A method and system for monitoring the operating state of an engine

Abstract

The application relates to the technical field of data processing, in particular to an engine operation state monitoring method and system, which comprises the following steps: acquiring vibration signals and a crankshaft rotating speed signal of an engine in real time; converting the vibration signals from a time domain to an angle domain based on the crankshaft rotating speed signal; performing order spectrum analysis on the vibration signals in the angle domain to generate real-time order spectrum, and comparing the real-time order spectrum with a benchmark order spectrum to determine the energy deviation of each order; meanwhile, extracting waveform segments of each order in continuous multiple period vibration signals, calculating the similarity between the waveform segments, evaluating the shape dispersion degree and quantifying the shape stability degree; finally, fusing the energy deviation and the shape stability degree to determine a comprehensive fault risk index of each order, and monitoring the operation state of the engine based on the comprehensive fault risk index of each order. The method improves the accuracy and reliability of engine state monitoring.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. Specifically, it relates to a method and system for monitoring engine operating status. Background Technology

[0002] As the core power source for various equipment such as industrial machinery, transportation, and power generation, real-time monitoring of the engine's operating status is a crucial link in ensuring the reliable operation of the entire system and mitigating safety risks.

[0003] Vibration signals reflect the mechanical motion state of an engine and contain a wealth of fault characteristic information. For example, imbalances in rotating parts can trigger vibrations at specific frequencies, wear on reciprocating parts can cause abnormal fluctuations in vibration amplitude, and even early, minute crack propagation can be manifested through distortion of the vibration waveform. Vibration signal analysis-based operational condition monitoring technology, with its non-invasive measurement advantage, has become a core means of monitoring engine operational conditions for fault diagnosis.

[0004] In the existing technology, spectrum analysis based on fast Fourier transform is the mainstream method for vibration signal processing. This method converts the time-domain vibration signal to the frequency domain, first identifies the characteristic frequencies related to the fault, such as the rotation frequency and harmonics of rotating parts, and then judges the fault by comparing the amplitude of the characteristic frequencies with a preset threshold.

[0005] However, the traditional Fast Fourier Transform (FFT) is based on the assumption of stationarity, which requires that the frequency characteristics of the signal remain constant during the analysis period. However, engines often face dynamic speed changes in actual operating conditions, such as acceleration, deceleration, and load fluctuations, which cause the vibration signal to exhibit significant non-stationary characteristics. This makes it difficult for the traditional FFT method to adapt. When the speed changes, the characteristic frequencies that are strongly correlated with the speed will drift over time, resulting in the broadening and attenuation of the characteristic spectrum peaks in the FFT result, forming spectral ambiguity. This makes it impossible to capture the true fault characteristics, causing missed or false alarms and affecting the accuracy of condition monitoring.

[0006] Therefore, there is an urgent need for a method that can adapt to engine operating conditions with varying speeds and loads, and accurately extract features from non-stationary vibration signals in order to accurately monitor the engine's operating status. Summary of the Invention

[0007] To address the problem that existing technologies, when analyzing vibration signals based on Fast Fourier Transform, cannot effectively track dynamic characteristic frequencies, are prone to spectral ambiguity and feature drift, resulting in low accuracy in engine operating status monitoring, this invention proposes an engine operating status monitoring method and system.

[0008] In a first aspect, the present invention provides a method for monitoring engine operating status, comprising:

[0009] The vibration signal and crankshaft speed signal of the engine are acquired in real time; using the crankshaft speed signal as a reference, the vibration signal is converted from the time domain to the angular domain to obtain the vibration signal in the angular domain.

[0010] The vibration signal in the angular domain is subjected to order spectrum analysis to generate a real-time order spectrum with each order as the abscissa and the vibration amplitude of each order as the ordinate. The real-time order spectrum is compared with the pre-acquired reference order spectrum to determine the energy deviation of each order in the real-time order spectrum.

[0011] For any order in the real-time order spectrum, extract all waveform segments corresponding to that order in multiple consecutive engine cycles as all waveform segments corresponding to that order, calculate the similarity between any two waveform segments, evaluate the morphological dispersion of all waveform segments corresponding to that order based on the statistical characteristics of the similarity between all any two waveform segments, and use the morphological dispersion as a negative index to quantify the morphological stability of all waveform segments corresponding to that order.

[0012] By integrating the energy deviation of each order in the real-time order spectrum and the morphological stability of all waveform segments corresponding to each order, the comprehensive failure risk index of each order is determined, and the engine operating status is monitored based on the comprehensive failure risk index of each order.

[0013] This technical solution cleverly transforms the analysis benchmark from the volatile time dimension to a physical benchmark—the crankshaft angle synchronized with the engine cycle. By employing order analysis, it fundamentally converts non-stationary vibration signals under varying engine speeds into stable order spectra, avoiding the inaccurate feature extraction problems caused by frequency drift and spectral ambiguity in traditional Fast Fourier Transform. Furthermore, it abandons fixed alarm thresholds and establishes a health benchmark model that dynamically changes with real-time output power, achieving condition-adaptive quantification of vibration energy deviation and reflecting abnormal conditions of the vibration signal under different load conditions. More importantly, it introduces the analysis of the morphological stability of the vibration signal, equivalent to quantifying the periodic stability of the vibration source. This effectively distinguishes between highly repeatable deterministic signals caused by real mechanical faults and random interference signals caused by combustion instability or external impacts. By nonlinearly fusing the energy deviation, which represents the severity of a fault, with the morphological stability, which represents the reliability of a diagnosis, a comprehensive fault risk index with both sensitivity and robustness is constructed. This makes the monitoring conclusion no longer a simple judgment based on a single dimension, but rather a comprehensive assessment that couples the energy significance of a fault with its periodic determinism by nonlinearly fusing the energy deviation index, which represents the severity of a fault, with the cyclic stability index, which represents the reliability of a diagnosis. This enhances the ability to distinguish between real mechanical fault sources and random disturbances and improves the accuracy of engine operating status monitoring.

[0014] Preferably, the vibration signal in the angular domain is determined based on the following method:

[0015] By integrating the crankshaft speed signal over time, the rotation angle of the crankshaft at each moment is calculated. Based on the time synchronization principle between the vibration signal and the crankshaft speed signal, the rotation angle and vibration amplitude corresponding to the vibration signal at each moment are determined. The vibration signal is divided into multiple periodic vibration signals with a rotation angle of 360 degrees as one period. The rotation angle from 0 to 360 degrees is used as the abscissa, and the average vibration amplitude corresponding to each rotation angle in all periodic vibration signals is used as the ordinate of the rotation angle to obtain the vibration signal in the angle domain.

[0016] This technical solution establishes a direct mapping relationship between vibration signals and the physical rotation angle of the engine, and performs synchronous averaging on multiple engine cycles within the angle domain. This effectively filters out non-periodic random noise interference, such as combustion instability, and enhances the deterministic mechanical vibration characteristics synchronized with the crankshaft's rotation angle.

[0017] Preferably, the real-time order spectrum is determined based on the following method: performing a Fourier transform on the vibration signal in the angular domain to decompose it into a series of vibration modes with specific periodic characteristics, each vibration mode corresponding to the vibration characteristics of a specific engine component; taking the number of repetitions of each vibration mode within every 360 degrees of crankshaft rotation as the order of the vibration mode, and characterizing each vibration mode through each order, taking the vibration amplitude of the vibration mode characterized by each order as the vibration amplitude of that order; generating a real-time order spectrum with each order as the abscissa and the vibration amplitude of each order as the ordinate.

[0018] Preferably, the energy deviation of each order in the real-time order spectrum is determined as follows: All reference order spectra under the real-time load conditions of the engine are obtained; for any order in the real-time order spectrum, the mean and standard deviation of the vibration amplitude of that order in all reference order spectra are used as the reference amplitude and the fluctuation value of the reference amplitude of that order, respectively; the absolute difference between the vibration amplitude and the reference amplitude of that order, and the cumulative value of the reference amplitude and the fluctuation value of the reference amplitude are calculated; the ratio of the absolute difference to the cumulative value is determined as the normalized deviation rate of the vibration amplitude of that order; the ratio of the vibration amplitude of that order to the reference amplitude is determined as the energy gain factor of that order; and the energy deviation of that order is determined by multiplying the normalized deviation rate of the vibration amplitude of that order and the energy gain factor of that order.

[0019] This technical solution implements a synergistic amplification mechanism. For small early deviations, the index shows an approximately linear response, while for significant fault symptoms, the energy gain factor acts as a powerful nonlinear amplifier, enabling the final energy deviation index to grow exponentially, thereby greatly improving the monitoring system's quantitative sensitivity to the initiation and development of faults.

[0020] Preferably, the reference order spectrum is determined as follows: the engine in a healthy state is operated under multiple different load conditions; under each load condition, multiple vibration signals and crankshaft speed signals are collected as reference vibration signals and reference crankshaft speed signals, respectively; a reference order spectrum is determined based on each reference vibration signal and the corresponding reference crankshaft speed signal, thus obtaining multiple reference order spectra under that load condition; wherein, the method for determining each reference order spectrum is consistent with the method for determining the real-time order spectrum.

[0021] Preferably, the similarity between any two waveform segments is determined as follows: for any two waveform segments among all waveform segments corresponding to any first order, they are respectively denoted as the first waveform segment and the second waveform segment; the first waveform segment and the second waveform segment are subjected to mean removal processing to eliminate DC offset and retain only waveform morphology features; the normalized cross-correlation coefficient of the first waveform segment and the second waveform segment after mean removal processing under different time delays is calculated; the peak value of the normalized cross-correlation coefficient is taken as the similarity between the first waveform segment and the second waveform segment.

[0022] Preferably, the morphological dispersion is determined as follows: the morphological difference between the first waveform segment and the second waveform segment is determined based on the similarity between the first waveform segment and the second waveform segment, wherein the morphological difference is negatively correlated with the similarity; the first waveform segment and the second waveform segment are combined, and the sum of the squares of the morphological differences corresponding to the combination is divided by the total number of combinations to obtain an average morphological difference; the average morphological difference is used as the morphological dispersion of all waveform segments corresponding to this order.

[0023] This technical solution constructs a statistical evaluation framework that compares all waveforms of the target order in a continuous cycle pairwise. By summing the squares of the morphological differences and then normalizing them, a discrete sum that depends on the total number of cycles is transformed into a standardized, horizontally comparable average morphological difference. This provides an objective quantitative indicator for quantifying whether the vibration of this order originates from the periodic reproducibility of deterministic mechanical faults or from the instability of random disturbances.

[0024] Preferably, the comprehensive failure risk index for each level is determined based on the following method:

[0025] For any order, the morphological stability of all waveform segments corresponding to that order is taken as input and transformed by a nonlinear activation function to generate a gating weight; the energy deviation of that order and the gating weight are weighted and fused to obtain the comprehensive fault risk index of that order.

[0026] This technical solution, by introducing a nonlinear gating fusion mechanism, physically simulates expert diagnostic logic. This design achieves a confidence-weighted assessment of fault severity; only when the system has a high confidence level in determining that the vibration originates from a persistent mechanical fault is the energy deviation representing fault severity allowed to be fully included in the final risk. This endows the monitoring method with the ability to distinguish between real faults and incidental disturbances, improving the robustness and reliability of diagnostic conclusions.

[0027] Preferably, the engine's operating status is monitored based on the comprehensive fault risk index of each order, including: pre-acquiring the correlation between the vibration mode corresponding to each order and specific components of the engine, as well as the normal range of the comprehensive fault risk index of each order; if the comprehensive fault risk index of a certain order exceeds the normal range of the comprehensive fault risk index of that order, it is determined that the operating status of the specific components of the engine associated with the vibration mode of that order is abnormal; if the comprehensive fault risk index of a certain order does not exceed the normal range of the comprehensive fault risk index of that order, it is determined that the operating status of the specific components of the engine associated with the vibration mode of that order is not abnormal.

[0028] Secondly, the present invention also provides an engine operating status monitoring system, the engine operating status monitoring system including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of any of the engine operating status monitoring methods.

[0029] The present invention has the following effects:

[0030] This invention transforms non-stationary vibration signals into stable order spectra by applying order spectrum analysis, overcoming the spectral distortion caused by the stationarity assumption of Fast Fourier Transform. Furthermore, by analyzing the health benchmark and waveform stability in relation to the real-time load, it can adaptively assess the energy deviation and reliability of fault severity, thereby improving the accuracy and reliability of engine condition monitoring. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0032] Figure 2 This is a schematic diagram of the vibration signal of the engine in this invention;

[0033] Figure 3This is a schematic diagram of the crankshaft speed signal of the engine in this invention;

[0034] Figure 4 This is the spectrum obtained by analyzing the vibration signal using traditional FFT in this invention;

[0035] Figure 5 This is a schematic diagram of the vibration signal in the angular domain of this invention;

[0036] Figure 6 This is a schematic diagram of the real-time order spectrum in this invention;

[0037] Figure 7 This is a schematic diagram comparing the morphology of the second-order vibration mode in multiple working cycles in this invention.

[0038] Figure 8 This is a schematic diagram of the final comprehensive fault risk index monitoring results generated in this invention. Detailed Implementation

[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0040] This invention provides a method for monitoring the operating status of an engine. By combining order analysis, adaptive operating conditions, and cyclic statistics, it achieves accurate and robust health status assessment of variable-speed engines. (Refer to...) Figure 1 The specific process of this method includes the following steps:

[0041] S1: Real-time acquisition of engine vibration signals and engine crankshaft speed signals.

[0042] This step forms the data foundation of the entire monitoring and analysis process. To achieve a comprehensive and accurate characterization of the engine's condition, it is essential to simultaneously acquire various physical quantities that reflect its mechanical vibration, motion state, and workload. Vibration signals are the core basis for diagnosing mechanical faults, crankshaft speed signals are a necessary benchmark for eliminating non-stationary influences and transitioning to the order domain, while output power is key data reflecting workload, enabling adaptive operation, and establishing a dynamic health benchmark.

[0043] High-frequency acceleration sensors are placed on key load-bearing components of the engine (such as the cylinder head or engine block) to collect vibration signals that reflect internal mechanical shocks. Simultaneously, photoelectric encoders or Hall effect sensors are installed on the engine crankshaft to acquire high-resolution angular pulse signals. By performing real-time calculations on these pulse signals, the instantaneous rotational speed of the crankshaft is obtained, yielding the crankshaft speed signal. An external power sensor acquires the engine's real-time output power, which directly reflects the engine's load intensity. Furthermore, relevant information about the engine cycle is obtained via the vehicle's CAN (Controller Area Network) bus. The CAN bus is a serial communication bus specifically designed for automobiles, allowing communication between various electronic control units and sensors within the vehicle without the need for a central host, thus reducing complex wiring within the vehicle.

[0044] The definition of an engine cycle is based on the crankshaft rotation angle. One cycle of a four-stroke engine corresponds to a crankshaft rotation of 720°, and one cycle of a two-stroke engine corresponds to a crankshaft rotation of 360°. This angle period is fixed. When the crankshaft speed changes, the time required to complete the same angle will inevitably change. The higher the speed, the shorter the time of a single engine cycle, and the lower the speed, the longer the time.

[0045] To monitor the engine's operating status in real time, 100 engine cycles prior to any given moment are acquired. This approach ensures sufficient historical data to reduce random interference while reflecting recent trends. The sample size was determined based on historical experience. Vibration and crankshaft speed signals corresponding to these 100 engine cycles are collected and analyzed for the same duration to monitor the engine's operating status.

[0046] In a four-stroke engine, the camshaft speed, which controls the opening and closing of the valves, is half the crankshaft speed. That is, the camshaft rotates once for every two crankshaft rotations. For the most common four-stroke engine, one engine cycle is equal to two crankshaft rotations. Therefore, the vibration signal and crankshaft speed signal corresponding to 100 engine cycles are equal to the vibration signal and crankshaft speed signal corresponding to 200 crankshaft rotations. This vibration signal and crankshaft speed signal can be used as the data basis for subsequent operations.

[0047] like Figure 2 and Figure 3 As shown, Figure 2 The collected vibration signal has a waveform that becomes increasingly dense over time, exhibiting typical non-stationary characteristics. Figure 3 The crankshaft speed signal acquired synchronously, whose speed increases smoothly over time with slight fluctuations, truly reflects the dynamic operation of the engine. These two synchronous signals, which have non-stationary characteristics, are the starting point for all subsequent analyses.

[0048] S2: Convert the vibration signal from the time domain to the angular domain and generate a real-time order spectrum.

[0049] This step is a prerequisite for eliminating the influence of speed fluctuations and achieving accurate monitoring of operating conditions. Essentially, it stems from the strong coupling between the frequency components of the vibration signal and the crankshaft motion, as well as the non-stationarity of the vibration signal.

[0050] From the perspective of mechanical vibration sources, almost all major engine vibrations are directly related to crankshaft rotation: the centrifugal force vibration frequency caused by crankshaft imbalance is exactly equal to the crankshaft's rotational speed; in a four-stroke engine, the reciprocating motion of the piston, completing a power cycle every two revolutions, results in an inertial force vibration frequency twice the crankshaft's rotational speed; in gear transmission systems, the meshing frequency is the product of the number of gear teeth and the crankshaft's rotational speed; even the impact vibration from valve opening and closing is proportional to the camshaft's rotational speed. These vibration sources do not exist in isolation, but rather change closely following the crankshaft's rotational speed.

[0051] When the crankshaft's rotational speed is stable, the instantaneous rotational speed is constant, and the aforementioned vibration frequencies also remain constant. At this time, the power spectral density of the vibration signal (reflecting the statistical characteristics of the frequency component distribution) exhibits a stable peak value, consistent with the characteristics of a stationary signal (whose statistical characteristics do not change with time). However, when the rotational speed changes (such as acceleration, deceleration, or idling fluctuations), the instantaneous rotational speed becomes a function of time, causing all vibration frequencies to synchronously become functions of time. For example, when the crankshaft speed increases from 1000 rpm to 3000 rpm, the centrifugal force vibration frequency increases from 16.7 Hz to 50 Hz, while the inertial force vibration frequency increases from 33.3 Hz to 100 Hz.

[0052] This continuous shift in frequency components over time causes the statistical characteristics of the vibration signal, such as the peak position and energy distribution, to change over time. The periodicity of the waveform is also broken, and the vibration mode within the same time interval no longer repeats. Therefore, changes in the crankshaft's rotational speed will alter the core frequency components of the vibration signal, causing its statistical characteristics to lose time stability. This results in the vibration signal exhibiting non-stationary characteristics. These non-stationary characteristics are not interference, but rather an inevitable manifestation of the engine's mechanical motion under dynamic operating conditions.

[0053] This non-stationary characteristic introduces errors into traditional Fast Fourier Transform (FFT) analysis methods that rely on frequency stability. Faced with frequency-drifted signals, the FFT cannot generate sharp, clear fault spectral peaks; instead, it produces a phenomenon known as spectral diffusion, which blurs the fault energy into a hazy, broadband energy blob. The direct consequence is that key fault features are diluted, the signal-to-noise ratio drops sharply, and the fault is ultimately submerged in background noise, leading to serious missed diagnoses and misdiagnoses.

[0054] like Figure 4 As shown in the diagram, as a comparative example, performing a traditional Fast Fourier Transform (FFT) on the vibration signal reveals that the characteristic frequencies related to the rotational speed drift over time due to changes in rotational speed. The FFT results fail to form clear, sharp spectral peaks; instead, the fault energy is obscured as a blurry, broadband energy cluster—a phenomenon known as spectral ambiguity. This dilutes key fault characteristics, drastically reduces the signal-to-noise ratio, and makes accurate condition assessment impossible.

[0055] Therefore, angle-domain resampling is necessary. The purpose is to transform the non-stationary vibration signal from the time domain to the angle domain, converting it into a pseudo-stationary signal with equal angular intervals that is unaffected by rotational speed. Only in this way can effective order analysis be performed, accurately capturing the true characteristics of the fault.

[0056] Specifically, this step achieves dimensional transformation of the analysis benchmark through order tracking technology. First, the instantaneous rotation angle of the crankshaft is calculated, then the vibration signal in the angle domain is obtained by resampling in the angle domain, and finally, the angle domain signal is subjected to fast Fourier transform to obtain the real-time order spectrum.

[0057] Specifically, it includes:

[0058] S21: Establish a precise mapping relationship between time and angle.

[0059] The crankshaft speed signal is integrated over time to calculate the crankshaft rotation angle at each moment. Based on the time synchronization principle of the vibration signal and the crankshaft speed signal, the rotation angle and vibration amplitude corresponding to the vibration signal at each moment are determined.

[0060] Since the instantaneous rotational speed of the crankshaft is the rate of change of the instantaneous rotational angle with respect to time (angular velocity), by integrating the instantaneous rotational speed over time, the total angle that the crankshaft has accumulated at any given moment can be obtained, which can be taken as the instantaneous rotational angle of the crankshaft at that moment.

[0061] For example, the crankshaft in instantaneous rotation angle at a given moment for:

[0062]

[0063] In this formula, It is the integral symbol, which represents the process of accumulation. It is the initial time; the entire integral formula represents the time from the initial time to... During this time period, every instantaneous tiny rotation angle is accumulated, thus obtaining the crankshaft's rotation angle at that moment. instantaneous rotation angle at a given moment . Indicates rotational speed. It is an integral variable, representing the time from the initial time to... Any instant within this time period. Indicates in The crankshaft speed at this instant is measured in revolutions per second. It's pi (π), and 2π is a key conversion factor here. In mathematics and physics, the standard unit of angle is the radian, and a full circle corresponds to... Radius. Due to The unit is revolutions per second, multiply it by Later, it was changed from "how many revolutions per second" to "how many arcs per second". It is the differential element of the integral.

[0064] In summary, by performing a definite integral of the instantaneous angular velocity over the time interval, it is possible to achieve the integration of each minute time interval. This involves the precise accumulation of minute angular displacements occurring within the crankshaft. The final result of this integration is the total angle rotated by the crankshaft from the initial moment to any other moment. This establishes a precise and continuous functional mapping from the time domain to the angle domain, yielding the crankshaft's rotation angle at each instant. The instantaneous rotation angle calculated using this method is in radians, the standard unit for angles in scientific calculations. For engineering purposes or visualization needs, it can be converted to degrees using the angle-radian conversion formula.

[0065] Next, since the vibration signal and crankshaft speed signal are acquired synchronously in time, the vibration amplitude of the vibration signal at each moment can be obtained, as well as the rotation angle of the crankshaft speed signal at each moment. Therefore, the rotation angle and vibration amplitude of the vibration signal at each moment can be obtained.

[0066] S22: Resampling yields vibration signals in the angular domain.

[0067] Taking a 360-degree rotation angle as one cycle, which is exactly one revolution of the crankshaft, the vibration signal is divided into multiple cycle vibration signals. The vibration signal in step S1 is 100 engine cycles, so there are a total of 200 crankshaft rotation cycles, that is, the crankshaft rotates 200 times.

[0068] Using rotation angles from 0 to 360 degrees (equal intervals of 1 degree) as the horizontal axis, and the average vibration amplitude corresponding to each rotation angle in all periodic vibration signals as the vertical axis, the vibration signal in the angle domain is obtained.

[0069] like Figure 5As shown, the vibration signal in the time domain is converted into a vibration signal in the angular domain. The inherent periodicity of the vibration signal is restored, exhibiting a stable shape. This operation not only eliminates waveform distortion caused by speed fluctuations but also forms a stable and reliable data foundation that can accurately characterize the core mechanical health of the engine, providing a prerequisite for subsequent precise order spectrum analysis.

[0070] This operation is a crucial step in transforming non-stationary time-domain signals into stable feature analysis. Its core logic lies in solving the challenge of vibration signal analysis under varying engine speeds through angle resampling and synchronous averaging. First, by forcibly converting the analysis reference from the volatile time axis to the crankshaft rotation angle that conforms to the physics of mechanical motion, it ensures that regardless of speed fluctuations, deterministic vibration events caused by mechanical structures such as imbalance and wear always reproduce at fixed angular positions, thus creating a stable analysis domain.

[0071] Based on this, the synchronous averaging step, which averages the vibration amplitude corresponding to each rotation angle in all cycles, realizes the function of a high-efficiency statistical filter. It can utilize the statistical characteristics of the signal: for periodic deterministic signals strongly correlated with mechanical faults, they are significantly enhanced by coherent superposition due to strict phase (rotation angle) alignment, while for asynchronous random noises such as combustion random fluctuations and background impacts, they cancel each other out by incoherent superposition during the averaging process due to the scattered phase (rotation angle).

[0072] Therefore, this operation ultimately yields a vibration signal in the angular domain with a high signal-to-noise ratio. This not only eliminates waveform distortion and spectral ambiguity caused by speed fluctuations, but also forms a stable and reliable data foundation that can accurately characterize the core mechanical health of the engine, providing a prerequisite for subsequent accurate order spectrum analysis.

[0073] S23: Perform a Fourier transform on the angle domain signal to generate a real-time order spectrum.

[0074] Performing a Fourier transform on the angle domain signal yields a real-time order spectrum, fundamentally eliminating analytical biases caused by non-stationarity and providing a stable and clear feature carrier for subsequent fault feature extraction.

[0075] Specifically, a Fourier transform is performed on the vibration signal in the angular domain to decompose it into a series of vibration modes with specific periodic characteristics. Each vibration mode corresponds to the vibration characteristics of a specific engine component. The number of repetitions of each vibration mode within every 360 degrees of crankshaft rotation is taken as the order of the vibration mode. Each order represents each vibration mode, and the vibration amplitude of the vibration mode represented by each order is taken as the vibration amplitude of that order. A real-time order spectrum is generated with each order as the abscissa and the vibration amplitude of each order as the ordinate. Each order corresponds to a vibration mode, and the vibration amplitude of each order reflects the energy intensity of that vibration mode.

[0076] In summary, performing a Fourier transform on a vibration signal in the angular domain essentially transforms the analytical benchmark of the vibration signal from the time domain to the order domain, thereby revealing the intrinsic relationship between vibration energy and engine rotational motion. The Fourier transform decomposes a vibration signal in the angular domain, which varies periodically with the crankshaft angle (e.g., 360°), into a linear superposition of a series of orders (harmonic components). The result is an energy distribution spectrum with the order as the abscissa, rather than the frequency energy distribution in traditional frequency domain analysis. The order is defined as the number of vibrations within one mechanical rotation cycle; it is a normalized physical quantity independent of rotational speed.

[0077] The advantage of this transformation is that it avoids the influence of speed variations on signal analysis. For example, the vibration mode corresponding to the first order precisely corresponds to the physical event that occurs once per revolution of the crankshaft. The order indicates that the frequency of the vibration mode is synchronized with the crankshaft's rotational frequency. Regardless of whether the engine speed is low or high, the physical law that this event occurs once per revolution of the crankshaft remains constant. Therefore, in the order spectrum, this vibration mode will always appear as a stable spectral peak at the first order, completely eliminating the characteristic frequency drift and spectral ambiguity problems caused by changes in engine speed in traditional Fourier transforms. Similarly, the vibration mode corresponding to the second order precisely corresponds to the physical event that occurs twice per revolution of the crankshaft; the frequency of this vibration mode is twice the crankshaft's rotational frequency, and so on.

[0078] Ultimately, the Fourier transform automatically identifies various vibration modes (such as ignition, piston movement, gear meshing, etc.) that are proportional to the engine speed. It represents these vibration modes by their order. Since the order characteristic is independent of engine speed, although the crankshaft rotation frequency and the frequency of the vibration modes change during engine acceleration or deceleration, their ratio (order) remains constant. This causes features that would otherwise drift and become blurred in the spectrum of the Fast Fourier Transform (FFT) result to form clear, sharp, and easily identifiable spectral peaks at fixed order positions in the order spectrum. This avoids the frequency drift and peak blurring problems encountered by traditional FFT spectral analysis under varying operating conditions, ensuring the stability and reliability of the diagnostic features.

[0079] like Figure 6 As shown, the real-time order spectrum generated after performing a Fourier transform on the signal in the angle domain is represented by the horizontal axis as the order, indicating the multiple relationship between the vibration frequency and the crankshaft speed, and the vertical axis as the amplitude, representing the energy intensity of that vibration mode (reflected by the vibration amplitude). The real-time order spectrum is the "vibration fingerprint" of the engine under the current operating conditions. Figure 6 A significant energy peak can be clearly seen at the second order, while the energies are lower at other orders. The real-time order spectrum provides a clear and stable characteristic carrier for subsequent quantification of energy deviation.

[0080] When a component malfunctions, it typically causes a significant increase in the energy of its corresponding order or harmonic order. Due to the stability of the spectral peak position, even during the dynamic process of engine acceleration or deceleration, the changes in the vibration amplitude of a specific order can be accurately tracked, thus clearly identifying the source of abnormal vibration.

[0081] In summary, the real-time order spectrum generated by performing Fourier transform on the angular domain signal is essentially a fingerprint spectrum of vibration energy distributed according to its order. It provides stable, clear, and physically meaningful characteristic evidence for engine condition monitoring and fault diagnosis under non-stationary operating conditions.

[0082] S3: Determine the energy deviation of each order in the real-time order spectrum.

[0083] Because engine vibration is not constant—for example, even a healthy engine will vibrate more violently when operating under full load (high load) than when idling (low load)—it is easy to make mistakes if the same fixed standard is used to judge whether the vibration is abnormal.

[0084] Therefore, in order to accurately and dynamically assess the engine's operating status, it is necessary to address the differences in vibration characteristics under different operating conditions. This requires systematically operating the engine under various typical load conditions and collecting vibration signals beforehand, once the engine is confirmed to be in a healthy state (e.g., after factory testing or major overhaul). By performing order analysis on these signals, a set of multiple benchmark order spectra covering the engine's main operating range can be constructed. These benchmark order spectra reflect and define the normal range of amplitude for each order under different loads, i.e., a healthy template.

[0085] In subsequent real-time monitoring, the system accurately compares the real-time order spectrum under the current operating condition (load conditions at the current moment) with the pre-acquired benchmark order spectrum under the same operating condition (same load conditions). This allows the system to determine the energy deviation of each order in the real-time order spectrum, thereby providing reliable data support for accurately identifying abnormal vibrations and judging the type and extent of faults.

[0086] Specifically, it includes:

[0087] S31: Determine the baseline order spectrum based on the engine's health status.

[0088] The engine in good condition is operated under multiple different load conditions, such as output power of 10kW, 20kW, 30kW, etc., to ensure that it covers common operating conditions;

[0089] Under each load condition, following the method in step S1, a vibration signal and a crankshaft speed signal are collected for each of 100 engine cycles, and used as the reference vibration signal and reference crankshaft speed signal under that load condition, respectively.

[0090] Based on the reference vibration signal and the corresponding reference crankshaft speed signal under the load condition, a reference order spectrum is determined according to the operation in step S2. That is, the method of determining each reference order spectrum under the load condition is the same as the method of determining the real-time order spectrum.

[0091] Considering that a single measurement may be affected by accidental factors, such as instantaneous fluctuations in oil quality, changes in ambient temperature, and random noise from the sensor, resulting in biased data that cannot fully represent the stable and healthy state under the operating condition, a total of 10 measurements are performed to obtain 10 reference order spectra under the load condition. This can effectively smooth out random errors and reduce the impact of random noise.

[0092] In summary, this method can yield 10 baseline order spectra of a healthy engine under each load condition.

[0093] S32: Compare the real-time order spectrum with the reference order spectrum to determine the energy deviation of each order.

[0094] Based on the engine's real-time load conditions (output power at the current moment), find all reference order spectra of the engine in a healthy state under the same load conditions; for any order in the real-time order spectrum, take the mean and standard deviation of the vibration amplitude of that order in all reference order spectra as the reference amplitude and the fluctuation value of the reference amplitude of that order, respectively.

[0095] For example, let the current time be... At any moment, the engine is The real-time load at any given moment is , The real-time order spectrum at time t is ,for The first in In each stage, the engine obtains its healthy state under load. All reference order spectra, the first in all reference order spectra The reference amplitude of each order and the Fluctuation value of the benchmark amplitude at each order .

[0096] It should be noted that if the real-time load condition is between two load conditions of an engine in a healthy state (in step S31), the reference amplitude of each order and the fluctuation value of the reference amplitude of each order under the real-time load condition can be calculated by methods such as linear interpolation, so as to achieve coverage of the entire operating range.

[0097] Calculate the absolute difference between the vibration amplitude of this order and the reference amplitude, as well as the cumulative value of the fluctuation values ​​of the reference amplitude and the reference amplitude; determine the normalized deviation rate of the vibration amplitude of this order by the ratio of the vibration amplitude of this order to the reference amplitude; determine the energy gain factor of this order by the ratio of the vibration amplitude of this order to the reference amplitude; and determine the energy deviation of this order by multiplying the normalized deviation rate of the vibration amplitude of this order and the energy gain factor of this order.

[0098] For example, Real-time order spectrum at time step The first in The energy deviation of each order is:

[0099]

[0100] In this formula, for Real-time order spectrum at time step The first in Energy deviation of each order As a single character, for The first in The amplitude of each order of vibration, For the first The reference amplitude of each order, For the first The fluctuation value of the baseline value of the vibration amplitude of each order. This is the absolute value symbol. In extreme cases, ... or If it is 0, then set it to a very small positive number, such as This is to avoid the denominator being 0, which would affect subsequent calculations.

[0101] In this formula, the first part is the first... The normalized deviation rate of the vibration amplitude of the th order reflects the th order in the real-time order spectrum. The deviation of each vibration amplitude from the reference value, and its proportion within the normal vibration amplitude range, achieves adaptive normalization of the deviation. For the first The absolute difference between the order vibration amplitude and the reference amplitude. For the first The sum of the baseline amplitude of the order and the fluctuation value of the baseline amplitude.

[0102] In this formula, the latter part is the energy gain factor for that order, reflecting the energy gain factor of the first order in the real-time order spectrum. The magnitude of each order of vibration is how many times the reference value, and the instantaneous energy gain is a nonlinear amplification factor.

[0103] The synergistic amplification effect was achieved by multiplying the normalized deviation rate and the instantaneous energy gain. When When it is only slightly exceeded, Slightly larger At that time, the instantaneous energy gain is approximately equal to 1. Primarily determined by the normalized deviation rate, it reflects approximately linearly... and The magnitude of the deviation. When Significantly higher than At this time, not only does the normalization bias increase, but the instantaneous energy gain also acts as a powerful gain coefficient, amplifying the result a second time. This makes... It exhibits extremely high sensitivity to drastic energy transitions.

[0104] In summary, this involves quantifying how much the real-time vibration energy deviates from its normal value under current operating conditions. This is a crucial step from qualitative comparison to quantitative assessment; a larger dynamic energy deviation indicates a more likely abnormal engine operating state, and vice versa. This energy deviation calculation formula is precisely designed for engine variable load and non-stationary operating conditions. Its core value lies in achieving sensitive capture of fault characteristics and full-condition adaptation through dual-dimensional collaborative quantification. The first part, the normalized deviation rate, incorporates the absolute deviation between the real-time amplitude and the reference amplitude into the framework of the reference value and the normal fluctuation range. This is achieved by dynamically adjusting the denominator (the higher the load, the greater the normal fluctuation). The larger the value, the better. This eliminates interference from differences in normal vibration amplitude under different loads, ensuring a unified standard for deviation evaluation under different operating conditions such as idling and full load, and avoiding false alarms or missed alarms caused by fixed thresholds. The energy gain factor in the latter part forms a non-linear amplification mechanism through the ratio of the real-time amplitude to the reference value. When the engine experiences early minor faults, the amplitude is slightly higher, and the energy gain factor is close to 1. The value primarily reflects linear deviation, ensuring that subtle anomalies are not ignored. As the fault worsens, the amplitude increases significantly, and the gain factor amplifies synchronously, creating a synergistic effect with the deviation rate, thus highlighting drastic energy jumps. This design not only adapts to the dynamic development process of engine faults from minor to severe but also ensures consistent assessment under different loads and orders, providing accurate quantitative basis for subsequent fault risk judgment.

[0105] S4: Determine the morphological stability of all waveform segments corresponding to each order in the real-time order spectrum.

[0106] In engine condition monitoring, changes in the order energy amplitude can provide a preliminary assessment of whether the engine's operating condition is abnormal. This step further considers that occasional factors such as unstable combustion or random external impacts can also cause a sudden increase in vibration energy, which can easily be confused with vibrations caused by real, persistent mechanical defects such as bearing pitting or gear tooth breakage.

[0107] Therefore, by analyzing the vibration characteristics of real mechanical faults, it can be seen that the vibration characteristics of genuine mechanical faults will stably reproduce in continuous cycles, and their waveform morphology exhibits periodic consistency. By quantifying the morphological stability of the waveform segments corresponding to each order in the real-time order spectrum, continuous verification of energy deviations can be provided, effectively distinguishing between incidental interference and genuine mechanical vibrations, and improving the reliability of operational status monitoring.

[0108] The specific process of this step is as follows: For any order in the real-time order spectrum, firstly, extract all waveform segments corresponding to that order in multiple consecutive engine cycles; then, perform mean removal processing on any two waveform segments to preserve morphological features, and determine the similarity between them by calculating the peak value of the normalized cross-correlation coefficient under different time delays; subsequently, based on the pairwise similarity between all waveform segments, convert the similarity into a negatively correlated morphological difference degree, and obtain the average morphological difference degree by calculating the average of the sum of squares of the morphological difference degrees of all combinations, thereby evaluating the morphological dispersion of the waveform segment of that order; finally, use the morphological dispersion degree as a negative indicator, the higher the dispersion degree, the lower the stability degree, and quantify the morphological stability degree of the waveform segment corresponding to that order.

[0109] Specifically, it includes:

[0110] S41: Extract all waveform segments corresponding to each order in the real-time order spectrum.

[0111] Since step S2 involves performing an angle domain transformation on the vibration signal at the current moment to obtain multiple periodic vibration signals with a period of 360 degrees per rotation of the crankshaft, and then performing frequency domain analysis on the periodic vibration signals through Fourier transform to extract each order and its corresponding characteristic parameters (such as vibration amplitude) from the vibration characteristics, this is to determine the order based on the vibration and obtain the real-time order spectrum.

[0112] In this step, the process of extracting all waveform segments corresponding to each order in the real-time order spectrum follows the technical logic of reverse positioning, specifically as follows:

[0113] Based on each order in the real-time order spectrum, the waveform segment matching the order (corresponding vibration mode) is reversely separated from each periodic vibration signal divided in step S2 using methods such as bandpass filtering, thus obtaining the waveform segment of the order in a single period; finally, all waveform segments extracted from all periodic vibration signals of the order are summarized to form all waveform segments corresponding to the order.

[0114] This operation, which first analyzes the order of the vibration signal and then reversely locates the waveform segment based on the order, is a conventional and reasonable technique used in order analysis to achieve refined feature extraction. The logic is consistent and conforms to the basic laws of signal processing.

[0115] like Figure 7As shown, this visually illustrates all waveform segments corresponding to the second-order (representative vibration mode) extracted from five consecutive vibration cycles. The five curves represent the changes in the waveform morphology of the second-order vibration mode across the five vibration cycles. It can be seen that these five curves largely overlap, exhibiting a highly consistent overall trend, yet with subtle differences. This aligns with the physical characteristics of real, persistent mechanical faults—"periodic recurrence" coupled with minor disturbances. By calculating the similarity between these waveform segments, their morphological stability can be quantified, providing a crucial basis for distinguishing between faults and disturbances.

[0116] S42: Calculate the similarity between any two waveform segments corresponding to each order.

[0117] After obtaining all waveform segments corresponding to each order, calculating the similarity between any two waveform segments is the core prerequisite for quantifying the degree of morphological stability and is irreplaceable. This is because the similarity of waveform segments directly reflects the consistency of the morphological characteristics of the same order vibration feature in different periods. If it is a persistent mechanical fault, its vibration waveform will stably reproduce in continuous cycles, and the similarity between any two waveform segments should be significantly high. If it is an occasional disturbance (such as unstable combustion or random impact), the waveform morphology will show low similarity due to irregular fluctuations.

[0118] By quantifying similarity, the subjective judgment of "whether the form is consistent" can be transformed into objective data, providing a basis for subsequent calculation of form difference and assessment of dispersion, and ultimately achieving accurate differentiation between stable recurring fault characteristics and random interference signals.

[0119] Specifically:

[0120] For any two waveform segments among all waveform segments corresponding to any first order, they are respectively denoted as the first waveform segment and the second waveform segment; the first waveform segment and the second waveform segment are subjected to mean removal processing to eliminate DC offset and retain only waveform morphology features; the normalized cross-correlation coefficient of the first waveform segment and the second waveform segment after mean removal processing under different time delays is calculated; the peak value of the normalized cross-correlation coefficient is taken as the similarity between the first waveform segment and the second waveform segment.

[0121] For example, specifically, the first Among all waveform segments corresponding to each order, the normalized cross-correlation coefficient between any two waveform segments is:

[0122]

[0123] In this formula, For the first The corresponding order of the first The waveform segment and the first Normalized cross-correlation coefficients of each waveform segment For the first The waveform segment and the first The time sequence number of each waveform segment. The range of values ​​is , For the first The waveform segment and the first The total number of moments in the waveform segment, due to the number of moments in the waveform segment The waveform segment and the first If the time lengths of the waveform segments are consistent, then the time length of the first waveform segment is consistent. At the [time], we will get the [number]th [time]. The vibration amplitude of the first waveform segment at that moment and the first The vibration amplitude of each waveform segment at that moment. For the first The corresponding order of the first The waveform segment in the first The vibration amplitude at each moment, For the first The corresponding order of the first The average vibration amplitude of each waveform segment at all times. For the first The corresponding order of the first The waveform segment in the first The vibration amplitude at each moment, The time delay values ​​range from 0.001 to 0.01. For the first The corresponding order of the first The average vibration amplitude of each waveform segment at all times.

[0124] The molecule is a quantitative carrier of trend consistency. By calculating the sum of the products of the two waveform segments after removing the mean, it directly reflects the trend consistency of the vibration pattern. and Representing the first , Each waveform segment at time... The fluctuation component (and the corresponding time after a delay of τ0) deviates from its own mean. When two waveform segments have similar shapes (simultaneous peaks and simultaneous decays), the product of the two fluctuation components is positive, and the sum is a large positive number; when the shapes are opposite (one rises while the other falls), the product of the two fluctuation components is negative, and the sum is a small negative number; when the shapes are irregular (one has a clear period while the other is chaotic), the positive and negative products cancel each other out, and the result approaches 0. This design accurately captures the essential difference between the stable reproduction of persistent fault waveforms and the random fluctuations of occasional interference waveforms.

[0125] The denominator is the normalization factor that eliminates amplitude differences. As a normalization factor, the denominator transforms the "absolute correlation" of the numerator into a "relative coefficient" by taking the square root of the product of the wave energies of the two waveform segments. This process solves the problem that "amplitude differences do not represent morphological differences" in engine vibration. For example, the waveform amplitude of the same order may be generally greater under high load than under low load, but as long as the morphology is consistent (e.g., the peak position and wave rhythm are the same), a high similarity result can still be obtained through the scaling effect of the denominator. Ultimately, this enables... Strictly limited to Within the interval, ensure that similarity results under different orders and different loads are directly comparable.

[0126] because For each time delay, a value ranging from 0.001 to 0.01 can be calculated. All peak As the first The corresponding order of the first The waveform segment and the first The normalized cross-correlation coefficient of the i-th waveform segment is what makes the i-th waveform segment... The waveform segment and the first The most similar time alignment for each waveform segment.

[0127] Introducing a time delay is an engineering optimization to address minute fluctuations in engine speed. Since the duration of different cycles may vary slightly due to speed fluctuations, adjusting the time delay allows for precise alignment of waveform segments, avoiding misjudgments of morphological differences due to time axis misalignment and accurately reflecting inherent similarity. For example, when two waveform segments have the same actual shape but a time difference of 0.002 seconds, in... The waveform will show a peak at certain times. By delaying the entire waveform by 0.002 seconds, the interference of duration difference can be eliminated, and the focus can be placed on the similarity of the waveform segments themselves.

[0128] when When, it indicates the first The waveform segment and the first The more similar the shapes of the waveform segments, the more consistent their fluctuation trends and peak positions, and the more consistent the shapes of the two waveform segments under optimal time alignment, the better. The vibration modes corresponding to each order exhibit stable and repetitive characteristics in these two waveform segments, consistent with the "periodic recurrence" characteristic of persistent mechanical faults; when When the value is negative, it indicates that the first... The corresponding order of the first The waveform segment and the first The large morphological differences in the waveform segments indicate that the first... The vibration modes corresponding to each order are unstable in these two waveform segments, with irregular shapes or opposite trends. They are more likely to be caused by occasional disturbances such as unstable combustion or random impacts. This quantitative result provides an objective basis for subsequent calculation of morphological differences and assessment of overall stability.

[0129] S43: Determine the degree of morphological dispersion of all waveform segments corresponding to each order.

[0130] Calculating the morphological dispersion of all waveform segments corresponding to each order is a crucial step in distinguishing between mechanical faults and random interference, and is therefore essential. This is because the similarity performance of individual waveform segments reflects the morphological correlation between pairs of segments, while the morphological dispersion, through statistical analysis of the similarity of all waveform segments, can provide an overall assessment of the stable reproducibility of the vibration mode at that order.

[0131] For persistent mechanical faults, the vibration waveform should exhibit high consistency and low overall dispersion in continuous cycles; for sporadic interference, the waveform shape is chaotic and the dispersion is significantly higher. By quantifying the dispersion of the waveform, the local similarity of multiple waveform segments can be transformed into a global stability index, providing a decisive basis for judging the nature of vibration anomalies and effectively improving the accuracy of fault diagnosis.

[0132] Therefore, the process of determining the degree of dispersion based on the statistical characteristics of the morphological similarity of all waveform segments in this step is as follows: first, the similarity of each pair of waveform segments is converted into a negatively correlated morphological difference degree, and then the average morphological difference degree is obtained by calculating the average of the sum of squares of the morphological difference degrees of all combinations, which is used as the morphological dispersion degree of all waveform segments of this order.

[0133] Specifically, for any order, the first and second waveform segments among all waveform segments (step S42) are determined based on their similarity, and the morphological difference between the first and second waveform segments is negatively correlated with their similarity. The first and second waveform segments are combined, and the sum of the squares of the morphological differences corresponding to the combination is divided by the total number of combinations to obtain an average morphological difference. The average morphological difference is used as the degree of morphological dispersion of all waveform segments corresponding to that order.

[0134] For example, the first order in the real-time order spectrum The morphological dispersion of all waveform segments corresponding to each order is calculated based on the following formula:

[0135]

[0136] In this formula, The first in the real-time order spectrum The degree of dispersion of the order in M ​​consecutive vibration signals quantifies the order from a global perspective. The degree of morphological dispersion of all waveform segments corresponding to each order; the smaller the value, the more dispersed the order. The higher the reproducibility of the vibration mode corresponding to the order, the more stable it is. For the first The corresponding order of the first The waveform segment and the first The morphological similarity of each waveform segment.

[0137] In this formula, the numerator is the cumulative quantification of the total degree of dispersion, first through... The morphological similarity is transformed into morphological difference. Each pair of waveform segments is grouped together, and the sum of the squares of the differences across all waveform segment combinations is then taken as the square root. The squaring operation amplifies the impact of significant differences (such as large morphological deviations caused by occasional interference) while suppressing interference from minor fluctuations (such as slight vibration deviations during normal operation), ensuring that the calculation of the total dispersion more closely matches the judgment requirement of "stable reproduction of fault characteristics." The square root operation keeps the dimensions of the result consistent with the morphological difference, facilitating an intuitive understanding of the overall dispersion level.

[0138] The denominator is a normalization calibration, where the denominator is the total number of combinations of all waveform segments. Its function is to transform the total dispersion into an average level. This processing eliminates the influence of sample size differences. Regardless of whether there are 5 or 500 waveform segments, the final result reflects the average dispersion of any two waveform segments, making the dispersion under different orders and sample sizes directly comparable. For example, for the same total dispersion, the larger the sample size, the lower the average dispersion, and the better it reflects the characteristics of stable reproducibility.

[0139] In short, The smaller the value, the smaller the average morphological difference among all waveform segments of that order, and the stronger the reproducibility of the vibration mode in continuous cycles, which is consistent with the characteristics of persistent faults such as bearing pitting and gear tooth breakage; conversely, The larger the value, the more chaotic and irregular the waveform morphology, which is more likely to be caused by occasional interference such as unstable combustion or random external impacts. This provides a core indicator for subsequent assessment of morphological stability, transforming stable reproducibility from a qualitative description into directly comparable quantitative data.

[0140] S44: Determine the morphological stability of all waveform segments corresponding to each order.

[0141] Determining the morphological stability of all waveform segments corresponding to each order is a crucial step in transforming dispersion into a direct diagnostic indicator. Morphological dispersion reflects the magnitude of waveform differences, while morphological stability directly links these differences to "fault reproducibility" through a clear functional relationship. For operational monitoring personnel, stability indicators can more directly determine whether the vibration mode conforms to the characteristics of a persistent fault. Higher stability indicates stronger reproducibility of the vibration pattern in continuous cycles, and a higher reliability of the fault; conversely, lower stability suggests more likely sporadic interference. This quantification process transforms abstract morphological consistency into a directly applicable diagnostic criterion.

[0142] The process of determining the degree of morphological stability is as follows: using the degree of morphological dispersion as a negative indicator, the degree of morphological stability of all waveform segments corresponding to each order is quantified through a function (such as an exponential function) that can reflect the negative correlation between the two, so as to obtain a more intuitive indicator in which the larger the value, the more stable the stability.

[0143] For example, the The morphological stability of all waveform segments corresponding to each order satisfies the following relationship:

[0144]

[0145] In this formula, For the first The degree of morphological stability of all waveform segments corresponding to each order. It is a natural constant. The first in the real-time order spectrum The degree of morphological dispersion of all waveform segments corresponding to each order. It is an adjustment coefficient greater than 0, with an empirical value of 0.5, used to adapt to the differences in vibration characteristics at different orders. For example, for lower orders such as the first and second orders, which are related to the core components, the dispersion of their normal vibrations should be extremely low, and can be appropriately increased. (e.g., 0.6) to improve sensitivity to small discreteness; for higher orders, due to more complex vibration modes, the normal dispersion is slightly higher, which can be reduced. (e.g., 0.4) to avoid oversensitivity. A default value of 0.5 ensures cross-sectional comparability of stability levels at different orders, allowing diagnosticians to assess stability differences at each order under the same criteria.

[0146] In summary, using an exponential function to construct a negative correlation between dispersion and stability is based on the characteristic of "nonlinear change in fault characteristics from stable to unstable." When the dispersion... When approaching 0 (waveform height is consistent), Approaching 1, it intuitively reflects the characteristics of high stability, which conforms to the vibration law of persistent faults such as bearing pitting and gear tooth breakage; when When the waveform difference increases, that is, when the waveform difference becomes larger, It exhibits exponential decay, rapidly approaching zero, accurately capturing the "sudden drop in stability" phenomenon caused by occasional disturbances. This nonlinear transformation better reflects the characteristics of actual fault development; small changes in morphological dispersion have little impact on stability, while a significant increase in morphological dispersion leads to a sharp decline in stability, making it easier to distinguish between acceptable normal fluctuations and abnormal morphological dispersion that requires vigilance.

[0147] at last, The range of values ​​for strictly falls within Within the range, a larger value indicates a higher degree of morphological stability: when When this occurs, it indicates that the vibration waveform of this order is highly consistent in continuous cycles, which is a typical characteristic of the periodic recurrence of mechanical faults and requires close attention; when If the waveform morphology is significantly different, it indicates that the problem is more likely due to occasional interference such as unstable combustion or external impact, and the priority of attention should be reduced. This intuitive quantitative result provides a clear decision-making basis for fault diagnosis, transforming the key characteristic of stable reproducibility from an abstract description into a quantifiable and comparable indicator.

[0148] S5: Determine the comprehensive failure risk index for each order by combining energy deviation and morphological stability.

[0149] Calculating the comprehensive fault risk index at each order is a crucial step in achieving fault decision-making from feature quantification. Energy deviation only reflects the degree of abnormality in vibration energy but cannot distinguish whether its source is a real fault or an occasional disturbance; morphological stability only reflects the reproducibility of vibration modes but cannot indicate the severity of the fault. Both have limitations when used alone. Relying solely on energy deviation may misjudge occasional disturbances such as combustion instability as faults; relying solely on morphological stability makes it difficult to quantify the actual impact of the fault. By integrating both, a synergistic assessment of severity and reliability can be achieved, ensuring that only when energy is significantly abnormal and morphological stability is reproducible is the fault deemed high-risk. This effectively improves the accuracy and reliability of fault diagnosis, providing accurate decision-making basis for engine condition monitoring.

[0150] Therefore, for any order, this step takes the morphological stability of all waveform segments corresponding to that order as input and generates a gating weight through a nonlinear activation function transformation; the energy deviation of that order and the gating weight are weighted and fused to obtain the comprehensive fault risk index of that order.

[0151] Specifically, adopt The function, as a nonlinear activation function, is used to achieve fusion based on the following formula to obtain the comprehensive fault risk index for each order:

[0152]

[0153] In this formula, It is the first The comprehensive failure risk index of each order, It is the first The order of energy deviation represents the severity of the fault. It is a natural constant. It is the first The degree of morphological stability of all waveform segments corresponding to each order. This is the activation threshold for morphological stability, set to 0.8. This value is determined based on statistical analysis of engine vibration characteristics. When the waveform morphology is highly consistent in continuous cycles, it conforms to the characteristics of periodic recurrence of mechanical faults, and at this time the gating weight begins to increase significantly; when At that time, the morphological stability is insufficient, the gating weight is low, and the interference of unstable features is effectively filtered out.

[0154] The core of the formula is to generate gating weights using the Sigmoid function. The latter half of the formula is a typical Sigmoid function; only when the morphological stability is sufficiently high will the risk indicated by the energy deviation be recognized. This design strictly adheres to the diagnostic logic that real faults must simultaneously satisfy both energy anomalies and morphological stability. Mechanical faults (such as bearing pitting) will simultaneously manifest as large energy deviations. High and high degree of morphological stability At this point, the gating weight approaches 1. Approximately equal to This accurately reflects high risks; while occasional disturbances (such as unstable combustion) may lead to large energy deviations, but their morphological stability is low. Low, the gating weight approaches 0. It is significantly suppressed, avoiding misjudgment.

[0155] at last, The magnitude of the value directly reflects the fault risk level corresponding to that order: when When the value is significantly higher (e.g., much higher than the historical normal threshold), it indicates that there is both a significant energy anomaly and high morphological stability at this order, which is highly likely to be a mechanical structural fault and requires immediate attention; when When the value is low, it may be due to a small energy deviation (normal state) or insufficient morphological stability (occasional interference), and the risk is low.

[0156] In summary, this quantitative result simplifies complex multi-feature analysis into intuitive risk indicators. It retains the characterization of fault severity by energy deviation and filters out unreliable interference signals by morphological stability, enabling the engine condition monitoring system to focus more accurately on real mechanical faults and significantly improving the efficiency and reliability of diagnostic decisions.

[0157] S6: Monitor the engine's operating status based on the comprehensive failure risk index at each level.

[0158] This step is the final output of the entire monitoring process, which achieves real-time monitoring of the engine's operating status through the comprehensive fault risk index of all orders in the real-time order spectrum.

[0159] Specifically, it includes:

[0160] S61: By analyzing the relationship between order and engine motion characteristics.

[0161] Each order corresponds to a specific vibration mode. The essence of the order is the "ratio of vibration frequency to crankshaft speed", which directly corresponds to the periodic vibration mode of specific engine components. For lower orders (such as the 1st, 2nd and 3rd orders), they are mostly related to the core motion system. For example, the 1st order corresponds to the vibration mode generated by rotating components such as crankshaft / flywheel every 1 revolution; the 2nd order corresponds to the vibration mode generated by reciprocating components such as piston / connecting rod every 2 revolutions; and the 3rd order corresponds to the vibration mode generated by gear meshing every 3 revolutions, and the vibration mode of multi-cylinder combustion pulse superposition.

[0162] Therefore, vibrations of orders 1-3 are the core indicators for engine vibration analysis. Their corresponding vibration excitation sources are directly related to fundamental moving components such as the rotating system, reciprocating system, and core accessories. Early faults in these components (such as crankshaft imbalance, piston sticking, and water pump wear) can be clearly presented through anomalies in low-order vibrations. This not only minimizes signal interference and provides clear characteristics, facilitating rapid fault location, but also covers the vast majority of basic fault types. While higher-order vibrations can reflect faults, they are often related to complex vibration superpositions such as high-frequency coupling of multi-cylinder combustion pulses and high-order resonance of components. Interpretation of these higher-order vibrations requires combining them with low-order anomalies, and they are primarily used as supplementary auxiliary analyses. Therefore, focusing on orders 1-3 allows for efficient capture of early fault signals while achieving a balance between practicality and complexity.

[0163] Therefore, the correlation between each order (corresponding vibration mode) and specific components of the engine is established as follows:

[0164] First order: This corresponds to rotating components, including the crankshaft and flywheel. When the first order vibration mode is abnormal, it usually means that the engine's rotating system is unbalanced (such as a bent crankshaft or an unbalanced flywheel), indicating that there is eccentric vibration in the core rotating components during operation.

[0165] Second-order: This corresponds to reciprocating components, including pistons, fuel injectors, etc. When the second-order vibration mode is abnormal, it usually means that the reciprocating system is stuck or the combustion is uneven (such as excessive clearance between the piston and cylinder liner, abnormal fuel injectors), indicating that there is impact vibration in the reciprocating components or a decrease in combustion stability during operation.

[0166] Third-order vibration: This corresponds to other accessories, including water pumps, gears, etc. When the third-order vibration mode is abnormal, it usually means that the accessories are unbalanced or poorly meshed (such as water pump bearing wear or gear tooth surface damage), indicating that there is high-frequency abnormal friction in the accessory system during operation.

[0167] S62: Determine the normal range of the comprehensive failure risk index for each order.

[0168] Beforehand, under the full operating conditions of the engine (without faults) including idling, partial load, and full load, at least 1000 engine cycles (10 times that in step S1) of vibration signals and crankshaft speed signals are collected. By dividing the engine into segments of 100 engine cycles, 10 segments of vibration signals and 10 segments of crankshaft speed signals can be obtained.

[0169] For each vibration signal segment, the real-time order spectrum of that vibration signal segment and the comprehensive fault risk index of each order in the real-time order spectrum of that vibration signal segment are calculated according to the method of steps S1-S6. The orders in the real-time order spectrum of these 10 vibration signals are the same.

[0170] For each order, calculate the mean and standard deviation of the comprehensive fault risk index for that order across 10 vibration signal segments. Based on the statistical principle of three standard deviations, determine the normal range of the comprehensive fault risk index for that order using the mean ± three standard deviations. For example: order The average comprehensive fault risk index of these 10 vibration signals is The standard deviation is Then its normal range is Using this method, the normal range of the comprehensive failure risk index for each order in the real-time order spectrum of the engine's vibration signal under healthy conditions can be obtained.

[0171] S63: Monitor the engine's operating status based on the comparison results of the comprehensive failure risk index at each level with the normal range.

[0172] The comprehensive fault risk index of each order in the real-time order spectrum acquired at the current moment is compared with its normal range. If the comprehensive fault risk index of a certain order exceeds the normal range, it is determined that the operating state of the specific engine component associated with the vibration mode of that order is abnormal; if the comprehensive fault risk index of a certain order does not exceed the normal range, it is determined that the operating state of the specific engine component associated with the vibration mode of that order is not abnormal. For example, if the comprehensive fault risk index of order 1 exceeds the normal range, it is determined that the operating state of the rotating parts of the engine is abnormal.

[0173] like Figure 8As shown, the final output of the entire process of this invention demonstrates the comprehensive failure risk index obtained after integrating energy deviation and morphological stability in step S5. Since the second-order vibration exhibits both high energy deviation and high morphological stability, its final calculated comprehensive failure risk index significantly exceeds the preset upper limit of the normal range, while the comprehensive failure risk indices of other orders are within the normal range. This result clearly indicates that the reciprocating components related to the second-order vibration exhibit abnormal operating conditions, verifying that the method of this invention can accurately and reliably monitor the engine's operating status.

[0174] The present invention also provides an engine operating status monitoring system, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the operations as described in steps S1-S6, thereby achieving accurate monitoring of the engine's operating status.

[0175] In summary, this invention overcomes the shortcomings of existing technologies in processing variable speed engine signals through a complete and logically rigorous technical solution, and provides a more accurate, intelligent and reliable new paradigm for condition monitoring.

[0176] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring engine operating status, characterized in that, include: Real-time acquisition of engine vibration signals and engine crankshaft speed signals; Using the crankshaft speed signal as a reference, the vibration signal is converted from the time domain to the angular domain to obtain the vibration signal in the angular domain; The vibration signal in the angular domain is subjected to order spectrum analysis to generate a real-time order spectrum with each order as the abscissa and the vibration amplitude of each order as the ordinate. The real-time order spectrum is compared with the pre-acquired baseline order spectrum to determine the energy deviations of each order in the real-time order spectrum, including: Based on the engine's real-time load conditions, obtain all reference order spectra under those load conditions. For any order in the real-time order spectrum, use the mean and standard deviation of the vibration amplitude of that order in all reference order spectra as the reference amplitude and fluctuation value of that order, respectively. Calculate the absolute difference between the vibration amplitude and the reference amplitude, as well as the cumulative sum of the reference amplitude and the fluctuation value of the reference amplitude. Determine the normalized deviation rate of the vibration amplitude of that order by the ratio of the vibration amplitude of that order to the reference amplitude. Determine the energy gain factor of that order by multiplying the normalized deviation rate of the vibration amplitude and the energy gain factor of that order. For any order in the real-time order spectrum, extract all waveform segments corresponding to that order across multiple consecutive engine cycles, and use these as the waveform segments corresponding to that order. Calculate the similarity between any two waveform segments, including: for any two waveform segments corresponding to any order, denoted as the first waveform segment and the second waveform segment, respectively; perform mean-removal processing on the first and second waveform segments to eliminate DC offset and retain only waveform morphology features; calculate the normalized cross-correlation coefficients of the mean-removed first and second waveform segments under different time delays; and use the peak value of the normalized cross-correlation coefficients as the similarity between the first and second waveform segments. The method assesses the morphological dispersion of all waveform segments corresponding to a given order based on the statistical characteristics of the similarity between any two waveform segments, and uses the morphological dispersion as a negative indicator to quantify the morphological stability of all waveform segments corresponding to that order. This includes: determining the morphological difference between the first and second waveform segments based on their similarity, with the morphological difference being negatively correlated with similarity; taking the first and second waveform segments as a combination, dividing the sum of the squares of the morphological differences corresponding to the combination by the total number of combinations to obtain an average morphological difference; and using the average morphological difference as the morphological dispersion of all waveform segments corresponding to that order. The comprehensive fault risk index for each order is determined by integrating the energy deviation of each order in the real-time order spectrum and the morphological stability of all waveform segments corresponding to each order. This includes: for any order, taking the morphological stability of all waveform segments corresponding to that order as input, and generating a gating weight through a nonlinear activation function; weighting and fusing the energy deviation of that order with the gating weight to obtain the comprehensive fault risk index for that order; and monitoring the engine's operating status based on the comprehensive fault risk index for each order.

2. The engine operating status monitoring method according to claim 1, characterized in that, The vibration signal in the angular domain is determined based on the following method: The crankshaft speed signal is integrated over time to calculate the crankshaft rotation angle at each moment. Based on the time synchronization principle of the vibration signal and the crankshaft speed signal, the rotation angle and vibration amplitude of the vibration signal at each moment are determined. The vibration signal is divided into multiple periodic vibration signals with a rotation angle of 360 degrees as one period. Using the rotation angle from 0 to 360 degrees as the horizontal axis and the average vibration amplitude corresponding to each rotation angle in all periodic vibration signals as the vertical axis, the vibration signal in the angle domain is obtained.

3. The engine operating status monitoring method according to claim 1, characterized in that, The real-time order spectrum is determined based on the following methods: Perform a Fourier transform on the vibration signal in the angular domain to decompose it into a series of vibration modes with specific periodic characteristics. Each vibration mode corresponds to the vibration characteristics of a specific engine component. The number of repetitions of each vibration mode within every 360 degrees of crankshaft rotation is taken as the order of the vibration mode, and each order is used to characterize each vibration mode. The vibration amplitude of the vibration mode characterized by each order is taken as the vibration amplitude of that order. A real-time order spectrum is generated by plotting each order on the x-axis and the vibration amplitude of each order on the y-axis.

4. The engine operating status monitoring method according to claim 1, characterized in that, The baseline order spectrum is determined based on the following method: Operating a healthy engine under multiple different load conditions; Under each load condition, multiple vibration signals and crankshaft speed signals are collected as reference vibration signals and reference crankshaft speed signals, respectively. A reference order spectrum is determined based on each reference vibration signal and the corresponding reference crankshaft speed signal, and multiple reference order spectra are obtained under this load condition. The method for determining each reference order spectrum is the same as the method for determining the real-time order spectrum.

5. The engine operating status monitoring method according to claim 1, characterized in that, The engine's operating status is monitored based on a comprehensive failure risk index at various levels, including: The correlation between the vibration modes corresponding to each order and specific components of the engine, as well as the normal range of the comprehensive failure risk index for each order, are obtained in advance. If the comprehensive failure risk index of a certain order exceeds the normal range of the comprehensive failure risk index of that order, it is determined that the operating status of the specific engine component associated with the vibration mode corresponding to that order is abnormal. If the comprehensive failure risk index of a certain order does not exceed the normal range of the comprehensive failure risk index of that order, it is determined that the operating status of the specific engine component associated with the vibration mode corresponding to that order is not abnormal.

6. An engine operating status monitoring system, characterized in that, The engine operating status monitoring system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the engine operating status monitoring method as described in any one of claims 1-5.

Images