Windshield Wiper Backswing Angle Measurement Method and System Based on Multimodal Dynamic Feature Fusion

Abstract

The present invention discloses a method and system for measuring the return sweep angle of a windshield wiper based on multi-modal dynamic feature fusion, which relates to the field of aircraft maintenance and detection, and includes: S1. When the windshield wiper is in the working state, the corresponding original signal and auxiliary signal are respectively obtained through a magnetic encoder module and a three-axis vibration accelerometer; S2. The magnetic signal with vibration interference removed is obtained through a preprocessing module; S3. The true angle value is obtained through a compensation module; S4. The motion state classification module determines whether the current motion is normal swinging or abnormal jitter, and then outputs the corresponding effective angle sequence; S5. The detection module determines whether the effective angle sequence is an effective return sweep angle; S6. The evaluation module gives the corresponding final return sweep angle and the credibility flag bit. The present invention provides a method and system for measuring the return sweep angle of a windshield wiper based on multi-modal dynamic feature fusion, which is used to capture the rotation angle of the windshield wiper drive shaft in real time, calculate the return sweep angle in dynamic, and realize high-precision and high-reliability real-time monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of aircraft maintenance and detection. More specifically, the present invention relates to a method and system for measuring the back-sweeping angle of a windshield wiper based on multi-modal dynamic feature fusion, which is applicable to accurately detecting the dynamic angle of windshield wipers of aircraft such as helicopters. Background Art

[0002] Qualified measurement of the back-sweeping angle of the windshield wiper can ensure that when the helicopter performs tasks in rainy or snowy weather, the pilot can clearly visualize the environmental conditions, which helps to anticipate or make adjustments to the air conditions, attitude, etc. of the helicopter in advance. Otherwise, it will have a greater impact on flight safety. In the prior art, the traditional method for measuring the back-sweeping angle of the windshield wiper of a helicopter in the factory is to indirectly obtain the back-sweeping angle by measuring the water marks generated when the windshield wiper sweeps back with an angle ruler. There are problems such as low accuracy and easy wear of the wiper mechanism, that is, there is a lack of tooling and accurate measurement methods. Summary of the Invention

[0003] An object of the present invention is to solve at least the above problems and / or defects and provide at least the advantages described hereinafter.

[0004] To achieve these objects and other advantages of the present invention, a method for measuring the back-sweeping angle of a windshield wiper based on multi-modal dynamic feature fusion is provided, including:

[0005] S1. When the windshield wiper is in the working state, obtain the corresponding original signal and auxiliary signal through a magnetic encoder module and a three-axis vibration accelerometer respectively;

[0006] S2. Perform multi-source signal preprocessing and time-domain alignment on the original signal and the auxiliary signal through a preprocessing module to obtain a magnetic signal with vibration interference removed;

[0007] S3. Perform secondary processing on the magnetic signal through a compensation module to obtain the corresponding true angle value;

[0008] S4. The motion state classification module classifies the motion state of the windshield wiper based on the true angle value to determine whether the current motion is normal swinging or abnormal jitter, and then outputs the corresponding effective angle sequence;

[0009] S5. The detection module performs back-sweeping angle trigger detection based on double-criterion fusion to determine whether the effective angle sequence is a valid back-sweeping angle;

[0010] S6. The evaluation module performs dynamic credibility evaluation based on the valid back-sweeping angle and historical data to give the corresponding final back-sweeping angle and the credibility flag bit.

[0011] Preferably, in S2, the processing flow of the preprocessing module includes:

[0012] S20. Normalize the original signal by the following formula:

[0013]

[0014] In the above formula, is the sine or cosine voltage signal output by the magnetic sensor in the magnetic encoder module, and i = 1, 2, is the sine or cosine voltage signal after normalization processing, is 's long-term statistical mean value, is 's standard deviation;

[0015] S21. Perform vibration compensation alignment on the auxiliary signal by the following formula:

[0016]

[0017] In the above formula, is the vibration transfer coefficient of each axis in the three-axis vibration accelerometer, k is the coordinate axis identifier of the three-axis vibration acceleration, is the synthesis model of vibration displacement noise, is the function of the acceleration of the k-th axis changing with time, t is the current moment variable of signal sampling, is the integration intermediate variable in the time dimension;

[0018] S22. Perform time-domain alignment based on cross-correlation calculation to obtain the corresponding time offset , and then obtain the magnetic signal removing vibration interference through the following formula :

[0019] .

[0020] Preferably, in S3, the working process of the compensation module includes:

[0021] S30. Perform non-linear angle calculation on the magnetic signal by the ellipse fitting method of the following formula:

[0022]

[0023] In the above formula, θ, h , k , a , b are all ellipse parameters, is the original angle after calculation;

[0024] S31. Establish a binary polynomial model of the angle error and temperature through the following formula to complete the Temperature drift compensation:

[0025]

[0026] In the above formula, is the angle after temperature compensation, , , , are the zero-temperature drift constant term, linear temperature coefficient, quadratic temperature coefficient, and cross-coupling temperature coefficient calibrated by the high and low temperature chamber experiment, T is the ambient temperature;

[0027] S32, the true angle value output after temperature drift compensation is characterized by the following formula:

[0028] .

[0029] Preferably, in S4, the working process of the motion state classification module is as follows:

[0030] S40. Decompose the true angle value into j intrinsic mode function components IMF j ( t );

[0031] S41. Analyze IMF j ( t ) through Hilbert transform to separate the corresponding instantaneous frequency and instantaneous amplitude:

[0032] S42. Extract features from the analysis result of S2 to obtain a 12-dimensional feature vector;

[0033] S43. Use the 12-dimensional feature vector as the input of the SVM classifier to judge the current motion state of the windshield wiper and output the corresponding effective angle sequence .

[0034] Preferably, S50. Calculate the central difference angular velocity based on the following formula :

[0035]

[0036] In the above formula, is the time step, is the effective angle value at one step after the current moment t, is the effective angle value at one step before the current moment t;

[0037] S51. When three consecutive sampling points satisfy , it is considered that the zero crossing point satisfies the triggering condition of criterion one;

[0038] S52. Calculate the curvature of the angle sequence by the following formula

[0039]

[0040] In the above formula, is the angular acceleration;

[0041] S53. When k ( t ) > k threshold , it is determined that k ( t ) is the triggering condition for the flyback inflection point to satisfy criterion two, k threshold is the curvature threshold;

[0042] S54. Within the time window τ = 10 ms, if the angle value in triggers both criterion one and criterion two, then it is determined as a valid flyback angle θ c .

[0043] Preferably, in S6, the working process of the evaluation module is as follows:

[0044] S60. Obtain the mean value θ c and the standard deviation μ θ of the most recent N valid flyback angles σ θ through sliding window statistics;

[0045] S61. If the valid flyback angle θ c satisfies the following formula, then start the voting mechanism:

[0046]

[0047] S62. Smoothly output the final flyback angle through the following formula:

[0048]

[0049] In the above formula, λ is the smoothing factor, is the smoothing output value at the previous moment.

[0050] A system, which is applied to the method for measuring the flyback angle of a windshield wiper based on multi-modal dynamic feature fusion, includes:

[0051] A triaxial vibration accelerometer and a magnetic encoder module for signal acquisition;

[0052] A signal processing unit communicatively connected to the triaxial vibration accelerometer and the magnetic encoder module;

[0053] Wherein, the magnetic encoder module includes:

[0054] A permanent magnet assembly coaxially fixed to the end of the wiper drive shaft;

[0055] A magnetic sensor non - contactingly mounted outside the magnetic ring;

[0056] The triaxial vibration accelerometer and the magnetic sensor are arranged on the aircraft through the same portable magnetic adsorption bracket;

[0057] The signal processing unit includes: a pre - processing module, a compensation module, a motion state classification module, a detection module, and an evaluation module.

[0058] The present invention has at least the following beneficial effects:

[0059] Firstly, the present invention simultaneously processes magnetic signals and vibration signals through the pre - processing module, avoiding the defect that single signals are vulnerable to interference and realizing multi - modal fusion;

[0060] Secondly, the present invention performs ellipse fitting through the compensation module to solve the eccentric error caused by the traditional arctangent method installation, so that no mechanical precision adjustment is required during the back - scan angle detection;

[0061] Thirdly, the present invention performs dynamic mode recognition on the wiper through the motion state classification module, that is, effectively distinguishes normal swing and abnormal jitter through the IMF + HHT + SVM classifier to avoid false triggering.

[0062] Fourthly, the present invention overcomes the high false alarm rate of the traditional single - criterion algorithm under vibration through the double - criterion triggering mechanism provided by the detection module.

[0063] Other advantages, objectives, and features of the present invention will be partially reflected by the following description and partially understood by those skilled in the art through the research and practice of the present invention. Brief Description of the Drawings

[0064] Figure 1 It is a schematic diagram of the cooperation between the magnetic adsorption bracket and the aircraft in the detection system of the present invention;

[0065] Figure 2 It is a schematic diagram of the cooperation between the magnetic encoder module and the wiper in the detection system of the present invention (i.e., Figure 1 The enlarged schematic diagram of part A in

[0066] Figure 3Schematic diagram of a wireless terminal in the detection system of the present invention;

[0067] Among them, 1 - permanent magnet assembly, 2 - magnetic sensor, 3 - magnetic adsorption bracket, 4 - wiper drive shaft, 5 - signal processing unit. Detailed implementation manners

[0068] The following further elaborates the present invention with reference to the accompanying drawings, so that those skilled in the art can implement it according to the description in the specification.

[0069] The present invention provides a method for measuring the wiper return sweep angle based on multi-modal dynamic feature fusion. First, the wiper motion features are extracted through empirical mode decomposition and Hilbert spectrum analysis, and the abnormal jitters are filtered by combining with an SVM classifier. Secondly, when determining the return sweep angle, it is necessary to simultaneously meet the dual conditions of the angular velocity sign change and the trajectory curvature extreme value. Finally, the final angle value is fused and output based on the sliding window statistics and the exponentially weighted moving average. In specific implementation, its operation steps include:

[0070] S1. When the wiper is in the working state, the corresponding original signal and auxiliary signal are obtained through the magnetic encoder module and the three-axis vibration accelerometer respectively; in this step, the original signal is the sine / cosine voltage signal (S1(t), S2(t)) output by the TMR sensor in the magnetic encoder module, and the auxiliary signal is the three-axis vibration accelerometer data (A x (t), A y (t), A z (t));

[0071] S2. The multi-source signal preprocessing and time domain alignment are performed on the original signal and the auxiliary signal through the preprocessing module to obtain the magnetic signal with vibration interference removed;

[0072] This step is mainly used for multi-source signal preprocessing and time domain alignment, and its specific processing flow includes:

[0073] S20. The signal normalization is performed through the following formula:

[0074]

[0075] In the above formula, is the sine or cosine voltage signal output by the magnetic sensor in the magnetic encoder module, and i = 1, 2, is the sine or cosine voltage signal after normalization processing, that is, the long-term statistical mean value of the sensor signal ; The sensor signal The standard deviation. In this step, first eliminate the sensor hardware bias (such as zero offset) through signal normalization, making the signal symmetrically distributed around 0, and then compress / expand the signal amplitude to the unit standard deviation to eliminate the sensitivity differences between different sensors;

[0076] S21. As shown in the following formula, calculate the vibration displacement noise model (i.e., perform vibration noise modeling) through the acceleration signal to achieve vibration compensation alignment;

[0077]

[0078] In the above formula, is the vibration transfer coefficient of each axis, which can be calibrated experimentally, is the combined model of vibration displacement noise, that is, convert acceleration to displacement through double integration, characterizing the mechanical offset of the sensor mounting base caused by vibration; k is the coordinate axis identifier of the three-axis vibration acceleration, corresponding to the three orthogonal directions of the helicopter (e.g., x-axis: front-back direction, y-axis: left-right direction, z-axis: up-down direction), to distinguish the vibration components from different directions and ensure independent modeling of the vibration of each axis. k is not a three-dimensional coordinate; t is the variable of the current moment of signal sampling; is the intermediate variable of integration in the time dimension (belonging to dummy variables); The function of the acceleration of the k-th axis (x / y / z axis) changing with time, that is , , ; The function of this step is: in the helicopter windshield wiper measurement system, the vibrations of the engine and rotor will be transmitted to the sensor through the fuselage, resulting in noise in the magnetic encoding signal. This modeling accurately depicts the impact of the complex vibration environment of the helicopter on the sensor through physical mechanism analysis and data-driven methods, thereby separating the vibration noise and the true signal and improving the accuracy of angle measurement;

[0079] S22. Align the noise model and the magnetic signal in the time domain based on the cross-correlation algorithm to obtain the time offset , that is, obtain the time offset through max{CrossCorr( ,N(t))}. The function of this step is: due to differences in the response speed, transmission delay, or sampling moment of different sensors, the collected signals may have a time offset. Time domain alignment ensures that these signals are synchronized in time for accurate analysis of the causal relationship between them. That is, eliminate the time delay between sensors and improve the compensation effect;

[0080] S23. Output the following magnetic signal after removing the vibration interference , and i =1,2:

[0081]

[0082] S3. Perform secondary processing on the magnetic signal through a compensation module to obtain the corresponding true angle value. This step is mainly used for non-linear angle calculation and error compensation, and its specific processing flow includes:

[0083] S30. Perform ellipse fitting (Levenberg-Marquardt iteration) through the following formula to eliminate the eccentricity error of sensor installation:

[0084] That is, through the following formula, a set of discrete two-dimensional data points ( , ) are scattered and fitted into an ellipse. By finding an ellipse equation with the best match, making the ellipse as close as possible to all data points, through this method, the errors caused by mechanical installation and electromagnetic characteristics in the magnetic encoder are eliminated, which belongs to error correction and restores the true angle information:

[0085]

[0086] In the above formula, θ, h , k , a , b are all ellipse parameters, and are solved iteratively through the Levenberg-Marquardt algorithm;

[0087] Further, the original angle after resolution is deduced from the ellipse parameters:

[0088]

[0089] S31. Establish the following binary polynomial model of angle error and temperature to achieve temperature drift compensation:

[0090]

[0091] In the above formula, , , , can be calibrated through experiments in a high and low temperature chamber, is the zero temperature drift constant term, representing the fixed deviation independent of temperature (such as sensor installation error); is the linear temperature coefficient, representing the angle deviation caused by the linear change of the compensated temperature; is the quadratic temperature coefficient, representing the non-linear change of the compensated temperature (such as the quadratic effect of material expansion); is the cross-coupling temperature coefficient, which characterizes the interaction effect between the compensation temperature and the angle value (such as the change of magnetic permeability with angle and temperature); T is the ambient temperature; the function of this step is to cancel or correct the system performance deviation caused by temperature changes, and ensure that the measurement or control results remain stable and accurate in different temperature environments. For example, the magnetic flux of a permanent magnet decreases with the increase of temperature (such as the temperature coefficient of neodymium iron boron magnet is about -0.12% / ℃); the resistance and sensitivity of a TMR sensor change with temperature; the thermal expansion and contraction of metal components cause the change of the distance between the magnetic ring and the sensor; the gain of the amplifier circuit, the reference voltage of AD conversion, etc. are affected by temperature, etc. Without temperature drift compensation, assuming the temperature rises by 50℃, the angle measurement error may exceed 5°;

[0092] S32. The output is:

[0093] S4. The motion state classification module classifies the motion state of the windshield wiper based on the true angle value to determine whether the current motion is normal swing or abnormal jitter, and then outputs the corresponding effective angle sequence;

[0094] The input of this step is , and its operation process is:

[0095] S40. Perform empirical mode decomposition (EMD) on , that is, IMF1, IMF2, IMF3 = EMD( ). This step decomposes the complex signal into a finite number of intrinsic mode functions (abbreviated as IMF), and each IMF represents the fluctuation components of different time scales in the signal. Empirical mode decomposition (EMD) effectively extracts the essential characteristics of the windshield wiper motion by adaptively decomposing the complex signal, providing key data support for the subsequent accurate determination of the return sweep angle;

[0096] Example 1: Taking the true angle value corresponding to the helicopter windshield wiper angle signal as an example, the decomposition process of EMD is as follows:

[0097] (1) Identify the extreme points

[0098] Locate all local maxima and minima in the signal.

[0099] (2) Construct the envelope

[0100] Use cubic spline curves to connect the maximum points (upper envelope) and minimum points (lower envelope) respectively.

[0101] (3) Extract the mean curve

[0102] Calculate the average value of the upper and lower envelopes to obtain the mean curve m1(t).

[0103] (4)Obtain the candidate IMF

[0104] Subtract the mean curve from the original signal: h1(t)= -m1(t). In this step, EMD aims to decompose a complex signal into multiple Intrinsic Mode Functions (IMFs). Each IMF needs to meet two conditions: First, the number of zero-crossing points is balanced with the number of extreme points, that is, the difference does not exceed 1. Second, local symmetry, that is, the mean value of the upper and lower envelopes at any point is zero. To achieve this goal, the high-frequency components and low-frequency trends in the signal need to be gradually separated through an iterative process.

[0105] h1(t) is the candidate IMF obtained after the first iteration. Its acquisition process includes:

[0106] 1. Envelope mean extraction m1(t):

[0107] Calculate the upper envelope (interpolated by local maximum points) and lower envelope (interpolated by local minimum points) of the original signal .

[0108] The envelope mean m1(t) reflects the low-frequency trend component of the signal (such as the baseline shift caused by slow drift or mechanical looseness).

[0109] 2. Candidate IMF generation h1(t):

[0110] Through h1(t)= -m1(t), remove the low-frequency trend and retain the high-frequency oscillation component.

[0111] If h1(t) meets the IMF conditions, it is output as the first IMF; otherwise, it is regarded as a new signal and the above steps are repeated until the conditions are met.

[0112] Repeat steps (1) to (3) for h1(t) until the IMF conditions are met: First, the difference between the number of extreme points and the number of zero-crossing points does not exceed 1. Second, the mean value of the upper and lower envelopes at any point is 0.

[0113] (5)Separate the remaining signal

[0114] Separate the first IMF component c1(t)=h1(t) from the original signal to obtain the remaining signal r1(t)= -c1(t).

[0115] Repeat the above steps for r1(t) until the remaining signal is a monotonic function or a constant value, and then obtain j intrinsic mode function components IMF j ( t )

[0116] S41. Calculate the instantaneous frequency for each IMF component through the Hilbert transform of the following formula f j ( t ) and amplitude A j ( t ) as follows:

[0117] H j ( t ) = Hilbert(IMF j ( t )) ( j = 1, 2, 3)

[0118] Example 2. When detecting the return sweep angle of the windshield wiper, after performing empirical mode decomposition on , complete the judgment of the motion state through the following steps:

[0119] (1) Perform Hilbert transform on the IMF components:

[0120] Transform the intrinsic mode function (IMF) c j (t) obtained by EMD decomposition through the following formula to generate an analytic signal:

[0121] Z j ( t ) = c i ( t ) + j · c j ( t )]

[0122] In the above formula, c j ( t ) is the j-th IMF component (real signal), obtained by EMD decomposition, that is, the intermediate result from EMD decomposition; c j ( t )] is the result (imaginary part) after performing Hilbert transform on c j ( t ). Among them, is the Hilbert transform operator, used to replace the real signal with an orthogonal imaginary component; Z j ( t ) is the analytic signal (complex form), composed of the IMF component and its Hilbert transform, used to further analyze the time-frequency characteristics of the IMF component.​​

[0123] (2)Extract instantaneous features. This step mainly reflects the dynamic changes in the wiper swing rate (such as frequency fluctuations caused by vibrations) through the instantaneous frequency f j ( t ) and characterizes the amplitude stability of the swing angle (abnormal jitter will cause amplitude mutations) through the instantaneous amplitude A j ( t ).

[0124] (3)Fault diagnosis and classification

[0125] a. Feature vector construction, that is, parameters such as the variance of the instantaneous frequency and the energy ratio of the first 3 IMF components are combined to form a 12-dimensional feature vector. The list of the 12-dimensional feature vector is shown in Table 1. IMF1, IMF2, and IMF3 in the table represent the first 3 IMF components:

[0126] Table 1

[0127]

[0128] It should be noted that: 1. The energy ratio is defined as the ratio of the energy of each IMF component to the total energy of the signal, reflecting the contribution weight of different frequency components to the overall movement. Abnormal jitter is usually accompanied by a sudden increase in the energy of high-frequency components (IMF1), also known as the normalized energy ratio;

[0129] 2. The variance of the instantaneous frequency is decomposed into mean, variance, skewness, and kurtosis;

[0130] 3. Permutation entropy is a non-linear feature for measuring complexity.

[0131] b. SVM classification decision, that is, by inputting the feature vector, a pre-trained SVM classifier is used to determine whether the current movement is normal swing (Class 0) or abnormal jitter to be filtered (Class 1).

[0132] To better illustrate the superiority of this step, a comparative description is given through the comparison of the differences between the processing steps of the present invention and the conventional method shown in Table 2:

[0133] Table 2

[0134]

[0135] It should be noted that the comparison of the differences between the processing steps of the present invention and the conventional method has the following innovative advantages:

[0136] 1. Energy proportion calculation: The conventional operation is to only calculate the energy proportion of each IMF component; while the MDFA algorithm (the present invention) introduces a time-varying energy gradient to calculate the energy proportion, which can sensitively capture the energy mutation caused by the mechanical jam of the wiper blade.

[0137] 2. Dynamic weighting of frequency variance: Different weights are assigned to the instantaneous frequency variances of the first three IMF components to suppress false frequency fluctuations caused by high-frequency vibrations.

[0138] Furthermore, from the perspective of classifier design, the processing steps of the present invention are different from those of conventional methods in the following aspects:

[0139] 1.SVM kernel function selection: using composite kernel function to improve the adaptability of classification boundaries;

[0140] 2. Sample imbalance compensation: To address the problem of scarce data on abnormal jitter of helicopter wipers, a 5-fold weight is applied to Class 1 samples during training to prevent the model from being biased towards the majority class.

[0141] S42, feature extraction is performed by the following formula

[0142] Feature Vector=[E1 / E_total,Var(f1),E2 / E_total,...]

[0143] In the above formula, E1 / E_total is the ratio of the first IMF energy to the total energy, Var(f1) is the variance of the first instantaneous frequency, and the subsequent writing is similar, which is used to extract 12-dimensional feature vectors such as the energy ratio and frequency variance of the first three IMF components;

[0144] S43. Use the pre-trained SVM classifier to determine the current motion state, that is, implement dynamic pattern recognition through the following code:

[0145] if SVM(Feature)==0→normal swing

[0146] else → Abnormal jitter (discard data points)

[0147] S44, output the valid angle sequence that excludes abnormal swing data points φ valid ( t ).

[0148] S5, the detection module triggers detection based on the retrace angle fusion of the double criteria to determine whether the valid angle sequence is a valid retrace angle;

[0149] In this step, to reduce the false trigger rate of a single criterion, a dual-criterion fusion method is adopted. The dual-criterion fusion path is: angular velocity detection → time window synchronization → curvature extreme value verification → effective trigger. The input in practical applications is a valid angle sequence φ valid ( t ), and it mainly completes the determination of the effective return sweep angle through the dual-criterion fusion logic, specifically including:

[0150] a. Criterion 1: Detection of angular velocity sign change

[0151] Calculate the central difference angular velocity through the following formula:

[0152]

[0153] In the above formula, is the time step, that is, the time interval between two adjacent sampling moments, which is determined by the sensor sampling frequency; is the current moment t the valid angle value at one step after the current moment, is the current moment t the valid angle value at one step before the current moment. Estimate the instantaneous angular velocity at t moment through the angle difference between the symmetric points before and after the current moment.

[0154] Criterion 1 is detected through the following code zero crossing point, and judge whether Criterion 1 is triggered:

[0155] ω ( t ) = φ ( t + Δ t ) - φ ( t - Δ t )] / (2Δ t )

[0156] if ω ( t ) × ω ( t - 1) < 0 lasts for 3 times → trigger flag F1 = 1.

[0157] b. Criterion 2: Detection of the curvature extreme value of the motion trajectory

[0158] Calculate the angle sequence curvature through the following formula:

[0159]

[0160] In the above formula, α ( t ) is the angular acceleration. When k( t ) > k threshold When it is determined to be a flyback inflection point, that is, criterion 2 is triggered, and its code is as follows:

[0161] k ( t ) = | ωα - ω ' α '| / ( ω ² + α ²)^(3 / 2)

[0162] if k (t) > k _th → trigger flag F2 = 1

[0163] It should be noted that k threshold is the curvature threshold (abbreviation k _th), that is, in the flyback angle detection, the curvature k ( t ) reflects the degree of bending of the wiper blade movement trajectory. When the wiper blade reaches the flyback point, its movement direction changes suddenly (from forward swing to reverse), and at this time the curvature k ( t ) will present a local maximum value. k threshold The role of

[0164] c. Dual-criterion fusion logic: Only when both criteria are triggered within the time window τ = 10ms, it is determined to be a valid flyback angle, and its code is as follows:

[0165] if F1 ∧ F2 within 10ms → θ candidate = φ(t trigger )

[0166] The output of this step is: flyback angle θ c and trigger timestamp t c , it should be noted that the trigger timestamp t c represents the exact time point when the fourth layer determines the flyback angle trigger (such as the millisecond-level timestamp of the system clock). Its core role is: 1. Historical data window alignment: In the sliding window statistics of the fifth layer, it is necessary to determine the time window range according to t c (such as the flyback angle data within the last 10 seconds). 2. Abnormal voting mechanism: When a certain θ cWhen marked as abnormal, it is necessary to check whether the data of adjacent timestamps meets continuity (such as whether the interval of 3 triggers is within a reasonable range).

[0167] S6. The evaluation module performs dynamic credibility evaluation based on the effective flyback angle and historical data to give the corresponding final flyback angle and the credibility flag bit.

[0168] This step is to ensure continuous and stable output of the angle. The credibility closed-loop control path is adopted, and its path is: current angle → historical statistics → anomaly suppression → smooth output. The input in actual application is the flyback angle θ c and historical data, and its processing flow mainly includes:

[0169] S60. Calculate the average value of the most recent N effective flyback angles through sliding window statistics θ c mean μ θ and standard deviation σ θ , and its code is as follows:

[0170] μ θ = mean( θ last_10 ), σ θ = std( θ last_10 )

[0171] S61. Perform outlier suppression. If the current , then start the voting mechanism: it is necessary to detect similar angles continuously for 3 times to confirm the output, and its code is as follows:

[0172] if | θ candidate - μ θ | > 3 σ θ

[0173] Start the voting mechanism (3 consecutive similar values are required);

[0174] S61. Realize smooth output of the angle through the exponential weighted moving average (EWMA) of the following formula:

[0175]

[0176] In the above formula, is the smooth output value of the previous moment. λ is the smoothing factor, and λ = 0.2. λ controls the current measured value θ c and the historical smooth value Weight ratio of:

[0177] 1. When λ approaches 1: The current measurement value dominates, with a fast response speed, but weak noise suppression ability;

[0178] 2. When λ approaches 0: Historical data dominates, with good smoothing effect, but significant response delay.

[0179] The basis for the value of λ (λ = 0.2) is as follows:

[0180] 1. Experimental verification: For the swing frequency of the helicopter wiper (usually 0.52 Hz), it is found through Monte Carlo simulation tests that when λ = 0.2, the response delay of the algorithm to angle mutations ≤ 20 ms (meeting real-time requirements); at the same time, it can suppress more than 90% of high-frequency noise (high-frequency noise is the sound greater than 50 Hz).

[0181] 2. Engineering compromise: In a vibrating environment, it is necessary to balance "smoothness" and "real-time performance", and λ = 0.2 is the empirically optimal value.

[0182] The specific code during processing is as follows:

[0183] θ out(t) = 0.2 × θ candidate + 0.8 × θ out(t-1)

[0184] This step is used to ensure both fast response and smoothness. The output of this step is the final retrace angle θ out and the credibility flag bit (0 / 1).

[0185] It should be noted that the symbol definitions involved in the code in this scheme are as follows:

[0186] (1) → represents the data flow or operation progression

[0187] (2) if represents the conditional judgment branch

[0188] (3) ∧ represents the logical "AND" operation

[0189] (4) θ last_10 represents the cache queue of the last 10 retrace angle data

[0190] This scheme calculates the retrace angle in dynamic through the retrace angle measurement method, dynamically distinguishes the normal swing and abnormal jitter of the wiper, and through multi-layer signal fusion and logical cross-verification, while ensuring the accuracy, it greatly improves the anti-interference ability, accurately determines the retrace angle trigger point, with an error < 0.1°, and realizes high-precision and high-reliability real-time monitoring.

[0191] Furthermore, in the actual application of the present invention, parameters such as the cut-off frequency of the FIR filter and the weight matrix of the SVM classifier of the present invention are dynamically adjusted through online learning and are not written into the firmware code, realizing dynamic self-adaptation of the parameters; at the same time, the present invention adopts multi-level cascaded non-linearity, that is, 3 non-linear operations (elliptical fitting, EMD decomposition, EWMA) are nested in five-layer processing, making it difficult for the algorithm to be reproduced through reverse engineering.

[0192] An aircraft wiper backscanning angle measurement system, as Figures 1 - 3 shown, includes:

[0193] A three-axis vibration accelerometer (not shown) for signal acquisition and a magnetic encoder module;

[0194] A signal processing unit 5 (such as a wireless terminal) communicatively connected to the three-axis vibration accelerometer and the magnetic encoder module;

[0195] Among them, the magnetic encoder module includes:

[0196] A permanent magnet assembly 1 coaxially fixed to the end of the wiper drive shaft 4. The permanent magnet assembly uses a ring-shaped multi-stage magnetic ring (N / S poles are alternately arranged, and the number of pole pairs ≥ 32). The magnetic ring uses a soft magnetic alloy shielding cover to reduce external magnetic interference (such as various communication devices of the helicopter). In this solution, a 32-pole pair magnetic ring and a 14-bit AD chip are used to achieve an angular resolution of 0.05°, reaching sub-angle grading resolution;

[0197] A magnetic sensor 2 mounted on the outside of the magnetic ring in a non-contact manner. The magnetic sensor uses a tunneling magnetoresistance (TMR) sensor array, is mounted on the outside of the magnetic ring in a non-contact manner, is spaced 0.52 mm from the magnetic ring, and the sensor signal line uses a twisted pair shielded wire with a grounding resistance < 1 Ω. In this solution, magnetic encoding is used for non-contact measurement, which can effectively avoid mechanical wear, adapt to the high-frequency vibration environment of the helicopter (> 5g acceleration), and does not require tools throughout the installation process, and the single installation time < 3 min;

[0198] The three-axis vibration accelerometer and the magnetic sensor 2 (TMR sensor) are installed inside or on the surface of the base of the same portable magnetic adsorption bracket 3, and are tightly coupled to the bracket base through a rigid mechanical structure or adhesive (such as metal screws) to ensure that the vibrations received by the accelerometer and the magnetic encoder are completely synchronized.

[0199] The signal processing unit 5 includes: a preprocessing module, a compensation module, a motion state classification module, a detection module, and an evaluation module.

[0200] In this solution, the original signal is obtained through the non-contact coupling of the magnetic sensor and the permanent magnet, and the auxiliary signal is obtained through the three-axis vibration accelerometer to capture the rotation angle of the windshield wiper drive shaft in real time.

[0201] Example:

[0202] The aircraft wiper backscanning angle measurement system was used to conduct tests on a certain type of helicopter. The measured backscanning angle was 61.2° ± 0.3°. The comparison error with the laser interferometer was <0.15°, and the data validity rate reached 99.7%.

[0203] Furthermore, the effectiveness of the application of this system is illustrated by the measurement data in Table 3 below:

[0204] Table 3

[0205]

[0206] It can be seen from the content in Table 3 that the algorithm of MDFA has a smaller error than the traditional algorithm, and the accuracy has been greatly improved.

[0207] The above solution is only an illustration of a preferred example, but it is not limited to this. When implementing the present invention, appropriate substitutions and / or modifications can be made according to the needs of users.

[0208] Although the embodiments of the present invention have been disclosed as above, it is not limited to the applications listed in the specification and embodiments. It can be fully applied to various fields suitable for the present invention. For those familiar with the field, additional modifications can be easily made. Therefore, without departing from the general concept defined by the claims and the equivalent scope, the present invention is not limited to the specific details and the illustrations shown and described here.

Claims

1. A method for measuring the return sweep angle of a windshield wiper based on multi-modal dynamic feature fusion, characterized in that, Including: S1. When the windshield wiper is in the working state, obtain the corresponding original signal and auxiliary signal through the magnetic encoder module and the three-axis vibration accelerometer respectively; S2. Through the preprocessing module, perform multi-source signal preprocessing and time-domain alignment on the original signal and the auxiliary signal to obtain a magnetic signal with vibration interference removed; S3. Through the compensation module, perform secondary processing on the magnetic signal to obtain the corresponding true angle value; S4. The motion state classification module classifies the motion state of the windshield wiper based on the true angle value to determine whether the current motion is normal swinging or abnormal jitter, and then outputs the corresponding effective angle sequence; S5. The detection module performs backscanning angle trigger detection based on double-criterion fusion to determine whether the effective angle sequence is a valid backscanning angle; S6. The evaluation module performs dynamic credibility evaluation based on the effective flyback angle and historical data to give the corresponding final flyback angle and the credibility flag bit; In S3, the working process of the compensation module includes: S30. Perform non-linear angle calculation on the magnetic signal through the ellipse fitting method of the following formula: In the above formula, θ, h , k , a , b are all elliptical parameters, is the original angle after calculation; S31. Establish a binary polynomial model of the angular error and temperature through the following formula to complete the temperature drift compensation for : In the above formula, is the angle after temperature compensation, , , , are the zero-temperature drift constant term, linear temperature coefficient, quadratic temperature coefficient, and cross-coupling temperature coefficient calibrated by the high and low temperature chamber experiment, T is the ambient temperature; The true angle value output after temperature drift compensation Is characterized by the following formula: In S5, the working process of the detection module is: S50. Calculate the central difference angular velocity based on the following formula :[[]]END]] In the above formula, is the time step, is the effective angle value at one step after the current time t, is the effective angle value at one step before the current time t; S51. When three consecutive sampling points satisfy , it is considered that the zero-crossing point meets the triggering condition of criterion one; S52. Calculate the curvature of the angle sequence through the following formula In the above formula, is the angular acceleration; S53. When k ( t )> k threshold , it is determined that k ( t ) meets the trigger condition of the second criterion for the flyback inflection point, k threshold is the curvature threshold; S54. Within the time window τ = 10 ms, if the angle value in causes both Criterion 1 and Criterion 2 to be triggered, then it is determined as a valid flyback angle θ c .

2. The method for measuring the back-sweeping angle of a windshield wiper based on multi-modal dynamic feature fusion according to claim 1, wherein In S2, the processing process of the preprocessing module includes: S20. Perform normalization processing on the original signal through the following formula: In the above formula, is the sine or cosine voltage signal output by the magnetic sensor in the magnetic encoder module, and i = 1, 2, is the sine or cosine voltage signal after normalization processing, is the long-term statistical mean of is the standard deviation of S21. Perform vibration compensation alignment on the auxiliary signal through the following formula: In the above formula, is the vibration transfer coefficient of each axis in the triaxial vibration accelerometer, k is the coordinate axis identifier of the triaxial vibration acceleration, is the synthesis model of the vibration displacement noise, is the function of the acceleration of the k-th axis changing with time, t is the current moment variable of the signal sampling, is the integral intermediate variable in the time dimension; S22. Perform time-domain alignment based on cross-correlation calculation to obtain the corresponding time offset , and then obtain the magnetic signal with vibration interference removed through the following formula : 。 3. The method for measuring the back-sweeping angle of a windshield wiper based on multi-modal dynamic feature fusion according to claim 1, wherein In S4, the working process of the motion state classification module is: S40. Decompose the true angle value into j intrinsic mode function components IMF j ( t ) S41. Parse the IMF through Hilbert transform j ( t ) to separate the corresponding instantaneous frequency and instantaneous amplitude: S42. Extract features from the analysis result of S2 to obtain a 12-dimensional feature vector; S43. Use the 12-dimensional feature vector as the input of the SVM classifier to judge the current motion state of the windshield wiper, so as to output the corresponding valid angle sequence .

4. The method for measuring the back - sweep angle of a windshield wiper based on multi - modal dynamic feature fusion according to claim 1, wherein, In S6, the working process of the evaluation module is: S60. Obtain the mean value θ c of the most recent N valid flyback angles obtained by sliding window statistics μ θ and the standard deviation σ θ ; S61. If the effective flyback angle θ c satisfies the following formula, the voting mechanism is started: S62. The final flyback angle is smoothed by the following formula: The smooth output is as follows: In the above formula, λ is the smoothing factor, is the smoothed output value at the previous moment.

5. A wiper reverse sweep angle measurement system based on multi-modal dynamic feature fusion, which is applied to the wiper reverse sweep angle measurement method based on multi-modal dynamic feature fusion according to any one of claims 1-4, and is characterized in that, Including: A three-axis vibration accelerometer and a magnetic encoder module for signal acquisition; A signal processing unit communicatively connected to the three-axis vibration accelerometer and the magnetic encoder module; Among them, the magnetic encoder module includes: A permanent magnet assembly coaxially fixed to the end of the windshield wiper drive shaft; A magnetic sensor mounted non-contact on the outside of the magnetic ring; The three-axis vibration accelerometer magnetic sensor is arranged on the aircraft through the same portable magnetic suction bracket; The signal processing unit includes: a preprocessing module, a compensation module, a motion state classification module, a detection module, and an evaluation module.

Images