A method and system for real-time monitoring of diesel engine operating status

Abstract

This invention relates to the field of electronic data processing technology and discloses a method and system for real-time monitoring of the operating status of a diesel engine. The method includes acquiring vibration signals from measuring points on the engine surface as the raw signal; performing mode decomposition to obtain the final intrinsic mode functions (IMFs). The decomposition process includes: performing multiple lumped operations, adding auxiliary noise signals to the raw signal to be decomposed in each lumped operation, and then repeating the filtering iteration process to sequentially separate a set of temporary IMFs; weighting and averaging the temporary IMFs of the same order obtained from all lumped operations to obtain the final IMF of that order; performing a Hilbert transform on the IMFs to construct the joint distribution of their instantaneous energy and instantaneous frequency, and extracting statistical and geometric features to form a feature vector. This invention can suppress mode aliasing, eliminate noise residue, and improve the accuracy of operating status monitoring results.

Description

Technical Field

[0001] This invention relates to the field of electronic data processing technology, specifically to a method and system for real-time monitoring of the operating status of a diesel engine. Background Technology

[0002] Diesel engines, as critical power equipment, are used in transportation, construction machinery, marine propulsion, and power generation. Real-time and accurate monitoring of their operating status is crucial for ensuring safe and reliable operation, preventing sudden failures, and reducing maintenance costs. Vibration signal analysis, a non-invasive and information-rich condition monitoring technology, has become a mainstream research direction in this field. The vibration signals of diesel engines are typically nonlinear and non-stationary signals, containing rich information about their operating status. Traditional signal processing methods, such as Fourier transform, are only suitable for analyzing stationary signals and struggle to reveal their time-varying characteristics. While wavelet transform possesses time-frequency analysis capabilities, its performance is limited by pre-selected wavelet basis functions, lacking adaptability and potentially leading to energy leakage or feature misjudgment. To overcome these limitations, Empirical Mode Decomposition (EMD) has been proposed as an adaptive signal processing method. It decomposes complex signals into a series of intrinsic mode functions (IMFs) based on the signal's own time-scale characteristics, exhibiting unique advantages in processing non-stationary signals.

[0003] However, traditional EMD methods suffer from "modal aliasing," where a single IMF contains components from different time scales, or components of the same scale are scattered across different IMFs. This severely impacts the accuracy of subsequent feature extraction and the interpretability of physical meaning. To suppress modal aliasing, Ensemble Empirical Mode Decomposition (EEMD) and its improved algorithms have been proposed. EEMD adds Gaussian white noise multiple times to the original signal, utilizing the uniform distribution of the white noise spectrum to assist in signal decomposition. After multiple decompositions, the noise is averaged to cancel out the added noise. Although EEMD alleviates mode mixing to some extent, it still has several shortcomings: First, when the lumped average number of iterations is limited, the auxiliary white noise cannot be completely neutralized, resulting in noise residue in the decomposition results and affecting the purity of the IMF components; Second, its core screening and iteration process still follows the framework of EMD, which has defects in the selection of extreme points, envelope construction, and the setting of iteration termination conditions. For example, it is sensitive to noise spikes, interpolation is prone to overshoot and undershoot, and the iteration stopping criterion is too simple. These problems limit the accuracy and efficiency of the decomposition; Third, the huge amount of computation limits its application in online monitoring applications that require high real-time performance. Summary of the Invention

[0004] This invention provides a method and system for real-time monitoring of the operating status of a diesel engine to solve the problem of inaccurate operating status monitoring results caused by modal aliasing and noise residue in the prior art.

[0005] In a first aspect, the diesel engine operating status real-time monitoring method of the present invention includes the following steps:

[0006] Vibration signals from measuring points on the body surface are acquired as the raw signal. Ensemble empirical mode decomposition (EMD) is performed on the raw signal to obtain the final intrinsic mode functions (IMFs). The EMD process includes: performing multiple lumped operations; in each lumped operation, an auxiliary noise signal matched with statistical characteristics is added to the raw signal to be decomposed; and a filtering iteration process is repeatedly performed on the signal after adding the auxiliary noise signal to sequentially separate a set of temporary IMFs. The filtering iteration process includes: a) adaptively filtering local extrema based on the local and global statistical characteristics of the signal to be processed; b) constructing upper and lower envelopes based on the retained extrema using a local adaptive interpolation method; and c) calculating the mean of the upper and lower envelopes and subtracting it from the signal to be processed. The mean is iterated from a to c repeatedly until a preset termination condition is met. The termination condition comprehensively considers the energy stability of the current component and its orthogonality with the separated temporary intrinsic mode functions. The temporary intrinsic mode functions of the same order obtained from all lumped operations are weighted and averaged to obtain the final intrinsic mode function of that order. The final intrinsic mode functions of all orders constitute the final intrinsic mode function set, where the weights are determined according to the signal-to-noise ratio or stability index of each temporary intrinsic mode function. Hilbert transform is performed on one or more selected intrinsic mode functions in the final intrinsic mode function set to construct the joint distribution of their instantaneous energy and instantaneous frequency, and the statistical and geometric features of the joint distribution are extracted to form a feature vector.

[0007] Preferably, adding an auxiliary noise signal that has been statistically matched to the original signal to be decomposed includes: generating a set of Gaussian white noise signals of the same length as the original signal to be decomposed, and adjusting the amplitude of the Gaussian white noise signals so that their standard deviation is a preset proportion of the global standard deviation of the original signal to be decomposed, wherein the preset proportion is 0.1 to 0.3; and then adding the adjusted Gaussian white noise signals to the original signal to be decomposed point by point.

[0008] Preferably, the step of adaptively selecting local extrema based on the local and global statistical characteristics of the signal to be processed includes: calculating the global standard deviation σ of the signal to be processed. g And set a global filtering threshold T g T g =α σ g For each extreme point, calculate the local standard deviation σ within its neighborhood window. l And set a local filtering threshold Tl =β σ l Only when the absolute magnitude of the extreme point is simultaneously greater than T g and T l When the extreme point is, it is preserved; where α is a coefficient between 0.3 and 0.7, and β is a coefficient between 1.0 and 1.5.

[0009] Preferably, the termination condition includes: setting an energy convergence threshold ε and an orthogonality exponent threshold ξ; during the screening iteration process, if the intermediate component obtained in the current iteration is h k (t), the intermediate component obtained in the previous iteration is h k-1 (t), then calculate the standard deviation criterion SD=Σ[(h k (t)-h k-1 (t)) 2 ] / Σ[h k-1 (t) 2 When SD is less than ε, and h k The iteration terminates when the cross-correlation coefficient between (t) and each of the separated temporary intrinsic mode functions in the current lumped operation is less than ξ.

[0010] Preferably, the step of weighted averaging of the temporary intrinsic mode functions of the same order obtained from all lumped operations includes: calculating the instantaneous frequency sequence of each temporary intrinsic mode function of the same order using Hilbert transform; calculating the standard deviation of the instantaneous frequency sequence and using the reciprocal of the standard deviation as a weight characterizing its stability; normalizing the weights of all temporary intrinsic mode functions of the same order and summing the corresponding temporary intrinsic mode functions accordingly to obtain the final intrinsic mode function of that order.

[0011] Preferably, the statistical and geometric features include: the statistical moments of the one-dimensional distribution obtained by projecting the joint distribution onto the instantaneous energy axis and the instantaneous frequency axis, wherein the statistical moments are selected from at least two of the mean, standard deviation, skewness, and kurtosis.

[0012] Preferably, the statistical and geometric features further include: the area and / or perimeter of the two-dimensional convex hull formed by the point set in the joint distribution on the energy-frequency plane, and the centroid coordinates of the joint distribution on the energy-frequency plane.

[0013] Preferably, the local adaptive interpolation method is a piecewise cubic Hermitian interpolation polynomial method.

[0014] Preferably, the step of acquiring the vibration signal of the measuring point on the surface of the engine as the raw signal includes: installing a piezoelectric accelerometer on the engine body near the cylinder head, acquiring the vibration acceleration signal through a data acquisition card, and using the acquired time series signal as the raw signal.

[0015] Secondly, the diesel engine operating status real-time monitoring system of the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the above-mentioned diesel engine operating status real-time monitoring method.

[0016] The beneficial effects of this invention are as follows: In the decomposition and screening iteration process, this invention introduces an extreme point discrimination mechanism based on the overall and local characteristics of the signal, effectively avoiding interference from noise spikes; and employs an optimized interpolation algorithm to construct the envelope, reducing waveform distortion. Simultaneously, the iteration termination condition integrates both energy stability and modal orthogonality indicators, ensuring that the physical meaning of each decomposed intrinsic mode function is clear and independent, suppressing mode aliasing. By weighted averaging the temporary modes obtained from multiple decompositions, residual auxiliary noise is further eliminated, resulting in final modal components with higher purity. Based on this, a joint distribution of instantaneous energy and frequency is constructed, and its morphological features are extracted to express subtle changes in the device's operating state. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the real-time monitoring method for the operating status of a diesel engine provided in an embodiment of the present invention. Detailed Implementation

[0018] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0019] like Figure 1 As shown, an embodiment of the real-time monitoring method for the operating status of a diesel engine provided by the present invention includes the following steps:

[0020] S1. Obtain the vibration signal of the measuring point on the surface of the diesel engine block as the raw signal.

[0021] Specifically, a piezoelectric accelerometer is installed on the engine block near the cylinder head of the diesel engine. Vibration acceleration signals are collected at a sampling frequency of 25.6 kHz using a data acquisition card, such as the NI 9234 board, and the collected time series signal is used as the raw signal.

[0022] S2. Perform an improved ensemble empirical mode decomposition on the original signal to obtain a set of final intrinsic mode functions. The decomposition process includes: performing multiple ensemble operations, adding an auxiliary noise signal matched by statistical characteristics to the original signal to be decomposed in each ensemble operation, and repeatedly performing a screening iteration process on the signal after adding noise to obtain a set of temporary intrinsic mode functions in sequence.

[0023] Specifically, the total number of lumped operations is set to, for example, 100 times. In the i-th operation, a Gaussian white noise sequence of the same length as the current signal to be decomposed is generated. The amplitude of the noise sequence is adjusted so that its standard deviation is 0.2 times the standard deviation of the signal to be decomposed. Then, the noise sequence is added to the signal to be decomposed to form a noisy signal. The noisy signal is then subjected to a filtering iteration.

[0024] The filtering iteration process is used to separate individual temporary intrinsic mode functions, and includes:

[0025] a. Based on the local and global statistical characteristics of the signal to be processed, adaptively filter local extrema. Specifically, identify all local maxima and minima in the signal, then calculate the overall standard deviation of the signal, and set a global threshold of 0.1 times this standard deviation. For each extrema, calculate the local standard deviation within its nearest small window (e.g., 50 data points), and set a local threshold. Only when the absolute value of an extrema is greater than both the global threshold and its corresponding local threshold is it retained as a valid extrema.

[0026] b. A local adaptive interpolation method is employed to construct upper and lower envelopes based on the retained extreme points. Specifically, a piecewise cubic Hermitian interpolation polynomial method is used to connect all retained maxima to form the upper envelope; then, all retained minima are connected to form the lower envelope. This method ensures the monotonicity of the interpolation curve between extreme points and avoids overshoot and oscillations that may occur with traditional spline interpolation.

[0027] c. Calculate the mean of the upper and lower envelopes and subtract this mean from the signal to be processed. Repeat steps a to c until a preset termination condition is met. The termination condition comprehensively considers the energy stability of the current component and its orthogonality with the separated temporary intrinsic mode functions. Specifically, add the upper and lower envelopes point by point and divide by two to obtain the mean envelope. Subtract this mean envelope from the signal to be processed to obtain a candidate component. Calculate the standard deviation ratio between this candidate component and the candidate component obtained in the previous iteration, and calculate the orthogonality index between this candidate component and all separated final intrinsic mode functions. When the standard deviation ratio is less than the energy convergence threshold (e.g., 0.01) and the orthogonality index is less than the orthogonality threshold (e.g., 0.05), the iteration terminates, and the current candidate component is confirmed as a temporary intrinsic mode function.

[0028] S3. Take a weighted average of the temporary intrinsic mode functions of the same order obtained from all lumped operations to obtain the final intrinsic mode function of that order. The final intrinsic mode functions of all orders constitute the final intrinsic mode function set, where the weights are determined based on the signal-to-noise ratio or stability index of each temporary intrinsic mode function.

[0029] For example, after 100 lumped operations, for the first order, 100 temporary first-order intrinsic mode functions are obtained; for each of these 100 temporary mode functions, the Pearson correlation coefficient with the original signal is calculated as its weight; then each temporary mode function is multiplied by its corresponding normalized weight and summed to obtain the final first-order intrinsic mode function; then this process is repeated for subsequent orders.

[0030] S4. Perform Hilbert transform on one or more selected intrinsic mode functions in the final intrinsic mode function set to construct a joint distribution of their instantaneous energy and instantaneous frequency, and extract the statistical and geometric features of the joint distribution to form a feature vector, which is used to characterize and monitor the operating status of the diesel engine.

[0031] Specifically, one or more final intrinsic mode functions (IMFs) are selected based on the energy proportion or the correlation with fault characteristic frequencies. A Hilbert transform is performed on each selected IMF to obtain its analytic signal, thereby calculating its instantaneous amplitude and instantaneous frequency over time. The instantaneous energy is the square of the instantaneous amplitude. This forms a three-dimensional energy-frequency-time distribution map. Features are extracted from this distribution, such as the mean and variance of instantaneous energy, the mean and variance of instantaneous frequency, the kurtosis and skewness of the energy distribution, and the area and centroid location of the main energy accumulation regions. These numerical features are combined into a feature vector, which is input into a pre-trained support vector machine classifier to output the specific operating state of the diesel engine, such as normal operation, misfire, or bearing wear.

[0032] In an optional embodiment, adding an auxiliary noise signal that has been statistically matched to the original signal to be decomposed includes: generating a set of Gaussian white noise signals of the same length as the original signal to be decomposed, and adjusting the amplitude of the Gaussian white noise signals so that their standard deviation is a preset proportion of the global standard deviation of the original signal to be decomposed, wherein the preset proportion is 0.1 to 0.3; and then adding the adjusted Gaussian white noise signals to the original signal to be decomposed point by point.

[0033] Adding Gaussian white noise helps distinguish signal components with similar amplitudes but large frequency differences, or intermittent signals with similar frequencies, ensuring the accuracy of subsequent decomposition. Statistical characteristic matching is crucial in this process, ensuring that the added noise both assists in decomposition and does not excessively obscure the useful information of the original signal. For example, assuming the vibration signal to be decomposed has 2048 data points, its global standard deviation is calculated to be 2.0 mV. A preset scaling factor of 0.2 is chosen, preferably within the range of 0.1 to 0.3. Therefore, the target standard deviation of the auxiliary noise signal to be generated is 0.4 mV. A Gaussian white noise sequence of the same length (2048 points) is generated, and its standard deviation is precisely adjusted to 0.4 mV through linear scaling. The adjusted noise sequence is then added one-to-one to each data point of the original vibration signal to form a new signal to be processed.

[0034] In an optional embodiment, the adaptive selection of local extrema based on the local and global statistical characteristics of the signal to be processed includes: calculating the global standard deviation σ of the signal to be processed. g And set a global filtering threshold T g T g =α σ g For each extreme point, calculate the local standard deviation σ within its neighborhood window. l And set a local filtering threshold T l =β σ l Only when the absolute magnitude of the extreme point is simultaneously greater than T g and T l When the extreme point is, it is preserved; where α is a coefficient between 0.3 and 0.7, and β is a coefficient between 1.0 and 1.5.

[0035] The dual threshold filtering mechanism intelligently identifies and preserves genuine signal fluctuations while effectively suppressing noise interference. Global threshold T g It is mainly used to filter out low-amplitude background noise that is prevalent throughout the entire signal, ensuring that only fluctuations with a certain energy are considered. The local threshold T... l Then, dynamic adjustments are made based on the local complexity of the signal. In the signal stationary region, the local standard deviation σ is... l Very small, T l It is also correspondingly small, able to preserve faint but true extreme points; in regions of drastic signal fluctuations, σ l It's very big, T l This also increases accordingly, effectively eliminating spurious extrema caused by noise and avoiding over-decomposition. Assume the global standard deviation σ of a signal... g With 5 units and α set to 0.6, the global threshold T is... gIt is 3 units. For a maximum value of 4.5 in the signal, its absolute amplitude is greater than T. g The first condition is met. Consider a neighborhood window around this point, for example, 10 data points before and after it, and calculate the local standard deviation σ within this window. l It is 2 units. Setting β to 1.2, the local threshold T... l The amplitude is 2.4 units. Since the amplitude of this maximum point, 4.5, is greater than both the global threshold of 3 and the local threshold of 2.4, this extreme point is retained as a valid point for constructing the envelope. Conversely, if there is another extreme point with an amplitude of 3.5, although it is greater than the global threshold, its neighborhood fluctuates wildly, and the local standard deviation σ is low. l The value is as high as 4, resulting in a local threshold T. l The value was 4.8, and this extreme point was removed because its amplitude was less than the local threshold.

[0036] In an optional embodiment, the termination condition includes: setting an energy convergence threshold ε and an orthogonality exponent threshold ξ; during the screening iteration process, if the intermediate component obtained in the current iteration is h k (t), the intermediate component obtained in the previous iteration is h k-1 (t), then calculate the standard deviation criterion SD=Σ[(h k (t)-h k-1 (t)) 2 ] / Σ[h k-1 (t) 2 When SD is less than ε, and h k The iteration terminates when the cross-correlation coefficient between (t) and each of the separated temporary intrinsic mode functions in the current lumped operation is less than ξ.

[0037] The termination condition ensures the convergence of the screening iteration process and the validity of the decomposition results. The standard deviation criterion SD is essentially the relative change in signal energy between two iterations. When this value is less than the preset energy convergence threshold ε, it indicates that the screening process has stabilized, and further iterations will not significantly change the components, so it can be stopped. This guarantees the stationarity of the intrinsic mode functions. Simultaneously, the second condition, the orthogonality check, avoids severe mode aliasing between newly extracted components and previously extracted components. For example, the energy convergence threshold ε is set to 0.001, and the orthogonality exponent threshold ξ is set to 0.05. After the k-th iteration, the calculated SD value is 0.0008, which is less than ε, satisfying the convergence condition. At this point, it is assumed that two temporary intrinsic mode functions, timf1 and timf2, have been separated in this lumped operation. The current component h needs to be calculated. k The cross-correlation coefficient between (t) and timf1 is 0.03, and the cross-correlation coefficient between (t) and timf2 is 0.04. Since both 0.03 and 0.04 are less than the orthogonality threshold ξ, it indicates that h...k (t) is sufficiently independent of the previously separated components. The screening iteration officially terminates only when both of these conditions are met, and h is set to... k (t) is output as a temporary intrinsic mode function.

[0038] In an optional embodiment, a weighted average is performed on all temporary intrinsic mode functions of the same order obtained from the lumped operation, including: calculating the instantaneous frequency sequence of each temporary intrinsic mode function of the same order using the Hilbert transform; calculating the standard deviation of the instantaneous frequency sequence and using the reciprocal of the standard deviation as a weight characterizing its stability; normalizing the weights of all temporary intrinsic mode functions of the same order, and then performing a weighted summation on the corresponding temporary intrinsic mode functions to obtain the final intrinsic mode function of that order.

[0039] A frequency-stability-based weighting strategy is used to optimize and integrate the results of multiple noise-assisted decompositions. Since the added white noise is random each time, the resulting temporary intrinsic mode functions (IMFs) of the same order will have slight differences. An ideal IMF should have a relatively stable instantaneous frequency. Therefore, the smaller the standard deviation of the instantaneous frequency sequence, the smaller the frequency fluctuation of the component, the clearer its physical meaning, and the higher its quality. Using the reciprocal of this standard deviation as the weight achieves the goal of giving greater weight to high-quality components.

[0040] For example, after performing 10 lumped operations, 10 first-order temporary intrinsic mode functions are obtained; for the i-th function, the standard deviation of its instantaneous frequency is calculated as σ. i Assume the standard deviations of the frequencies of the first three functions are σ1 = 2 Hz, σ2 = 5 Hz, and σ3 = 1.5 Hz, respectively. Then their corresponding original weights are w1 = 1 / 2 = 0.5, w2 = 1 / 5 = 0.2, and w3 = 1 / 1.5 ≈ 0.67. Adding all 10 original weights together gives the total weight W. total Each original weight is normalized by dividing it by the total weight to obtain the final weighting coefficients. The third function, with the most stable frequency, has the largest weight in the final synthesis of the first-order intrinsic mode functions, while the second function, with greater frequency fluctuations, has a smaller weight.

[0041] In an optional embodiment, the statistical and geometric features include: the statistical moments of the one-dimensional distribution obtained by projecting the joint distribution onto the instantaneous energy axis and the instantaneous frequency axis, wherein the statistical moments are selected from at least two of the mean, standard deviation, skewness, and kurtosis.

[0042] The intrinsic mode functions (IMFs) obtained after decomposition are transformed into numerical features that can be used for quantitative analysis and machine learning models. Through the Hilbert transform, each IMF can be mapped into a three-dimensional energy-frequency-time space. Its joint distribution on the energy-frequency plane reveals the intrinsic state of the signal. Projecting the two-dimensional joint distribution onto the instantaneous frequency axis yields a one-dimensional probability distribution with respect to frequency. Calculating the mean of this distribution reveals the center frequency of this mode component, the standard deviation represents its bandwidth, the skewness describes the symmetry of the frequency distribution, and the kurtosis reflects the degree of concentration or dispersion of the frequency distribution.

[0043] Similarly, projecting the joint distribution onto the instantaneous energy axis yields a one-dimensional probability distribution of energy. Calculating its mean reveals the average energy intensity of this mode, while the standard deviation reflects the magnitude of energy fluctuations. For example, during normal operation of a gearbox, the mean of the frequency distribution of a key intrinsic mode function might be stable at 100 Hz, with a standard deviation of 5 Hz and a skewness close to 0. When an early crack failure occurs, the mean frequency might remain unchanged, but the standard deviation might increase to 15 Hz, and the skewness becomes positive, reflecting the occurrence of nonlinear modulation. These statistical moments collectively constitute a multi-dimensional feature vector, capable of sensitively capturing changes in equipment status.

[0044] In an optional embodiment, the statistical and geometric features further include: the area and / or perimeter of the two-dimensional convex hull formed by the point set in the joint distribution on the energy-frequency plane, and the centroid coordinates of the joint distribution on the energy-frequency plane.

[0045] On the energy-frequency plane, all data points form a point cloud, and the two-dimensional convex hull of this point cloud is the smallest convex polygon that can enclose all data points. The area of ​​this convex hull directly reflects the combined magnitude of the signal energy and frequency variation range. In a healthy and stable system, the energy-frequency point cloud corresponding to the modal components is usually relatively clustered, and the convex hull area is small. When the system malfunctions or experiences abnormal conditions, new frequency components or energy abrupt changes may appear in the signal, causing the point cloud to become dispersed, and the convex hull area to increase significantly. Furthermore, the centroid coordinates of the point cloud give the position of the centroid of the entire distribution on the energy-frequency plane. For example, when a bearing is operating normally, the centroid of its first mode's energy-frequency distribution may be located at coordinate point, with an energy of 0.1 Joules and a frequency of 30 Hz. When the bearing experiences wear, increased friction, and slight fluctuations in rotational speed, the centroid may drift to a position with an energy of 0.3 Joules and a frequency of 35 Hz. By monitoring changes in the convex hull area and centroid coordinates, changes in the system's dynamic characteristics can be detected very intuitively.

[0046] The implementation principle of the real-time monitoring method for diesel engine operating status in this invention is as follows: During the decomposition and screening iteration process, this invention effectively avoids interference from noise spikes by introducing an extreme point discrimination mechanism based on the overall and local characteristics of the signal; simultaneously, an optimized interpolation algorithm is used to construct the envelope, reducing waveform distortion. Moreover, the iteration termination condition integrates both energy stability and modal orthogonality indicators, ensuring that the physical meaning of each decomposed intrinsic modal function is clear and independent, suppressing modal aliasing. In addition, by weighted averaging the temporary modes obtained from multiple decompositions, auxiliary noise residue is further eliminated, resulting in a final modal component with higher purity. Based on this, a joint distribution of instantaneous energy and frequency is constructed, and its morphological features are extracted, which can better express the subtle changes in the equipment operating status.

[0047] An embodiment of the diesel engine operating status real-time monitoring system provided by the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the diesel engine operating status real-time monitoring method in the above embodiment.

[0048] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for real-time monitoring of the operating status of a diesel engine, characterized in that, include: Vibration signals from measuring points on the body surface are used as raw signals; The original signal is subjected to ensemble empirical mode decomposition (EMD) to obtain the final intrinsic mode functions (IMFs). The EMD process includes: performing multiple lumped operations; in each lumped operation, an auxiliary noise signal matched with statistical characteristics is added to the original signal to be decomposed; and the filtering iteration process is repeated on the signal after adding the auxiliary noise signal to sequentially separate a set of temporary IMFs. The filtering iteration process includes: a) adaptively filtering local extrema based on the local and global statistical characteristics of the signal to be processed; b) constructing upper and lower envelopes based on the retained extrema using a local adaptive interpolation method; c) calculating the mean of the upper and lower envelopes and subtracting the mean from the signal to be processed, and repeating steps a to c. The process iterates until a preset termination condition is met. The termination condition comprehensively considers the energy stability of the current component and its orthogonality with the separated temporary intrinsic mode functions. The temporary intrinsic mode functions of the same order obtained from all lumped operations are weighted and averaged to obtain the final intrinsic mode function of that order. The final intrinsic mode functions of all orders constitute the final intrinsic mode function set, where the weights are determined based on the signal-to-noise ratio or stability index of each temporary intrinsic mode function. Hilbert transform is performed on one or more selected intrinsic mode functions in the final intrinsic mode function set to construct the joint distribution of their instantaneous energy and instantaneous frequency, and the statistical and geometric features of the joint distribution are extracted to form a feature vector.

2. The method for real-time monitoring of diesel engine operating status according to claim 1, characterized in that, Adding an auxiliary noise signal that has been statistically matched to the original signal to be decomposed includes: generating a set of Gaussian white noise signals of the same length as the original signal to be decomposed, and adjusting the amplitude of the Gaussian white noise signals so that their standard deviation is a preset proportion of the global standard deviation of the original signal to be decomposed, wherein the preset proportion is 0.1 to 0.3; and then adding the adjusted Gaussian white noise signals to the original signal to be decomposed point by point.

3. The method for real-time monitoring of diesel engine operating status according to claim 1, characterized in that, The adaptive selection of local extrema based on the local and global statistical characteristics of the signal to be processed includes: calculating the global standard deviation σ of the signal to be processed. g And set a global filtering threshold T g T g =α σ g For each extreme point, calculate the local standard deviation σ within its neighborhood window. l And set a local filtering threshold T l =β σ l Only when the absolute magnitude of the extreme point is simultaneously greater than T g and T l When the extreme point is, it is preserved; where α is a coefficient between 0.3 and 0.7, and β is a coefficient between 1.0 and 1.

5.

4. The method for real-time monitoring of diesel engine operating status according to claim 1, characterized in that, The termination conditions include: setting an energy convergence threshold ε and an orthogonality exponent threshold ξ; during the screening iteration process, if the intermediate component obtained in the current iteration is h k (t), the intermediate component obtained in the previous iteration is h k-1 (t), then calculate the standard deviation criterion SD=Σ[(h k (t)-h k-1 (t)) 2 ] / Σ[h k-1 (t) 2 When SD is less than ε, and h k The iteration terminates when the cross-correlation coefficient between (t) and each of the separated temporary intrinsic mode functions in the current lumped operation is less than ξ.

5. The method for real-time monitoring of diesel engine operating status according to claim 1, characterized in that, The step of weighted averaging of the temporary intrinsic mode functions of the same order obtained from all lumped operations includes: calculating the instantaneous frequency sequence of each temporary intrinsic mode function of the same order using Hilbert transform; calculating the standard deviation of the instantaneous frequency sequence and using the reciprocal of the standard deviation as a weight characterizing its stability; normalizing the weights of all temporary intrinsic mode functions of the same order and summing the corresponding temporary intrinsic mode functions accordingly to obtain the final intrinsic mode function of that order.

6. The method for real-time monitoring of the operating status of a diesel engine according to claim 1, characterized in that, The statistical and geometric features include: the statistical moments of the one-dimensional distribution obtained by projecting the joint distribution onto the instantaneous energy axis and the instantaneous frequency axis, wherein the statistical moments are selected from at least two of the mean, standard deviation, skewness, and kurtosis.

7. The method for real-time monitoring of the operating status of a diesel engine according to claim 6, characterized in that, The statistical and geometric features also include: the area and / or perimeter of the two-dimensional convex hull formed by the point set in the joint distribution on the energy-frequency plane, and the centroid coordinates of the joint distribution on the energy-frequency plane.

8. The method for real-time monitoring of the operating status of a diesel engine according to claim 1, characterized in that, The local adaptive interpolation method is a piecewise cubic Hermitian interpolation polynomial method.

9. The method for real-time monitoring of the operating status of a diesel engine according to claim 1, characterized in that, The process of acquiring vibration signals from measuring points on the engine body surface as raw signals includes: installing a piezoelectric accelerometer on the engine body near the cylinder head, acquiring vibration acceleration signals through a data acquisition card, and using the acquired time-series signals as raw signals.

10. A real-time monitoring system for the operating status of a diesel engine, characterized in that, It includes a memory and a processor. The memory stores computer instructions. When the processor executes the computer instructions, it implements the real-time monitoring method for the operating status of a diesel engine as described in any one of claims 1-9.

Images