Vehicle wheel-side redundant brake-by-wire system and method based on EMB system

Abstract

The application provides a vehicle wheel edge redundant line control brake system and method based on an EMB system, relates to the technical field of vehicle wheel edge redundant line control brake, and adopts different frequencies to collect main and standby channel torque, wheel speed and longitudinal acceleration signals in parallel, and adopts time stamp and sliding window alignment technology to ensure data synchronization; secondly, the application is based on Kalman filter residual hysteresis determination and combines acceleration correlation to calculate main and standby channel credibility weight, realizes rapid and accurate fault identification, fuses torque according to the weight, and predicts the next period demand torque by means of a short time sequence model; then, the application estimates the road adhesion coefficient on line, modifies the dynamic abnormal threshold and fusion weight in real time to adapt to different road conditions; finally, the application continuously monitors the main road credibility, and completes the main and standby switching in milliseconds through a smooth transition strategy, so that the brake output is continuous and smooth.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle wheel edge redundant brake-by-wire technology, in particular to a vehicle wheel edge redundant brake-by-wire system and method based on an EMB system. BACKGROUND

[0002] As the core development direction of brake-by-wire technology, EMB (electronic mechanical brake system) directly drives brake calipers through motors to achieve precise torque control, and has important application value in new energy vehicles and automatic driving fields. The wheel edge redundancy architecture improves system reliability through main and backup actuators, which is a key design to meet the functional safety requirements of ISO 26262 ASIL-D level. However, in the process of multi-channel cooperative control, three technical problems of real-time anomaly detection of main channel signals, timing synchronization of multi-source heterogeneous data, and smoothness control of redundancy switching have not been effectively solved.

[0003] The prior art usually realizes wheel edge redundancy braking in two ways: one is hardware-level redundancy, that is, a mechanical switch or a static switching valve is used to switch to the backup channel when the main channel fails; the other is software-level redundancy, that is, after the deviation of the main channel feedback signal or the communication anomaly is monitored in real time, the vehicle controller switches the output to the backup channel. However, both of these two ways have limitations: hardware switching often accompanies mechanical hysteresis and impact, resulting in a sudden drop in instantaneous braking force; software switching relies on simple threshold judgment and fixed delay strategy, and cannot balance fault prediction, smooth transition and multi-channel signal fusion, making it difficult to meet the needs of millisecond-level response and high-precision distribution.

[0004] The existing scheme still has shortcomings in brake channel failure detection, main and backup channel data reliability evaluation, and seamless switching smoothness, and cannot guarantee continuous and stable braking output under variable road conditions and high dynamic conditions.

[0005] The above information disclosed in the background section is only used to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide a vehicle wheel edge redundant brake-by-wire system and method based on an EMB system to solve the problems raised in the background.

[0007] To achieve the above purpose, the present application provides the following technical solutions:

[0008] A vehicle wheel edge redundant brake-by-wire system based on an EMB system, comprising the following specific steps:

[0009] A signal acquisition module for asynchronously acquiring and aligning the main and backup channel motor torque signals, wheel speed signals and vehicle body longitudinal acceleration signals of the vehicle to be detected;

[0010] anomaly detection module, configured to predict a main passage motor torque signal theoretical value based on a Kalman filter, calculate a residual value of the theoretical value and a measured value thereof, determine an anomaly when the residual value exceeds a dynamic anomaly threshold value in a sliding window, and calculate a reliability weight of the main passage and the standby passage according to a statistical correlation of the residual value in the same window and a vehicle body longitudinal acceleration signal;

[0011] an adjustment parameter module, configured to estimate a road surface friction adhesion coefficient by using a recursive least square algorithm through a wheel speed signal and a vehicle body longitudinal acceleration signal, correct the dynamic anomaly threshold value according to the adhesion coefficient, and adjust a fusion torque distribution coefficient in the redundancy switching module;

[0012] a fusion calculation module, configured to perform weighted fusion on the main passage motor torque signal and the standby passage motor torque signal according to the reliability weights of the main passage and the standby passage, generate a fusion torque, and output a required torque value of a next control period through a short-time sequence prediction model by taking data signals in a sliding window as input;

[0013] a redundancy switching module, configured to execute a full required torque value by the main passage when the reliability weight of the main passage is higher than a safety threshold value, start switching when the reliability weight of the main passage is lower than the safety threshold value, gradually reduce a distribution proportion of the main passage motor torque according to a fusion torque distribution coefficient, synchronously increase a distribution proportion of the standby passage motor torque, monitor a fluctuation intensity of the fusion torque in real time, and dynamically adjust a transition time window to suppress mutation.

[0014] Further, the signal acquisition module asynchronously receives the following signals through a vehicle CAN / CAN-FD bus:

[0015] a main passage motor torque signal fed back by a wheel-side main actuator and a standby passage motor torque signal fed back by a wheel-side standby actuator, both of which have a sampling rate of 500 Hz, a wheel speed signal output by a wheel speed sensor, which has a sampling rate of 100 Hz, and a vehicle body longitudinal acceleration signal output by a vehicle body inertia unit, which has a sampling rate of 1 kHz;

[0016] After the asynchronously acquired data is time-stamped with an ECU local time, the data is stored in a bidirectional circular buffer that can buffer at least W continuous control period data, and time sequence alignment is performed through a sliding window mechanism: the sliding window contains W continuous control periods, each control period has a length of , the window step length is one control period length, and the total window length covers at least a maximum braking response delay of the system;

[0017] The timing alignment mechanism takes the vehicle body longitudinal acceleration signal as the reference time axis, adopts piecewise cubic Hermite interpolation for the wheel speed signal, and adopts linear interpolation for the main and standby path motor torque signals. After interpolation, the timestamp synchronization error of each signal is less than or equal to 5μs. After interpolation, the abnormality detection module reads the aligned signal data in the sliding window.

[0018] Further, the abnormality detection module predicts the theoretical value of the main path motor torque signal based on the Kalman filter, and calculates the difference between the measured value and the theoretical value as a residual value:

[0019] Let i represent the wheel number, and k be the index of the control period in the sliding window. The residual value of each wheel in the abnormality detection module at the kth control period is calculated according to the time sequence. The calculation formula is:

[0020]

[0021] Wherein, is the Kalman filter theoretical value of the main path motor torque signal of the i-th wheel of the vehicle at the k-th control period, is the corresponding measured torque value, and the difference between the two is the residual value ; The average residual value in the sliding window is: ;

[0022] Only when the average residual value of the main path of the i-th wheel exceeds the dynamic abnormal threshold in the sliding window, it is determined that the main path signal of the wheel is abnormal, otherwise the main path signal is normal.

[0023] Further, according to the residual value and the vehicle body longitudinal acceleration signal in the same time window, the logic for calculating the confidence weight of the main and standby paths using statistical correlation analysis is as follows:

[0024] The calculation formula of the confidence weight of the i-th wheel main path in the k-th control period is as follows:

[0025]

[0026] Wherein, is the residual sensitivity coefficient at the k-th control period, is the Pearson correlation coefficient of the residual sequence and the vehicle body longitudinal acceleration sequence in the same window;

[0027] The calculation formula of the confidence weight of the i-th wheel standby path in the k-th control period is as follows: ;

[0028] ​​Average confidence weight within the sliding window: Similarly .

[0029] Furthermore, the fusion calculation module calculates the fusion torque of the i-th wheel during the k-th control cycle. Calculate using the following formula:

[0030]

[0031] in, and These are the measured torque values ​​of the main path and backup path of the i-th wheel during the k-th control cycle, respectively.

[0032] The fused torque sequence of W control cycles within the sliding window, along with the vehicle longitudinal acceleration signal sequence and wheel speed signal sequence, are input into the LSTM neural network model to predict the required torque value for the next control cycle.

[0033] Furthermore, the adjustment parameter module, based on wheel speed signals and vehicle longitudinal acceleration signals, identifies the road surface friction coefficient online using a recursive least squares algorithm, and corrects the dynamic anomaly threshold and fused torque distribution coefficient according to the following formula:

[0034] The correction formula for the dynamic anomaly threshold is:

[0035]

[0036] in, Based on the threshold, For road surface sensitivity coefficient, The road surface friction adhesion coefficient is estimated based on the recursive least squares method in the k-th control cycle. For the first Dynamic anomaly threshold for each control cycle;

[0037] The adjustment of the fusion torque distribution coefficient satisfies:

[0038]

[0039] in, As the benchmark allocation coefficient, As an attenuation factor, it automatically reduces the proportion of the main path when the road surface has low adhesion. For the predicted first The fusion torque distribution coefficient for each control cycle For the first The road surface friction adhesion coefficient for each control cycle;

[0040] In the calculation of the credibility weight, the dynamic adjustment of the residual sensitivity coefficient is as follows: , is the residual sensitivity coefficient of the first control cycle, and the smaller the dynamic abnormal threshold value is, the higher the residual sensitivity is.

[0041] Further, the redundancy switching module monitors the credibility weight of the main channel and calculates the average credibility weight thereof, and when the average credibility weight is lower than a preset safety threshold, a transition stage is entered.

[0042] Let the total transition duration be T, The relative progress of the switching process is defined as , is a time variable, and an exponential smoothing transition function is defined as :

[0043]

[0044] wherein, is a smoothing rate coefficient, , is the current transition progress, indicates the start of switching, indicates the completion of switching;

[0045] In the transition period, the demand torque value is proportionally distributed between the main channel and the backup channel, wherein the main channel torque distribution proportion is , the backup channel torque distribution proportion is , so that the main channel torque proportion gradually decays from to 0, and the backup channel proportion synchronously rises from 0 to ;

[0046] The fusion torque is monitored in real time, the standard deviation of the fusion torque sequence is calculated to represent the fluctuation intensity, and is increased or decreased according to the fluctuation intensity to balance the switching speed and smoothness and suppress mutation;

[0047] When the credibility weight of the main channel recovers to the safety threshold and above, the redundancy switching module switches the braking torque distribution back to the main channel with an opposite exponential smoothing transition function and the same , realizing bidirectional seamless switching.

[0048] The application further provides a vehicle wheel edge redundancy line control braking method based on an EMB system, which is used for executing the vehicle wheel edge redundancy line control braking system based on the EMB system.

[0049] Step 1: asynchronously collecting and aligning the main channel motor torque signal, the wheel speed signal and the vehicle body longitudinal acceleration signal of the vehicle to be detected;

[0050] ​Step 2: The main passage motor torque signal theoretical value is predicted based on the Kalman filter, the residual value of the theoretical value and the measured value is calculated, when the residual value exceeds the dynamic abnormal threshold in the sliding window, the abnormality is determined, and the reliability weight of the main passage and the standby passage is calculated according to the statistical correlation of the residual value in the same window and the vehicle body longitudinal acceleration signal;

[0051] Step 3: The road surface friction adhesion coefficient is estimated by using the recursive least square algorithm through the wheel speed signal and the vehicle body longitudinal acceleration signal, the dynamic abnormal threshold is corrected according to the adhesion coefficient, and the fusion torque distribution coefficient in the redundancy switching module is adjusted;

[0052] Step 4: The main passage and the standby passage motor torque signal is weighted and fused according to the reliability weight of the main passage and the standby passage, the fusion torque is generated, the data signal in the current sliding window is taken as the input, and the demand torque value in the next control period is output through the short sequence prediction model;

[0053] Step 5: When the main passage reliability is higher than the safety threshold, the main passage executes the full demand torque value, when it is lower than the safety threshold, the switching is started, the distribution proportion of the main passage motor torque is gradually reduced according to the fusion torque distribution coefficient, the distribution proportion of the standby passage motor torque is synchronously improved, the fluctuation intensity of the fusion torque is monitored in real time, and the transition time window is dynamically adjusted to suppress the mutation.

[0054] Compared with the prior art, the beneficial effects of the present application are:

[0055] The present application effectively eliminates the acquisition error caused by the time sequence deviation of the signal through the high-precision synchronous acquisition and alignment of multiple source signals, ensures the accuracy of the pedal force signal, the main passage torque signal, the standby passage torque signal, the wheel speed and the acceleration signal under the same time reference, the double abnormality detection based on the Kalman filter residual hysteresis judgment and the statistical correlation evaluation not only shortens the fault detection delay, but also avoids the misjudgment caused by the transient interference, the residual and mutual information weighted fusion is used to generate the fusion torque, and the LSTM short sequence prediction model is combined to output the demand torque value in the next control period in advance, so that the brake force distribution can accurately track the driver's intention and predict the future dynamic demand, and the torque distribution precision is greatly improved;

[0056] The road surface friction adhesion coefficient identified online by the recursive least square algorithm is used to dynamically correct the abnormality judgment threshold and the fusion torque distribution coefficient, and the adaptive control in various road conditions is realized; when the main passage fails, the exponential smoothing transition function and the adjustable transition time window enable the main passage and the standby passage to switch in a short time, the fusion torque fluctuation amplitude is extremely small, the switching impact and the jerk are eliminated, and the above technical features cooperate to not only solve the lag and mutation problems of the existing mechanical or software switching, but also significantly improve the safety reliability and passenger comfort of the redundant brake-by-wire system. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 The whole method flowchart of the application;

[0058] Figure 2 The residual value-Pearson correlation coefficient-main path reliability weight 3D color mapping surface chart of the application;

[0059] Figure 3 The residual value-main path reliability weight vertical chart of the application;

[0060] Figure 4 The residual value-residual sensitivity coefficient fitting curve chart of the application;

[0061] Figure 5 The residual value-main path reliability weight vertical chart of the application;

[0062] Figure 6 The whole system module flowchart of the application. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical scheme and advantages of the application more clear, the application is further described in detail below in combination with specific embodiments.

[0064] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the application should be understood as the general meaning understood by those skilled in the art to which the application belongs. The "first", "second" and similar words used in the application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0065] Embodiment:

[0066] Please refer to Figures 1-5 The application provides a technical scheme:

[0067] A vehicle wheel edge redundant line control brake system based on an EMB system, the specific steps comprising:

[0068] The signal acquisition module is used for asynchronously acquiring and aligning the main and standby path motor torque signals, wheel speed signals and vehicle body longitudinal acceleration signals of the vehicle to be detected.

[0069] The signal acquisition module asynchronously receives the following signals through the vehicle CAN / CAN-FD bus:

[0070] The main channel motor torque signal fed back through the wheel-side main actuator and the backup channel motor torque signal fed back through the wheel-side backup actuator both have a sampling rate of 500 Hz, the wheel speed signal output by the wheel speed sensor has a sampling rate of 100 Hz, and the vehicle body longitudinal acceleration signal output by the vehicle body inertia unit has a sampling rate of 1 kHz;

[0071] The system uses a 500 Hz sampling rate to collect the main and backup channel torque signals in real time, synchronizes with the EMB motor control period, and ensures accurate capture of dynamic responses such as stall and overload; at the same time, the ABS wheel speed sensor standard frequency is matched to collect the wheel speed signal, which optimizes the calculation load while ensuring the accuracy of the slip rate detection; and a 20-50 Hz high-frequency sampling channel is set to capture the vehicle body longitudinal acceleration, which is used to identify the braking pitch oscillation characteristics and provide data support for road identification and fault diagnosis;

[0072] After the asynchronously collected data is timestamped with the ECU local time, it is stored in a bidirectional circular buffer that can cache at least W consecutive control period data, and time alignment is performed through a sliding window mechanism: the sliding window contains W consecutive control periods, each control period has a duration of , and the window step is one control period long, ensuring that the total window duration covers at least the maximum braking response delay of the system ;

[0073] The time alignment mechanism uses the vehicle body longitudinal acceleration signal as the reference time axis, uses piecewise cubic Hermite interpolation for the wheel speed signal, and uses linear interpolation for the main and backup channel motor torque signals. After interpolation, the timestamp synchronization error of each signal is less than or equal to 5μs. After interpolation, the abnormality detection module reads the aligned signal data in the sliding window;

[0074] Linear interpolation only requires one addition and one multiplication (proportion calculation), which has a very small amount of calculation. This is crucial for ECU systems that require high real-time performance, especially when processing main and backup channel signals, which can save a lot of calculation resources. Interpolation uses existing discrete sampling point data close in time to the target time to estimate the signal value at the target time, i.e., a specific time on the reference acceleration signal time axis;

[0075] Based on the relatively flat characteristics of the torque signal, a straight line is simply connected between the two adjacent original sampling points, and then the torque values of the two endpoints are mixed in proportion according to the position of the target time on the straight line to obtain the torque estimate at the target time;

[0076] Given the fact that torque signal usually varies smoothly, in a short sampling interval (relative to signal variation rate), linear interpolation between adjacent samples has been able to provide sufficiently accurate intermediate value estimation to meet the requirement of less than or equal to 5μs timestamp alignment and subsequent anomaly detection, in the area of smooth variation, the benefit of higher order interpolation is small;

[0077] Motor torque signal can reflect the size of the motor output torque, torque command usually issued by the controller, its change relative to the mechanical response of the wheel speed is usually more smooth, more linear, especially in the control cycle, the physical inertia of the motor itself and the control response also make it not like the wheel speed as instant dramatic change; torque command update usually occurs at the control cycle boundary, may remain constant or slowly change according to the instruction in the control cycle;

[0078] For the fast changing, nonlinear strong wheel speed signal, use more complex but more accurate, can reflect the rate of piecewise cubic hermite interpolation, for the relatively smooth motor torque signal, use simple and efficient linear interpolation, the target time point of all interpolation operations comes from the timestamp of the vehicle body longitudinal acceleration signal, ensures that all signals are finally unified to the same time reference;

[0079] Bidirectional circular buffer and sliding window mechanism to ensure that the latest, continuous W cycle data are processed, and the total time covers the key delay , so that the aligned data can be effectively used to detect the delay characteristics of the anomaly.

[0080] The anomaly detection module is configured to predict a theoretical value of the motor torque signal of the main channel based on a Kalman filter, calculate a residual value of the theoretical value and a measured value of the motor torque signal, determine an anomaly when the residual value exceeds a dynamic anomaly threshold in a sliding window, and calculate a reliability weight of the main channel and the backup channel according to a statistical correlation between the residual value in the same window and the vehicle body longitudinal acceleration signal.

[0081] The anomaly detection module predicts a theoretical value of the motor torque signal of the main channel based on a Kalman filter, and calculates the difference between the theoretical value and the measured value as a residual value:

[0082] Let i represent the wheel number, k be the index of the control cycle in the sliding window, and the index be sequentially indexed in time order. The residual value of each wheel in the kth control cycle in the anomaly detection module is calculated according to the following formula:

[0083]

[0084] wherein, is the Kalman filter theoretical value of the motor torque signal of the i th wheel of the vehicle in the k th control cycle, ​For the corresponding measured torque value, the difference between the two is the residual value ; the average residual value within the sliding window: ;

[0085] represents the deviation between the theoretical prediction value of the main path motor torque on the i-th wheel in the k-th control cycle and the actual measured value , which directly quantifies the inconsistency between the main path torque sensor reading and the torque that should be generated based on the Kalman filter prediction under normal fault-free conditions ;

[0086] represents that the measured value and the model prediction value are highly consistent, which is usually a strong indication that the system is working normally, the sensor is accurate, and the model matches the actual working condition; The larger the value, the more likely it indicates that: a. the main path torque sensor is faulty, such as drift, jam, excessive noise, b. the actual output torque of the motor or related drive circuit does not match the instruction or model expectation, such as motor winding short circuit, open circuit, power device failure, c. the vehicle model itself has deviation, especially in extreme or unmodeled working conditions, d. the Kalman filter parameter setting is not appropriate, such as inaccurate process noise or measurement noise covariance;

[0087] The larger the value, the more likely it indicates that the actual measured torque is significantly smaller than the model prediction value , for example, sensor failure causes the reading to be low, or the motor actually outputs insufficient torque, The larger the absolute value of the negative value, the more likely it indicates that the actual measured torque is significantly larger than the model prediction value , for example, sensor failure causes the reading to be high, or the motor unexpectedly outputs excessive torque, such as short circuit causing locked-rotor torque; in general, The larger the value, the higher the likelihood of abnormal main path torque signal, and the greater the deviation between the system operating state and the expected model;

[0088] When is fixed, increasing will cause to increase, when is fixed, increasing will cause to decrease,

[0089] The size and sign of If the dynamic abnormality threshold is exceeded within the sliding window, the main path signal of the wheel is determined to be abnormal; otherwise, the main path signal is maintained to be normal.

[0090] This average value determination avoids false alarms caused by individual instantaneous spikes, while ensuring that the detected deviations are continuous and can truly reflect the potential impact on the braking process.

[0091] This dynamic anomaly threshold is not fixed but dynamically adjusted. It may change in real time based on factors such as vehicle speed, load, road conditions, and estimated signal-to-noise levels. The purpose is to adapt the anomaly detection sensitivity to different operating conditions. For example, under low-adhesion road surfaces or severe dynamic conditions, the model prediction error, i.e. The threshold value may be relatively high, in which case increasing the threshold can avoid false alarms; under stable operating conditions, the threshold can be decreased to improve detection sensitivity.

[0092] The logic for calculating the reliability weights of the primary and backup paths using statistical correlation analysis based on the residual values ​​and longitudinal acceleration signals of the vehicle body within the same time window is as follows:

[0093] The reliability weight of the i-th wheel main path in the k-th control cycle The calculation formula is as follows:

[0094]

[0095] in, Let be the residual sensitivity coefficient in the k-th control period, used to adjust the strength of the influence of the residual value on the confidence weight. The Pearson correlation coefficient between the residual sequence and the vehicle longitudinal acceleration sequence within the same window;

[0096] The specific practical meaning and the situation it reflects: It is a value between 0 and 1, representing the system's level of confidence in the torque signal of the i-th wheel's main path during the k-th control cycle, and it incorporates the current residual magnitude. Statistical correlation between residuals and vehicle longitudinal acceleration within the sliding window Two factors are used to dynamically assess the reliability of the main pathway signal;

[0097] A value close to 1 indicates that the system highly trusts the main path signal, which occurs when the residual is very small. The residual is small and highly correlated with the vehicle's acceleration, i.e. When the value is close to 1; A value close to 0.5 indicates that the system has neutral or moderate confidence in the main path signal. This typically occurs when: a. the residuals are small but the correlation is also low, i.e. When the value is close to 0, b. the residual is large but the correlation is very high, i.e. A value close to 1 indicates a large residual, which is likely caused by real vehicle dynamics, such as sudden acceleration or deceleration, rather than a malfunction of the sensor itself.

[0098] A value close to 0 indicates that the system has extremely low trust in the main path signal, which occurs when the residual is very large. The residual is large and has a very low correlation with vehicle acceleration, i.e. When the value is close to 0, this is a strong indication that a sensor or actuator malfunction has rendered the signal completely unreliable.

[0099] The larger the value, the more reliable and trustworthy the main path torque signal is considered in the k-th control cycle. Subsequent systems, such as torque distribution and fault-tolerant control, will tend to use the main path signal more. It is a follow A function that increases and then rapidly decays; The larger, Even a slight increase will lead to Sharply decreased; The larger, The smaller, The smaller the value, the larger the residual, and the lower the confidence level in the main pathway. The larger the value, the more sensitive the system is to residuals;

[0100] The larger, The larger, The larger the value, the stronger the correlation between the residual and the vehicle acceleration, and the higher the confidence in the main path.

[0101] In the calculation of the credibility weight, the dynamic adjustment of the residual sensitivity coefficient is as follows: , For the first The residual sensitivity coefficient for each control cycle; the smaller the dynamic anomaly threshold, the higher the residual sensitivity.

[0102] Its specific practical significance and the situation it reflects: It is used for the next control cycle. In calculating the credibility weight of the main pathway When adjusting the absolute value of the residual The coefficient representing the influence of weights reflects the system's ability to operate at different detection sensitivities. In this regard, it is necessary to adjust the sensitivity of the residual size in the weight calculation accordingly so that the weight changes are consistent with the anomaly detection logic;

[0103] high Good road sections have a wide threshold. small, in the calculation , decreases slower as the residual increases, meaning even if the residual is relatively large, it can still be smaller than the relaxed dynamic abnormal threshold , the penalty on the weight is relatively light, the master channel weight drops less drastically, which matches the relaxed abnormality criterion on good roads;

[0104] low , for slippery road segments, the threshold is relatively large, decreases extremely fast as the residual increases, meaning even if the residual is not large in absolute value, it can be close to or exceed the tightened dynamic abnormal threshold , it will also cause a significant penalty on the weight , leading to a rapid drop of the master channel weight, which matches the tightened abnormality criterion and higher detection sensitivity requirement on slippery roads, ensuring that when the residual is close to or exceeds the low threshold, the weight can quickly reflect the drop of signal credibility;

[0105] larger, means the system is more sensitive to the change of residual , a smaller increase of residual will cause a significant drop of the master channel weight, smaller, means the system is less sensitive to the change of residual , the residual needs to become larger to cause a significant drop of the weight.

[0106] The calculation formula of the credibility weight of the ith wheel backup channel in the kth control cycle is: ;

[0107] is the complement of the master channel credibility weight, also a value between 0 and 1, representing the degree of trust of the system to the ith wheel backup channel torque signal in the kth control cycle, the trust degree of the system to the master and backup channels is a trade-off relationship, when the master channel credibility is high, the backup channel credibility is naturally low; when the master channel credibility is low, it means the backup channel can be more reliable, and its credibility is correspondingly improved;

[0108] close to 1, means the system highly trusts the backup channel signal, which directly corresponds to the case close to 0, i.e. the master channel signal is considered extremely untrustworthy; close to 0.5, means the system has neutral or moderate trust in the backup channel signal, which corresponds​ close to 0.5; close to 0, indicating that the system has extreme distrust in the backup path signal, which corresponds to close to 1, indicating that the primary path signal is considered very reliable;

[0109] the larger, the more reliable the backup path torque signal is considered in the kth control cycle, and the more trustworthy, when the primary path is determined to be abnormal or the weight is very low, the system will tend to use the signal of the backup path for control or diagnosis; the average reliability weight in the sliding window: , and the same ;

[0110] The correlation coefficient or residual mutation of a single cycle may be a spike caused by transient noise or short pulse interference, and using it to make immediate decisions can easily lead to excessive switching or false positives. The average weight can smooth out these short-term spikes and only be sensitive to persistent deviations;

[0111] In a window of length W, aggregate information from multiple cycles, better represent the overall health level of the primary path in this period of time, if the average value is always high, it means that the signal is reliable most of the time; if the average value decreases, it can be judged as a trend of failure risk; compared with the maximum value strategy, the average value is not too sensitive to single extreme error; compared with the minimum value strategy, it also does not ignore multiple slight abnormalities, it can not only capture the cumulative abnormalities of multiple cycles in time, but also avoid excessive reaction to accidental errors;

[0112] The average weight is directly used for "switching" or "restoring" decisions, which is consistent with the way of using LSTM model based on each sequence input in the sliding window, both rely on data statistics in the same period, logically more unified and easy to implement.

[0113] The specific data of the primary path reliability weight in some control cycles is shown in Table 1.

[0114] Table 1

[0115]

[0116] As the residual value increases, the confidence weight of the main path exhibits an exponential decay trend. For example, when the residual value is 0.12, the confidence weight of the main path is 0.86 in the first control cycle, indicating high confidence. When the residual value is 1.52, the confidence weight of the main path drops to 0.12 in the tenth control cycle, approaching failure. When the residual value is 1.68, the confidence weight of the main path further drops to 0.09 in the fourteenth control cycle, reaching critical failure. This demonstrates the system's sensitive response to abnormal signals: the greater the deviation between the measured torque and the theoretical prediction value of the main path, the lower its confidence assessment value. This relationship reflects the exponential term in the formula. The core function is that as the residual increases, the weights decay exponentially.

[0117] Pearson correlation coefficient absolute value There is a positive correlation with the weight: when Higher, that is In the first control period, even with a residual value of 0.12, the confidence weight of the main pathway still reached 0.86. Lower, that is In the sixth control period, even with a residual value of 0.89, the confidence weight of the main pathway decreased to 0.32; this reflects the relevance reward item. The compensation mechanism is as follows: When the residual is highly correlated with the change in vehicle acceleration, the system determines that the deviation may be due to actual working conditions rather than sensor failure, thereby slowing down the weight decay. This design effectively distinguishes between real fault signals and model prediction deviations.

[0118] Residual sensitivity coefficient As a dynamically adjustable parameter, it significantly affects the decay rate: low Scene, In the first control period, the decay is gradual, and a significant reduction in weighting (main pathway confidence weight) is only triggered when the residual value reaches 1.0; high Scene, In the 10th control period, the decay was steep, and when the residual value was 1.52, the confidence weight of the main pathway had dropped to 0.12; when Upon exceeding the critical threshold of 2.0, at the start of the 7th control cycle, the median confidence weight of the main pathway decreased from 0.57 to 0.16, with the deweighting rate increasing by 2.4 times; this change stemmed from... Inversely proportional to the dynamic anomaly threshold: In low-adhesion road conditions, Automatically increases in size to improve detection sensitivity; in high-adhesion road conditions, To avoid misjudgment, the system's adaptive capability to complex operating conditions is demonstrated. The system does not simply rely on a single threshold, but rather assesses signal reliability through a dynamic balance of residuals, correlation, and environmental sensitivity, significantly improving the accuracy of fault diagnosis.

[0119] The parameter adjustment module is used to estimate the adhesion coefficient of road friction using the wheel speed signal and the longitudinal acceleration signal of the vehicle body through the recursive least squares algorithm, correct the dynamic anomaly threshold based on the adhesion coefficient, and adjust the fusion torque distribution coefficient in the redundancy switching module.

[0120] The fusion calculation module calculates the fusion torque of the i-th wheel during the k-th control cycle. Calculate using the following formula:

[0121]

[0122] in, and These are the measured torque values ​​of the main path and backup path of the i-th wheel during the k-th control cycle, respectively.

[0123] The specific practical meaning and the situation it reflects: It represents the optimal estimated torque value for the i-th wheel in the k-th control cycle, which is ultimately used for vehicle control or subsequent prediction. This is based on the measured torque of the main path. Actual torque measured in the spare circuit The result of weighted averaging, weights , The real-time reliability of the primary and backup paths is directly derived from the anomaly detection module's calculation.

[0124] when Close to 1 When it is close to 0, close to This means that the system has a high degree of trust in the main path, and the fusion torque is almost equal to the measured value of the main path, which is the ideal state when the main path is normal.

[0125] when Close to 0 When it is close to 1, close to This means that the system highly trusts the backup path, and the fusion torque is almost equal to the measured value of the backup path. This indicates that the main path may have been judged to be abnormal or extremely unreliable, and the system has essentially switched to the backup path.

[0126] when , When all are between 0 and 1, is the weighted fusion of the master and standby channel torque values, which occurs when the system has medium or uncertain confidence in the master channel, for example, large residual but strong correlation with acceleration large, or small residual but weak correlation small, the system adopts a soft switching / fusion strategy, smooth transition or comprehensive information of both;

[0127] The size of itself represents the size of the wheel fused torque output instruction or estimated value, which depends on the specific vehicle control requirements, such as acceleration, braking, maintaining speed, and the formula itself does not directly indicate The larger or smaller represents the good or bad of a certain system state, which represents the torque value selected by the system based on the confidence judgment;

[0128] The larger, The greater the contribution to , The larger, The greater the contribution to ; when or itself increases, if its corresponding weight is not zero, tends to increase; the fusion torque dynamically and smoothly follows the torque signal of the channel currently considered more reliable by the system ( or ), or balances between the two, which isolates the influence of the untrusted signal and provides a more robust input for subsequent control or prediction.

[0129] The LSTM neural network model is inputted with the fusion torque sequence of W control cycles in the sliding window and the vehicle body longitudinal acceleration signal sequence and the wheel speed signal sequence to predict the required torque value of the next control cycle;

[0130] The LSTM neural network is a model good at learning long-term dependencies in time series data, which uses the historical data sequence of W consecutive control cycles in the sliding window to input the fusion torque sequence of W control cycles in the sliding window and the vehicle body longitudinal acceleration signal sequence and the wheel speed signal sequence, the purpose is to predict the required torque value of the next control cycle , which may be the torque required by the driver request, the target of the upper controller or to meet a certain performance index;

[0131] The prediction model synthesizes historical control input (fused torque), vehicle dynamic response (acceleration), actuator / system state (wheel speed) to infer future demand, which captures the dynamic response pattern of the vehicle to these inputs, the predicted demand torque value will be used as feedforward or desired input for subsequent control modules, such as torque distribution, motor control, to help improve the response speed and accuracy of the system;

[0132] Using fused torque as input ensures that the torque data input to the prediction model is data that has been evaluated for reliability and possibly fault-tolerant processing, eliminating the influence of unreliable sensor readings, improving the accuracy and robustness of the prediction model, there is a complex time correlation and hysteresis effect between vehicle torque demand and historical state, especially acceleration response and wheel speed change, such as inertia, using LSTM can effectively model this time sequence dynamic characteristic, more accurate than model prediction using only current time data.

[0133] The adjustment parameter module identifies the road surface friction coefficient based on the wheel speed signal and the vehicle body longitudinal acceleration signal through recursive least squares algorithm online And corrects the dynamic abnormal threshold and the fused torque distribution coefficient according to the following formula:

[0134] The specific practical significance and reflection of the situation: it is a value between 0 (such as ice) and 1 (such as dry asphalt), which represents the maximum available friction coefficient between the tire and the road estimated by the system at the kth control period, the ultimate performance of the vehicle's longitudinal dynamics (acceleration, braking) is fundamentally limited by the road adhesion conditions, is the real-time online estimation of this key physical environmental parameter.

[0135] The wheel speed signal reflects the wheel speed and slip rate, and the vehicle body longitudinal acceleration signal reflects the actual acceleration and deceleration of the vehicle, and the recursive least squares algorithm iteratively updates The core idea is to use the real-time relationship between wheel speed and acceleration to continuously adjust the value of According to a model describing the tire force-slip rate relationship, such as the simplified form of the magic formula, so that the model predicted vehicle body acceleration is as consistent as possible with the actual measured vehicle body acceleration;

[0136] When the actual measured acceleration deviates from the predicted acceleration based on the current and wheel speed, the recursive least squares algorithm adjusts To reduce this deviation, the algorithm has the ability to "forget" old data (through the forgetting factor), and pays more attention to recent dynamics;

[0137] is the key environmental parameter that the system real-time infers based on wheel behavior (wheel speed) and vehicle overall response (acceleration), and is the cornerstone of all subsequent parameter adjustments;

[0138] The correction formula of the dynamic abnormal threshold is:

[0139]

[0140] wherein, is the basic threshold, is the road surface sensitivity coefficient, is the road surface friction coefficient estimated based on the recursive least square method in the kth control cycle, is the dynamic abnormal threshold in the kth control cycle;

[0141] The specific practical significance and reflection of are as follows: it is the threshold set for the next control cycle for judging whether the main path residual exceeds the threshold, and the threshold is dynamically adjusted according to the currently estimated road surface adhesion coefficient, which reflects that the system has different tolerance to torque signal abnormalities on different road surfaces. is set for the next control cycle, is the threshold for judging whether the main path residual exceeds the threshold, and the threshold is dynamically adjusted according to the currently estimated road surface adhesion coefficient, which reflects that the system has different tolerance to torque signal abnormalities on different road surfaces. is high,

[0142] is large, such as a dry good road section, the system allows a larger deviation between the measured torque of the main path and the model predicted torque to trigger an abnormality determination, because on a good road, the model is usually more accurate, the vehicle dynamics is more stable, and a smaller deviation is more likely to be a problem of the sensor or actuator itself, and at the same time, a larger threshold can avoid false positives in the allowed normal operation fluctuations, such as heavy acceleration / braking. is low, is small, such as an icy and slippery road section, the system has a higher requirement for the consistency of the main path signal, and even a small deviation can trigger attention or abnormality determination, because on a low adhesion road: a. the vehicle dynamics is complex and intense (easy to slip and lock), the model prediction itself may be less accurate, resulting in a larger baseline of the residual absolute value, if a fixed threshold for high adhesion roads is still used, it will lead to a large number of "false normal" and miss real faults, b. any torque sensor failure or actuator abnormal output, even if the absolute value of the deviation is not large, is easy to cause the vehicle to lose control on a low adhesion road, and a lower threshold is needed to improve the detection sensitivity and find potential problems as soon as possible.

[0143]

[0144] ​​​​​​The greater, the higher the tolerance of the system to the main channel torque signal residual in the next cycle, the higher the threshold of abnormal judgment (more difficult to trigger abnormal judgment), The smaller, the lower the tolerance of the system to the residual, the lower the threshold of abnormal judgment (easier to trigger abnormal judgment), and the higher the detection sensitivity;

[0145] is a reference threshold set under typical or calibrated conditions, such as medium adhesion dry road, is a positive coefficient, controlling the strength of threshold adjustment, The greater, the greater the influence of road adhesion change on the threshold; loosen the threshold to prevent false positives on good roads, and tighten the threshold to improve sensitivity and safety on slippery roads, Adjust the degree of aggressiveness of this adaptive strategy;

[0146] The adjustment of the fusion torque distribution coefficient satisfies:

[0147]

[0148] Wherein, is the reference distribution coefficient, is the decay factor, which automatically reduces the main channel proportion in low adhesion road, is the predicted fusion torque distribution coefficient of the control cycle, is the predicted road friction adhesion coefficient of the control cycle;

[0149] The specific actual meaning and reflection of : it is a value between 0 and , used to dynamically adjust the degree of tendency to use the main channel torque in the fusion strategy in the next predicted control cycle , which reflects the system's trust in the main channel under low adhesion road conditions and reduces the basis;

[0150] It is obtained by recursive least squares algorithm online identification, which has built-in historical data weight decay or Kalman filter mechanism, etc. The algorithm itself has done a certain smoothing to noise and mutation, therefore, directly updating the threshold and distribution coefficient with the latest , can balance robustness and real-time performance, without the need for external window averaging;

[0151] And are control parameters for adaptive system, they are not directly used for instantaneous switching judgment, but as next cycle target threshold or distribution ratio, they should reflect the latest working condition to keep accurate matching to system error level or road condition, not too much lag; if both smoothed within the identification algorithm, and window averaged again, it will form two layers of filtering, leading to too slow response of parameters to road condition changes, losing the ability to adapt to rapid changes of road adhesion in real time, reducing safety; and window averaged again, it will form two layers of filtering, leading to too slow response of parameters to road condition changes, losing the ability to adapt to rapid changes of road adhesion in real time, reducing safety;

[0152] and depend on the latest online estimation results to ensure that the control parameters are highly consistent with the current road condition, the functional positioning and time requirements of the two are completely different, so in the present invention, only the residual and weight are averaged, and the latest value is directly used for dynamic threshold and distribution coefficient;

[0153] high such as dry good road, close to , the system tends to maintain or allow a higher basic weight of the main path, on good roads, the main path is usually more reliable and should be used preferentially;

[0154] low such as icy and slippery road, significantly less than , the system actively reduces the reference weight or upper limit of the main path, because: a. on low adhesion road, vehicle stability is sensitive and fault tolerance is low, b. the main path as the main path, once it fails, the damage is greater, c. even if the main path does not report a failure, the performance of its sensors or actuators in extreme working conditions may be more susceptible, d. by reducing , the system pre-biases the backup path at the architecture level, or limits the maximum weight of the main path in fusion, which is equivalent to making a pre-plan for the failure of the main path on the slippery road, improving the level of redundant safety;

[0155] the larger, the higher the basic setting or allowed weight or proportion of the main path torque in the next cycle fusion strategy, the smaller, the lower the basic setting or allowed weight or proportion of the main path, more inclined to rely on the backup path;

[0156] is the ideal reference weight or proportion of the main path when the road adhesion is high, i.e. close to 1, is a positive coefficient, controlling the speed of decay when the road adhesion decreases, the speed of decay, The greater, the faster the decay, the more aggressive the response to low adhesion road surface; represents the degree of loss of road adhesion, The lower, The greater, the faster the decrease with the increase of ;

[0157] On the slip road, the system strategically reduces the default dependence on the main channel, prepares for possible failures, and improves the overall robustness of the system in harsh conditions, Control the strength of this strategy.

[0158] Redundancy switching module, when the credibility weight of the main channel is higher than the safety threshold, the full demand torque value is executed by the main channel, when it is lower than the safety threshold, the switching is started, the distribution proportion of the main channel motor torque is gradually reduced according to the fusion torque distribution coefficient, the distribution proportion of the standby channel motor torque is simultaneously increased, the fluctuation strength of the fusion torque is monitored in real time, and the transition time window is dynamically adjusted to suppress mutation

[0159] The redundancy switching module monitors the credibility weight of the main channel and calculates the average credibility weight, when the average credibility weight is lower than the preset safety threshold, it enters the transition stage; This indicates that the system judges that the reliability of the main channel torque signal of the i-th wheel has been low to a dangerous level, which is not sufficient to safely undertake the full torque control task, reflecting that the main channel may have failed (according to the continuous over-standard judgment of the abnormal detection module), or its credibility weight Suddenly dropped due to large residual error and low correlation, indicating that the signal quality is very poor;

[0160] This is a hard signal, indicating that the redundancy switching mechanism must be started, and the control responsibility is transferred from the main channel to the standby channel, the preset safety threshold is a set threshold value, for example, 0.3 or 0.4, below this value means that it is too risky to continue to rely on the main channel; The lower the preset safety threshold, the higher the tolerance of the system to the failure of the main channel (the switching is less sensitive), the higher the preset safety threshold, the easier it is to trigger switching;

[0161] Let the total transition time be , the relative progress of the switching process is defined as , is a time variable, and the exponential smoothing transition function is defined as :

[0162]

[0163] Where, is the smoothing rate coefficient, , is the current transition progress, represents the start of switching, indicates the completion of switching;

[0164] indicates the continuous time elapsed since the start of the current switching action, when the switching trigger is the switching start point, when indicates the planned total switching time is exhausted, the switching end point;

[0165] The specific practical meaning and reflection of : it is a value between 0 and asymptotically 1, in the fixed transition total time , it defines the completion or intensity of the switching process over time, reflecting the system's adoption of a nonlinear, slow-to-fast transition strategy, which is small at the beginning and changes slowly, and close to 1 at the end and changes quickly;

[0166] close to 0, that is, close to 0, the switching just starts, the change to the system is very small, gradually increases, the switching process continues, and the impact on the system gradually deepens, when close to 1, at the end of the set transition period , the switching has completed most of it and is close to completion;

[0167] the larger, the higher the intensity of the current switching progress, that is, the system state is closer to the state of completely switching to the standby path, which directly controls the rate of change of the torque distribution ratio; is the smooth rate coefficient, which controls the shape and rate of change of the transition curve, the larger, the more gentle the initial rise of the curve and the steeper the later rise, is the calibration value, which ensures that it reaches a near-complete state within ; indicates the proportion of the current time to the planned total switching time ;

[0168] increases, decreases, increases, and the growth is nonlinear, small (initial switching), very slow growth (small slope), large (late switching), relatively fast growth (slope increases); in the initial switching period, the main path torque decreases slowly, giving the actuator and vehicle dynamics response time and avoiding initial mutations; in the later period, the switching is confirmed to be feasible, and the speed is increased to complete the switching as soon as possible to reduce the risk window period;

[0169] In the transition period, the demand torque value is proportionally distributed between the main and backup channels, where the main channel torque distribution proportion is , and the backup channel torque distribution proportion is , so that the main channel torque proportion gradually decays from to 0, and the backup channel proportion synchronously rises from 0 to ;

[0170] The specific practical meaning and reflected situation of : it indicates that in the switching transition period, the proportion of the demand torque of the i-th wheel allocated to the main channel for execution, which reflects that the responsibility of the main channel is being gradually unloaded;

[0171] Switching starts , , the main channel assumes its baseline distribution proportion of torque (from the adjustment parameter module), which is consistent with the basis of the fusion strategy before the switching trigger or in the normal state; in the switching , , as increases, decreases linearly, and the proportion assumed by the main channel gradually decreases; switching ends , , the main channel completely exits torque execution and no longer assumes any torque output task;

[0172] The greater , the greater the proportion of torque that the main channel still needs to assume in the transition period, decreases to 0 indicating that the main channel is completely unloaded;

[0173] The specific practical meaning and reflected situation of : it indicates that in the switching transition period, the proportion of the demand torque of the i-th wheel allocated to the backup channel for execution, which reflects that the responsibility of the backup channel is being gradually loaded;

[0174] Switching starts , , the main channel assumes its baseline distribution proportion of torque, which is consistent with the basis of the fusion strategy before the switching trigger or in the normal state; in the switching , , as increases, increases linearly, and the proportion assumed by the backup channel gradually increases; switching ends , , the backup channel assumes the entire torque execution task; is the reference weight of the main path at normal or switching start, The larger means the higher proportion of the main path at normal, and the starting proportion of the backup path The lower;

[0175] The larger, the greater the proportion of torque borne by the backup path during the transition period, Increased to 1 means the backup path takes over completely;

[0176] Although And Directly controlled by Linear, but It is linear time progress And Is an exponential function, which means that the change rate of torque distribution ratio / In time It is linear, uniform, Here mainly used to define the concept of transition intensity or possibly used in adaptive adjustment The logic, but not directly involved in torque distribution calculation, torque distribution ratio is a linear function of time progress ;

[0177] The core logic is to linearly reduce the torque execution ratio of the main path from To 0 at a constant rate within the fixed planned transition time , While increasing the execution ratio of the backup path from To 1, which ensures the smoothness of torque command transfer between paths;

[0178] Real-time monitoring of fusion torque, calculating the standard deviation of fusion torque sequence to represent fluctuation intensity, and increasing or decreasing To balance the switching speed and smoothness, and suppress mutation;

[0179] The specific practical significance and reflection of the standard deviation of fusion torque sequence: it is calculated in the latest time window, for example, several control periods, within the switching transition period, which reflects the size of torque command fluctuation or oscillation caused by the current switching process. It quantifies the disturbance intensity of switching action on system output;

[0180] The smaller the standard deviation of the fusion torque sequence, the smoother the switching process, and the smaller the fusion torque output fluctuation, the smoother the vehicle dynamic feeling;

[0181] The larger the standard deviation of the fusion torque sequence, the greater the torque fluctuation or oscillation caused by the switching process, which may be due to: a. Large differences in main and backup path torque sensor / actuator characteristics, b. Switching speed Too fast, beyond the response capability of the actuator or vehicle dynamics, c. External disturbance, such as road bumps, exacerbates the switching impact;

[0182] The greater the standard deviation of the fusion torque sequence, the more intense the disturbance caused by the current switching process to the system, the worse the smoothness, and the higher the potential risk;

[0183] Adjusted total transition time Specific actual meaning and reflection: it is the new planned total transition time for subsequent switching or the remaining part of the current switching that the system dynamically calculates according to the currently monitored standard deviation of the fusion torque sequence, which reflects the system's adaptive balance between switching speed and smoothness;

[0184] The standard deviation of the fusion torque sequence is small, It can be reduced or maintained, indicating that the current switching speed Can be accepted to be slower, and the system attempts to speed up the switching speed To complete the switching faster to reduce the time window of the main path unreliable state;

[0185] The standard deviation of the fusion torque sequence is large, Increase, indicating that the current switching speed Is too fast, beyond the smooth response capability of the system, and the system actively prolongs the switching time Slow down the torque transfer rate to suppress fluctuations and improve smoothness;

[0186] The greater, the longer the system plans to take to complete the subsequent switching process, or the remaining part of the current switching, the slower the switching speed, in order to obtain better smoothness, The smaller, the faster the switching speed;

[0187] When the credibility weight of the main path returns to the safety threshold and above, the redundant switching module uses the opposite exponential smoothing transition function and the same The braking torque distribution is switched back to the main path, realizing bidirectional seamless switching;

[0188] The specific actual meaning and reflection of the main path credibility weight recovery, that is, the main path credibility weight is greater than or equal to the preset safety threshold: this indicates that the main path torque signal of the ith wheel has recovered to a safe level at the current time, which reflects that the main path may recover from failure, such as transient disturbance disappearing, failure being eliminated, or its signal quality significantly improving, residual error becoming small, correlation improving; it indicates that this is a signal that indicates that it is safe to consider switching the control back to the main path;

[0189] The switching back process uses the same basis or current adjusted As the total duration of the plan, it ensures that the return process and the cut-out process are symmetrical in time scale;

[0190] The most reasonable and safe implementation of the opposite exponential smoothing transition function is: a. Define a new return progress , whose value is from 0 to 1, b. Define the return transition function , in order to achieve the effect of "opposite", the more common engineering practice is: A. Load the main path: the torque distribution ratio of the main path Gradually increases from 0 to , which can be defined as , so that the load is slow at the beginning of the return , and fast later , which conforms to the general engineering principle of slow first and fast later;

[0191] B. Unload the standby path: the torque distribution ratio of the standby path Gradually decreases from 1 to , which can be defined as , so that the unloading is slow at the beginning of the return , and fast later ;

[0192] No matter scheme A or B, the form of the transition function Itself does not change, only the object or starting point of its application changes, that is, the main load and the standby unloading, so "opposite" is more likely to refer to the reversal of the operation object, that is, from the cut-out "unload the main load the standby" to the return "load the main unload the standby", rather than the mathematical function form negation, the function form remains unchanged, which ensures the similarity of the switching dynamics;

[0193] The meaning of two-way seamless switching: refers to the mechanism of triggering based on credibility weight, exponential smoothing transition, torque ratio linear transfer, and time length adaptive adjustment, whether it is from the main to the standby or from the standby to the main, to ensure that the switching in both directions is smooth, impact-free and controlled, like "seamless" connection.

[0194] Please refer to Figure 6 , the application further provides a vehicle wheel edge redundant line control brake method based on an EMB system, the method is used for executing the above-mentioned vehicle wheel edge redundant line control brake system based on the EMB system, comprising:

[0195] Step 1: asynchronously collect and align the main and standby path motor torque signals, wheel speed signals and vehicle body longitudinal acceleration signals of the vehicle to be detected;

[0196] Step 2: The Kalman filter is used to predict the main channel motor torque signal, and the residual value of the theoretical value and the measured value is calculated. When the residual value exceeds the dynamic abnormal threshold in the sliding window, the abnormality is determined. According to the statistical correlation between the residual value in the same window and the body longitudinal acceleration signal, the reliability weight of the main and standby channels is calculated;

[0197] Step 3: The road friction adhesion coefficient is estimated by using the recursive least square algorithm through the wheel speed signal and the body longitudinal acceleration signal. The dynamic abnormal threshold is corrected according to the adhesion coefficient, and the fusion torque distribution coefficient in the redundancy switching module is adjusted.

[0198] Step 4: The main and standby channel motor torque signals are weighted and fused according to the reliability weight of the main and standby channels to generate a fusion torque. The data signals in the current sliding window are used as input, and the demand torque value in the next control period is output through a short-time sequence prediction model.

[0199] Step 5: When the main channel reliability is higher than the safety threshold, the main channel executes the full demand torque value. When it is lower than the safety threshold, the switching is started. The distribution proportion of the main channel motor torque is gradually reduced according to the fusion torque distribution coefficient, and the distribution proportion of the standby channel motor torque is simultaneously increased. The fluctuation intensity of the fusion torque is monitored in real time, and the transition window is dynamically adjusted to suppress mutation.

[0200] The above formulas are dimensionless values, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the latest real situation. The preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0201] The above embodiments can be realized by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software depends on the specific application and design constraints of the technical solutions.

[0202] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0203] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application.

Claims

1. A vehicle wheel-side redundant brake-by-wire system based on an EMB system, characterized in that, Specifically comprising: A signal acquisition module for asynchronously acquiring and aligning the main and backup channel motor torque signals, wheel speed signals and vehicle body longitudinal acceleration signals of the vehicle to be detected; An anomaly detection module for predicting the theoretical value of the main channel motor torque signal based on a Kalman filter, calculating the residual value of the theoretical value and the measured value, determining an anomaly when the residual value exceeds a dynamic anomaly threshold within a sliding window, and calculating the reliability weight of the main and backup channels according to the statistical correlation of the residual value and the vehicle body longitudinal acceleration signal within the same window; An adjustment parameter module for estimating the road friction adhesion coefficient by using the wheel speed signal and the vehicle body longitudinal acceleration signal, correcting the dynamic anomaly threshold according to the adhesion coefficient, and adjusting the fusion torque distribution coefficient in the redundancy switching module; A fusion calculation module for weighting and fusing the main and backup channel motor torque signals according to the reliability weight of the main and backup channels to generate a fusion torque, taking the data signals within the sliding window as input, and outputting the demand torque value of the next control period through a short-time sequence prediction model. A redundancy switching module for executing the full demand torque value from the main channel when the reliability weight of the main channel is higher than a safety threshold, starting switching when the reliability weight is lower than the safety threshold, gradually reducing the distribution proportion of the main channel motor torque according to the fusion torque distribution coefficient, synchronously increasing the distribution proportion of the backup channel motor torque, and real-time monitoring the fluctuation intensity of the fusion torque to dynamically adjust the transition window to suppress mutations.

2. The vehicle wheel-side redundant brake-by-wire system based on an EMB system according to claim 1, characterized in that: The signal acquisition module asynchronously receives the following signals through the vehicle CAN / CAN-FD bus: The main channel motor torque signal fed back by the wheel-side main actuator and the backup channel motor torque signal fed back by the wheel-side backup actuator, both with a sampling rate of 500Hz, the wheel speed signal output by the wheel speed sensor with a sampling rate of 100Hz, and the vehicle body longitudinal acceleration signal output by the vehicle body inertia unit with a sampling rate of 1kHz; The asynchronously collected data is time-stamped with the ECU local time and stored in a double-direction circular buffer capable of storing at least W consecutive control cycle data, and time alignment is performed through a sliding window mechanism: the sliding window contains W consecutive control cycles, each control cycle has a duration of , the window step is one control cycle duration, and the total window duration is at least equal to the maximum brake response delay of the system . The time sequence alignment mechanism takes the vehicle body longitudinal acceleration signal as the reference time axis, uses piecewise cubic Hermite interpolation for the wheel speed signal, and uses linear interpolation for the main and backup channel motor torque signals. After interpolation, the timestamp synchronization error of each signal is less than or equal to 5μs. After interpolation, the anomaly detection module reads the aligned signal data within the sliding window.

3. The vehicle wheel edge redundancy brake-by-wire system based on the EMB system according to claim 2, characterized in that: The anomaly detection module predicts the theoretical value of the main channel motor torque signal based on the Kalman filter, and calculates the difference between the theoretical value and the measured value as the residual value: The set i represents the wheel number, k is the index of the control period in the sliding window, and each wheel in the abnormality detection module is sequentially indexed in time sequence The calculation formula of the residual value of each wheel in the kth control period is: ； wherein, is the Kalman filtered value of the motor torque signal of the master path for the i-th wheel of the vehicle in the k-th control cycle, is the corresponding measured torque value, the difference between the two being the residual value ; the average residual value within the sliding window: ; only when the average residual value of the main passage of the i-th wheel exceeds the dynamic abnormality threshold value When the dynamic abnormality threshold value is exceeded within the sliding window, the wheel main passage signal is determined to be abnormal, otherwise the main passage signal is maintained normal.

4. The vehicle wheel edge redundancy brake-by-wire system based on the EMB system according to claim 3, characterized in that: According to the statistical correlation analysis of the residual value and the vehicle body longitudinal acceleration signal within the same time window, the logic for calculating the reliability weight of the main and backup channels is: a weight of credibility of the ith wheel main passage in the kth control cycle The calculation formula is as follows: ； wherein, is the residual sensitivity coefficient at the kth control cycle, is the Pearson correlation coefficient of the residual sequence and the body longitudinal acceleration sequence within the same window. The reliability weight of the ith wheel in the kth control cycle The calculation formula is: ; Average confidence weight within a sliding window: , and similarly .

5. The vehicle wheel edge redundancy brake-by-wire system based on the EMB system according to claim 4, characterized in that: The fusion computing module calculates the fusion torque of the ith wheel in the kth control cycle is calculated as follows: ； wherein, with respectively the torque measured values of the i-th wheel main passage and the standby passage in the k-th control cycle. The LSTM neural network model is input with the fusion torque sequence, vehicle body longitudinal acceleration signal sequence and wheel speed signal sequence within the sliding window of W control periods to predict and output the demand torque value of the next control period.

6. The vehicle wheel edge redundancy brake-by-wire system based on the EMB system according to claim 5, characterized in that: The adjustment parameter module identifies the road friction adhesion coefficient online based on the wheel speed signal and the vehicle body longitudinal acceleration signal through the recursive least squares algorithm, and corrects the dynamic anomaly threshold and the fusion torque distribution coefficient according to the following formula: The correction formula of the dynamic anomaly threshold is: ； in, Based on the threshold, For road surface sensitivity coefficient, The road surface friction adhesion coefficient is estimated based on the recursive least squares method in the k-th control cycle. For the first Dynamic anomaly threshold for each control cycle; The adjustment of the fusion torque distribution coefficient satisfies: ； wherein, is a reference distribution coefficient, is a decay factor, automatically reduces the main passage ratio when the low adhesion road surface, is a predicted fusion torque distribution coefficient of the control cycle, is a road surface friction adhesion coefficient of the control cycle; In the calculation of the credibility weight, the dynamic adjustment of the residual sensitivity coefficient is as follows: , For the first The residual sensitivity coefficient for each control cycle; the smaller the dynamic anomaly threshold, the higher the residual sensitivity.

7. The vehicle wheel edge redundancy brake-by- wire system based on the EMB system according to claim 6, characterized in that: The redundancy switching module monitors the reliability weight of the main channel and calculates the average reliability weight thereof, and when the average reliability weight is lower than a preset safety threshold, enters a transition stage; Let the total transition duration be , the relative progress of the handover procedure is defined as , where is the time variable, and the exponential smoothing transition function is defined as : ； wherein is a smoothing rate coefficient, , is a current transition progress, denotes a switch start, denotes a switch completion; During the transition period, the demand torque value is proportionally distributed between the primary and secondary paths, with the primary path torque distribution proportion being and the secondary path torque distribution proportion being , such that the primary path torque proportion is gradually decayed from to 0 and the secondary path proportion is simultaneously ramped up from 0 to ; Real-time monitoring of fusion torque, calculating fusion torque sequence standard deviation to represent fluctuation intensity, and increasing or decreasing according to fluctuation intensity To balance the switching speed and smoothness, and suppress mutation; When the confidence weight of the primary channel recovers to the safety threshold and above, the redundancy switching module transitions back to the primary channel with an inverse exponential smoothing transition function and the same The brake torque distribution is switched back to the primary channel, enabling bi-directional seamless switching.

8. A vehicle wheel-side redundant brake-by-wire method based on an EMB system, characterized in that: The method is used for executing the vehicle wheel edge redundancy line control brake system based on the EMB system according to any one of claims 1-7, and the specific steps include: Step 1: asynchronously collecting and aligning the main and standby channel motor torque signals, wheel speed signals and vehicle body longitudinal acceleration signals of the vehicle to be detected; Step 2: predicting the theoretical value of the main channel motor torque signal based on the Kalman filter, calculating the residual value of the theoretical value and the measured value thereof, determining the abnormality when the residual value exceeds the dynamic abnormal threshold in the sliding window, and calculating the reliability weight of the main and standby channels according to the statistical correlation between the residual value in the same window and the vehicle body longitudinal acceleration signal; Step 3: estimating the road friction adhesion coefficient by using the recursive least square algorithm through the wheel speed signal and the vehicle body longitudinal acceleration signal, correcting the dynamic abnormal threshold according to the adhesion coefficient, and adjusting the fusion torque distribution coefficient in the redundancy switching module; Step 4: weighting and fusing the main and standby channel motor torque signals according to the reliability weight of the main and standby channels to generate the fusion torque, taking the data signals in the current sliding window as the input, and outputting the required torque value in the next control period through the short-time sequence prediction model; Step 5: when the reliability of the main channel is higher than the safety threshold, executing the full required torque value by the main channel, and when the reliability is lower than the safety threshold, starting the switching, gradually reducing the distribution proportion of the main channel motor torque according to the fusion torque distribution coefficient, synchronously improving the distribution proportion of the standby channel motor torque, monitoring the fluctuation intensity of the fusion torque in real time, and dynamically adjusting the transition window to suppress the mutation.

Images