Fault model training and diagnosis method, device and equipment for turbine pump rotor system

Abstract

The invention relates to the technical field of turbine pump rotor system fault diagnosis, and discloses a turbine pump rotor system fault model training and diagnosis method, device and equipment, and the fault model training method comprises the steps: obtaining stress wave signals and vibration signals of a turbine pump rotor system under different faults; according to the method, the stress wave signals and the vibration signals are combined to carry out model training, the fault mode recognition and positioning capability is improved, and the diagnosis accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of turbine pump condition monitoring and fault diagnosis technology, specifically to a method, apparatus, and equipment for fault diagnosis and model training of a turbine pump rotor system. Background Technology

[0002] As the heart of a pump-fed liquid rocket engine, the performance and reliability of the turbopump directly determine the success or failure of the entire propulsion system. However, turbopumps operate in extremely harsh environments of cryogenic temperatures, high pressures, high speeds, and strong vibrations, making them one of the most frequently failing components in the engine. Domestic and international research data indicate that turbopump-related failures account for a very high proportion of failure modes in liquid rocket engines. Once a turbopump fails, it can rapidly trigger a chain reaction, leading to catastrophic consequences.

[0003] In recent years, extensive research has been conducted both domestically and internationally on fault diagnosis of rocket engine turbopumps. The main methods fall into three categories: model-based, knowledge-based, and data-driven. However, most existing research utilizes a single signal source. Currently, vibration signal analysis is the most common technique in rotating machinery fault diagnosis. However, in complex systems like turbopumps with strong background noise, the subtle impact characteristics caused by early faults are easily masked by intense overall vibration, leading to insufficient sensitivity and a high false alarm rate in traditional vibration analysis methods. This has become a technical bottleneck for achieving high-reliability fault diagnosis of turbopumps. Summary of the Invention

[0004] This invention provides a method and apparatus for training and diagnosing fault models of a turbopump rotor system, in order to solve the problems of insufficient fault mode identification and location capabilities, low diagnostic accuracy, and the lack of application of stress wave technology in this field.

[0005] In a first aspect, the present invention provides a fault model training method for a turbopump rotor system, the method comprising: Acquire stress wave and vibration signals of the turbopump rotor system under different fault conditions; The signal features of the stress wave signal and the vibration signal are extracted, and a sample dataset is constructed based on the signal features and the fault type labels corresponding to the signal features; Construct an initial diagnostic model; The sample dataset is input into the initial diagnostic model, and the initial diagnostic model is trained to obtain a fault diagnosis model for identifying fault categories based on stress wave signals and vibration signals.

[0006] In one optional implementation, the step of acquiring stress wave signals and vibration signals of the turbopump rotor system under different faults includes: Stress wave sensors and vibration sensors are respectively installed at the rolling bearing positions of the turbopump rotor system; The stress wave sensor and the vibration sensor respectively collect stress wave signals and vibration signals during the operation of the turbine pump rotor system.

[0007] In one optional implementation, the step of acquiring stress wave signals and vibration signals of the turbine pump rotor system during operation via the stress wave sensor and the vibration sensor, respectively, includes: Fault simulations are performed on the turbine pump rotor system according to different faults; stress wave signals and vibration signals of the turbine pump rotor system are collected respectively under a single simulated fault using the stress wave sensor and the vibration sensor.

[0008] In an optional implementation, before extracting the signal features of the stress wave signal and the vibration signal, the method further includes: The fault impact envelope component in the stress wave signal is extracted using an envelope demodulation method.

[0009] In one optional implementation, extracting the signal features of the stress wave signal and the vibration signal includes: Time-domain features, frequency-domain features, and time-frequency-domain features are extracted from the stress wave signal and the vibration signal, respectively. The time-domain features include signal amplitude statistical parameters, the frequency-domain features include spectral energy distribution parameters, and the time-frequency-domain features include frequency band features obtained by wavelet packet decomposition.

[0010] In one optional implementation, the initial diagnostic model includes a first sub-model and a second sub-model, and the step of inputting the sample dataset into the initial diagnostic model and training the initial diagnostic model includes: The first feature of the stress wave signal is input into the first sub-model to obtain the first classification result of the first sub-model. Based on the error between the first classification result and the first fault type label, the model parameters of the first sub-model are adjusted. The first fault type label is the fault type label in the sample dataset that corresponds to the first feature. The first fault type label is used to represent bearing faults, non-bearing faults, and health status. The second feature of the vibration signal is input into the second sub-model to obtain the second classification result of the second sub-model; Based on the error between the second classification result and the second fault type label, the model parameters of the second sub-model are adjusted. The second fault type label is the fault type label in the sample dataset that corresponds to the second feature. The second fault type label is used to represent non-bearing faults and health status.

[0011] Secondly, the present invention provides a fault diagnosis method for a turbopump rotor system, the method comprising: Acquire the stress wave signal and vibration signal to be detected from the turbine pump rotor system; The signal features of the stress wave signal and the vibration signal to be detected are extracted to obtain the sample to be detected; The sample to be tested is input into the fault diagnosis model trained by the first arbitrary method, and the fault category of the turbopump rotor system is output.

[0012] In one optional implementation, the sample to be detected is input into a fault diagnosis model trained by any method of the first aspect, and the fault category of the turbopump rotor system is output, including: The first sample in the sample to be detected, which corresponds to the stress wave signal to be detected, is input into the first sub-model; The output of the first sub-model is used to determine whether the fault is a bearing-related fault. If the fault is a bearing-related fault, then the specific bearing fault type is determined, and the first detection result is output. If the fault is not a bearing-related fault, the second sample corresponding to the vibration signal to be detected in the sample to be detected is input into the second sub-model; The specific non-bearing fault type or health status is determined by the second sub-model, and the second detection result is output.

[0013] Thirdly, the present invention provides a fault diagnosis model training device for a turbopump rotor system, comprising: The training signal acquisition module is used to acquire stress wave signals and vibration signals of the turbopump rotor system under different faults; The training feature extraction module is used to extract the signal features of the stress wave signal and the vibration signal, and to construct a sample dataset based on the signal features and the fault type labels corresponding to the signal features; The initial model building module is used to build the initial diagnostic model; The training module is used to input the sample dataset into the initial diagnostic model, train the initial diagnostic model, and obtain a fault diagnosis model for identifying fault categories based on stress wave signals and vibration signals.

[0014] Fourthly, the present invention provides a fault diagnosis device for a turbine pump rotor system, comprising: The detection signal acquisition module is used to acquire the detection stress wave signal and the detection vibration signal, which are acquired by a sensor installed at the rolling bearing position of the turbine pump rotor system. The detection feature extraction module is used to extract the signal features of the stress wave signal and the vibration signal to be detected, respectively, to obtain the sample to be detected; The fault identification module is used to input the sample to be tested into the fault diagnosis model provided by any one of the first aspects, so as to output the fault category of the turbine pump rotor system through the fault diagnosis model.

[0015] Fifthly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the fault model training method for the turbopump rotor system of the first aspect or any corresponding embodiment thereof, or to perform the fault diagnosis method for the turbopump rotor system of the second aspect.

[0016] In a sixth aspect, the present invention provides a computer-readable storage medium storing computer instructions, the computer instructions being used to cause a computer to execute the fault model training method for a turbopump rotor system according to the first aspect or any corresponding embodiment thereof, or to execute the fault diagnosis method for a turbopump rotor system according to the second aspect.

[0017] In a seventh aspect, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the fault model training method for a turbopump rotor system according to the first aspect or any corresponding embodiment thereof, or to execute the fault diagnosis method for a turbopump rotor system according to the second aspect.

[0018] The technical solution provided by this invention has the following advantages: Based on the aforementioned technical methods, stress wave and vibration signals under different faults in the turbine pump rotor system are acquired simultaneously. Features are extracted and a sample dataset is constructed to train the diagnostic model. This not only applies stress wave technology to this field for the first time but also enhances the model's ability to identify fault categories by combining dual-signal features, laying the foundation for subsequent accurate diagnosis. This effectively solves the problems of insufficient fault identification and localization and low diagnostic accuracy in traditional methods. Furthermore, by strategically deploying stress wave and vibration sensors at the high-fault location of the rolling bearing, dual signals from the same location can be acquired simultaneously and accurately, ensuring the consistency and correlation of signal sources. This provides a reliable data foundation for subsequent extraction of effective features and improvement of model training quality, further enhancing the accuracy and specificity of fault diagnosis. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a fault diagnosis model training method for a turbine pump rotor system according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of an experimental apparatus for a turbine pump rotor system according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating a fault diagnosis method for a turbine pump rotor system according to an embodiment of the present invention. Figure 4 This is another schematic flowchart of a fault diagnosis method for a turbopump rotor system according to an embodiment of the present invention; Figure 5 This is a schematic diagram of a fault diagnosis model training device for a turbine pump rotor system according to an embodiment of the present invention; Figure 6 This is a schematic diagram of a fault diagnosis device for a turbine pump rotor system according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0023] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0024] In related technologies, the fault diagnosis of turbopumps mostly relies on a single vibration signal, which is insufficient in identifying early faults in strong noise environments, has low accuracy, and does not apply stress wave technology. To address the aforementioned issues, this invention proposes a fault model training and diagnosis method for a turbopump rotor system, as well as related devices.

[0025] According to an embodiment of the present invention, a method for training a fault diagnosis model for a turbine pump rotor system is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0026] This embodiment provides a method for training a fault diagnosis model for a turbopump rotor system, which can be executed in a computer system such as a set of computer-executable instructions. Figure 1 This is a flowchart of a fault diagnosis model training method for a turbine pump rotor system according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain stress wave signals and vibration signals of the turbine pump rotor system under different faults; Step S102: Extract the signal features of the stress wave signal and the vibration signal, and construct a sample dataset based on the signal features and the fault type labels corresponding to the signal features; Step S103: Construct the initial diagnostic model; Step S104: Input the sample dataset into the initial diagnostic model, train the initial diagnostic model to obtain a fault diagnosis model for identifying fault categories based on stress wave signals and vibration signals.

[0027] The present invention aims to provide a trainable fault diagnosis model for a turbopump rotor system based on dual-modal signals. This model is used to accurately identify typical fault modes of the turbopump rotor system, such as bearing outer ring fault, bearing inner ring fault, rotor imbalance, rotor rubbing, and assembly loosening.

[0028] In the fault diagnosis model training method provided in this embodiment of the invention, stress wave signals and vibration signals of the turbopump rotor system under different faults are first acquired. The turbopump rotor system is the core component for energy conversion and power transmission in the turbopump, consisting of key parts such as rotors, rolling bearings, and shafts. For example, the rotor system of a liquid rocket engine turbopump needs to operate at high speeds above 15000 r / min and in low-temperature environments below -183℃. Its operating state directly affects the overall reliability of the turbopump and is prone to various faults due to wear, fatigue, and other factors. Therefore, it is necessary to collect characteristic signals that reflect its operating state.

[0029] Stress wave signals are mechanical wave signals generated when localized stress changes abruptly due to component defects (such as bearing raceway scratches or rolling element damage) during the operation of a turbopump rotor system. These signals propagate through the structure in the form of elastic waves. This signal exhibits unique sensitivity to the early impact characteristics of bearing-related failures, capturing even impacts from micron-level surface damage. Vibration signals, on the other hand, are physical signals generated by vibrations caused by uneven mass distribution, abnormal component clearances, or faults during rotor system rotation. They are characterized by displacement, velocity, or acceleration. These signals effectively reflect the periodic vibration characteristics of non-bearing-related failures such as rotor imbalance and rotor-stator rubbing. For example, when the rotor has mass eccentricity, its vibration spectrum will show a significant characteristic synchronized with the rotational speed frequency. In practice, signal acquisition involves collecting and storing target signals using specialized equipment. For typical fault types in turbopump rotor systems, fault scenarios must be reproduced in a reasonable manner, and corresponding stress wave and vibration signals must be collected simultaneously. Signals from a healthy state (signals from a fault-free state, used for comparison with fault signals) should also be collected as baseline data to ensure signal coverage of both the target fault type and the normal state, providing a comprehensive and reliable raw data foundation for subsequent steps. In this embodiment, by accurately collecting two types of signals that are sensitive to different faults, the problem of one-sided information acquisition caused by traditional single signal acquisition is avoided, laying a data foundation for high-accuracy fault diagnosis.

[0030] After acquiring the original signals, the signal features of the stress wave signal and vibration signal are extracted (quantitative parameters extracted from the original signals that reflect the essence of the fault, which are the core basis for the model to identify the fault). Based on the signal features and the corresponding fault type labels (fault category identifiers labeled with signal features, establishing a "feature-fault" correspondence), a sample dataset is constructed.

[0031] "Feature extraction" refers to the operation of calculating fault-sensitive parameters from the original signal using algorithms. This requires targeted extraction based on the physical characteristics of both types of signals. For example, features reflecting the statistical regularity of signal amplitude (such as mean and root mean square, which can be calculated using MATLAB's mean() and rms() functions), features reflecting the frequency distribution characteristics of the signal (such as root mean square frequency and center of gravity frequency; the center of gravity frequency of rotor imbalance fault vibration signals needs to be calculated after converting the spectrum using the fft() function), and features reflecting the joint variation of the signal in the time and frequency domains (capturing the time-frequency differences of non-stationary fault signals, suitable for dynamic fault identification).

[0032] After feature extraction, each set of features needs to be labeled (adding fault category identifiers to the feature data), such as bearing outer ring fault, rotor imbalance, health status, etc., to ensure a unique correspondence between features and fault categories. Then, the "feature-label" pairs are divided into training and test sets according to a preset ratio to form a well-structured sample dataset. In this embodiment, multi-dimensional feature extraction and structured sample construction enable the model to fully learn the correlation between faults and features, avoiding diagnostic bias caused by a single feature.

[0033] In the fault diagnosis model training method provided in this embodiment of the invention, after the sample dataset is constructed, an initial diagnostic model is built. This model needs to be adapted to the scenario of fusion diagnosis of stress wave and vibration signal, and has the ability to process multi-dimensional features. It integrates the feature information of the two types of signals through built-in algorithm logic (such as the calculation rules for classification functions such as support vector machines and neural networks) to realize the mapping from features to fault categories.

[0034] In terms of model composition, it needs to include a feature input module (receiving feature data, with dimensions matching the number of extracted features; for example, if 50 features are extracted, the number of input layer nodes is set to 50), a feature processing module (the core computing unit, carrying a classification algorithm), and a classification output module. Building the model involves setting the architecture and initial parameters. Basic parameters need to be configured reasonably based on the feature dimensions and the number of fault categories (e.g., initial hyperparameters of the algorithm are set based on domain experience) to ensure the model initially possesses basic feature processing and classification capabilities, while reserving space for subsequent parameter optimization. In this embodiment, the model architecture design adapted to dual-modal signals provides structural support for subsequent accurate diagnosis.

[0035] Finally, the sample dataset is input into the initial diagnostic model, and the model is trained by iteratively adjusting parameters and improving model performance using sample data to obtain a fault diagnosis model that can identify fault categories based on stress waves and vibration signals.

[0036] During training, the feature vectors of the training set are first input into the model. The model performs calculations according to its internal algorithm and outputs a prediction result for the fault category. Then, this prediction result is compared with the actual fault type label of the sample. A pre-defined evaluation metric is used to calculate the diagnostic error, quantifying the degree of deviation between the predicted and actual results. Based on the calculated error, key parameters in the model are adjusted according to a pre-defined parameter optimization algorithm to correct the model's decision boundary. Through repeated iterations of the "input-prediction-error calculation-parameter adjustment" process, the model continuously learns the inherent correlation between fault features and categories, thereby gradually reducing the diagnostic error.

[0037] During training, the model performance needs to be validated periodically using independent test set data to evaluate its generalization ability, with the core purpose of preventing overfitting. The training process terminates when the model's performance on the test set reaches a preset convergence criterion. The final model can stably and accurately output fault categories based on the input feature vectors, thereby achieving effective identification of faults in the turbopump rotor system. This embodiment uses a strategy combining iterative training and generalization validation to ensure that the final diagnostic model can balance its fitting ability to training samples with its generalization ability to unknown samples, meeting the needs of practical engineering applications. The turbopump rotor system fault diagnosis model training method provided in this embodiment completes signal acquisition, feature extraction, model construction, and training in steps, achieving effective fusion of dual-modal signals. Compared to traditional single-signal diagnosis, its feature extraction dimensions are more comprehensive, and the model training takes into account both learning effect and generalization ability, resulting in higher model recognition accuracy and stability. Moreover, it does not rely on complex fault mechanism modeling, but achieves accurate diagnosis through data-driven methods, reducing the dependence on expert experience. It is suitable for fault diagnosis under harsh operating conditions of turbopumps, providing reliable technical support for the safe operation of liquid rocket engine turbopumps and reducing economic losses and safety risks caused by failure to identify faults in a timely manner.

[0038] In some optional implementations, step S101 includes: Step a1: Stress wave sensors and vibration sensors are respectively installed at the rolling bearing positions of the turbine pump rotor system; Step a2: Stress wave signals and vibration signals of the turbine pump rotor system during operation are collected by stress wave sensors and vibration sensors, respectively.

[0039] Specifically, such as Figure 2As shown, the rolling bearing is the core support component of the turbine pump rotor shaft system. A typical turbine pump rotor shaft system test bench includes two rolling bearings. This location is not only the direct occurrence area for faults in the outer and inner rings of the bearing, but also the concentrated signal area after rotor imbalance, rubbing, and other faults are transmitted through the shaft system. Placing sensors here minimizes signal attenuation and ensures distortion-free transmission of fault information. The stress wave sensor is a high-frequency response sensor sensitive to early defect impact characteristics. Its core function is to capture elastic wave signals generated by bearing component defects (such as outer ring scratches or inner ring damage). The vibration sensor uses a radial vibration acquisition type to obtain periodic vibration signals caused by rotor rotation (such as centrifugal force vibration caused by rotor imbalance). During setup, it is essential to ensure that both types of sensors are in close contact with the bearing housing surface. The stress wave sensor should be fixed with bolts, while the vibration wave sensor should be fixed with magnets or bolts. A combination of "one stress wave sensor and one vibration sensor" should be formed at each of the two rolling bearing locations, constructing a total of four signal acquisition channels. This lays the hardware foundation for the subsequent synchronous acquisition of two stress wave signals and two vibration signals. This approach not only aligns with the core design of "stress wave and vibration composite testing" but also allows for the synergistic effect of the dual-modal sensors to capture the high-frequency impact characteristics of bearing-related faults and the low-frequency vibration characteristics of non-bearing-related faults, avoiding the problem of one-sided information from a single sensor.

[0040] The data acquisition operation relies on the experimental setup. First, the sensor is connected to the signal acquisition device via a coaxial cable. The signal acquisition device converts the analog electrical signal output by the sensor into a digital signal according to preset parameters (assuming the stress wave signal sampling frequency is adapted to its 200kHz frequency, and the vibration signal sampling frequency is adapted to its 50kHz frequency). Then, communication between the signal acquisition device and the computer is established via Ethernet protocol, and the computer software controls the synchronous acquisition and storage of four signals. The acquisition process needs to be combined with a fault simulation scheme, targeting typical faults such as bearing outer ring faults (customized outer ring grooved bearings), bearing inner ring faults (customized inner ring grooved bearings), rotor imbalance (adding unbalanced counterweights to the rotor disc), rotor rubbing (installing rubbing components), and loose assembly (loosening bearing housing bolts), as well as healthy conditions, and data acquisition is carried out under a single simulated fault. For example, when simulating a bearing outer ring failure, only a custom-designed grooved outer ring bearing is installed, and the rotor system is started to its rated operating speed. Two stress wave signals and two vibration signals under this failure condition are simultaneously acquired. Data is continuously acquired in 1-second frames for each type of failure, ensuring that sufficient signal data containing fault characteristics is accumulated for each type of failure. This embodiment of the invention strictly adheres to the principle of single fault simulation and synchronous signal acquisition. It solves the problem of obtaining fault samples from liquid rocket engine turbopumps and ensures a unique correspondence between each signal and the fault type. This provides accurate raw data for subsequent fault type labeling and sample dataset construction. Furthermore, the synchronous acquisition of stress wave and vibration signals fully leverages the complementary advantages of these two types of signals, laying a data foundation for improving the accuracy of subsequent fault identification.

[0041] In some alternative implementations, step a2 above includes: Step a21: Perform fault simulation for the turbine pump rotor system according to different faults. Under a single simulated fault, stress wave signal and vibration signal of the turbine pump rotor system during operation are collected by stress wave sensor and vibration sensor respectively.

[0042] Specifically, fault simulation refers to the operation of reproducing real fault states through physical modification or component adjustment for typical fault types of turbine pump rotor systems. It needs to cover core fault categories such as bearing outer ring faults, bearing inner ring faults, rotor imbalance, rotor rubbing, and assembly loosening. For example, when simulating bearing outer ring faults, a bearing with custom-designed outer ring grooves can be used, with the groove depth and width adapted to common actual fault dimensions to ensure the representativeness of the fault characteristics. When simulating rotor imbalance faults, metal counterweights of a preset mass (e.g., 5g counterweight, corresponding to the mass eccentricity commonly encountered in actual operation) are attached symmetrically to the edge of the rotor disk. When simulating assembly loosening faults, the bearing housing fixing bolts are loosened according to a preset torque (e.g., loosening by 1 / 3 turn to reproduce the shaft instability caused by bolt loosening). Each type of fault simulation must ensure a single fault state, that is, only one fault type is set at a time, avoiding the confusion of signal characteristics caused by multiple faults overlapping, ensuring that the collected signals uniquely correspond to a certain type of fault, and providing a clear basis for subsequent fault type labeling.

[0043] Signal acquisition under a single simulated fault involves starting the turbine pump rotor system to its rated operating speed after the fault simulation is completed. Stress wave sensors and vibration sensors, already positioned at the rolling bearing locations, simultaneously acquire signals under this fault condition. The stress wave sensors capture high-frequency elastic wave signals induced by the fault (such as impact stress waves generated by indentations in the bearing inner ring), while the vibration sensors acquire periodic vibration signals caused by the fault (such as low-frequency vibrations generated by rotor rubbing). The acquisition process relies on the collaboration between the signal acquisition device and a computer to convert the analog electrical signals output by the sensors into digital signals, storing them in 1-second data frames (ensuring each frame contains the complete fault characteristic cycle). Sufficient data frames must be continuously acquired for each type of single fault. Simultaneously, stress wave and vibration signals under a healthy system condition (without any fault simulation) are also acquired as baseline data for fault identification, forming a complete "health-fault" signal comparison system.

[0044] In some optional implementations, prior to step S102, the method includes: extracting the fault impact envelope component in the stress wave signal using an envelope demodulation method.

[0045] Specifically, stress wave signals are extremely sensitive to the early impact of bearing failures, but their raw signals are easily interfered with by high-frequency background noise, resulting in poor feature extraction performance when performed directly. Therefore, a preprocessing operation of envelope extraction is required. This operation aims to demodulate the low-frequency envelope component that clearly reflects the impact of the failure from the high-frequency raw stress wave signal, thereby suppressing noise, highlighting the essential characteristics of the failure, and laying the foundation for subsequent accurate extraction of signal features.

[0046] In practice, Hilbert transform can be used to extract the envelope. The process is as follows: The acquired raw stress wave signal is converted into a complex analytic signal using Hilbert transform. The magnitude of this analytic signal is then calculated to obtain the envelope curve of the stress wave signal. This envelope curve smooths out high-frequency noise fluctuations in the original signal and clearly presents the pulse sequence generated by the fault impact.

[0047] It should be noted that in this embodiment of the invention, the envelope extraction operation is performed only on the stress wave signal. The vibration signal mainly reflects the overall periodic vibration law of the system. The analysis of the vibration signal mainly focuses on its low-frequency band below 10kHz, which concentrates the characteristic frequencies of faults such as rotor imbalance and misalignment. By extracting the spectral features of this band, the impact of high-frequency noise interference can be effectively reduced strategically. Therefore, the vibration signal does not require envelope extraction and can be directly used for subsequent feature extraction. Using the extracted stress wave envelope signal and the original vibration signal together as the data source for feature engineering can effectively improve the discriminative power and reliability of the subsequently extracted features, providing key support for building a high-quality sample dataset and training a high-accuracy fault diagnosis model.

[0048] It should be noted that in this embodiment of the invention, envelope extraction preprocessing is an essential step for stress wave signals, while vibration signals do not require this process. This is determined by the physical nature and diagnostic mechanism of the two signals. Stress waves originate from instantaneous impacts within the material (such as bearing defect contact) and propagate in the form of high-frequency stress waves. Their effective fault characteristics are implicit in the high-frequency resonant response after attenuation through a complex transmission path, and are easily submerged by various background noises. Therefore, this embodiment uses Hilbert transform for envelope demodulation to separate the low-frequency envelope component synchronized with the fault impact from the high-frequency carrier, thereby making the implicit fault information explicit. In contrast, vibration signals reflect the overall structural vibration of the system, and their fault characteristics (such as characteristic frequencies) are directly reflected in the main vibration energy bands. Therefore, vibration signals can directly extract features from the original signal without the need for the specific processing of envelope demodulation.

[0049] In some optional implementations, step S102 includes: extracting time-domain features, frequency-domain features, and time-frequency-domain features from the stress wave signal and the vibration signal, respectively. The time-domain features include signal amplitude statistical parameters, the frequency-domain features include spectral energy distribution parameters, and the time-frequency-domain features include frequency band features obtained by wavelet packet decomposition.

[0050] Specifically, time-domain, frequency-domain, and time-frequency-domain features are extracted from the stress wave signal and the vibration signal, respectively. These three types of features reflect the nature of the fault from different dimensions. Time-domain features focus on the amplitude variation law of the signal on the time axis, with the core being the statistical parameters of the signal amplitude, that is, quantifying the distribution characteristics of the signal amplitude through mathematical statistical methods. For the stress wave signal after extracting the fault impact envelope component, parameters such as root mean square (RMS) and kurtosis are extracted. The stress wave signal mainly extracts the peak value, which directly reflects the fault impact intensity. The RMS reflects the average level of signal energy, and the value increases significantly under fault conditions. Kurtosis mainly describes the steepness of the signal amplitude distribution, and the kurtosis value will increase abnormally in the early stage of the fault. For the vibration signal, in addition to the above parameters, it is also necessary to extract the mean (reflecting the DC component of the signal, the mean will be offset in the case of rotor imbalance fault) and variance (reflecting the dispersion of the signal amplitude, the variance increases significantly in the case of rubbing fault). For example, the MATLAB max() function is used to calculate the peak value and the rms() function is used to calculate the RMS. These parameters can intuitively distinguish the amplitude difference between the fault state and the healthy state.

[0051] Frequency domain features are derived by converting the time-domain signal to the frequency domain using Fourier transform, focusing on spectral energy distribution parameters, i.e., the distribution pattern of signal energy across different frequency components. For stress wave signals, the focus is on analyzing the characteristic frequencies generated by fault impacts (e.g., the formula for calculating the characteristic frequency of a bearing outer ring fault is: outer ring characteristic frequency = 0.5 × rotational speed frequency × (1 - rolling element diameter / pitch circle diameter) × number of rolling elements), extracting parameters such as the energy value at the characteristic frequency and the energy proportion within the characteristic frequency bandwidth (reflecting the concentration of fault impact energy). For vibration signals, the focus is on frequency components related to rotor rotation (e.g., rotational speed frequency and its harmonics), extracting parameters such as the dominant frequency energy (corresponding to the energy of the main vibration component) and the spectral centroid (reflecting the frequency location of energy concentration; when the rotor is unbalanced, the centroid frequency is consistent with the rotational speed frequency). After performing a Fourier transform using the fft() function, the energy of a specific frequency band can be calculated through integration, achieving quantitative extraction of frequency domain features.

[0052] The time-frequency domain features employ wavelet packet decomposition to capture the joint distribution characteristics of the signal in the time and frequency domains. To ensure consistency in the feature extraction process, both the stress wave envelope signal and the vibration signal are uniformly decomposed using the db4 wavelet basis function at three levels, resulting in eight frequency bands. The energy entropy of each frequency band is calculated, forming eight time-frequency domain features for each set of stress wave envelope signals and vibration signals to be analyzed. Energy entropy quantifies the uniformity or uncertainty of the signal energy distribution across the eight frequency bands. When the system state changes, whether the energy distribution becomes concentrated or dispersed, it causes a significant change in the information entropy value, thus making this feature a sensitive indicator of the fault state. For example, after wavelet decomposition using the wavedec() function, the energy of each frequency band can be calculated sequentially, and the energy entropy can be further obtained. This method provides consistent and discriminative time-frequency domain features for different fault modes.

[0053] This multi-domain feature extraction method assumes that each signal can generate 20-30 quantized features (e.g., 10 time-domain features, 10 frequency-domain features, and 8 time-frequency-domain features). The two classes of signals, totaling four signals, form a set of 80-120 features, comprehensively covering the manifestations of faults in different signal domains. After subsequent screening, these features are associated with corresponding fault type labels to construct a complete sample dataset. Compared to single-domain features, the fusion of multi-domain features allows for a larger feature space distance between different fault types, significantly improving the classification accuracy of subsequent model training and providing key feature support for the accurate identification of faults in turbopump rotor systems.

[0054] In some optional implementations, the initial diagnostic model includes a first sub-model and a second sub-model. Step S104 includes: Step b1: Input the first feature of the stress wave signal into the first sub-model to obtain the first classification result of the first sub-model; Specifically, the first feature of this embodiment of the invention refers to the multi-domain feature set extracted from the stress wave signal (impact envelope signal after Hilbert transform envelope decoupling) in the sample dataset. Specifically, it includes time-domain features (mean, root mean square, skewness, peak factor, impulse factor, etc.), frequency-domain features (root mean square frequency, centroid frequency, frequency variance, frequency standard deviation, mean square frequency, and the five maximum energy frequencies and corresponding peak values ​​in the spectrum), and time-frequency-domain features (information entropy of eight frequency bands after three-level wavelet packet decomposition using the db wavelet function). These features can accurately capture the high-frequency impact patterns generated by bearing-type faults (bearing outer ring faults and bearing inner ring faults), and are the core basis for the first sub-model to identify bearing-type faults.

[0055] In one example, the first sub-model is built based on the Support Vector Machine (SVM) algorithm. Its core function is to classify input samples into six categories, including: bearing outer ring fault, bearing inner ring fault, rotor imbalance, rotor rubbing, loose assembly, and health status. The model incorporates a radial basis function kernel to accommodate the classification requirements of nonlinear features. After the first feature is input into the first sub-model, the model maps it to one of the six categories through algorithmic calculation and outputs the corresponding first classification result.

[0056] Step b2: Adjust the model parameters of the first sub-model based on the error between the first classification result and the first fault type label. The first fault type label is the fault type label in the sample dataset that corresponds to the first feature. The first fault type label is used to represent bearing faults, non-bearing faults, and health status. Specifically, the first fault type label is a fault category identifier that uniquely corresponds to the first feature in the sample dataset, and its value is one of the six specific fault modes mentioned above (bearing outer ring fault, bearing inner ring fault, rotor imbalance, rotor rubbing, assembly looseness, and health status). This label is completely consistent with the six-classification output dimension of the first sub-model, providing an accurate benchmark for error calculation.

[0057] In the error calculation stage, the bias of the model in multi-class scenarios is quantified by comparing the first classification result with the first fault type label. Commonly used indicators include multi-class accuracy, macro-average F1 score, and confusion matrix. Based on the calculated error, optimization algorithms such as grid search are used to adjust the SVM parameters of the first sub-model (such as the penalty parameter C and kernel function parameter γ). Through repeated iterative training, the comprehensive classification accuracy of the first sub-model for the six states is stabilized at a preset threshold, thus completing model optimization.

[0058] Step b3: Input the second feature of the vibration signal into the second sub-model to obtain the classification result of the second sub-model; Specifically, the second feature is a set of multi-domain features extracted from vibration signals in the sample dataset. It also covers time-domain, frequency-domain, and time-frequency-domain features. Time-domain features include mean, root mean square, peak factor, etc. (reflecting the statistical law of vibration signal amplitude). Frequency-domain features include root mean square frequency, center of gravity frequency, and the five maximum energy frequencies and corresponding peak values ​​in the spectrum (reflecting the frequency distribution characteristics of vibration signals, such as the peak characteristics of rotor imbalance faults at the frequency corresponding to the rotational speed). The time-frequency-domain features are the information entropy of the eight frequency bands after the three-level wavelet packet decomposition of the db wavelet function (capturing the time-frequency variation law of non-stationary vibration). These features can accurately reflect the periodic vibration and dynamic variation law of non-bearing faults and adapt to the classification requirements of the second sub-model.

[0059] In an optional implementation, the second sub-model is also built based on the SVM algorithm, but its classification objective differs from that of the first sub-model. Its core function is to subdivide the specific categories of non-bearing faults, namely, distinguishing between rotor imbalance, rotor rubbing, loose assembly, and healthy state. After the second feature is input into the second sub-model, the model performs multi-class mapping on the feature using the algorithm and outputs a second classification result. This result must clearly indicate the specific non-bearing fault type or healthy state corresponding to the input feature. For example, when the input second feature originates from a rotor imbalance fault simulated by adding an unbalanced counterweight to the rotor disk, the second sub-model analyzes the periodic vibration peaks in the feature that are consistent with the rotational speed frequency and outputs a classification result of "non-bearing fault - rotor imbalance"; if the feature originates from a loose assembly fault simulated by "loose bearing housing bolts," then the classification result is "non-bearing fault - loose assembly."

[0060] Step b4: Adjust the model parameters of the second sub-model based on the error between the second classification result and the second fault type label. The second fault type label is the fault type label in the sample dataset that corresponds to the second feature. The second fault type label is used to represent non-bearing faults and health status.

[0061] Specifically, the second fault type label is a fault category identifier that uniquely corresponds to the second feature in the sample dataset, and its value is one of the three non-bearing faults and health statuses mentioned above. This label is consistent with the four-class classification output dimension of the second sub-model.

[0062] The error calculation and parameter adjustment strategies are similar to those of the first sub-model, aiming to improve the accuracy of the second sub-model on the four-class classification task through iterative optimization.

[0063] Through the training process from steps b1 to b4, the first and second sub-models are optimized separately to adapt to their respective diagnostic tasks. The first sub-model is trained to be a "full-category recognizer" capable of identifying all six states, while the second sub-model is specifically optimized to be a "refined classifier" for four non-bearing faults. This collaborative training strategy aligns with the core concept of "stress wave and vibration composite": it utilizes the sensitivity of stress wave signals to bearing impact faults to train a general preliminary judgment model, while leveraging the advantages of vibration signals in characterizing rotor dynamics faults to train a dedicated confirmation model. The trained diagnostic model follows an efficient "two-step diagnosis" strategy when performing the identification task: the test sample is first pre-judged by the first sub-model; if its output is one of the two bearing faults, it is directly adopted; if its output is a non-bearing fault or a healthy state, the conclusion is considered "pending confirmation," and the second sub-model is activated to make the final decision using vibration signal characteristics. This method effectively avoids potential misjudgments by a single model in complex fault modes by dividing the work among models and linking decisions, thereby systematically improving the accuracy and reliability of the overall diagnosis.

[0064] This embodiment provides a fault diagnosis method for a turbopump rotor system, which can be used in the aforementioned computer equipment, etc. Figure 3 This is a flowchart of a fault diagnosis method for a turbine pump rotor system according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S301: Obtain the stress wave signal and vibration signal to be detected from the turbine pump rotor system.

[0065] Step S302: Extract the signal features of the stress wave signal and the vibration signal to be detected to obtain the sample to be detected.

[0066] Step S303: Input the sample to be tested into the trained turbopump rotor fault diagnosis model and output the fault category of the turbopump rotor system.

[0067] In this embodiment of the invention, the turbine pump rotor fault diagnosis model is trained using the turbine pump rotor fault diagnosis model training method of any one of the above embodiments.

[0068] In one example, for a trained fault diagnosis model and the signal data I = {x1, x2, x3, x4} of the turbine pump rotor system to be detected (where x1 and x2 are the stress wave signals to be detected, and x3 and x4 are the vibration signals to be detected), the fault category is determined according to the following steps: The stress wave signals (x1, x2) and vibration signals (x3, x4) to be detected are preprocessed separately. The stress wave signals (x1, x2) are decoupled using Hilbert transform to extract the fault impact envelope component. The two vibration signals (x3, x4) retain their original periodic vibration characteristics, ensuring that the fault-related information is highlighted in both types of signals. Based on the preprocessed signals, multi-domain signal features are extracted. For the preprocessed stress wave signals x1 and x2, time-domain features (such as mean, root mean square, peak factor) and frequency-domain features (such as root mean square frequency, centroid frequency, and five values ​​in the spectrum) are extracted. The system extracts multi-domain features of the same dimension (time-domain features such as impulse factor, frequency-domain features such as mean square frequency, and time-frequency domain features such as the 8 frequency band information entropy after 3-level wavelet packet decomposition of the db wavelet function) from the vibration signals of x3 and x4. These two types of features are integrated to form the sample to be detected. The sample to be detected is input into the fault diagnosis model. The model analyzes and calculates the features through the built-in algorithm and directly outputs the fault category of the turbine pump rotor system, including bearing outer ring fault, bearing inner ring fault, rotor imbalance, rotor rubbing, assembly looseness or health status. For example, when the sample contains high-frequency impact features and corresponds to a specific frequency distribution of bearing inner ring fault, the model outputs "bearing inner ring fault". When the sample shows periodic vibration features consistent with the rotational speed, the model outputs "rotor imbalance".

[0069] The fault diagnosis method for turbopump rotor systems provided in this embodiment acquires composite stress wave and vibration signals, extracts multi-domain signal features, and relies on a stepwise diagnosis strategy based on a twin model. This method can fully leverage the sensitivity of stress wave signals to the early impact characteristics of bearing-related faults, and accurately capture the periodic vibration patterns of non-bearing-related faults using vibration signals. It effectively solves the problem of low diagnostic accuracy caused by the multiple fault modes and indistinct boundaries of turbopump rotor systems. It can accurately identify typical faults such as bearing outer ring faults, bearing inner ring faults, rotor imbalance, rotor rubbing, and assembly loosening, providing reliable fault diagnosis support for the safe operation of liquid rocket engine turbopumps.

[0070] In some optional implementations, step S303 above includes: Step d1: Input the first sample corresponding to the stress wave signal to be detected into the first sub-model. Step d2: Determine whether the fault is a bearing-related fault based on the output of the first sub-model; Step d3: If the fault is a bearing-related fault, determine the specific bearing fault type and output the first detection result. Step d4: If the fault is not a bearing, input the second sample corresponding to the vibration signal to be detected into the second sub-model. Step d5: Determine the specific non-bearing fault type using the second sub-model and output the second detection result.

[0071] Specifically, such as Figure 4 As shown, the first sample is a multi-domain feature set extracted from the stress wave signal to be detected. It includes time-domain, frequency-domain, and time-frequency-domain features that reflect the impact characteristics of bearing-type faults, such as peak factor, centroid frequency, and wavelet packet decomposition information entropy. These features are standardized and then used to form structured data input to the first sub-model. The first sub-model is built based on the support vector machine algorithm. Through previous training, it has learned the feature differences between bearing-type faults (outer ring faults and inner ring faults) and non-bearing-type faults. Its output is a binary judgment. When the output value meets the preset bearing-type fault threshold, it is judged as a bearing-type fault; otherwise, it is a non-bearing fault.

[0072] When the first sub-model determines that the fault is a bearing, the model will further determine whether it is a fault in the outer ring or the inner ring of the bearing based on the matching degree between the feature vector and the feature templates of the two types of bearing faults, and output the detection result.

[0073] If the first sub-model determines the fault to be non-bearing type, it indicates that the signal to be detected is more likely to contain non-bearing fault characteristics such as rotor imbalance and rubbing, because stress wave signals are not sensitive to non-bearing faults. Therefore, this embodiment considers the non-bearing faults detected by the first sub-model to be unreliable and may have a problem with type identification error, and thus continues to input the second sample into the second sub-model. The second sample is a set of multi-domain features extracted from the vibration signal to be detected, including parameters such as pulse factor and mean square frequency that can reflect the periodic vibration characteristics of non-bearing faults. The second sub-model is also built based on the support vector machine algorithm. After training, it can achieve multi-class identification of rotor imbalance, rotor rubbing, assembly looseness, and health status. By calculating the distance between the input features and the feature space of each fault type, the fault type with the highest matching degree is output as the second detection result.

[0074] This step-by-step diagnostic approach fully leverages the expertise of the first sub-model in identifying bearing-related faults and the advantages of the second sub-model in classifying non-bearing-related faults. It avoids misjudgments caused by cross-interference of different types of fault characteristics and significantly improves the accuracy of fault diagnosis for turbine pump rotor systems.

[0075] This embodiment also provides a fault diagnosis model training device for a turbopump rotor system. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0076] This embodiment provides a fault diagnosis model training device for a turbine pump rotor system, such as... Figure 5 As shown, it includes: The training signal acquisition module 501 is used to acquire stress wave signals and vibration signals of the turbine pump rotor system under different faults.

[0077] The training feature extraction module 502 is used to extract the signal features of the stress wave signal and the vibration signal, and to construct a sample dataset based on the signal features and the fault type labels corresponding to the signal features.

[0078] Initial model building module 503 is used to build the initial diagnostic model.

[0079] The training module 504 is used to input the sample dataset into the initial diagnostic model, train the initial diagnostic model, and obtain a fault diagnosis model for identifying fault categories based on stress wave signals and vibration signals.

[0080] The turbine pump rotor fault model training device provided in this embodiment of the invention can execute the turbine pump rotor fault model training method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method execution. Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments, and will not be repeated here.

[0081] This embodiment provides a fault diagnosis device for a turbine pump rotor system, such as... Figure 6 As shown, it includes: The signal acquisition module 601 is used to acquire the stress wave signal and the vibration signal to be detected. The stress wave signal and the vibration signal to be detected are acquired by a sensor installed at the rolling bearing position of the turbine pump rotor system.

[0082] The feature extraction module 602 is used to extract the signal features of the stress wave signal and the vibration signal to be detected, respectively, to obtain the sample to be detected. The fault identification module 603 is used to input the sample to be detected into the fault diagnosis model trained by the aforementioned method embodiment, so as to output the fault category of the turbine pump rotor system through the fault diagnosis model.

[0083] The turbine pump rotor fault diagnosis device provided in this embodiment of the invention can execute the turbine pump rotor fault model diagnosis method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments, and will not be repeated here.

[0084] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0085] The following is a detailed reference. Figure 7 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 701, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 702 or a program loaded from memory 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the electronic device. The processor 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0086] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic devices to exchange data via wireless or wired communication with other devices. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0087] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a memory 708, or installed from a ROM 702. When the computer program is executed by the processor 701, it performs the functions defined in the fault model training method and the fault diagnosis method for a turbopump rotor system according to embodiments of the present invention.

[0088] Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0089] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, it implements the fault model training method and fault diagnosis method for a turbopump rotor system shown in the above embodiments.

[0090] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0091] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for training a fault diagnosis model for a turbine pump rotor system, characterized in that, The method includes: Acquire stress wave and vibration signals of the turbopump rotor system under different fault conditions; The signal features of the stress wave signal and the vibration signal are extracted, and a sample dataset is constructed based on the signal features and the fault type labels corresponding to the signal features; Construct an initial diagnostic model; The sample dataset is input into the initial diagnostic model, and the initial diagnostic model is trained to obtain a fault diagnosis model for identifying fault categories based on stress wave signals and vibration signals.

2. The method according to claim 1, characterized in that, The acquisition of stress wave signals and vibration signals of the turbine pump rotor system under different fault conditions includes: Stress wave sensors and vibration sensors are respectively installed at the rolling bearing positions of the turbopump rotor system; The stress wave sensor and the vibration sensor respectively collect stress wave signals and vibration signals during the operation of the turbine pump rotor system.

3. The method according to claim 2, characterized in that, The process of acquiring stress wave signals and vibration signals of the turbine pump rotor system during operation via the stress wave sensor and vibration sensor, respectively, includes: Fault simulations were performed on the turbopump rotor system based on different faults; The stress wave sensor and the vibration sensor respectively collect stress wave signals and vibration signals during the operation of the turbine pump rotor system under a single simulated fault.

4. The method according to claim 1, characterized in that, Before extracting the signal features of the stress wave signal and the vibration signal, the method further includes: The fault impact envelope component in the stress wave signal is extracted using an envelope demodulation method.

5. The method according to claim 1, characterized in that, The extraction of signal features from the stress wave signal and the vibration signal includes: Time-domain features, frequency-domain features, and time-frequency-domain features are extracted from the stress wave signal and the vibration signal, respectively. The time-domain features include signal amplitude statistical parameters, the frequency-domain features include spectral energy distribution parameters, and the time-frequency-domain features include frequency band features obtained by wavelet packet decomposition.

6. The method according to claim 1, characterized in that, The initial diagnostic model includes a first sub-model and a second sub-model. The step of inputting the sample dataset into the initial diagnostic model and training the initial diagnostic model includes: The first feature of the stress wave signal is input into the first sub-model to obtain the first classification result of the first sub-model. Based on the error between the first classification result and the first fault type label, the model parameters of the first sub-model are adjusted. The first fault type label is the fault type label in the sample dataset that corresponds to the first feature. The first fault type label is used to represent bearing faults, non-bearing faults, and health status. The second feature of the vibration signal is input into the second sub-model to obtain the second classification result of the second sub-model; Based on the error between the second classification result and the second fault type label, the model parameters of the second sub-model are adjusted. The second fault type label is the fault type label in the sample dataset that corresponds to the second feature. The second fault type label is used to represent non-bearing faults and health status.

7. A fault diagnosis method for a turbine pump rotor system, characterized in that, The method includes: Acquire the stress wave signal and vibration signal to be detected from the turbine pump rotor system; The signal features of the stress wave signal and the vibration signal to be detected are extracted to obtain the sample to be detected; Input the sample to be tested into the fault diagnosis model provided by any one of claims 1-6, and output the fault category of the turbine pump rotor system.

8. The method according to claim 7, characterized in that, The step of inputting the sample to be tested into the fault diagnosis model provided by any one of claims 1-6, so as to output the fault category of the turbine pump rotor system through the fault diagnosis model, includes: The first sample in the sample to be detected, which corresponds to the stress wave signal to be detected, is input into the first sub-model; The output of the first sub-model is used to determine whether the fault is a bearing-related fault. If the fault is a bearing-related fault, then the specific bearing fault type is determined, and the first detection result is output. If the fault is not a bearing-related fault, the second sample corresponding to the vibration signal to be detected in the sample to be detected is input into the second sub-model; The second sub-model determines specific non-bearing faults and health conditions, and outputs the second detection result.

9. A fault diagnosis model training device for a turbine pump rotor system, characterized in that, The device includes: The training signal acquisition module is used to acquire stress wave signals and vibration signals of the turbopump rotor system under different faults; The training feature extraction module is used to extract the signal features of the stress wave signal and the vibration signal, and to construct a sample dataset based on the signal features and the fault type labels corresponding to the signal features; The initial model building module is used to build the initial diagnostic model; The training module is used to input the sample dataset into the initial diagnostic model, train the initial diagnostic model, and obtain a fault diagnosis model for identifying fault categories based on stress wave signals and vibration signals.

10. A fault diagnosis device for a turbine pump rotor system, characterized in that, The device includes: The detection signal acquisition module is used to acquire the detection stress wave signal and the detection vibration signal, which are acquired by a sensor installed at the rolling bearing position of the turbine pump rotor system. The detection feature extraction module is used to extract the signal features of the stress wave signal and the vibration signal to be detected, respectively, to obtain the sample to be detected; The fault identification module is used to input the sample to be tested into the fault diagnosis model provided by any one of claims 1-6, so as to output the fault category of the turbine pump rotor system through the fault diagnosis model.

11. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the fault diagnosis model training method provided in any one of claims 1 to 6 or the fault diagnosis method in claim 7.

Images