Water pump flow and head on-line detection device and method using acoustic emission technology

Abstract

The application discloses a water pump flow and lift online detection device and method using acoustic emission technology, and relates to the technical field of water pump performance detection. The water pump flow and lift online detection device using acoustic emission technology comprises an acoustic emission sensor acquisition module, a transient event processing module and a water pump flow and lift detection module. The application judges whether to optimize the acoustic emission signal by the quality parameter of the emission signal, labels the transient event, classifies the transient event when the transient event is detected, feeds back a prompt to a preset person, extracts acoustic characteristics from the acoustic emission signal of the non-transient event to obtain the water pump flow and lift if the transient event is not detected, finally reflects the evaluation of the mapping accuracy of the water pump flow and lift, and judges whether to perform mapping optimization. The application improves the online detection accuracy of the water pump flow and lift, and solves the problem of low online detection accuracy of the water pump flow and lift in the prior art.

Description

Technical Field

[0001] This invention relates to the field of water pump performance testing technology, and in particular to an online testing device and method for water pump flow rate and head using acoustic emission technology. Background Technology

[0002] Acoustic emission (AE) refers to the phenomenon where materials or components, under the influence of external or internal forces, undergo irreversible plastic deformation due to deformation, fracture, or internal stress exceeding the yield limit, releasing strain energy in the form of transient elastic waves. These elastic waves propagate through the equipment structure and can be captured by highly sensitive sensors and converted into electrical signals, thereby enabling real-time monitoring of the equipment's condition. During pump operation, flow rate and head are crucial parameters reflecting the pump's operating conditions, providing a basis for energy efficiency assessment, condition monitoring, and fault diagnosis.

[0003] Existing detection methods involve collecting acoustic emission signals using sensors placed in key parts of the water pump (such as impellers, bearings, or pipes). The collected acoustic emission signals are then filtered, amplified, and feature extracted (e.g., energy rate) to identify characteristic parameters related to flow rate and head. Finally, by establishing a mapping relationship between acoustic emission characteristics and flow rate and head, online parameter calculation is achieved.

[0004] For example, Chinese patent application CN116104750B discloses a method and apparatus for testing the flow rate and head of a water pump, including: a sampling step, testing the water resistance curve of the test system under different valve openings of the regulating valve, then testing the hydraulic performance (head-flow rate) curve of a qualified water pump, obtaining the pressure sensor value, and calculating the relationship coefficient x between the pressure parameter and the water pump head; a testing step, installing the water pump to be tested into the test system, adjusting the valve opening, recording the pressure parameter at the test point, using the relationship coefficient between the pressure parameter and the water pump head to obtain the head of the test water pump, and then calculating the flow rate of the test water pump based on the water resistance curve and the head; and judging whether the water pump performance meets the requirements based on the obtained water pump performance parameters, if it is unqualified, an alarm is triggered, and if it is qualified, it passes.

[0005] For example, Chinese patent application CN105673474B discloses a method and system for detecting pump efficiency, flow rate, and head, comprising: acquiring the output frequency f and output power Pf,output of a frequency converter, wherein the frequency converter is connected to the pump signal; and calculating the pump flow rate Qf, head Hf, and efficiency ηf based on the frequency converter's output frequency f and output power Pf,output. According to the method and system for detecting pump efficiency, flow rate, and head provided by this invention, the controller reads the frequency converter's output power and output frequency, and the pump flow rate, head, and efficiency can be calculated based on the frequency converter's output power and output frequency.

[0006] The above-mentioned technology has at least the following technical problems:

[0007] In existing technologies, under high load and high pressure conditions, industrial water pumps experience dynamic changes in flow rate and head due to variations in user demand and pipeline resistance. This not only induces cavitation (where bubbles or "cavities" form inside the liquid when the local pressure is lower than its saturated vapor pressure, releasing physical effects such as high temperature, high pressure, shock waves, and noise) and generates high-frequency acoustic emission signals, but also causes the signal amplitude and frequency distribution to exhibit nonlinear characteristics, significantly increasing the difficulty of signal processing. Simultaneously, the coupling interference from multiple factors in the industrial environment, such as electromagnetic noise, mechanical vibration, and hydrodynamic noise, significantly reduces the signal-to-noise ratio of the acoustic emission signal. Especially in variable frequency pumps, although the system operates more smoothly, frequency adjustments and load changes still exacerbate signal complexity and interference.

[0008] Furthermore, changes in flow rate and head can simultaneously affect the characteristics of acoustic emission signals, making decoupling analysis difficult. For example, both can cause an increase in signal amplitude, further increasing the complexity of data analysis due to this multivariate coupling problem. Meanwhile, in industries such as chemical and petroleum, different production stages may require the transport of fluids with varying viscosities, densities, or temperatures to meet reaction or processing needs. Changes in the viscosity, density, and temperature of the fluid medium can also affect turbulence intensity and acoustic emission propagation characteristics, causing calibration curves to become invalid when the medium changes. At high flow rates, the intensified cavitation effect can easily cause the generated acoustic emission signals to overlap with turbulence signals, making them difficult to distinguish and leading to overestimation of flow rate. This results in low accuracy in online detection of pump flow rate and head. Summary of the Invention

[0009] To address the low accuracy of online pump flow and head detection in existing technologies, this invention provides a method and apparatus for online pump flow and head detection utilizing acoustic emission technology. The technical solution is as follows:

[0010] On one hand, an online detection device for water pump flow rate and head using acoustic emission technology is provided. This device includes: an acoustic emission sensor acquisition module, a transient event processing module, and a water pump flow rate and head detection module. The acoustic emission sensor acquisition module collects acoustic emission signals caused by various operating conditions using a pre-deployed number of acoustic emission sensors. Based on the quality parameters of the acoustic emission signals, it determines whether acoustic emission signal optimization is needed. If so, transient events reflecting cavitation phenomena are marked after acoustic emission signal optimization; otherwise, transient events are directly marked. The transient event processing module classifies transient events when they are detected and feeds back the classification results and abnormal detection alerts to pre-determined personnel. If no transient event is detected, the water pump flow rate and head are obtained by mapping the acoustic features extracted from the acoustic emission signals of non-transient events. The water pump flow rate and head detection module evaluates the accuracy of the mapped water pump flow rate and head and determines whether mapping optimization is needed based on the evaluation results.

[0011] On the other hand, an online detection method for water pump flow rate and head using acoustic emission technology is provided. This method is implemented by an online detection device for water pump flow rate and head using acoustic emission technology. The method includes: S1, collecting acoustic emission signals caused by various operating conditions through a preset number of deployed acoustic emission sensors, determining whether to optimize the acoustic emission signals based on the quality parameters of the acoustic emission signals, if so, marking transient events reflecting cavitation phenomena after acoustic emission signal optimization, otherwise directly marking the transient events; S2, classifying transient events when they are detected, and feeding back the classification results and detection anomaly prompts to preset personnel, if no transient events are detected, mapping the acoustic features extracted from the acoustic emission signals of non-transient events to obtain the water pump flow rate and head; S3, evaluating the accuracy of the mapping based on the mapped water pump flow rate and head, and determining whether to optimize the mapping based on the evaluation results.

[0012] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0013] 1. By comprehensively collecting acoustic emission signals through deployed acoustic emission sensors, and determining whether to optimize the acoustic emission signals based on signal quality parameters, low-quality signals are prevented from directly entering subsequent analysis stages, reducing false alarms and missed alarms, thereby improving the stability and accuracy of signal processing. When signal quality is poor, acoustic emission signal optimization is performed first, followed by marking transient events; when signal quality is high, transient events are marked directly. This ensures efficient identification of cavitation events under different conditions, improving the sensitivity and accuracy of transient event detection. Furthermore, detected transient events are classified, and the classification results are fed back to designated personnel, which can be used for early warning of cavitation, preventing severe cavitation from escalating. The process addresses the issue of pump impeller or casing damage caused by chemical reactions, improving operational response speed. It then maps the acoustic features extracted from the acoustic emission signals of non-transient events to obtain pump flow rate and head. Compared to traditional mechanical or electrical sensors, this method eliminates the need for additional pressure sensors or flow meters, reducing hardware costs and avoiding poor acoustic emission signal quality stability due to temperature and pressure shocks. Finally, it evaluates the accuracy of the current mapping results and determines whether mapping optimization is triggered. This system adapts to different fluid media and operating conditions, maintaining higher accuracy and better adaptability in the pump flow rate and head mapping mechanism, thus improving the accuracy of pump flow rate and head mapping.

[0014] 2. By introducing a reference signal-to-noise ratio (SNR) mechanism, if the acoustic emission signal SNR is high, transient event discrimination is performed directly, ensuring real-time performance. Otherwise, the first acoustic emission signal is optimized before transient event discrimination, preventing low-quality signals from directly entering the detection stage and reducing the false alarm rate. Then, transient event discrimination is performed by combining kurtosis and transient SNR to ensure effective differentiation between real transient events and random noise. Simultaneously, the accuracy of transient event discrimination is verified using multiple sensors. If the verification result is a non-transient event, the second acoustic emission signal is optimized; otherwise, it is marked as a transient event, further eliminating local interference and improving the accuracy and reliability of transient event marking. The first acoustic emission signal optimization uses different filters based on kurtosis; if the kurtosis is low... Low-pass filtering is performed to quickly remove high-frequency noise; otherwise, variational mode decomposition (MODED) is performed first, followed by integrated empirical mode decomposition (IDED) to achieve deep denoising of non-stationary signals. Variational mode decomposition introduces an adaptive mapping between the number of decomposed modes and the penalty factor, ensuring the stability of the acoustic emission signal decomposition effect. IDED, combined with noise amplitude adjustment and correlation coefficient stability judgment, can automatically select the optimal stopping point, avoid over-decomposition, and improve the physical interpretability of intrinsic mode function components. By prioritizing signal reconstruction to select the optimal IDE component, the energy retention of the denoised signal is higher. The second acoustic emission signal optimization is performed differently for different noise types to improve the anti-interference capability in complex noise environments and ensure the filtering effect under different conditions.

[0015] 3. By setting a reference mapping to accurately evaluate scores, no optimization is performed when the accuracy meets the standard, ensuring the real-time online detection of pump flow and head. Otherwise, mapping optimization is triggered. Then, the acoustic emission signal is decoupled segment by segment by combining the sliding window length in independent component analysis to separate the influence of various interference factors on the acoustic emission signal. The decoupling result of the previous window is dynamically inherited as the initial signal of the next window, which significantly reduces the randomness of parameter initialization in independent component analysis and improves the convergence speed and stability of independent component analysis. Furthermore, after each sliding window processing, the reconstructed acoustic emission signal is compared with the acoustic emission signal before decoupling. The residual signal is obtained, and the parameters in the independent component analysis are dynamically corrected by combining the residual signal. This can gradually approach the optimal decoupling effect, thereby improving the detection accuracy during long-term operation. Then, the mutual information criterion is introduced, and the parameters in the independent component analysis are dynamically adjusted by the gradient descent method to minimize the mutual information. This allows the model to adapt to changes in different fluid media and pump operating conditions, ensuring the robustness of the detection results. Finally, the parameters output by the optimized model are weighted and coupled with the parameters obtained by mapping. The influence of single model error is reduced through the weight allocation mechanism, thereby improving the stability and reliability of the output of flow rate and head parameters. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is one of the flowcharts for online detection of water pump flow rate and head provided in the embodiments of the present invention;

[0018] Figure 2 This is the second flowchart of online detection of water pump flow rate and head provided in the embodiments of the present invention;

[0019] Figure 3 This is a schematic diagram of the structure of the online detection device for water pump flow and head using acoustic emission technology provided in an embodiment of the present invention;

[0020] Figure 4 This is a waveform diagram of the received signal of the acoustic emission signal provided in an embodiment of the present invention;

[0021] Figure 5 This is a flowchart of an online detection method for water pump flow rate and head using acoustic emission technology, provided in an embodiment of the present invention. Detailed Implementation

[0022] This application provides an online pump flow and head detection device and method utilizing acoustic emission technology, solving the problem of low accuracy in online pump flow and head detection in the prior art. It determines whether acoustic emission signal optimization is needed by using the quality parameters of the emitted signal, and simultaneously marks transient events. When a transient event is detected, it is classified and feedback is sent to preset personnel. If no transient event is detected, the pump flow and head are obtained by mapping the acoustic features extracted from the acoustic emission signal of non-transient events. Finally, this is used to evaluate the accuracy of the pump flow and head mapping and determine whether mapping optimization is needed, thus improving the accuracy of online pump flow and head detection.

[0023] To better understand the above technical solutions, the technical solutions of the present invention will be described below with reference to the accompanying drawings.

[0024] like Figure 1 One of the flowcharts shown is for online detection of water pump flow rate and head, such as... Figure 2The second flowchart of the online detection process for water pump flow and head, as shown, has the following specific logic: Acoustic emission signals are collected. If the signal-to-noise ratio (SNR) is greater than the reference SNR, a transient event is directly determined; otherwise, a transient event is determined after optimizing the first acoustic emission signal. If the kurtosis of the acoustic emission signal is greater than the reference kurtosis, and the transient SNR of the set target frequency band is greater than the transient reference SNR, it is marked as a candidate transient event, and the corresponding acoustic emission sensor is marked as a candidate acoustic emission sensor. If the number of candidate acoustic emission sensors is greater than the preset detection number, and the coherence coefficient between the acoustic emission signals collected by the candidate acoustic emission sensors is greater than the reference coherence coefficient, the candidate transient event is marked as a transient event; otherwise, it is marked as noise, and a second acoustic emission signal is optimized. After the second acoustic emission signal is optimized, acoustic features are extracted. If the kurtosis of the acoustic emission signal is not greater than the reference kurtosis, or the transient SNR of the set target frequency band is not greater than the transient reference SNR, feature extraction is directly performed on the collected acoustic emission signal. When a transient event is detected, it is classified, and the classification results and abnormal detection prompts are fed back to preset personnel. If the cavitation category output is "no cavitation," then the classifier predicts the probability. If the cavitation category output is "Level 1 cavitation," then a cavitation alert for the water pump is sent to the preset personnel at the Level 1 alert frequency. If the cavitation category output is "Level 2 cavitation," then a cavitation alert for the water pump is sent to the preset personnel at the Level 2 alert frequency. If the predicted probability is greater than the reference predicted probability, then a transient event judgment anomaly alert is sent to the preset personnel. Otherwise, if the number of prediction anomalies within the preset time period is less than the reference prediction anomaly number, then a classification ambiguity alert is sent to the preset personnel. If the variance of the predicted probability in each fold of the cross-validation is greater than the variance of the reference predicted probability, the number of folds in the cross-validation is increased by the amount of fold increase. Otherwise, the transient event is reclassified. If the output result of the cavitation category is still no cavitation after the transient event is reclassified, a prompt indicating that the transient event judgment is abnormal is sent to the preset personnel. If no transient event is detected, the pump flow rate and head are obtained by mapping the extracted acoustic features, and the mapping accuracy score is obtained. If the mapping accuracy score is greater than the reference mapping accuracy score, no mapping optimization is performed; otherwise, mapping optimization is performed.

[0025] This invention provides an online detection device for water pump flow rate and head using acoustic emission technology. For example... Figure 3 The diagram shows the structure of an online pump flow and head detection device using acoustic emission technology. The device includes: an acoustic emission sensor acquisition module, a transient event processing module, and a pump flow and head detection module.

[0026] The acoustic emission sensor acquisition module is used to collect acoustic emission signals caused by cavitation, flow fluctuations, and abnormal impacts through a preset number of acoustic emission sensors (set by preset personnel). This allows for comprehensive acquisition of acoustic emission signals from different locations during pump operation, improving the integrity and accuracy of signal acquisition and providing a more complete reflection of the pump's operating status. Based on the quality parameters of the acoustic emission signals, it determines whether to optimize the signals. If so, transient events reflecting cavitation phenomena are marked after optimization; otherwise, transient events are directly marked. This removes noise and other interference factors, improves the signal-to-noise ratio, and ensures accurate marking and analysis of transient events, providing reliable basic data for accurate detection of pump flow and head. The acoustic emission sensor is a device used to detect acoustic emission signals, converting weak elastic waves generated within materials into electrical signals for subsequent acquisition, processing, and analysis.

[0027] It should be added that, such as Figure 4 The waveform diagram of the received acoustic emission signal shown is as follows: the sampling rate of the waveform is 2MHz, the center frequency is 150kHz, the peak amplitude is about 5mV, and there is a waveform with obvious transient events around 1ms.

[0028] The transient event handling module is used to classify transient events when they are detected. This allows for a deeper understanding of the cavitation situation inside the water pump. Different types of cavitation may have different degrees of impact on the pump's performance. The classification results and abnormal detection prompts are fed back to designated personnel, which helps to identify problems in a timely manner and take measures to prevent the water pump from experiencing performance degradation or damage due to cavitation and other issues, thus ensuring the safe and stable operation of the water pump.

[0029] If no transient events are detected, it indicates that the water pump is in a relatively stable operating state. Acoustic features are extracted from the acoustic emission signals of non-transient events, and through a pre-established mapping relationship, these acoustic features are mapped to the water pump's flow rate and head data. This enables the real-time acquisition of the water pump's key performance parameters without interrupting its operation, providing an important basis for water pump operation monitoring and optimization. Non-transient events, in contrast to transient events, refer to the relatively stable acoustic emission signal state during water pump operation, without sudden changes, reflecting the water pump's operation under relatively stable conditions. Acoustic features represent parameters extracted from the acoustic emission signals that reflect the signal characteristics, such as frequency, amplitude, and duration, and can be used for the classification, analysis, and mapping of acoustic emission signals.

[0030] The pump flow rate and head detection module is used to map the acoustic features extracted from the acoustic emission signals of non-transient events to obtain the pump flow rate and head. It can promptly detect problems in the mapping process, such as inaccurate mapping relationships caused by dynamic changes in pump flow rate and head, providing direction for subsequent optimization. Based on the evaluation results, it determines whether to perform mapping optimization, which can continuously improve the accuracy of the mapping relationship between acoustic features and pump flow rate and head, thereby improving the accuracy and reliability of the entire device in detecting pump flow rate and head.

[0031] Furthermore, the process of determining whether to optimize the acoustic emission signal based on its quality parameters is as follows:

[0032] If the signal-to-noise ratio (SNR) of the acoustic emission signal is greater than the reference SNR, it is directly determined whether a transient event exists. Otherwise, after optimizing the first acoustic emission signal used to remove noise signals from the acoustic emission signal, it is determined whether a transient event exists. The SNR is the ratio of the signal power (such as acoustic emission characteristics related to the water pump's operating state) to the noise power of the acoustic emission signal. It can quickly screen out high-quality acoustic emission signals, avoid complex optimization processing of all acoustic emission signals, improve processing efficiency, and optimize poor-quality signals to effectively remove noise, enhance useful information in the signal, and lay the foundation for accurate judgment of transient events in the future.

[0033] To determine whether a transient event exists, the specific steps are as follows:

[0034] If the kurtosis of the acoustic emission signal is greater than the reference kurtosis, and the transient signal-to-noise ratio (SNR) of the set target frequency band is greater than the transient reference SNR, it is marked as a candidate transient event, and the corresponding acoustic emission sensor is marked as a candidate acoustic emission sensor. Judgment is based on the candidate acoustic emission sensor. Kurtosis is a statistical measure describing the steepness of the data distribution pattern, used to measure the sharpness of the acoustic emission signal amplitude distribution. For acoustic emission signals containing transient events (such as sudden signals generated by cavitation), the kurtosis is usually larger because transient events cause obvious peaks in the signal amplitude. The target frequency band is a pre-set frequency range based on the characteristics of acoustic emission signals that may be generated during pump operation, set by pre-defined personnel. Different operating conditions or fault types (such as cavitation) may generate stronger acoustic emission signals within a specific frequency band. By focusing on the target frequency band, related events can be detected and analyzed more effectively. The transient SNR represents the SNR of the acoustic emission signal within the target frequency band. The reference kurtosis, transient reference SNR, preset detection quantity, and reference coherence coefficient are set by pre-defined personnel.

[0035] If the kurtosis of the acoustic emission signal is not greater than the reference kurtosis, or the transient signal-to-noise ratio of the set target frequency band is not greater than the transient reference signal-to-noise ratio, then the features of the acquired acoustic emission signal are directly extracted for subsequent pump flow and head detection; thereby improving the efficiency of the entire pump flow and head detection process, while ensuring that the feature information of the signal can be obtained in a timely manner for subsequent analysis.

[0036] It is important to understand that comprehensively judging the existence of transient events from both the signal morphology and the signal strength in a specific frequency band improves the accuracy and reliability of transient event detection and reduces the possibility of false positives and false negatives.

[0037] The determination is based on candidate acoustic emission sensors, and the specific process is as follows:

[0038] If the number of candidate acoustic emission sensors is greater than the preset detection number, and the coherence coefficient between the acoustic emission signals collected by the candidate acoustic emission sensors is greater than the reference coherence coefficient, then the candidate transient event is marked as a transient event. Otherwise, it indicates that the misjudgment may be caused by noise interference, and the candidate transient event is marked as noise and a second acoustic emission signal optimization is performed to improve the quality of the acoustic emission signal. The coherence coefficient is used to measure the similarity between the signals collected by two acoustic emission sensors. If the coherence coefficient is larger, it indicates that the correlation between the signals collected by the two acoustic emission sensors is stronger.

[0039] It should be added that the specific method for obtaining kurtosis is as follows: subtract the average value of the acoustic emission signal to obtain a zero-mean signal, then calculate the fourth power of each data point in the zero-mean signal and sum them to obtain the fourth-order central moment, then calculate the variance of the zero-mean signal, and then take the square root to obtain the standard deviation, and finally substitute the fourth-order central moment and the standard deviation into the kurtosis formula to obtain the kurtosis; the specific method for obtaining coherence coefficient is as follows: perform fast Fourier transform on the acoustic emission signals collected by the two acoustic emission sensors respectively, then substitute the complex conjugate result of the fast Fourier transform into the cross-power spectral density formula to obtain the cross-power spectral density, and then substitute the result of the fast Fourier transform into the auto-power spectral density formula to obtain the auto-power spectral density, and finally substitute the cross-power spectral density and the auto-power spectral density into the coherence coefficient formula to obtain the coherence coefficient.

[0040] By statistically analyzing the number of candidate acoustic emission sensors and calculating the coherence coefficient between signals, the authenticity of transient events can be further confirmed from the perspective of the detection results and signal correlation of multiple sensors. This can effectively eliminate the influence of misjudgment by a single sensor or local noise interference, and improve the accuracy of transient event detection. For signals judged as noise, marking and optimizing the second acoustic emission signal can continuously improve signal quality, reduce the interference of noise on subsequent pump flow and head detection analyses, and ensure the stability and reliability of the entire pump flow and head detection process.

[0041] Furthermore, the first sound transmission signal was optimized, specifically as follows:

[0042] If the kurtosis of the acoustic emission signal is less than the reference kurtosis, the acoustic emission signal undergoes a first filtering process; otherwise, it undergoes a second filtering process. The change in the reference correlation coefficient is set by a pre-defined operator. Selecting different filtering methods based on the comparison between the kurtosis and the reference kurtosis allows for the adoption of more suitable processing methods for acoustic emission signals with different characteristics, improving the targeting and effectiveness of signal processing.

[0043] Specifically, the cutoff frequency of the first filter differs from that of the low-pass filter. The cutoff frequency of the first filter represents the result of mapping the center frequency of the acoustic emission signal into a cutoff frequency mapping set. This cutoff frequency mapping set is a collection of data obtained from a pre-set database representing the mapping relationship between the center frequency and cutoff frequency of the acoustic emission signal. The cutoff frequency mapping set is trained using cutoff frequency training data, which includes the center frequencies of acoustic emission signals from historical data over a specific time period, as well as cutoff frequencies set by professionals based on empirical rules. By determining the cutoff frequency through the cutoff frequency mapping set, the cutoff frequency of the first filter can adaptively adjust according to the center frequency of the acoustic emission signal, better adapting to signals with different center frequencies and improving the filtering effect.

[0044] Specifically, a second filtering process is performed, and the specific steps are as follows:

[0045] The first step is to perform variational mode decomposition on the acoustic emission signal by combining the number of decomposed modes and the penalty factor to obtain the acoustic emission signal components of each frequency band, so as to achieve the separation of signal components of different frequency bands. Variational mode decomposition is used to decompose non-stationary signals into a set of mode functions. By constructing and solving variational problems, the signal is decomposed into mode components with different center frequencies and bandwidths, thereby achieving the separation of signal components of different frequency bands.

[0046] The decomposed mode number represents the result obtained by mapping the number of peak values ​​of the acoustic emission signal into the decomposed mode number mapping set. The decomposed mode number mapping set is a collection obtained from a preset database representing the mapping relationship between the number of peak values ​​of the acoustic emission signal and the decomposed mode number. The decomposed mode number mapping set is trained using decomposed mode number training data, which includes the number of peak values ​​of the acoustic emission signal based on historical data for a historical time period, as well as the decomposed mode number set by professional technicians based on empirical rules. The penalty factor represents the result obtained by mapping the signal-to-noise ratio (SNR) of the acoustic emission signal into the penalty factor mapping set. The penalty factor mapping set is a collection obtained from a preset database representing the mapping relationship between the SNR of the acoustic emission signal and the penalty factor. The penalty factor mapping set is trained using penalty factor training data, which includes the SNR of the acoustic emission signal based on historical data for a historical time period, as well as the penalty factor set by professional technicians based on empirical rules. The peak position and number can be automatically detected using the scipy.signal.find_peaks() function in SciPy.

[0047] By using the decomposition mode number mapping set and the penalty factor mapping set, the decomposition mode number and penalty factor can be adaptively determined according to the peak number and signal-to-noise ratio of the acoustic emission signal. This makes the variational mode decomposition more consistent with the actual situation of the signal, improves the accuracy of signal component separation in different frequency bands, and enables accurate and effective processing of acoustic emission signals with different characteristics.

[0048] The second step involves integrating empirical mode decomposition (EMD) with the noise amplitude to obtain intrinsic mode function (EMF) components from the acoustic emission signal components of each frequency band after variational mode decomposition. This improves the stability and accuracy of the decomposition and reduces reconstruction errors. Integrated EMD is an improved method of EMD, which involves introducing adaptive noise into the original signal, performing multiple EMD decompositions, and finally averaging the decomposition results to improve the signal decomposition effect. The intrinsic mode function (EMF) represents the basic components obtained after signal decomposition. Each EMF has the same oscillation frequency and amplitude in a local range and satisfies certain conditions, such as the number of extrema being equal to or differing from the number of zero-crossings by no more than 1.

[0049] Among them, the noise amplitude represents the result obtained by mapping the signal-to-noise ratio of the signal into the noise amplitude mapping set. The noise amplitude mapping set is a collection obtained from a preset database that represents the mapping relationship between the signal-to-noise ratio of the acoustic emission signal and the noise amplitude. The noise amplitude mapping set is trained by noise amplitude training data, which includes the signal-to-noise ratio of the acoustic emission signal based on historical data of historical time periods, as well as the noise amplitude set by professional technicians based on empirical rules.

[0050] The noise amplitude mapping set determines the noise amplitude based on the signal-to-noise ratio, making the introduced adaptive noise more reasonable, improving the stability and accuracy of integrated empirical mode decomposition, effectively reducing reconstruction errors, and enabling more accurate decomposition of intrinsic mode function components, thus improving the pertinence, accuracy and stability of signal processing.

[0051] Third, if the change in the correlation coefficient of the extracted intrinsic mode function components is greater than the change in the reference correlation coefficient, then stop the integrated empirical mode decomposition and proceed to the fourth step; otherwise, continue the integrated empirical mode decomposition. The change in correlation coefficient represents the difference between the correlation coefficients obtained from two adjacent decompositions, and is used to quantify the degree of change in the correlation between the intrinsic mode function components and the acquired acoustic emission signal in each integrated empirical mode decomposition. Substitute each intrinsic mode function component and the acoustic emission signal into the covariance calculation formula to obtain the covariance corresponding to each intrinsic mode function component. Then, substitute the covariance corresponding to each intrinsic mode function component and the standard deviation between each intrinsic mode function component and the acoustic emission signal into the Pearson correlation coefficient calculation formula to obtain the correlation coefficient corresponding to each intrinsic mode function component. Finally, the average value of the correlation coefficients corresponding to all intrinsic mode function components is recorded as the correlation coefficient obtained in this decomposition.

[0052] By comparing the change in the reference correlation coefficient, it is possible to determine in a timely manner whether the decomposition has reached an appropriate level, avoid over-decomposition or under-decomposition, and improve decomposition efficiency.

[0053] The fourth step is to select the integrated empirical mode decomposition components with a signal reconstruction priority greater than that of the reference signal as effective integrated empirical mode decomposition components, and then reconstruct the selected effective integrated empirical mode decomposition components to obtain the first denoised transmitted signal.

[0054] The signal reconstruction priority refers to the result obtained by mapping the correlation coefficient, component energy ratio, and kurtosis corresponding to each intrinsic mode function component into the signal reconstruction priority mapping set. The signal reconstruction priority mapping set is a collection obtained from a preset database representing the mapping relationship between the correlation coefficient, component energy ratio, and kurtosis corresponding to each intrinsic mode function component and the signal reconstruction priority. The signal reconstruction priority mapping set is trained using signal reconstruction priority training data, which includes the correlation coefficient, component energy ratio, and kurtosis corresponding to each intrinsic mode function component based on historical data of historical time periods, as well as the signal reconstruction priority set by professional technicians based on empirical rules. The component energy ratio refers to the ratio of the energy of each intrinsic mode function component to the total signal energy, used to measure the proportion of energy of that component in the signal.

[0055] The signal reconstruction priority mapping set comprehensively considers the correlation coefficient, component energy ratio, and kurtosis to determine the signal reconstruction priority. It can screen out the intrinsic mode function components that contribute greatly to the signal reconstruction and have good quality, thereby improving the quality of the reconstructed signal and obtaining a more accurate first denoised emission signal. This provides a more reliable basis for subsequent analysis and application of acoustic emission signals.

[0056] Furthermore, the second acoustic transmission signal is optimized as follows:

[0057] If the first denoised transmitted signal (if the first acoustic transmitted signal has been optimized, then it is the first denoised transmitted signal; otherwise, it is the acoustic transmitted signal, and the same applies below) contains broadband noise but no periodic noise, then wavelet denoising is performed on the first denoised transmitted signal to suppress the broadband noise. Because broadband noise energy is dispersed, the multi-scale decomposition and processing method of wavelet denoising can effectively remove this noise dispersed in a wide frequency band, retain the main features of the signal, improve the signal-to-noise ratio, and make the subsequent analysis of the acoustic transmitted signal more accurate. Broadband noise means that the energy in the first denoised transmitted signal is not concentrated in the noise interference frequency band set by the preset personnel, but is widely distributed in a wide frequency range. This noise does not have obvious frequency concentration characteristics and will interfere with the detection and analysis of the effective acoustic transmitted signal. Periodic noise means that the energy is concentrated in the noise interference frequency band set by the preset personnel and has periodic repetition characteristics. It appears in a fixed frequency pattern and interferes with the effective information in the acoustic transmitted signal.

[0058] If the first denoised transmitted signal contains periodic noise but not broadband noise, then adaptive filtering is applied to the first denoised transmitted signal to remove the periodic noise. The adaptive filtering can dynamically adjust the filter parameters according to the characteristics of the periodic noise and in combination with parameters such as the reference noise signal, thereby effectively removing the periodic noise, reducing interference to the effective signal, and improving the signal quality.

[0059] If there is no broadband or periodic noise in the first denoised transmitted signal, then noise frequency fluctuation is judged. Specifically, if the noise frequency fluctuation value is not greater than the reference noise frequency fluctuation value (set by preset personnel), then the first denoised transmitted signal is subjected to a third filter (such as bandpass filtering); otherwise, the first denoised transmitted signal is subjected to a fourth filter (such as Kalman filtering). The noise frequency fluctuation value represents the standard deviation of the center frequency over a preset time period, used to quantify the degree of fluctuation of the noise signal over time in the frequency domain. By judging the noise frequency fluctuation and selecting an appropriate filtering method, it is possible to better adapt to noise with different characteristics, further improve the purity of the signal, and provide more reliable data for subsequent signal analysis.

[0060] If the first denoised transmitted signal contains broadband noise and periodic noise, then performing adaptive filtering followed by wavelet denoising on the first denoised transmitted signal can fully leverage the advantages of both adaptive filtering and wavelet denoising. This approach first addresses the periodic noise problem and then processes the broadband noise, thereby improving the overall denoising effect and comprehensively removing noise from the signal.

[0061] Wavelet denoising specifically involves: multi-scale decomposition of the first denoised transmitted signal using the optimal wavelet basis, decomposing the signal into different frequency sub-bands, processing each sub-band (such as thresholding), and finally reconstructing the signal to obtain the second denoised transmitted signal. The optimal wavelet basis refers to the wavelet basis with the largest correlation coefficient between the first denoised transmitted signal and the approximation coefficient selected from the candidate wavelet basis set (such as dbN, symN, coifN, biorN, etc.), which can better preserve the characteristics of the signal and improve the denoising effect. The approximation coefficient represents the result obtained by performing a single-level decomposition on each wavelet basis, which contains the main characteristic information of the signal. Here, the correlation coefficient refers to the Pearson correlation coefficient.

[0062] Choosing the optimal wavelet basis for multi-scale decomposition and processing can better preserve the detailed features of the signal and avoid signal distortion caused by excessive denoising. Through multi-scale decomposition, targeted processing can be carried out on the noise characteristics of different frequency sub-bands, improving the accuracy and effectiveness of denoising and ultimately obtaining a denoised transmission signal with a higher signal-to-noise ratio.

[0063] Adaptive filtering, specifically, involves combining a reference noise signal (set by pre-defined personnel), a step size factor, a filter order, and an adaptive filtering algorithm to filter a first denoised transmitted signal (if wavelet denoising has been performed, it becomes the second denoised transmitted signal; otherwise, it remains the first denoised transmitted signal) to obtain a second denoised transmitted signal. The step size factor represents the result of mapping the error rate of the acoustic transmitted signal into a step size factor mapping set. This step size factor mapping set is a collection obtained from a pre-defined database representing the mapping relationship between the error rate of the acoustic transmitted signal and the step size factor. The step size factor mapping set is trained using step size factor training data. The long-factor training data includes the error rate of change of acoustic emission signals based on historical data of historical time periods, and the step size factor set by professional technicians based on empirical rules. The filter order represents the result obtained by mapping the noise period into the filter order mapping set. The filter order mapping set is a set obtained from a preset database that represents the mapping relationship between the noise period of the acoustic emission signal and the filter order. The filter order mapping set is obtained by training the filter order training data, which includes the noise period of the acoustic emission signal based on historical data of historical time periods, and the filter order set by professional technicians based on empirical rules.

[0064] Among them, the error change rate is used to measure the dynamic change speed of error during acoustic emission signal processing. In acoustic emission signal processing, the error is defined as the difference between the original signal and the processed signal (such as the filtered signal). The error change rate is calculated by first-order difference. In order to reduce noise interference, a moving average window can be used to smooth the difference sequence to finally obtain the error change rate of the acoustic emission signal. The noise period refers to the repetition frequency of periodic noise (such as mechanical vibration and electrical interference). The acoustic emission signal is processed by removing the DC component (subtracting the mean) and normalizing. Then, the time domain signal is converted into the frequency domain by fast Fourier transform to obtain the frequency corresponding to the significant peak. Finally, the reciprocal of the frequency corresponding to the significant peak is recorded as the noise period.

[0065] Adaptive filtering, which combines a reference noise signal, a step size factor, and a filter order, can dynamically adjust filter parameters according to real-time noise changes, improving the adaptability and accuracy of filtering. Reasonable determination of the step size factor and filter order can ensure that the filter converges to the optimal solution quickly, while avoiding unstable conditions such as oscillations, resulting in a denoised transmission signal that is closer to the real signal.

[0066] In this embodiment, through the above steps, the most suitable filtering method can be used to process the signal according to the different types of noise (broadband noise, periodic noise) and the frequency fluctuation characteristics of the noise, so that the subsequent analysis of the acoustic emission signal is more accurate and reliable, and the effective information in the signal can be extracted better.

[0067] Furthermore, the specific process for classifying transient events is as follows:

[0068] Short-time Fourier transform (SFT) is used to extract features from acoustic emission signals in transient events. This involves segmenting the signal along the time axis and performing a Fourier transform on each segment to obtain frequency components at different time points, simultaneously reflecting both time and frequency domain characteristics. This effectively captures the time-frequency characteristics of acoustic emission signals in transient events. Since the frequency components of transient signals change rapidly over time, SFT provides information on the frequency distribution of the signal at different time points, offering rich features for subsequent classification and helping to more accurately distinguish different cavitation categories.

[0069] Instantaneous signal features are input into a random forest classifier (which classifies data by constructing multiple decision trees and combining their predictions. Each decision tree is trained on a random subset of the dataset and a subset of features, and the classification result is determined by voting. It can output an importance score for each feature in the classification process to evaluate the contribution of each feature to the classification). The random forest classifier outputs importance scores for each instantaneous signal feature (e.g., based on Gini impurity or information gain). This allows for the evaluation of the importance of each instantaneous signal feature to the cavitation classification. By understanding which features contribute significantly to the classification and which contribute less, it provides a basis for subsequent feature selection, helps to remove redundant features, and improves the efficiency and performance of the classification model.

[0070] The instantaneous signal features are sorted in descending order of importance score, and features ranked after the reference ranking (set by preset personnel) are removed to generate dimensionality-reduced instantaneous signal features. Reducing the number of features can reduce computational complexity, improve the training speed and prediction efficiency of the classifier, and remove unimportant features to avoid these features interfering with the classification results, thereby improving the accuracy and generalization ability of the classification.

[0071] The dimensionality-reduced instantaneous signal features are input into the AdaBoost classifier (Adaptive Boosting, which iteratively trains a series of weak classifiers, i.e., classifiers with slightly better classification performance than random guessing, and adjusts the sample weights based on the results of each training iteration, so that subsequent weak classifiers pay more attention to previously misclassified samples, and finally combine these weak classifiers into a strong classifier to improve classification accuracy) to obtain the cavitation category output. At the same time, the output also shows the predicted probability reflecting the classification accuracy of the AdaBoost classifier. The cavitation category output includes no cavitation, first-level cavitation, and second-level cavitation, with the anomaly degree of first-level cavitation being less than that of second-level cavitation.

[0072] The AdaBoost classifier improves classification accuracy by integrating multiple weak classifiers, thereby enhancing overall classification performance. The output predicted probabilities provide more information, helping to determine the credibility of the classification results. In practical applications, different decision-making strategies can be adopted based on the predicted probabilities to ensure accurate identification of different cavitation categories in complex transient events.

[0073] Furthermore, the cavitation category output is obtained, which then includes:

[0074] If the output result of the cavitation category is no cavitation, the judgment is not directly considered correct. Instead, the classifier predicts the probability to check for potential anomalies. This avoids ignoring potential cavitation problems due to classifier misjudgment and improves the accuracy of the judgment.

[0075] If the cavitation category output result is Level 1 cavitation, a cavitation alert for the water pump will be sent to the preset personnel at the Level 1 alert frequency. If the cavitation category output result is Level 2 cavitation, a cavitation alert for the water pump will be sent to the preset personnel at the Level 2 alert frequency, with the Level 1 alert frequency being lower than the Level 2 alert frequency. The Level 1 alert frequency, Level 2 alert frequency, reference prediction probability, reference prediction anomaly count, and reference prediction probability variance are set by the preset personnel. Different frequencies of alerts are provided based on the severity of cavitation, enabling relevant personnel to promptly understand the water pump's operating status and rationally arrange maintenance and repair work.

[0076] The classifier prediction probability determination is as follows:

[0077] If the predicted probability is greater than the reference predicted probability, a prompt indicating an anomaly in the transient event judgment is sent to a designated person. Otherwise, a judgment is made on the number of prediction anomalies. Specifically, if the number of prediction anomalies within a preset time period is less than the reference prediction anomaly number, a prompt indicating fuzzy classification is sent to a designated person. Otherwise, a judgment is made on the variance of the prediction probability. The number of prediction anomalies represents the number of times the classifier's predicted probability is not greater than the reference predicted probability. By statistically analyzing multiple prediction anomalies, the performance of the classifier can be evaluated more comprehensively, avoiding misjudgments caused by random factors. This allows for the rapid detection of abnormal prediction results from the classifier and timely feedback of problems, helping to ensure the reliability of pump flow and head detection.

[0078] The determination of the variance of the predicted probability is as follows:

[0079] If the variance of the predicted probability in each tradeoff of cross-validation (i.e., the variance of the predicted probability in different cross-validation tradeoffs, which measures the predictive instability of the classifier in different cross-validation tradeoffs) is greater than the reference predicted probability variance, then the number of cross-validation folds is increased by the fold increment. The fold increment represents the result obtained by mapping the predicted probability variance into the fold increment mapping set. The fold increment mapping set is a set obtained from a preset database that represents the mapping relationship between the predicted probability variance and the fold increment. The fold increment mapping set is trained using fold increment training data, which includes the predicted probability variance based on historical data for historical time periods, as well as the fold increment set set by professional technicians based on empirical rules.

[0080] If the variance of the predicted probability in each fold of cross-validation is not greater than the variance of the reference predicted probability, the transient event is reclassified. When the variance is too large, the number of folds is increased to improve the accuracy of model evaluation, thereby obtaining a more reliable cavitation category judgment.

[0081] If the cavitation category output result is still "no cavitation" after the transient event is reclassified, a prompt indicating an abnormal transient event judgment will be sent to the designated personnel. This alerts relevant personnel to the potential complex issues, prompting further in-depth investigation and ensuring the safe operation of the online pump flow and head monitoring device. Otherwise, based on the cavitation category output result, a pump cavitation prompt will be sent to the designated personnel at the appropriate frequency. Prompts will be sent promptly based on the new accurate results, enabling relevant personnel to take timely measures to address cavitation issues. This provides strong support for the normal operation and maintenance of the pump and effectively reduces the risk of pump failure due to cavitation problems.

[0082] Furthermore, the pump flow rate and head are obtained by mapping the acoustic features extracted from the acoustic emission signals of non-transient events, as detailed below:

[0083] The slope consistency score and the reference slope consistency score are processed by relative deviation to obtain the slope consistency error. The slope consistency score is used to quantify the accuracy of the pump's operating characteristic mapping. The pump's operating characteristics can usually be represented by a flow-head curve. The slope of this curve at different points reflects the changing characteristics of the pump's operating state. The slope consistency score is obtained by calculating the ratio of the slope of the actual mapped flow-head curve to the slope of the reference curve. The reference curve slope, reference slope consistency score, reference head overshoot, and reference flow-head correlation coefficient are set by preset personnel. In this application, relative deviation processing means calculating the ratio of the absolute value of the deviation between the actual value and the reference value to the reference value. By calculating the slope consistency error, the degree of deviation between the slope of the pump's operating characteristic curve obtained by actual mapping and the reference slope can be intuitively understood. The smaller the slope consistency error, the closer the slope of the actual mapped curve is to the reference slope, that is, the more accurate the pump's operating characteristic mapping is, and the more accurately it can reflect the pump's head change characteristics under different flow rates.

[0084] The head overshoot is compared with the reference head overshoot to obtain a head overshoot comparison score. The head overshoot represents the maximum deviation of the predicted head value from the theoretical head (set by preset personnel) during flow rate changes, reflecting the stability of the mapping; in this application, the comparison processing represents a ratio calculation. The head overshoot comparison score reflects the degree to which the stability of the head prediction during dynamic processes closely approximates the ideal situation. The higher the score, the closer the actual predicted head overshoot is to the reference overshoot, and the more stable the head prediction is during flow rate changes, thus better responding to dynamic changes during pump operation.

[0085] The flow-head correlation coefficient and the reference flow-head correlation coefficient are processed for relative deviation to obtain the flow-head correlation error. The flow-head correlation coefficient reflects the statistical correlation between the output flow rate and the head, and is quantified using the Pearson correlation coefficient. In statistics, the correlation coefficient is used to measure the strength of the linear relationship between two variables, and its value ranges from -1 to 1. The flow-head correlation error can measure the degree of deviation between the actual mapping of the relationship between flow rate and head and the ideal situation. The smaller the error, the closer the statistical correlation between flow rate and head in the actual mapping is to the reference situation.

[0086] A balance factor reflecting the influence of each parameter on the accuracy of pump flow and head mapping is introduced. Inverse assignment coupling is performed on the slope consistency error, head overshoot comparison score, and flow-head correlation error to obtain a mapping accuracy assessment score. This score is used to quantitatively evaluate the accuracy of pump flow and head mapping. The specific constraint expression for the mapping accuracy assessment score is as follows:

[0087] ;

[0088] In the formula, M represents the mapping accuracy assessment score, M1 represents the slope consistency error, M2 represents the head overshoot comparison score, M3 represents the flow-head correlation error, α1 represents the first mapping accuracy balance factor, α2 represents the second mapping accuracy balance factor, and α3 represents the third mapping accuracy balance factor.

[0089] The mapping accuracy balancing factors involved are obtained from a preset database, specifically including a first mapping accuracy balancing factor, a second mapping accuracy balancing factor, and a third mapping accuracy balancing factor; the sum of the three is 1. For example, the slope consistency error and the preset first mapping accuracy balancing factor form a first mapping accuracy balancing factor mapping set. The real-time slope consistency error is input into the first mapping accuracy balancing factor mapping set to obtain the first mapping accuracy balancing factor. The first mapping accuracy balancing factor represents the degree of influence of the slope consistency error on the mapping accuracy assessment score; the second mapping accuracy balancing factor represents the degree of influence of the head overshoot comparison score on the mapping accuracy assessment score; and the third mapping accuracy balancing factor represents the degree of influence of the flow-head correlation error on the mapping accuracy assessment score.

[0090] In this embodiment, by introducing a balance factor, multiple factors such as slope consistency, dynamic stability, and flow-head correlation are comprehensively considered. This allows for a more comprehensive and reasonable quantitative assessment of the accuracy of pump flow and head mapping, providing a basis for pump mapping optimization and selection.

[0091] Furthermore, based on the evaluation results, a decision is made on whether to perform mapping optimization. The specific process is as follows:

[0092] If the mapping accuracy score is greater than the reference mapping accuracy score (set by preset personnel), it means that the current mapping effect is good and no mapping optimization is needed to avoid unnecessary mapping optimization; otherwise, mapping optimization should be performed.

[0093] The mapping optimization specifically involves: combining the sliding window length in independent component analysis (used to divide continuous signal data into multiple windows for processing, the window length determining the amount of data processed each time), decoupling the second denoised emission signal (if the second acoustic emission signal has been optimized, it is the second denoised emission signal; if the second acoustic emission signal has not been optimized but the first acoustic emission signal has been optimized, it is the first denoised emission signal; if the first acoustic emission signal has not been optimized, it is the acoustic emission signal, and so on) segment by segment within each window to separate the influence of various interference factors on the acoustic emission signal; the use of the sliding window can transform continuous signal processing into segmented signal processing, reducing the complexity of signal processing, while independent component analysis can effectively separate interference factors, improve signal purity, and provide more accurate data for subsequent signal analysis and processing.

[0094] If a previous sliding window exists, it means that this is not the first window. Using the decoupled acoustic emission signal of the previous sliding window as the initial acoustic emission signal of the current sliding window helps to improve the stability and accuracy of the current window processing. If no previous sliding window exists, the second denoised emission signal is used as the initial acoustic emission signal. Using the decoupling result of the previous sliding window as the initial signal of the current window can improve the accuracy of the starting point of the current window processing, reduce the accumulation of errors in the processing process, and make the decoupling process more stable and efficient.

[0095] After each sliding window processing, signal reconstruction is performed based on the decoupled acoustic emission signal and the second denoised emission signal to recombine the decoupled signal into a form close to the original signal. The parameters in the independent component analysis (including the learning rate, elements in the mixing matrix, and parameters of the nonlinear function) are optimized according to the parameter correction ratio to improve the convergence speed of the independent component analysis.

[0096] The parameter correction ratio represents the result obtained by mapping the residual signal and the reference mapping accuracy score into the parameter correction ratio mapping set. The parameter correction ratio mapping set is a collection obtained from a preset database that represents the mapping relationship between the residual signal, the reference mapping accuracy score, and the parameter correction ratio. The parameter correction ratio mapping set is trained using parameter correction ratio training data, which includes the residual signal and the reference mapping accuracy score based on historical data from historical time periods, as well as the parameter correction ratio set by professional technicians based on empirical rules. The residual signal is obtained by performing residual processing on the acoustic emission signal obtained from signal reconstruction and the second denoised emission signal, reflecting the difference between the acoustic emission signal obtained from signal reconstruction and the second denoised emission signal.

[0097] Signal reconstruction allows for the evaluation of decoupling effectiveness. By comparing the signal with the original signal, problems in the decoupling process can be identified. Parameter optimization, based on the residual signal and reference mapping, accurately assesses the scores and adjusts the parameters of independent component analysis, thereby improving the convergence speed of independent component analysis, enabling the decoupling process to reach the optimal solution more quickly, and improving the efficiency and quality of signal processing.

[0098] If the mutual information of the decoupled acoustic emission signals is greater than the reference mutual information threshold, it indicates that the decoupled signals are highly correlated and a real-time feedback mechanism is required; otherwise, no real-time feedback mechanism is adopted. By judging the correlation of the decoupled signals through mutual information, the changes in the signals can be detected in a timely manner. When the mutual information is greater than the reference mutual information threshold, a real-time feedback mechanism is adopted, which can dynamically adjust the parameters of independent component analysis to adapt to changes in the signals and ensure the stability of the decoupling effect.

[0099] It should be added that the specific method for obtaining mutual information is as follows: the value ranges of the two decoupled acoustic emission signals are divided into several small intervals, a two-dimensional histogram is constructed to statistically analyze the joint probability distribution and marginal probability distribution, the joint probability distribution and marginal probability distribution are substituted into the mutual information calculation formula to obtain the mutual information of the two signals, and finally the average value of all the obtained mutual information is recorded as the mutual information of the decoupled acoustic emission signals, which is used to measure the correlation between the decoupled acoustic emission signals.

[0100] The real-time feedback mechanism is as follows:

[0101] By combining gradient descent with the mutual information method, the parameters of independent component analysis are dynamically adjusted to minimize mutual information, adapt to signal changes, make the decoupled signals more independent, improve the signal processing effect, and adapt to the characteristics of signals at different times.

[0102] The optimized acoustic features are input into a pump parameter detection model used to evaluate pump parameters in real time. The pump parameters include pump flow rate and head, providing an important basis for pump operation monitoring and control.

[0103] By combining the pump parameters output by the model with the pump parameters obtained from mapping, weighted coupling is performed using weights set by pre-defined personnel to reflect the importance of each pump parameter output method, thus obtaining the actual pump parameters. This weighted coupling comprehensively considers both the model output and the mapped pump parameters, and by incorporating weights to reflect the importance of different parameter output methods, it yields more accurate pump parameters that better reflect actual conditions, thereby improving the reliability of pump parameter detection.

[0104] It should be added that the pump parameter detection model is a dynamic model based on a long short-term memory network. The input data is the time series of acoustic emission signals, and the output data is the pump parameters. It is suitable for processing dynamically changing time series data, such as the nonlinear characteristics of signals caused by dynamic changes in flow rate and head.

[0105] Furthermore, if the actual pump flow deviation is greater than the reference pump flow deviation, and the actual pump head deviation is not greater than the reference pump head deviation, then the current adjustment will be to correct the inverter frequency. The reference pump flow deviation, reference pump head deviation, reference pump flow, and reference pump head are set by preset personnel. The inverter frequency refers to the target frequency or set frequency that the inverter is expected to achieve. It exists in the inverter's control circuit and is a digital or analog signal. The inverter output frequency refers to the frequency of the AC power actually applied to the motor terminals by the inverter unit. It exists on the output side of the inverter and is the power supply frequency that actually drives the motor to rotate.

[0106] Pump flow deviation represents the result of processing the relative deviation between the actual pump flow and the reference pump flow. Pump head deviation represents the result of processing the relative deviation between the actual pump head and the reference pump head. Corrected inverter frequency represents the result of mapping the actual pump flow deviation and the current inverter frequency into the inverter frequency mapping set. The inverter frequency mapping set is a collection obtained from a preset database that represents the mapping relationship between the actual pump flow deviation, the current inverter frequency, and the corrected inverter frequency. The inverter frequency mapping set is trained using inverter frequency training data, which includes the actual pump flow deviation and the current inverter frequency based on historical data for a historical time period, as well as the corrected inverter frequency set by professional technicians based on empirical rules.

[0107] By adjusting the frequency of the inverter to change the speed of the water pump motor, the water pump flow rate is adjusted to be closer to the reference flow rate. While keeping the head basically unchanged, the flow output of the water pump is optimized, improving the accuracy and stability of the water pump operation and meeting the flow requirements.

[0108] If the actual pump flow deviation is not greater than the reference pump flow deviation, but the actual pump head deviation is greater than the reference pump head deviation, then the current inverter output frequency is adjusted to the corrected inverter output frequency. The corrected inverter output frequency represents the result obtained by mapping the actual pump flow deviation and the current inverter output frequency into the inverter output frequency mapping set. The inverter output frequency mapping set is a collection obtained from a preset database that represents the mapping relationship between the actual pump flow deviation, the current inverter output frequency, and the corrected inverter output frequency. The inverter output frequency mapping set is obtained through training with inverter output frequency training data, which includes the actual pump flow deviation and the current inverter output frequency based on historical data for a historical time period, as well as the corrected inverter output frequency set by professional technicians based on empirical rules.

[0109] By adjusting the frequency of the inverter to change the operating state of the water pump, the pump head is brought closer to the reference head. Under the premise of ensuring that the flow rate basically meets the requirements, the pump head performance is optimized to ensure that the pump can operate at a suitable head.

[0110] If the actual pump flow rate deviation is greater than the reference pump flow rate deviation, and the actual pump head deviation is greater than the reference pump head deviation, a pump malfunction alert will be sent to the designated personnel. The abnormal pump operation will be promptly reported to relevant personnel so that maintenance personnel can quickly identify the problem and take corresponding measures, such as checking the pump equipment and troubleshooting the cause of the malfunction, to avoid more serious consequences caused by abnormal pump operation and ensure the safe and stable operation of the pump.

[0111] If the actual pump flow rate deviation is not greater than the reference pump flow rate deviation, and the actual pump head deviation is not greater than the reference pump head deviation, then online monitoring of pump flow rate and head continues. Under the condition that the pump is operating normally, continuous monitoring of pump parameters can promptly identify potential problems, enabling preventative maintenance, ensuring long-term stable pump operation, and avoiding unnecessary frequency adjustments, thereby improving the operating efficiency of the pump flow rate and head monitoring device.

[0112] like Figure 5 The flowchart shown illustrates a method for online detection of water pump flow rate and head using acoustic emission technology. This embodiment of the invention provides a method for online detection of water pump flow rate and head using acoustic emission technology, the method comprising:

[0113] S1. Acoustic emission signals caused by various operating conditions are collected by a predetermined number of deployed acoustic emission sensors. Based on the quality parameters of the acoustic emission signals, it is determined whether acoustic emission signal optimization is required. If so, transient events reflecting cavitation phenomena are marked after acoustic emission signal optimization; otherwise, transient events reflecting cavitation phenomena are directly marked. By collecting acoustic emission signals and judging their quality, the reliability and accuracy of the signals used in subsequent analysis can be ensured. Optimizing the signals can further improve signal quality, reduce noise interference, and thus more accurately mark transient events of cavitation phenomena, providing reliable basic data for subsequent analysis of the pump's operating status.

[0114] S2 classifies transient events upon detection and feeds back the classification results and anomaly alerts to designated personnel. If no transient event is detected, the pump flow rate and head are obtained by mapping the acoustic features extracted from the acoustic emission signals of non-transient events. Classifying and feeding back transient events allows relevant personnel to promptly grasp abnormal pump operation and take proactive maintenance or adjustment measures to avoid equipment damage and performance degradation. In cases where no transient event is detected, extracting the acoustic features of non-transient events and mapping them to obtain pump flow rate and head enables online monitoring of pump operating parameters without the need for additional complex flow and head measurement equipment, reducing costs and improving the convenience and real-time performance of the monitoring.

[0115] S3 uses acoustic features extracted from the acoustic emission signals of non-transient events to map the pump flow rate and head, determining whether to optimize the mapping based on the evaluation results. By evaluating and optimizing the mapping accuracy, the precision of mapping pump flow rate and head using acoustic emission signals can be continuously improved, making the detection results more reliable and accurate. This helps to more accurately grasp the pump's operating status, providing strong data support for optimized pump operation and maintenance, and further improving the efficiency and reliability of online pump flow rate and head detection.

[0116] In summary, this application embodiment comprehensively collects acoustic emission signals through deployed acoustic emission sensors. Based on signal quality parameters, it determines whether acoustic emission signal optimization is necessary. This avoids low-quality signals directly entering subsequent analysis stages, reducing false alarms and missed alarms, thereby improving the stability and accuracy of signal processing. When signal quality is poor, acoustic emission signal optimization is performed first, followed by marking transient events. When signal quality is high, transient events are marked directly. This ensures efficient identification of cavitation events under different conditions, improving the sensitivity and accuracy of transient event detection. Furthermore, the detected transient events are classified, and the classification results are fed back to designated personnel. This can be used for early warning of cavitation and prevention. To prevent damage to the pump impeller or casing caused by severe cavitation and improve maintenance response speed, the acoustic features extracted from the acoustic emission signal of non-transient events are then mapped to obtain the pump flow rate and head. Compared with traditional mechanical or electrical sensors, no additional pressure sensor or flow meter is required, thereby reducing hardware costs and avoiding poor stability of acoustic emission signal quality due to temperature and pressure shocks. Finally, the accuracy of the current mapping result is evaluated, and it is determined whether to trigger mapping optimization. It can adapt to different fluid media and operating conditions, thus maintaining higher accuracy and good adaptability in the pump flow rate and head mapping mechanism, and improving the accuracy of pump flow rate and head mapping.

[0117] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0118] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

Claims

1. An online pump flow rate and head detection device utilizing acoustic emission technology, characterized in that, The device includes: an acoustic emission sensor acquisition module, a transient event processing module, and a water pump flow and head detection module; The acoustic emission sensor acquisition module is used to acquire acoustic emission signals caused by various operating conditions through a preset number of acoustic emission sensors. Based on the quality parameters of the acoustic emission signals, it determines whether to optimize the acoustic emission signals. If so, the transient events reflecting cavitation phenomena are marked after the acoustic emission signals are optimized; otherwise, the transient events are marked directly. The transient event processing module is used to classify transient events when they are detected, and to feed back the classification results and abnormal detection prompts to preset personnel. If no transient event is detected, the water pump flow rate and head are obtained by mapping the acoustic features extracted from the acoustic emission signals of non-transient events. The pump flow rate and head detection module is used to evaluate the accuracy of the mapping based on the mapped pump flow rate and head, and to determine whether to optimize the mapping based on the evaluation results.

2. The online detection device for water pump flow rate and head using acoustic emission technology according to claim 1, characterized in that, The specific process for determining whether to optimize the acoustic emission signal based on its quality parameters is as follows: If the signal-to-noise ratio of the acoustic emission signal is greater than the reference signal-to-noise ratio, it is directly determined whether a transient event exists; otherwise, after optimizing the first acoustic emission signal used to remove noise signals from the acoustic emission signal, it is determined whether a transient event exists. The determination of whether a transient event exists specifically involves: If the kurtosis of the acoustic emission signal is greater than the reference kurtosis, and the transient signal-to-noise ratio of the set target frequency band is greater than the transient reference signal-to-noise ratio, then it is marked as a candidate transient event, and the corresponding acoustic emission sensor is marked as a candidate acoustic emission sensor, and the judgment is made based on the candidate acoustic emission sensor; If the kurtosis of the acoustic emission signal is not greater than the reference kurtosis, or the transient signal-to-noise ratio of the set target frequency band is not greater than the transient reference signal-to-noise ratio, then feature extraction is performed directly on the acquired acoustic emission signal. The specific process for determining based on candidate acoustic emission sensors is as follows: If the number of candidate acoustic emission sensors is greater than the preset detection number, and the coherence coefficient between the acoustic emission signals collected by the candidate acoustic emission sensors is greater than the reference coherence coefficient, then the candidate transient event is marked as a transient event; otherwise, it is marked as noise and a second acoustic emission signal optimization is performed to improve the acoustic emission signal quality.

3. The online detection device for water pump flow rate and head using acoustic emission technology according to claim 2, characterized in that, The first acoustic emission signal is optimized as follows: If the kurtosis of the acoustic emission signal is less than the reference kurtosis, the acoustic emission signal is subjected to the first filtering; otherwise, the acoustic emission signal is subjected to the second filtering. The cutoff frequency of the first filter is different from that of the low-pass filter. The cutoff frequency of the first filter represents the result obtained by mapping the center frequency of the acoustic emission signal into the cutoff frequency mapping set. The second filtering process is as follows: The first step is to perform variational mode decomposition on the acoustic emission signal by combining the number of decomposition modes and the penalty factor to obtain the acoustic emission signal components of each frequency band, so as to achieve the separation of signal components of different frequency bands; The number of decomposed modes represents the result obtained by mapping the number of peak values ​​of the acoustic emission signal into the decomposed mode number mapping set, and the penalty factor represents the result obtained by mapping the signal-to-noise ratio of the acoustic emission signal into the penalty factor mapping set. The second step involves integrating empirical mode decomposition on the acoustic emission signal components of each frequency band after variational mode decomposition, based on the noise amplitude, to obtain intrinsic mode function components, thereby improving the stability and accuracy of the decomposition and reducing reconstruction errors. The noise amplitude represents the result obtained by mapping the signal-to-noise ratio of the signal into a noise amplitude mapping set; The third step is to stop the integrated empirical mode decomposition if the change in the correlation coefficient of the extracted intrinsic mode function components is greater than the change in the reference correlation coefficient, and then proceed to the fourth step; otherwise, continue the integrated empirical mode decomposition. The change in the correlation coefficient is used to quantify the degree of change in the correlation between the intrinsic mode function components and the acquired acoustic emission signal in each integrated empirical mode decomposition. The fourth step is to select the integrated empirical mode decomposition components with a signal reconstruction priority greater than that of the reference signal as effective integrated empirical mode decomposition components, and then reconstruct the selected effective integrated empirical mode decomposition components to obtain the first denoised transmitted signal. The signal reconstruction priority refers to the result obtained by mapping the correlation coefficient, component energy ratio and kurtosis of the input signal reconstruction priority mapping set corresponding to each intrinsic mode function component.

4. The online detection device for water pump flow rate and head using acoustic emission technology according to claim 3, characterized in that, The second acoustic transmission signal optimization is as follows: If there is broadband noise in the first denoised transmission signal, and there is no periodic noise with energy concentrated in the set noise interference frequency band, then wavelet denoising is performed on the first denoised transmission signal to suppress broadband noise. The broadband noise indicates that the energy in the first denoised transmission signal is not concentrated in the set noise interference frequency band. If there is periodic noise in the first denoised transmitted signal but no broadband noise, then adaptive filtering is performed on the first denoised transmitted signal to remove the periodic noise. If there is no broadband noise or periodic noise in the first denoised transmission signal, then noise frequency fluctuation judgment is performed. The noise frequency fluctuation judgment is specifically as follows: if the noise frequency fluctuation value is not greater than the reference noise frequency fluctuation value, then the first denoised transmission signal is subjected to third filtering; otherwise, the first denoised transmission signal is subjected to fourth filtering. The noise frequency fluctuation value is used to quantify the degree of fluctuation of the center frequency of the noise signal in the frequency domain over time. If the first denoised transmitted signal contains broadband noise and periodic noise, then the first denoised transmitted signal is subjected to adaptive filtering and then wavelet denoising. The wavelet denoising specifically involves: performing multi-scale decomposition and processing on the first denoised transmitted signal using the optimal wavelet basis to reconstruct the signal and obtain the second denoised transmitted signal. The optimal wavelet basis refers to the wavelet basis with the largest correlation coefficient between the first denoised transmitted signal selected from the candidate wavelet basis set and the approximation coefficient. The approximation coefficient represents the result obtained by performing a single-level decomposition on each wavelet basis. The adaptive filtering specifically involves processing the first denoised transmission signal to obtain the second denoised transmission signal by combining the reference noise signal, step size factor, filter order, and adaptive filtering algorithm. The step size factor represents the result obtained by mapping the error change rate of the acoustic transmission signal into the step size factor mapping set, and the filter order represents the result obtained by mapping the noise period into the filter order mapping set.

5. The online detection device for water pump flow rate and head using acoustic emission technology according to claim 1, characterized in that, The specific process for classifying transient events is as follows: Short-time Fourier transform is used to extract features from acoustic emission signals in transient events to obtain instantaneous signal features; Input the instantaneous signal features into the random forest classifier and output the importance score of each instantaneous signal feature; The instantaneous signal features are sorted in descending order of importance score, and features ranked after the reference ranking are removed to generate the dimensionality-reduced instantaneous signal features. The instantaneous signal features after dimensionality reduction are input into the AdaBoost classifier to obtain the cavitation class output result, and the predicted probability reflecting the classification accuracy of the AdaBoost classifier is also output. The cavitation category output includes no cavitation, first-level cavitation, and second-level cavitation, with the degree of anomalousness of first-level cavitation being less than that of second-level cavitation.

6. The online detection device for water pump flow rate and head using acoustic emission technology according to claim 5, characterized in that, After obtaining the cavitation category output result, the following is also included: If the output result of the cavitation category is no cavitation, then the classifier prediction probability is determined. If the cavitation category output result is Level 1 cavitation, then send a water pump cavitation notification to the preset personnel at the Level 1 notification frequency; If the cavitation category output result is level two cavitation, then a water pump cavitation prompt is sent to the preset personnel at the level two prompt frequency, wherein the level one prompt frequency is less than the level two prompt frequency; The classifier prediction probability determination is specifically as follows: If the predicted probability is greater than the reference predicted probability, a prompt indicating an abnormal transient event is sent to a preset personnel; otherwise, a prediction anomaly count is performed. Specifically, if the prediction anomaly count is less than the reference prediction anomaly count within a preset time period, a prompt indicating fuzzy classification is sent to a preset personnel; otherwise, a prediction probability variance is performed. The prediction anomaly count represents the number of times the classifier's predicted probability is not greater than the reference predicted probability. The determination of the predicted probability variance is specifically as follows: If the variance of the predicted probability in each fold of cross-validation is greater than the variance of the reference predicted probability, then the number of folds in cross-validation is increased by the fold increase amount, where the fold increase amount represents the result obtained by mapping the predicted probability variance into the fold increase amount mapping set; otherwise, the transient event is reclassified. If the cavitation category output result is still no cavitation after the transient event is reclassified, a prompt indicating an abnormal transient event determination will be sent to the preset personnel; otherwise, a prompt indicating pump cavitation will be sent to the preset personnel according to the prompt frequency based on the cavitation category output result.

7. The online detection device for water pump flow rate and head using acoustic emission technology according to claim 1, characterized in that, The method for mapping the acoustic features extracted from the acoustic emission signals of non-transient events to obtain the pump flow rate and head is as follows: The slope consistency score and the reference slope consistency score are processed to obtain the slope consistency error. The slope consistency score is used to quantify the accuracy of the pump's operating characteristic mapping. The head overshoot is compared with the reference head overshoot to obtain a head overshoot comparison score. The head overshoot represents the maximum deviation of the predicted head value from the theoretical head during the flow rate change process. The flow-head correlation coefficient and the reference flow-head correlation coefficient are processed to obtain the flow-head correlation error. The flow-head correlation coefficient reflects the statistical correlation between the output flow rate and the head. A balance factor reflecting the influence of each parameter on the accuracy of pump flow and head mapping is introduced. The slope consistency error, head overshoot comparison score and flow-head correlation error are inversely assigned and coupled to obtain the mapping accuracy evaluation score. The mapping accuracy evaluation score is used to quantitatively evaluate the accuracy of pump flow and head mapping.

8. The online detection device for water pump flow rate and head using acoustic emission technology according to claim 7, characterized in that, The specific process for determining whether to perform mapping optimization based on the evaluation results is as follows: If the accurate mapping score is greater than the reference accurate mapping score, then no mapping optimization is performed; otherwise, mapping optimization is performed. Specifically, the mapping optimization is as follows: By combining the sliding window length in independent component analysis, the second denoised emission signal is decoupled segment by segment to separate the influence of each interference factor on the acoustic emission signal; Use the decoupled acoustic emission signal of the previous sliding window as the initial acoustic emission signal of the current sliding window; After each sliding window processing is completed, the signal is reconstructed based on the decoupled acoustic emission signal and the second denoised emission signal, and the parameters in the independent component analysis are optimized according to the parameter correction ratio to improve the convergence speed of the independent component analysis. The parameter correction ratio represents the result obtained by mapping the residual signal and the reference mapping accurate evaluation score into the parameter correction ratio mapping set. The residual signal is obtained by performing residual processing on the acoustic emission signal obtained by signal reconstruction and the second denoised emission signal, reflecting the difference between the acoustic emission signal obtained by signal reconstruction and the second denoised emission signal. If the mutual information of the decoupled acoustic emission signals is greater than the reference mutual information threshold, a real-time feedback mechanism is adopted; otherwise, a real-time feedback mechanism is not adopted. The real-time feedback mechanism is specifically as follows: By combining gradient descent with dynamic adjustment of independent component analysis parameters to minimize mutual information and adapt to signal changes; The optimized acoustic features are input into a pump parameter detection model used to evaluate pump parameters in real time to obtain pump parameters, which include pump flow rate and head. By combining the weights that reflect the importance of the output methods of each pump parameter, the pump parameters output by the model and the pump parameters obtained by mapping are weighted and coupled to obtain the actual pump parameters.

9. The online detection device for water pump flow rate and head using acoustic emission technology according to claim 1, characterized in that, Also includes: If the actual pump flow deviation is greater than the reference pump flow deviation, and the actual pump head deviation is not greater than the reference pump head deviation, then the current inverter frequency will be adjusted to the corrected inverter frequency. The pump flow deviation represents the result of processing the relative deviation between the actual pump flow and the reference pump flow; the pump head deviation represents the result of processing the relative deviation between the actual pump head and the reference pump head; and the corrected inverter frequency represents the result of mapping the actual pump flow deviation and the current inverter frequency input inverter frequency mapping set. If the actual pump flow deviation is not greater than the reference pump flow deviation, and the actual pump head deviation is greater than the reference pump head deviation, then the current inverter output frequency is adjusted to the corrected inverter output frequency. The corrected inverter output frequency represents the result obtained by mapping the actual pump flow deviation and the current inverter output frequency into the inverter output frequency mapping set. If the actual pump flow rate deviation is greater than the reference pump flow rate deviation, and the actual pump head deviation is greater than the reference pump head deviation, a pump malfunction alert will be sent to the designated personnel. If the actual pump flow deviation is not greater than the reference pump flow deviation, and the actual pump head deviation is not greater than the reference pump head deviation, then continue to perform online pump flow and head monitoring.

10. A method for online detection of water pump flow rate and head using acoustic emission technology, characterized in that, The method includes: S1. Acoustic emission signals caused by various operating conditions are collected by deploying a preset number of acoustic emission sensors. Based on the quality parameters of the acoustic emission signals, it is determined whether to optimize the acoustic emission signals. If so, the transient events reflecting cavitation phenomena are marked after the acoustic emission signals are optimized; otherwise, the transient events are marked directly. S2, when a transient event is detected, the transient event is classified, and the classification result and the detection anomaly prompt are fed back to the preset personnel. If no transient event is detected, the water pump flow rate and head are obtained by mapping the acoustic features extracted from the acoustic emission signal of the non-transient event. S3. An evaluation is performed based on the pump flow rate and head obtained from the mapping to reflect the accuracy of the mapping, and a decision is made on whether to optimize the mapping based on the evaluation results.

Images