A pump unit abnormality detection method and system based on voiceprint recognition

Abstract

The present invention relates to the technical field of pump unit abnormality detection, and discloses a pump unit abnormality detection method and system based on voiceprint recognition, the method comprising: forming a voiceprint signal set to extract high-frequency signals; determining a time window, splitting the high-frequency signals, and obtaining sub-high-frequency signals; determining the signal frequency of the sub-high-frequency signals, determining the characterization frequency, comparing with the standard frequency range, and determining the initial detection result; when it is determined to be a suspected abnormality, obtaining the signal strength and obtaining the comprehensive frequency value of the voiceprint signal to determine whether there is an abnormality in the circulating water pump; when there is an abnormal circulating water pump in the pump unit to be detected, using a pre-trained curled neural model to determine the abnormality type of each abnormal circulating water pump, and determining the warning level according to the abnormality type and the number of abnormal water pumps. The present application realizes the early identification of equipment abnormalities and timely warning, reduces equipment downtime and maintenance costs, and enhances the stability and operation efficiency of the pump unit.

Description

Technical Field

[0001] The present invention relates to the technical field of pump unit abnormality detection, and in particular to a pump unit abnormality detection method and system based on voiceprint recognition. Background Art

[0002] As an important component of the pump unit, the circulating water pump is widely used in chemical industry, metallurgy, electric power and other fields, and undertakes key tasks such as cooling, heating and fluid transportation. However, since the pump unit may encounter various mechanical failures during long-term operation, such as impeller wear, bearing damage, pump body eccentricity, etc., its working efficiency decreases, energy consumption increases, and even serious equipment damage occurs. In order to ensure the stable operation of the pump unit and reduce the equipment failure rate, it is particularly important to perform timely and effective abnormality detection.

[0003] At present, most pump unit fault diagnosis methods mainly rely on monitoring based on physical parameters such as vibration, temperature, and pressure. However, due to the complexity of the equipment, diverse environmental factors, and changing failure modes, traditional methods have certain limitations. For example, although vibration sensors and temperature sensors can provide some fault information, they can only detect direct signs of mechanical failure. It is difficult to effectively identify early equipment failures, minor abnormal changes, or complex failure modes. Voiceprint recognition, as an emerging intelligent detection technology, is gradually being applied to the field of equipment fault monitoring. However, the complexity and dynamics of voiceprint signals in different environments and working conditions make their analysis more difficult, and traditional processing methods have poor adaptability to voiceprint processing.

[0004] Therefore, it is necessary to design a pump unit abnormality detection method and system based on voiceprint recognition to solve the problems existing in the current technology. Summary of the invention

[0005] In view of this, the present invention proposes a pump unit abnormality detection method and system based on voiceprint recognition, aiming to solve the current problems of low efficiency, difficult detection and poor detection reliability of early fault detection of pump units.

[0006] In one aspect, the present invention proposes a method for detecting abnormalities of a pump unit based on voiceprint recognition, comprising:

[0007] Collect the voiceprint signal of each circulating water pump in the pump unit to be detected within a preset time period, form a voiceprint signal set, and perform wavelet transform on each voiceprint signal in the voiceprint signal set to extract high-frequency signals;

[0008] Determine a time window according to the frequency fluctuation of the high-frequency signal, and split the high-frequency signal according to the time window to obtain a plurality of sub-high-frequency signals;

[0009] Determine the signal frequency of the sub-high frequency signal at each moment, use Gaussian mixture distribution to determine the characterization frequency of the sub-high frequency signal, compare all the characterization frequencies with the standard frequency range, and determine the initial detection result according to the comparison result, wherein the initial detection result includes abnormal, non-abnormal and suspected abnormal;

[0010] When it is determined that the initial detection result is suspected to be abnormal, the signal strength of each of the sub-high-frequency signals is obtained, and the comprehensive frequency value of the voiceprint signal is obtained according to the signal strengths and the characteristic frequencies of all the sub-high-frequency signals, and whether the circulating water pump is abnormal is determined according to the comprehensive frequency value;

[0011] When there is an abnormal circulating water pump in the pump unit to be detected, a pre-trained curled neural model is used to determine the abnormal type of each abnormal circulating water pump, and the warning level is determined according to the abnormal type and the number of abnormal water pumps.

[0012] Furthermore, when performing wavelet transform on each voiceprint signal in the voiceprint signal set to extract the high-frequency signal, it includes:

[0013] Using Daubechies 6 wavelet to perform discrete wavelet transform, the voiceprint signal is decomposed into 6 layers of wavelet;

[0014] The soft threshold method is used to extract the detail signal, and the threshold is set to 1.5 times the signal standard deviation.

[0015] Furthermore, when determining the time window according to the frequency fluctuation of the high-frequency signal, it includes:

[0016] Collect the frequency fluctuation value of each unit time in the high-frequency signal, determine the number of times the frequency fluctuation value is greater than the fluctuation threshold, compare the number with a first preset number and a second preset number respectively, and determine the time window according to the comparison result; the first preset number is less than the second preset number;

[0017] When the number is less than or equal to a first preset number, the time window is determined to be a first time period; when the number is greater than the first preset number and less than or equal to a second preset number, the time window is determined to be a second time period; when the number is greater than the second preset number, the time window is determined to be a third time period; the first time period is greater than the second time period, and the second time period is greater than the third time period.

[0018] Furthermore, when the Gaussian mixture distribution is used to determine the representation frequency of the sub-high frequency signal, it includes:

[0019] The kernel density function expression is:

[0020]

[0021] Where n represents the number of signal frequencies in the sub-high frequency signal, h represents the smoothing bandwidth, represents the frequency of the ith signal in the sub-high-frequency signal, and x represents the value of a certain frequency point to be estimated;

[0022] The frequency corresponding to the highest value of the kernel density is taken as the characterization frequency.

[0023] Further, all the characterization frequencies are compared with a standard frequency range, and the standard frequency range is obtained by:

[0024] Acquiring standard operating frequency data corresponding to the circulating water pump;

[0025] Collecting historical operation data, extracting non-abnormal operation data, and extracting a corresponding non-abnormal historical frequency range from the non-abnormal operation data;

[0026] The non-abnormal historical frequency range is compared with the standard operating frequency data, and a standard load range is generated according to the numerical value relationship, wherein the standard load range includes a left boundary value and a right boundary value.

[0027] Furthermore, when determining the initial test result based on the comparison result, it includes:

[0028] When all the characteristic frequencies are within the standard frequency range, the initial detection result is determined to be normal, and the circulating water pump is determined to be normal;

[0029] When all the characteristic frequencies are not within the standard frequency range, the initial detection result is determined to be abnormal, and the circulating water pump is determined to be abnormal;

[0030] When there is the characteristic frequency within the standard frequency range, and there is the characteristic frequency not within the standard frequency range, the initial detection result is determined to be a suspected abnormality.

[0031] Furthermore, when obtaining the comprehensive frequency value of the voiceprint signal according to the signal strength and the characterization frequency of all the sub-high-frequency signals, it includes:

[0032]

[0033] Wherein, F represents the comprehensive frequency value, fi represents the representation frequency of the i-th sub-high frequency signal, Ai represents the signal strength of the i-th sub-high frequency signal, and w represents the standard deviation of the signal strength.

[0034] Further, judging whether the circulating water pump is abnormal according to the comprehensive frequency value includes:

[0035] Comparing the comprehensive frequency value with the standard frequency range, and judging whether the circulating water pump has an abnormality according to the comparison result;

[0036] When the comprehensive frequency value is within the standard frequency range, it is determined that there is no abnormality in the circulating water pump;

[0037] When the comprehensive frequency value is not within the standard frequency range, it is determined that the circulating water pump is abnormal.

[0038] Furthermore, the pre-trained curled neural model is used to determine the abnormal type of each abnormal circulating water pump, and the warning level is determined according to the abnormal type and the number of abnormal water pumps, including:

[0039] Acquire historical abnormal data according to the historical operation data, determine the abnormal type of each historical abnormal data, and build a historical comparison database according to the historical abnormal data and the abnormal type;

[0040] Sampling the historical comparison database according to a preset ratio to obtain a training subset and a test subset;

[0041] Acquire a pre-selected neural network model, perform iterative training on the neural network model according to the training subset, and evaluate the iteratively trained neural network model according to the test subset to obtain the curled neural model;

[0042] The curled neural model is used to input the characteristic frequency or comprehensive frequency value of the circulating water pump with the current abnormality to determine the abnormality type;

[0043] The abnormal type and the abnormal water pump are input into the fuzzy algorithm, and the warning level is output according to the fuzzy rules.

[0044] Compared with the prior art, the beneficial effects of the present invention are: extracting the high-frequency components of the pump unit soundprint signal through wavelet transform, dynamically determining the time window using the frequency fluctuation of the high-frequency signal, and performing signal splitting to accurately capture the slight changes of the abnormality. The frequency of each sub-high-frequency signal is modeled using Gaussian mixture distribution to obtain the characterization frequency, and compared with the standard frequency to achieve a preliminary assessment of the pump unit status. When the preliminary detection result is a suspected abnormality, the signal strength and frequency characteristics are further combined to calculate the comprehensive frequency value and make an abnormality judgment. The types of abnormal water pumps are classified by a pre-trained convolutional neural network model, so as to dynamically determine the warning level. The ability to recognize complex fault modes is improved, and equipment abnormalities can be identified at an early stage and early warnings can be issued in a timely manner, which reduces equipment downtime and maintenance costs, and enhances the stability and operating efficiency of the pump unit.

[0045] On the other hand, the present application also provides a pump unit abnormality detection system based on voiceprint recognition, which is used to apply the above-mentioned pump unit abnormality detection method based on voiceprint recognition, including:

[0046] A collection unit is configured to collect the voiceprint signal of each circulating water pump in the pump unit to be detected within a preset time period, form a voiceprint signal set, and perform wavelet transform on each voiceprint signal in the voiceprint signal set to extract a high-frequency signal;

[0047] an analysis unit, configured to determine a time window according to the frequency fluctuation of the high-frequency signal, and split the high-frequency signal according to the time window to obtain a plurality of sub-high-frequency signals;

[0048] A judgment unit is configured to determine the signal frequency of the sub-high frequency signal at each moment, determine the characterization frequency of the sub-high frequency signal using a Gaussian mixture distribution, compare all the characterization frequencies with a standard frequency range, and determine a preliminary detection result according to the comparison result, wherein the preliminary detection result includes abnormal, non-abnormal, and suspected abnormal;

[0049] a processing unit configured to, when determining that the initial detection result is suspected to be abnormal, obtain the signal strength of each of the sub-high-frequency signals, obtain a comprehensive frequency value of the voiceprint signal according to the signal strengths and the characterization frequencies of all the sub-high-frequency signals, and determine whether the circulating water pump is abnormal according to the comprehensive frequency value;

[0050] The early warning unit is configured to use a pre-trained curled neural model to determine the abnormal type of each abnormal circulating water pump when there is an abnormal circulating water pump in the pump unit to be detected, and determine the early warning level according to the abnormal type and the number of abnormal water pumps.

[0051] It can be understood that the above-mentioned pump unit abnormality detection method and system based on voiceprint recognition have the same beneficial effects, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] 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:

[0053] Figure 1 A flow chart of a pump unit abnormality detection method based on voiceprint recognition provided by an embodiment of the present invention;

[0054] Figure 2 A structural block diagram of a pump unit abnormality detection system based on voiceprint recognition provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] 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.

[0056] In some embodiments of the present application, see Figure 1 As shown, a pump unit abnormality detection method based on voiceprint recognition includes:

[0057] S100: Collect the voiceprint signal of each circulating water pump in the pump unit to be detected within a preset time period, form a voiceprint signal set, and perform wavelet transform on each voiceprint signal in the voiceprint signal set to extract high-frequency signals.

[0058] S200: determining a time window according to the frequency fluctuation of the high-frequency signal, and splitting the high-frequency signal according to the time window to obtain a plurality of sub-high-frequency signals.

[0059] S300: Determine the signal frequency of the sub-high frequency signal at each moment, use Gaussian mixture distribution to determine the characterization frequency of the sub-high frequency signal, compare all the characterization frequencies with the standard frequency range, and determine the initial detection result based on the comparison result. The initial detection result includes abnormal, non-abnormal and suspected abnormal.

[0060] S400: When it is determined that the initial detection result is a suspected abnormality, the signal strength of each sub-high-frequency signal is obtained, and the comprehensive frequency value of the voiceprint signal is obtained according to the signal strength of all sub-high-frequency signals and the characterization frequency, and the circulating water pump is judged whether there is an abnormality according to the comprehensive frequency value.

[0061] S500: When there is an abnormal circulating water pump in the pump unit to be detected, a pre-trained curled neural model is used to determine the abnormal type of each abnormal circulating water pump, and the warning level is determined according to the abnormal type and the number of abnormal water pumps.

[0062] Specifically, in S100, the operating status of the pump unit is obtained by collecting soundprint signals. Soundprint signals are sound signals generated by the pump unit when it is working, which contain the frequency characteristics of various mechanical activities inside the pump unit. Wavelet transform is used to extract high-frequency signals to help capture minor abnormalities of the pump unit. High-frequency signals can reflect the details of mechanical failures, such as impeller wear and bearing damage. In S200, high-frequency signals often fluctuate over time, so a time window is determined to observe the changes in the signal within the window. The time window is divided by frequency fluctuations, and the signal is cut into shorter time segments (i.e., sub-high-frequency signals), thereby improving the accuracy and sensitivity of detection. In S300, the frequency of each sub-high-frequency signal is analyzed and the Gaussian mixture distribution model is used to determine the characterization frequency of the signal. This frequency is used as the "feature" of the signal for comparison with the preset standard frequency range. According to the comparison results, the preliminary detection results are judged, including "abnormal", "non-abnormal" and "suspected abnormal". In S400, when the preliminary detection result is "suspected abnormality", the signal strength of each sub-high-frequency signal is obtained, combined with the characteristic frequency of each sub-signal, to calculate the comprehensive frequency value. If the value exceeds the standard frequency range, it is determined that the pump unit is abnormal. In S500, once an abnormality is detected in a circulating water pump in the pump unit, the pre-trained convolutional neural network model (CNN) is used to classify the abnormal pump type. Determine what type of fault it is (such as impeller wear, bearing damage, etc.), and evaluate the operating status of the overall equipment based on the number of abnormal pumps and the type of fault, give an early warning level, and prompt maintenance personnel to intervene in time.

[0063] It is understandable that voiceprint recognition can capture tiny changes in the operation of the pump unit, especially in the high-frequency part, and can detect potential faults earlier, improving the predictive maintenance capability of the equipment. The voiceprint signal not only monitors the health status of the equipment, but also can adapt to different operating environments and working conditions, solving the problem of poor monitoring results of traditional methods due to changes in environmental factors. Combining wavelet transform, Gaussian mixture model, convolutional neural network and other technologies, data collection, signal analysis, abnormality judgment and fault classification are completed automatically, improving the efficiency and accuracy of detection. By comprehensively classifying frequency values ​​and abnormality types, potential problems can be discovered and warned in time before obvious faults occur in the equipment, preventing major damage and downtime caused by delayed maintenance of the equipment. Through multi-dimensional analysis, the operating status of the pump unit is evaluated, improving the comprehensiveness and accuracy of fault diagnosis.

[0064] In some embodiments of the present application, when performing wavelet transform on each voiceprint signal in the voiceprint signal set to extract the high-frequency signal, it includes:

[0065] Daubechies 6 wavelets are used for discrete wavelet transform, and the voiceprint signal is decomposed into 6 layers of wavelets.

[0066] The soft threshold method is used to extract the detail signal, and the threshold is set to 1.5 times the signal standard deviation.

[0067] Specifically, by applying 6 layers of wavelet decomposition to the voiceprint signal, the signal is decomposed into detail parts (detail signals) and approximate parts of different frequencies. Each layer of decomposition will split the signal into high-frequency and low-frequency components. The final detail signal can help analyze the operation of the device in each frequency band. For the detail signal part, if the signal amplitude is less than the set threshold, it is considered to be noise and suppressed; if the amplitude is greater than the threshold, the signal amplitude is retained and reduced. Through soft threshold denoising, the interference from environmental noise can be effectively reduced, the quality of the signal can be improved, and the sensitivity of anomaly detection can be improved. The threshold is set to 1.5 times the standard deviation of the signal.

[0068] It can be understood that by using Daubechies 6 wavelets for six-layer wavelet decomposition, the high-frequency components in the voiceprint signal can be effectively extracted, providing more refined signal features for subsequent fault detection. Wavelet transform not only helps the multi-scale analysis of the signal, but also removes noise while retaining effective information, thereby improving the accuracy of detection. The soft threshold method is used to further optimize the detail signal, effectively remove environmental noise and irrelevant signal components, thereby improving the accuracy of fault identification. This embodiment can more sensitively capture early faults and minor anomalies of the equipment, improving the timeliness and reliability of anomaly detection.

[0069] In some embodiments of the present application, when determining a time window based on the frequency fluctuation of a high-frequency signal, the method includes: collecting the frequency fluctuation value of each unit time in the high-frequency signal, determining the number of times the frequency fluctuation value is greater than a fluctuation threshold, comparing the number with a first preset number and a second preset number, respectively, and determining the time window based on the comparison result. The first preset number is less than the second preset number.

[0070] Specifically, when the number is less than or equal to the first preset number, the time window is determined to be the first time period. When the number is greater than the first preset number and less than or equal to the second preset number, the time window is determined to be the second time period. When the number is greater than the second preset number, the time window is determined to be the third time period. The first time period is greater than the second time period, and the second time period is greater than the third time period.

[0071] It can be understood that the frequency fluctuation value refers to the change of signal frequency in a short period of time, which is closely related to the mechanical operation of the equipment. Large frequency fluctuations indicate that the equipment has some abnormalities, such as imbalance, excessive vibration or component wear. By analyzing the frequency fluctuations of high-frequency signals, the size of the time window can be determined more flexibly, thereby improving the sensitivity and accuracy of abnormality detection. By comparing the number of fluctuations with the preset number of fluctuations, the length of the time window can be dynamically adjusted according to the strength of the signal fluctuation, and potential equipment abnormalities can be accurately captured according to different situations. When the frequency fluctuation is small and the equipment runs relatively smoothly, a longer time window can reduce overreaction to small fluctuations; when the frequency fluctuation is large, a shorter time window can quickly identify abnormal signals. The response speed and accuracy of abnormality detection are improved, and the adaptability and flexibility of the system are enhanced.

[0072] In some embodiments of the present application, when a Gaussian mixture distribution is used to determine the representation frequency of the sub-high frequency signal, it includes:

[0073] The kernel density function expression is:

[0074]

[0075] Where n represents the number of signal frequencies in the sub-high frequency signal, h represents the smoothing bandwidth, represents the frequency of the ith signal in the sub-high-frequency signal, and x represents the value of a certain frequency point to be estimated.

[0076] The frequency corresponding to the highest value of the kernel density is taken as the characterization frequency.

[0077] It is understandable that by using Gaussian mixture distribution and kernel density estimation methods, the most representative frequencies can be accurately extracted from complex frequency distributions. Compared with traditional simple frequency analysis methods, it can better cope with multi-peak situations in frequency distributions, effectively remove noise interference through smoothing, and extract the main frequencies that reflect the health status of the equipment. By determining the characterization frequency of each sub-high-frequency signal, the accuracy of fault diagnosis can be improved, especially in the early stages of equipment failure or minor abnormalities, key frequency features can be captured in a timely manner to ensure accurate monitoring and early warning of equipment status.

[0078] In some embodiments of the present application, all the characterization frequencies are compared with the standard frequency range, and the standard frequency range is obtained by: acquiring the standard operation frequency data corresponding to the circulating water pump. Collecting the historical operation data, extracting the non-abnormal operation data, and extracting the corresponding non-abnormal historical frequency range from the non-abnormal operation data. Comparing the non-abnormal historical frequency range with the standard operation frequency data, and generating the standard load range according to the numerical value relationship, wherein the standard load range includes a left boundary value and a right boundary value.

[0079] Specifically, when determining the initial test result based on the comparison result, it includes: when all the characterization frequencies are within the standard frequency range, the initial test result is determined to be non-abnormal, and the circulating water pump is determined to be normal. When all the characterization frequencies are not within the standard frequency range, the initial test result is determined to be abnormal, and the circulating water pump is determined to be abnormal. When there is a characterization frequency within the standard frequency range, and there is a characterization frequency that is not within the standard frequency range, the initial test result is determined to be suspected abnormal.

[0080] It is understandable that, assuming that there is no fault, the frequency fluctuation of the circulating water pump will be within a stable range. This range is obtained through the analysis of historical operation data and reflects the standard operation state of the pump unit. The standard operation frequency data is extracted from the historical operation data of the equipment. The typical operating frequency range of the equipment is determined by collecting and analyzing the data under normal operation. These data include the frequency fluctuation of the equipment under different loads, and based on long-term operation accumulation, they can accurately reflect the normal working state of the pump unit. By comparing the characterization frequency with the standard frequency range, it is determined whether the pump unit is in normal operation. The acquisition of the standard frequency range is based on a large amount of historical operation data, which is highly representative and reliable. Therefore, the comparison result can truly reflect the health status of the equipment. When the characterization frequency of the equipment falls completely into the standard range, it is judged as "non-abnormal", and when the frequency deviates, the abnormality can be quickly and effectively identified. Especially when some characterization frequencies are abnormal, the "suspected abnormality" judgment method can provide more detailed judgment for fault warning and avoid misjudgment or missed judgment due to single signal abnormality. The detection method based on the standard frequency range improves the accuracy and reliability of pump unit fault detection.

[0081] In some embodiments of the present application, obtaining the comprehensive frequency value of the voiceprint signal according to the signal strength and the characterization frequency of all sub-high-frequency signals includes:

[0082]

[0083] Wherein, F represents the comprehensive frequency value, fi represents the representation frequency of the i-th sub-high frequency signal, Ai represents the signal strength of the i-th sub-high frequency signal, and w represents the standard deviation of the signal strength.

[0084] In some embodiments of the present application, when judging whether the circulating water pump has an abnormality according to the comprehensive frequency value, it includes: comparing the comprehensive frequency value with the standard frequency range, and judging whether the circulating water pump has an abnormality according to the comparison result.

[0085] Specifically, when the comprehensive frequency value is within the standard frequency range, it is determined that the circulating water pump has no abnormality. When the comprehensive frequency value is not within the standard frequency range, it is determined that the circulating water pump has an abnormality.

[0086] It is understandable that the comprehensive frequency value is calculated by combining the signal strength and the characterization frequency. The comprehensive frequency value can take into account the influence of each sub-high frequency signal, making the analysis result more accurate and reliable, especially when the signal strength changes greatly or there are multiple frequency components, which can better reflect the overall operating status of the equipment. The comprehensive frequency value is compared with the standard frequency range to avoid the problem of misjudgment or missed judgment caused by relying solely on a single frequency or signal strength. Potential faults can be discovered earlier and early warnings can be issued in time.

[0087] In some embodiments of the present application, a pre-trained curled neural model is used to determine the abnormal type of each abnormal circulating water pump, and the warning level is determined according to the abnormal type and the number of abnormal water pumps, including:

[0088] Obtain historical abnormal data based on historical operation data, determine the abnormal type of each historical abnormal data, and build a historical comparison database based on the historical abnormal data and abnormal type.

[0089] The historical comparison database is sampled according to a preset ratio to obtain a training subset and a test subset.

[0090] A pre-selected neural network model is obtained, and the neural network model is iteratively trained according to a training subset, and the neural network model after iterative training is evaluated according to a test subset to obtain a curled neural model.

[0091] The curled neural model is used to input the characteristic frequency or comprehensive frequency value of the circulating water pump with the current abnormality to determine the abnormality type.

[0092] The abnormal type and abnormal water pump are input into the fuzzy algorithm, and the warning level is output according to the fuzzy rules.

[0093] Specifically, by analyzing the historical operation data, all the historical records of abnormalities are extracted, and the abnormality type of each record is marked. The abnormality types include different types of equipment failures, such as bearing wear, impeller damage, pump body eccentricity, etc. According to the abnormality type and data characteristics, a historical comparison database is constructed. The database contains the annotation information of various abnormality types. According to the historical comparison database, the data set is extracted from it according to the preset ratio to generate the training subset and the test subset. The training subset is used for learning the network model, and the test subset is used to verify the effect of the model to ensure that the model can be effectively generalized to new data. The pre-selected curled neural network model is selected, and the model is iteratively trained using the training subset. Through training, the network learns the relationship between the abnormality type and its corresponding frequency characteristics. The test subset is used to evaluate the accuracy and effect of the trained network model. After the model training is completed, the trained curled neural network model is used to infer the abnormal water pump to be detected. The input data includes the characterization frequency or comprehensive frequency value of the circulating water pump to determine the abnormality type of the water pump. The abnormality type and number of each abnormal water pump are input into the fuzzy algorithm. The fuzzy algorithm outputs the final warning level based on fuzzy rules (such as the number of anomalies, the severity of the anomaly type, etc.). The warning level reflects the severity of the current equipment failure, thereby helping to formulate subsequent maintenance strategies.

[0094] Specifically, assume several common fault types of water pumps, for example: Bearing wear: represents a mild abnormality. Impeller damage: represents a moderate abnormality. Pump body eccentricity: represents a severe abnormality. A small number (1-2) of abnormal water pumps: indicates that the system is in an early fault state. A medium number (3-5) of abnormal water pumps: indicates that there is a moderate fault risk and the operation is unstable. A large number (more than 6) of abnormal water pumps: indicates that it is in a serious fault state and requires emergency treatment. Fuzzy rule definition: Input parameters: abnormality type (mild, moderate, severe), number of abnormal water pumps (small, medium, large). Output parameters: warning level (low, medium, high). Fuzzy rules: Rule 1: Condition: The abnormality type is "mild abnormality" and the number of abnormal water pumps is "small". Output: The warning level is "low".

[0095] Rule 2: Condition: The abnormality type is "mild abnormality" and the number of abnormal pumps is "medium". Output: The warning level is "medium".

[0096] Rule 3: Condition: The abnormality type is "mild abnormality" and the number of abnormal pumps is "large". Output: The warning level is "high".

[0097] It is understandable that the curled neural network can effectively identify and classify various types of anomalies by automatically learning complex patterns in historical data, and can show high accuracy and robustness when processing complex signal data. By combining the type of anomaly and the number of water pumps with a fuzzy algorithm to determine the warning level, this embodiment can not only identify the type of fault, but also output the corresponding warning level according to the actual situation, thereby achieving accurate equipment maintenance and fault response. The intelligent level of anomaly detection is improved.

[0098] In the above embodiment, the high-frequency components of the pump unit soundprint signal are extracted by wavelet transform, the frequency fluctuation of the high-frequency signal is used to dynamically determine the time window, and the signal is split to accurately capture the slight changes of the abnormality. The frequency of each sub-high-frequency signal is modeled using Gaussian mixture distribution to obtain the characterization frequency, and compared with the standard frequency to achieve a preliminary assessment of the state of the pump unit. When the preliminary detection result is a suspected abnormality, the signal strength and frequency characteristics are further combined to calculate the comprehensive frequency value and make an abnormality judgment. The types of abnormal water pumps are classified by a pre-trained convolutional neural network model, so as to dynamically determine the warning level. The ability to recognize complex fault modes is improved, and equipment abnormalities can be identified at an early stage and early warning can be issued in a timely manner, which reduces equipment downtime and maintenance costs, and enhances the stability and operation efficiency of the pump unit.

[0099] In another preferred embodiment based on the above embodiment, refer to Figure 2 As shown, this embodiment provides a pump unit abnormality detection system based on voiceprint recognition, which is used to apply the above-mentioned pump unit abnormality detection method based on voiceprint recognition, including:

[0100] A collection unit is configured to collect the voiceprint signal of each circulating water pump in the pump unit to be detected within a preset period of time, form a voiceprint signal set, and perform wavelet transform on each voiceprint signal in the voiceprint signal set to extract a high-frequency signal;

[0101] An analysis unit is configured to determine a time window according to the frequency fluctuation of the high-frequency signal, and split the high-frequency signal according to the time window to obtain a plurality of sub-high-frequency signals;

[0102] A judgment unit is configured to determine the signal frequency of the sub-high frequency signal at each moment, determine the representation frequency of the sub-high frequency signal using a Gaussian mixture distribution, compare all the representation frequencies with the standard frequency range, and determine the initial detection result according to the comparison result, wherein the initial detection result includes abnormal, non-abnormal, and suspected abnormal;

[0103] The processing unit is configured to obtain the signal strength of each sub-high frequency signal when it is determined that the initial detection result is suspected to be abnormal, obtain the comprehensive frequency value of the voiceprint signal according to the signal strength and the characterization frequency of all sub-high frequency signals, and determine whether the circulating water pump is abnormal according to the comprehensive frequency value;

[0104] The early warning unit is configured to use a pre-trained curled neural model to determine the abnormal type of each abnormal circulating water pump when there is an abnormal circulating water pump in the pump unit to be detected, and determine the early warning level according to the abnormal type and the number of abnormal water pumps.

[0105] It can be understood that the high-frequency components of the pump unit soundprint signal are extracted through wavelet transform, and the frequency fluctuations of the high-frequency signal are used to dynamically determine the time window, and the signal is split to accurately capture the slight changes of the abnormality. The frequency of each sub-high-frequency signal is modeled using Gaussian mixture distribution to obtain the characterization frequency, and compared with the standard frequency to achieve a preliminary assessment of the state of the pump unit. When the preliminary detection result is suspected to be abnormal, the signal strength and frequency characteristics are further combined to calculate the comprehensive frequency value for abnormal judgment. The types of abnormal water pumps are classified through the pre-trained convolutional neural network model, so as to dynamically determine the warning level. The ability to recognize complex fault modes has been improved, and equipment abnormalities can be identified at an early stage and early warning can be issued in a timely manner, which reduces equipment downtime and maintenance costs, and enhances the stability and operation efficiency of the pump unit.

[0106] 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.

[0107] 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.

[0108] 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 comprising an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0109] 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. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0110] 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. A pump unit abnormality detection method based on voiceprint recognition, characterized in that: include: Collect the voiceprint signal of each circulating water pump in the pump unit to be detected within a preset time period, form a voiceprint signal set, and perform wavelet transform on each voiceprint signal in the voiceprint signal set to extract high-frequency signals; Determine a time window according to the frequency fluctuation of the high-frequency signal, and split the high-frequency signal according to the time window to obtain a plurality of sub-high-frequency signals; Determine the signal frequency of the sub-high frequency signal at each moment, use Gaussian mixture distribution to determine the characterization frequency of the sub-high frequency signal, compare all the characterization frequencies with the standard frequency range, and determine the initial detection result according to the comparison result, wherein the initial detection result includes abnormal, non-abnormal and suspected abnormal; When it is determined that the initial detection result is suspected to be abnormal, the signal strength of each of the sub-high-frequency signals is obtained, and the comprehensive frequency value of the voiceprint signal is obtained according to the signal strengths and the characteristic frequencies of all the sub-high-frequency signals, and whether the circulating water pump is abnormal is determined according to the comprehensive frequency value; When there is an abnormal circulating water pump in the pump unit to be detected, a pre-trained curled neural model is used to determine the abnormal type of each abnormal circulating water pump, and the warning level is determined according to the abnormal type and the number of abnormal water pumps; All of the characterization frequencies are compared to a standard frequency range, which is obtained by: Acquiring standard operating frequency data corresponding to the circulating water pump; Collecting historical operation data, extracting non-abnormal operation data, and extracting a corresponding non-abnormal historical frequency range from the non-abnormal operation data; Comparing the non-abnormal historical frequency range with the standard operating frequency data, and generating a standard load range according to the numerical value relationship, wherein the standard load range includes a left boundary value and a right boundary value; When determining the initial test results based on the comparison results, it includes: When all the characteristic frequencies are within the standard frequency range, the initial detection result is determined to be normal, and the circulating water pump is determined to be normal; When all the characteristic frequencies are not within the standard frequency range, the initial detection result is determined to be abnormal, and the circulating water pump is determined to be abnormal; When there is the characteristic frequency within the standard frequency range, and there is the characteristic frequency not within the standard frequency range, determining that the initial detection result is suspected abnormal; When the comprehensive frequency value of the voiceprint signal is obtained according to the signal strength and the characterization frequency of all the sub-high frequency signals, it includes: ； Wherein, F represents the comprehensive frequency value, fi represents the representation frequency of the i-th sub-high frequency signal, Ai represents the signal strength of the i-th sub-high frequency signal, and w represents the standard deviation of the signal strength.

2. The method for detecting abnormality of a pump unit based on voiceprint recognition according to claim 1, characterized in that: The step of performing wavelet transform on each voiceprint signal in the voiceprint signal set to extract a high-frequency signal includes: Using Daubechies 6 wavelet to perform discrete wavelet transform, the voiceprint signal is decomposed into 6 layers of wavelet; The soft threshold method is used to extract the detail signal, and the threshold is set to 1.5 times the signal standard deviation.

3. The method for detecting abnormality of a pump unit based on voiceprint recognition according to claim 1, characterized in that: When the time window is determined according to the frequency fluctuation of the high-frequency signal, it includes: Collect the frequency fluctuation value of each unit time in the high-frequency signal, determine the number of times the frequency fluctuation value is greater than the fluctuation threshold, compare the number with a first preset number and a second preset number respectively, and determine the time window according to the comparison result; the first preset number is less than the second preset number; When the number is less than or equal to a first preset number, the time window is determined to be a first time period; when the number is greater than the first preset number and less than or equal to a second preset number, the time window is determined to be a second time period; when the number is greater than the second preset number, the time window is determined to be a third time period; the first time period is greater than the second time period, and the second time period is greater than the third time period.

4. The method for detecting abnormality of a pump unit based on voiceprint recognition according to claim 1, characterized in that: When the Gaussian mixture distribution is used to determine the representation frequency of the sub-high frequency signal, it includes: The kernel density function expression is: ； Where n represents the number of signal frequencies in the sub-high frequency signal, h represents the smoothing bandwidth, represents the frequency of the ith signal in the sub-high-frequency signal, and x represents the value of a certain frequency point to be estimated; The frequency corresponding to the highest value of the kernel density is taken as the characterization frequency.

5. The method for detecting abnormality of a pump unit based on voiceprint recognition according to claim 1, characterized in that: When judging whether the circulating water pump is abnormal according to the comprehensive frequency value, it includes: Comparing the comprehensive frequency value with the standard frequency range, and judging whether the circulating water pump has an abnormality according to the comparison result; When the comprehensive frequency value is within the standard frequency range, it is determined that there is no abnormality in the circulating water pump; When the comprehensive frequency value is not within the standard frequency range, it is determined that the circulating water pump is abnormal.

6. The method for detecting abnormality of a pump unit based on voiceprint recognition according to claim 1, characterized in that: The pre-trained curled neural model is used to determine the abnormal type of each abnormal circulating water pump. When determining the warning level according to the abnormal type and the number of abnormal water pumps, it includes: Acquire historical abnormal data according to the historical operation data, determine the abnormal type of each historical abnormal data, and build a historical comparison database according to the historical abnormal data and the abnormal type; Sampling the historical comparison database according to a preset ratio to obtain a training subset and a test subset; Acquire a pre-selected neural network model, perform iterative training on the neural network model according to the training subset, and evaluate the iteratively trained neural network model according to the test subset to obtain the curled neural model; The curled neural model is used to input the characteristic frequency or comprehensive frequency value of the circulating water pump with the current abnormality to determine the abnormality type; The abnormal type and the abnormal water pump are input into the fuzzy algorithm, and the warning level is output according to the fuzzy rules.

7. A pump unit abnormality detection system based on voiceprint recognition, used for applying the pump unit abnormality detection method based on voiceprint recognition as described in any one of claims 1 to 6, characterized in that: include: A collection unit is configured to collect the voiceprint signal of each circulating water pump in the pump unit to be detected within a preset time period, form a voiceprint signal set, and perform wavelet transform on each voiceprint signal in the voiceprint signal set to extract a high-frequency signal; an analysis unit, configured to determine a time window according to the frequency fluctuation of the high-frequency signal, and split the high-frequency signal according to the time window to obtain a plurality of sub-high-frequency signals; A judgment unit is configured to determine the signal frequency of the sub-high frequency signal at each moment, determine the characterization frequency of the sub-high frequency signal using a Gaussian mixture distribution, compare all the characterization frequencies with a standard frequency range, and determine a preliminary detection result according to the comparison result, wherein the preliminary detection result includes abnormal, non-abnormal, and suspected abnormal; a processing unit configured to, when determining that the initial detection result is suspected to be abnormal, obtain the signal strength of each of the sub-high-frequency signals, obtain a comprehensive frequency value of the voiceprint signal according to the signal strengths and the characterization frequencies of all the sub-high-frequency signals, and determine whether the circulating water pump is abnormal according to the comprehensive frequency value; The early warning unit is configured to use a pre-trained curled neural model to determine the abnormal type of each abnormal circulating water pump when there is an abnormal circulating water pump in the pump unit to be detected, and determine the early warning level according to the abnormal type and the number of abnormal water pumps.

Images