Suspension system sensor fault diagnosis and signal redundancy fault-tolerant method and system

Abstract

The present application relates to the technical field of suspension system detection, and discloses a kind of suspension system sensor fault diagnosis and signal redundancy fault-tolerant method and system, comprising: first, non-sprung acceleration signal and vehicle speed signal are collected, and non-sprung acceleration power spectrum is obtained after pre-processing and power spectrum estimation;Determine the knee point frequency by piecewise fitting, then calculate the speed normalized roll-off frequency and determine the final roll-off frequency, combine vehicle dynamics model transfer function to establish piecewise power law road displacement power spectrum prior, obtain the expected output power spectrum.Based on roll-off frequency, the analysis band is demarcated, the single-channel normalized energy statistics is calculated, and the fault threshold is set according to the statistical distribution and significance level to determine whether the sensor is faulty.If fault, power spectrum prior is used as constraint to carry out frequency domain regularization inversion, obtain road displacement power spectrum estimation and generate target channel virtual output;Finally, according to the difference between the statistics and the fault threshold, output fault-tolerant result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of suspension system detection, more particularly, it relates to a suspension system sensor fault diagnosis and signal redundancy fault-tolerant method and system. BACKGROUND

[0002] In vehicle dynamics, the excitation of the road to the suspension is often expressed in the form of power spectral density. Although the standard gives a statistical definition of the road roughness grade, the spatial structure of the real road often presents a multi-scale feature. Factors such as the paving material, the maximum aggregate particle size, and the patch length of different road sections directly change the energy distribution of the road spatial spectrum. When the vehicle travels at different speeds, the road spatial unevenness is converted into time-frequency excitation, and the position of the energy peak value changes proportionally with the speed. This phenomenon of the joint action of geometry and motion parameters makes the same suspension model present completely different output forms at different speeds or different road sections.

[0003] The traditional suspension sensor fault diagnosis algorithm is mostly based on the Gaussian assumption, which assumes that the statistical properties of the road excitation remain consistent in time. However, in actual situations, the change of the road geometry main scale makes this assumption no longer valid. After the vehicle enters a road section with different materials or repair structures, the frequency domain center of gravity of the suspension acceleration and displacement signals will shift, resulting in the failure of the residual or energy criterion based on the standard power distribution. As a result, the diagnosis system cannot distinguish between environmental changes and sensor misalignment, resulting in frequent false alarms or missed detections.

[0004] When the vehicle speed increases, the interaction frequency between the tire and the road increases; while on the road section with rough paving texture, the proportion of low-frequency energy rises. The superposition of the two makes the vibration input of the suspension system present a spectrum that changes dynamically with speed and geometric scale. If the diagnosis model still evaluates the normality of the sensor with fixed parameters, a systematic deviation will occur between the statistical expectation and the measured signal. This deviation is manifested in the frequency domain as energy distribution misalignment, and in the time domain as amplitude drift. In addition, road repair and local structure discontinuity introduce significant energy concentration points, and this structural excitation will make the output of a single sensor deviate from the model prediction, but it does not mean that the element has failed. The existing algorithm cannot identify this mode difference caused by environmental geometry changes, so when facing dynamic driving conditions across road sections and speeds, the stability and repeatability of the diagnosis system obviously decrease. SUMMARY

[0005] The present application provides a suspension system sensor fault diagnosis and signal redundancy fault-tolerant method and system, which solves the technical problem raised in the background art: how to make the suspension system sensor fault diagnosis and signal redundancy processing reliable and comparable in the case of frequent changes in driving conditions and road spatial scale, so as to avoid systematic misjudgment caused by environmental scale changes.

[0006] In a first aspect, a suspension system sensor fault diagnosis and signal redundancy fault-tolerant method comprises:

[0007] Collecting driving signals, including: unsprung acceleration signals and vehicle speed signals; preprocessing and power spectrum estimation of the driving signals to obtain unsprung acceleration power spectrum;

[0008] Determining the knee point frequency in the unsprung acceleration power spectrum by piecewise fitting, and calculating the speed-normalized roll-off frequency based on the knee point frequency and the vehicle speed signal;

[0009] Determining the roll-off frequency based on the speed-normalized roll-off frequency, establishing a piecewise power-law road displacement power spectrum prior with the roll-off frequency as the boundary, combining the power spectrum prior with the vehicle dynamics model transfer function to obtain the expected output power spectrum;

[0010] Determining the analysis frequency band based on the roll-off frequency, calculating the statistics of single-channel normalized energy in the analysis frequency band, determining the fault threshold based on the statistical distribution and the significance level, and determining that the sensor is in a fault state when the statistics exceeds the fault threshold;

[0011] When it is determined that the sensor is in a fault state, frequency domain regularization inversion is performed with the power spectrum prior as a constraint to obtain road displacement power spectrum estimation, and virtual output of the target channel is generated based on the road displacement power spectrum estimation;

[0012] According to the difference between the statistics and the fault threshold, the measured output and the virtual output of the target channel are implemented weighted soft replacement, and the fault-tolerant result is output.

[0013] In a second aspect, a suspension system sensor fault diagnosis and signal redundancy fault-tolerant system is applied to any one of the suspension system sensor fault diagnosis and signal redundancy fault-tolerant methods, comprising:

[0014] A data acquisition module acquires driving signals, including: unsprung acceleration signals and vehicle speed signals; preprocessing and power spectrum estimation of the driving signals to obtain unsprung acceleration power spectrum;

[0015] A data processing module determines the knee point frequency in the unsprung acceleration power spectrum by piecewise fitting, and calculates the speed-normalized roll-off frequency based on the knee point frequency and the vehicle speed signal;

[0016] An expected output power spectrum module determines the roll-off frequency based on the speed-normalized roll-off frequency, establishes a piecewise power-law road displacement power spectrum prior with the roll-off frequency as the boundary, and combines the power spectrum prior with the vehicle dynamics model transfer function to obtain the expected output power spectrum;

[0017] A fault determination module determines an analysis frequency band based on the roll-off frequency, calculates a statistic of single-channel normalized energy in the analysis frequency band, determines a fault threshold based on a statistical distribution and a significance level, and determines that the sensor is in a fault state when the statistic exceeds the fault threshold;

[0018] A virtual output generation module performs frequency domain regularization inversion with a power spectrum prior as a constraint to obtain a road displacement power spectrum estimate when the sensor is determined to be in the fault state, and generates a virtual output of the target channel based on the road displacement power spectrum estimate;

[0019] A fault-tolerant calculation module performs weighted soft replacement on the measured output and the virtual output of the target channel according to the difference between the statistic and the fault threshold, and outputs a fault-tolerant result.

[0020] The beneficial effects of the present application include: by introducing the speed-normalized roll-off frequency and the segmented power-law road displacement power spectrum prior model, adaptive fault diagnosis and signal redundancy fault tolerance of the suspension system under different vehicle speeds and road conditions are realized, the environment change and sensor failure can be accurately distinguished, and the false alarm and missed detection rates are significantly reduced; when the sensor fails, the virtual output is generated by frequency domain regularization inversion and weighted fusion with the measured signal to realize smooth fault-tolerant output, and the stability and reliability of the system are improved, thereby effectively guaranteeing the safety and performance consistency of the vehicle suspension system under complex working conditions. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flowchart of the present application;

[0022] Figure 2 is a module diagram of the present application. DETAILED DESCRIPTION

[0023] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is merely meant to provide a better understanding of the subject matter described herein and can include changes to the function and arrangement of the elements discussed without departing from the scope of the content of this specification. Various examples can omit, substitute, or add various procedures or components as desired. In addition, features described in relation to some examples can also be combined in other examples.

[0024] Example One: As shown in the figure, a suspension system sensor fault diagnosis and signal redundancy fault tolerance method includes: Figure 1

[0025] Collecting driving signals, including: unsprung acceleration signals and vehicle speed signals; pre-processing and power spectrum estimation of the driving signals to obtain unsprung acceleration power spectrum;

[0026] ​The knee frequency was determined in the unsprung acceleration power spectrum by piecewise fitting, and the speed-normalized roll-off frequency was calculated based on the knee frequency and the vehicle speed signal.

[0027] The roll-off frequency is determined based on the speed-normalized roll-off frequency. A piecewise power-law road displacement power spectrum prior is established with the roll-off frequency as the boundary. The power spectrum prior is combined with the vehicle dynamics model transfer function to obtain the desired output power spectrum.

[0028] The analysis frequency band is determined based on the roll-off frequency. Within the analysis frequency band, the statistics of the normalized energy of a single channel are calculated. The fault threshold is determined based on the statistical distribution and significance level. When the statistics exceed the fault threshold, the sensor is determined to be in a fault state.

[0029] When the fault state is determined, frequency domain regularization inversion is performed with power spectrum prior as constraint to obtain the road displacement power spectrum estimate, and the virtual output of the target channel is generated based on the road displacement power spectrum estimate.

[0030] Based on the difference between the statistics and the fault threshold, a weighted soft substitution is applied to the measured output and the virtual output of the target channel to output the fault-tolerant result.

[0031] In one embodiment of the present invention, the driving signal is preprocessed and its power spectrum is estimated to obtain the unsprung acceleration power spectrum, including:

[0032] With a fixed sampling frequency Acquire discrete signals of unsprung acceleration at discrete times. The value is Unsprung acceleration discrete signals include Discrete moments;

[0033] Calculate the full-length mean of the unsprung acceleration discrete signal. ;in, for Index;

[0034] Subtracting the mean value from the discrete signal of unsprung acceleration along the entire length yields the sub-mean signal. ;

[0035] The mean-removing signal eliminates the DC component (constant offset) in the signal, thus avoiding interference from the DC component in the energy estimation of the low-frequency region of the power spectrum.

[0036] Set the number of discrete points to The sliding step of the Hanning window is... ;

[0037] The Hanning window is set to smooth the transition of the two ends of the signal segment and suppress the spectral leakage (i.e. the false diffusion of frequency energy) caused by signal truncation in the discrete Fourier transform;

[0038] The sliding step length is set to achieve reasonable overlap in signal segmentation and balance the calculation efficiency and statistical stability of spectrum estimation;

[0039] Segmenting the de-meaned signal , the starting sample position of the de-meaned signal in the th segment is

[0040] ;

[0041] Applying the Hanning window to the de-meaned signal in the th segment, the windowed signal of the de-meaned signal in the th segment is :

[0042] ; wherein, is the index of ;

[0043] The Hanning window is applied to strengthen the smoothness of the endpoints of the single-segment signal and completely eliminate the spectral distortion caused by signal truncation in the single-segment signal;

[0044] Discrete Fourier transform is performed on the windowed signal to obtain the transform result :

[0045] ; wherein, is the frequency point index, the physical frequency corresponding to the th frequency point is , the value of is 1 to , is the base number of the natural exponential function, represents the th sample value of the discrete time signal after the Hanning window processing in the

[0046] th segment;

[0047] The window energy normalization coefficient is calculated;

[0048] The window energy normalization coefficient is calculated to correct the influence of the Hanning window on the signal energy, i.e. the energy of the Hanning window is not 1, which will change the original signal energy, and the window energy normalization coefficient can compensate for this energy deviation;

[0049] According to the transformation result and the window energy normalization coefficient, the single-segment power spectrum of the segment signal at the first frequency point is calculated The single-segment power spectrum of the segment signal at the first frequency point is calculated The modulus of the complex spectrum in the frequency domain is converted into power spectrum data capable of directly representing energy, that is, the energy density distribution of each segment signal at different frequency points is obtained by taking the modulus square of the complex spectrum and combining the normalization coefficient;

[0050] The single-segment power spectrum of each segment of the de-meaned signal is summed and averaged to obtain the non-sprung acceleration power spectrum

[0051] The number of segments of the de-meaned signal is represented.

[0052] The Welch power spectrum estimation method is used to reduce the statistical variance (random error) of the single-segment power spectrum, improve the stability and reliability of the spectrum estimation result, and ultimately obtain a power spectrum that can accurately reflect the frequency energy characteristics of the non-sprung acceleration signal.

[0053] In an embodiment of the present application, the knee point frequency is determined in the non-sprung acceleration power spectrum by segment fitting, and the speed normalized roll-off frequency is calculated based on the knee point frequency and the vehicle speed signal, including:

[0054] The physical frequency of each frequency point and the corresponding power spectrum value are respectively subjected to natural logarithmic conversion to obtain the logarithmic frequency of the first frequency point and the logarithmic power ; wherein, is a natural logarithmic function;

[0055] The non-sprung acceleration power spectrum often presents a power law distribution (non-linear exponential relationship between frequency and power) in different frequency bands, and the logarithmic conversion can convert this non-linear relationship into a linear relationship.

[0056] The value of the candidate boundary frequency point is selected to be 1 to ;

[0057] The knee point frequency is the boundary point of the change from one change rule (fractal power law segment) to another rule (roll-off segment), and its position cannot be directly predicted; traversing all frequency points as candidates can ensure that the real knee point position is not missed;

[0058] For the candidate boundary frequency point , a left segment linear model and a right segment linear model are established: ​​​​​​​

[0059] The left segment linear model is ;

[0060] The right segment linear model is ;

[0061] wherein, is an intercept of the left segment linear model, is a slope of the left segment linear model, is an intercept of the right segment linear model, is a slope of the right segment linear model;

[0062] The power spectrum has different change characteristics on both sides of the knee point (for example, the power slowly decays with the increase of frequency on the left side, and the decay is faster on the right side), and the frequency-power correlation on both sides is described by two linear models respectively, so that the actual spectrum shape of each segment can be matched, and the overall deviation caused by fitting two different characteristic spectrum lines with a single model is avoided;

[0063] The left segment residual sum of squares is calculated as :

[0064] ;

[0065] The right segment residual sum of squares is calculated as :

[0066] ;

[0067] The sum of the left segment residual sum of squares and the right segment residual sum of squares is taken as the total residual;

[0068] The residual sum of squares is an index for quantifying the deviation between the predicted value of the model and the actual power spectrum value, and the total residual (the sum of the left segment residual sum of squares and the right segment residual sum of squares) can comprehensively reflect the fitting accuracy of the overall spectrum shape by the two models on both sides of the candidate point;

[0069] The candidate knee point frequency corresponding to the candidate knee point frequency point that minimizes the total residual is taken as the knee point frequency; ;

[0070] According to the logic of minimizing the fitting error, when the total residual is minimized, the two models on both sides corresponding to the candidate point can best fit the actual change rule of the power spectrum, and the knee point frequency is ensured to be the real and accurate demarcation of the transition of the two characteristics;

[0071] The vehicle speed signal at the i-th discrete time is obtained ;

[0072] The average value of the vehicle speed signals at all discrete times is taken as the average vehicle speed ;

[0073] ​​​​The instantaneous vehicle speed fluctuates due to driving operation (such as acceleration and deceleration), and direct use can lead to unstable results; taking the average value of all discrete time vehicle speeds can eliminate the influence of instantaneous fluctuations and obtain the representative vehicle speed of the road section;

[0074] The ratio of the knee frequency to the average vehicle speed is taken as the speed-normalized roll-off frequency .

[0075] The knee frequency is affected by the vehicle speed (on the same road surface, the knee frequency changes proportionally with the vehicle speed), and the ratio to the average vehicle speed can eliminate the interference of the exogenous variable of the vehicle speed, and obtain a pure environmental characteristic quantity determined only by the road surface geometric characteristics (such as aggregate particle size and repair block size). This characteristic quantity has comparability between different vehicle speeds and road sections.

[0076] In one embodiment of the present application, the roll-off frequency is determined based on the speed-normalized roll-off frequency, and a segmented power-law road surface displacement power spectrum prior is established with the roll-off frequency as the boundary. The power spectrum prior is combined with the vehicle dynamics model transfer function to obtain the expected output power spectrum, including:

[0077] The speed-normalized roll-off frequency is based on the roll-off frequency and the average vehicle speed ; ;

[0078] The speed-normalized roll-off frequency is a pure road surface geometric characteristic quantity that strips away the influence of the vehicle speed, and when combined with the average vehicle speed, it can convert the pure environmental characteristic into a specific boundary frequency (roll-off frequency) of the power spectrum in the time domain at the current driving speed.

[0079] A segmented power-law road surface displacement power spectrum prior is constructed , specifically:

[0080] When the physical frequency satisfies , ;

[0081] When the physical frequency satisfies , ;

[0082] wherein is a relative amplitude calibration constant, is a reference frequency, is an autoregressive power index, is an additional attenuation index after roll-off, and ;

[0083] The spatial spectrum of the actual road surface follows different variation laws at different frequency bands (corresponding to different scale road features), that is, the power attenuation characteristics of the low frequency band (large scale road, such as a repair block) and the high frequency band (small scale road, such as aggregate texture) are different. The segmented power law model can fit this actual characteristic, and the additional attenuation exponent can reflect the measured law that the power of the high frequency band decays faster; the relative amplitude calibration constant and the reference frequency can calibrate the amplitude level of the spectrum, so that it matches the specific road level (such as ISO8608 road classification), and ensures that the prior spectrum can truly reflect the frequency energy distribution of the actual road surface;

[0084] The vehicle dynamics model transfer function that defines the displacement of the road surface to the unsprung acceleration is as follows:

[0085] ;

[0086] wherein, is the Laplace complex frequency, is the sprung mass, is the unsprung mass, is the suspension damping coefficient, is the suspension stiffness coefficient, is the tire stiffness coefficient;

[0087] The transfer function is to establish the frequency domain correlation between the road surface displacement input and the unsprung acceleration output, that is, to quantify the dynamic response characteristics of the suspension system (including the sprung mass, the unsprung mass, the damping, the stiffness and the like) to different frequency road excitation, and to clearly show how the road surface displacement excitation is converted into the unsprung acceleration signal through the vibration transmission of the suspension components. The introduction of the Laplace complex frequency can represent the amplitude-frequency and phase-frequency characteristics of the system at different frequencies, avoiding the problem that the time domain analysis is difficult to quantify the dynamic transmission relationship;

[0088] The transfer function is combined with the road surface displacement power spectrum prior to calculate the expected output power spectrum of the unsprung acceleration ; wherein, is the modulus value of the transfer function at .

[0089] The road surface displacement power spectrum prior is the input energy distribution, and the square of the modulus value of the transfer function represents the transmission efficiency of the system to different frequency energy, and the combination of the two can obtain the energy distribution (expected output power spectrum) of the unsprung acceleration under normal working conditions.

[0090] In an embodiment of the present application, the fault threshold is determined based on the statistical distribution and the significance level, and when the statistical quantity exceeds the fault threshold, the sensor is determined to be in a fault state, comprising:

[0091] The degrees of freedom of the chi-square distribution are calculated ; wherein, is the frequency point set Number of mid-frequency points;

[0092] Calculate the fault threshold ;in, Describing the degrees of freedom as The quantile function of the chi-square distribution, Indicates the pre-set significance level. ;

[0093] If the statistics of single-channel normalized energy > Fault Threshold If so, the sensor is determined to be in a faulty state.

[0094] The roll-off frequency is the boundary where the road surface spectrum transitions from the fractal power-law segment to the roll-off segment. By setting upper and lower limits around this frequency using preset coefficients, the analysis frequency band can be accurately covered to ensure that the most significant road surface characteristics are captured. This approach includes the roll-off transition region to capture core environmental information while avoiding the inclusion of low-frequency interference (such as vehicle body steady-state vibration) or high-frequency noise (such as electronic interference).

[0095] By selecting frequency points that fall within the analysis band and forming a dedicated calculation dataset, interference from irrelevant frequency points outside the band can be eliminated. This allows statistical calculations to focus only on frequency components that are strongly correlated with the road environment and sensor response, reducing the dilution of statistical results by invalid data and ensuring that the statistics can truly reflect the sensor's output deviation in the core frequency band.

[0096] The inclusion of a minimum value prevents the denominator from approaching zero and causing abnormal distortion of the statistics when the expected output power spectrum value is too small. Normalization by dividing the measured energy by the expected energy eliminates the influence of differences in the absolute energy value under different road surfaces (such as roughness) and different vehicle speeds, allowing the statistics to quantify only the degree of deviation between the measured output and the normal expectation. The deviation is small under normal operating conditions, but significant when the sensor malfunctions.

[0097] In one embodiment of the present invention, when a fault state is determined, frequency domain regularization inversion is performed with power spectrum prior as a constraint to obtain a road surface displacement power spectrum estimate. Based on the road surface displacement power spectrum estimate, a virtual output of the target channel is generated, including:

[0098] Calculate the first Discrete spectrum estimation of road displacement at each frequency point :

[0099] ;in, Indicates the transfer function in The complex conjugate at the location, For the prior of the road displacement power spectrum at physical frequency The value of , Represents the regularization coefficient;

[0100] Calculate the road displacement power spectrum estimation of the first frequency point :

[0101] ; wherein, is the modulus value of the road displacement discrete frequency spectrum estimation :

[0102] Calculate the target channel frequency domain virtual output of the first frequency point :

[0103] ; wherein, is the transfer function from road displacement to target channel :

[0104] .

[0105] The complex conjugate of the transfer function is used to establish the inverse mapping of the non-sprung acceleration output to the road displacement input, ensuring that the inversion direction conforms to the system dynamics logic; the road displacement power spectrum prior is used as a constraint, which can make the inversion result conform to the frequency energy distribution law of the actual road (avoiding the inversion result deviating from the true road characteristics); the regularization coefficient can suppress the divergence of the inversion result caused by noise or measurement error (such as abnormal increase of high frequency values), so as to obtain the road displacement spectrum that takes into account the true road characteristics and numerical stability.

[0106] The road displacement discrete frequency spectrum is in complex form, and the modulus square can convert the frequency spectrum information in the complex domain into a power spectrum that can directly represent the energy density of road displacement at different frequencies, the purpose being to clarify the excitation intensity distribution of the road at each frequency band, to provide quantitative road excitation input data for generating the target channel virtual output, and to ensure that the virtual output can match the actual road excitation level.

[0107] The transfer function from road displacement to target channel quantifies the dynamic characteristics (transmission efficiency of different frequency excitations) of the road excitation transmitted to the target channel (such as the vehicle body acceleration channel) through the suspension system; combining this transfer function with the road displacement frequency spectrum can simulate the frequency domain signal that the target channel should output under normal road excitation, the purpose being to provide a virtual replacement signal for the fault channel that matches the actual working conditions (road excitation, system characteristics).

[0108] In an embodiment of the present application, according to the difference between the statistical quantity and the fault threshold, the measured output and the virtual output of the target channel are implemented weighted soft replacement, and the fault tolerance result is output, including:

[0109] Calculate the difference ratio of the statistical quantity relative to the fault threshold ; wherein, is a positive function, ​​;

[0110] calculating the weighting coefficient of the target channel ; wherein, represents a coefficient for adjusting the replacement intensity;

[0111] determining the measured discrete spectrum of the target channel and the virtual discrete spectrum ;

[0112] calculating the fusion discrete spectrum of the target channel at the physical frequency ;

[0113] ; wherein, is the fusion spectrum value of the target channel at the physical frequency ;

[0114] the total number of preset frequency points , calculating the time domain fault tolerance output of the target channel at the th discrete moment .

[0115] The positive part function only retains the part of the statistical quantity that exceeds the fault threshold (the difference ratio is 0 when it does not exceed the threshold). The purpose is to quantify the severity of sensor failure, that is, the more the statistical quantity exceeds the threshold, the greater the difference ratio.

[0116] The coefficient for adjusting the replacement intensity can control the intervention of the virtual output. When the difference ratio is large (the fault is severe), the weighting coefficient is small, and the proportion of the virtual output increases. When the difference ratio is small (the fault is slight), the weighting coefficient is large, and the proportion of the measured output increases. The purpose is to dynamically allocate the weights of the measured and virtual signals according to the severity of the fault, avoid the sudden change of the signal caused by the one-size-fits-all hard replacement, and improve the fault tolerance smoothness.

[0117] The measured spectrum may retain some effective information in some frequency bands although it contains faults, and the virtual spectrum is generated based on road excitation and system characteristics (high reliability).

[0118] The measured and virtual spectra are combined according to the weights. The purpose is to correct the fault anomalies in the measured spectrum with virtual data, while retaining the effective components in the measured spectrum that are not affected by faults, to obtain a fusion signal in the frequency domain that not only fits the actual working condition but also eliminates fault interference.

[0119] The frequency domain fusion signal is converted into a time domain signal. The purpose is to output time domain data that can be directly adapted to practical application scenarios such as suspension system control and monitoring (systems usually rely on time domain signals to work), complete the conversion from frequency domain fusion to actual usable fault tolerance signal, and realize the final landing of the fault tolerance function.

[0120] Embodiment two: as Figure 2 ​As shown, a suspension system sensor fault diagnosis and signal redundancy fault-tolerant system is applied to any one of the suspension system sensor fault diagnosis and signal redundancy fault-tolerant method, comprising:

[0121] A data acquisition module acquires driving signals, including: unsprung acceleration signals and vehicle speed signals; the driving signals are preprocessed and power spectrum is estimated to obtain unsprung acceleration power spectrum;

[0122] A data processing module determines the knee point frequency in the unsprung acceleration power spectrum by piecewise fitting, and calculates the speed normalized roll-off frequency based on the knee point frequency and the vehicle speed signal;

[0123] An expected output power spectrum module determines the roll-off frequency based on the speed normalized roll-off frequency, establishes a piecewise power law road displacement power spectrum prior with the roll-off frequency as the boundary, and combines the power spectrum prior with the vehicle dynamics model transfer function to obtain the expected output power spectrum;

[0124] A fault determination module determines the analysis band based on the roll-off frequency, calculates the statistic of single-channel normalized energy in the analysis band, determines the fault threshold based on the statistical distribution and the significance level, and determines that the sensor is in a fault state when the statistic exceeds the fault threshold;

[0125] A virtual output generation module performs frequency domain regularization inversion with the power spectrum prior as the constraint to obtain road displacement power spectrum estimation when it is determined to be in a fault state, and generates the virtual output of the target channel based on the road displacement power spectrum estimation;

[0126] A fault-tolerant calculation module performs weighted soft replacement on the measured output and the virtual output of the target channel according to the difference between the statistic and the fault threshold, and outputs the fault-tolerant result.

[0127] The above describes the embodiments of the present embodiment, but the present embodiment is not limited to the specific embodiments described above, and the specific embodiments described above are only illustrative and not limiting, and those skilled in the art can make many forms under the inspiration of the present embodiment, which are all within the protection scope of the present embodiment.

Claims

1. A method for sensor fault diagnosis and signal redundancy fault tolerance in a suspension system, characterized in that, include: Acquire driving signals, including unsprung acceleration signals and vehicle speed signals; preprocess the driving signals and estimate their power spectrum to obtain the unsprung acceleration power spectrum; The knee frequency was determined in the unsprung acceleration power spectrum by piecewise fitting, and the speed-normalized roll-off frequency was calculated based on the knee frequency and the vehicle speed signal. The roll-off frequency is determined based on the speed-normalized roll-off frequency. A piecewise power-law road displacement power spectrum prior is established with the roll-off frequency as the boundary. The power spectrum prior is combined with the vehicle dynamics model transfer function to obtain the desired output power spectrum. The analysis band is determined based on the roll-off frequency. Within the analysis band, the statistics of the normalized energy of a single channel are calculated, including: Preset analysis frequency band lower limit coefficient Analysis of the upper limit coefficient of the frequency band Calculate and analyze the lower limit of the frequency band Analysis of the upper limit of the frequency band ;in, For roll-off frequency; Construct a set of frequency points , For all satisfied The set of frequency points; Statistical calculation of single-channel normalized energy ;in, , The desired output power spectrum at the physical frequency for unsprung acceleration The value at that location, This is the power spectrum of unsprung acceleration. The fault threshold is determined based on the statistical distribution and significance level. When the statistic exceeds the fault threshold, the sensor is determined to be in a fault state. When the fault state is determined, frequency domain regularization inversion is performed with power spectrum prior as constraint to obtain the road displacement power spectrum estimate, and the virtual output of the target channel is generated based on the road displacement power spectrum estimate. Based on the difference between the statistics and the fault threshold, a weighted soft substitution is applied to the measured output and virtual output of the target channel, and the fault-tolerant output results are output, including: Calculate the proportion of the difference between the statistic and the fault threshold. ;in, It is a positive part function. ; Calculate the weighting coefficients of the target channel ;in, A coefficient indicating the adjustment of the substitution intensity; Determine the measured discrete spectrum of the target channel With virtual discrete spectrum ; Calculate the target channel at physical frequency Fusion discrete spectrum : ;in, For the target channel at physical frequency The fused spectral value; Total number of preset frequency points Calculate the target channel in the 1st... Time-domain fault-tolerant output at discrete moments ;in, This is the fault threshold.

2. The method for sensor fault diagnosis and signal redundancy tolerance in a suspension system according to claim 1, characterized in that, The driving signal is preprocessed and its power spectrum is estimated to obtain the unsprung acceleration power spectrum, including: With a fixed sampling frequency Acquire discrete signals of unsprung acceleration at discrete times. The value is Unsprung acceleration discrete signals include Discrete moments; Calculate the full-length mean of the unsprung acceleration discrete signal. ;in, for The index; Subtracting the mean value from the discrete signal of unsprung acceleration along the entire length yields the sub-mean signal. ; Set the number of discrete points to The sliding step of the Hanning window is... ; Segmenting the mean-reduced signal , No. The starting sample position of the segment mean-removed signal is : ； For the first Applying a Hanning window to the mean signal of segment 1, the first segment The first segment of the mean signal The windowing signal for the mean signal is : ;in, for The index; Perform a Discrete Fourier Transform on the windowed signal to obtain the transform result. : ;in, For frequency point index, the first The physical frequency corresponding to each frequency point , The value ranges from 1 to , The imaginary unit, is the base of the natural exponential function. This represents the q-th sample value of the discrete-time signal after the p-th segment has been processed with a Hanning window; Calculate the window energy normalization coefficient ; Based on the transformation result and the window energy normalization coefficient, calculate the first... The signal segment in the first Single-segment power spectrum at each frequency point , express The modulus; The unsprung acceleration power spectrum is obtained by summing and averaging the power spectra of each segment of the mean-de-mean signal. ; This indicates the number of segments in the mean-removed signal.

3. The method for sensor fault diagnosis and signal redundancy tolerance in a suspension system according to claim 2, characterized in that, The knee frequency is determined by piecewise fitting in the unsprung acceleration power spectrum. Based on the knee frequency and the vehicle speed signal, the speed-normalized roll-off frequency is calculated, including: Perform a natural logarithmic transformation on the physical frequency and the corresponding power spectrum value at each frequency point to obtain the th... Logarithmic frequency at each frequency point With logarithmic power ;in, It is the natural logarithm function; Selecting candidate boundary frequency points , The value ranges from 1 to ; For candidate boundary frequency points Establish the left-segment linear model and the right-segment linear model: The left-side linear model is ; The right-hand linear model is ; in, The intercept of the left-side linear model. The slope of the left segment of the linear model. The intercept of the right-hand linear model. The slope of the right segment of the linear model; Calculate the sum of squares of the left segment residuals : ； Sum of squared residuals on the right segment : ； Will and The sum of these values ​​is used as the total residual; The candidate boundary frequency point that will cause the total residual to be minimized As knee frequency ; Get the Vehicle speed signal at discrete time points ; The average vehicle speed is the average value of all discrete-time vehicle speed signals. ; The ratio of knee frequency to average vehicle speed is used as the speed-normalized roll-off frequency. .

4. The method for sensor fault diagnosis and signal redundancy tolerance in a suspension system according to claim 3, characterized in that, The roll-off frequency is determined based on the speed-normalized roll-off frequency. A piecewise power-law road displacement power spectrum prior is established with the roll-off frequency as the boundary. The power spectrum prior is combined with the vehicle dynamics model transfer function to obtain the desired output power spectrum, including: Based on velocity-normalized roll-off frequency With average vehicle speed Calculate the roll-off frequency ; Constructing a priori pavement displacement power spectrum in piecewise power law form Specifically: When physical frequency satisfy hour, ; When physical frequency satisfy hour, ; in, This is the relative amplitude calibration constant. As the reference frequency, It is a self-affine power exponent. The additional decay exponent after roll-off and ; The vehicle dynamics model transfer function from road surface displacement to unsprung acceleration is defined as follows: ； in, For the Laplace complex frequency, For the sprung mass, For unsprung mass, This is the suspension damping coefficient. This is the suspension stiffness coefficient. This refers to the tire stiffness coefficient. By combining the transfer function with the prior knowledge of the road displacement power spectrum, the expected output power spectrum of unsprung acceleration is calculated. ;in, For the transfer function in The modulus at that point.

5. The method for sensor fault diagnosis and signal redundancy tolerance in a suspension system according to claim 4, characterized in that, The fault threshold is determined based on statistical distribution and significance level. When the statistic exceeds the fault threshold, the sensor is determined to be in a faulty state, including: Calculate the degrees of freedom of the chi-square distribution. ;in, Set of frequency points Number of mid-frequency points; Calculate the fault threshold ;in, Describing the degrees of freedom as The quantile function of the chi-square distribution, Indicates the pre-set significance level. ; If the statistics of single-channel normalized energy > Fault Threshold If so, the sensor is determined to be in a faulty state.

6. The method for sensor fault diagnosis and signal redundancy tolerance in a suspension system according to claim 5, characterized in that, When a fault condition is identified, frequency domain regularization inversion is performed with power spectrum prior as a constraint to obtain a road surface displacement power spectrum estimate. Based on the road surface displacement power spectrum estimate, a virtual output of the target channel is generated, including: Calculate the first Discrete spectrum estimation of road displacement at each frequency point : ;in, Indicates the transfer function in Complex conjugate at the location, For the prior of the road displacement power spectrum at physical frequency The value of , Represents the regularization coefficient; Calculate the first Road displacement power spectrum estimation at each frequency point : ;in, Represents the discrete spectrum estimation of road displacement The modulus; Calculate the first Virtual output of target channel frequency domain at each frequency point : ;in, The transfer function from road surface displacement to the target channel : 。 7. A suspension system sensor fault diagnosis and signal redundancy fault tolerance system, applied in the suspension system sensor fault diagnosis and signal redundancy fault tolerance method according to any one of claims 1-6, characterized in that, include: The data acquisition module collects driving signals, including unsprung acceleration signals and vehicle speed signals; it preprocesses and estimates the power spectrum of the driving signals to obtain the unsprung acceleration power spectrum. The data processing module determines the knee frequency in the unsprung acceleration power spectrum through piecewise fitting, and calculates the speed-normalized roll-off frequency based on the knee frequency and the vehicle speed signal. The desired output power spectrum module determines the roll-off frequency based on the speed-normalized roll-off frequency, establishes a piecewise power-law road displacement power spectrum prior with the roll-off frequency as the boundary, and combines the power spectrum prior with the vehicle dynamics model transfer function to obtain the desired output power spectrum. The fault determination module determines the analysis frequency band based on the roll-off frequency and calculates the statistics of the normalized energy of a single channel within the analysis frequency band, including: Preset analysis frequency band lower limit coefficient Analysis of the upper limit coefficient of the frequency band Calculate and analyze the lower limit of the frequency band Analysis of the upper limit of the frequency band ;in, For roll-off frequency; Construct a set of frequency points , For all satisfied The set of frequency points; Statistical calculation of single-channel normalized energy ;in, , The desired output power spectrum at the physical frequency for unsprung acceleration The value at that location, This is the power spectrum of unsprung acceleration. The fault threshold is determined based on the statistical distribution and significance level. When the statistic exceeds the fault threshold, the sensor is determined to be in a fault state. When a fault condition is determined, the virtual output generation module performs frequency domain regularization inversion with power spectrum prior as a constraint to obtain the road displacement power spectrum estimate, and generates the virtual output of the target channel based on the road displacement power spectrum estimate. The fault tolerance calculation module, based on the difference between statistics and fault thresholds, performs weighted soft substitution on the measured and virtual outputs of the target channel, and outputs the fault tolerance results, including: Calculate the proportion of the difference between the statistic and the fault threshold. ;in, It is a positive part function. ; Calculate the weighting coefficients of the target channel ;in, A coefficient indicating the adjustment of the substitution intensity; Determine the measured discrete spectrum of the target channel With virtual discrete spectrum ; Calculate the target channel at physical frequency Fusion discrete spectrum : ;in, For the target channel at physical frequency The fused spectral value; Total number of preset frequency points Calculate the target channel in the 1st... Time-domain fault-tolerant output at discrete moments ;in, This is the fault threshold.

Images