficiently detailed signal data. The sampling rate is preferably 20khz, and the number of acquisition points is preferably 5, which can be adjusted according to actual needs. The collected sound signal is analyzed to obtain the soundprint index of the pump in operation. By determining the soundprint index, the system can determine whether the pump has an abnormal signal, which lays a data foundation for subsequent processing. When the processing module identifies the second or third indicator, the abnormal signal is further processed to obtain the characteristic spectrum of the pump. The characteristic spectrum can reflect the structural changes of the pump. By analyzing the characteristic spectrum of the pump, the abnormal position of the pump can be located, and the abnormal soundprint features can be extracted, and then the specific abnormal type can be identified, avoiding the judgment and error of human experience, and improving the reliability and accuracy of the judgment of the abnormal type. The verification module confirms the accuracy of the abnormal type based on the abnormal position and abnormal soundprint features. First, the verification module determines whether the abnormal position is located inside the pump or on the wall. If the abnormal position is located inside the pump, the historical soundprint data will be used for comparison to obtain the abnormal soundprint similarity. The historical soundprint data is obtained by long-term monitoring of the pump operation status, and provides the system with soundprint reference data during normal operation. When the abnormal type is judged to be inaccurate, the clustering algorithm is used for further analysis to re-determine the abnormal type of the pump. The reliability and accuracy of the abnormal type are improved. Secondly, if the abnormal position is located on the wall of the pump, the convolutional neural network (CNN) model is used to predict the abnormal type. The convolutional neural network can extract complex feature information from the signal data, identify the abnormal type through deep learning, and re-determine the abnormal type of the pump accordingly, ensuring the accuracy of the abnormal type. The early warning module is responsible for issuing corresponding alarms based on the detected abnormal position and the verified abnormal type, reducing the probability of major failures in the pump.

[0022] It is understandable that by collecting the sound signal of the pump and analyzing its soundprint indicators, various abnormal conditions in the operation of the pump can be accurately identified. The synergy of the processing module, verification module and early warning module can detect the abnormal position of the pump and confirm the abnormal type at the first time, issue the corresponding alarm and promptly remind relevant personnel to perform maintenance, thereby improving the service life and work efficiency of the pump and reducing maintenance costs and risks. The system can not only provide efficient and accurate abnormal diagnosis, but also provide strong support for subsequent equipment maintenance and management.

[0023] In some embodiments of the present application, when parsing the sound signal, generating a soundprint index for the pump according to the parsing result and determining an abnormal signal, it includes: The acquisition module obtains the standard sound waveform corresponding to the sound signals of all acquisition points, and determines the sound waveform corresponding to the sound signals of all acquisition points; When the sound waveforms are all equal to the standard sound waveform, a first indicator is generated for the pump; When the sound waveforms are not equal to the standard sound waveform, a second indicator is generated for the pump, and the sound signal of any collection point is determined as an abnormal signal; When there are one or more sound waveforms that are not equal to the standard sound waveform, a third indicator is generated for the pump, and it is determined whether the sound waveform that is not equal to the standard sound waveform is unique; When it is determined that the sound waveform that is not equal to the standard sound waveform is unique, any sound signal corresponding to the unique sound waveform is determined as an abnormal signal; When it is determined that the sound waveform that is not equal to the standard sound waveform is not unique, any sound signal corresponding to the maximum peak value of the sound waveform is determined as an abnormal signal.

[0024] In some embodiments of the present application, when the second indicator or the third indicator is identified, the abnormal signal is processed, and the characteristic spectrum of the pump is obtained based on the processing result and the abnormal position of the pump is determined, including: When the processing module identifies that the second indicator or the third indicator exists, retaining valid information of the abnormal signal based on the first signal algorithm and removing noise of the abnormal signal; Based on the result of noise removal, the abnormal position of the pump is determined according to the time delay difference of the abnormal signal. The abnormal signal is decomposed by a sine wave function based on the second signal algorithm, and the frequency data of the abnormal signal and the amplitude data of the abnormal signal are extracted. The abnormal signal is converted from the time domain to the frequency domain, and the characteristic spectrum of the pump is established according to the frequency data and the amplitude data. The horizontal axis of the characteristic spectrum is represented by frequency, and the vertical axis of the characteristic spectrum is represented by amplitude.

[0025] Specifically, the sound signals at different acquisition points are obtained and compared with the standard sound waveform. Through this comparison, it is judged whether the sound waveform of each acquisition point is normal, and the abnormal signal is determined according to whether the sound waveform is unique and the peak value of the sound waveform, avoiding the error of accidental signals, thereby improving the accuracy of abnormal signal determination. The first signal algorithm includes the sliding average method and the wavelet denoising method, which reduces the interference of noise by smoothing or removing irrelevant high-frequency components, thereby smoothing the abnormal signal, suppressing short-term fluctuations and noise, and improving the stability of abnormal signal determination. Specifically, one of them can be adopted according to the actual situation. The second signal algorithm includes the fast Fourier transform (FFT) and the short-time Fourier transform (STFT), and specifically one of them can be adopted according to the actual situation. The abnormal signal is decomposed by the sine wave function, and the frequency data and amplitude data of the abnormal signal are extracted, and then the characteristic spectrum of the pump is constructed, which improves the automation of the system, avoids the judgment of voiceprint detection based on human experience, and improves the reliability and accuracy of processing abnormal voiceprints, so as to timely discover the fault or abnormality of the pump. According to the characteristic spectrum, the abnormal position of the pump can be accurately located, thereby providing data support for subsequent analysis and improving the efficiency of pump detection.

[0026] In some embodiments of the present application, when extracting abnormal soundprint features of the characteristic spectrum and determining the abnormal type of the pump according to the abnormal soundprint features, it includes: The processing module pre-sets a first preset frequency and a second preset frequency, and extracts an amplitude peak value of the characteristic spectrum; When the first preset frequency has an amplitude peak, the processing module determines the abnormal type of the pump as oxidation; When the second preset frequency has an amplitude peak, the processing module determines the abnormal type of the pump as expansion; When both the first preset frequency and the second preset frequency have amplitude peaks, the processing module determines the abnormal type of the pump as fracture.

[0027] Specifically, when an abnormality occurs in the pump, its amplitude peak will fluctuate on the frequency of the characteristic spectrum. By finding the amplitude peak between different preset frequencies through the pre-set frequency, the abnormal situation of the pump can be timely classified to obtain the abnormality type, reducing the processing and analysis time of complex data, enabling the system to respond quickly, and improving the reliability and comprehensiveness of the system's processing of abnormal sound pattern features, providing data support for the maintenance and management of the pump.

[0028] In some embodiments of the present application, when comparing abnormal voiceprint features with historical voiceprint data to determine the similarity of abnormal voiceprints, it includes: The verification module determines the historical voiceprint features of the same abnormal type as the abnormal voiceprint features based on the historical voiceprint data, where the historical voiceprint features include the historical amplitude peak, the historical first preset frequency, the historical second preset frequency, and the number of historical amplitude peaks; The abnormal voiceprint similarity is obtained by the following formula: ; in, Indicates the abnormal voiceprint similarity, , , and represents the weight coefficient, and , represents the peak amplitude, Indicates the historical amplitude peak value, Indicates the first preset frequency, Indicates the first preset frequency in history, Indicates the second preset frequency, Indicates the second preset frequency in history, Indicates the number of amplitude peaks, Indicates the number of historical amplitude peaks.

[0029] In some embodiments of the present application, when judging whether the abnormal type is accurate according to the abnormal voiceprint similarity, it includes: The verification module pre-sets a similarity threshold, compares the abnormal voiceprint similarity with the similarity threshold, and determines whether the abnormal type is accurate based on the comparison result; When the abnormal voiceprint similarity is greater than or equal to the similarity threshold, the abnormal type is judged to be accurate; When the abnormal voiceprint similarity is less than the similarity threshold, the abnormal type is judged inaccurately.

[0030] Specifically, by comparing with historical voiceprint data, it can accurately identify and classify abnormal types, avoiding the error of finding the amplitude peak by simply setting the frequency, and improving the judgment ability and adaptability. By using the similarity of abnormal voiceprints and automatically judging whether the abnormal type is accurate, the reliability of abnormal voiceprint feature processing is improved.

[0031] In some embodiments of the present application, when the abnormality type is determined to be inaccurate, the abnormality type of the pump is re-determined using a clustering algorithm, including: According to the historical voiceprint data, representative data of each abnormal type is obtained, and the representative data and abnormal voiceprint features are combined to establish an aggregated data set. The number of clusters k is set to 3, and the feature vector of each data in the aggregated data set is extracted and the important features are determined. The parameters of the Gaussian distribution are initialized, and the probability that each data in the aggregated data set belongs to each Gaussian distribution is calculated to obtain the responsibility value. The cluster with the largest responsibility value is taken as the abnormal type of the data.

[0032] Specifically, representative data of each abnormal type is extracted based on historical voiceprint data. An aggregated data set is established by combining representative data and abnormal voiceprint features. The operation of the pumps in different periods is used to guide the subsequent clustering process, providing a reliable classification basis for redetermining the abnormal type. The Gaussian mixture model allows the system to automatically adjust the classification parameters according to the natural distribution of the data, thereby adapting to different abnormal types and data patterns, thereby reducing misjudgments and improving the accuracy of abnormal types.

[0033] In some embodiments of the present application, when the abnormal position is located on the wall of the pump, determining the abnormal prediction type of the pump based on the convolutional neural network model, verifying whether the abnormal type is accurate according to the abnormal prediction type, and re-determining the abnormal type of the pump based on the abnormal prediction type includes: The verification module sets the historical voiceprint data as the model training set, uses grid search to find the establishment parameters of the convolutional layer, pooling layer, fully connected layer and output layer of the convolutional neural network model, obtains the convolutional neural network model based on cross-validation, and uses the model training set to fit the convolutional neural network model; When the abnormal position is located on the wall of the pump, the abnormal soundprint features are input into the convolutional neural network model to obtain the abnormal prediction type of the pump, and the abnormal type is compared with the abnormal prediction type to determine whether the abnormal type is accurate based on the comparison result; When the abnormality type is equal to the abnormality prediction type, the abnormality type is judged to be accurate; When the abnormality type is not equal to the abnormality prediction type, it is determined that the abnormality type is inaccurate, and the abnormality type is deleted, and the abnormality prediction type is used as the abnormality type of the pump.

[0034] Specifically, the historical voiceprint data records the voiceprint data of the pump in different time periods. The historical voiceprint data is set as the model training set to improve the generalization ability of the convolutional neural network model. The grid search exhaustively searches the establishment parameters of the convolutional layer, pooling layer, fully connected layer and output layer of the convolutional neural network model in the parameter space. The cross-validation trains the model multiple times to verify its stability and performance. The model training set is used to fit the convolutional neural network model, which reduces the risk of overfitting of the convolutional neural network model, improves the accuracy and stability of the convolutional neural network model, and effectively eliminates human errors and deviations from preliminary judgments. The convolutional neural network model can automatically learn from historical voiceprint data, and finally obtains a convolutional neural network model with deep learning capabilities, which not only improves the accuracy of abnormal type prediction, but also improves the reliability of pump safety management.

[0035] In some embodiments of the present application, when determining a graded alarm according to the abnormal location and the verified abnormal type, it includes: The early warning module issues a first-level alarm based on the abnormal location being located on the wall of the pump and the abnormal type of the pump being oxidation or expansion; Since the abnormality is located inside the pump and the abnormality is oxidation or expansion, a secondary alarm is issued; A level 3 alarm is issued when the abnormality is located inside or on the wall of the pump and the abnormality is fracture; The urgency level of the alert increases from level one, level two, and level three.

[0036] Specifically, by combining the abnormal location and abnormal type, the system can automatically judge its severity according to the specific situation and issue corresponding alarms, helping operators to quickly identify and handle faults of different abnormal types, avoiding one-size-fits-all judgments, and reducing unnecessary intervention and misoperation. The graded alarms ensure the timeliness and reliability of response measures, improve the automation level of the system, ensure the early identification and processing of abnormal voiceprint features, ensure the safe operation of the pump and timely detect abnormal situations, reduce response time, and thus improve the safety and reliability of the pump.

[0037] In summary, the beneficial effects of the present invention are: by setting the sampling rate, the sound signal of the pump can be fully acquired from multiple collection points, accurate voiceprint indicators can be generated, and abnormal signals can be accurately identified. The abnormal position of the pump body is obtained based on the abnormal signal, which ensures the accuracy and comprehensiveness of the detection. By analyzing the characteristic spectrum and extracting the abnormal voiceprint features, the reliability of the abnormal voiceprint feature analysis and comprehensive processing is improved, and then the abnormal type of the pump is preliminarily judged, the accuracy of the judgment is improved, and the dependence on human experience is avoided. Based on the comparison with the historical voiceprint data, the accuracy of the abnormal type can be verified according to the abnormal position. When the abnormal position of the pump is located inside, the accuracy of the abnormal type is judged by comparing with the historical voiceprint data. When the abnormal position of the pump is located on the wall, the convolutional neural network model is used to obtain the abnormal prediction type and compare it with the abnormal type, which further improves the accuracy of the abnormal type and ensures the reliability of the detection. The graded alarm is automatically generated based on the abnormal position and the verified abnormal type. It ensures that personnel can respond in a timely and effective manner, reduces the downtime of the pump, and improves the safety management of the pump.

[0038] In another preferred embodiment based on the above embodiment, refer to Figure 2 As shown, this embodiment provides an abnormal voiceprint detection method based on a pump characteristic spectrum, which is used to apply the above-mentioned abnormal voiceprint detection system based on a pump characteristic spectrum, including: S100: setting a sampling rate, acquiring sound signals of multiple collection points of the pump machine based on the sampling rate, analyzing the sound signals, generating a soundprint index for the pump machine according to the analysis result and determining an abnormal signal, wherein the soundprint index includes a first index, a second index and a third index; S200: Identify the voiceprint indicator. When the second indicator or the third indicator is identified, process the abnormal signal, obtain the characteristic spectrum of the pump based on the processing result and determine the abnormal position of the pump, extract the abnormal voiceprint feature of the characteristic spectrum, and determine the abnormal type of the pump according to the abnormal voiceprint feature; S300: verify the abnormality type according to the abnormal position. When the abnormal position is located inside the pump, compare the abnormal voiceprint features with the historical voiceprint data to determine the abnormal voiceprint similarity. Determine whether the abnormality type is accurate based on the abnormal voiceprint similarity. When the abnormality type is inaccurate, use a clustering algorithm to re-determine the abnormality type of the pump. When the abnormal position is located on the wall of the pump, determine the abnormality prediction type of the pump based on the convolutional neural network model. Verify whether the abnormality type is accurate based on the abnormality prediction type. Determine the abnormality type of the pump based on the abnormality prediction type. S400: Determine a graded alarm according to the abnormal location and the verified abnormal type.

[0039] Specifically, the sampling rate is first set, and the sound signals of multiple collection points of the pump are obtained based on the sampling rate, and these sound signals are analyzed. The results of the analysis are used to generate the voiceprint index of the pump. Setting a suitable sampling rate can improve the accuracy of the sound signal and ensure the accuracy and reliability of the voiceprint index, thereby providing a data basis for the subsequent judgment of the status of the pump. If the second or third indicator is identified, the abnormal signal is processed, and the characteristic spectrum of the pump is obtained according to the processing result. The characteristic spectrum helps to determine the abnormal position of the pump and further extract the abnormal voiceprint features, which helps to narrow the abnormal judgment range and improve the efficiency of positioning. Extracting the characteristic spectrum and analyzing the abnormal voiceprint features can distinguish the abnormal type and improve the reliability of comprehensive processing and analysis of the abnormal voiceprint features. In addition, when verifying the abnormal type, if the abnormal position is located inside the pump, the abnormal voiceprint features are compared with the historical voiceprint data to obtain the abnormal voiceprint similarity, so as to determine whether the abnormal type is accurate. If the abnormal type is inaccurate, the abnormal type of the pump is re-determined by the clustering algorithm. If the abnormal location is located on the wall of the pump, the abnormal prediction type is obtained based on the convolutional neural network (CNN) model, and the accuracy of the abnormal type is verified based on the abnormal prediction type, which further improves the accuracy of the abnormal type and reduces the possibility of misjudgment. The clustering algorithm determines the accurate abnormal type through representative data in the historical voiceprint data. Whether it is a clustering algorithm based on historical voiceprint data or a convolutional neural network model based on deep learning, it can avoid human errors, reasonably analyze and comprehensively process abnormal voiceprint features, and improve the operational stability of the pump. The hierarchical alarm can ensure that alarms with high urgency are handled first, further improving the safety and reliability of the pump.

[0040] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0041] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0042] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0043] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An abnormal voiceprint detection system based on pump characteristic spectrum, characterized in that: include: A collection module, used to set a sampling rate, obtain sound signals of multiple collection points of the pump based on the sampling rate, analyze the sound signals, generate a soundprint index for the pump according to the analysis result and determine an abnormal signal, wherein the soundprint index includes a first index, a second index and a third index; a processing module, for identifying the voiceprint indicator, and when the second indicator or the third indicator is identified, processing the abnormal signal, obtaining a characteristic spectrum of the pump based on the processing result and determining the abnormal position of the pump, extracting the abnormal voiceprint feature of the characteristic spectrum, and determining the abnormal type of the pump according to the abnormal voiceprint feature; a verification module, for verifying the abnormality type according to the abnormal position; when the abnormal position is located inside the pump, comparing the abnormal voiceprint feature with the historical voiceprint data to determine the abnormal voiceprint similarity; judging whether the abnormality type is accurate according to the abnormal voiceprint similarity; when judging that the abnormality type is inaccurate, re-determining the abnormality type of the pump by using a clustering algorithm; when the abnormal position is located on the wall of the pump, determining the abnormality prediction type of the pump based on a convolutional neural network model; verifying whether the abnormality type is accurate according to the abnormality prediction type; and re-determining the abnormality type of the pump based on the abnormality prediction type; The early warning module is used to determine a graded alarm according to the abnormal location and the verified abnormal type.

2. The abnormal voiceprint detection system based on pump characteristic spectrum according to claim 1 is characterized in that: When the sound signal is analyzed, a soundprint index is generated for the pump according to the analysis result, and an abnormal signal is determined, the method includes: The acquisition module acquires the standard sound waveform corresponding to the sound signals of all acquisition points, and determines the sound waveform corresponding to the sound signals of all acquisition points; When the sound waveforms are all equal to the standard sound waveform, generating the first indicator for the pump; When the sound waveforms are not equal to the standard sound waveform, the second index is generated for the pump, and the sound signal of any collection point is determined as an abnormal signal; When there are one or more sound waveforms that are not equal to the standard sound waveform, generating the third indicator for the pump, and determining whether the sound waveform that is not equal to the standard sound waveform is unique; When it is determined that the sound waveform that is not equal to the standard sound waveform is unique, any sound signal corresponding to the unique sound waveform is determined as an abnormal signal; When it is determined that the sound waveform that is not equal to the standard sound waveform is not unique, any sound signal corresponding to the maximum peak value of the sound waveform is determined as an abnormal signal.

3. The abnormal voiceprint detection system based on pump characteristic spectrum according to claim 2 is characterized in that: When the second indicator or the third indicator is identified, the abnormal signal is processed, and a characteristic spectrum of the pump is obtained based on the processing result and the abnormal position of the pump is determined, including: When the processing module identifies that the second indicator or the third indicator exists, retaining valid information of the abnormal signal based on the first signal algorithm and removing noise of the abnormal signal; Based on the result of noise removal, the abnormal position of the pump is determined according to the time delay difference of the abnormal signal, the abnormal signal is decomposed by a sine wave function based on the second signal algorithm, and the frequency data of the abnormal signal and the amplitude data of the abnormal signal are extracted, the abnormal signal is converted from the time domain to the frequency domain, and a characteristic spectrum of the pump is established according to the frequency data and the amplitude data, wherein the horizontal axis of the characteristic spectrum is represented by frequency, and the vertical axis of the characteristic spectrum is represented by amplitude.

4. The abnormal voiceprint detection system based on pump characteristic spectrum according to claim 3 is characterized in that: When the abnormal voiceprint feature of the characteristic spectrum is extracted and the abnormal type of the pump is determined according to the abnormal voiceprint feature, it includes: The processing module pre-sets a first preset frequency and a second preset frequency, and extracts an amplitude peak value of the characteristic spectrum; When the amplitude peak value appears at the first preset frequency, the processing module determines the abnormal type of the pump as oxidation; When the amplitude peak value appears at the second preset frequency, the processing module determines the abnormal type of the pump machine as expansion; When the amplitude peaks appear at both the first preset frequency and the second preset frequency, the processing module determines the abnormality type of the pump as fracture.

5. The abnormal voiceprint detection system based on pump characteristic spectrum according to claim 4 is characterized in that: When comparing the abnormal voiceprint features with the historical voiceprint data to determine the abnormal voiceprint similarity, it includes: The verification module determines, based on the historical voiceprint data, a historical voiceprint feature of the same abnormal type as the abnormal voiceprint feature, wherein the historical voiceprint feature includes a historical amplitude peak value, a historical first preset frequency, a historical second preset frequency, and a historical amplitude peak number; The abnormal voiceprint similarity is obtained by the following formula: ； in, Indicates the abnormal voiceprint similarity, , , and represents the weight coefficient, and , represents the peak amplitude, Indicates the historical amplitude peak value, Indicates the first preset frequency, Indicates the first preset frequency in history, Indicates the second preset frequency, Indicates the second preset frequency in history, Indicates the number of amplitude peaks, Indicates the number of historical amplitude peaks.

6. The abnormal voiceprint detection system based on pump characteristic spectrum according to claim 5 is characterized in that: When judging whether the abnormal type is accurate according to the abnormal voiceprint similarity, it includes: The verification module pre-sets a similarity threshold, and compares the abnormal voiceprint similarity with the similarity threshold, and determines whether the abnormal type is accurate based on the comparison result; When the abnormal voiceprint similarity is greater than or equal to the similarity threshold, the abnormal type is determined to be accurate; When the abnormal voiceprint similarity is less than the similarity threshold, it is determined that the abnormal type is inaccurate.

7. The abnormal voiceprint detection system based on pump characteristic spectrum according to claim 6 is characterized in that: When it is determined that the abnormality type is inaccurate, a clustering algorithm is used to re-determine the abnormality type of the pump, including: According to the historical voiceprint data, representative data of each abnormal type is obtained, and the representative data and the abnormal voiceprint features are combined to establish an aggregated data set. The number of clusters k is set to 3, and the feature vector of each data in the aggregated data set is extracted and important features are determined. The parameters of the Gaussian distribution are initialized, and the probability that each data in the aggregated data set belongs to each Gaussian distribution is calculated to obtain the responsibility value, and the cluster with the largest responsibility value is taken as the abnormal type of the data.

8. The abnormal voiceprint detection system based on pump characteristic spectrum according to claim 7 is characterized in that: When the abnormal position is located on the wall of the pump, determining the abnormality prediction type of the pump based on a convolutional neural network model, verifying whether the abnormality type is accurate according to the abnormality prediction type, and re-determining the abnormality type of the pump based on the abnormality prediction type, including: The verification module sets the historical voiceprint data as a model training set, uses grid search to find the establishment parameters of the convolution layer, pooling layer, fully connected layer and output layer of the convolutional neural network model, obtains the convolutional neural network model based on cross-validation, and uses the model training set to fit the convolutional neural network model; When the abnormal position is located on the wall of the pump, the abnormal soundprint feature is input into the convolutional neural network model to obtain the abnormal prediction type of the pump, and the abnormal type is compared with the abnormal prediction type, and whether the abnormal type is accurate is determined according to the comparison result; When the abnormality type is equal to the abnormality prediction type, it is determined that the abnormality type is accurate; When the abnormality type is not equal to the abnormality prediction type, it is determined that the abnormality type is inaccurate, and the abnormality type is deleted, and the abnormality prediction type is used as the abnormality type of the pump.

9. The abnormal voiceprint detection system based on pump characteristic spectrum according to claim 8, characterized in that: When determining the graded alarm according to the abnormal location and the verified abnormal type, it includes: The early warning module issues a first-level alarm according to the abnormal position being located on the wall of the pump machine and the abnormal type of the pump machine being oxidation or expansion; According to the abnormal position being located inside the pump machine and the abnormal type of the pump machine being oxidation or expansion, a secondary alarm is issued; According to the abnormal position being located inside or on the wall of the pump machine, and the abnormal type of the pump machine being fracture, a third-level alarm is issued; The urgency levels of the first level alarm, the second level alarm, and the third level alarm are increased sequentially.

10. A method for detecting abnormal voiceprint based on pump characteristic spectrum, used for applying the abnormal voiceprint detection system based on pump characteristic spectrum as claimed in any one of claims 1 to 9, characterized in that: include: Setting a sampling rate, acquiring sound signals of multiple collection points of a pump machine based on the sampling rate, analyzing the sound signals, generating a soundprint index for the pump machine according to the analysis result and determining an abnormal signal, wherein the soundprint index includes a first index, a second index and a third index; Identify the voiceprint indicator, and when the second indicator or the third indicator is identified, process the abnormal signal, obtain the characteristic spectrum of the pump based on the processing result and determine the abnormal position of the pump, extract the abnormal voiceprint feature of the characteristic spectrum, and determine the abnormal type of the pump according to the abnormal voiceprint feature; The abnormality type is verified according to the abnormal position. When the abnormal position is located inside the pump, the abnormal voiceprint feature is compared with the historical voiceprint data to determine the abnormal voiceprint similarity. Whether the abnormality type is accurate is determined according to the abnormal voiceprint similarity. When the abnormality type is determined to be inaccurate, the abnormality type of the pump is re-determined using a clustering algorithm. When the abnormal position is located on the wall of the pump, the abnormality prediction type of the pump is determined based on a convolutional neural network model. Whether the abnormality type is accurate is verified according to the abnormality prediction type, and the abnormality type of the pump is re-determined based on the abnormality prediction type. A graded alarm is determined according to the abnormal location and the verified abnormal type.

Images