Axial piston pump fault diagnosis method based on sound-temperature-pressure multi-source information fusion

Abstract

The application discloses an axial plunger pump fault diagnosis method based on sound-temperature-pressure multi-source information fusion, and belongs to the technical field of hydraulic component fault diagnosis. The method synchronously collects pressure pulsation signals, oil discharge port temperature signals and acoustic signals; extracts pressure waveform distortion features, temperature rise rate features and acoustic high-frequency energy features respectively; constructs a fusion decision model based on D-S evidence theory, maps the three features into basic probability distribution functions by using a Sigmoid membership function; obtains joint confidence by fusion calculation and sets double thresholds to realize graded early warning; when the joint confidence exceeds the threshold, a feature sensitivity matrix is constructed based on the distribution difference of the three types of evidence bodies to determine the suction, friction pair wear or plunger jam fault type. The application utilizes the difference and complementarity of the fault sensitivity of multi-physical field signals, solves the problems that single signals are not obvious in the weak fault stage and are easily disturbed, and realizes accurate identification and graded early warning of the plunger pump fault.

Description

Technical Field

[0001] This invention relates to the technical field of hydraulic component fault diagnosis and health management (PHM), and in particular to the technical field of axial piston pump fault diagnosis method based on multi-source information fusion of sound, temperature and pressure. Background Technology

[0002] Axial piston pumps, as the "heart" of hydraulic systems, are widely used in high-end equipment such as aerospace, marine power, and engineering machinery. With the increasing demands for power density in modern industry, piston pumps are evolving towards high-speed and high-pressure operating conditions.

[0003] Top-level failures in plunger pumps are primarily triggered by the following failure modes: friction pair wear, cavitation, and plunger jamming. Friction pair wear is often caused by oil contamination, deteriorated lubrication conditions, or uneven load, manifesting as increased internal leakage, decreased efficiency, and increased temperature and vibration. Caustic pull usually originates from insufficient inlet oil supply or pipeline leaks, leading to cavitation and flow / pressure pulsations. Plunger jamming is often caused by contaminant particles, scoring, seizing, or thermal deformation, causing sudden torque increases, abnormal pressure, and even pump seizure.

[0004] In existing technologies, common methods for monitoring and diagnosing the operating status of plunger pumps include: threshold alarms based on operating parameters such as pressure / flow rate, frequency domain characteristic analysis based on vibration signals, energy or spectral characteristic judgment based on acoustic noise, and indirect assessment based on temperature, efficiency, or leakage. However, the early stages of plunger pump failure often manifest as weak abnormalities such as local lubrication film rupture and short-term impacts in the friction pair. These abnormalities are transient, intermittent, and non-stationary, and are easily masked by background noise, load fluctuations, speed changes, and other normal operating shocks. Monitoring methods using a single signal source typically have the following shortcomings: 1. Pressure signal monitoring is extremely sensitive to changes in operating conditions and is easily affected by system load disturbances and pipeline resonance, making it difficult to distinguish between abnormal pulsations caused by faults and normal fluctuations caused by changes in operating conditions; 2. Temperature signal monitoring has advantages such as simple system structure, low cost, good real-time performance and convenient data processing. However, its monitoring information is limited to a single measuring point, its spatial characterization ability is insufficient, and it is easily affected by the arrangement of measuring points and sensor errors. Therefore, it is more suitable for monitoring local temperature changes in key parts.

[0005] 3. Vibration signal monitoring is sensitive to installation location, structural transmission path and environmental vibration. Different faults or operating conditions may produce similar spectral characteristics, resulting in false alarms or missed alarms. 4. Acoustic signal monitoring is susceptible to environmental noise and attenuation, and high-frequency acoustic emission attenuates significantly when propagating in complex structures, making it difficult to obtain consistent characteristics stably.

[0006] To improve identification accuracy, some solutions attempt to employ multi-sensor or multi-parameter joint diagnosis. Using only pressure, temperature, and acoustic signals for fault diagnosis is primarily based on the following reasons: pressure signals directly reflect the hydraulic system's operating conditions, accurately capturing minute abnormal pulsations caused by faults, and are less susceptible to environmental interference, exhibiting high repeatability and accuracy; temperature signals can capture abnormal temperature changes caused by overheating or friction; acoustic signals are sensitive to minute impact anomalies, promptly capturing high-frequency, brief impacts in the early stages of a fault, and are easy to deploy, non-invasive, and can quickly respond to early fault symptoms. Combining these three signals complements each other, avoids redundancy and inconsistency between multiple signal sources, and improves the accuracy and real-time performance of fault diagnosis. However, vibration sensors are prone to damage after prolonged use and have limited measurement ranges. Therefore, selecting pressure and acoustic signals can effectively improve the sensitivity and stability of fault diagnosis. However, multi-source data often presents challenges such as difficulty in time alignment and complex information coupling. Therefore, there is an urgent need for a fault diagnosis method for axial piston pumps that can make full use of multi-source information such as pressure, temperature and sound under multiple operating conditions to reliably extract and effectively integrate early signs of faults, thereby improving timeliness and accuracy. Summary of the Invention

[0007] The purpose of this invention is to address the problems in existing axial piston pump fault monitoring, such as difficulty in feature extraction from a single signal source, high false alarm rate, and difficulty in achieving early warning of weak faults. This invention provides an axial piston pump fault diagnosis method based on the fusion of multi-source information from acoustic, temperature, and pressure sources. This method fuses pressure, outlet temperature, and acoustic signals, utilizes Dempster-Shafer (DS) evidence theory to improve the confidence and robustness of the diagnosis, and uses a feature sensitivity matrix to classify and diagnose faults, achieving real-time monitoring and graded early warning of multiple fault states.

[0008] To achieve the above objectives, this invention proposes a fault diagnosis method for axial piston pumps based on the fusion of acoustic, temperature, and pressure multi-source information, comprising the following steps: Step S1: Multi-source signal synchronous acquisition, synchronously acquire pressure pulsation signal, oil outlet temperature signal and acoustic signal during the operation of axial piston pump; Step S2: Multi-source signal preprocessing, the preprocessing including at least time alignment of the multi-source signals based on a unified time reference to obtain time-aligned synchronization signal data; Step S3: Heterogeneous feature extraction: extract pressure waveform distortion features from the preprocessed pressure pulsation signal, extract temperature rise rate features from the temperature signal, and extract high-frequency acoustic energy features from the acoustic signal. Step S4: Construct a fusion decision model. Based on the DS evidence theory, construct a fusion decision model and use the Sigmoid membership function to map the three features extracted in step S3 to the basic probability assignment (BPA) function for the three health states of normal, mild failure and severe failure. Step S5: Fusion calculation and hierarchical early warning. The basic probability assignment functions of the three evidence bodies are fused to obtain the joint confidence level, and hierarchical early warning is carried out based on the joint confidence level. Step S6: Fault classification and diagnosis. When the joint confidence exceeds the preset threshold, a feature sensitivity matrix is ​​constructed based on the difference in fault confidence distributions corresponding to the three evidence bodies to determine the specific fault type and output the diagnosis result.

[0009] As a preferred method, one extraction method for the pressure waveform distortion feature in step S3 is as follows: convert the time-domain pressure pulsation signal to the angle domain through equal-angle resampling, extract the pressure waveform of a single plunger working cycle, calculate the dynamic time warping (DTW) distance between it and the pre-stored standard healthy waveform, and use this distance as the pressure waveform distortion feature.

[0010] As a preferred method, another extraction method for the pressure waveform distortion feature described in step S3 is as follows: The pressure signal of the current window and the healthy baseline pressure reference waveform are standardized and subjected to Fast Fourier Transform (FFT) respectively. The normalized norm of the difference between their amplitude spectra is then calculated as the pressure waveform distortion feature. This feature is used to capture abnormal flow pulsations and flow shocks caused by various faults.

[0011] Preferably, the extraction of the temperature rise rate feature in step S3 is performed on the high-pressure oil outlet. The extraction method is as follows: using least squares linear fitting based on a sliding window, the rise rate of the temperature signal within the window is calculated as the temperature rise rate feature. This algorithm can effectively suppress the quantization noise of the temperature sensor and minor environmental disturbances, and extract the true temperature change trend.

[0012] As a preferred embodiment, the method for extracting the acoustic high-frequency energy features in step S3 is as follows: Wavelet Packet Decomposition (WPD) is used to decompose the acoustic signal into multiple layers, the ratio of the reconstructed signal energy to the full-band signal energy is calculated, and this ratio is used as the acoustic high-frequency energy feature to characterize the micro-wear of the metal.

[0013] Preferably, the mapping in step S4 uses the Sigmoid membership function to normalize and map each feature value to the [0,1] interval, thereby obtaining the basic probability allocation of each feature under the three states of normal, mild fault and severe fault.

[0014] Preferably, the graded early warning in step S5 specifically involves: setting a first preset threshold and a second preset threshold, wherein the second preset threshold is greater than the first preset threshold; outputting an early warning signal when the confidence level of a minor fault state is higher than the first preset threshold; and outputting an alarm signal when the confidence level of a severe fault state is higher than the second preset threshold.

[0015] Preferably, the fault classification diagnosis in step S6 specifically includes: calculating the pressure fault confidence level corresponding to the pressure evidence body based on a preset feature sensitivity matrix. Temperature fault confidence level corresponding to temperature evidence body Acoustic fault confidence corresponding to acoustic evidence body ;like Significantly higher than and ,and If it is at a relatively low position, the fault type is determined to be air suction; if and Significantly higher than ,and No significant sudden changes were observed, and the fault type was determined to be friction pair wear; if , and All values ​​are at high levels and close to each other, indicating that the fault type is plunger jamming.

[0016] Preferably, the pressure pulsation signal and the oil outlet temperature signal are collected from the oil outlet of the plunger pump, and the acoustic signal is collected from the surface of the pump housing or the surrounding space.

[0017] Preferably, the preprocessing in step S2 includes signal alignment based on a unified time reference and truncating data segments of fixed length using a sliding window method for feature extraction.

[0018] The beneficial effects of this invention are: 1. Multi-source complementarity and strong robustness: This invention comprehensively utilizes three physical field information: pressure, temperature, and sound characteristics. Compared with single signal monitoring, it can effectively avoid misjudgments caused by sensor failure or a single interference source, significantly improving the robustness of the monitoring system.

[0019] 2. Highly targeted features and high early identification rate: Pressure spectrum distortion, temperature rise rate, and high-frequency sound energy features were designed to address different manifestations of various faults. In particular, the DS fusion algorithm can identify fault trends early through evidence accumulation even when a single feature is not yet significant, enabling early warning for multiple faults.

[0020] 3. Quantitative Assessment and Fault Diagnosis: Qualitative fault characteristics are transformed into a continuous confidence curve of 0-1, which intuitively reflects the health degradation process of the plunger pump. When the health is abnormal, the conflict and distribution patterns of three types of evidence—pressure, acoustic, and temperature—are analyzed, and the fault is classified and diagnosed using a feature sensitivity matrix.

[0021] The features and advantages of the present invention will be described in detail through embodiments and in conjunction with the accompanying drawings. Attached Figure Description

[0022] Figure 1 This is the fault tree for the plunger pump of the present invention.

[0023] Figure 2 This is a flowchart of the calculation process of the present invention.

[0024] Figure 3 The graph shows the overall sticking confidence level when the plunger sticks out. The horizontal axis represents time (s) and the vertical axis represents the sticking confidence level [0,1].

[0025] Figure 4 This is a schematic diagram of the fault classification and diagnosis results of an axial piston pump based on a feature sensitivity matrix. The diagram represents a three-dimensional decision space with three-source feature sensitivity. The horizontal axis represents the confidence level of temperature faults, the vertical axis represents the confidence level of pressure faults, and the vertical bar represents the confidence level of acoustic faults. The color bar on the right indicates the sample number. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Figure 2 Flowchart of the calculation: Example 1: Early Warning Method for Plunger Sticking Based on Multi-Sensor Information Fusion in Plunger Pumps 1. Signal Acquisition and Input Definition: The experimental subject was a 9-plunger axial piston pump with a displacement of 1.8 cc / r and a maximum pressure of 28 MPa. Pressure sensors, temperature sensors, and acoustic sensors were installed to collect the following signals: Pressure pulsation signal: pressure(t).

[0027] Temperature signal: temperature(t).

[0028] Acoustic noise signal: noise(t).

[0029] The sampling rate for the pressure signal is set to 30 kHz, the sampling rate for the temperature signal is set to 1 kHz, and the sampling rate for the sound signal is set to 20 kHz. The sampling duration is T = 10 s (expandable in engineering). Input data is uniformly converted to column vector and double-precision format to meet the accuracy and consistency requirements of subsequent calculations.

[0030] Feature extraction is performed using a sliding window approach: Window length Winlen = 0.5 s × fs (i.e., 0.5 seconds, where fs is the sampling rate) Step size Step = 0.1 s × fs (i.e., 0.1 seconds) The total number of windows is: (1) Where L is the total length of the sample points. Each window outputs one feature point and records the timeline: Simultaneously establish a health baseline pressure reference waveform. The initial 0.5 s window data (i.e., the first window) is preferred as the health reference.

[0031] 2. Feature Extraction Feature A: Pressure waveform distortion (spectral difference) Current window pressure signal With reference pressure Standardize each value separately, then compare the FFT amplitude spectra: standardization: (2) (3) Amplitude spectrum: (4) (5) in, This is a Fourier transform.

[0032] Distortion is defined as the normalized norm of the spectral difference: (6) This feature is used to capture pressure pulsation waveform changes caused by abnormal flow pulsation due to plunger jamming, flow shock, etc.

[0033] Feature B: Temperature rise rate characteristic The temperature rise rate was calculated using the least squares method based on a sliding window. The data was collected within the window. A sequence of temperature sampling points Constructing a linear regression model ,in This is the desired rate of temperature rise.

[0034] By minimizing the mean square error objective function The optimal slope estimate is obtained by solving for: (7) This algorithm can effectively suppress the quantization noise of the temperature sensor and minor environmental disturbances, and extract the true temperature change trend.

[0035] Feature C: Acoustic high-frequency energy entropy Wavelet packet decomposition is preferred for extracting high-frequency sound node energy. right Perform 3-level wavelet packet decomposition to obtain the energy vector e of each node; take the sum of the energy proportions of the high-frequency end nodes as the feature.

[0036] (8) To suppress the effects of inter-window jitter and random noise, the three features can be smoothed by motion to enhance trend discernibility.

[0037] 3. Feature to BPA mapping Features are normalized to [0,1] and a lower bound bias is added to avoid zero confidence that could cause fusion distortion. The following mapping is used: (9) "Fault confidence assignment" with evidence obtained: (10) (11) (12) And construct the BPA for each piece of evidence in the two propositional spaces of "normal / fault": (13) 4. DS Evidence Fusion The first step is fusion (pressure + temperature). Calculate the conflict factor: (14) To avoid extreme conflicts that could cause the denominator to approach 0, an engineering protection is introduced: when K1 > 0.99, K1 = 0.99. The result is: (15) Then, m12 is normalized twice (to ensure numerical stability and a sum of 1).

[0038] The second step is the fusion of (pressure + temperature) and acoustics. Similarly, calculate the conflict factor K2 and merge: (16) (17) The final overall confidence level is defined as the normalized support of the fault proposition: (19) The output is a continuous curve from 0 to 1, which can be directly used for early trend warning and control linkage.

[0039] 5. Tiered early warning strategy Implement tiered alarms by setting dual thresholds: Warning threshold: δ1 (minor fault trend) Alarm threshold: δ2 (High risk of severe fault) when (i) When ≥ δ1, output an orange warning; when (i) A red alarm is output when ≥δ2. As shown in the figure: during the period from 0 to approximately 7 seconds, the axial piston pump is in a basically normal or slightly abnormal state; around 7 to 8 seconds, a consistent abrupt change occurs in the multi-source characteristics; after 8 seconds, the overall jamming confidence obtained by DS fusion rapidly exceeds the δ2 ​​red alarm threshold, indicating that the piston has entered a high-risk state of severe jamming. This result shows that the method of this invention can provide a reliable judgment of jamming trend in advance through multi-source evidence fusion when a single signal is not yet significantly abnormal, and stably trigger an alarm when the fault worsens.

[0040] Example 2: Implementation details of fault classification based on the difference in sensitivity of acoustic-pressure characteristics When the joint fault confidence level output in step S5 ( Exceeding the preset alarm threshold (e.g.) At this point, the system triggers the classification and diagnostic module. Although the DS fusion result indicates that the plunger pump is in an "unhealthy state," further identification of the fault source is necessary to guide subsequent maintenance.

[0041] 1. Data preparation: Read the pressure failure confidence score obtained by mapping with the Sigmoid function within the current time window. Temperature fault confidence Acoustic Fault Confidence These three parameters represent the severity of "fluid anomaly", "temperature anomaly", and "solid contact dynamics anomaly", respectively.

[0042] 2. Classification and Decision Logic: Scenario 1: Air Suction Judgment Characteristic features: Significantly higher than , .

[0043] Physical mechanism: When cavitation or cavitation occurs, the collapse of bubbles in the oil mainly causes severe pressure pulsation and waveform distortion, leading to a sharp deterioration of the pressure spectrum characteristics; at this time, the solid structure has not yet experienced direct metal friction, and although the high-frequency energy of the acoustic signal fluctuates, the amplitude is relatively small.

[0044] Conclusion: If the system detects that the pressure confidence level dominates the fault trend, it is determined to be cavitation.

[0045] Scenario 2: Wear Judgment of Friction Pairs Characteristic features: , Significantly higher than .

[0046] Physical mechanism: During wear, microscopic surface spalling and increased roughness will excite strong high-frequency acoustic emission signals, leading to a rapid increase in acoustic entropy characteristics; a large amount of heat will be generated in the solid contact area, but at this time the internal leakage has not increased significantly, and the macroscopic waveform of the oil outlet pressure remains relatively stable.

[0047] Conclusion: If the system detects that acoustic confidence level dominates the fault trend, it is determined to be wear of the friction pair.

[0048] Scenario 3: Judgment based on plunger jamming Characteristic features: , and All are at high levels and their values ​​are similar.

[0049] Physical Mechanism: Plunger jamming is a serious failure of fluid-structure interaction. Jamming leads to a sharp increase in mechanical friction resistance, generating huge mechanical noise; it also generates a lot of heat. At the same time, the obstruction of plunger movement directly causes instantaneous pressure changes and abnormal flow pulsations in the oil discharge chamber.

[0050] Conclusion: If both types of evidence point to high risk and the combined confidence level after fusion is close to 1, the case is determined to be plunger jamming.

[0051] Figure 4 A three-dimensional decision plane was constructed, and by integrating the sensitivity differences of the three physical fields of sound, temperature, and pressure, this space was divided into several typical three-dimensional regions, revealing the decoupling and evolution logic of hydraulic pump failures more comprehensively and three-dimensionally. Normal region (lower left corner, light green cube): This region is located near the origin of the coordinate system and is characterized by... , and All three are at low levels. The high concentration of sample points here indicates that the system has no abnormalities in the three dimensions of sound, temperature, and pressure, and is in a healthy and stable operating state.

[0052] The wear zone of the friction pair (top, pale yellow cube): This region exhibits outstanding acoustic and temperature confidence levels, but relatively low confidence levels in the pressure dimension. This indicates that high-frequency acoustic signals are most sensitive to early mechanical friction, capturing wear characteristics before pressure waveform distortion and significant temperature rise. At this point, the acoustic emission signal generated by solid friction dominates, indicating wear of the friction pair.

[0053] Void Suction Zone (left side, light blue cube): This region extends along the Y-axis (pressure confidence level) and is characterized by a significantly higher pressure confidence level than the acoustic and temperature confidence levels. At this point, it primarily causes severe pressure pulsations and waveform distortion on the fluid side, but has not yet caused significant mechanical vibration noise or heat accumulation; the system identifies this as a cavitation suction fault.

[0054] Piston jamming zone (center / deep, pale red cube): This region is located at a higher-dimensional intersection of space, manifesting as... (Stress) and (Acoustic) levels are all high, and are usually accompanied by An increase in temperature corresponds to the most severe fault condition: abnormal fluid pressure (fluid obstruction), explosive solid friction noise (mechanical jamming), and abnormal temperature (intense frictional heat generation) occur simultaneously, which is determined to be plunger jamming.

[0055] By leveraging the sensitivity differences of different faults in the three-source characteristics of "sound-temperature-pressure" (such as air suction being sensitive to pressure, wear being sensitive to temperature and acoustics, and jamming being sensitive to the entire field), more accurate clustering and identification of the health status of hydraulic pumps can be achieved.

[0056] The scatter points in the figure represent monitoring samples collected at different times, and the color bar on the right indicates the sample number. The distribution trajectory of the sample points demonstrates that this invention can clearly map different types of faults to specific sensitivity ranges, achieving effective separation and accurate diagnosis of complex fault modes.

[0057] The above embodiments are illustrative of the present invention and are not intended to limit the present invention. Any simple modifications to the present invention are within the scope of protection of the present invention.

Claims

1. A fault diagnosis method for axial piston pumps based on multi-source information fusion of acoustic, temperature, and pressure sources, characterized in that, Includes the following steps: Step S1: Multi-source signal synchronous acquisition, synchronously acquire pressure pulsation signal, oil outlet temperature signal and acoustic signal during the operation of axial piston pump; Step S2: Multi-source signal preprocessing, the preprocessing including at least time alignment of the multi-source signals based on a unified time reference to obtain time-aligned synchronization signal data; Step S3: Heterogeneous feature extraction: extract pressure waveform distortion features from the preprocessed pressure pulsation signal, extract temperature rise rate features from the temperature signal, and extract high-frequency acoustic energy features from the acoustic signal. One method for extracting the pressure waveform distortion feature is as follows: the time-domain pressure pulsation signal is converted to the angle domain through equal-angle resampling, the pressure waveform of a single plunger working cycle is extracted, the dynamic time warping distance between it and the pre-stored standard healthy waveform is calculated, and this distance is used as the pressure waveform distortion feature. The extraction of the temperature rise rate feature is performed on the high-pressure oil outlet. The extraction method is as follows: using least squares linear fitting based on a sliding window, the rise rate of the temperature signal within the window is calculated as the temperature rise rate feature. The method for extracting the acoustic high-frequency energy features is as follows: wavelet packet decomposition is used to decompose the acoustic signal into multiple layers, the ratio of the reconstructed signal energy to the full-band signal energy is calculated, and this ratio is used as the acoustic high-frequency energy features. Step S4: Construct a fusion decision model. Based on the DS evidence theory, construct a fusion decision model and use the Sigmoid membership function to map the three features extracted in step S3 into basic probability assignment functions for the three health states of normal, mild failure and severe failure, respectively. Step S5: Fusion calculation and hierarchical early warning. The basic probability allocation functions of the three evidence bodies are fused to obtain the joint confidence level, and hierarchical early warning is performed based on the joint confidence level. The hierarchical early warning specifically involves: setting a first preset threshold and a second preset threshold, wherein the second preset threshold is greater than the first preset threshold; when the confidence level of the minor fault state is higher than the first preset threshold, an early warning signal is output. An alarm signal is output when the confidence level of a severe fault condition is higher than the second preset threshold. Step S6: Fault classification and diagnosis. When the joint confidence exceeds the preset threshold, a feature sensitivity matrix is ​​constructed based on the difference in fault confidence distributions corresponding to the three evidence bodies to determine the specific fault type and output the diagnosis result. The fault classification and diagnosis specifically includes: calculating the confidence level of the pressure fault corresponding to the pressure evidence based on a preset feature sensitivity matrix. Temperature fault confidence level corresponding to temperature evidence body Acoustic fault confidence corresponding to acoustic evidence body ;like Significantly higher than and ,and If it is at a relatively low position, the fault type is determined to be air suction; if and Significantly higher than ,and No significant sudden changes were observed, and the fault type was determined to be friction pair wear; if , and All values ​​are at high levels and close to each other, indicating that the fault type is plunger jamming.

2. The axial piston pump fault diagnosis method based on multi-source information fusion of acoustic, temperature, and pressure as described in claim 1, is characterized in that, Another method for extracting the pressure waveform distortion features is to perform standardization and fast Fourier transform on the pressure pulsation signal and the standard healthy waveform in the current window, respectively, and calculate the normalized norm of the difference between their amplitude spectra as the pressure waveform distortion features.

3. The axial piston pump fault diagnosis method based on multi-source information fusion of acoustic, temperature, and pressure as described in claim 1, is characterized in that, The mapping described in step S4 uses the Sigmoid membership function to normalize and map each feature value to the [0,1] interval, thereby obtaining the basic probability allocation of each feature under the three states of normal, mild fault and severe fault.

4. The axial piston pump fault diagnosis method based on multi-source information fusion of acoustic, temperature, and pressure as described in claim 1, is characterized in that, The pressure pulsation signal and the oil outlet temperature signal are collected from the oil outlet of the plunger pump, and the acoustic signal is collected from the surface of the pump housing or the surrounding space.

5. The axial piston pump fault diagnosis method based on multi-source information fusion of acoustic, temperature, and pressure as described in claim 1, characterized in that, The preprocessing described in step S2 includes signal alignment based on a unified time reference and truncating fixed-length data segments using a sliding window method for feature extraction.

Images