An intelligent control method and system for an automotive glass lifter

Abstract

The application relates to the technical field of automobile electronic control, and discloses an intelligent control method and system for an automobile glass lifter, aiming to solve the technical problem that existing anti-pinch control relies on a fixed threshold, cannot adapt to environmental and aging changes, and can only be passively responded. The method comprises the following steps: acquiring multi-modal data of lifter states, indoor and outdoor environments, and short-range obstacles of a window in real time; constructing and updating a dynamic digital twin model representing physical characteristics; generating a multi-dimensional state deviation instruction based on the deviation between model prediction and actual measurement data; fusing the instruction and short-range detection data to generate a hierarchical decision containing early warning, pre-contact and collision type anti-pinch; and dynamically adjusting the motor driving strategy accordingly to realize intelligent lifting and ice breaking functions. The application significantly improves the anti-pinch reliability, safety and intelligent level through predictive protection and full life cycle adaptive calibration.

Description

Technical Field

[0001] This invention belongs to the field of automotive electronic control technology, specifically relating to an intelligent control method and system for automotive window regulators. Background Technology

[0002] With the continuous improvement of automotive electronics and intelligence, the control method of window regulators, as an important component of human-vehicle interaction, has gradually evolved from traditional mechanical switch operation to intelligent sensing and automatic response. Modern vehicles are generally equipped with electric window regulators, where users trigger the raising or lowering action with a button, and the system completes the raising or lowering process according to preset logic. However, such traditional control strategies are essentially still open-loop or simple closed-loop control, lacking the comprehensive perception capability of the in-vehicle environment, user intentions, and external risk factors, making it difficult to achieve coordinated optimization of safety, comfort, and convenience in complex usage scenarios.

[0003] Among these, intelligent control of automotive window regulators focuses on achieving adaptive adjustment of the glass movement process through multi-source sensor information fusion and dynamic decision-making mechanisms. This technological direction aims to overcome the limitations of fixed logic control, enabling the lifting action to autonomously adjust its operating speed, stopping timing, or anti-pinch sensitivity based on real-time operating conditions, thereby improving user experience and avoiding potential safety hazards. Its core lies in building an intelligent control framework capable of understanding contextual situations, identifying abnormal states, and making appropriate responses.

[0004] While some high-end vehicles have incorporated basic anti-pinch functions in existing technologies, these typically rely on sudden current changes or position encoder thresholds to detect obstacles. However, this method suffers from high false alarm rates, slow response times, and an inability to differentiate between various obstacle types, such as children's arms, pets, or flexible fabrics. Furthermore, existing systems generally lack the ability to proactively recognize user intent; for example, they cannot determine the true purpose of short and long presses, nor can they dynamically adjust the lifting strategy based on factors like temperature differences between the vehicle's interior and exterior, rainfall, vehicle speed, or occupant status (e.g., whether there are passengers in the rear seats). Moreover, in emerging interaction modes such as multi-window linkage, voice commands, or remote control, traditional architectures, due to isolated modules and rigid logic, struggle to achieve coordinated responses and priority arbitration, leading to control delays, action conflicts, and even functional failures. These deficiencies are particularly pronounced in critical scenarios such as extreme weather, emergency escapes, or child misoperation, severely restricting the safety and intelligence of window lifting systems. Therefore, a smart control method and system with environmental perception, intent understanding, and adaptive execution capabilities is urgently needed to address these issues. Summary of the Invention

[0005] The technical problem this invention aims to solve is to provide an intelligent control method and system for automotive window regulators, aiming to overcome the technical defects of existing automotive window regulator anti-pinch control strategies, which rely on fixed thresholds, cannot adapt to environmental changes and mechanical aging, and can only perform passive post-collision responses. The control logic in existing technologies is prone to misjudgment or missed judgment when faced with dynamic changes in friction caused by temperature changes, dust contamination, and aging of sealing strips, thereby reducing the reliability and safety of the anti-pinch function. This invention, by constructing a dynamic digital twin model of the regulator system, integrates information from vehicle interior and exterior environmental perception and window proximity detection, achieving a fundamental shift from passive response to predictive protection, and possessing adaptive calibration capabilities throughout the entire lifecycle.

[0006] To address the aforementioned technical problems, this invention provides an intelligent control method for automotive window regulators, comprising the following steps:

[0007] Real-time acquisition of multimodal fusion perception data, including elevator assembly status data, vehicle interior and exterior environment status data, and near-range obstacle data in the window area;

[0008] Based on the state data of the lift assembly and the state data of the vehicle's internal and external environment, a dynamic digital twin model characterizing the physical characteristics of the lift assembly is constructed and updated in real time. The dynamic digital twin model is used to output predicted operating parameters corresponding to the current system state.

[0009] Based on the predicted operating parameters output by the dynamic digital twin model, a deviation quantification analysis is performed between the real-time acquired elevator assembly status data and the parameters to generate a real-time multi-dimensional status deviation command.

[0010] By integrating multi-dimensional state deviation commands with near-range obstacle data in the window area, the final elevator control decision command is generated.

[0011] Based on the control decision command of the lifter, the drive strategy of the motor driving the lifter assembly is dynamically adjusted to achieve intelligent lifting control of the automotive windows.

[0012] As one embodiment of the present invention, real-time acquisition of multimodal fusion sensing data specifically includes:

[0013] The real-time armature current data of the drive motor is acquired by a Hall effect current sensor integrated inside the elevator drive motor or its circuit, at a sampling frequency of not less than 1 kilohertz.

[0014] The real-time terminal voltage data of the drive motor is obtained by a voltage acquisition unit connected in parallel with the power supply circuit of the drive motor.

[0015] The real-time angular position and angular velocity data of the motor rotor are obtained by an orthogonal encoder magnetic rotary encoder installed on the output shaft of the drive motor. The resolution of the rotary encoder is no less than 1,200 pulses per revolution.

[0016] Real-time temperature and humidity data of the vehicle's interior environment are obtained by temperature and humidity sensors installed in the vehicle's cockpit.

[0017] Rainfall and ambient temperature data of the external environment are obtained by integrating rain sensors and external temperature sensors on the windshield or exterior rearview mirrors of the vehicle.

[0018] An array of capacitive or active infrared proximity sensors embedded in the top sealing strip of the window glass acquires short-range detection data on the presence and proximity of obstacles along the window's closing path.

[0019] Furthermore, after acquiring the aforementioned data, the method also includes a data preprocessing step, specifically including aligning all collected data with timestamps, filtering out noise using a Kalman filter algorithm, and filling missing data points with linear interpolation based on adjacent data, thereby forming a multidimensional state vector in a unified time domain for use by the subsequent dynamic digital twin model.

[0020] As one embodiment of the present invention, constructing and updating a dynamic digital twin model in real time specifically includes:

[0021] First, during the system initialization phase, a mechanism-based state-space model of the lifting assembly is established. This model includes voltage balance equations describing the electrical characteristics of the motor and torque balance equations describing the mechanical characteristics of the motor and load. The voltage balance equations take into account armature resistance, armature inductance, back electromotive force constant, and motor angular velocity. The torque balance equations take into account motor moment of inertia, viscous friction coefficient, load torque, and motor electromagnetic torque. The load torque is modeled as a nonlinear function of glass position, running speed, sealing strip friction coefficient, and guide rail friction coefficient.

[0022] Secondly, during operation, the preprocessed multidimensional state vector is input into an extended Kalman filter. The extended Kalman filter uses the state-space model as the prediction model to estimate and correct the system's state variables in real time. The state variables include the glass's precise linear displacement, linear velocity, motor angular velocity, and armature current, thereby completing the real-time state update of the digital twin model.

[0023] Next, an adaptive calibration unit for model parameters is established. After each complete lifting cycle, this unit analyzes the cumulative error between the model prediction and the actual measurement value throughout the entire stroke. The key parameters in the state space model, such as the friction coefficient of the sealing strip and the back electromotive force constant of the motor, are finely adjusted and updated using the gradient descent algorithm to compensate for the drift of system characteristics caused by mechanical wear, material aging or changes in ambient temperature.

[0024] Furthermore, the specific process of generating multi-dimensional state deviation commands by performing deviation quantification analysis on the predicted operating parameters output by the dynamic digital twin model and the real-time acquired elevator assembly status data includes:

[0025] In each control cycle, the dynamic digital twin model predicts the expected armature current, expected motor angular velocity, and expected glass linear velocity at the next moment based on the current state and the motor drive command at the next moment.

[0026] The armature current, motor angular velocity, and glass linear velocity obtained by converting the motor angular velocity after real-time measurement and preprocessing are compared one by one with the expected values ​​predicted by the model, and the normalized residuals of each physical quantity are calculated.

[0027] The normalized residuals are linearly combined using preset weighting coefficients to calculate a comprehensive state deviation scalar value. This scalar value is the multidimensional state deviation command, and its magnitude directly reflects the degree to which the actual operating state of the elevator assembly deviates from its normal physical model.

[0028] Furthermore, the specific process of integrating multi-dimensional state deviation commands with near-range obstacle data in the window area to generate the final elevator control decision command includes:

[0029] A dynamic anti-pinch decision threshold is set. The baseline value of the dynamic anti-pinch decision threshold is dynamically adjusted based on the ambient temperature outside the vehicle and the system aging factor output by the adaptive calibration unit of the model parameters. Under low temperature or high aging factor conditions, the baseline value of the threshold is appropriately increased to reduce the probability of false triggering.

[0030] The state deviation scalar value is compared with the dynamic anti-pinch decision threshold. If the state deviation scalar value is continuously greater than the dynamic anti-pinch decision threshold for a preset time window, a collision-type anti-pinch command is generated.

[0031] In parallel, the output signals of the proximity sensor array are processed in real time. If the signal strength indicates that an object has entered the preset warning distance, a warning anti-pinch command is generated; if the signal strength indicates that the object has entered the danger distance, a contact-before-pinch command is generated.

[0032] The above instructions are arbitrated by a decision logic unit. The pre-contact anti-pinch instruction has the highest priority, followed by the collision-type anti-pinch instruction, and then the warning-type anti-pinch instruction. The decision logic unit outputs a unique and final elevator control decision instruction.

[0033] As one embodiment of the present invention, dynamically adjusting the drive strategy of the drive motor according to the lifting device control decision command specifically includes:

[0034] When a regular lifting command is received, a closed-loop speed control strategy is adopted, which adjusts the duty cycle of the pulse width modulation signal to make the glass lift and lower in a smooth speed curve.

[0035] When a warning anti-pinch command is received, immediately switch to low-speed crawling mode and activate audible or visual warning signals;

[0036] When a collision-type anti-pinch command or a contact-before-pinch command is received, an emergency reversal strategy is immediately executed, controlling the H-bridge of the motor drive circuit to conduct in reverse with maximum current, so that the glass drops at maximum acceleration by a preset safe distance, such as 150 millimeters.

[0037] When the vehicle's external temperature sensor detects that the temperature is below freezing, and the initial current and state deviation scalar values ​​when the motor starts are both higher than a certain multiple of the normal values, the ice-breaking mode is automatically activated. This mode outputs a drive signal consisting of a high-amplitude, narrow-pulse-width, alternating polarity voltage pulse sequence to effectively break the ice layer between the glass and the sealing strip without damaging the motor.

[0038] To achieve the above method, the present invention also provides an intelligent control system for automotive window regulators, the system comprising:

[0039] A multimodal perception module, which integrates a lifter status perception unit, an in-vehicle and out-of-vehicle environment perception unit, and a window proximity perception unit, is used to comprehensively collect real-time data of the lifter assembly, vehicle environment, and window area.

[0040] A central processing unit that connects to the multimodal sensing module for data processing;

[0041] A memory that stores instructions that can be executed by a central processing unit;

[0042] When the central processing unit executes instructions, it implements any of the above-mentioned intelligent control methods for automotive window regulators.

[0043] In one specific embodiment, the system is configured with multiple functional modules to perform a specific task, the functional modules including:

[0044] The multimodal perception module is configured to acquire in real time the armature current, terminal voltage, angular position, and angular velocity of the lift assembly, as well as the temperature and humidity of the vehicle's interior and exterior environment, rainfall, and obstacle proximity information in the window area through a Hall effect current sensor, voltage acquisition unit, rotary encoder, temperature and humidity sensor, rain sensor, and proximity sensor array.

[0045] A dynamic digital twin modeling module, connected to a multimodal sensing module, is configured to receive and process sensing data. Internally, it runs a state estimation algorithm based on extended Kalman filtering and a parameter adaptive calibration algorithm based on gradient descent to construct, update and calibrate the state space model of the elevator assembly in real time, and output predicted system operating parameters.

[0046] A predictive anti-pinch decision module, connected to a dynamic digital twin modeling module and a multimodal perception module, is configured to compare model prediction parameters with actual measurement parameters in real time, calculate a scalar value of state deviation, and fuse obstacle information from the near-field perception unit. Based on a set of dynamic threshold and priority arbitration logic, it generates the final control decision command.

[0047] An adaptive motor drive module receives instructions from a predictive anti-pinch decision module. It has a built-in pulse width modulation signal generator and a motor drive strategy library. It is configured to select and execute specific drive strategies, including regular lifting, low-speed creeping, emergency reversal, and ice-breaking operation, according to different received instructions. It precisely controls the rotation of the lift motor through an H-bridge drive circuit.

[0048] The dynamic digital twin modeling module, predictive anti-pinch decision module, and adaptive motor drive module all function through a central processing unit (CPU) executing specific program segments stored in memory. The CPU employs an automotive-grade microcontroller with a floating-point unit, multiple timers, and analog-to-digital converter channels. Data exchange between modules occurs via a controller area network (CLAN) bus or an internal integrated circuit bus.

[0049] In summary, this application includes at least one of the following beneficial technical effects:

[0050] First, it achieves a paradigm shift from passive collision detection to active predictive protection. By predicting the normal behavior of the system through a dynamic digital twin model, this invention can identify anomalies through state deviation and short-range perception when an obstacle exerts a small force, or even before physical contact occurs. This greatly improves the sensitivity and response speed of anti-pinch protection and significantly reduces the risk of pinching injuries.

[0051] Secondly, it possesses a high degree of environmental and state adaptability. The digital twin model and its parameter adaptive calibration mechanism of this invention can compensate in real time for changes in system characteristics caused by factors such as changes in ambient temperature, voltage fluctuations, mechanical wear, and aging of sealing strips, ensuring that the accuracy of anti-pinch decisions remains consistent throughout the entire vehicle lifecycle, and completely solving the problem of low reliability of traditional fixed threshold methods.

[0052] Third, it expands the intelligent functions of the window regulator. Based on a deep understanding of the system status and multimodal perception, this invention can realize advanced functions such as intelligent ice-breaking operation, improving the user experience in adverse weather conditions, which is impossible with existing technologies.

[0053] Fourth, predictive health management capabilities have been introduced. By tracking the drift trends of digital twin model parameters over a long period, this system can assess and predict the health status of the lift assembly, issuing early warnings before critical components such as motors or transmission mechanisms experience severe wear. This provides data support for preventative maintenance and improves the overall reliability of the vehicle. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0055] Figure 2 This is a schematic diagram of the core principle framework for the construction and adaptive calibration of the dynamic digital twin model of this invention;

[0056] Figure 3 This is a logical flowchart of the predictive anti-pinch decision-making and multimodal perception fusion of the present invention; Detailed Implementation

[0057] This invention provides an intelligent control method and system for automotive window regulators, aiming to address the technical shortcomings of existing anti-pinch control strategies, such as reliance on fixed thresholds, inability to adapt to environmental changes and mechanical aging, and the limitation to only passive post-collision responses. Existing control logic is prone to misjudgment or missed detection when faced with dynamic changes in friction caused by temperature variations, dust contamination, and aging of sealing strips, thereby reducing the reliability and safety of the anti-pinch function. This invention constructs a dynamic digital twin model of the regulator system, integrating information from both inside and outside the vehicle's environment and near-window detection, achieving a fundamental shift from passive response to predictive protection, and possessing adaptive calibration capabilities throughout its entire lifecycle.

[0058] Reference Figures 1 to 3The intelligent control method for automotive window regulators in this application includes the following steps: real-time acquisition of multimodal fusion sensing data; construction and real-time updating of a dynamic digital twin model characterizing the physical characteristics of the regulator assembly based on the multimodal fusion sensing data; deviation quantification analysis of the predicted operating parameters output by the model and the real-time acquired regulator assembly state data to generate multidimensional state deviation commands; fusion of the multidimensional state deviation commands with near-range obstacle data in the window area to generate the final regulator control decision command; and dynamic adjustment of the drive strategy of the motor driving the regulator assembly according to the control decision command to achieve intelligent lifting control of the automotive window.

[0059] First, real-time acquisition of multimodal fusion sensing data specifically includes: acquiring real-time armature current data of the drive motor through a Hall effect current sensor integrated inside the lifter drive motor or its circuitry, at a sampling frequency of not less than 1kHz. Specifically, real-time terminal voltage data of the drive motor is acquired through a voltage acquisition unit connected in parallel with the drive motor power supply circuit; real-time angular position and angular velocity data of the motor rotor are acquired through an orthogonal encoder magnetic rotary encoder mounted on the drive motor output shaft, with a resolution of not less than 1200 pulses per revolution; and real-time temperature and humidity data of the vehicle interior environment are acquired through temperature and humidity sensors installed in the vehicle's cockpit.

[0060] In addition, rain and ambient temperature data of the outside environment are obtained by rain sensors and external temperature sensors integrated into the windshield or exterior rearview mirrors of the vehicle; and a set of capacitive or active infrared proximity sensor arrays embedded in the top sealing strip of the window glass are used to obtain short-range detection data on whether there are obstacles in the window closing path and their proximity.

[0061] After acquiring the aforementioned data, the intelligent control method for automotive window regulators also includes a data preprocessing step. This specifically involves timestamping all collected data, filtering out noise using a Kalman filter algorithm, and filling missing data points with linear interpolation based on adjacent data, thereby forming a multidimensional state vector in a unified time domain for use in subsequent dynamic digital twin models.

[0062] Timestamp alignment is achieved through hardware synchronization trigger signals, ensuring that all sensor data are tagged at the same physical moment; the Kalman filter algorithm adopts the standard discrete-time form, and its process noise covariance matrix and observation noise covariance matrix are pre-calibrated according to the sensor type; linear interpolation filling is only enabled when consecutive missing points do not exceed three sampling periods, otherwise it is judged as a sensor fault and a diagnostic flag is triggered.

[0063] The construction and real-time updating of the dynamic digital twin model specifically includes three sub-processes. First, in the system initialization phase, a mechanism-based state-space model of the elevator assembly is established. This model includes voltage balance equations describing the electrical characteristics of the motor and torque balance equations describing the mechanical characteristics of the motor and load. The voltage balance equations are expressed as: ,in, This is the motor terminal voltage. For armature resistance, For armature inductance, For armature current, The back electromotive force constant is... This represents the angular velocity of the motor.

[0064] The torque balance equation is expressed as: ,in, The moment of inertia of the motor. The torque constant is The coefficient of viscous friction, For load torque, For glass line displacement, For the linear velocity of the glass, The coefficient of friction of the sealing strip. The coefficient of friction of the guide rail is given.

[0065] Load torque It is modeled as a nonlinear function, and its form is: ,in, Characterizing position-dependent static friction components, Characterizing the velocity-dependent kinetic friction components, A coupling term characterizing the aging of materials and their environmental impact.

[0066] Secondly, during operation, the preprocessed multidimensional state vector is input into an extended Kalman filter. This extended Kalman filter uses the state-space model as the prediction model to estimate and correct the system's state variables in real time. State variables include the precise linear displacement and linear velocity of the glass, the angular velocity of the motor, and the armature current. The state prediction step of the extended Kalman filter is executed based on the discretized state-space model, and the observation update step uses measured armature current, terminal voltage, and angular velocity data for correction. The Jacobian matrix of the filter is calculated online in each control cycle to ensure the linearization accuracy of the nonlinear system. Through this process, the real-time state update of the digital twin model is completed, ensuring that the model output always closely approximates the actual behavior of the physical system.

[0067] Next, an adaptive calibration unit for model parameters is established. This unit analyzes the cumulative error between the model's predicted values ​​and the actual measured values ​​throughout the entire stroke after each complete rise and fall cycle. The cumulative error is defined as the root mean square error between the predicted current and the measured current over the entire stroke. If this error exceeds a preset threshold, the parameter fine-tuning process is initiated.

[0068] The key parameters in the state-space model are updated using the gradient descent algorithm, and the update formula is as follows: ,in, This represents the set of parameters to be calibrated, including the friction coefficient of the sealing strip. With the back electromotive force constant of the motor , For learning rate, The loss function is defined as the integral of the squared error between the predicted current and the measured current.

[0069] Learning rate The system is dynamically adjusted based on its aging rate. The initial value is set to 0.01, and it decays exponentially with the cumulative number of adjustments to prevent noise introduced by over-calibration in later stages. This mechanism compensates for system characteristic drift caused by mechanical wear, material aging, or changes in ambient temperature, maintaining the long-term effectiveness of the model.

[0070] The specific process of generating multi-dimensional state deviation commands by performing deviation quantification analysis on the predicted operating parameters output by the dynamic digital twin model and the real-time acquired elevator assembly status data is as follows: In each control cycle, the dynamic digital twin model, based on the current state (including glass line displacement), performs a multi-dimensional state deviation command generation analysis. linear velocity Motor angular velocity and armature current The system uses the discretized state-space equations to perform a one-step forward simulation, based on the motor drive command (i.e., PWM duty cycle or target voltage) and the next time step, to predict the desired armature current, desired motor angular velocity, and desired glass linear velocity at the next time step. Specifically, the prediction process is achieved through the following steps:

[0071] 1. Input the current state variable into the state-space model;

[0072] 2. Substitute the motor drive command from the next moment into the voltage balance equation as the input. Calculate the current change at the next moment;

[0073] 3. Using the torque balance equation Calculate the angular acceleration at the next moment;

[0074] 4. Integrate the angular velocity to obtain the angular velocity at the next moment, and convert it into the glass linear velocity through the transmission ratio.

[0075] The normalized residual is defined as: ,in, Let i be the measured value of the i-th physical quantity. For its predicted value, Its historical standard deviation is used to eliminate the influence of dimensions.

[0076] The normalized residuals are linearly combined using preset weighting coefficients to calculate a comprehensive state deviation scalar value: The formula for calculating the comprehensive state deviation scalar value D is as follows: ,in:

[0077] The current residual reflects the deviation between the measured current and the predicted current.

[0078] The velocity residual reflects the deviation between the measured angular velocity (or linear velocity) and the predicted velocity.

[0079] The positional residual reflects the deviation between the glass displacement calculated from the angular position and the predicted displacement.

[0080] , , For the weighting coefficients, satisfying .

[0081] The calculation methods for each residual are as follows:

[0082] 1. Normalized residuals ,in This represents the standard deviation of the physical quantity over a statistical window in the past.

[0083] 2. Current residual Calculation based on armature current;

[0084] 3. Velocity residual Calculations based on motor angular velocity or glass linear velocity;

[0085] 4. Location residuals Linear displacement calculation based on angular position.

[0086] The dynamic adjustment logic for the weighting coefficients is as follows:

[0087] 1. In low-temperature environments, such as below 0°C, increase If the weight is increased to 0.5, the sealing strip may harden and friction may increase due to low temperature, making the current response more sensitive.

[0088] 2. In environments with high aging factors (such as...) Under these conditions, appropriately increase Weights, which may become sluggish in mechanical response due to aging;

[0089] 3. Improve during the start-up or load change phase. Weights are added to enhance position tracking capabilities.

[0090] 4. Weight adjustment is achieved by querying a preset working condition-weight mapping table, which is obtained based on experimental calibration or simulation optimization.

[0091] The specific process of generating the final lifter control decision command by integrating multi-dimensional state deviation commands and near-range obstacle data in the window area includes: setting a dynamic anti-pinch decision threshold. The baseline value of this threshold is dynamically adjusted based on the external ambient temperature and the system aging factor output by the model parameter adaptive calibration unit. The system aging factor is defined as the ratio of the current friction coefficient of the sealing strip to its initial value, and its value ranges from 1 to 3. When the ambient temperature is below zero degrees Celsius or the aging factor is greater than 1.5, the threshold baseline value is increased by 20% to reduce the probability of false triggering under low temperature and high friction conditions.

[0092] The state deviation scalar value is compared with the dynamic anti-pinch decision threshold. If the state deviation scalar value is continuously greater than the threshold for five consecutive control cycles, a collision-type anti-pinch command is generated.

[0093] The output signals of the proximity sensor array are processed in parallel and in real time. This array consists of six capacitive sensors evenly distributed along the top of the window frame. Each sensor outputs an analog voltage signal, the amplitude of which monotonically increases as the distance to the obstacle decreases. Two distance thresholds are set: a warning distance of 30 mm and a danger distance of 10 mm. If the signal strength of any sensor indicates that an object has entered the warning distance, a pre-contact anti-pinch command is generated; if the signal strength indicates that the object has entered the danger distance, a pre-contact anti-pinch command is generated.

[0094] The aforementioned instructions are arbitrated by a decision logic unit. This unit employs a priority encoding mechanism, with the pre-contact anti-pinch instruction having the highest priority, followed by collision-type anti-pinch instructions, and then warning-type anti-pinch instructions. When multiple instructions exist simultaneously, only the instruction with the highest priority is output as the final elevator control decision instruction. This decision logic unit also has an instruction debouncing function, requiring that the same type of instruction persist for at least 20 milliseconds before confirmation, in order to filter out transient interference.

[0095] The dynamic adjustment of the drive motor's drive strategy based on the lift control decision command specifically includes: when a regular lift command is received, a closed-loop speed control strategy is adopted. This strategy adjusts the duty cycle of the pulse width modulation signal through a proportional-integral controller, so that the glass lifts and lowers with a smooth speed curve. The target speed curve is S-shaped, with the rising and falling segments symmetrical, and the maximum speed is limited to 150 millimeters per second.

[0096] Upon receiving a warning-type anti-pinch command, the system immediately switches to low-speed crawl mode. At this time, the target speed decreases to 30 millimeters per second, and a yellow warning icon on the dashboard is activated, accompanied by a short beep lasting two seconds. If the obstacle is removed during crawling, normal lifting and lowering automatically resumes; if the obstacle remains, manual intervention from the user is required.

[0097] Upon receiving a collision-type anti-pinch command or a pre-contact anti-pinch command, the emergency reversal strategy is immediately executed. The H-bridge controlling the motor drive circuit is reverse-biased with the maximum permissible current, causing the glass to descend by 150 mm at maximum acceleration. After the reversal is complete, the system enters a locked state, preventing automatic reassignment until the user presses and holds the descent button for more than three seconds to confirm safety.

[0098] When the vehicle's external temperature sensor detects a temperature below freezing, and the initial current and state deviation scalar values ​​upon motor startup are both higher than 1.8 times the normal values, the ice-breaking mode is automatically activated. This mode outputs a drive signal consisting of a high-amplitude, narrow-pulse-width, alternating polarity voltage pulse sequence. Each pulse is 5 milliseconds wide, with an amplitude of 120% of the rated voltage, alternating between positive and negative pulses at 10-millisecond intervals, and a duration of 500 milliseconds. This design utilizes the combined effect of thermal stress and mechanical impact to break up the ice layer while preventing motor overheating.

[0099] To implement the above method, the present invention also provides an intelligent control system for automotive window regulators. This system includes a multimodal sensing module, a central processing unit (CPU), and a memory. The multimodal sensing module integrates a regulator status sensing unit, an in-vehicle / outside-vehicle environment sensing unit, and a window proximity sensing unit, for comprehensively collecting real-time data on the regulator assembly, the vehicle environment, and the window area. The CPU is connected to the multimodal sensing module, and the memory stores instructions executable by the CPU. When the CPU executes the instructions, it implements any of the aforementioned intelligent control methods for automotive window regulators.

[0100] In one specific embodiment, the system is configured with multiple functional modules to perform specific tasks. These functional modules include: a multimodal perception module, configured to acquire in real-time armature current, terminal voltage, angular position, and angular velocity of the lift assembly, as well as ambient temperature and humidity, rainfall, and obstacle proximity information in the window area, through a Hall effect current sensor, voltage acquisition unit, rotary encoder, temperature and humidity sensor, rain sensor, and proximity sensor array; and a dynamic digital twin modeling module, connected to the multimodal perception module, configured to receive and process the perceived data, internally running a state estimation algorithm based on extended Kalman filtering and a parameter adaptive calibration algorithm based on gradient descent to construct, update, and calibrate the state-space model of the lift assembly in real time, and output predicted system operating parameters. The system comprises: a predictive anti-pinch decision module, connected to the dynamic digital twin modeling module and the multimodal perception module; a predictive anti-pinch decision module, configured to compare model prediction parameters with actual measurement parameters in real time, calculate the state deviation scalar value, and fuse obstacle information from the near-field perception unit; and an adaptive motor drive module, which receives instructions from the predictive anti-pinch decision module, has a built-in pulse width modulation signal generator and motor drive strategy library, and is configured to select and execute specific drive strategies, including regular lifting, low-speed crawling, emergency reversal, and ice-breaking operation, according to different received instructions, and precisely control the rotation of the lift motor through an H-bridge drive circuit.

[0101] The dynamic digital twin modeling module, predictive anti-pinch decision module, and adaptive motor drive module all function through the central processing unit (CPU) executing specific program segments stored in memory. The CPU employs an automotive-grade microcontroller with a floating-point unit, multiple timers, and analog-to-digital converter channels. Data interaction between modules is conducted via a controller area network (CLAN) bus at a baud rate of 500 kilobits per second. Message frames use a standard identifier format to ensure real-time performance and reliability.

[0102] This embodiment achieves intelligent, predictive, and adaptive control of automotive window regulators through the aforementioned methods and system architecture. Its core lies in the construction and maintenance of a dynamic digital twin model, which not only reflects the current physical state of the system but also tracks long-term degradation trends through a parameter adaptation mechanism. Deep fusion of multimodal perception data enables anti-pinch decisions to possess both high sensitivity and robustness, effectively avoiding the failure problems of traditional methods under complex operating conditions. Furthermore, the introduction of advanced functions such as ice-breaking mode significantly enhances user experience and system usability. The entire solution fully complies with automotive-grade safety and reliability requirements and can be deployed in existing automotive electronic architectures without significantly increasing hardware costs.

[0103] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0104] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A smart control method for an automotive window regulator, characterized in that, include: Real-time acquisition of multimodal fusion sensing data, including real-time armature current data of the drive motor, real-time terminal voltage data of the drive motor acquired by the voltage acquisition unit, real-time angular position and angular velocity data of the motor rotor, real-time temperature and humidity data of the in-vehicle environment, rainfall and ambient temperature data of the outside environment, and short-range detection data on whether there are obstacles and their proximity on the path of closing the windows. The multimodal fusion sensing data is preprocessed to form a multidimensional state vector in a unified time domain. During the system initialization phase, a state-space model of the lifting assembly based on voltage balance equations and torque balance equations is established. During operation, based on the state-space model and the multi-dimensional state vector, the system's state variables are estimated and corrected in real time using an extended Kalman filter. After each complete lifting cycle, the cumulative error between the model's predicted values ​​and the actual measured values ​​is analyzed using a model parameter adaptive calibration unit. The key parameters in the state-space model, including the sealing strip friction coefficient and the motor back EMF constant, are fine-tuned and updated using a gradient descent algorithm. This constructs and updates a dynamic digital twin model characterizing the physical properties of the lifting assembly in real time. This dynamic digital twin model is used to output predicted operating parameters corresponding to the current system state. Based on the predicted operating parameters output by the dynamic digital twin model, a deviation quantification analysis is performed on the armature current, motor angular velocity and glass linear velocity acquired in real time. A comprehensive state deviation scalar value is generated by linearly combining the normalized residuals, which serves as a multidimensional state deviation command. The state deviation scalar value is compared with the dynamic anti-pinch decision threshold, and combined with the warning signal and danger signal output by the proximity sensor array, the final elevator control decision command is generated after priority arbitration. Based on the control decision command of the lifter, the drive strategy of the motor driving the lifter assembly is dynamically adjusted to realize intelligent lifting control of the automotive glass. The specific process of integrating multi-dimensional state deviation commands with near-range obstacle data in the window area to generate the final elevator control decision commands includes: A dynamic anti-pinch decision threshold is set, and the baseline value of the dynamic anti-pinch decision threshold is dynamically adjusted according to the system aging factor output by the adaptive calibration unit of the external ambient temperature and model parameters. The state deviation scalar value is compared with the dynamic anti-pinch decision threshold. If the state deviation scalar value is continuously greater than the dynamic anti-pinch decision threshold within a continuous preset time window, a collision-type anti-pinch command is generated. In parallel, the output signals of the proximity sensor array are processed in real time. If the signal strength indicates that an object has entered the preset warning distance, a warning anti-pinch command is generated; if the signal strength indicates that the object has entered the danger distance, a contact-before-pinch command is generated. The above instructions are arbitrated by a decision logic unit. The pre-contact anti-pinch instruction has the highest priority, followed by the collision-type anti-pinch instruction, and then the warning-type anti-pinch instruction. The decision logic unit outputs a unique and final elevator control decision instruction.

2. The intelligent control method for automotive window regulators according to claim 1, characterized in that, After acquiring the multimodal fusion sensing data, a data preprocessing step is also included, which includes: Timestamp alignment is performed on all collected data; Noise removal is achieved using the Kalman filter algorithm; Missing data points are filled using linear interpolation based on adjacent data. A multi-dimensional state vector is formed under a unified time domain, which is then used by the dynamic digital twin model.

3. The intelligent control method for automotive window regulators according to claim 1, characterized in that, Building and updating dynamic digital twin models in real time specifically includes: During the system initialization phase, a state space model of the elevator assembly based on voltage balance equations and torque balance equations is established. This model includes voltage balance equations describing the electrical characteristics of the motor and torque balance equations describing the mechanical characteristics of the motor and load. During operation, a multidimensional state vector is input into an extended Kalman filter. The extended Kalman filter uses the state-space model as the prediction model to estimate and correct the system's state variables in real time. The state variables include the glass's precise linear displacement, linear velocity, motor angular velocity, and armature current. After each complete rise and fall cycle, the cumulative error between the model prediction and the actual measurement is analyzed by the model parameter adaptive calibration unit, and the key parameters in the state space model are fine-tuned and updated by the gradient descent algorithm.

4. The intelligent control method for automotive window regulators according to claim 1, characterized in that, The specific process of generating multi-dimensional state deviation commands by performing deviation quantification analysis on the predicted operating parameters output by the dynamic digital twin model and the real-time acquired elevator assembly status data includes: In each control cycle, the dynamic digital twin model predicts the expected armature current, expected motor angular velocity, and expected glass linear velocity at the next moment based on the current state and the motor drive command at the next moment. The armature current, motor angular velocity, and glass linear velocity obtained by converting the motor angular velocity after real-time measurement and preprocessing are compared one by one with the expected values ​​predicted by the model, and the normalized residuals of each physical quantity are calculated. The normalized residuals are linearly combined using preset weighting coefficients to calculate a comprehensive state deviation scalar value, which serves as the multidimensional state deviation instruction.

5. The intelligent control method for automotive window regulators according to claim 1, characterized in that, The dynamic adjustment of the drive motor's drive strategy based on the lift control decision command specifically includes: When a regular lifting command is received, a closed-loop speed control strategy is adopted, which adjusts the duty cycle of the pulse width modulation signal to make the glass lift and lower in a smooth speed curve. When a warning anti-pinch command is received, immediately switch to low-speed crawling mode and activate audible or visual warning signals; When a collision-type anti-pinch command or a contact-before-pinch command is received, an emergency reversal strategy is immediately executed, controlling the H-bridge of the motor drive circuit to conduct in reverse with maximum current, so that the glass drops a preset safe distance with maximum acceleration. When the vehicle's external temperature sensor detects that the temperature is below freezing, and the initial current and state deviation scalar values ​​when the motor starts are both higher than 1.8 times the normal values, the ice-breaking mode is automatically activated. This mode outputs a drive signal consisting of a high-amplitude, narrow-pulse-width, alternating polarity voltage pulse sequence.

6. An intelligent control system for an automotive window regulator, applied to the intelligent control method for an automotive window regulator as described in any one of claims 1 to 5, characterized in that, include: The multimodal perception module is used to acquire multimodal fusion perception data in real time. The multimodal fusion perception data includes elevator assembly status data, vehicle interior and exterior environment status data, and near-range obstacle data in the window area. The dynamic digital twin modeling module is used to construct and update a dynamic digital twin model that characterizes the physical characteristics of the lift assembly based on the lift assembly status data and the vehicle's internal and external environmental status data. The dynamic digital twin model is used to output predicted operating parameters corresponding to the current system status. The predictive anti-pinch decision module is used to perform deviation quantification analysis between the predicted operating parameters output by the dynamic digital twin model and the real-time acquired elevator assembly status data to generate a real-time multi-dimensional status deviation command. The multi-dimensional status deviation command is then fused with the window area near-range obstacle data to generate the final elevator control decision command. An adaptive motor drive module is used to dynamically adjust the drive strategy of the motor driving the lift assembly according to the lift control decision command, so as to realize intelligent lifting control of the automotive glass.

7. The intelligent control system for automotive window regulators according to claim 6, characterized in that, The multimodal sensing module includes: Hall effect current sensor, used to acquire real-time armature current data of drive motor at a sampling frequency of not less than 1kHz; voltage acquisition unit, used to acquire real-time terminal voltage data of drive motor; orthogonal encoder magnetic rotary encoder, used to acquire real-time angular position and angular velocity data of motor rotor, with a resolution of not less than 1200 pulses per revolution; in-vehicle temperature and humidity sensor, used to acquire real-time temperature and humidity data of in-vehicle environment; rain sensor and external temperature sensor, used to acquire rainfall and ambient temperature data of external environment; proximity sensor array, embedded in the top sealing strip of window glass, used to acquire short-range detection data of whether there are obstacles in the window closing path and their proximity.

Images