Method and system for monitoring operating state of driverless vehicle

Abstract

The present invention relates to the technical field of electrical signal processing, and specifically relates to a method and system for monitoring the operating state of a driverless vehicle. Residual amplitudes, which do not have Gaussianity, are screened out by means of iteration, a vibration element is designed on the basis of the largest residual amplitude and the smallest residual amplitude among the screened-out residual amplitudes, and a second original signal is obtained by performing convolution on a first original signal. Then, segmentation and matching are performed on an upper envelope and a lower envelope of the second original signal, and the symmetry and the maximum amplitude difference of a matching signal segment are calculated; and then an estimated intensity of residual noise is obtained on the basis of the correlation between the symmetry and the maximum amplitude difference, so as to generate Gaussian filtering having the same intensity, and smoothing is performed on component signals obtained after the second original signal has been subjected to LMD, so as to filter out deep noise. In the present invention, a monitoring data pre-processing system is optimized by means of an adaptive vibration element and an adaptive Gaussian filtering kernel, thereby effectively improving the precision of monitoring data.

Description

A method and system for monitoring the working status of an unmanned vehicle Technical Field

[0001] The present invention relates to the technical field of electrical signal processing, and in particular to a method and system for monitoring the working status of an unmanned vehicle. Background Art

[0002] Unmanned driving system technology uses a variety of on-board sensors (such as cameras, lidar, millimeter-wave radar, GPS, inertial sensors, etc.) to identify the vehicle's surrounding environment and status, and independently analyzes and judges based on the environmental information obtained, thereby controlling the vehicle to achieve unmanned driving. The VCU (vehicle control unit) reads the operating status of control units such as the vehicle's engine, transmission, and accelerator pedal and controls them. However, during driving, the information received or transmitted by the VCU is always subject to various types of noise interference. The processing logic of traditional pre-processing units is often indiscriminate and smooth, unable to eliminate deep noise, which may cause the vehicle to be allergic to commands, such as sudden braking and sharp turns. Or, due to the influence of noise, it is impossible to diagnose and respond to abnormal or fault information of each control unit in a timely manner, posing a major safety hazard.

[0003] Summary of the Invention

[0004] The present invention provides a method and system for monitoring the working status of an unmanned vehicle to solve the problem that the existing traditional preprocessing unit often performs indiscriminate smoothing on the processing logic of the working status monitoring signals of each control unit of the unmanned vehicle, is unable to eliminate deep noise, and the residual noise may cause the vehicle to be allergic to commands.

[0005] The present invention provides a method and system for monitoring the working status of an unmanned vehicle using the following technical solutions:

[0006] In one aspect, an embodiment of the present invention provides a method for monitoring the operating status of an unmanned vehicle, the method comprising the following steps:

[0007] Obtaining a vibration signal of an unmanned vehicle engine as a first original signal, and obtaining maximum and minimum weighted reference amplitudes within a vibration element based on a residual amplitude of the first original signal, obtaining a weighted reference amplitude change rate within the vibration element based on the maximum and minimum weighted reference amplitudes, obtaining a weighted reference amplitude at each sampling point of the vibration element based on the maximum and minimum weighted reference amplitudes and the weighted reference amplitude change rate, obtaining a weight value at each sampling point of the vibration element based on the weighted reference amplitude at each sampling point of the vibration element, obtaining a vibration element based on the weighted values, and convolving the vibration element with the first original signal to obtain a second original signal;

[0008] Obtaining upper and lower envelopes of the second original signal, performing signal segmentation on the upper and lower envelopes to obtain upper and lower envelope signal segment sets, respectively, matching signal segments in the upper and lower envelope signal segment sets to obtain a matched signal segment set, and determining the symmetry and maximum amplitude difference of all matched signal segments in the matched signal segment set;

[0009] Obtaining a Pearson correlation coefficient between the symmetry and the maximum amplitude difference of all matching signal segments in the set of matching signal segments based on the symmetry and the maximum amplitude difference, obtaining a degree of interference of the residual noise based on the Pearson correlation coefficient between the symmetry and the maximum amplitude difference of all matching signal segments, obtaining an adaptive Gaussian filter kernel based on the interference degree, and performing Gaussian filtering and smoothing on the LMD component signal of the second original signal using the adaptive Gaussian filter kernel to obtain a smoothed component signal;

[0010] The smoothed component signals are reconstructed to obtain an engine vibration signal without noise interference.

[0011] Preferably, the specific steps of obtaining the vibration element are as follows:

[0012] The preset vibration element length is L;

[0013] Subtracting the amplitude average from all amplitudes of the first original signal to obtain a residual amplitude set, and obtaining the filtered residual amplitude set by constructing an objective function;

[0014] Obtaining the maximum residual amplitude and the minimum residual amplitude in the screened residual amplitude set, using the maximum residual amplitude and the minimum residual amplitude as the weight reference amplitudes at both ends of the vibration element, obtaining the weight reference amplitude change rate of adjacent sampling point positions based on the weight reference amplitudes at both ends of the vibration element and the vibration element length, obtaining the weight reference amplitude of each sampling point position based on the weight reference amplitude change rate, and finally normalizing all weight reference amplitudes to obtain the weight value of each sampling point position of the vibration element;

[0015] The vibration element is obtained according to the length of the vibration element and the weight value of each sampling point position of the vibration element.

[0016] Preferably, the specific calculation method for filtering out some residual amplitudes by constructing an objective function and concentrating the vibration signal and the modulation signal with less noise interference in the filtered residual amplitude set is as follows:

[0017] The exhaustive method is used to iteratively screen out any number of residual amplitudes in all residual amplitude sets. The objective function is:

[0018] Among them, g i Represents the value of the i-th residual amplitude in the residual amplitude set remaining during the screening process, represents the average value in the set of residual amplitudes, N represents the number of residual amplitudes in the exhaustive iteration process; g i' represents the value of the i'th residual amplitude in the residual amplitude set when any residual amplitude is screened out, and M represents the total number of amplitudes when no residual amplitude is screened out;

[0019] As the exhaustive iteration proceeds, when the output value of the objective function E is minimized, a set of residual amplitudes that are screened out is obtained.

[0020] Preferably, the specific steps of normalizing all weight reference amplitudes to obtain the weight value of each sampling point position of the vibration element are as follows:

[0021] Among them, j v Represents the weighted reference amplitude of the vth sampling point position of the vibration element, v represents any sampling point position in the vibration element that does not contain the left and right endpoints, j max 、j min Respectively represent the maximum and minimum values ​​of the screened residual amplitude, and serve as the weighted reference amplitudes at the left vibration convex endpoint and the right vibration attenuation cutoff end in the vibration element;

[0022] in, is the rate of change of the weight reference amplitude of adjacent sampling points within the vibration element;

[0023] Among them, s represents any sampling point position including the left and right endpoints, j s Represents the weight reference amplitude of the s-th position, That is, the weighted reference amplitude normalization of the s-th sampling point position of the vibration element, ρ s Represents the weight value of the sth sampling point position;

[0024] The weight values ​​of all sampling point positions of the vibration element are obtained according to the weight reference amplitudes of all sampling point positions.

[0025] Furthermore, the specific steps of convolving the vibration element with the first original signal to obtain the second original signal are as follows:

[0026] Take each sampling point of the first original signal as the convolution target point, make the convolution target point coincide with the sampling point position with the highest weight in the vibration element, the sampling point position with the highest weight is the leftmost endpoint of the vibration element, and align the remaining sampling point positions on the right with the sampling points on the right side of the convolution target point. Multiply the weight of each sampling point position in the vibration element with the signal amplitude of the corresponding overlapping sampling point, add the products of all sampling point positions in the vibration element to obtain the convolution result, replace the amplitude of the original convolution target point with the convolution result, and then perform convolution processing on all signal sampling points to obtain the second original signal.

[0027] Furthermore, the specific steps of performing signal segmentation on the upper and lower envelopes are as follows:

[0028] Obtain all maximum and minimum points of the second original signal and use them to fit the upper and lower envelopes respectively, and then obtain all maximum and minimum points on the upper envelope and all maximum and minimum points on the lower envelope based on the fitted upper and lower envelopes;

[0029] For the upper envelope, intercept the signal segments between all the maximum points and the next adjacent minimum points on the upper envelope, and divide the upper envelope into several signal segments between the maximum points and the minimum points, which are recorded as upper envelope signal segments;

[0030] For the lower envelope, intercept the signal segments between all the minimum points and the next adjacent maximum points on the lower envelope, and divide the lower envelope into several signal segments between the minimum point and the maximum point, which are recorded as the lower envelope signal segments;

[0031] Then, the signal segment sets of the upper and lower envelopes are obtained respectively.

[0032] Furthermore, the specific steps of obtaining the matching signal segment are as follows:

[0033] The midpoints of all upper and lower envelope signal segments are extracted, and then the timing difference between the midpoints of any two upper and lower envelope signal segments is calculated. The two upper and lower envelope signal segments with the smallest midpoint timing difference are taken as a group of matching signal segments.

[0034] Furthermore, the specific method for obtaining the symmetry of the matching signal segment and the maximum amplitude difference is as follows:

[0035] According to the amplitudes of the upper and lower envelope signal segments of each group of matching signal segments, the amplitude variance of the upper envelope signal segment and the amplitude variance of the lower envelope signal segment are obtained, and according to the absolute value of the difference between the amplitude variance of the upper envelope signal segment and the amplitude variance of the lower envelope signal segment, the symmetry of the matching signal segment is obtained;

[0036] The maximum amplitude difference of each match is obtained by subtracting the maximum amplitude in the upper envelope signal segment from the minimum amplitude in the lower envelope signal segment in each group of matching signal segments.

[0037] Preferably, the interference degree of the residual noise is obtained according to the Pearson correlation coefficient of the symmetry and the maximum amplitude difference of all matching signal segments, an adaptive Gaussian filter kernel is obtained according to the interference degree, and the LMD component signal of the second original signal is Gaussian filtered and smoothed using the adaptive Gaussian filter kernel to obtain a smoothed component signal, including the specific calculation method as follows:

[0038] Among them, B represents the interference degree of residual noise, Kz Represents the symmetry of the zth group of matching signal segments, with a total of U groups of matching signal segments. represents the average symmetry of all matching signal segments, C z Represents the maximum amplitude difference of the zth group of matching signal segments, Represents the average maximum amplitude difference of all matching signal segments, σ(K z ) represents the standard deviation of the symmetry of the matching signal segment, σ(C z ) represents the standard deviation of the maximum amplitude difference of the matching signal segments;

[0039] is the Pearson correlation coefficient of the symmetry and maximum amplitude difference of all matching signal segments;

[0040] Perform LMD decomposition on the second original signal to obtain all PF component signals;

[0041] The interference degree B of the residual noise is used as the standard deviation of the Gaussian filter kernel to obtain an adaptive Gaussian filter kernel. The adaptive Gaussian filter kernel is used to smooth the decomposed PF component signals to obtain the denoised PF component signals. The original signal is then reconstructed using a time synchronization and component superposition method to obtain a noise-free engine vibration signal.

[0042] On the other hand, an embodiment of the present invention further provides an unmanned vehicle operating status monitoring system, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor to implement the aforementioned method.

[0043] The present invention has at least the following beneficial effects:

[0044] This invention addresses the problem of deep-seated noise that cannot be eradicated in current multi-sensor monitoring systems used in autonomous vehicles. A method and system for monitoring the operating status of autonomous vehicles is proposed. Based on the existing LMD algorithm, this method iteratively removes residual amplitudes that are not Gaussian, concentrating signals less susceptible to noise in the removed residual amplitudes. A vibration element is then designed based on the maximum and minimum residual amplitudes, and convolved with the first original signal to obtain a second original signal. This method initially suppresses prominent noise while preserving the vibration signal characteristics to the greatest extent possible. The upper and lower envelopes of the second original signal are then segmented and matched, and the symmetry and maximum amplitude difference of the matched signal segments are calculated. Based on the correlation between the two, the interference level of the residual noise is determined and used as an estimated noise intensity. A Gaussian filter of the same intensity is then generated to smooth the component signals of the second original signal after LMD decomposition to remove deep-seated noise. Compared to traditional indifferential smoothing processing logic, the adaptive vibration element in this invention effectively preserves vibration characteristics, while the adaptive Gaussian filter kernel is set based on the estimated residual noise intensity, minimizing residual noise. It avoids noise interference while retaining detailed information of the vibration signal, greatly improving the accuracy of sensor monitoring data and providing reliable data support for subsequent analysis, processing and feedback of the vehicle control unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] FIG1 is a flowchart of a method for monitoring the working status of an unmanned vehicle according to the present invention;

[0047] FIG2 is a schematic diagram of the vibration characteristics of the engine vibration signal. DETAILED DESCRIPTION

[0048] To further illustrate the technical means and effectiveness of the present invention in achieving its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effectiveness of a method and system for monitoring the operating status of an unmanned vehicle proposed in accordance with the present invention. In the following description, different references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0049] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0050] Please refer to FIG1 , which shows a flowchart of a method for monitoring the working status of an unmanned vehicle provided by this embodiment. The method includes the following steps:

[0051] S001. Obtain an engine vibration signal using a vibration sensor to obtain a first original signal.

[0052] The operating status of an autonomous vehicle can be reflected in the sensor monitoring signals of its various control units. Take the engine vibration signal as an example. As the engine generates power, it generates high-frequency vibrations. The vibration sensor on the surface of the engine cylinder wall collects the vibration signal at a sampling frequency of Fs = 1000Hz.

[0053] This signal is then transmitted to the VCU, which calculates the engine's operating status and outputs a control signal for adjustment. Prior to this, the acquired vibration signal undergoes noise reduction preprocessing to improve the accuracy of the subsequent feedback control signal. This vibration signal is referred to as the first raw signal.

[0054] S002. Preset the vibration element length, subtract the amplitude average from all amplitudes in the first original signal to obtain a residual amplitude set, iteratively filter out some residual amplitudes, and use the standard Gaussian distribution characteristics to constrain the iterative process, so as to concentrate the vibration signals and modulation signals with small noise interference in the filtered residual amplitudes.

[0055] Engine vibration pulse signals generally exhibit significant non-stationary characteristics. Due to the rotation of the engine crankshaft and the bumps of the road, the various frequency components in the vibration signal will exhibit varying degrees of modulation. Modulation refers to the process of embedding information from one signal into another, thereby altering the characteristic parameters of the embedded signal. To address this signal modulation phenomenon, the LMD algorithm is currently used to decompose the different modulation components.

[0056] The decomposition process of LMD is as follows:

[0057] The LMD method essentially decomposes a complex original signal into the sum of several PF components. It requires continuously extracting and gradually removing the high-frequency components in the original signal. First, all local maximum and local minimum points of the original signal are obtained. Then, a sliding average method is used to obtain the local mean function and envelope of the signal. The local mean function is removed from the original signal and demodulated with the envelope until a standard pure frequency modulation function is obtained. The loop iteration is terminated. All envelopes generated during the iterative process are accumulated to obtain the envelope function, which is multiplied by the final pure frequency modulation function to obtain the first-order PF component. After separating the first-order PF component from the original signal, the above steps are repeated to decompose the PF component signals of each order in turn.

[0058] It can be seen that the decomposition process of each modulated component is related to the envelope. In the absence of noise, LMD effectively decomposes the vibration signal and each modulated signal, enabling the VCU to more accurately monitor the engine status during vehicle operation. However, noise distorts the envelope estimation function, causing noise modes to appear in the modulated component signals. This makes it difficult to effectively suppress noise in the presence of modulated signals. This embodiment proposes an optimization method to address this problem by designing convolutional vibration elements and adaptive Gaussian filtering.

[0059] It can be seen that a complete vibration process of the engine vibration signal is a convex phenomenon of the signal, the amplitude increases instantly, and then the amplitude gradually decays until it is cut off. This is because at the moment of each explosion, the inertial force of the fuel explosion on the engine cylinder wall is the largest, and the inertial force is gradually eliminated through high-frequency vibration. Then the real, unmodulated, and noise-free vibration signal is theoretically shown in Figure 2.

[0060] Whether it's engine vibration, crankshaft rotation, or road bumps, even if they're modulated, they're still considered vibration signals, whereas noise isn't. Therefore, a vibration signal model, called a vibrometer, can be used to convolve the original signal to initially suppress noise. A vibrometer, similar to signal filtering, has only length but no height. Therefore, to determine a suitable vibrometer, two hyperparameters must be determined: its length and its weight distribution.

[0061] Specifically, first, based on empirical data on the duration of an engine fuel burst, the vibration element length is set to 20ms. This can be adjusted based on different engine models. Then, based on a sampling frequency of Fs = 1000Hz, the vibration element needs to have L = 20 sampling point weights.

[0062] Furthermore, the first original signal is normalized by subtracting the average amplitude from all amplitudes in the first original signal to obtain a set of residual amplitudes. The residual amplitudes contain engine vibration signals, modulation signals, and noise signals. By filtering out some residual amplitudes, the remaining residual amplitude set is made Gaussian in the most robust way. This allows signals with significant noise interference to be retained in the residual amplitude set, while vibration signals and modulation signals with minimal noise interference are concentrated in the filtered residual amplitudes.

[0063] The iterative elimination of residual amplitudes can be performed using an exhaustive approach. This involves eliminating any residual amplitude, or any number of residual amplitudes, from the residual amplitude set, and obtaining the optimal elimination result based on the objective function. Because it is impossible to directly predict noise intensity from complex mixed signals, the elimination process can only be constrained by a standard Gaussian distribution with a mean of 0 and a variance of 1.

[0064] Furthermore, when exhaustive method is used to iteratively screen out any number of residual amplitudes, the objective function is:

[0065] Among them, g i Represents the value of the i-th residual amplitude in the residual amplitude set remaining during the screening process, represents the average value in the set of residual amplitudes, N represents the number of residual amplitudes in the exhaustive iteration process; g i' represents the value of the i'th residual amplitude in the residual amplitude set when any residual amplitude is screened out, and M represents the total number of amplitudes when no residual amplitude is screened out;

[0066] in, Both represent the average value of the residual amplitude. is the variance of the residual amplitude. When the residual amplitude set conforms to the standard Gaussian distribution, the mean value is close to 0 and the variance is close to 1. Therefore, the objective function The closer it is to 0. But it is necessary to limit the screening scale, Represents the absolute value of the total residual amplitude before any residual amplitude is filtered out. Represents the absolute value of the total residual amplitude after iteratively screening out some residual amplitudes. The larger the difference between the two, the more scale anomalies are screened out. That is, in order to make the objective function converge, more residual amplitudes are screened out, resulting in data distortion. Therefore, it needs to be used as a penalty term. It represents the absolute value of the total residual amplitude before screening divided by the absolute value of the total residual amplitude after iterative screening. The more abnormal the screening scale is, the larger the ratio is and the greater the penalty is. Therefore, the objective function is

[0067] Furthermore, as the iterative screening proceeds, the objective function will output an E value each time the residual amplitude is screened out. When the output value of the objective function E is the smallest, it means that the objective function has converged. At this time, the remaining residual amplitude set has the best Gaussianity, which means that the screened part is the residual amplitude with little noise interference.

[0068] S003. Obtain the maximum residual amplitude and the minimum residual amplitude in the screened residual amplitude set, and use them as the weight reference amplitudes at both ends of the vibration element. According to the weight reference amplitude and the vibration element length, obtain the rate of change of the adjacent sampling point positions, and then obtain the weight reference amplitude of each sampling point position. Finally, normalize all the weight reference amplitudes to obtain the weight value of each sampling point position of the vibration element, and then obtain the vibration element. Use the vibration element to convolve the first original signal to obtain the second original signal.

[0069] According to the convergence result of the objective function in step S002, a set of filtered residual amplitudes can be obtained. The filtered residual amplitudes are most likely the engine vibration signals and modulation signals with the least noise interference and the most prominent features in the original signal. The vibration sensor originally collects the engine signal, so other modulation signals such as crankshaft rotation and turbulence are necessarily weaker than the engine signal. Therefore, these filtered residual amplitudes are restored to the original signal, and the maximum residual amplitude and the minimum residual amplitude are taken as the weight reference amplitudes of the left vibration convex end and the right attenuation cutoff end of the convolution element. Then, according to the length of the vibration element, the weight reference amplitude change rate of each adjacent sampling point position between the two end points is obtained, and then the weight reference amplitude and weight of all positions of the vibration element are obtained. The specific calculation method is as follows:

[0070] Among them, j v Represents the weighted reference amplitude of the vth sampling point position of the vibration element, v represents any sampling point position in the vibration element that does not contain the left and right endpoints, j max 、j min Respectively represent the maximum and minimum values ​​of the screened residual amplitude, and serve as the weighted reference amplitudes at the left vibration convex endpoint and the right vibration attenuation cutoff end in the vibration element;

[0071] in, is the weight reference amplitude change rate of adjacent sampling points in the vibration element, j v The calculation method is That is, starting from the weight reference amplitude of the left endpoint, the weight reference amplitude of each sampling point position of the vibration element is obtained by attenuating it v times according to the weight reference amplitude change rate of the adjacent sampling point position;

[0072] Among them, s represents any sampling point position including the left and right endpoints, j sRepresents the weight reference amplitude of the s-th position, is the sum of the weighted reference amplitudes of all sampling points of the vibration element, That is, the weight reference amplitude of the vibration element at the s-th sampling point position is normalized to obtain the weight value ρ at the s-th sampling point position s .

[0073] The weight values ​​of all sampling point positions of the vibration element are obtained according to the weight reference amplitudes of all sampling point positions.

[0074] The vibration element can be obtained according to the obtained vibration element length and the weight values ​​of all sampling point positions of the vibration element.

[0075] Furthermore, the first original signal is convolved with the vibration element. The traditional convolution method is to coincide the midpoint of the convolution kernel with the convolution target point. However, in order to ensure the convexity and attenuation characteristics of the vibration signal itself, the sampling point of each first original signal is still used as the convolution target point in the present invention, so that the convolution target point coincides with the sampling point position with the highest weight in the vibration element. The sampling point position with the highest weight is the leftmost endpoint of the vibration element, and the remaining sampling point positions on the right are aligned with the sampling points on the right side of the convolution target point. The weights of the sampling point positions in the vibration element are multiplied by the signal amplitudes of the corresponding coincident sampling points, and the products of all the sampling point positions in the vibration element are added to obtain the convolution result, and the amplitudes of the original convolution target points are replaced by the convolution results.

[0076] Furthermore, all signal sampling points are convolved to obtain a second original signal. The second original signal after preliminary noise reduction can suppress the influence of abnormal noise while retaining the vibration signal.

[0077] S004. Segment the upper and lower envelopes of the second original signal into signal segments, then match the upper and lower envelope signal segments based on the minimum timing difference, and calculate the symmetry and maximum amplitude difference of each matched signal segment.

[0078] While the use of vibration elements suppressed some prominent noise, some noise was still deeply mixed with the modulation signal and the engine vibration signal, making it difficult to filter out directly. However, the envelope of the suppressed signal already exhibited distinct convex-decay characteristics and good symmetry. Regarding the modulation of the crankshaft drive, the normal vibration of the engine should have a certain correlation with the crankshaft drive speed. Simply put, the engine injection-combustion-explosion process must positively promote the drive system, and the modulation of the engine vibration signal by the crankshaft rotation must also have a significant correlation. Modulation means that the crankshaft rotation causes the engine vibration signal to have the changing characteristics of the crankshaft rotation signal. When the two are highly correlated, since they are both periodic signals, they do not affect the symmetry of the engine vibration signal.

[0079] This embodiment calculates the correlation between the symmetry of the characteristic signal segments and the amplitude difference of the second original signal, feeds back the degree of residual noise interference, and then obtains hyperparameters for adjusting Gaussian filtering to smooth the residual noise in the decomposed signal to completely eliminate the noise.

[0080] Specifically, all maximum and minimum points of the second original signal are obtained, and upper and lower envelopes are fitted. Then, all maximum and minimum points of the upper and lower envelopes are obtained.

[0081] It should be noted that the maximum and minimum points used to fit the upper and lower envelopes are all extreme points of the second original signal itself, while the maximum and minimum points on the upper and lower envelopes are not the same as the maximum and minimum points on the second original signal.

[0082] For the upper envelope, intercept the signal segments between all the maximum points and the next adjacent minimum points on the upper envelope, and divide the upper envelope into several signal segments between the maximum points and the minimum points, which are recorded as upper envelope signal segments;

[0083] For the lower envelope, intercept the signal segments between all the minimum points and the next adjacent maximum points on the lower envelope, and divide the lower envelope into several signal segments between the minimum point and the maximum point, which are recorded as the lower envelope signal segments;

[0084] The purpose of segmenting the envelope curve according to the above method in this embodiment is to extract the vibration signal characteristics in the second original signal, namely the amplitude spike-decay characteristics of the vibration signal, and perform signal segment matching based on the intercepted upper and lower envelope curve signal segments.

[0085] Furthermore, the upper and lower signal segments are matched according to the closest timing difference, that is, the midpoints of all upper and lower envelope signal segments are extracted, and then the two upper and lower envelope signal segments with the smallest timing difference between the midpoints of any two upper and lower envelope signal segments are selected as a group of matching signal segments. The matching method is as follows: D(p, q) = |t p -t q |

[0086] Where p and a represent any signal segment of the upper and lower envelope respectively, D(p,q) represents the timing difference between the upper envelope signal segment p and the lower envelope signal segment q, t p , t q They represent the midpoint time values ​​of the two signal segments respectively. p -t q| is the absolute value of the difference between the midpoint time values ​​of the pth upper envelope signal segment and the qth lower envelope signal segment. When the timing difference between the pth upper envelope signal segment and the qth lower envelope signal segment is minimal, segments p and q are matched together into a group, called a matched signal segment.

[0087] According to the above logic, the signal segment contains the amplitude spike-attenuation characteristics of the vibration signal, and the purpose of matching the signal segment is to match the upper and lower envelope signal segments within the same vibration cycle, thereby creating conditions for the next step of processing.

[0088] Furthermore, after all signal segments are matched, the symmetry of each set of matched signal segments can be calculated. The calculation method is as follows: K z =|σ 2 (W a )-σ 2 (W b )|

[0089] Among them, z represents the zth group of matching signal segments, K z represents the symmetry of the zth group of matching signal segments, a and b represent the upper and lower envelope signal segments of the group of matching signal segments, and W a 、W b Represents any amplitude in the a and b signal segments, σ 2 (W a ),σ 2 (W b ) represent the variance of the amplitude of signal segments a and b respectively.

[0090] Among them, when the matching signal segment is well symmetrical, that is, the distance from the upper envelope to the zero line is equal to the distance from the lower envelope to the zero line, then the absolute value of the amplitude variance of the upper and lower envelopes should also be equal, |σ 2 (W a )-σ 2 (W b )| is the absolute value of the difference between the amplitude variance of the upper envelope signal segment and the amplitude variance of the lower envelope signal segment. The smaller the value, the better the symmetry.

[0091] Furthermore, the maximum amplitude of the upper envelope signal segment in each set of matching signal segments is subtracted from the minimum amplitude of the lower envelope signal segment to obtain the maximum amplitude difference of each matching signal segment: C z =maxW a -minW b

[0092] Where z represents the zth group of matching signal segments, C z Represents the maximum amplitude difference of the zth group of matching signal segments, a and b represent the upper and lower envelope signal segments of this group of matching signal segments, maxW aRepresents the maximum amplitude of the upper envelope signal segment, minW b Represents the minimum amplitude of the lower envelope, maxW a -minW b The difference between the two is the maximum amplitude difference.

[0093] Then, the maximum amplitude difference of all matching signal segments can be obtained.

[0094] Since both the engine vibration signal and the modulation signal generated by the crankshaft speed have good symmetry in each vibration cycle, the purpose of calculating the symmetry of the matching signal segment and the maximum amplitude difference in this embodiment is to subsequently estimate the interference degree of the residual noise through the co-frequency modulation relationship between the vibration signal and the modulation signal, combined with the matching signal segment, that is, the symmetry change of the upper and lower envelope signal segments in the same vibration cycle.

[0095] S005. Calculate the maximum amplitude difference and the Pearson correlation coefficient of symmetry of all matching signal segments to obtain the residual noise interference degree.

[0096] Based on S004, the Pearson correlation coefficient between the two is measured based on the maximum amplitude difference and symmetry of all matching signal segments. That is, when the amplitude difference is large during a single vibration, it means that the injection volume and the acceleration provided by the fuel burst are greater. At this time, the crankshaft speed and vibration signal will also increase, and the modulation signal generated by it on the engine vibration signal will also increase synchronously. The two are in a co-frequency modulation relationship, so the symmetry of the vibration signal characteristic signal segment can be maintained. Even in high-frequency vibration, some interference signals will be submerged, and the vibration characteristics should be more prominent and the symmetry should be better. If not, it is considered to be interference caused by residual noise. The specific implementation method is as follows:

[0097] Among them, B represents the interference degree of residual noise, K z Represents the symmetry of the zth group of matching signal segments, with a total of U groups of matching signal segments. represents the average symmetry of all matching signal segments; C z Represents the maximum amplitude difference of the zth group of matching signal segments, Represents the average maximum amplitude difference of all matching signal segments.

[0098] in That is, the covariance of the symmetry of the matching signal segment and the maximum amplitude difference, σ(K z ),σ(C z ) represent the standard deviation of the symmetry and maximum amplitude difference of all matching signals, It represents the absolute value of the Pearson correlation coefficient between the symmetry and maximum amplitude difference of all matching signal segments. The purpose of adding the absolute value is to constrain the Pearson correlation coefficient between 0 and 1. The closer it is to 1, the stronger the correlation. It represents the degree of interference of the residual noise on the symmetry of the matching signal segment on the second original signal.

[0099] S006. Perform LMD decomposition on the second original signal, use the residual noise interference level as a noise intensity reference value for adaptive Gaussian filter kernel strength, smooth the decomposed PF component signals, and then superimpose and reconstruct the smoothed PF component signals to obtain a denoised engine vibration signal.

[0100] Furthermore, the second original signal is subjected to LMD decomposition. This decomposition process is not the focus of this invention; it is a well-known technique that does not require hyperparameter configuration and will not be described in detail. All PF component signals are obtained. At this point, the decomposed PF component signals need to be smoothed to eliminate residual noise.

[0101] The residual noise interference degree B is used as the standard deviation of the Gaussian filter kernel, and the decomposed PF component signals are smoothed with the filter size set to 7. Generally, the filter strength of the Gaussian filter kernel is controlled by adjusting the standard deviation of the Gaussian filter kernel. In this embodiment, the residual noise interference degree is set to the standard deviation of the Gaussian filter kernel. That is, when the residual noise interference is greater, the Gaussian filter kernel strength is greater, and when the residual noise interference is smaller, the Gaussian filter kernel filtering strength is smaller.

[0102] The Gaussian kernel is used to smooth the signals of each PF component, thereby effectively removing the residual noise in each PF component. Denoising is performed on the modulation component decomposed by LMD, which can eliminate the noise signal that is deeply mixed with the original signal and greatly improve the accuracy of engine monitoring data.

[0103] The original signal is then reconstructed by time-series synchronization and component superposition to obtain a noise-free engine vibration signal. Reconstructing all PF components from the LMD decomposition involves simple addition of the component signals. This method is well-known in LMD decomposition algorithms and will not be detailed here.

[0104] The effective information in the denoised vibration signal is retained and input into the vehicle control unit for analysis and diagnosis by the vehicle fault diagnosis system (ECU). The unmanned vehicle control unit generates feedback instructions based on the diagnosis results, such as stopping, slowing down, reporting errors to the control center, etc.

[0105] This embodiment aims to optimize the data accuracy of the monitoring signals of each operating unit of the automobile received by the automobile control unit, mainly for effectively filtering out the deeply mixed noise when the engine vibration signal is mixed with the modulation signal and noise, and providing reliable data support for the control unit to analyze the vibration signal and obtain accurate automobile operation status results and fault diagnosis results. The specific feature diagnosis of various types of faults of the vibration signal is not the focus of this invention, and there are many types of engine abnormalities and faults, which cannot be analyzed and diagnosed by a single method. However, there are a large number of known technologies and literatures for the current research on the diagnosis of various types of engine faults using engine vibration signals. For example, the author is Mao Zhiwei, and the title of the paper is "Research on Typical Fault Diagnosis and Unstable Working Condition Monitoring and Evaluation Methods for Piston Engines". It discloses a feature extraction method based on the engine vibration signal, and is used to diagnose engine valve abnormalities, connecting rod faults, misfire faults, etc.

[0106] This embodiment also provides a system for monitoring the working status of an unmanned vehicle, including:

[0107] The memory, the processor, and the computer program stored in the memory and executable on the processor, and the computer program executed by the processor, i.e., the processing logic for the engine vibration signal have been described in detail in the above content and will not be described in detail here.

[0108] 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 in the scope of protection of the present invention.

Claims

1. A method for monitoring the working status of an unmanned vehicle, characterized in that: The method comprises the following steps: Acquire the vibration signal of the engine of the unmanned vehicle as the first original signal, and obtain the maximum and minimum weight reference amplitudes in the vibration element according to the residual amplitude of the first original signal, obtain the weight reference amplitude change rate in the vibration element according to the maximum and minimum weight reference amplitudes, obtain the weight reference amplitude of each sampling point position of the vibration element according to the maximum and minimum weight reference amplitudes and the weight reference amplitude change rate, obtain the weight value of each sampling point position of the vibration element according to the weight reference amplitude of each sampling point position of the vibration element, obtain the vibration element according to the weight value, and convolve the vibration element with the first original signal to obtain the second original signal; Obtaining upper and lower envelopes of the second original signal, performing signal segmentation on the upper and lower envelopes to obtain upper and lower envelope signal segment sets, respectively, matching signal segments in the upper and lower envelope signal segment sets to obtain a matching signal segment set, and determining the symmetry of all matching signal segments in the matching signal segment set and the maximum amplitude drop; According to the symmetry of all matching signal segments in the matching signal segment set and the maximum amplitude difference, the Pearson correlation coefficient between the symmetry of all matching signal segments and the maximum amplitude difference is obtained, according to the Pearson correlation coefficient between the symmetry of all matching signal segments and the maximum amplitude difference, the interference degree of the residual noise is obtained, according to the interference degree, an adaptive Gaussian filter kernel is obtained, and the LMD component signal of the second original signal is Gaussian filtered and smoothed by using the adaptive Gaussian filter kernel to obtain a smoothed component signal; The smoothed component signals are reconstructed to obtain an engine vibration signal without noise interference.

2. The method for monitoring the working status of an unmanned vehicle according to claim 1, characterized in that: The specific steps of obtaining the vibration element are as follows: The preset vibrating element length is L; Subtracting the amplitude average from all the amplitudes of the first original signal to obtain a residual amplitude set, and obtaining a screened residual amplitude set by constructing an objective function; Obtain the maximum residual amplitude and the minimum residual amplitude in the screened residual amplitude set, use the maximum residual amplitude and the minimum residual amplitude as the weight reference amplitudes at both ends of the vibration element, obtain the weight reference amplitude change rate of adjacent sampling point positions according to the weight reference amplitudes at both ends of the vibration element and the vibration element length, obtain the weight reference amplitude of each sampling point position according to the weight reference amplitude change rate, and finally normalize all weight reference amplitudes to obtain the weight value of each sampling point position of the vibration element; The vibration element is obtained according to the length of the vibration element and the weight value of each sampling point position of the vibration element.

3. The method for monitoring the working status of an unmanned vehicle according to claim 2, characterized in that: The specific calculation method of filtering out some residual amplitudes by constructing an objective function and concentrating the vibration signal and the modulation signal with small noise interference in the filtered residual amplitude set is as follows: The exhaustive method is used to iteratively screen out any number of residual amplitudes in all residual amplitude sets. The objective function is: Among them, g i Represents the value of the ith residual amplitude in the residual amplitude set remaining during the screening process, represents the average value in the set of residual amplitudes, N represents the number of residual amplitudes in the exhaustive iteration process; g i' represents the value of the i'th residual amplitude in the residual amplitude set when any residual amplitude is screened out, and M represents the total number of amplitudes when no residual amplitude is screened out; As the exhaustive iteration proceeds, when the output value of the objective function E is the minimum, a set of residual amplitudes that are screened out is obtained.

4. The method for monitoring the working status of an unmanned vehicle according to claim 2, characterized in that: The specific steps of normalizing all weight reference amplitudes to obtain the weight values ​​of each sampling point position of the vibration element are as follows: Among them, j v represents the weighted reference amplitude of the vth sampling point position of the vibration element, v represents any sampling point position in the vibration element that does not contain the left and right endpoints, j max 、j min Respectively represent the maximum and minimum values ​​of the screened residual amplitudes, and serve as the weighted reference amplitudes at the left vibration convex end point and the right vibration attenuation cutoff end in the vibration element; in, is the weight reference amplitude change rate of adjacent sampling points in the vibration element; Among them, s represents any sampling point position including the left and right endpoints, j s represents the weight reference amplitude of the sth position, That is, the weighted reference amplitude normalization of the sth sampling point position of the vibration element, ρ s Represents the weight value of the sth sampling point position; The weight values ​​of all sampling point positions of the vibration element are obtained according to the weight reference amplitudes of all sampling point positions.

5. The method for monitoring the working status of an unmanned vehicle according to claim 1, characterized in that: The specific steps of convolving the vibration element with the first original signal to obtain the second original signal are as follows: Take each sampling point of the first original signal as the convolution target point, make the convolution target point coincide with the sampling point position with the highest weight in the vibration element, the sampling point position with the highest weight is the leftmost end point of the vibration element, and align the remaining right sampling point positions with the sampling points on the right side of the convolution target point. Compare the weights of the sampling point positions in the vibration element with the signal amplitudes of the corresponding coincident sampling points. Multiply, add the products of all sampling point positions in the vibration element to obtain the convolution result, replace the amplitude of the original convolution target point with the convolution result, and then convolve all signal sampling points to obtain the second original signal.

6. The method for monitoring the working status of an unmanned vehicle according to claim 1, characterized in that: The specific steps of signal segmentation by the upper and lower envelopes are as follows: Obtain all the maximum and minimum points of the second original signal and use them to fit the upper and lower envelopes respectively, and then obtain all the maximum and minimum points on the upper envelope and all the maximum and minimum points on the lower envelope according to the fitted upper and lower envelopes; For the upper envelope, intercept the signal segments between all the maximum points on the upper envelope and the next adjacent minimum points, and divide the upper envelope into several signal segments between the maximum points and the minimum points, which are recorded as the upper envelope signal segments; For the lower envelope, intercept the signal segments between all the minimum points on the lower envelope and the next adjacent maximum points, and divide the lower envelope into several signal segments between the minimum points and the maximum points, which are recorded as the lower envelope signal segments; Then, the signal segment sets of the upper and lower envelopes are obtained respectively.

7. The method for monitoring the working status of an unmanned vehicle according to claim 1, characterized in that: The specific steps of obtaining the matching signal segment are as follows: The midpoints of all upper and lower envelope signal segments are extracted, and then the timing difference between the midpoints of any two upper and lower envelope signal segments is calculated, and the two upper and lower envelope signal segments with the smallest midpoint timing difference are taken as a group of matching signal segments.

8. The method for monitoring the working status of an unmanned vehicle according to claim 1, characterized in that: The specific method for obtaining the symmetry of the matching signal segment and the maximum amplitude difference is as follows: According to the amplitudes of the upper and lower envelope signal segments of each group of matching signal segments, the amplitude variance of the upper envelope signal segment and the amplitude variance of the lower envelope signal segment are obtained, and according to the absolute value of the difference between the amplitude variance of the upper envelope signal segment and the amplitude variance of the lower envelope signal segment, the symmetry of the matching signal segment is obtained; The maximum amplitude difference of each match is obtained by subtracting the maximum amplitude in the upper envelope signal segment from the minimum amplitude in the lower envelope signal segment in each group of matching signal segments.

9. The method for monitoring the working status of an unmanned vehicle according to claim 1, characterized in that: The interference degree of the residual noise is obtained according to the Pearson correlation coefficient of the symmetry of all matching signal segments and the maximum amplitude difference, and an adaptive Gaussian filter kernel is obtained according to the interference degree. The LMD component signal of the second original signal is Gaussian filtered and smoothed by using the adaptive Gaussian filter kernel to obtain a smoothed component signal. The specific calculation method is as follows: Among them, B represents the interference degree of residual noise, K z Represents the symmetry of the zth group of matching signal segments, with a total of U groups of matching signal segments. represents the average symmetry of all matching signal segments, C z Represents the maximum amplitude difference of the zth group of matching signal segments, represents the average maximum amplitude difference of all matching signal segments, σ(K z ) represents the standard deviation of the symmetry of the matching signal segment, σ(C z ) represents the standard deviation of the maximum amplitude difference of the matching signal segment; is the Pearson correlation coefficient of the symmetry and maximum amplitude difference of all matching signal segments; Perform LMD decomposition on the second original signal to obtain all PF component signals; The interference degree B of the residual noise is taken as the standard deviation obeyed by the Gaussian filter kernel to obtain an adaptive Gaussian filter kernel, and the adaptive Gaussian filter kernel is used to smooth the decomposed PF component signals to obtain the PF component signal after noise reduction; then the original signal is reconstructed in a time synchronization and component superposition manner to obtain an engine vibration signal without noise interference.

10. A system for monitoring the working status of an unmanned vehicle, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images