Engine fault signal feature extraction and diagnosis method and device and storage medium

Abstract

The application discloses an engine fault signal feature extraction and diagnosis method and device and a storage medium, and belongs to the technical field of mechanical fault diagnosis and signal processing. The application introduces a Newton-Raphson optimization algorithm NRBO to jointly and adaptively optimize modal number, bandwidth and filter bank parameters of FMD, combines relevant kurtosis and modal correlation coefficients to screen fault sensitive modes, and then constructs a fault feature vector through wavelet threshold denoising and multi-domain feature extraction. Finally, the engine fault type recognition and working condition discrimination are realized by inputting the classifier, so that the accuracy and robustness of fault diagnosis under complex working conditions are improved.

Description

Technical Field

[0001] This invention relates to the field of mechanical fault diagnosis and signal processing technology. Specifically, it relates to a method, system and storage medium for extracting and diagnosing engine fault signal features, which is applicable to the condition monitoring and intelligent diagnosis of fuel engine power units under multiple operating conditions and multiple fault modes. Background Technology

[0002] As the core power unit of vehicles and construction machinery, the engine's operating status directly affects the overall performance and safety of the machine. During actual operation, the engine is affected by multiple sources of excitation, including combustion impact, reciprocating inertial forces, rotational imbalance, valve train collisions, and coupled vibrations from accessory systems. Its vibration signals exhibit significant characteristics of strong non-stationarity, high noise, and multi-component coupling. Simultaneously, factors such as uneven road surfaces, tire-ground interaction, chassis resonance, and environmental noise further reduce the signal-to-noise ratio of any single sensor's signal, making traditional filtering methods based on the linear time-invariant assumption ineffective in highlighting fault-related components.

[0003] To extract features helpful for fault identification from complex vibration signals, researchers have proposed several adaptive decomposition methods. Empirical Mode Decomposition (EMD) obtains eigenmode functions through spline interpolation at local extrema points, exhibiting some adaptability to non-stationary signals. However, it is prone to mode aliasing, endpoint effects, and high sensitivity to noise when processing broadband engine vibration signals, affecting the physical interpretability of the decomposition results. Variational Mode Decomposition (VMD) decomposes the signal into several sub-signals with finite bandwidth using a variational framework, mitigating some mode aliasing. However, its decomposition results are highly sensitive to hyperparameters such as the number of modes, penalty factor, and center frequency; slight inappropriate parameter settings can lead to decomposition instability, limiting its application in engine scenarios with frequent changes in speed and load.

[0004] Feature Mode Decomposition (FMD) is a signal decomposition method based on adaptive filters. This method designs a set of Finite Impulse Response (FIR) filters, convolves them with the input signal to generate a series of modal components, and optimizes the filter bank parameters using indices such as correlation kurtosis to obtain the characteristic modes highlighting impulse or modulation components. Compared to EMD and VMD, FMD exhibits better stability and robustness in extracting features from impact-type faults and rotating machinery faults. However, the decomposition effect of FMD largely depends on multiple parameters such as the number of modes, filter bandwidth, and filter bank parameters. Currently, these parameters are generally set empirically or ergonomically, resulting in low parameter optimization efficiency, susceptibility to local optima, and difficulty in adapting to complex operating conditions.

[0005] Meanwhile, although many current engine fault diagnosis methods have introduced intelligent algorithms such as support vector machines and convolutional neural networks, the front-end signal preprocessing and the back-end classification model are often designed separately: the front-end noise reduction and decomposition methods have not been optimized for fault feature enhancement, resulting in insufficient sensitivity of the extracted features to faults, and making it difficult to further improve the diagnostic accuracy and robustness.

[0006] Therefore, there is an urgent need for a fault diagnosis method that can adaptively optimize FMD parameters and highlight engine fault components while suppressing background noise, so as to improve the accuracy and stability of engine fault identification under complex operating conditions. Summary of the Invention

[0007] The main objective of this invention is to address the problems in existing engine fault diagnosis methods, such as the difficulty in adaptively determining FMD parameters, the susceptibility of decomposition results to changes in operating conditions and noise, insufficient prominence of fault features, and inadequate diagnostic robustness. This invention proposes an engine fault diagnosis and feature extraction method based on Newton-Raphson optimized eigenmode decomposition. By introducing the Newton-Raphson Based Optimization (NRBO) algorithm, the number of modes, bandwidth, and filter bank parameters of the FMD are jointly and adaptively optimized. Fault-sensitive modes are screened using correlation kurtosis and modal correlation coefficients. Then, fault feature vectors are constructed through wavelet threshold denoising and multi-domain feature extraction. Finally, these vectors are input into a classifier to achieve engine fault type identification and operating condition discrimination, thereby improving the accuracy and robustness of fault diagnosis under complex operating conditions.

[0008] To achieve the above objectives, in a first aspect, this application provides a method for extracting and diagnosing engine fault signal features, comprising the following steps:

[0009] S1: Acquire the engine's vibration signal to be diagnosed, and record it as the original signal;

[0010] S2: Preprocess the original signal to obtain the preprocessed vibration signal x(n);

[0011] S3: Construct an adaptive parameter optimization space for Eigenmode Decomposition (FMD) to address the issue that FMD decomposition performance relies on manual experience. The parameters include at least the number of modes K and the center frequency f of each modal subband. k Bandwidth BW k and filter length L h ;

[0012] S4: Establish the mapping relationship between FMD parameters and modal components, and construct the parameter vector. As an optimization decision variable;

[0013] S5: Construct a multi-objective integrated fitness function F(θ) with the goal of minimization as an indicator to evaluate the quality of modal component decomposition and provide guidance for parameter optimization;

[0014] S6: The Newton-Raphson optimization algorithm (NRBO) is used to iteratively update the parameter vector θ until the termination condition is met, and the optimal parameter θ that minimizes the fitness function F(θ) is output. * In order to obtain the FMD parameter combination that can achieve the best decomposition quality;

[0015] S7: Based on the optimal parameter θ * The preprocessed vibration signal x(n) is subjected to final FMD decomposition to obtain K optimal modal components y. k (n);

[0016] S8: Based on the correlation kurtosis and modal correlation coefficient, from the optimal modal component y k Select the set of fault-sensitive modes S from (n) * This allows for the removal of background noise components unrelated to the fault and the elimination of redundant elements, thus making the fault-sensitive mode set S * It contains key fault information; S9: For the fault-sensitive mode set S * The modal components in the signal are subjected to wavelet threshold denoising (to eliminate residual noise interference) and reconstructed to obtain the fault component enhanced signal x with improved signal-to-noise ratio. f (n); S10: Extract time-domain, frequency-domain, and / or time-frequency-domain features from the enhanced fault component signal, construct a feature vector, input it into the fault classifier, and output the engine fault diagnosis result.

[0017] A further technical solution, wherein step S1 specifically includes: setting at least one vibration sensor at at least one location in the engine block, cylinder head, crankcase or valve cover, and collecting real-time vibration data of the engine in operation as the vibration signal to be diagnosed.

[0018] In a further technical solution, the preprocessing in step S2 includes at least DC removal, bandpass filtering, and normalization.

[0019] The bandpass filter band covers the engine's main operating frequency and its harmonics / order components, as well as the fault characteristic frequency band; preferably, the bandpass filter band is 50 Hz to 5 kHz or an effective frequency band determined according to the sampling frequency.

[0020] In step S3, the number of modes K is a positive integer and K∈[Kmin,Kmax]; the filter length L h The elements are in the discrete set of values, preferably {64,128,256}.

[0021] In a further technical solution, the formula for the multi-objective integrated fitness function in step S5 is as follows:

[0022] (Formula 1)

[0023] Among them, E rec The reconstruction error term, E, characterizes the signal integrity. orth For the orthogonality constraint term characterizing modal independence, E spar The sparsity penalty term is used to characterize the significance of fault impact, and w1, w2, and w3 are weighting coefficients.

[0024] The weighting coefficients satisfy w1>w2 and w1>w3 to ensure that the decomposition results have high reconstruction accuracy.

[0025] In a further technical solution, the termination condition in step S6 includes reaching the maximum number of iterations or the change in the optimal fitness in multiple consecutive iterations being less than a preset threshold.

[0026] The optimization algorithm used is the Newton-Raphson optimization algorithm (NRBO), which enhances global search capabilities through adaptive search rules and trap avoidance operators.

[0027] In a further technical solution, step S8 specifically includes: Calculate the correlation kurtosis value of each optimal mode component, and select candidate modes based on a preset threshold; Calculate the modal correlation coefficients between candidate modes, and remove redundant modes based on the correlation coefficient thresholds to obtain the fault-sensitive mode set.

[0028] The relevant kurtosis value is based on a preprocessed vibration signal.

[0029] x(n) and modal components y k The cross-correlation sequence between (n) is calculated and used as a threshold ρ. th Perform filtering; the threshold ρ th Satisfying ρ th =μ+λσ, where μ and σ are the mean and standard deviation of the correlation kurtosis, respectively, and λ is an empirical coefficient, λ∈[0.5,1.5];

[0030] The modal correlation coefficient ρ i,j When |ρ i,j When |>ρ0, determine the mode y i (n) and y j (n) Redundancy; and retain the mode with larger decorrelation kurtosis as the representative mode, ρ0 is the redundancy judgment threshold, ρ0∈[0.8,0.95].

[0031] In a further technical solution, the wavelet threshold denoising in step S9 adopts a soft threshold or hard threshold rule, preferably using the db8 wavelet basis, and the number of decomposition layers is 4 to 6.

[0032] The features in step S10 include at least one of the following: root mean square, peak value, kurtosis, spectral kurtosis, energy ratio, spectral entropy, or wavelet packet energy.

[0033] The classifier is a support vector machine (SVM), which uses a radial basis kernel function and employs a one-to-one or one-to-many strategy for multi-class discrimination.

[0034] Secondly, this application provides an apparatus for extracting and diagnosing engine fault signal features. The apparatus includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program, when executed by the processor, implements one or more of the steps of the engine fault signal feature extraction and diagnosis method described in the application.

[0035] Thirdly, this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of one or more of the engine fault signal feature extraction and diagnosis methods described in this application.

[0036] Beneficial effects

[0037] 1. The method of this invention can adaptively optimize FMD parameters, avoiding manual parameter tuning. This invention utilizes the Newton-Raphson Optimization Algorithm (NRBO), which includes the Newton-Raphson Search Rule (NRSR) and the Trap Avoidance Operator (TAO), to jointly and adaptively optimize key parameters of FMD such as the number of modes, center frequency, bandwidth, and filter length. Compared with traditional methods, this invention uses NRSR to improve the convergence speed of parameter search and uses TAO to enhance the ability to escape local extrema, completely overcoming the problem that the FMD decomposition effect is heavily dependent on manual experience settings, and achieving consistency and stability of signal decomposition quality under multiple engine speeds and multiple fault conditions.

[0038] 2. The method of this invention can significantly reduce mode aliasing and highlight fault-sensitive modes. This invention constructs a mode based on reconstruction error (E... rec Orthogonality constraint (E) orth ) and sparsity penalty (E spar The multi-objective integrated fitness function is composed of a reconstruction error term that ensures high-fidelity signal restoration, an orthogonality constraint term that effectively suppresses aliasing between different modes, and a sparsity penalty term that enhances the saliency of fault impact features. This multi-dimensional analysis enables FMD to decompose modal components with good frequency band separation and energy concentration, significantly improving the signal-to-noise ratio of fault features.

[0039] 3. The integrated design of denoising and feature extraction in this invention improves diagnostic accuracy. This invention establishes a dual screening mechanism consisting of correlation kurtosis indices and modal correlation coefficients, combined with wavelet threshold denoising technology, to achieve a "selection of the best among the best" feature extraction process. First, correlation kurtosis is used to accurately locate modes rich in fault impact; then, modal correlation coefficients are used to remove redundant interference; finally, denoising processing makes the feature vector input to the classifier purer. Compared with EMD-wavelet, VMD-wavelet, and traditional FMD-wavelet methods, this invention has significant advantages in feature signal-to-noise ratio, root mean square error, and final fault identification accuracy.

[0040] 4. The NRBO-FMD algorithm of this invention can run efficiently on general-purpose industrial control computers or embedded platforms, and the output feature vectors are standardized and can be flexibly coupled with pre-trained intelligent diagnostic modules such as support vector machines (SVM) and convolutional neural networks. This invention not only reduces the computational load of online monitoring, but also facilitates rapid integration and application in engine test benches, on-board diagnostic systems (OBD), and remote monitoring platforms. Attached Figure Description

[0041] To more clearly illustrate the technical solution of the present invention, the specific embodiments of the present invention will be further described below with reference to the accompanying drawings, but the present invention is not limited to the drawings shown.

[0042] Figure 1 This is a flowchart illustrating the engine fault diagnosis and feature extraction method in an embodiment of the present invention.

[0043] Figure 2 The following is a comparison of the time-domain waveforms of the original vibration signals collected by the engine under different typical operating conditions in the embodiments of the present invention, wherein: a is the time-domain waveform of the vibration signal under normal operating conditions; b is the time-domain waveform of the vibration signal under crankshaft bearing wear conditions; c is the time-domain waveform of the vibration signal under intake valve leakage conditions; and d is the time-domain waveform of the vibration signal under piston knocking conditions.

[0044] Figure 3(a) is a time-domain diagram of the EMD-wavelet method after noise reduction in the case of crankshaft bearing wear in an embodiment of the present invention.

[0045] Figure 3(b) is a time-domain diagram of the VMD-wavelet method after noise reduction under crankshaft bearing wear conditions in an embodiment of the present invention.

[0046] Figure 3(c) is a time-domain diagram of the traditional FMD-wavelet method after noise reduction under crankshaft bearing wear conditions in an embodiment of the present invention.

[0047] Figure 3(d) is a time-domain diagram of the method of the present invention after noise reduction under crankshaft bearing wear conditions in an embodiment of the present invention;

[0048] Figure 4 Line graph showing the signal-to-noise ratio (SNR) and root mean square error (RMSE) of the method of the present invention and the comparative method on engine test data;

[0049] Figure 5(a) shows the time-frequency distribution of the original vibration signal in the case of crankshaft bearing wear.

[0050] Figure 5(b) shows the STFT time-frequency distribution of the signal after EMD-wavelet method processing under crankshaft bearing wear conditions.

[0051] Figure 5(c) shows the STFT time-frequency distribution of the signal after processing by the VMD-wavelet method under crankshaft bearing wear conditions;

[0052] Figure 5(d) shows the STFT time-frequency distribution of the signal after processing by the traditional FMD-wavelet method under crankshaft bearing wear conditions;

[0053] Figure 5(e) shows the STFT time-frequency distribution of the signal processed by the method of the present invention under crankshaft bearing wear conditions. Detailed Implementation

[0054] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0055] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or apparatus.

[0056] Example 1

[0057] like Figure 1 As shown in the figure, this application provides a method for extracting and diagnosing engine fault signal features, including the following steps:

[0058] S1: Acquire the engine's vibration signal to be diagnosed, and record it as the original signal;

[0059] S2: The original signal is preprocessed to obtain the preprocessed vibration signal x(n);

[0060] S3: Construct an adaptive parameter optimization space for Eigenmode Decomposition (FMD) to address the issue that FMD decomposition performance relies on manual experience. The parameters include at least the number of modes K and the center frequency f of each modal subband. k Bandwidth BW k and filter length L h ;

[0061] S4: Establish the mapping relationship between FMD parameters and modal components, and construct the parameter vector. As an optimization decision variable;

[0062] S5: Construct a multi-objective integrated fitness function F(θ) with the goal of minimization as an indicator to evaluate the quality of modal component decomposition and provide guidance for parameter optimization;

[0063] S6: The Newton-Raphson optimization algorithm (NRBO) is used to iteratively update the parameter vector θ until the termination condition is met, and the optimal parameter θ that minimizes the fitness function F(θ) is output. * In order to obtain the FMD parameter combination that can achieve the best decomposition quality;

[0064] S7: Based on the optimal parameter θ * The preprocessed vibration signal x(n) is subjected to the final FMD decomposition to obtain K optimal modal components y. k (n);

[0065] S8: Based on the correlation kurtosis and modal correlation coefficient, from the optimal modal component y k Select the set of fault-sensitive modes S from (n) * This allows for the removal of background noise components unrelated to the fault and the elimination of redundant elements, thus making the fault-sensitive mode set S * It contains key fault information; S9: For the fault-sensitive mode set S * The modal components in the signal are subjected to wavelet threshold denoising (to eliminate residual noise interference) and reconstructed to obtain the fault component enhanced signal x with improved signal-to-noise ratio. f (n); S10: Extract time-domain, frequency-domain, and / or time-frequency-domain features from the enhanced fault component signal, construct a feature vector, input it into the fault classifier, and output the engine fault diagnosis result.

[0066] Step S1 specifically includes: setting at least one vibration sensor at at least one location in the engine block, cylinder head, crankcase or valve cover, and collecting real-time vibration data of the engine in operation as the vibration signal to be diagnosed.

[0067] The preprocessing in step S2 includes at least DC removal, bandpass filtering, and normalization.

[0068] The bandpass filter band covers the engine's main operating frequency and its harmonics / order components, as well as the fault characteristic frequency band; preferably, the bandpass filter band is 50 Hz to 5 kHz or an effective frequency band determined according to the sampling frequency.

[0069] In step S3, the modality number K is a positive integer and K∈[K]. min ,K max ]; Filter length L h The elements are in the discrete set of values, preferably {64,128,256}.

[0070] In step S5, the formula for the multi-objective integrated fitness function is as follows:

[0071] (Formula 1)

[0072] Among them, E rec The reconstruction error term, E, characterizes the signal integrity. orth For the orthogonality constraint term characterizing modal independence, E spar The sparsity penalty term characterizes the significance of fault impact, and w1, w2, and w3 are weighting coefficients. Among them, the reconstruction error term ensures signal integrity, the orthogonality constraint term suppresses mode aliasing, the sparsity penalty term can enhance the significance of fault impact, and the weighting coefficients ensure decomposition accuracy and mode separability.

[0073] The weighting coefficients satisfy w1>w2 and w1>w3 to ensure that the decomposition results have high reconstruction accuracy.

[0074] Before iteratively updating, step S6 requires setting the population size NP and the maximum number of iterations MaxIter for the Newton-Raphson optimization algorithm (NRBO). The termination conditions include reaching the maximum number of iterations or the change in the optimal fitness in multiple consecutive iterations being less than a preset threshold.

[0075] The optimization algorithm used is the Newton-Raphson optimization algorithm (NRBO), which enhances global search capabilities through adaptive search rules and trap avoidance operators.

[0076] Step S8 specifically includes: Calculate the correlation kurtosis value of each optimal mode component, and select candidate modes based on a preset threshold; Calculate the modal correlation coefficients between candidate modes, and remove redundant modes based on the correlation coefficient thresholds to obtain the fault-sensitive mode set.

[0077] The relevant kurtosis value is based on the preprocessed vibration signal x(n) and modal components y k The cross-correlation sequence between (n) is calculated and used as a threshold ρ. th Perform filtering; the threshold ρ th Satisfying ρ th =μ+λσ, where μ and σ are the mean and standard deviation of the correlation kurtosis, respectively, and λ is an empirical coefficient, λ∈[0.5,1.5];

[0078] The modal correlation coefficient ρ i,j When |ρ i,j When |>ρ0, determine the mode y i (n) and y j (n) Redundancy; and retain the mode with larger decorrelation kurtosis as the representative mode, ρ0 is the redundancy judgment threshold, ρ0∈[0.8,0.95].

[0079] The wavelet threshold denoising in step S9 adopts either soft or hard threshold rules, preferably using the db8 wavelet basis, and the number of decomposition layers is 4 to 6.

[0080] The features in step S10 include at least one of the following: root mean square, peak value, kurtosis, spectral kurtosis, energy ratio, spectral entropy, or wavelet packet energy.

[0081] The classifier is a support vector machine (SVM), which uses a radial basis kernel function and employs a one-to-one or one-to-many strategy for multi-class discrimination.

[0082] This embodiment verifies the above method on an engine test bench. The test bench includes the engine body, a dynamometer / loading device, a speed and load controller, and a data acquisition system. To ensure the representativeness of the vibration data, vibration sensors can be placed at at least one location, such as the engine block, cylinder head, crankcase, or valve cover. In this embodiment, a piezoelectric accelerometer is placed near the main bearing of the cylinder block and fixed with bolts or strong adhesive to obtain a stable and reliable vibration signal. With the sensor installation position and data acquisition link parameters remaining unchanged, the engine is operated under different operating / fault conditions at different times, and vibration data for each condition is collected.

[0083] The engine was operated under normal operating conditions and at least one typical fault condition, and vibration data were collected. To verify the effectiveness of the method of this invention, this embodiment selected four typical operating conditions for experimental data collection: normal operating condition, crankshaft bearing wear condition, intake valve leakage condition, and piston knocking condition; the time-domain waveforms of the vibration signals collected under each condition are shown below. Figure 2 As shown in the figures, where: a) is the time-domain waveform of the vibration signal under normal operating conditions; b) is the time-domain waveform of the vibration signal under crankshaft bearing wear conditions; c) is the time-domain waveform of the vibration signal under intake valve leakage conditions; and d) is the time-domain waveform of the vibration signal under piston knocking conditions. It can be seen from the figures that the amplitude fluctuations and impact patterns of the original vibration signals under these four different fault conditions differ, but they also contain significant background vibration and noise, causing the fault impact and modulation characteristics to be easily obscured. This phenomenon indicates that relying solely on the original waveform is insufficient to reliably extract discriminative features, necessitating adaptive decomposition and enhancement processing.

[0084] Each fault condition can be achieved by replacing or processing parts to change the fit clearance, sealing condition or impact strength, or by using an equivalent fault simulation device on a test bench.

[0085] Regarding the acquisition parameters, the sampling frequency is set to fs (Hz), and the acquisition duration for each segment is T (s). To ensure coverage of the engine's main operating frequencies and fault characteristic frequency bands, anti-aliasing filtering can be set in the acquisition link, and signals can be intercepted during stable operation as subsequent analysis objects. Multiple samples are repeatedly acquired for each type of operating condition to construct a dataset, and the samples are divided into training and test sets for classification verification.

[0086] The basic process of the method of this invention is "decomposition → mode selection → wavelet threshold denoising → reconstruction → feature → classification". To verify the effectiveness of the NRBO-FMD-wavelet method described in this invention in denoising and fault diagnosis, a comparative experiment was constructed under the same data and the same classifier settings. The comparison methods include: EMD-wavelet, VMD-wavelet, and traditional FMD-wavelet, and are compared with the method of this invention.

[0087] The EMD-wavelet method involves performing Empirical Mode Decomposition (EMD) on the preprocessed vibration signal to obtain multiple intrinsic mode components, selecting fault-related modes according to a preset mode selection criterion, and then reconstructing the denoised signal after wavelet threshold denoising on the selected modes. The VMD-wavelet method involves using Variational Mode Decomposition (VMD) to obtain multiple mode components and selecting fault-related modes, then performing wavelet threshold denoising on the selected modes and reconstructing the signal. The traditional FMD-wavelet method involves using Eigenmode Decomposition (FMD) to decompose the signal and select modes without adaptive parameter optimization, followed by wavelet threshold denoising and reconstruction.

[0088] The comparative experiment was conducted as follows: Under the same sampling data conditions, the above methods were applied to each working condition sample to obtain denoised / enhanced signals, and features were extracted to construct feature vectors. These vectors were then input into a pre-trained fault classifier to obtain diagnostic results. Noise reduction performance was evaluated using signal-to-noise ratio (SNR) and root mean square error (RMSE), while diagnostic performance was evaluated using indicators such as classification accuracy.

[0089] Under crankshaft bearing wear conditions, the time-domain results of noise reduction of the present invention and the comparative method are shown in Figures 3(a)-(d), where Figure 3(a) is the signal waveform after noise reduction by the EMD-wavelet method; Figure 3(b) is the signal waveform after noise reduction by the VMD-wavelet method; Figure 3(c) is the signal waveform after noise reduction by the traditional FMD-wavelet method; and Figure 3(d) is the fault component enhanced signal waveform obtained by the NRBO-FMD-wavelet method of the present invention.

[0090] As can be seen from the figure, different methods differ in terms of residual noise, impact retention, and waveform distortion. The method of this invention can better highlight the fault impact component and suppress background interference, making the impact position of the enhanced signal clearer and the energy more concentrated, thus providing an effective basis for subsequent feature extraction.

[0091] The experimental results for signal-to-noise ratio (SNR) and root mean square error (RMSE) are as follows: Figure 4 As shown in the figure, the method of the present invention has a higher average SNR and a lower average RMSE, indicating that the present invention has advantages in increasing the proportion of effective fault information and reducing reconstruction error, and the noise reduction and enhancement effect is better than the comparative method.

[0092] Under crankshaft bearing wear conditions, the time-frequency distribution diagrams of the enhanced fault component signal extracted by this invention and those of the comparison method are shown in Figures 5(a)-(e), where: Figure 5(a) is the short-time Fourier transform (STFT) time-frequency distribution of the original vibration signal; Figure 5(b) is the STFT time-frequency distribution of the signal after processing by the traditional FMD-wavelet method; Figure 5(c) is the STFT time-frequency distribution of the signal after processing by the traditional FMD-wavelet method; Figure 5(d) is the STFT time-frequency distribution of the signal after processing by the traditional FMD-wavelet method; Figure 5(e) is the STFT of the signal after processing by the NRBO-FMD-wavelet method of this invention. The time-frequency distribution is used to compare the effectiveness of different methods in highlighting fault impact and modulation components. As can be seen from the figure, the fault-related energy stripes and modulation structure are not prominent enough in the time-frequency distribution of the original signal. EMD-wavelet and VMD-wavelet can enhance the correlation band to a certain extent, but energy diffusion and noise residue still exist. Traditional FMD-wavelet further improves the signal but still has background interference. The NRBO-FMD-wavelet of this invention presents clearer impact stripes and more concentrated energy accumulation areas in the time-frequency domain, which can more effectively highlight fault impact and modulation components, further proving the effectiveness of "fault-sensitive mode screening + noise reduction and reconstruction".

[0093] On a dataset containing four states—normal, crankshaft bearing wear, intake valve leakage, and piston knocking—the samples are first divided into training and test sets according to a preset ratio. A fault classifier is then trained on the training set. Subsequently, predictions are made on the test set, and the overall recognition accuracy is calculated. The accuracy is calculated as Acc = N_correct / N_total × 100%, where N_correct is the number of correctly predicted samples, and N_total is the total number of samples in the test set. To ensure fairness in the comparison, this invention uses the same data partitioning and classifier configuration as the comparison method, differing only in the signal decomposition / denoising and feature construction stages. The four typical faults are classified as follows:

[0094] Category Number Fault type illustrate 0 Normal operating conditions All components have normal clearances, vibration is stable, and there is no obvious impact. 1 Crankshaft bearing wear conditions Increased bearing clearance leads to periodic impacts and mid-to-high frequency noise. 2 Intake valve leakage condition Intake valve not sealing properly, uneven combustion, and low-frequency energy redistribution. 3 Piston knocking condition Excessive clearance between the piston and cylinder wall causes a strong impact near top dead center.

[0095] The overall recognition accuracy results of the present invention and the comparison method are shown in the table below:

[0096] method Overall recognition accuracy / % EMD-Wavelet + Features + SVM 74.3 VMD-Wavelet + Feature + SVM 80.7 Traditional FMD-Wavelet + Features + SVM 82.1 This invention utilizes NRBO-FMD wavelet + features + SVM 96.3

[0097] As can be seen from the table above, the overall recognition accuracy of the method of this invention reaches 96.3%, which is significantly higher than that of the comparison method. Further analysis from the perspective of error rate shows that the error rate of this invention is 3.7%, a decrease of 14.2 percentage points compared to the 17.9% of the traditional FMD-wavelet method. This corresponds to a reduction of approximately 79.3% in the proportion of misjudged / missed samples, indicating that the method of this invention, which combines "parameter adaptive optimization decomposition + fault-sensitive mode screening + denoising reconstruction + multi-domain features," can improve the stability and reliability of multi-category working condition recognition.

[0098] Example 2

[0099] This application provides an apparatus for extracting and diagnosing engine fault signal features. The apparatus includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the method described in Embodiment 1.

[0100] It should be understood that in this embodiment, the memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory. For example, the memory may also store device type information.

[0101] The processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0102] A computing program (also called a program, software, software application, or code) includes a machine with a programmable processor.

[0103] Instructions can be used to implement these computational programs using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages.

[0104] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.

[0105] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor. The software modules can be located in mature storage media in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc.

[0106] Example 3

[0107] This application provides a non-transitory computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method described in Embodiment 1.

[0108] Storage media include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks, and other media that can store program code.

[0109] Those skilled in the art will recognize that the methods described in connection with the embodiments herein can be implemented using electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0110] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for extracting and diagnosing engine fault signal features, characterized in that, Includes the following steps: S1: Acquire the engine's vibration signal to be diagnosed, and record it as the original signal; S2: Preprocess the original signal to obtain the preprocessed vibration signal x(n); S3: Construct an adaptive parameter optimization space for Eigenmode Decomposition (FMD), wherein the parameters include at least the number of modes K and the center frequency f of each modal subband. k Bandwidth BW k and filter length L h , ; S4: Establish the mapping relationship between FMD parameters and modal components, and construct the parameter vector. As an optimization decision variable; S5: Construct a multi-objective integrated fitness function F(θ) with the goal of minimization, as an indicator to evaluate the quality of modal component decomposition; S6: The Newton-Raphson optimization algorithm (NRBO) is used to iteratively update the parameter vector θ until the termination condition is met, and the optimal parameter θ that minimizes the fitness function F(θ) is output. * ; S7: Based on the optimal parameter θ * The preprocessed vibration signal x(n) is subjected to final FMD decomposition to obtain K optimal modal components y. k (n); S8: Based on the correlation kurtosis and modal correlation coefficient, from the optimal modal component y k Select the set of fault-sensitive modes S from (n) * ; S9: For the fault-sensitive mode set S * The modal components in the signal are denoised using wavelet thresholding and then reconstructed to obtain the fault component enhanced signal x. f (n); S10: Extract time-domain, frequency-domain, and / or time-frequency-domain features from the fault component enhancement signal, construct a feature vector, input it into the fault classifier, and output the engine fault diagnosis result.

2. The method for extracting and diagnosing engine fault signal features according to claim 1, characterized in that, Step S1 specifically includes: setting at least one vibration sensor at at least one location in the engine block, cylinder head, crankcase or valve cover, and collecting real-time vibration data of the engine in operation as the vibration signal to be diagnosed.

3. The method for extracting and diagnosing engine fault signal features according to claim 1, characterized in that, The preprocessing in step S2 includes at least DC removal, bandpass filtering, and normalization.

4. The method for extracting and diagnosing engine fault signal features according to claim 1, characterized in that, In step S5, the formula for the multi-objective integrated fitness function is as follows: (Official 1) Among them, E rec The reconstruction error term, E, characterizes the signal integrity. orth For the orthogonality constraint term characterizing modal independence, E spar The sparsity penalty term is used to characterize the significance of fault impact, and w1, w2, and w3 are weighting coefficients. The weighting coefficients satisfy w1>w2 and w1>w3.

5. The method for extracting and diagnosing engine fault signal features according to claim 1, characterized in that, In step S6, before iterative updates, the population size NP and the maximum number of iterations MaxIter of the Newton-Raphson optimization algorithm (NRBO) need to be set; the termination conditions include reaching the maximum number of iterations or the change in the optimal fitness in multiple consecutive iterations being less than a preset threshold.

6. The method for extracting and diagnosing engine fault signal features according to claim 1, characterized in that, Step S8 specifically includes: Calculate the correlation kurtosis value of each optimal mode component, and select candidate modes based on a preset threshold; Calculate the modal correlation coefficients between candidate modes, and remove redundant modes based on the correlation coefficient thresholds to obtain the fault-sensitive mode set.

7. The method for extracting and diagnosing engine fault signal features according to claim 6, characterized in that, The relevant kurtosis value is based on the preprocessed vibration signal x(n) and modal components y k The cross-correlation sequence between (n) is calculated and used as a threshold ρ. th Perform filtering; the threshold ρ th Satisfying ρ th =μ+λσ, where μ and σ are the mean and standard deviation of the correlation kurtosis, respectively, and λ is an empirical coefficient, λ∈[0.5,1.5]; The modal correlation coefficient ρ i,j When |ρ i,j When |>ρ0, determine the mode y i (n) and y j (n) redundancy; The mode with the larger decorrelation kurtosis is retained as the representative mode, and ρ0 is the redundancy determination threshold, ρ0∈[0.8,0.95].

8. The method for extracting and diagnosing engine fault signal features according to claim 1, characterized in that, The wavelet threshold denoising in step S9 adopts either soft or hard threshold rules, preferably using the db8 wavelet basis, and the number of decomposition layers is 4 to 6. The features in step (10) include at least one of the following: root mean square, peak value, kurtosis, spectral kurtosis, energy ratio, spectral entropy, or wavelet packet energy; The classifier is a support vector machine (SVM), which uses a radial basis kernel function and employs a one-to-one or one-to-many strategy for multi-class discrimination.

9. An apparatus for extracting and diagnosing engine fault signal features, the apparatus comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the method according to any one of claims 1 to 8.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 8.

Images