A tailgate anti-pinch system and method based on multi-modal sensor fusion

Abstract

The application provides a tailgate anti-pinch system and method based on multi-modal sensor fusion. The system synchronously collects the phase current of the tailgate motor, the acceleration of the push rod, and the acoustic signal of the tailgate cavity, and obtains the absolute opening angle of the tailgate by backward difference integration of angular velocity. The absolute opening angle is used as a spatial domain index to map the collected time domain signal to the corresponding spatial domain slot, and the Mahalanobis distance evaluation is performed with the preset dynamic baseline manifold. When the distance evaluation value exceeds the dynamic chi-square distribution threshold, the anti-pinch reverse action is triggered. The system eliminates the judgment delay caused by mechanical elastic deformation.

Description

Technical Field

[0001] This application relates to the field of automotive electronic control technology, specifically to a vehicle body closure anti-pinch control technology, and more particularly to a tailgate anti-pinch system and method based on multimodal sensor fusion. Background Technology

[0002] Current tailgate anti-pinch technologies are generally trapped in a theoretical dead end based on current thresholds. When faced with low-rigidity obstacles such as human soft tissue, single-lumen sensors inevitably produce millisecond-level physical response delays caused by the absorption of initial kinetic energy by the elastic deformation of the mechanical transmission mechanism.

[0003] Meanwhile, to avoid fluctuations in the environmental background load caused by oncoming gusts and vehicle slope deflection, traditional time-domain threshold comparison schemes are forced to compromise by significantly raising the warning threshold. This not only exacerbates the risk of instantaneous crushing injury in real-world trapped scenarios but also causes the system to lose its ability to detect early, minor resistance. These issues indicate that single-dimensional absolute threshold control methods have reached their performance ceiling. Summary of the Invention

[0004] In a first aspect, this application provides a tailgate anti-pinch method based on multimodal sensor fusion. The method includes acquiring a phase current signal, an inertial measurement signal, and an acoustic signal with a unified timestamp; extracting an absolute opening angle based on the inertial measurement signal and discretizing the absolute opening angle into a spatial domain slot index; fusing the feature extraction results of the phase current signal, the acceleration signal in the inertial measurement signal, and the acoustic signal into a current state vector based on the spatial domain slot index; acquiring the historical baseline mean vector and the historical covariance inverse matrix corresponding to the spatial domain slot index; determining a spatial distance evaluation value based on the current state vector, the historical baseline mean vector, and the historical covariance inverse matrix; and triggering a tailgate anti-pinch reversal action based on a comparison result between the spatial distance evaluation value and an anti-pinch trigger threshold.

[0005] Optionally, the analog-to-digital converter, serial peripheral interface, and integrated circuit built-in audio bus are concurrently triggered at a fixed frequency using a hardware timer; the phase current signal output by the analog-to-digital converter, the inertial measurement signal output by the serial peripheral interface, and the acoustic signal output by the integrated circuit built-in audio bus are transferred to a ring buffer in static random access memory via a direct memory access channel.

[0006] Optionally, the angular velocity component in the inertial measurement signal is extracted; an angular velocity history queue containing the angular velocity component at the current moment and the historical angular velocity component of the previous sampling period is constructed; based on the angular velocity history queue, the absolute opening angle is updated using the second-order Adams-Bashfors explicit numerical integration algorithm; and the spatial domain slot index is generated based on the modulo operation result of the absolute opening angle and the preset discretization step size.

[0007] Optionally, a fast Fourier transform is performed on the acoustic signal to extract the acoustic energy integral value of a preset high-frequency contact band; the acceleration signal along the push rod axis is extracted from the inertial measurement signal; and the phase current signal, the acceleration signal, and the acoustic energy integral value are vector-concatenated to generate the current state vector consisting of three dimensions.

[0008] Optionally, the state residual vector between the current state vector and the historical baseline mean vector is obtained; the transpose of the state residual vector, the inverse historical covariance matrix, and the state residual vector are multiplied by matrix to obtain the spatial distance evaluation value as a dimensionless pure scalar.

[0009] Optionally, the chi-square distribution information interval value corresponding to the three degrees of freedom is obtained as the anti-pinch trigger threshold; when the spatial distance evaluation value is greater than the anti-pinch trigger threshold, the general-purpose input / output pin of the microcontroller is set to a high level to drive the motor to control the full-bridge circuit to reverse; when the spatial distance evaluation value is not greater than the anti-pinch trigger threshold, the current state vector is pushed into the safety history queue.

[0010] Optionally, when the tailgate completes its closing motion and the tailgate anti-pinch reversal action is not triggered globally, the current state vector corresponding to each spatial domain slot index in the safety history queue is read; using an exponentially weighted moving average algorithm, the historical baseline mean vector is updated based on the current state vector to generate a new mean vector; a residual outer product matrix is ​​generated based on the new mean vector and the current state vector; using the exponentially weighted moving average algorithm, the historical covariance matrix is ​​updated based on the residual outer product matrix to generate a new covariance matrix; the Cholsky decomposition inversion operation is performed on the new covariance matrix to generate a new covariance inverse matrix and overwrite the non-volatile memory.

[0011] Secondly, this application provides a tailgate anti-pinch system based on multimodal sensor fusion. The system includes a multimodal hard-synchronous acquisition module configured to acquire phase current signals, inertial measurement signals, and acoustic signals with a unified timestamp; a spatial domain phase-locked resampling module configured to extract the absolute opening angle based on the inertial measurement signal and discretize the absolute opening angle into a spatial domain slot index; the spatial domain phase-locked resampling module is further configured to fuse the feature extraction results of the phase current signal, the acceleration signal in the inertial measurement signal, and the acoustic signal into a current state vector based on the spatial domain slot index; and a Mahalanobis distance evaluation and decision module configured to acquire the historical baseline mean vector and the historical covariance inverse matrix corresponding to the spatial domain slot index; the Mahalanobis distance evaluation and decision module is further configured to determine a spatial distance evaluation value based on the current state vector, the historical baseline mean vector, and the historical covariance inverse matrix, and trigger a tailgate anti-pinch reversal action based on a comparison result between the spatial distance evaluation value and the anti-pinch trigger threshold.

[0012] Optionally, the system further includes a dynamic baseline adaptive evolution module; the dynamic baseline adaptive evolution module is configured to update the historical baseline mean vector and the historical covariance inverse matrix using an exponentially weighted moving average algorithm when the tailgate completes its closing motion and no anti-pinch action is triggered globally.

[0013] Thirdly, this application provides a non-volatile computer-readable storage medium. When the computer instructions are executed by a microcontroller, the microcontroller performs the tailgate anti-pinch method based on multimodal sensor fusion as described in the first aspect or any optional method of the first aspect.

[0014] This application offers the following advantages: It elevates the underlying logic of anti-pinch detection from one-dimensional absolute threshold comparison in the time domain to dynamic calculation of the Mahalanobis distance probability envelope in the multi-dimensional spatial domain. By deeply integrating the microsecond-level time advantage of high-frequency broadband acoustic sensors in capturing structural stress waves at the moment of minute physical contact, and the three-dimensional spatial representation of the vehicle's pitch and roll states by a six-axis inertial measurement unit, this application achieves decoupling of the drag characteristics of complex nonlinear environments from the actual drag characteristics of being trapped at the algorithm level.

[0015] Based on real-time inverse covariance matrix projection calculation, the system can adaptively remove motor overload artifacts induced by sudden changes in vehicle tilt angle or surges in rear wind resistance. This enables the multi-modal fusion system to eliminate the mechanical buffer hysteresis present in traditional single-phase current anti-pinch when dealing with extremely low stiffness elastic body compression conditions. Attached Figure Description

[0016] Figure 1This is a flowchart illustrating a tailgate anti-pinch method based on multimodal sensor fusion, provided in an embodiment of this application.

[0017] Figure 2 This is a structural block diagram of a tailgate anti-pinch system based on multimodal sensor fusion, provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0020] This embodiment provides a tailgate anti-pinch system and method based on multimodal sensor fusion. In a specific implementation, this method employs hard-synchronous phase-locked resampling and dimensionality reduction of electrical, acoustic, and inertial attitude signals within the absolute opening space of the tailgate, and executes a Mahalanobis distance anomaly boundary verification mechanism based on real-time covariance matrix inversion. This allows for the extraction and isolation of long-term common-mode slowly varying components caused by wear, aging, or temperature changes between sensor signals at fixed physical topology nodes independent of time variables. This method solves the technical problems of existing single-point time-domain determination strategies, which are prone to microsecond-level decision dead zones due to nonlinear deformation of mechanical components absorbing kinetic energy, and the forced compromise of increased trigger thresholds due to the lack of interpretability for slope parking and strong wind interference caused by a single representation dimension. It achieves the goal of reducing the reverse protection action trigger force value to within the critical safety standard while eliminating the delay caused by the buffering effect and maintaining the background disturbance resistance stability in complex environments.

[0021] This embodiment of the tailgate anti-pinch system based on multimodal sensor fusion is physically deployed within a standard microcontroller. The microcontroller includes an instruction set central processing unit (CPU), a first memory block configured with static random access memory (SRAM), and a second memory block configured with non-volatile flash memory. The system further integrates a sixteen-channel direct memory access controller operating independently of the CPU, a twelve-bit successive approximation analog-to-digital converter (ADC) peripheral module, a high-frequency serial peripheral interface (HFPI) main control module, and an integrated circuit (IC) built-in audio bus module. At the external device level, the input pins of the ADC peripheral module are connected to a milliohm-level precision shunt resistor located in the low-side loop of the tailgate electric push rod drive circuit; the HFPI main control module is connected to a six-axis inertial measurement unit (IMU) embedded in the middle of the push rod's sealed cavity; and the IC built-in audio bus module is connected to an automotive-grade miniature microphone installed behind the tailgate's sheet metal cavity interior panel.

[0022] The following detailed description of the methods and processing units provided in the embodiments of this application, with reference to specific steps, provides a detailed explanation.

[0023] S000, System initialization and cold start transition period degradation control implementation example established.

[0024] When the microcontroller experiences a cold start state that results in the loss of volatile memory data, such as a power outage restart, a calibration reset after vehicle assembly line production, or a deep discharge and reconnection of the vehicle battery, the dynamic baseline manifold parameters stored in the second memory block are invalid or deviate from the current actual mechanical conditions. The system then executes an initialization self-built baseline process. During this transition phase, directly enabling an algorithm based on probability distribution distance may lead to unpredictable control states due to the non-convergence of the covariance matrix. Therefore, the method controls the system to enter a degradation protection operation mode.

[0025] S001. Perform hardware connectivity and memory structure integrity checks.

[0026] The microcontroller initiates a self-test sequence within its internal reset exception handling routine. The microcontroller sends a read command to the device identifier register of the six-axis inertial measurement unit to verify the serial bus timing; it also measures an internal known bandgap reference voltage to verify the conversion accuracy of the analog-to-digital converter. Simultaneously, the microcontroller allocates a contiguous physical memory space within the first memory block and performs a full-address step-by-step read / write test. After confirming the absence of bad blocks, it establishes this space as a protected circular buffer for high-speed data stream caching.

[0027] S002, activate the independent hard threshold degradation anti-pinch strategy during the cold start transition period.

[0028] When the system detects that it is in the state of needing to re-establish the baseline after the first three complete tailgate opening and closing strokes, the microcontroller sets the execution flag of the Mahalanobis distance calculation module in the main process to a sleep state. The microcontroller then starts the underlying watchdog absolute current comparison subroutine in parallel. This subroutine is directly attached to the end of the interrupt service routine of the analog-to-digital converter to perform one-dimensional absolute value monitoring of the phase current signal.

[0029] For example, the underlying watchdog absolute current comparison subroutine includes a fixed safety fuse threshold constant. For example, by taking into account the stall current characteristics of the tailgate motor, a fixed safety fuse threshold constant is established. Set as (Ampere). During the first closing cycle after the system is powered on, the tailgate moves to the middle position, and the real-time sampled motor phase current is... (Ampere). The watchdog absolute current comparison subroutine evaluates the conditions. If the error is false, the motor will continue to close the gate. During the second closing cycle of data collection, if the tailgate impedance suddenly increases, the motor phase current will... Within (milliseconds) from Rise to (Ampere). The watchdog absolute current comparison subroutine captures... Once the conditions are met, the general-purpose input / output pins are manipulated to flip the high and low side switch bridge arm control bits of the driver chip, thus executing the tailgate reversal.

[0030] S003. Perform non-interventional full-journey baseline data multidimensional feature harvesting and parameter solidification.

[0031] During the period when the system maintains the cold start degradation strategy, the system performs multimodal feature information extraction during the subsequent three complete normal closed loops that do not trigger the aforementioned safety fuse threshold. The microcontroller accumulates the current, acceleration, and acoustic energy samples collected in each discretized spatial opening slot during these three loops. After the third loop completely ends and the door lock microswitch is activated, the microcontroller calculates all spatial domain slot indices for these three independent sample sets. Initial baseline mean vector and the initial covariance inverse matrix The data is then written into the second memory block. The microcontroller then removes the sleep status flag for the distance algorithm module and enters the multi-dimensional protection mode.

[0032] S100: Acquire phase current signals, inertial measurement signals, and acoustic signals with a unified timestamp.

[0033] After the system enters the adaptive anti-pinch monitoring cycle, the multimodal hard synchronization acquisition module ensures that the three sensors with independent sampling frequencies and communication protocols are aligned on the time axis through data routing of the underlying hardware interconnect matrix, so as to eliminate the time misalignment error in the distance assessment model.

[0034] S110 uses a hardware timer to concurrently trigger the analog-to-digital converter, serial peripheral interface, and integrated circuit built-in audio bus at a fixed frequency.

[0035] The microcontroller's internal first-level timer peripheral is configured as a time base generator. This generator routes the generated overflow pulse directly to the microcontroller's hardware trigger multiplexer module. This multiplexer module fans out this clock overflow event and simultaneously triggers three hardware control ports: the first trigger line writes to the control register of the analog-to-digital converter peripheral module to enable sequence scan conversion; the second trigger line writes to the control bits of the high-frequency serial peripheral interface master module to activate register exchange timing; and the third trigger line sends a receive request signal to the direct memory access controller in the integrated circuit's built-in audio bus controller.

[0036] For example, the auto-reload register value of the hardware timer is set to the system bus clock divided by the target sampling rate. For example, the target sampling frequency... Set as (Hertz), producing strict Events are triggered at (millisecond) intervals. At time nodes. (milliseconds), three independent acquisition chains are activated simultaneously. to During the interval, the analog-to-digital converter converts the voltage drop across the shunt resistor into a digital quantity; after receiving the chip select signal, the six-axis inertial measurement unit outputs acceleration and angular velocity frames through the serial clock line.

[0037] S120. The phase current signal output by the analog-to-digital converter, the inertial measurement signal output by the serial peripheral interface, and the acoustic signal output by the integrated circuit's built-in audio bus are transferred to a ring buffer in the static random access memory via a direct memory access channel.

[0038] The multimodal hard-synchronous acquisition module allocates independent direct memory access (DMI) hardware request channels to the three peripherals, allowing the data transfer process to bypass the central processing kernel. Each channel is configured with a peripheral source register base address and a continuous target address pool located within the first memory block. The multimodal hard-synchronous acquisition module sets the target address generator register of the channel to a cyclic incrementing mode. When the peripheral's internal data register is ready, the peripheral sends a signal to the corresponding DMI hardware request channel, and the DMI controller concatenates the multiple data streams and writes them into the designated static random access memory (SRAM) unit via a bus matrix.

[0039] For example, the circular buffer maintained in the static random access memory contains a basic data structure entity SyncRawFrame_t. This structure contains: an unsigned 32-bit integer timestamp, an unsigned 16-bit integer current-to-digital converter value, three signed 16-bit integer acceleration values, three signed 16-bit integer angular velocity values, and a signed 16-bit integer acoustic pulse-code modulation sample. The structure is aligned to four-byte boundaries in memory, and each unit occupies [a certain amount of space / area]. (Byte) contiguous physical space. After the direct memory access controller completes one cycle of data transfer, the memory address pointer jumps from the starting address 0x20000000 to 0x20000014.

[0040] S200. Extract the absolute opening angle based on the inertial measurement signal, and discretize the absolute opening angle into a spatial domain slot index.

[0041] The spatial domain phase-locked resampling module resamples and projects continuous data containing the time dimension onto a spatial domain coordinate system with the physical absolute opening angle of the tailgate as the horizontal axis, in order to solve the problem of misalignment of time series comparison data caused by the non-uniform mechanical movement of the tailgate.

[0042] S210. Extract the angular velocity component from the inertial measurement signal.

[0043] The spatial domain phase-locked resampling module locates the memory address of the SyncRawFrame_t structure and reads the signed 16-bit integer raw register data corresponding to the rotation around the vehicle's lateral axis. The spatial domain phase-locked resampling module converts this data into a single-precision floating-point number, multiplies it by a sensitivity coefficient pre-written into non-volatile memory, subtracts the stationary zero-bias constant, and obtains the debiased angular velocity component.

[0044] For example, if the hexadecimal reading of the Y-axis angular velocity register is 0xFEA0, the corresponding signed integer is Let the sensitivity coefficient be a constant. Let the resting zero bias constant be... The spatial domain phase-locked resampling module performs floating-point operations: The output angular velocity components (degrees per second).

[0045] S220. Construct an angular velocity history queue containing the angular velocity component at the current moment and the historical angular velocity component of the previous sampling period, and update the absolute opening angle based on the angular velocity history queue using the second-order Adams-Bashforth explicit numerical integration algorithm.

[0046] The spatial domain phase-locked resampling module declares and maintains a two-dimensional circular floating-point array structure in the static random access memory, which is the angular velocity history queue. New angular velocity components are then acquired. At that time, the spatial domain phase-locked resampling module pushes it to the head of the queue and updates the index of the value remaining in the previous cycle. The spatial domain phase-locked resampling module calls the floating-point arithmetic unit to explicitly recursively update the position information based on the second-order Adams-Bashfors difference equation.

[0047] For example, the second-order Adams-Bashfors explicit numerical integration algorithm defines a step size constant. for (seconds). The iterative equation is: Let the first... One cycle, the previous absolute opening angle for (degrees). Read the angular velocity history queue, current moment. angular velocity (degrees per second), historical moment angular velocity (degrees per second). The spatial domain phase-locked resampling module performs algebraic substitution operations: Analyzing the combination terms yields... Multiplying them gives the step size increment. (degrees). Update the output of the absolute opening angle. (Spend).

[0048] S230. Based on the modulo operation result of the absolute opening angle and the preset discretization step size, generate the spatial domain slot index.

[0049] The spatial domain phase-locked resampling module extracts the discretization step size constant from the configuration register. The spatial domain phase-locked resampling module performs a division operation, dividing the absolute opening angle by the constant, and calls a floor function to truncate the decimal part. The extracted integer value constitutes the spatial domain slot index.

[0050] For example, the preset discretization step size Set as (degrees). Let the absolute aperture angle be obtained using the difference equation. for (degree). The spatial domain phase-locked resampling module executes the formula. Except for legal business Output the integer after rounding. This integer That is, the spatial domain slot index .

[0051] S300. Based on the spatial domain slot index, the feature extraction results of the phase current signal, the acceleration signal in the inertial measurement signal, and the acoustic signal are fused into the current state vector.

[0052] The spatial domain phase-locked resampling module performs frequency domain transformation, downsampling filtering, and structural splicing operations on the three heterogeneous signals respectively, and encapsulates the pure numerical features after removing environmental noise into a column vector in a fixed order to generate feature input entities that characterize the macroscopic physical state and meet the requirements of the multidimensional spatial operation interface.

[0053] S310. Perform a fast Fourier transform on the acoustic signal to extract the acoustic energy integral value of the preset high-frequency contact band.

[0054] The spatial domain phase-locked resampling module inputs a package to the digital signal processing coprocessor. A sliding time frame window for each acoustic signal sampling point. The coprocessor performs a radix-2 fast Fourier transform to convert the time-domain sound pressure sequence into an amplitude-frequency information array. The spatial domain phase-locked resampling module accumulates and sums the squares of the frequency point amplitudes within a preset frequency window, outputting an integral value of the acoustic energy characterizing the contact transient energy.

[0055] For example, let the audio time frame length be... for There are 10 sampling points. The frequency domain amplitude array is denoted as . The preset lower limit frequency of the high-frequency contact band. for (Hertz), corresponding index point Upper bound frequency for (Hertz), corresponding index point The spatial domain phase-locked resampling module executes the accumulation instruction: In the normal state without encountering obstacles, Scalar output is (Dimensionless). When the tailgate contacts an obstacle, the audio signal generates high-frequency oscillations. Scalar output jumps to (Dimensionless).

[0056] S320. Extract the acceleration signal along the push rod axis from the inertial measurement signal.

[0057] The spatial domain phase-locked resampling module inputs the corresponding Z-axis acceleration component from the inertial measurement signal into the iterative register queue of a second-order infinite impulse response digital low-pass filter. After iterative equation calculation, the filter outputs the truncated low-frequency gravity projection component, which is the acceleration signal.

[0058] For example, the time series difference equation of a second-order Butterworth digital low-pass filter is: The forward coefficient is , The feedback coefficient is , .enter After filtering, high-frequency vibrations are suppressed, and the output value is... The acceleration signal is measured in meters per second squared.

[0059] S330. The phase current signal, the acceleration signal, and the acoustic energy integral value are concatenated into vectors to generate the current state vector consisting of three dimensions.

[0060] The spatial domain phase-locked resampling module allocates a continuous 12-byte physical space in the working stack area of ​​the static random access memory to construct a standard floating-point column vector data entity, satisfying the data specification form of the multidimensional statistical distance interface. The spatial domain phase-locked resampling module sequentially writes the phase current signal, acceleration signal, and acoustic energy integral value into this structural space to form the current state vector.

[0061] For example, the current state vector The structure is represented as The system acquires the motor phase current. for (Ampere), acceleration for (meters per second squared), acoustic energy integral value for (Dimensionless). The spatial domain phase-locked resampling module generates column vectors. .

[0062] It should be noted that the current state vector and mean vector Defined as a dimension The column vector. The internal dimensional vector is: In this system, no scalar normalization operation with artificial weights is performed on the different physical field data before merging.

[0063] S400: Obtain the historical baseline mean vector and the historical covariance inverse matrix corresponding to the spatial domain slot index.

[0064] The Mahalanobis distance evaluation and decision module extracts the spatial domain slot index, substitutes it into the linear memory mapping formula, and generates the physical addressing pointer within the second memory block. The Mahalanobis distance evaluation and decision module initiates a bus read request, extracts three mean floating-point numbers and nine inverse matrix floating-point number sequences from the contiguous memory block, and loads them into the high-speed register group to form the historical baseline mean vector and the historical covariance inverse matrix.

[0065] For example, the base address macro is defined as BASE_ADDR = 0x08040000. Let the slot index... for The Mahalanobis distance evaluation and decision-making module calculates the absolute target address of the pointer: TARGET_ADDR = 0x08040000 + 144 × 48. The historical baseline mean vector is then extracted from this address. Simultaneously extract The historical covariance inverse matrix Its internal main diagonal elements contain Off-diagonal elements contain Equal values.

[0066] S500. Determine the spatial distance assessment value based on the current state vector, the historical baseline mean vector, and the historical covariance inverse matrix.

[0067] The Mahalanobis distance evaluation and decision module calls the underlying linear algebra computation library to perform high-order tensor compression and dimensionality reduction operations, so as to eliminate boundary false alarms caused by using independent single scalar comparisons in a multidimensional heterogeneous parameter space, thereby obtaining a normalized pure probability metric.

[0068] S510. Obtain the state residual vector between the current state vector and the historical baseline mean vector.

[0069] The Mahalanobis distance assessment and decision-making module uses Single Instruction Multiple Data (SIMD) extension technology to perform floating-point vector subtraction operation, subtracting the historical baseline mean vector from the current state vector to generate a pure differential entity after extracting and stripping the background physical load, namely the state residual vector.

[0070] For example, let the current state vector be... for Historical baseline mean vector for The Mahalanobis distance evaluation and decision-making module performs subtraction row by row to generate a column vector. The generated column vector is the state residual vector.

[0071] S520. Perform matrix multiplication on the transpose of the state residual vector, the inverse of the historical covariance matrix, and the state residual vector to obtain the spatial distance evaluation value as a dimensionless pure scalar.

[0072] The Mahalanobis distance evaluation and decision module performs register-optimized ternary matrix multiplication operations. The transposed... Row vector entity left multiplication The historical covariance inverse matrix generates the transition. The intermediate row vector; this intermediate row vector is then multiplied right by the original vector. The state residual vector outputs the final single-precision floating-point metric, namely the spatial distance evaluation value.

[0073] For example, the state residual vector Dimensions are Historical covariance inverse matrix First element of the main diagonal Dimensions are .implement At that time, the product of independent components in the first dimension Substituting the dimensions, we can deduce the following: Cross-fusion product term Substituting the dimensions as The dimensions of each multiplication term cancel each other out. (Input) The spatial distance assessment value is output after calculation. for (Dimensionless).

[0074] S600. Based on the comparison result between the spatial distance evaluation value and the anti-pinch trigger threshold, the tailgate anti-pinch reversing action is triggered.

[0075] The Mahalanobis distance assessment and decision module compares the generated spatial distance assessment value with the system's fixed tolerance boundary constant, and uses Boolean logic to output the level of the control pin or trigger the background data update channel to achieve adaptive control response to different operating conditions.

[0076] S610. Obtain the chi-square distribution information interval value corresponding to the three degrees of freedom as the anti-pinch trigger threshold.

[0077] The system maps the three independent parameter dimensions contained in the state vector to the degree of freedom parameter. The chi-square distribution model is used. The Mahalanobis distance evaluation and decision module extracts a numerical constant obtained in advance by solving the cumulative probability integral limit of the chi-square distribution from the static constant segment of the code storage area, and uses this constant as the anti-pinch trigger threshold.

[0078] For example, the cumulative probability confidence level is set to... Solving the integral equation makes the integral equation Limit value that holds The query retrieves this constant as... The anti-pinch trigger threshold In the code, a static assignment of a floating-point constant is performed. (Dimensionless).

[0079] S620. When the spatial distance assessment value is greater than the anti-pinch trigger threshold, the general-purpose input / output pin of the microcontroller is set to a high level to drive the motor to control the full-bridge circuit to reverse.

[0080] The Mahalanobis distance evaluation and decision module calculates the evaluation and judgment statements. When the condition is met, the Mahalanobis distance evaluation and decision module generates a hardware interrupt, writes the control instruction word into the microcontroller's port set / reset register GPIOPORTCBSRR corresponding to the bits for motor enable and direction control, outputs a high level to the integrated dual half-bridge circuit drive control chip, changes the conduction state of the field-effect transistor, and flips the motor closing instruction.

[0081] For example, for the calculation output for The physical state point is compared and confirmed. True. The microcontroller interrupt writes to the PORTC_BSRR register, causing the control port level to toggle. After a very short delay, the motor performs anti-pinch retraction, and the system controls the peak force to be below... Within the (Newton's) safety threshold.

[0082] S630. When the spatial distance evaluation value is not greater than the anti-pinch trigger threshold, the current state vector is pushed into the safety history queue.

[0083] If the conditional statement If true, the Mahalanobis distance evaluation and decision-making module does not send any change commands to the motor drive port. The Mahalanobis distance evaluation and decision-making module triggers a pointer auto-increment operation, updating the current state vector... and the spatial domain slot index The encapsulated data is stored in the security history queue buffer entity located in the first memory block.

[0084] In some embodiments, the dynamic baseline adaptive evolution module reads historical data during system idle periods and uses an exponential smoothing model to update the mean and tolerance covariance parameters in order to adaptively eliminate the cumulative error of the background parameters caused by long-term physical wear and tear of vehicle door hinges and sudden changes in ambient temperature.

[0085] S710. When the tailgate completes its closing motion and the tailgate anti-pinch reversal action is not triggered globally, read the current state vector corresponding to each spatial domain slot index in the safety history queue.

[0086] After the tailgate physical latch is fully closed and a level confirmation signal is issued, a low-priority background thread wakes up the dynamic baseline adaptive evolution module. This module traverses the memory partitions of the safety history queue and extracts a series of current state vectors stored in the current loop according to the associated spatial domain slot index. Data blocks are transferred to the kernel computing region.

[0087] S720. Using the exponentially weighted moving average algorithm, perform a mean update operation on the historical baseline mean vector based on the current state vector to generate a new mean vector.

[0088] The dynamic baseline adaptive evolution module reads the specified learning rate constant parameter, performs a linear combination operation, reduces the extracted old mean vector components by a specific ratio, introduces the new sampled state components by a complementary ratio, and outputs the numerically fine-tuned new mean vector.

[0089] For example, configuring constant learning rate coefficients Read the historical baseline mean vector of slot 119. for Extract the current state vector. for Execute the formula: The first dimension result was calculated independently. (Ampere). Combined output of the new mean vector .

[0090] S730. Generate a residual outer product matrix based on the new mean vector and the current state vector.

[0091] The dynamic baseline adaptive evolution module performs subtraction to obtain the difference column vector, and then performs algebraic operations on the difference column vector using matrix outer product (Kronecker product). Column vectors and The row vectors are multiplied term by term and instantiated in the stack as a vector of size. The residual outer product matrix is ​​output in the form of a floating-point square matrix.

[0092] S740. Using the exponentially weighted moving average algorithm, update the historical covariance matrix based on the residual outer product matrix to generate a new covariance matrix.

[0093] The dynamic baseline adaptive evolution module loads the historical covariance matrix corresponding to the index. Reuse the learning rate coefficient For two sizes Perform item-by-item scaling and addition operations on the matrix entities: The new covariance matrix is ​​then generated to reflect the recent fluctuation width. .

[0094] S750. Perform the Cholsky decomposition and inversion operation on the new covariance matrix to generate a new covariance inverse matrix and overwrite the non-volatile memory.

[0095] The dynamic baseline adaptive evolution module, during its offline background phase, invokes a linear algebra application library algorithm function to perform matrix inversion. It calls the Cholesky numerical decomposition algorithm kernel to decompose the positive definite new covariance matrix into lower triangular matrix factors, performs a backward inference operation to calculate and output the new covariance inverse matrix. The dynamic baseline adaptive evolution module manipulates the page erase / programming timing logic of the flash memory controller to overwrite the original flash memory sector addresses with the generated vector and matrix data entities. The system then completes the cognitive parameter update and immediately resets to hibernation, awaiting the next interrupt wake-up.

[0096] This application provides a tailgate anti-pinch system based on multimodal sensor fusion. In a specific physical implementation, the system includes a multimodal hard synchronization acquisition module, a spatial domain phase-locked resampling module, a Mahalanobis distance evaluation and decision module, and an optional dynamic baseline adaptive evolution module.

[0097] The entire system is physically deployed in a dedicated electronic control unit integrated inside the tailgate electric push rod assembly. Its core is an automotive-grade 32-bit microcontroller based on the ARM® Cortex®-M4F core. This microcontroller integrates a floating-point unit, a sixteen-channel direct memory access controller, and 1 megabyte of non-volatile flash memory and 128 kilobytes of static random access memory.

[0098] The multimodal hard synchronization acquisition module is composed of hardware and firmware. Its hardware includes a milliohm-level precision shunt resistor connected to the low side of the motor drive circuit, a six-axis inertial measurement unit connected to the microcontroller via a serial peripheral interface bus, and a microelectromechanical system (MEMS) microphone that outputs a pulse-code modulated digital stream to the microcontroller via an integrated circuit's built-in audio bus. To achieve strict alignment of the multimodal signals at the timestamp level, the firmware within the module configures an internal hardware timer as the system synchronization heartbeat. The overflow event of this timer is hardware-routed to concurrently trigger the microcontroller's analog-to-digital converter, serial peripheral interface, and integrated circuit's built-in audio bus to perform synchronous sampling. Simultaneously, the direct memory access controller, bypassing the central processing core, automatically transports and encapsulates the raw digital samples of phase current, inertial measurement, and acoustic signals generated by each peripheral into a unified data frame structure, and sequentially writes them into a preset circular buffer in the static random access memory.

[0099] The spatial domain phase-locked resampling module is a functional entity consisting of a library of algorithm functions running on the microcontroller kernel. It receives raw data frames with a unified timestamp provided by the circular buffer. To reconstruct the time-domain signal sequence into a stable physical spatial domain, this module first calls the floating-point unit to perform a second-order Adams-Bashfors explicit numerical integration algorithm on the angular velocity component of the inertial measurement signal to update the absolute opening angle of the tailgate in real time. Next, a modulo operation and rounding operation based on this absolute opening angle and a preset discretization step size are performed to generate a spatial domain slot index for indexing historical baselines. To generate feature inputs for risk assessment, the module performs a fast Fourier transform on the acoustic signal to extract the acoustic energy integral value of a preset high-frequency contact band and applies a digital low-pass filter to the acceleration component of the inertial measurement signal. Finally, the processed phase current signal, acceleration signal, and acoustic energy integral value are concatenated to form a three-dimensional current state vector.

[0100] Based on the spatial domain slot index and current state vector generated by the spatial domain phase-locked resampling module, the Mahalanobis distance assessment and decision module performs core risk determination. This module first uses the spatial domain slot index as an address pointer to read the corresponding historical baseline mean vector and historical covariance inverse matrix from the non-volatile flash memory. Then, the module calls the underlying linear algebra library to determine a dimensionless scalar spatial distance assessment value by performing matrix multiplication operations on the transpose of the state residual vector between the current state vector and the historical baseline mean vector, the historical covariance inverse matrix, and the state residual vector. This assessment value is then compared with a preset anti-pinch trigger threshold based on a three-degree-of-freedom chi-square distribution signal interval. If the spatial distance assessment value is greater than the anti-pinch trigger threshold, the module sends a reversal command to the motor drive circuit by setting the level state of a general-purpose input / output pin; otherwise, the current state vector is pushed into a safe history queue in the static random access memory for subsequent model adaptive updates.

[0101] In one embodiment, after the tailgate completes a full closing motion without triggering the anti-pinch reversal, the dynamic baseline adaptive evolution module is activated as a background firmware task. This module traverses the safety history queue and, for each spatial domain slot index, uses an exponentially weighted moving average algorithm to iteratively update the historical baseline mean vector and historical covariance matrix stored in the non-volatile flash memory based on the current state vector recorded during that stroke. To generate parameters that can be directly used for the next decision calculation, the module further calls a numerical computation library to perform a Cholsky decomposition inversion operation on the updated covariance matrix, and then writes the finally generated new mean vector and new covariance inverse matrix back to the corresponding storage address in the non-volatile flash memory, thereby completing the adaptive evolution closed loop of the entire system in response to vehicle component aging, mechanical wear, and changes in ambient temperature.

[0102] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.

[0103] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A tailgate anti-pinch method based on multimodal sensor fusion, characterized in that, The method includes: Acquire phase current signals, inertial measurement signals, and acoustic signals with a unified timestamp; The absolute opening angle is extracted based on the inertial measurement signal, and the absolute opening angle is discretized into a spatial domain slot index. Based on the spatial domain slot index, the feature extraction results of the phase current signal, the acceleration signal in the inertial measurement signal, and the acoustic signal are fused into the current state vector; Obtain the historical baseline mean vector and historical covariance inverse matrix corresponding to the spatial domain slot index; The spatial distance assessment value is determined based on the current state vector, the historical baseline mean vector, and the historical covariance inverse matrix. Based on the comparison between the spatial distance assessment value and the anti-pinch trigger threshold, the tailgate anti-pinch reversing action is triggered.

2. The tailgate anti-pinch method based on multimodal sensor fusion according to claim 1, characterized in that, The steps of acquiring phase current signals, inertial measurement signals, and acoustic signals with a unified timestamp include: The analog-to-digital converter, serial peripheral interface, and integrated circuit built-in audio bus are triggered concurrently at a fixed frequency using a hardware timer. The phase current signal output by the analog-to-digital converter, the inertial measurement signal output by the serial peripheral interface, and the acoustic signal output by the integrated circuit's built-in audio bus are transferred to a ring buffer in the static random access memory via a direct memory access channel.

3. The tailgate anti-pinch method based on multimodal sensor fusion according to claim 1, characterized in that, The step of extracting the absolute opening angle based on the inertial measurement signal and discretizing the absolute opening angle into a spatial domain slot index includes: Extract the angular velocity component from the inertial measurement signal; Construct an angular velocity history queue that includes the angular velocity component at the current moment and the historical angular velocity components from the previous sampling period; Based on the angular velocity history queue, the absolute opening angle is updated using the second-order Adams-Bashfors explicit numerical integration algorithm; The spatial domain slot index is generated based on the modulo operation result of the absolute opening angle and the preset discretization step size.

4. The tailgate anti-pinch method based on multimodal sensor fusion according to claim 1, characterized in that, The step of fusing the feature extraction results of the phase current signal, the acceleration signal in the inertial measurement signal, and the acoustic signal into a current state vector based on the spatial domain slot index includes: Perform a fast Fourier transform on the acoustic signal to extract the acoustic energy integral value of the preset high-frequency contact band; Extract the acceleration signal along the push rod axis from the inertial measurement signal; The phase current signal, the acceleration signal, and the acoustic energy integral value are concatenated into vectors to generate the current state vector consisting of three dimensions.

5. The tailgate anti-pinch method based on multimodal sensor fusion according to claim 1, characterized in that, The step of determining the spatial distance assessment value based on the current state vector, the historical baseline mean vector, and the historical covariance inverse matrix includes: Obtain the state residual vector between the current state vector and the historical baseline mean vector; The transpose of the state residual vector, the inverse of the historical covariance matrix, and the state residual vector are multiplied by matrix to obtain the spatial distance evaluation value as a dimensionless pure scalar.

6. The tailgate anti-pinch method based on multimodal sensor fusion according to claim 5, characterized in that, The step of triggering the tailgate anti-pinch reversing action based on the comparison result between the spatial distance assessment value and the anti-pinch trigger threshold includes: The chi-square distribution interval value corresponding to the three degrees of freedom is obtained as the anti-pinch trigger threshold; When the spatial distance assessment value is greater than the anti-pinch trigger threshold, the general-purpose input / output pin of the microcontroller is set to a high level to drive the motor to control the full-bridge circuit to reverse. When the spatial distance assessment value is not greater than the anti-pinch trigger threshold, the current state vector is pushed into the safety history queue.

7. The tailgate anti-pinch method based on multimodal sensor fusion according to claim 6, characterized in that, The method further includes: When the tailgate completes its closing motion and the tailgate anti-pinch reversal action is not triggered globally, read the current state vector corresponding to each spatial domain slot index in the safety history queue; Using the exponentially weighted moving average algorithm, the historical baseline mean vector is updated based on the current state vector to generate a new mean vector; Generate a residual outer product matrix based on the new mean vector and the current state vector; Using the exponentially weighted moving average algorithm, the historical covariance matrix is ​​updated based on the residual outer product matrix to generate a new covariance matrix; Perform the Cholsky decomposition inversion operation on the new covariance matrix to generate a new covariance inverse matrix and overwrite the non-volatile memory.

8. A tailgate anti-pinch system based on multimodal sensor fusion, characterized in that, The system includes: The multimodal hard synchronization acquisition module is configured to acquire phase current signals, inertial measurement signals, and acoustic signals with a unified timestamp. The spatial domain phase-locked resampling module is configured to extract the absolute opening angle based on the inertial measurement signal and discretize the absolute opening angle into a spatial domain slot index; the spatial domain phase-locked resampling module is also configured to fuse the feature extraction results of the phase current signal, the acceleration signal in the inertial measurement signal, and the acoustic signal into a current state vector based on the spatial domain slot index; The Mahalanobis distance assessment and decision module is configured to obtain the historical baseline mean vector and the historical covariance inverse matrix corresponding to the spatial domain slot index; the Mahalanobis distance assessment and decision module is also configured to determine the spatial distance assessment value based on the current state vector, the historical baseline mean vector and the historical covariance inverse matrix, and trigger the tailgate anti-pinch reversal action based on the comparison result of the spatial distance assessment value and the anti-pinch trigger threshold.

9. The tailgate anti-pinch system based on multimodal sensor fusion according to claim 8, characterized in that, The system also includes a dynamic baseline adaptive evolution module; The dynamic baseline adaptive evolution module is configured to update the historical baseline mean vector and the historical covariance inverse matrix using an exponentially weighted moving average algorithm when the tailgate completes its closing motion and no anti-pinch action is triggered globally.

10. A non-volatile computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instructions are executed by the microcontroller, the microcontroller performs the tailgate anti-pinch method based on multimodal sensor fusion as described in any one of claims 1 to 7.

Images