An aero-engine fault monitoring method and device and a storage medium

Abstract

The present application relates to the technical field of aero-engine state monitoring and fault diagnosis, and particularly relates to a fault monitoring method and device of an aero-engine and a storage medium. The method comprises: acquiring vibration signals of each part in the working process of the aero-engine; dividing the vibration signals into steady-state signals, periodic impact signals or non-steady-state transient signals based on short-time energy and zero-crossing rate; when the vibration signals are steady-state signals, determining whether there is a fault by using time-domain feature extraction and a classification model; when the vibration signals are periodic impact signals, determining whether there is a fault by using Fourier transform and envelope spectrum analysis; and when the vibration signals are non-steady-state transient signals, determining whether there is a fault by using time-frequency analysis. In the present application, compared with using a single analysis algorithm, the method matches different algorithms based on the characteristics of the vibration signals for fault monitoring, thereby improving the accuracy of monitoring and reducing the false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of aircraft engine condition monitoring and fault diagnosis technology, specifically to an aircraft engine fault monitoring method, device, and storage medium. Background Technology

[0002] As the core power unit of aircraft, the operating status of aero-engines directly affects flight safety and equipment lifespan. Vibration signals, as key parameters characterizing engine mechanical performance, can reflect early signs of various typical faults such as bearing wear, blade fracture, rotor imbalance, and surge. Therefore, real-time monitoring and intelligent analysis technology of vibration signals has become a research focus in the field of aero-engine health management.

[0003] However, when using vibration signals for fault monitoring, traditional monitoring methods often rely on single analysis algorithms, such as frequency domain feature extraction based on spectrum analysis or time domain statistical methods based on indicators such as root mean square value and peak value. These methods do not incorporate the characteristics of the vibration signal for analysis and processing, resulting in problems such as high false alarm rates and response delays. In addition, existing alarm systems typically only provide simple audible and visual alerts and lack the ability to differentiate processing for different types of anomalies. Summary of the Invention

[0004] In view of this, the present invention provides a method, device and storage medium for fault monitoring of aero-engines, in order to solve the problem of high false alarm rate in the fault monitoring of the prior art.

[0005] In a first aspect, the present invention provides a fault monitoring method for an aero-engine, the method comprising: acquiring vibration signals from various parts of the aero-engine during operation; classifying the vibration signals into steady-state signals, periodic impact signals, or non-steady-state transient signals based on short-time energy and zero-crossing rate; when the vibration signal is a steady-state signal, determining whether a fault exists by employing time-domain feature extraction and classification models; when the vibration signal is a periodic impact signal, determining whether a fault exists by employing Fourier transform and envelope spectrum analysis; and when the vibration signal is a non-steady-state transient signal, determining whether a fault exists by employing time-frequency analysis.

[0006] In this invention, vibration signals are first classified based on short-time energy and zero-crossing rate. Then, for each type of vibration signal, a corresponding data mining algorithm is used for processing, achieving dynamic matching of signal features. Furthermore, compared to using a single analysis algorithm, this method uses different algorithms to match vibration signal features for fault monitoring, improving monitoring accuracy and reducing false alarm rate.

[0007] In one optional implementation, the vibration signal is classified into a steady-state signal, a periodic impact signal, or a non-steady-state transient signal based on short-time energy and zero-crossing rate, including: classifying the vibration signal as a steady-state signal when the short-time energy is less than or equal to a short-time energy baseline and the zero-crossing rate is less than or equal to a first zero-crossing rate threshold; classifying the vibration signal as a periodic impact signal when the short-time energy is greater than the short-time energy baseline and meets the periodic detection condition; and classifying the vibration signal as a non-steady-state transient signal when the short-time energy is greater than a transient event energy threshold and the zero-crossing rate is greater than a second zero-crossing rate threshold, wherein the transient event energy threshold is greater than the short-time energy baseline and the second zero-crossing rate threshold is greater than the first zero-crossing rate threshold.

[0008] In this invention, by combining short-time energy and zero-crossing rate for determination, steady-state, periodic impact, and non-steady-state transient signals can be accurately distinguished. Furthermore, by classifying signals based on different signal characteristic thresholds, differences in engine operating states can be effectively captured, providing accurate initial classification for fault diagnosis.

[0009] In one optional implementation, when the vibration signal is a steady-state signal, a time-domain feature extraction and classification model is used to determine whether a fault exists, including: extracting the peak value, kurtosis, impulse factor, and root mean square of the vibration signal as time-domain features; standardizing the time-domain features; and inputting the standardized time-domain features into a support vector machine classifier to obtain an output result indicating whether a fault exists.

[0010] In this invention, for steady-state vibration signals, time-domain features such as peak value and kurtosis are first extracted, and then input into a support vector machine (SVM) classifier after standardization. This scheme utilizes time-domain features to directly reflect the essential characteristics of the signal, standardization eliminates the influence of dimensions, and SVM excels at small-sample and nonlinear classification, accurately identifying steady-state signal faults. It is efficient and accurate, providing a reliable means for fault diagnosis, helping to promptly detect potential hazards, and ensuring stable operation.

[0011] In one optional implementation, when the vibration signal is a periodic impact signal, Fourier transform and envelope spectrum analysis are used to determine whether a fault exists. This includes: performing a Fourier transform on the vibration signal to determine the characteristic frequency; extracting the envelope of the vibration signal and performing a Fourier transform to obtain the envelope spectrum; and determining whether a fault exists based on the relationship between the peak value exceeding three standard deviations from the baseline in the envelope spectrum and the characteristic frequency and its harmonics.

[0012] This invention employs Fourier transform combined with envelope spectrum analysis for periodic impact signals. Fourier transform precisely locates characteristic frequencies, while envelope spectrum analysis highlights the periodicity of the impact. Faults are determined by correlating the peak value (three times the standard deviation of the baseline) with the harmonics of the characteristic frequencies. This effectively captures the subtle impact characteristics of periodic faults, overcoming the limitations of conventional spectrum analysis, improving the accuracy and reliability of fault identification, and providing an efficient and accurate technical approach for the diagnosis of periodic faults.

[0013] In one optional implementation, when the vibration signal is a non-steady-state transient signal, time-frequency analysis is used to determine whether a fault exists, including: using continuous wavelet transform to determine the wavelet coefficients of the vibration signal at different scales and time points; determining a dynamic threshold based on the modulus square, mean, and standard deviation of the wavelet coefficients; determining whether a fault exists based on whether multiple scales at the same time point exceed the dynamic threshold; and verifying the determined fault using the time-frequency energy distribution obtained by Hilbert-Huang transform.

[0014] In this invention, for non-steady-state transient vibration signals, continuous wavelet transform is used to capture wavelet coefficients at different scales and time points. The dynamic threshold is determined by the energy distribution represented by the square of the coefficient modulus, and faults are identified by exceeding the threshold at multiple scales, with Hilbert-Huang transform used for verification. This allows for accurate identification of short-term, abrupt non-steady-state faults, adapting to transient anomalies such as surge, overcoming the limitations of traditional methods, and improving the timeliness and accuracy of fault diagnosis under complex working conditions.

[0015] In an optional implementation, before acquiring vibration signals from various parts of the aero-engine during operation, the method further includes: acquiring the parts of the aero-engine to be monitored and the types of sensors used for vibration signal acquisition, wherein the parts include the compressor, turbine, rotor bearing, combustion chamber, tail nozzle, and engine mounting bracket, and the sensor types include piezoelectric accelerometers, microelectromechanical system accelerometers, fiber optic grating vibration sensors, and capacitive vibration sensors; and determining the type of sensor installed at each part based on the environment of each part and the frequency band of the vibration signals generated by each part.

[0016] In this invention, multiple sensors are combined to form a multimodal data acquisition network based on monitoring needs and environmental conditions. Through the synergy of piezoelectric, MEMS, fiber optic, and capacitive sensors, vibration signals across the entire frequency band are covered, ensuring monitoring reliability under complex operating conditions.

[0017] In one optional implementation, after acquiring vibration signals from various parts of the aero-engine during operation, the method further includes: using a wavelet threshold denoising algorithm to remove high-frequency noise and low-frequency drift from the vibration signals; and performing normalization and signal segmentation processing on the vibration signals.

[0018] In this invention, the collected aero-engine vibration signals are first processed using a wavelet threshold denoising algorithm to accurately filter out high-frequency noise and low-frequency drift, restoring the original signal. Then, normalization and segmentation are applied to unify signal characteristics and adapt to batch processing requirements. This effectively improves signal quality, laying a solid foundation for subsequent classification and fault diagnosis of steady-state, periodic impact, and non-steady-state transient signals.

[0019] In an optional implementation, the method further includes: when a fault is determined to exist, controlling the buzzer to sound an alarm at a preset frequency and controlling the red light to flash; when a fault is determined not to exist, controlling the buzzer to be silent and controlling the green light to remain on.

[0020] In this invention, during a fault, the buzzer sounds an alarm at a preset frequency and the red light flashes, triggering an immediate warning and facilitating a rapid response from maintenance personnel. When there is no fault, the buzzer is silent and the green light remains constantly on, indicating that the equipment is in normal working order. This achieves enhanced alarm identification through coordinated audio-visual feedback.

[0021] Secondly, the present invention provides a fault monitoring device for an aero-engine, the device comprising: a signal acquisition module for acquiring vibration signals from various parts of the aero-engine during operation; a signal classification module for classifying the vibration signals into steady-state signals, periodic impact signals, or non-steady-state transient signals based on short-time energy and zero-crossing rate; a first fault judgment module for determining whether a fault exists when the vibration signal is a steady-state signal using time-domain feature extraction and a classification model; a second fault judgment module for determining whether a fault exists when the vibration signal is a periodic impact signal using Fourier transform and envelope spectrum analysis; and a third fault judgment module for determining whether a fault exists when the vibration signal is a non-steady-state transient signal using time-frequency analysis.

[0022] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the aircraft engine fault monitoring method of the first aspect or any corresponding embodiment described above.

[0023] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the fault monitoring method for an aero-engine described in the first aspect or any corresponding embodiment thereof.

[0024] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the fault monitoring method for an aero-engine described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0025] 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.

[0026] Figure 1 This is a flowchart illustrating a fault monitoring method for an aero-engine according to an embodiment of the present invention.

[0027] Figure 2 This is a structural block diagram of an aircraft engine fault monitoring device according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of the working process of an aircraft engine fault monitoring device according to an embodiment of the present invention;

[0029] Figure 4 This is a schematic flowchart of an implementation method for a signal acquisition module according to an embodiment of the present invention;

[0030] Figure 5 This is a schematic flowchart illustrating the implementation method of the preprocessing module according to an embodiment of the present invention;

[0031] Figure 6 This is a schematic flowchart of the implementation method of the signal classification module according to an embodiment of the present invention;

[0032] Figure 7 This is a schematic flowchart illustrating the implementation method of the data mining module according to an embodiment of the present invention;

[0033] Figure 8 This is a schematic flowchart illustrating the implementation method of the alarm and display module according to an embodiment of the present invention;

[0034] Figure 9 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0035] As described in the background section, traditional aero-engine fault monitoring methods suffer from high false alarm rates due to the use of a single analysis algorithm. Specifically, under actual aero-engine operating conditions, vibration signals exhibit multimodal characteristics, and a single algorithm struggles to simultaneously address the feature extraction requirements of different signal types. For instance, spectral analysis lacks sufficient time-domain resolution for transient signals, and time-domain statistical methods have low sensitivity to the fault frequencies of periodic impacts, leading to increased false alarm rates (especially in scenarios with complex background noise). Furthermore, existing systems generally employ fixed threshold alarm mechanisms, relying solely on uniform audible and visual signals to indicate anomalies. This lack of tiered fault type assessment and targeted response strategies makes it difficult for maintenance personnel to promptly locate the fault source, prolonging emergency response time.

[0036] Furthermore, with the rise of multimodal data mining technology in the field of industrial equipment monitoring, it achieves adaptive feature extraction under complex working conditions by integrating signal processing and machine learning algorithms. For example, there is a wavelet packet decomposition-based method for classifying vibration signals of rotating machinery, but it does not consider the real-time switching of signal modes under dynamic working conditions. Existing methods still have bottlenecks in terms of algorithm switching delay (usually >200ms) and multi-source feature fusion efficiency, making it difficult to meet the real-time requirements of high-speed (>10,000rpm) scenarios for aero-engines.

[0037] Therefore, the existing technology has the following core defects:

[0038] 1. Insufficient algorithm adaptability: Single analysis methods are difficult to cover the characteristics of multimodal vibration signals, leading to missed detection of key fault information or misjudgment due to noise interference;

[0039] 2. Real-time bottleneck: Traditional signal classification and algorithm matching processes have computational delays, which cannot meet the millisecond-level response requirements of high-speed engine operation;

[0040] 3. Rigid alarm mechanism: The lack of anomaly type classification and visualization-assisted decision-making functions restricts the efficiency of fault handling.

[0041] In view of this, this embodiment provides a fault monitoring method for aero-engines, which integrates intelligent monitoring with dynamic signal classification, adaptive algorithm matching and multi-level alarm mechanism to achieve accurate detection and rapid response of aero-engine vibration signals.

[0042] 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.

[0043] According to an embodiment of the present invention, a fault monitoring method for an aircraft engine 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.

[0044] This embodiment provides a fault monitoring method for aircraft engines, which can be used in electronic devices such as computers, mobile phones, and tablets. Figure 1 This is a flowchart of a fault monitoring method for an aero-engine according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0045] Step S101: Acquire vibration signals from various parts of the aero-engine during operation. Specifically, in practical applications, sensors can be installed on key parts of the aero-engine to collect vibration signals from those parts. To ensure data accuracy, this embodiment uses high-precision vibration sensors, such as piezoelectric accelerometers, to collect vibration signals at a specific sampling frequency, for example, no less than 20kHz, thereby acquiring more data points and improving data quality. The key parts for installing vibration sensors can be determined based on actual conditions; for example, they can be installed on bearings, blades, and rotors. Furthermore, to facilitate the analysis of transient events, after acquiring the vibration signals, they are first divided into time windows, i.e., the vibration signals are processed in segments, and fault judgment is then performed on the vibration signals within each window.

[0046] Step S102: Based on short-time energy and zero-crossing rate, the vibration signal is classified into steady-state signal, periodic impact signal, or non-steady-state transient signal. Specifically, short-time energy (STE) represents the sum of squares of the signal amplitude within a small interval (a certain time window), used to measure the energy intensity of the signal within the time window. The short-time energy is calculated using the following formula:

[0047]

[0048] In the formula, x k (n) represents the nth sampling point of the kth signal window; N represents the window length (e.g., N = f s ×T, T=50ms, corresponding to N=20kHz×0.05s=1000 points, f s (This represents the sampling frequency, and T represents the sampling period).

[0049] Zero-crossing rate represents the number of times a signal crosses a zero point within a window, reflecting the frequency components of the signal. The zero-crossing rate (ZCR) is calculated using the following formula:

[0050]

[0051] In the formula, sign(x) represents the sign function (takes 1 when x>0, otherwise takes -1); the denominator 2N is the normalization coefficient, ensuring that ZCR is in the range [0,1].

[0052] Vibration signals typically include multimodal signals such as steady-state signals, periodic impact signals, and non-steady-state transient signals. Steady-state signals include low-frequency vibrations caused by bearing wear, shaft misalignment, and poor lubrication; periodic impact signals include periodic impact faults caused by sudden increases in harmonic components due to blade breakage, gear tooth breakage, and rolling element spalling; and non-steady-state transient signals include time-frequency energy abrupt changes caused by transient anomalies such as surge and combustion instability. Based on this characteristic, the short-time energy and zero-crossing rate of different types of signals also differ. Therefore, short-time energy and zero-crossing rate can be used to classify vibration signals within each window.

[0053] Step S103: When the vibration signal is a steady-state signal, a time-domain feature extraction and classification model are used to determine whether a fault exists. Specifically, for a steady-state signal, the time-domain features of the signal, such as peak values, are extracted first, and based on the extracted features, a pre-determined classification model is used to determine whether a fault has occurred within the time window.

[0054] Step S104: When the vibration signal is a periodic impact signal, Fourier transform and envelope spectrum analysis are used to determine whether a fault exists. Specifically, for periodic impact signals, periodic impact features need to be extracted for fault judgment. In this embodiment, Fourier transform is used to first locate characteristic frequencies that may be strongly correlated with the fault (i.e., the fault can be understood as causing energy anomalies at specific frequencies), then envelope spectrum analysis is used to extract the impact envelope, i.e., focusing on the periodic impact features; finally, the relationship between the peak value of the impact envelope and the characteristic frequency is used to determine whether a fault exists.

[0055] Step S105: When the vibration signal is a non-steady-state transient signal, time-frequency analysis is used to determine whether a fault exists. Specifically, non-steady-state transient signals are characterized by being short-lived, sudden, and having variable frequencies. This embodiment uses the time-frequency focusing capability of wavelet transform to accurately locate the time point and frequency range of the anomaly. Therefore, time-frequency analysis can accurately locate whether a fault has occurred.

[0056] The fault monitoring method for aero-engines provided in this invention first classifies vibration signals based on short-time energy and zero-crossing rate. Then, for each type of vibration signal, a corresponding data mining algorithm is used for processing, achieving dynamic matching of signal features. Compared with using a single analysis algorithm, this method uses different algorithms to monitor faults based on the feature matching of vibration signals, improving the accuracy of monitoring and reducing the false alarm rate.

[0057] This embodiment provides a fault monitoring method for an aircraft engine, the process of which includes the following steps:

[0058] Step S201: Obtain the parts of the aero-engine to be monitored and the types of sensors used for vibration signal acquisition. The parts include the compressor, turbine, rotor bearing, combustion chamber, tail nozzle, and engine mounting bracket. The sensor types include piezoelectric accelerometers, microelectromechanical system accelerometers, fiber optic grating vibration sensors, and capacitive vibration sensors. Determine the type of sensor installed on each part based on the environment of each part and the frequency band of the vibration signal generated by each part.

[0059] Specifically, to accurately acquire vibration signals from various parts of an aero-engine, the vibration sensors installed in the aero-engine need to meet the requirements of high temperature, high frequency response, high sensitivity, and strong anti-interference capability. In addition, the sensors need to support analog (4-20mA) or digital (IEPE) output for easy integration with the acquisition system. Based on this, the vibration sensors used in this embodiment include piezoelectric accelerometers, microelectromechanical system accelerometers, fiber Bragg grating vibration sensors, and capacitive vibration sensors.

[0060] Piezoelectric accelerometers utilize the piezoelectric effect of piezoelectric materials (such as quartz and ceramics) to convert mechanical vibrations into electrical signals, thereby acquiring vibration signals. These accelerometers feature a high-frequency response (0.5Hz–20kHz), making them suitable for capturing transient impact signals; high sensitivity (e.g., 100mV / g); excellent signal-to-noise ratio; and compact size, making them easy to embed and install. Therefore, these accelerometers can be installed in high-frequency vibration areas such as engine compressors and turbine blades; they are also suitable for operation in high-temperature environments (requiring a high-temperature resistant housing, with an operating temperature range of -50℃ to +260℃). For example, they can be installed in engine bearing housings to monitor vibration signals generated by rotor imbalance or abnormal gear meshing.

[0061] The core principle of a Micro-Electro-Mechanical System (MEMS) accelerometer is to output an electrical signal by detecting changes in capacitance caused by the deformation of a micromechanical structure (such as a cantilever beam). This accelerometer features a wide bandwidth (DC to 5kHz), making it suitable for steady-state signal monitoring. It also boasts high integration, allowing direct embedding into data acquisition boards, and low cost, making it suitable for multi-point deployment. Therefore, this accelerometer can be installed in low-to-medium frequency vibration areas such as engine casings and accessory gearboxes, or in applications requiring low power consumption and integrated monitoring. For example, it can be used to monitor the overall vibration level of an engine, helping to diagnose problems such as loose mounting bases.

[0062] The core working principle of a fiber Bragg grating (FBG) vibration sensor is to use the wavelength offset of the fiber grating to reflect the strain changes caused by external vibration. This vibration sensor features high temperature resistance (up to 600°C or higher), suitability for extreme environments, no signal attenuation over long distances, and the ability to cascade and multiplex multiple sensors, reducing wiring complexity. Therefore, this vibration sensor is suitable for high-temperature and high-pressure environments (such as near a turbine); or for installation in areas with strong electromagnetic interference (due to the inherent electromagnetic interference resistance of optical fibers). For example, this vibration sensor can be deployed at the root of a turbine blade to monitor blade vibration and crack propagation at high temperatures.

[0063] The core working principle of a capacitive vibration sensor is to detect changes in the distance between capacitor plates caused by vibration and output an electrical signal. This vibration sensor features high resolution (down to nanometer-level displacement detection), a wide dynamic range (0.1Hz–1kHz), high stability, and minimal long-term drift. Therefore, this vibration sensor can be used for monitoring low-frequency micro-vibrations (such as during engine idle); or installed in applications requiring high linearity and high resolution. For example, it can be used to monitor low-frequency vibration anomalies during engine start / stop.

[0064] Based on the different types of vibration sensors mentioned above, the following characteristics of the sensor can be considered when determining which vibration sensor to install in different locations: First, the frequency response range, i.e., the selected vibration sensor can cover the characteristic frequency band of the engine vibration signal (e.g., 0.1Hz~20kHz); second, environmental adaptability, i.e., whether the corresponding location requires the vibration sensor to have characteristics such as high temperature resistance, oil resistance, or impact resistance; and third, sensitivity and range, i.e., the selected vibration sensor can match the vibration amplitude range (e.g., 0.1g~1000g).

[0065] In this embodiment, the vibration sensors installed in various parts of the engine are shown in Table 1 below:

[0066] Table 1

[0067] Engine section Sensor type Monitoring targets compressor rotor piezoelectric accelerometer Blade passing frequency, rotor imbalance Turbine disk Fiber Bragg Grating (FBG) Sensor Blade vibration and thermal fatigue at high temperatures Main bearing housing MEMS accelerometer Bearing wear and lubrication failure Engine mounting bracket Capacitive sensor Analysis of vibration transmission characteristics of the whole machine

[0068] In this invention, the vibration sensors installed in the aero-engine are not of a single type, but rather a combination of multiple sensors used according to monitoring needs and environmental conditions to form a multimodal data acquisition network. Through the synergy of piezoelectric, MEMS, fiber optic, and capacitive sensors, vibration signals across the entire frequency band are covered, ensuring monitoring reliability under complex operating conditions.

[0069] Step S202: Acquire vibration signals from various parts of the aero-engine during operation. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0070] Step S203: Use wavelet threshold denoising algorithm to remove high-frequency noise and low-frequency drift from the vibration signal; normalize and segment the vibration signal.

[0071] Specifically, among wavelet bases, the Daubechies wavelet family (such as db6) possesses characteristics such as tight support and approximate symmetry, making it suitable for transient signal analysis. Therefore, this implementation uses the Daubechies wavelet family as the wavelet base. After determining the wavelet base, the number of decomposition levels is first determined based on the signal's dominant frequency range. Specifically, the number of decomposition levels can be determined using the following formula:

[0072]

[0073] In the formula, J represents the number of decomposition levels, f s f represents the sampling rate. max (Indicates the highest effective frequency of the signal).

[0074] Then, the selected wavelet basis is used to perform J-level discrete wavelet transform on the original signal, i.e., the acquired vibration signal, to obtain the approximation coefficients a of each level. j and detail coefficient d j Because the obtained detail coefficients contain transient details of the real signal and random fluctuations in noise, they need to be denoised. Specifically, the following soft thresholding function can be used to denoise the detail coefficients:

[0075]

[0076] In the formula, λ represents the universal threshold, which can be obtained from... Determined, where σ is the noise standard deviation (through the highest level detail coefficient d). J (Estimated), where N is the signal length. This universal threshold can distinguish between noise and signal details; therefore, after processing the detail coefficients using the above formula, denoising of the detail coefficients is achieved. Finally, inverse wavelet transform is performed using the denoised detail coefficients and the approximate coefficients obtained from the layering to achieve denoising of the vibration signal.

[0077] Normalizing the denoised vibration signal eliminates the effects of sensor range differences and amplitude fluctuations, facilitating unified processing in subsequent algorithms. During signal normalization, the signal is first linearly scaled; specifically, the signal amplitude is normalized to the [-1, 1] interval using the following formula:

[0078]

[0079] If the dynamic range of the signal is too large, the standard deviation can be normalized using the following formula:

[0080]

[0081] In the formula, μ is the mean and σ is the standard deviation. After linear amplitude scaling, a threshold (such as ±3σ) is set to forcibly truncate amplitudes that exceed the range to the boundary value, i.e., outlier truncation, to prevent normalization distortion.

[0082] Furthermore, the acquired vibration signals are typically continuous, so they are further segmented into fixed-duration analysis windows to meet the needs of real-time monitoring and batch processing algorithms. First, the windows are divided using a preset window length and overlap rate. The window length is set based on the signal's characteristic frequency; for example, it could be 1 second. Therefore, the number of sampling points within a single window is N = f. s ×T = 20kHz × 1s = 20,000 points. Additionally, to reduce edge effects, a 50% overlap rate can be used, meaning adjacent windows overlap by N / 2 points. After windowing, the signal in each window is processed using a Hanning window based on the weight w(n) calculated according to the following formula to suppress spectral leakage:

[0083]

[0084] Subsequently, the amplitude is corrected for the signal processed by the Hanning window using the window energy compensation coefficient calculated by the following formula:

[0085]

[0086] Specifically, in practical applications, a ring buffer can be used to store the latest N data points, and segmentation processing can be triggered every N / 2 points to achieve real-time processing. Alternatively, historical data can be segmented using a sliding window with a fixed step size to achieve offline processing.

[0087] Step S204: Based on short-time energy and zero-crossing rate, the vibration signal is divided into steady-state signal, periodic impact signal, or non-steady-state transient signal.

[0088] Specifically, step S204 includes:

[0089] Step S2041: When the short-time energy is less than or equal to the short-time energy baseline and the zero-crossing rate is less than or equal to the first zero-crossing rate threshold, the vibration signal is divided into a steady-state signal.

[0090] Step S2042: When the short-time energy is greater than the short-time energy baseline and meets the periodic detection conditions, the vibration signal is divided into periodic impact signals.

[0091] Step S2043: When the short-time energy is greater than the transient event energy threshold and the zero-crossing rate is greater than the second zero-crossing rate threshold, the vibration signal is divided into a non-steady-state transient signal. The transient event energy threshold is greater than the short-time energy baseline, and the second zero-crossing rate threshold is greater than the first zero-crossing rate threshold.

[0092] Different types of vibration signals exhibit varying short-time energy and zero-crossing rates. Specifically, steady-state signals are characterized by low-frequency steady-state vibration and dominance of low-frequency components, resulting in low short-time energy and a low zero-crossing rate. Periodic impact signals, as impact events, contain high-frequency components within the periodic impact, leading to periodic spikes in short-time energy and a moderate zero-crossing rate. Non-steady-state transient signals, as transient events, contain broadband noise, resulting in sudden surges in short-time energy and a relatively high zero-crossing rate. Therefore, signals can be classified based on the magnitude of their short-time energy and zero-crossing rate within each window.

[0093] Before classification, it is necessary to determine the thresholds for distinguishing between short-time energy and zero-crossing rate. For short-time energy, two thresholds need to be determined: a short-time energy baseline (steady-state signal energy baseline) and a transient event energy threshold. The 90th percentile of the steady-state signal energy baseline (STE) from historical data (historical vibration signals) can be used as the steady-state signal energy baseline (STE). base ; with 3×STE base STE as the energy threshold of transient events thresh_high For the zero-crossing rate, the first zero-crossing rate ZCR is determined. low Second zero-crossing rate ZCR high Two thresholds, the size of which can be determined based on the actual situation, such as ZCR. low =0.1, ZCR high =0.3. Furthermore, the threshold determined here can be updated in real time based on the acquired data; for example, the STE can be recalculated every 24 hours based on the latest data. base With ZCR low ZCR high To adapt to changes in working conditions.

[0094] In determining the STE of the current window k With ZCR k Then, the signal can be divided using the following process:

[0095] If STE k ≤STE base And ZCR k ≤ZCR low Then the vibration signal of the current window will be classified as a Class A signal (steady-state signal).

[0096] If STE k STE baseIf the periodicity detection passes, the vibration signal in the current window will be classified as a Class B signal (periodic impact signal). The method for periodicity detection of the signal includes: performing autocorrelation analysis on the STE sequence. If there is a significant periodic peak (peak height > 0.7 times the maximum autocorrelation value), it is determined to be periodic.

[0097] If STE k STE thresh_high And ZCR k >ZCR high Then the vibration signal of the current window will be classified into Class C signal (non-steady transient signal).

[0098] It should be noted that the window length in step S203 can be determined by considering both STE and ZCR. If the window time is too short, such as 10ms, it will lead to unstable ZCR statistics and make STE susceptible to noise interference; if the window time is too long, such as 200ms, it will fail to capture transient events (such as surge duration of approximately 50ms). Therefore, this embodiment sets the window time length to T = 50ms, thus balancing feature stability and transient response.

[0099] Step S205: When the vibration signal is a steady-state signal, time-domain feature extraction and classification models are used to determine whether a fault exists. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0100] Step S206: When the vibration signal is a periodic impact signal, Fourier transform and envelope spectrum analysis are used to determine whether a fault exists. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0101] Step S207: When the vibration signal is a non-steady-state transient signal, time-frequency analysis is used to determine whether a fault exists. For details, please refer to [link to relevant documentation]. Figure 1 Step S105 of the illustrated embodiment will not be described again here.

[0102] Step S208: When a fault is confirmed, control the buzzer to sound an alarm at a preset frequency and control the red light to flash; when no fault is confirmed, control the buzzer to be silent and control the green light to remain on.

[0103] Specifically, to implement the fault alarm function in this embodiment, hardware and software design are required. The hardware components include a microcontroller (such as an embedded development platform like Arduino, STM32, or Raspberry Pi), a buzzer (supporting PWM control, 800Hz frequency), red and green LEDs (with 220Ω current-limiting resistors), a signal input interface (such as a GPIO or ADC interface for receiving abnormal signals), and a communication module (such as a USB serial port or Wi-Fi / Bluetooth module for communication with a host computer). The hardware components are connected as follows: the buzzer is connected to the microcontroller's PWM output pin (such as Arduino's Pin 9). The red LED is connected to the GPIO output pin (such as Pin 10), connected in series with a 220Ω resistor to ground. The green LED is connected to the GPIO output pin (such as Pin 11), connected in series with a 220Ω resistor to ground. The abnormal detection signal input is connected to the GPIO input pin (such as Pin 2) or the ADC pin (analog signal).

[0104] In addition to the aforementioned hardware components, software design is also required. Specifically, this software design includes microcontroller firmware development and host computer interface development. The development environment for microcontroller firmware development can be Arduino IDE, PlatformIO, or STM32CubeIDE. The specific development logic flow includes:

[0105] while True:

[0106] If an abnormal signal is detected:

[0107] Turn off green LED

[0108] Start the buzzer (800Hz, 0.2-second sound, 0.5-second silence cycle).

[0109] Activate the red LED (it lights up in 0.25 seconds, turns off in 0.25 seconds, with a cycle of 0.5 seconds).

[0110] Send abnormal data (including the abnormality level) to the host computer.

[0111] else:

[0112] Turn off the buzzer and red LED.

[0113] Turn on the green LED (always on).

[0114] Send normal data to the host computer

[0115] The development tools used for the host computer interface include Python (PyQt5 / Tkinter + Matplotlib) or LabVIEW. The developed host computer interface can specifically implement the following functions: First, time-frequency plot display, such as converting real-time signal data using FFT and dynamically updating it using Matplotlib. Second, anomaly levels and handling suggestions; that is, matching preset rules based on received data (e.g., Level 1 - Minor anomaly → Recommend checking equipment). In addition, the host computer receives data via serial port (UART) during operation, with an example format: {status:"abnormal",level:2}.

[0116] The hardware structure and software design defined above still require debugging and testing to ensure their proper functioning. The debugging process specifically includes the following steps:

[0117] Step a1, Hardware Setup. Solder the buzzer, LED, and current-limiting resistor to the microcontroller development board. Connect the abnormal signal input source (such as a sensor or analog signal generator).

[0118] Step a2, firmware debugging. Test the buzzer frequency (verify the 800Hz waveform with an oscilloscope). And calibrate the LED blinking period (adjust delay() or use a timer interrupt).

[0119] Step a3, Host Computer Development. Write the serial port data parsing module. Implement dynamic time-frequency graphs (using Matplotlib animation). Design the UI layout (displaying charts, exception levels, and handling suggestions in different areas).

[0120] Step a4: System integration and debugging. Simulate abnormal signals to verify the synchronous response of the buzzer, LED, and interface. Optimize communication latency (to ensure real-time data transmission).

[0121] After the above debugging process, any problems encountered during debugging need to be resolved before further testing and verification. Specifically, for the synchronization issue between the buzzer and LED, a hardware timer interrupt (non-blocking code) should be used to avoid delay() causing stuttering. For the stuttering issue in the time-frequency graph, the data sampling rate should be reduced or an efficient graphing library (such as PyQtGraph) should be used. For the packet loss issue, a checksum (such as CRC) and a retransmission mechanism can be added.

[0122] After resolving issues encountered during debugging, testing and verification can be performed. First, a normal state test is conducted, where vibration signal monitoring confirms the absence of a fault; this is indicated by a constantly lit green light, a silent buzzer, and a "normal" display on the interface. Next, an abnormal state test is performed, where vibration signal monitoring confirms a fault; this is indicated by a flashing red light, a buzzer sounding at a set frequency, and an updated time-frequency graph and abnormal information on the interface. Additionally, a stress test can be performed, continuously triggering abnormal states to verify the system's stability.

[0123] This embodiment provides a fault monitoring method for an aircraft engine, which includes the following steps:

[0124] Step S301: Acquire vibration signals from various parts of the aero-engine during operation; for details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0125] Step S302: Based on short-time energy and zero-crossing rate, the vibration signal is classified into steady-state signal, periodic impact signal, or non-steady-state transient signal; for details, please refer to [link to relevant documentation]. Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0126] Step S303: When the vibration signal is a steady-state signal, a time-domain feature extraction and classification model is used to determine whether a fault exists.

[0127] Specifically, step S303 includes:

[0128] Step S3031: Extract the peak value, kurtosis, impulse factor, and root mean square of the vibration signal as time-domain features; wherein, the peak value of the vibration signal x(t) is extracted using the following formula:

[0129] Peak = max(|x(t)|)

[0130] The kurtosis K of the vibration signal is extracted using the following formula:

[0131]

[0132] In the formula, N represents the number of sampling points within the window, xi represents the i-th signal within the window, μ represents the mean of the signals within the window, and σ represents the standard deviation of the signals within the window. This kurtosis further characterizes the signal pulse intensity; under normal conditions, K is approximately equal to 0, and under abnormal conditions, K is greater than 3.

[0133] The impulse factor IF of the vibration signal is extracted using the following formula:

[0134]

[0135] The root mean square is extracted using the following formula:

[0136]

[0137] Step S3032: Standardize the time-domain features; specifically, the extracted time-domain features F = [Peak, K, IF, RMS] can be Z-score standardized using the following formula:

[0138]

[0139] Step S3033: The standardized temporal features are input into the support vector machine classifier to obtain the output result indicating whether a fault exists. Specifically, before using the support vector machine classifier for classification, the kernel function to be used is determined and trained. In this embodiment, the radial basis function (RBF) kernel shown in the following formula is used as the kernel function, which can adapt to nonlinear classification boundaries.

[0140] K(F i F j )=exp(-γ||F i -F j || 2 )

[0141] During the training of the Support Vector Machine (SVM), the main optimizations are to the penalty coefficient and the kernel parameter. The penalty coefficient can be determined through hyperparameter search. The kernel parameter γ can be selected based on cross-validation. In practical applications, the extracted temporal features can be input into the trained SVM classifier, which outputs the probability of a fault. This fault probability P... fault ∈[0,1], if P fault A value >0.8 is considered abnormal, indicating a fault.

[0142] Step S304: When the vibration signal is a periodic impact signal, Fourier transform and envelope spectrum analysis are used to determine whether a fault exists.

[0143] Specifically, step S304 includes:

[0144] Step S3041 involves performing a Fourier transform on the vibration signal to determine its characteristic frequencies. Specifically, a Fast Fourier Transform (FFT) is applied to the vibration signal to convert the time-domain signal into the frequency domain, yielding a spectrum X(f). This spectrum provides a visual representation of the distribution of different frequency components within the signal. Then, theoretical characteristic frequencies are determined for potential fault conditions in different structures. For example, for turbine blades, the blade passage frequency (i.e., the theoretical characteristic frequency) can be determined using its fundamental frequency and the number of blades. Based on this theoretical characteristic frequency, the spectrum X(f) is then observed to identify any significant peaks near these frequencies. If any are found, the corresponding blade passage frequency is determined. Furthermore, the frequency resolution of the spectrum is calculated using the formula Δf = f. s / N, where f s Here, N represents the sampling frequency, and N represents the number of sampling points. In practical applications, the frequency resolution can be adjusted based on this calculation formula to accurately observe the peak value corresponding to the theoretical characteristic frequency in the spectrum.

[0145] Step S3042: Extract the envelope of the vibration signal and perform a Fourier transform to obtain the envelope spectrum. Specifically, when a fault occurs, such as a bearing or gear crack, the vibration signal will contain an impact component, but a direct FFT may be masked by other frequencies. Therefore, envelope extraction is performed first, followed by FFT of the envelope signal to highlight the periodicity of the impact. Hilbert transform can be used to extract the signal envelope. In the formula, This represents the Hilbert transform; this step extracts the "amplitude variation pattern" of the impulse signal, filtering out carrier frequency interference. Then, a Fourier transform is performed on the extracted envelope, i.e., an FFT is applied to the envelope signal e(t) to obtain the envelope spectrum E(f). The peak values ​​in this spectrum correspond to the periodic frequency of the impulse signal.

[0146] Step S3043: Determine whether a fault exists based on the relationship between peak values ​​exceeding three times the standard deviation of the baseline in the envelope spectrum and the characteristic frequency and its harmonics. Specifically, for the obtained envelope spectrum, first determine the baseline of the envelope spectrum based on the mean and standard deviation of the amplitudes at all frequency points in the envelope spectrum, where μ+3σ (μ represents the mean, σ represents the standard deviation) can be used to represent the baseline; then, identify the peak values ​​in the envelope spectrum whose amplitudes exceed the baseline. Determine whether the peak value corresponds to the theoretical characteristic frequency and the peak value of the theoretical characteristic frequency harmonics (i.e., perform harmonic matching). For example, if the theoretical characteristic frequency is 4000Hz, and there are peak values ​​exceeding the baseline at both 4000Hz and 8000Hz in the envelope spectrum, then a fault is determined to exist.

[0147] Step S305: When the vibration signal is a non-steady-state transient signal, time-frequency analysis is used to determine whether a fault exists.

[0148] Specifically, step S305 includes:

[0149] Step S3051 involves using continuous wavelet transform to determine the wavelet coefficients of the vibration signal at different scales and time points. Specifically, when using continuous wavelet transform, wavelet basis selection and scale selection are performed first. In this embodiment, the Morlet wavelet is used as the wavelet basis, thus achieving both time and frequency resolution. The Morlet wavelet is represented by the following formula:

[0150]

[0151] The scale selection needs to cover the dominant frequency range of the signal. For example, selecting a scale a∈[0.1,10] corresponds to f=f c / (aΔt), f c (where the wavelet center frequency is). Then, based on the chosen wavelet basis and scale, the wavelet coefficients are calculated using the following formula:

[0152]

[0153] In the formula, b represents time, and x(t) represents the vibration signal.

[0154] Step S3052: Determine the dynamic threshold based on the modulus square, mean, and standard deviation of the wavelet coefficients; specifically, transient anomalies such as surge are short-lived and can trigger energy mutations simultaneously across multiple frequency ranges. Based on this, this embodiment uses the modulus square |W(a,b)| of the wavelet coefficients. 2 The energy distribution at scale a-time b, i.e., the energy matrix, is obtained. Then, the local mean and standard deviation of this energy distribution at each time point are calculated using a sliding window, and the dynamic threshold is determined using the following formula:

[0155] Threshold(a,b)=μ |W| +k·σ |W|

[0156] In the formula, k is a coefficient, which can take the value 3, and μ ∣W∣ Represents local mean and σ ∣W∣ It represents the standard deviation.

[0157] Step S3053: Determine whether a fault exists based on whether multiple scales at the same time point exceed the dynamic threshold; specifically, when at a certain time point such as b0, the energy corresponding to multiple different scales a simultaneously exceeds the threshold, it is determined to be a transient event, i.e., a fault exists.

[0158] Step S3054: The time-frequency energy distribution obtained by Hilbert-Huang Transform is used to verify the identified fault. Specifically, when using Hilbert-Huang Transform, the signal is first decomposed into IMF (Intrinsic Mode Function) components through Empirical Mode Decomposition (EMD), where each IMF is a fluctuation of a single frequency component. Then, a Hilbert Transform is performed on each IMF to obtain the instantaneous frequency-time-energy distribution (horizontal axis: time, vertical axis: frequency, color represents energy), i.e., the Hilbert spectrum. In the Hilbert spectrum, if the energy concentration areas (highlighted areas) of multiple IMFs near time b0 cover a multi-scale frequency range, it indicates that the time-frequency characteristics of the two are consistent, confirming that it is a real transient anomaly.

[0159] This embodiment also provides a fault monitoring device for an aircraft engine, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs 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.

[0160] This embodiment provides a fault monitoring device for an aircraft engine, such as... Figure 2 As shown, it includes:

[0161] The signal acquisition module 21 is used to acquire vibration signals from various parts of the aero-engine during operation.

[0162] The signal classification module 22 is used to classify the vibration signal into steady-state signal, periodic impact signal or non-steady-state transient signal based on short-time energy and zero-crossing rate.

[0163] The first fault judgment module 23 is used to determine whether a fault exists when the vibration signal is a steady-state signal by using time-domain feature extraction and classification model.

[0164] The second fault judgment module 24 is used to determine whether a fault exists when the vibration signal is a periodic impact signal by using Fourier transform and envelope spectrum analysis.

[0165] The third fault judgment module 25 is used to determine whether a fault exists by using time-frequency analysis when the vibration signal is a non-steady-state transient signal.

[0166] Further functional descriptions of the above modules are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0167] As a specific application embodiment of the present invention, such as Figure 3As shown, the fault monitoring device of the aeroengine specifically includes the following modules:

[0168] The signal acquisition module is used to install high-precision vibration sensors at key parts of the aeroengine, such as the compressor, turbine, and bearing housing, etc., to collect vibration data (vibration signals).

[0169] The preprocessing module is used to input the original vibration signal, first adopt wavelet threshold denoising, then normalize the denoised signal, that is, scale the signal amplitude to the interval [-1, 1], and then segment the normalized signal, that is, divide the signal window according to a fixed time length (such as 1 second), and output the segmented signal after cleaning.

[0170] The signal classification module is used to calculate the short-time energy and zero crossing rate (ZCR) through feature extraction, and classify the signal by comparing ZCR with the threshold. For example, the ZCR thresholds are 1 and 2. Among them, if 1 < ZCR < 2, the vibration signal is classified as a B-type signal, that is, a periodic impact signal, which has a periodic sudden increase in energy; if ZCR < 1, the vibration signal is classified as an A-type signal, that is, a steady-state signal, which has low-energy fluctuations; if ZCR > 2, the vibration signal is classified as a C-type signal, that is, a non-steady-state transient signal, which has a high-energy short-time burst.

[0171] The data mining module is used to match the algorithm for the classified signals. For A-type signals, algorithm 1 is adopted, that is, extract time-domain features (peak value and kurtosis, etc.) and input them into the SVM classifier to determine whether there is a fault; for B-type signals, algorithm 2 is adopted, that is, use FFT and envelope spectrum analysis to extract and determine whether there is a fault by marking the fault frequency; for C-type signals, algorithm 3 is adopted, that is, use wavelet transform time-frequency such as marking energy mutation detection; finally, output the abnormal score (such as 0-1 probability value) and the fault type label through the judgment result.

[0172] The alarm and display module is used to trigger an 800Hz buzzing sound (beeping for 0.2 seconds and cycling with a 0.5-second interval) and the red light flashes synchronously for 0.5 seconds when there is a fault, that is, an abnormality. When there is no fault, that is, normal, the green light is always on. For the display interface, a real-time time-frequency diagram (the horizontal axis is time and the vertical axis is frequency) is displayed in area 1, and the LED status indicator (green / red) is in area 2, and the alarm level and maintenance suggestions (such as blade crack alarm, code B-203) are pop-up displayed in area 3.

[0173] In an optional implementation manner, such as Figure 4As shown, the signal acquisition module is specifically used to: install high-precision vibration sensors at vibration monitoring locations on the aero-engine, such as the compressor, turbine, rotor bearing housing, combustion chamber, exhaust nozzle, engine mounting bracket, and other locations requiring vibration monitoring to collect vibration data. These high-precision vibration sensors include piezoelectric accelerometers, microelectromechanical system accelerometers, fiber optic grating vibration sensors, capacitive vibration sensors, and other types of high-precision sensors. During installation, the sensors are selected based on their characteristics and the location of the aero-engine being monitored. After installation, the vibration sensors are used to collect vibration signals from key parts of the engine in real time, with a sampling frequency of not less than 20kHz, and the raw vibration signals are output after acquisition.

[0174] In one alternative implementation, such as Figure 5 As shown, the preprocessing module is specifically used for: wavelet threshold denoising through wavelet basis selection, signal decomposition, threshold processing, and signal reconstruction; then, signal normalization is performed using linear amplitude scaling and outlier truncation, where ±3σ is used as the threshold, and truncation is performed when the threshold is exceeded, otherwise no truncation is performed; finally, signal segmentation is performed using window partitioning, windowing functions, and segmented data management (real-time mode and offline mode).

[0175] Among them, when the relevant parameter values ​​in the preprocessing module are as shown in Table 2 below, the implementation effect is evaluated and verified.

[0176] Table 2

[0177]

[0178]

[0179] The evaluation begins with assessing the noise reduction effect. Specifically, the signal-to-noise ratio improvement is calculated using the following formula:

[0180]

[0181] In the formula, x noisy x represents the vibration signal before denoising. clean This represents the vibration signal after denoising. When the signal-to-noise ratio improvement is greater than or equal to 10dB, it indicates that the denoising effect meets the requirements.

[0182] In addition, for the vibration signal segmentation window, a simulated fault pulse signal can be injected to verify whether the segmented window can complete the capture of transient events and perform segmented testing (if the pulse position deviation is less than 5ms, the test is considered passed).

[0183] In one alternative implementation, such as Figure 6As shown, the signal classification module is specifically used to: input the segmented signals from the preprocessing module, calculate short-time energy and zero-crossing rate. Steady-state signals (Class A signals) are characterized by low-frequency steady-state vibration and low-frequency components, resulting in low short-time energy and a low zero-crossing rate. Periodic impact signals (Class B signals), as impact events, contain high-frequency components in periodic impacts, thus exhibiting periodic spikes in short-time energy and a moderate zero-crossing rate. Non-steady-state transient signals (Class C signals), as transient events, contain broadband noise, resulting in sudden spikes in short-time energy and a relatively high zero-crossing rate. Therefore, signals can be classified based on the magnitude of the short-time energy and zero-crossing rate of the vibration signals in each window. For short-time energy, a short-time energy baseline (steady-state signal energy baseline TE) needs to be determined. base Two thresholds: ) and transient event energy threshold. Using 3×STE base STE as the energy threshold of transient events thresh_high For determining the zero-crossing rate, the first zero-crossing rate ZCR is used. low Second zero-crossing rate ZCR high Two thresholds, the size of which can be determined based on the actual situation, such as ZCR. low =0.1, ZCR high =0.3. Furthermore, the threshold determined here can be updated in real time based on the acquired data; for example, the STE can be recalculated every 24 hours based on the latest data. base With ZCR low ZCR high Adapt to changes in operating conditions. If STE k ≤STE base And ZCR k ≤ZCR low Then the vibration signal of the current window will be classified as a Class A signal (steady-state signal). If STE k STE base If the periodicity detection passes, the vibration signal in the current window is classified as a Class B signal (periodic impact signal). The method for periodicity detection includes performing autocorrelation analysis on the STE sequence; if a significant periodic peak exists (peak height > 0.7 times the maximum autocorrelation value), it is determined to be periodic. If the STE... k STE thresh_high And ZCR k >ZCR high Then the vibration signal of the current window will be classified into Class C signal (non-steady transient signal).

[0184] In practical applications, the values ​​of each parameter in the signal classification module are shown in Table 3 below:

[0185] Table 3

[0186]

[0187] Specifically, the classification process of this signal classification module can be illustrated and verified using an aero-engine bearing wear monitoring scenario: First, input the preprocessed segmented signal (Class A signal, 50ms window); then calculate the STE of the current window. k =0.05×10 -3 ZCR k =0.08; then determine STE k <STE base (0.05×10 -3 <0.1×10 -3 ZCR k <ZCR low (0.08 < 0.1); This classifies the signal into class A, triggering SVM time-domain feature analysis. Therefore, the classification process is relatively accurate.

[0188] The classification effect of the signal classification module can also be verified. The specific classification results are shown in Table 4 below:

[0189] Table 4

[0190]

[0191] Alternatively, the classification results can be used to plot the ROC (Receiver Operating Characteristic) curve. When the AUC (Area Under Curve) > 0.95, it proves the effectiveness of the classification results.

[0192] In one alternative implementation, such as Figure 7 As shown, the data mining module is specifically used for: For Class A signals, determining the fault probability based on time-domain feature extraction, feature standardization, SVM classifier training, and inference; if the fault probability is greater than 0.8, a fault type alarm is triggered; otherwise, it is considered normal. This signal processing is suitable for low-frequency steady-state fault scenarios such as bearing wear, shaft misalignment, and poor lubrication. For Class B signals, determining whether to trigger a fault type alarm is based on Fast Fourier Transform, envelope spectrum analysis, fault frequency (feature frequency) identification, peak detection, and harmonic matching. This signal processing is suitable for periodic impact fault scenarios such as blade cracks, gear tooth breakage, and rolling element spalling. For Class C signals, determining the presence of a fault, i.e., whether to trigger a fault type alarm, is based on continuous wavelet transform, time-frequency energy mutation detection, and Hilbert-Huang transform-assisted verification (HHT verification).

[0193] Specifically, this processing procedure can be illustrated using different application scenarios. For example, when the classified vibration signal is a Class A signal (such as bearing vibration data), time-domain features are extracted with kurtosis K = 4.2 and impulse factor IF = 5.0. These features are then input into an SVM classifier, which outputs the fault P. fault =0.92; then a bearing wear alarm is triggered. When the classified vibration signal is a Class B signal (turbine blade vibration data), Fourier transform analysis is performed to determine the fundamental frequency f. rotor =200Hz, number of blades N blade =20, theoretical f BPF =4kHz; then envelope spectrum analysis was performed, and 2×f was detected. BPF If the amplitude increases by 40% at 8kHz, a blade crack alarm is triggered. When the classified vibration signal is a Class C signal (compressor surge data), a continuous wavelet transform is first performed. Within a 50ms window, multiple scale energy surges (amplitude exceeding the threshold by 5 times) are detected. Then, HHT verification is performed. If the energy concentration area of ​​the first IMF component is consistent with the CWT result, then a "surge warning" alarm is triggered.

[0194] The parameters of the data mining module are shown in Table 5 below:

[0195] Table 5

[0196]

[0197] In addition, the processing algorithms for various signals can be validated. For Class A algorithms, using the bearing life-cycle dataset, the SVM classification accuracy is ≥95% (F1-score); the fault early warning time is ≥2 hours (based on kurtosis trend analysis). For Class B algorithms, the sensitivity of envelope spectrum detection for gear tooth breakage faults is 98% (false negative rate ≤2%); the frequency resolution error is ≤0.5Hz. The surge event detection delay is ≤10ms; the time-frequency positioning accuracy (synchronized with high-speed camera) error is ≤1ms.

[0198] In one alternative implementation, such as Figure 8 As shown, the alarm and display module is specifically used for: hardware and software design. Hardware design includes determining the hardware component list and circuit connections, while software design includes microcontroller firmware development and host computer interface development. Based on the designed hardware and software, it performs hardware assembly, firmware debugging, and system integration debugging. Problems encountered during debugging, such as buzzer and LED synchronization issues, time-frequency graph stuttering, and communication packet loss, are resolved before re-debugging. After debugging, normal state testing, abnormal state testing, and stress testing are performed. If no problems are found during the tests, debugging is complete. The system is then used for alarm and display functions.

[0199] This invention also provides a computer device having the above-described features. Figure 2 The image shows a fault monitoring device for an aircraft engine.

[0200] Please see Figure 9 , Figure 9 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 9 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 9 Take a processor 10 as an example.

[0201] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0202] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0203] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device as shown by a landing page for an app. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0204] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0205] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0206] 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, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0207] 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.

[0208] 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 fault monitoring method for an aero-engine, characterized in that, The method includes: To acquire vibration signals from various parts of an aero-engine during operation; The vibration signal is classified into steady-state signal, periodic impact signal, or non-steady-state transient signal based on short-time energy and zero-crossing rate. When the vibration signal is a steady-state signal, time-domain feature extraction and classification models are used to determine whether a fault exists. When the vibration signal is a periodic impact signal, Fourier transform and envelope spectrum analysis are used to determine whether a fault exists. When the vibration signal is a non-steady-state transient signal, time-frequency analysis is used to determine whether a fault exists; The vibration signal is classified into steady-state signal, periodic impact signal, or non-steady-state transient signal based on short-time energy and zero-crossing rate, including: When the short-time energy is less than or equal to the short-time energy baseline and the zero-crossing rate is less than or equal to the first zero-crossing rate threshold, the vibration signal is classified as a steady-state signal. When the short-time energy is greater than the short-time energy baseline and meets the periodic detection conditions, the vibration signal is divided into a periodic impact signal. When the short-time energy is greater than the transient event energy threshold and the zero-crossing rate is greater than the second zero-crossing rate threshold, the vibration signal is classified as a non-steady-state transient signal. The transient event energy threshold is greater than the short-time energy baseline, and the second zero-crossing rate threshold is greater than the first zero-crossing rate threshold. When the vibration signal is a steady-state signal, time-domain feature extraction and classification models are used to determine whether a fault exists, including: The peak value, kurtosis, impulse factor, and root mean square of the vibration signal are extracted as time-domain features; The time-domain features are standardized. The standardized temporal features are input into a support vector machine classifier to obtain the output result indicating whether a fault exists. When the vibration signal is a periodic impact signal, Fourier transform and envelope spectrum analysis are used to determine whether a fault exists, including: Perform a Fourier transform on the vibration signal to determine the characteristic frequency; The envelope of the vibration signal is extracted and subjected to Fourier transform to obtain the envelope spectrum; The presence of a fault is determined based on the relationship between the peak values ​​and characteristic frequencies exceeding three standard deviations from the baseline in the envelope spectrum and their harmonics. When the vibration signal is a non-steady-state transient signal, time-frequency analysis is used to determine whether a fault exists, including: Continuous wavelet transform was used to determine the wavelet coefficients of the vibration signal at different scales and time points; The dynamic threshold is determined based on the squared modulus, mean, and standard deviation of wavelet coefficients; The presence of a fault is determined based on whether multiple scales at the same point in time exceed the dynamic threshold. The time-frequency energy distribution obtained by Hilbert-Huang transform is used to verify the existence of the fault.

2. The method according to claim 1, characterized in that, Before acquiring vibration signals from various parts of the aero-engine during operation, the method further includes: The parts of the aero-engine to be monitored and the types of sensors used for vibration signal acquisition are obtained. The parts include the compressor, turbine, rotor bearing, combustion chamber, tail nozzle and engine mounting bracket. The sensor types include piezoelectric accelerometers, microelectromechanical system accelerometers, fiber optic grating vibration sensors and capacitive vibration sensors. The type of sensor to be installed in each part is determined based on the environment in which each part is located and the frequency band of the vibration signal generated by each part.

3. The method according to claim 1, characterized in that, After acquiring vibration signals from various parts of the aero-engine during operation, the method further includes: Wavelet threshold denoising algorithm is used to remove high-frequency noise and low-frequency drift from vibration signals; The vibration signal is normalized and segmented.

4. The method according to claim 1, characterized in that, The method further includes: When a fault is detected, the buzzer is controlled to sound an alarm at a preset frequency, and the red light is controlled to flash. When it is determined that there is no fault, the buzzer is silenced and the green light remains on.

5. A fault monitoring device for an aircraft engine, characterized in that, The device includes: The signal acquisition module is used to acquire vibration signals from various parts of the aero-engine during operation. A signal classification module is used to classify the vibration signal into steady-state signals, periodic impact signals, or non-steady-state transient signals based on short-time energy and zero-crossing rate. Classifying the vibration signal into steady-state signals, periodic impact signals, or non-steady-state transient signals based on short-time energy and zero-crossing rate includes: classifying the vibration signal as a steady-state signal when the short-time energy is less than or equal to a short-time energy baseline and the zero-crossing rate is less than or equal to a first zero-crossing rate threshold; classifying the vibration signal as a periodic impact signal when the short-time energy is greater than the short-time energy baseline and meets the periodic detection condition; and classifying the vibration signal as a non-steady-state transient signal when the short-time energy is greater than a transient event energy threshold and the zero-crossing rate is greater than a second zero-crossing rate threshold, wherein the transient event energy threshold is greater than the short-time energy baseline and the second zero-crossing rate threshold is greater than the first zero-crossing rate threshold. The first fault determination module is used to determine whether a fault exists when the vibration signal is a steady-state signal by employing a time-domain feature extraction and classification model. This process includes: extracting the peak value, kurtosis, impulse factor, and root mean square of the vibration signal as time-domain features; standardizing the time-domain features; and inputting the standardized time-domain features into a support vector machine classifier to obtain an output result indicating whether a fault exists. The second fault diagnosis module is used to determine whether a fault exists when the vibration signal is a periodic impact signal, using Fourier transform and envelope spectrum analysis. This includes: performing a Fourier transform on the vibration signal to determine the characteristic frequency; extracting the envelope of the vibration signal and performing a Fourier transform to obtain the envelope spectrum; and determining whether a fault exists based on the relationship between the peak value exceeding three standard deviations from the baseline in the envelope spectrum and the characteristic frequency and its harmonics. The third fault determination module is used to determine whether a fault exists when the vibration signal is a non-steady-state transient signal using time-frequency analysis. This includes: using continuous wavelet transform to determine the wavelet coefficients of the vibration signal at different scales and time points; determining a dynamic threshold based on the squared modulus, mean, and standard deviation of the wavelet coefficients; determining whether a fault exists based on whether multiple scales at the same time point exceed the dynamic threshold; and verifying the determined fault using the time-frequency energy distribution obtained by Hilbert-Huang transform.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the fault monitoring method for an aero-engine as described in any one of claims 1 to 4.

Images