Noise data processing method based on cavitation flow characteristics of propeller

Abstract

The invention relates to the technical field of state monitoring and fault diagnosis of ship power devices, and provides a noise data processing method based on cavitation flow characteristics of a propeller. According to the method, observable dynamic frequencies such as leaf frequency and cavitation bubble falling main frequency and statistical non-stationary quantities such as a frequency spectrum broadening factor, self-correlation time, broadband energy and kurtosis are unified in the same analysis framework, a mapping model of a cavitation state and signal processing parameters is constructed, analysis parameters such as FFT, STFT and CWT are adaptively configured, and the analysis parameters of the cavitation state and the signal processing parameters are analyzed. And multi-scale feature extraction of non-stationary pressure pulsation signals under different cavitation types and cavitation intensities is realized. According to the method, the cavitation state and type are identified and evaluated on the basis of the extracted multi-domain features, the threshold and mapping parameters are reversely updated according to the identification result, a closed loop of state judgment, parameter mapping, multi-domain analysis, feature output, cavitation identification and parameter updating is formed, and the precision and engineering applicability of propeller cavitation monitoring and diagnosis are improved.

Description

Technical Field

[0001] This invention belongs to the field of marine power plant condition monitoring and fault diagnosis technology, specifically relating to an adaptive signal analysis method for processing propeller cavitation flow noise data. Background Technology

[0002] The propulsion performance, radiated noise level, and structural safety of ships and marine engineering equipment are closely related to propeller cavitation behavior. Cavitation not only leads to a decrease in propulsion efficiency but also causes blade cavitation erosion, structural vibration, and high-intensity underwater noise, posing a serious threat to the stealth and reliability of ships. Therefore, accurate identification and early diagnosis of propeller cavitation are crucial for propulsion system optimization, condition monitoring, and condition-based maintenance.

[0003] Existing cavitation identification technologies mainly rely on two approaches: one is the empirical threshold method based on macroscopic operating parameters such as cavitation number, rotational speed, and torque; the other is the pattern recognition method based on acoustic signals, extracting features such as broadband sound level, blade frequency amplitude, or cavitation shedding frequency. However, propeller cavitation flow exhibits strong non-stationary and multi-scale dynamic evolution characteristics, and the fixed-parameter signal processing flow relied upon by the above methods is severely mismatched with the physical process, specifically manifested in the following ways:

[0004] 1. Insufficient coupling between feature extraction and physical mechanism: Fixed analysis parameters such as window length and time-frequency resolution cannot simultaneously meet the differentiated requirements of different physical mechanisms such as tip vortex oscillation, sheet cavitation shedding, and cloud cavitation collapse for frequency domain / time domain resolution, resulting in low feature discrimination.

[0005] 2. Mismatch between parameters and signal dynamic characteristics: During cavitation evolution, the signal statistical characteristic spectrum broadens and autocorrelation time changes drastically. Fixed parameters cannot remain optimal throughout the process, resulting in unstable main frequency identification and distorted broadband energy estimation.

[0006] 3. Limited input feature quality of the recognition model: Distortion and instability in front-end feature extraction directly restrict the performance ceiling and robustness of the cavitation recognition model.

[0007] In summary, there is an urgent need for a core method capable of adaptively adjusting signal processing strategies based on cavitation state. This invention aims to construct an adaptive signal processing and recognition framework driven by propeller cavitation state. By dynamically correlating cavitation features with signal analysis parameters, it achieves a paradigm shift from fixed strategies to state-adaptive approaches. This allows the extracted high-quality time-frequency features to be used independently for cavitation identification, or embedded as enhancement features into existing monitoring systems, thereby systematically improving the accuracy and stability of cavitation diagnosis. Summary of the Invention

[0008] To address the problem that existing propeller cavitation identification technologies typically rely on fixed-parameter spectral and time-frequency analysis to obtain key features, leading to instability in feature extraction (such as cavitation shedding frequency, spectral broadening, autocorrelation time, and broadband energy) and insufficient coupling with cavitation physical mechanisms under strongly non-stationary and multi-scale evolutionary conditions, this invention aims to provide an adaptive noise data processing and cavitation identification method based on propeller cavitation flow characteristics. By unifying observable dynamic frequencies such as blade frequency and cavitation shedding frequency with statistical non-stationary quantities such as spectral broadening factor, autocorrelation time, broadband energy, and kurtosis within the same analytical framework, a parameter mapping model based on cavitation state is constructed. This model achieves a closed-loop logic of "cavitation state discrimination—parameter mapping—multi-domain signal analysis—feature output—cavitation identification—parameter update," used for discriminating propeller cavitation state and type. Simultaneously, it is compatible with existing identification methods as feature inputs, improving the accuracy, stability, and engineering applicability of propeller cavitation identification and monitoring.

[0009] The objective of this invention is achieved through the following technical solution:

[0010] This invention discloses a noise data processing method based on propeller cavitation flow characteristics, comprising the following steps:

[0011] Step 1: Calculate the cavitation number and spectral broadening factor based on the collected signals and corresponding operating parameters. Based on the value range of the cavitation number and spectral broadening factor, divide the propeller operating conditions into at least one of cavitation state A, cavitation state B, or cavitation state C to achieve preliminary identification of cavitation state.

[0012] Step 2: Based on the cavitation state initially identified in Step 1 and the non-stationary characteristics of the signal, construct a family of signal analysis parameter mapping functions based on the cavitation state, and adaptively determine the key parameters for spectrum analysis, time-frequency analysis, and wavelet analysis using the family of mapping functions.

[0013] Step 3: Based on the key parameter set obtained in Step 2, perform joint signal processing of FFT, STFT and CWT on the standardized propeller cavitation noise signal to extract multi-dimensional noise time-frequency features characterizing the propeller cavitation intensity and cavitation evolution characteristics. The noise time-frequency features are used to evaluate the propeller cavitation state, or the noise time-frequency features are used to form a noise time-frequency feature set for identification.

[0014] The process also includes step four: using at least one or more noise time-frequency features from the noise time-frequency feature set obtained in step three, combined with propeller speed, cavitation number, and blade frequency (BPF), to construct a cavitation identification input vector. Based on this cavitation identification input vector, the propeller cavitation type is classified and identified. The cavitation type includes at least one or more of the following: no cavitation, tip vortex cavitation, sheet cavitation, cloud-like cavitation, and combinations thereof.

[0015] Furthermore, the preliminary identification and classification of cavitation states in step one is based on the following criteria:

[0016] based on The cavitation state is divided into:

[0017]

[0018] threshold , , Determined by actual operating conditions. Cavitation number Spectral broadening factor Characterizing the degree of spectral spread, it is used to measure the spectral broadening characteristics of cavitation signals and is expressed as:

[0019]

[0020] This represents the power spectral density.

[0021] Furthermore, in step one, when the cavitation number is greater than the first critical value and the spectral broadening factor is less than the first broadening threshold, the broadband noise energy in the pressure signal spectrum is not significantly increased, and the energy is highly concentrated in the blade frequency and its harmonics, exhibiting a discrete spectral structure. This corresponds to a large autocorrelation time scale and a small spectral broadening factor. Above the critical value, the cavitation state is determined to be state A, which corresponds to the propeller without cavitation, providing a benchmark state for subsequent cavitation type determination.

[0022] Furthermore, in step one, when the cavitation number is between the first critical value and the second critical value, and the spectral broadening factor is between the first broadening threshold and the second broadening threshold, and the pressure signal spectrum simultaneously exhibits broadband energy components of both low and high frequencies, and the cavitation intensity is significantly increased compared to state A and exhibits non-stationary characteristics, When the cavitation level drops below the critical value, the cavitation state is determined to be state B, which corresponds to a mixed cavitation state in which tip vortex cavitation, sheet cavitation and cloud cavitation coexist.

[0023] Furthermore, in step one, when the cavitation number is less than the second critical value and the spectral broadening factor is greater than the second broadening threshold, the cavitation shedding frequency and its harmonic energy in the pressure signal spectrum are significantly enhanced, and the cavitation intensity reaches the preset high intensity threshold. Further reducing the cavitation state, the cavitation state is determined to be state C, which corresponds to the cavitation state dominated by sheet-like cavitation and cloud-like cavitation.

[0024] Furthermore, the implementation method for step two is as follows:

[0025] Based on the cavitation states and nonstationarity characteristics identified in step one, we define a family of parameter mapping functions M:

[0026]

[0027] The time scale for autocorrelation is expressed as follows:

[0028]

[0029] in Based on the first zero or threshold and the autocorrelation function A fixed upper limit for points.

[0030] The cavitation shedding frequency is represented as follows:

[0031]

[0032] Among them, calibration coefficient This is used to ensure the continuous definition of the main frequency index under different cavitation states.

[0033] For kurtosis, it is expressed as follows:

[0034]

[0035] in This is the instantaneous pressure value. For average pressure, The number of samples. When the frequency is large, the system automatically increases the acoustic resolution per octave to improve the ability to capture transient events.

[0036] according to Determine the key parameters for FFT analysis, namely the window length. Target frequency resolution Based on the cavitation state setting:

[0037]

[0038] in, For fixed sampling frequency. To distinguish the proportionality constant, and with Consistency check, For correction constants:

[0039]

[0040] The window function selection for FFT is determined based on the cavitation state initially identified in step one:

[0041] State A: When the signal is close to a periodic signal, a rectangular window is preferred to obtain the best frequency resolution.

[0042] State B / C: Non-stationary enhancement with significant random and transient components. To effectively suppress spectral leakage, the Hanning window is preferred.

[0043] according to Determine the key parameters for STFT analysis, namely the time-frequency window length. overlap rate .

[0044] Time-frequency window length The adaptive definition based on non-stationary characteristics is:

[0045]

[0046] in, As a reference value for spectral broadening, the window length and Conversely, when non-stationarity is enhanced, the window length is automatically shortened to improve temporal resolution. As the periodic coverage factor, This is a correction constant.

[0047] Overlap rate Adaptive is defined as:

[0048]

[0049] The larger the value, the more intense the cavitation, and the automatic increase in the temporal spectrum overlap rate, maintaining continuity and information integrity.

[0050] The window length of the STFT is determined based on the cavitation state initially identified in step one.

[0051] State A: To obtain high frequency resolution to confirm spectral stability and observe for a sufficiently long period, a long window is used. The window length is set to cover multiple leaf frequency cycles.

[0052] State B and State C: The analysis objective is to capture the dynamic evolution of cavitation, and short-window analysis is used.

[0053] according to Determine the key parameters for CWT analysis, namely, sound part resolution (VPO) and number of scales. .

[0054] The sound resolution VPO is expressed as follows:

[0055]

[0056] It is a proportionality constant.

[0057] Scale-Pseudo-Frequency Correspondence :

[0058]

[0059] Sampling time interval, scale Covering the target frequency band according to logarithmic distribution:

[0060]

[0061]

[0062] The number of scales per octave controls the resolution in the scale direction.

[0063] Based on the cavitation state initially identified in step one and the obtained kurtosis Determine the wavelet basis functions in CWT :

[0064]

[0065] State A: The analysis objective is to detect potential, submerged, weak nonstationarities; Morlet wavelets are recommended because they have a good response to periodic signals and can be used to verify the purity of signals within the frequency band.

[0066] State B: The analysis targets are diverse, including quasi-periodic vortex oscillations and transient collapses. Morlet wavelets are recommended, as they are similar to the waveform of oscillating signals and can provide good resolution in the scale-frequency domain.

[0067] State C: Focus on impact events caused by cavitation detachment and collapse. Daubechies wavelets are recommended because their tight support characteristics and vanishing moment are conducive to accurately capturing and locating transient impacts.

[0068] Furthermore, the implementation method for step three is as follows:

[0069] Based on the parameter mapping results from step two, the non-stationary feature parameter set is called. and analysis parameter set , , For standardized signals Perform FFT, STFT, and CWT signal processing. Through this adaptively optimized analysis workflow, extract multidimensional features corresponding to the physical mechanism of the current cavitation state from the analysis results, including but not limited to: frequency domain spectral broadening factor and broadband energy characteristics; autocorrelation time scale and time domain statistical characteristics; time-frequency modulation intensity and energy intermittency characteristics; wavelet multi-scale impact and high-frequency concentration characteristics.

[0070] The multidimensional features constitute a noise time-frequency feature set, which is used for subsequent identification of cavitation state and cavitation type, and serves as part of the feature index in existing cavitation identification methods.

[0071] Furthermore, the cavitation type identification results and their confidence levels from step four are fed back to steps one and two to adjust the threshold for dividing the cavitation number and the spectral broadening factor, as well as the parameter weights in the family of signal analysis parameter mapping functions. This ensures that when the noise pressure signal acquired subsequently re-executes the initial cavitation state identification and parameter mapping, the signal analysis parameters adaptively converge to the current propeller cavitation state, thereby forming a closed-loop data processing flow that includes adaptive parameter updates.

[0072] Beneficial effects:

[0073] 1. Uniformity: The noise data processing method based on propeller cavitation flow characteristics disclosed in this invention unifies the cavitation shedding main frequency with non-stationary statistical characteristics such as spectral broadening, autocorrelation time, broadband energy, and kurtosis into the same parameterization framework. Through the parameter mapping function driven by cavitation state, frequency domain analysis, time-frequency analysis, and wavelet multi-scale analysis can work in a coordinated manner under the same system.

[0074] 2. Continuity: The noise data processing method based on propeller cavitation flow characteristics disclosed in this invention, when the dominant frequency cannot be robustly identified, uses the equivalent dominant frequency definition and autocorrelation time scale and other indicators to jointly constrain the analysis parameters, so as to achieve a continuous transition of parameters from no cavitation to mixed cavitation and then to the dominant stage of sheet / cloud cavitation, and avoid the discontinuity of analysis results due to the jump of dominant frequency.

[0075] 3. Adaptability: The noise data processing method based on propeller cavitation flow characteristics disclosed in this invention automatically adjusts the analysis parameters according to the cavitation state and non-stationarity. The FFT window length, STFT window length and overlap rate, wavelet basis and scale density are all adaptively set according to the characteristics of spectrum broadening, autocorrelation time and broadband energy, covering the multi-scale information needs of different cavitation stages.

[0076] 4. Robustness: The noise data processing method based on propeller cavitation flow characteristics disclosed in this invention improves the stability and comparability of characteristic indicators under different operating conditions and reduces the impact of measurement noise and operating condition fluctuations by means of unified frequency band setting, blade frequency and harmonic narrowband suppression, and autocorrelation integral upper limit strategy.

[0077] 5. Engineering Applicability and Compatibility: The noise data processing method based on propeller cavitation flow characteristics disclosed in this invention can be directly used for cavitation state and cavitation type discrimination in this method, or it can be used as one of the feature indicators in existing cavitation identification technologies. It can be combined with traditional operating condition parameters, sound pressure level indicators, etc. to form an extended feature vector, which can significantly improve the identification accuracy and robustness without changing the existing identification framework. At the same time, through the closed-loop parameter update mechanism, online adaptive optimization can be achieved in water tunnel tests and actual ship monitoring, which has good engineering portability.

[0078] 6. This invention discloses a noise data processing method based on propeller cavitation flow characteristics. By establishing a mapping relationship between cavitation state and signal processing parameters, and adaptively configuring the analysis parameters of Fast Fourier Transform (FFT), Short-Time Fourier Transform (STFT), and Continuous Wavelet Transform (CWT), it achieves multi-scale feature extraction of non-stationary pressure pulsation signals under different cavitation types and intensities. This method is further used for the identification and diagnosis of propeller cavitation state and cavitation type, providing a high-quality data foundation and feature support for cavitation monitoring and early warning in engineering practice. Attached Figure Description

[0079] Figure 1 The flowchart of a noise data processing method based on propeller cavitation flow characteristics disclosed in this invention shows the complete closed-loop process from signal acquisition to feature extraction, state discrimination, parameter mapping and analysis output.

[0080] Figure 2 The logic diagram for cavitation state discrimination and adaptive parameter mapping shows the relationship between FFT / STFT / CWT parameters as inputs and outputs after logical discrimination, using non-stationary feature parameters as inputs.

[0081] Figure 3 A schematic diagram of the integrated feature parameter system architecture: illustrating the hierarchical relationship and fusion mechanism between the dominant frequency feature and the non-stationary feature parameter family.

[0082] Figure 4 Adaptive adjustment relationship diagram for signal analysis parameters: showing the adaptive change trend of FFT window length, overlap rate and wavelet resolution under cavitation state evolution (A→B→C).

[0083] Figure 5 The result diagram for implementing Case 1.

[0084] Figure 6 The result diagram for implementing Case 2.

[0085] Figure 7 The result diagram for implementing Case 3. Detailed Implementation

[0086] Example 1

[0087] like Figure 1 As shown in this embodiment, a noise data processing method based on propeller cavitation flow characteristics is disclosed and applied to a closed cavitation water tunnel under high cavitation number conditions. This method is used to obtain "cavitation-free" baseline characteristics and verify the stability of the method under cavitation-free conditions. In this engineering scenario, a high-speed camera and hydrophone are deployed in the test section to achieve synchronous acquisition of propeller flow field images and noise signals. The cross-sectional dimensions of the water tunnel test section in this embodiment are approximately 0.6m × 0.6m, and the length is approximately 1.0m; the propeller model has a diameter D = 0.25m and a number of blades... In this embodiment, the rotational speed is selected. By increasing the static pressure in the test section, the cavitation number is reduced. The cavitation number was controlled within the range of 4.6-8.0, at which point there were virtually no visible cavitation bubbles on the blade surface in the high-speed camera images. This was used as the baseline for cavitation-free operation. The high-speed camera frame rate was 10,000 fps, and the hydrophone sampling frequency was... The analysis frequency band is uniformly set to 80Hz-80kHz.

[0088] Step S1: Preliminary identification of cavitation state

[0089] Under the aforementioned high-altitude operating conditions, propeller noise and pressure signals were collected using a hydrophone. The original signal is subjected to DC removal, bandpass filtering (80Hz-80kHz), and amplitude normalization to obtain the preprocessed signal. Leaf frequency BPF Calculations show that the Hz frequency under this operating condition is approximately 105 Hz. The preprocessed signal is then divided according to the window length. The time window is divided into segments, with each time window corresponding to approximately 5,000 frames of high-speed video.

[0090] The cavitation number was calculated based on the static pressure, flow velocity, and water temperature of the test section. For the high cavitation number condition in this embodiment, the cavitation number The calculated value is approximately in the range of 4.6-8.0. The power spectral density is calculated for the signal within each time window. A narrowband region is constructed around the leaf frequency and its first three harmonics, while the remaining frequency bands are defined as broadband regions. The spectral broadening factor is calculated according to the definitions given earlier in the manual. With broadband energy index , Used to characterize the broadening of the power spectrum relative to the center frequency near the blade frequency. Used to characterize the target frequency band Broadband energy after deducting leaf frequency and its harmonics.

[0091] On the other hand, with preprocessed signals Calculate the autocorrelation function And obtain the autocorrelation time scale according to the definition in the instruction manual. Cavitation shedding frequency index According to the optimal definition in the manual, based on the autocorrelation time scale... and calibration coefficients Determined jointly.

[0092] Under the operating conditions of this embodiment, based on the statistical results, we can obtain: Actual measurements are typically less than 0.2, indicating a wideband energy index. With cavitation-free noise baseline compared to The autocorrelation timescale satisfies ,and It does not exhibit a clear concentrated peak. (Comprehensive eigenvector) According to the state discrimination rules in the specification, this embodiment can be determined to be state A, that is, the non-cavitation reference state.

[0093] Step S2: Signal analysis parameter mapping based on cavitation state

[0094] Under state A, based on the non-stationary eigenvectors and parameter mapping function family The key analysis parameters for FFT, STFT, and CWT are adaptively configured. In this embodiment, kurtosis... The value is close to the Gaussian level, approximately between 3 and 3.5, indicating fewer transient impact events; therefore, the acoustic resolution VPO is set to a base value of 12. To accurately confirm the purity and stability of the blade frequencies and their harmonic spectral lines, the FFT window length... Set as The order of magnitude is 262,144 points, corresponding to a frequency resolution. The window function uses a rectangular window to avoid artificial spectral broadening.

[0095] STFT window length With overlap rate according to The functional relationship given in the middle follows and Adaptive change. For state A, since Smaller size, weaker non-stationarity. The larger value is automatically selected so that multiple leaf frequency periods are covered within each window. In this embodiment, the typical window length is about 4096 points, and the overlap rate is about 0.5. In the CWT analysis, the scale range is set according to the frequency corresponding to 0.5–20 times the leaf frequency. The preferred wavelet basis is Morlet, and the wavelet scale is uniformly distributed on the logarithmic axis. The number of scales per octave is set to VPO=12 to examine whether there are potential weak non-stationary structures under cavitation-free conditions.

[0096] Step S3: Multi-domain joint signal processing and noise time-frequency feature extraction

[0097] Under the above parameter configuration, a combined analysis of FFT, STFT, and CWT was performed on the standardized noise signal, and the results are as follows. Figure 5 As shown, in the frequency domain, it can be observed that the power spectrum energy is highly concentrated in the BPF and its 2nd–3rd harmonics, with a low broadband energy level and a low spectral broadening factor. Within the aforementioned small range; in the time-frequency domain, the STFT energy along the BPF and harmonic directions maintains a narrow-band ridge shape over time, indicating that the signal is close to steady-state periodic noise; in the wavelet time-scale plane, the high-frequency scale energy is extremely weak, and no obvious transient impulse mode maxima chain is observed. Therefore, the obtained... , , , The relevant statistical quantities constitute the noise time-frequency characteristic baseline under vacancy conditions, and the characteristics of subsequent Examples 2 and 3 are compared with this baseline.

[0098] Step S4: Propeller cavitation type identification

[0099] Based on the feature vector and operating parameters obtained in step S3, the operating condition in this embodiment is marked as "no cavitation (NC)" in the cavitation identification model. These samples serve as the benchmark category in the training set, used to constrain the identification model under high cavitation number and low cavitation parameters. ,Small ,big The combined features output the non-cavitation type, providing a reference for the subsequent identification of cavitation conditions.

[0100] Step S5: Recognition Result Feedback and Closed-Loop Parameter Update

[0101] In Example 1, the noise identification results were completely consistent with the manual interpretation from the high-speed camera, both indicating that the propeller was in a cavitation-free state. During closed-loop updates, this example primarily uses it as a family of parameter mapping functions. Baseline calibration condition: Set the upper limit of the cavitation number corresponding to state A. Upper limit of spectrum broadening factor Autocorrelation time scale reference value and broadband energy baseline This will provide a clear benchmark for comparison in subsequent operating conditions where cavitation exists.

[0102] Example 2

[0103] This embodiment uses the same water tunnel and measurement layout and processing methods as Embodiment 1, such as... Figure 1As shown, the difference lies in adjusting the static pressure of the test section to allow the propeller to operate within a medium cavitation number range, thereby achieving a "mixed cavitation" condition where tip vortex cavitation (TVC), sheet cavitation (SC), and cloud cavitation (CC) coexist. The cavitation number... The pressure is controlled within the range of 2.5-3.6. At this time, there is a stable tip vortex cavitation zone near the blade tip, and periodically generated cavitation patches appear on the suction surface of the blade. These patches then entrain and form cavitation clouds at a certain chord length position, and the clouds collapse intermittently downstream.

[0104] In this embodiment, the processing flow of step S1 is the same as in embodiment 1, and the cavitation number is calculated. Spectrum broadening factor Broadband energy index Autocorrelation time scale and cavitation shedding frequency index Within this interval, statistical results show... , Approximately 0.4–0.6, with relatively high broadband energy. The relevant time scale is shortened to approximately kurtosis Slightly higher than in Example 1, the cavitation shedding frequency index is... A clear peak appears near a multiple of BPF. According to the state discriminant, this operating condition satisfies the conditions of state B, i.e. and Therefore, it was judged to be a mixed cavitation state with moderate cavitation intensity.

[0105] In step S2, the parameter mapping function family The analysis parameters are automatically adjusted based on the feature vector of state B. The FFT window length is determined by Example 1. shortened to (65536 points), the window function is switched to Hanning window to balance frequency resolution and spectral leakage control; Approximately 2048 points, overlap rate The value was increased to approximately 0.7, which allows for sufficient time-frequency slices to be obtained within a single cloud formation-collapse cycle. CWT continues to use Morlet wavelets, but the number of scales per octave (VPO) is increased from 12 to approximately 14–16, reflecting the multi-scale resolution requirements brought about by increased kurtosis and broadband energy.

[0106] In step S3, multi-domain joint analysis is performed on the signal under the above parameter configuration. The results are as follows: Figure 6 As shown, in the FFT spectrum, except for the BPF and its harmonics, the low-frequency and mid-to-high-frequency broadband energy is significantly increased, compared to Example 1. There is a significant increase; in the STFT time-scale spectrum, the energy bands near the leaf frequency and its harmonics broaden and modulate over time, clearly corresponding to the periodic changes in sheet cavitation length and cloud formation and collapse cycles in high-speed imaging; intermittent energy fringes appear in the mesoscale region of the CWT time-scale map, consistent with cloud scale variations. This results in... , , , Features such as multi-scale modulation intensity constitute the noise feature vector under mixed cavitation conditions with medium cavitation number.

[0107] In step S4, the aforementioned feature vectors and operating condition parameters are combined to form a cavitation recognition input vector, which is then input into a multi-class classifier based on a support vector machine. After training with the "non-cavitation NC" samples from Example 1 and the "hybrid cavitation (TVC+SC+CC)" samples from Example 2, the recognition model can identify cavitation under unknown operating conditions. , , , By combining features, stable operating conditions are identified as mixed cavitation types. Compared with the baseline method that uses fixed-parameter FFT+STFT to extract features and identify cavitation types, on the same sample set, the cavitation shedding frequency identification error obtained by the adaptive method of this invention is reduced from about 10% to below 5%, and the accuracy of identifying multiple cavitation types is improved by about 8 to 10 percentage points.

[0108] In the closed-loop update stage of step S5, the noise identification result is compared with the high-speed camera interpretation result to fine-tune the weight of state B, so that state B more accurately corresponds to the cavitation level of "clearly visible in engineering but not yet in the strong impact stage" in the parameter space, while maintaining a clear distinction from the cavitation-free baseline of Example 1.

[0109] Example 3

[0110] This embodiment still uses Figure 1 The experimental method shown further reduces the static pressure in the test section and increases the incoming flow velocity, causing the propeller to operate in a low cavitation number range, with a typical cavitation number. Controlled The blade's suction surface is largely covered by sheet-like cavitation. The sheet cavitation periodically falls off at the blade tip and mid-diameter area to form large-volume clouds. These clouds collapse within a short distance downstream, exhibiting obvious strong unsteady state and strong impact characteristics in the image, corresponding to the "sheet-like + cloud-like cavitation dominance" situation in cavitation state C.

[0111] Under this operating condition, the non-stationary eigenvectors are calculated according to the method in step S1. The results show... Spectrum broadening factor Greater than 0.8, relative broadband energy Typically greater than 4, autocorrelation timescale kurtosis Significantly improved to above 6, cavitation shedding frequency index Its harmonics exhibit sharp spectral peaks in the FFT spectrum. According to the state criterion, this operating condition simultaneously satisfies… and The condition is met, therefore it is determined to be in cavitation state C.

[0112] In step S2, the parameter mapping function family The FFT window length was further shortened to (16384 points) To enhance the time response capability to broadband broadening and high-frequency changes, the window function adopts the Hanning window; Shortened to approximately 1024 points, overlap rate The value was increased to approximately 0.85 to obtain higher temporal resolution and continuity within a single cloud shedding-collapse cycle; the CWT analysis was switched to Daubechies 4th or 5th order wavelets, and the wavelet acoustic resolution VPO was increased to 18–20, making the energy peak of a single impact event appear localized and sharp on the time-scale plane.

[0113] In step S3, the signal is subjected to multi-domain analysis using the above parameter configuration, and the results are as follows. Figure 7 As shown, the broadband energy in the FFT spectrum is significantly increased, and the leaf frequency and its harmonic spectral lines are submerged by the high-energy background. and All were significantly higher than in Example 2; the STFT time-scale spectrum showed significantly broadened energy patches in multiple frequency bands, highly consistent with the temporal locations of cloud formation and collapse in high-speed camera images; the CWT time-scale map showed numerous energy ridges and isolated peaks at high-frequency scales, corresponding to intense impact events. Multi-scale impact intensity, high-frequency energy concentration, and impact event counts were extracted from these analytical results and compared with… , , , The combination of equal statistics forms the noise time-frequency feature vector under state C.

[0114] In step S4, the aforementioned feature vectors are input into the cavitation identification model. After joint training with the samples from Examples 1 and 2, the model can stably output the "sheet-like cavitation + cloud-like cavitation dominance (SC+CC)" type under low cavitation number conditions. In 15 sets of low cavitation number condition tests, compared with the results of manual interpretation by high-speed cameras, the false negative rate of the method of the present invention in judging the risk level of strong cavitation decreased from about 18% for the fixed parameter method to about 6%, significantly improving the robustness and early warning capability of identifying strong cavitation conditions.

[0115] In the closed-loop update process of step S5, this embodiment utilizes the consistency and differences between the "noise identification result + image interpretation result" in state C to correct the weights, making the configuration of the parameter mapping function family more closely match the actual impact characteristics under strong unsteady conditions. At the same time, it does not destroy the non-cavitation and mixed cavitation parameter ranges that have been calibrated in embodiments 1 and 2, thereby completing the self-consistent realization of the entire closed-loop process of "state discrimination - parameter mapping - multi-domain analysis - cavitation identification - parameter update" in an engineering scenario.

[0116] After a period of testing and calibration, the method of the present invention can identify the cavitation state and type of subsequent working conditions through the noise channel without relying on continuous manual frame-by-frame video interpretation, and automatically mark the corresponding high-speed camera segments. This realizes the transformation from "manual identification + noise assistance" to "noise-driven identification + image spot check verification" in engineering applications, significantly improving experimental efficiency and data utilization.

[0117] In summary, this embodiment organically combines high-frame-rate image observation of cavitation water tunnels with hydrophone noise signal processing in engineering applications using real-time high-speed cameras. Through state-driven adaptive parameter mapping and closed-loop update mechanisms, it constructs a cavitation noise data processing and identification method that can be run online and calibrated on-site. This not only improves the accuracy and robustness of cavitation identification but also reduces the reliance on manual image interpretation in engineering experiments.

[0118] The method described in this embodiment is applicable to the processing and feature recognition of propeller cavitation flow noise signals under different cavitation intensities. It is particularly suitable for engineering applications in cavitation water tunnel test rigs equipped with real-time high-speed camera systems. While a high-speed camera is deployed in the water tunnel test section to capture the propeller cavitation evolution process in real time, a hydrophone collects underwater noise signals near the propeller. The noise signals are then processed online or quasi-online using the method of this invention, enabling automatic labeling and condition screening of cavitation morphology in the high-speed camera footage. This reduces the workload of manual frame-by-frame interpretation and provides quantitative indicators for subsequent cavitation dynamics analysis and propeller scheme optimization.

[0119] In this embodiment, the method is deployed in the experimental data acquisition and processing system to perform real-time or post-processing analysis on synchronous noise data, and to align the cavitation state identification results with high-speed camera images, which helps in the coordinated engineering diagnosis of "image-acoustics-operating conditions".

[0120] The above detailed description further illustrates the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A noise data processing method based on propeller cavitation flow characteristics, characterized in that: Includes the following steps: Step 1: Calculate the cavitation number and spectrum broadening factor based on the collected signals and corresponding operating parameters. Based on the value range of the cavitation number and spectrum broadening factor, divide the propeller operating condition into at least one of cavitation state A, cavitation state B, or cavitation state C to achieve preliminary identification of cavitation state. Step 2: Based on the cavitation state initially identified in Step 1 and the non-stationary characteristics of the signal, construct a family of signal analysis parameter mapping functions based on the cavitation state, and adaptively determine the key parameters for spectrum analysis, time-frequency analysis and wavelet analysis using the family of mapping functions. Step 3: Based on the key parameter set obtained in Step 2, perform joint signal processing of FFT, STFT and CWT on the standardized propeller cavitation noise signal to extract multi-dimensional noise time-frequency features characterizing the propeller cavitation intensity and cavitation evolution characteristics. The noise time-frequency features are used to evaluate the propeller cavitation state, or the noise time-frequency features are used to form a noise time-frequency feature set for identification.

2. The method as described in claim 1, characterized in that, It also includes step four: Based on at least one or more noise time-frequency features obtained in step three, and the propeller speed, cavitation number, and blade frequency (BPF) are combined to form a cavitation identification input vector; the propeller cavitation type is classified and identified based on the cavitation identification input vector, and the cavitation type includes at least one or more of the following types: no cavitation, tip vortex cavitation, sheet cavitation, cloud cavitation, and combinations thereof.

3. The method as described in claim 2, characterized in that: The cavitation type identification results and their confidence levels in step four are fed back to steps one and two to adjust the threshold for dividing the cavitation number and the spectral broadening factor, as well as the parameter weights in the family of signal analysis parameter mapping functions. This ensures that when the noise pressure signal acquired subsequently re-executes the initial cavitation state identification and parameter mapping, the signal analysis parameters adaptively converge to the current propeller cavitation state, thus forming a closed-loop data processing flow that includes adaptive parameter updates.

4. The method as described in claim 1, characterized in that: The preliminary identification and classification of cavitation states in step one is based on the following criteria: based on The cavitation state is divided into: threshold , , Determined by actual operating conditions; cavitation number Spectrum broadening factor Characterizing the degree of spectral spread, it is used to measure the spectral broadening characteristics of cavitation signals and is expressed as: This represents the power spectral density.

5. The method as described in claim 1, characterized in that: The implementation method for step two is as follows: Based on the cavitation states and nonstationarity characteristics identified in step one, we define a family of parameter mapping functions M: The time scale for autocorrelation is expressed as follows: in Based on the first zero or threshold and the autocorrelation function A fixed upper limit for points; The cavitation shedding frequency is represented as follows: Among them, calibration coefficient Used to ensure the continuous definition of the main frequency index under different cavitation states; For kurtosis, it is expressed as follows: in This is the instantaneous pressure value. For average pressure, For the sample size; when When the frequency is large, the system automatically increases the acoustic resolution per octave to improve the ability to capture transient events; according to Determine the key parameters for FFT analysis, namely the window length. Target frequency resolution Based on the cavitation state setting: in, For fixed sampling frequency; To distinguish the proportionality constant, and with Consistency check, For correction constants: The window function selection for FFT is determined based on the cavitation state initially identified in step one: State A: The signal is close to a periodic signal, so a rectangular window is preferred to obtain the best frequency resolution; State B / C: Non-stationary enhancement with significant random and transient components. To effectively suppress spectral leakage, the Hanning window is preferred. according to Determine the key parameters for STFT analysis, namely the time-frequency window length. overlap rate ; Time-frequency window length The adaptive definition based on non-stationary characteristics is: in, As a reference value for spectral broadening, the window length and Inversely, when nonstationarity is enhanced, the window length is automatically shortened to improve temporal resolution; As the periodic coverage factor, This is a correction constant; Overlap rate Adaptive is defined as: The larger the value, the more intense the cavitation, and the time-spectral overlap rate automatically increases, maintaining continuity and information integrity; The window length of the STFT is determined based on the cavitation state initially identified in step one. State A: To obtain high frequency resolution to confirm spectral stability and observe a sufficiently long period of time, a long window is used; the window length is set to cover multiple leaf frequency cycles. State B and State C: The analysis objective is to capture the dynamic evolution of cavitation, and short-window analysis is used; according to Determine the key parameters for CWT analysis, namely, sound part resolution (VPO) and number of scales. ; The sound resolution VPO is expressed as follows: It is a proportionality constant; Scale-Pseudo-Frequency Correspondence : Sampling time interval, scale Covering the target frequency band according to logarithmic distribution: The number of scales per octave controls the resolution in the scale direction; Based on the cavitation state initially identified in step one and the obtained kurtosis Determine the wavelet basis functions in CWT : State A: The analysis objective is to detect potential, submerged, weak nonstationarities; Morlet wavelet analysis is used to verify the purity of the signal within the frequency band. State B: The analysis targets are diverse, including quasi-periodic vortex oscillations and transient collapse, and Morlet wavelet analysis is used; State C: Focus on impact events caused by cavitation detachment and collapse, using Daubechies wavelet analysis.

6. The method as described in claim 1, characterized in that: The implementation method for step three is as follows: Based on the parameter mapping results from step two, the non-stationary feature parameter set is called. and analysis parameter set , , For standardized signals Perform FFT, STFT, and CWT signal processing; through this adaptively optimized analysis process, extract multidimensional feature quantities corresponding to the physical mechanism of the current cavitation state from the analysis results, including but not limited to: frequency domain spectrum broadening factor and broadband energy characteristics; Autocorrelation time scale and time domain statistical characteristics; Time-frequency modulation intensity and energy intermittent characteristics; wavelet multi-scale impact and high-frequency concentration characteristics; The multidimensional features constitute a noise time-frequency feature set, which is used for subsequent identification of cavitation state and cavitation type, and serves as part of the feature index in existing cavitation identification methods.

7. The method as described in claim 1, characterized in that: In step one, when the cavitation number is greater than the first critical value and the spectral broadening factor is less than the first broadening threshold, the broadband noise energy in the pressure signal spectrum is not significantly increased, and the energy is highly concentrated in the blade frequency and its harmonics, exhibiting a discrete spectral structure. This corresponds to a large autocorrelation time scale and a small spectral broadening factor. Above the critical value, the cavitation state is determined to be state A, which corresponds to the propeller without cavitation, providing a benchmark state for subsequent cavitation type determination.

8. The method as described in claim 1, characterized in that: In step one, when the cavitation number is between the first and second critical values ​​and the spectral broadening factor is between the first and second broadening thresholds, and the pressure signal spectrum simultaneously exhibits broadband energy components of both low and high frequencies, and the cavitation intensity is significantly increased compared to state A and shows non-stationary characteristics, When the cavitation level drops below the critical value, the cavitation state is determined to be state B, which corresponds to a mixed cavitation state in which tip vortex cavitation, sheet cavitation and cloud cavitation coexist.

9. The method as described in claim 1, characterized in that: In step one, when the cavitation number is less than the second critical value and the spectral broadening factor is greater than the second broadening threshold, the cavitation detachment main frequency and its harmonic energy in the pressure signal spectrum are significantly enhanced, and the cavitation intensity reaches the preset high intensity threshold. Further reducing the cavitation state, the cavitation state is determined to be state C, which corresponds to the cavitation state dominated by sheet-like cavitation and cloud-like cavitation.

Images