Intelligent control method and system for motorcycle calipers

Abstract

The invention discloses an intelligent control method and system for motorcycle calipers. According to the method, multi-modal data such as vehicle speed, wheel speed, piston displacement, brake disc temperature, inclination angle and transverse acceleration are collected in real time through a multi-source sensor and are preprocessed; calculating a braking demand and an adhesion coefficient according to the data, and recognizing a straight line, a curve or an emergency braking scene in combination with the inclination angle and the transverse acceleration; inputting the data into a pre-trained brake control machine learning model to generate a target brake force change curve; controlling the driving motor and the calipers to brake based on the curve, and regularly optimizing the model according to brake feedback and historical data; predicting a braking efficiency attenuation trend by combining a part degradation model, and compensating the braking force to update the model; the motor current is adjusted in real time through closed-loop control so as to follow the target braking force; and continuously monitoring the temperature of the brake disc, and triggering thermal attenuation compensation when the temperature exceeds a threshold value. According to the invention, self-adaptive, accurate and reliable intelligent braking of the motorcycle is realized.

Description

Technical Field

[0001] This application relates to the field of vehicle braking control technology, and in particular to an intelligent control method and system for motorcycle calipers. Background Technology

[0002] The motorcycle braking system is a core component ensuring riding safety, and its performance directly affects the vehicle's handling and passenger safety. Traditional motorcycle braking, especially caliper braking control, mainly relies on the rider's direct operation and the response of mechanical or basic electronic control systems.

[0003] Chinese Patent, Publication No. CN117644772A, Publication Date: March 5, 2024, discloses a method for controlling the drive and braking system of an electric motorcycle, comprising: receiving battery pack data indicating the current state of a battery pack connected to an electric drive motor of the electric motorcycle and a target value of a motor torque that should be generated by the drive motor to brake the electric motorcycle; checking whether the motor torque can be generated according to the target value based on the battery pack data; if the motor torque cannot be generated according to the target value, then: using the target value, determining a compensation torque that should be generated by the braking system of the electric motorcycle to brake the electric motorcycle, and generating a control command for manipulating the braking system to generate the compensation torque.

[0004] The drawback of the above-mentioned technical solution is that, when regenerative braking is insufficient, it relies on the driver's active manipulation of the braking system (such as the brake lever or pedal) to compensate for the braking force ("the driver must additionally manipulate the braking system"). This manual intervention not only increases the complexity of operation but may also lead to risks due to reaction delays or operational errors, especially in emergency braking scenarios. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application aims to provide an intelligent control method and system for motorcycle calipers. Through intelligent multimodal data fusion, machine learning models, and adaptive compensation mechanisms, it effectively solves the problem of how to achieve adaptive and predictive control of the motorcycle braking system in complex riding scenarios, thereby improving braking accuracy, reliability, and personalized experience.

[0006] To achieve the above objectives, this application adopts the following technical solution: This application provides an intelligent control method for motorcycle calipers, the method comprising the following steps: S101, through multi-source sensors integrated into the motorcycle, collects raw multimodal data in real time, including vehicle speed, wheel speed, piston displacement of motorcycle calipers, brake disc temperature, motorcycle lean angle and lateral acceleration. It performs time alignment and filtering on the raw multimodal data to obtain preprocessed multimodal data. S102, calculate the current braking demand level and road adhesion coefficient based on the preprocessed multimodal data, and identify real-time braking scenarios by combining the time-aligned and filtered motorcycle tilt angle and lateral acceleration. Real-time braking scenarios include straight-line braking, cornering braking or emergency braking. S103, the preprocessed multimodal data is input into the pre-trained braking control machine learning model, and the braking control machine learning model generates the corresponding target braking force change curve based on the identified real-time braking scenario. S104, based on the target braking force change curve, controls the motorcycle drive motor and motorcycle caliper to brake, collects braking effect feedback data, and periodically optimizes and trains the braking control machine learning model according to historical driving habit data and collected braking effect feedback data, updates the parameters of the braking control machine learning model, and obtains the optimized braking control machine learning model. S105, based on the wear state of the brake disc and the hysteresis characteristics of the piston displacement, the attenuation trend of the caliper braking efficiency is predicted through a preset component degradation model, and the braking force control quantity is compensated based on the attenuation trend of the caliper braking efficiency, and the optimized braking control machine learning model is updated to obtain the final braking control machine learning model. S106, based on the final braking control machine learning model, obtains the updated target braking force change curve. Based on the updated target braking force change curve, the piston mechanism of the motorcycle caliper is driven by the motorcycle drive motor. At the same time, based on the feedback piston displacement, the current of the motorcycle drive motor is adjusted in real time using a closed-loop control algorithm to adapt to the target braking force. S107 continuously monitors the brake disc temperature during braking. If the brake disc temperature exceeds the preset brake disc temperature threshold, a heat fade compensation strategy is triggered to dynamically correct the output braking force.

[0007] As a preferred technical solution, in step S101, the raw multimodal data, including vehicle speed, wheel speed, piston displacement of motorcycle calipers, brake disc temperature, motorcycle lean angle, and lateral acceleration, are collected in real time using multi-source sensors integrated into the motorcycle. The raw multimodal data is then time-aligned and filtered to obtain preprocessed multimodal data, including: Vehicle speed and wheel speed signals are acquired via CAN bus. Linear displacement sensors are used to measure the piston displacement of the motorcycle calipers, and non-contact infrared temperature sensors are used to monitor the brake disc temperature. A six-axis IMU integrated into the motorcycle body is used to collect the motorcycle's lean angle and lateral acceleration to obtain raw multimodal data. A data fusion buffer is set up in the central processing unit, and the raw multimodal data is time-aligned using hardware clock synchronization and interpolation algorithms. A low-pass filter is used to filter the wheel speed to suppress road noise, a Kalman filter is used to filter the motorcycle lean angle and lateral acceleration, and a moving average filter is used to filter the piston displacement and brake disc temperature. Finally, the preprocessed multimodal data is output.

[0008] As a preferred technical solution, in step S102, the current braking demand level and road adhesion coefficient are calculated based on the preprocessed multimodal data. Combined with the time-aligned and filtered motorcycle lean angle and lateral acceleration, real-time braking scenarios are identified. These real-time braking scenarios include straight-line braking, cornering braking, or emergency braking. Based on the preprocessed piston displacement and piston displacement change rate, the current braking demand level is calculated using a preset mapping function. The wheel slip ratio is calculated using the preprocessed wheel speed and vehicle speed, and the current road surface adhesion coefficient is iteratively calculated based on the vehicle dynamics model using the ratio of the wheel-end braking force to the tire vertical load derived in real time. A priority state machine is defined, with emergency braking as the highest priority, cornering braking as the second highest priority, and straight-line braking as the lowest priority. When the braking demand level exceeds the preset braking demand level, regardless of the motorcycle lean angle and lateral acceleration values, it is immediately determined to be emergency braking. If it is not emergency braking, then when the absolute value of the motorcycle lean angle continuously exceeds the preset motorcycle lean angle threshold and the lateral acceleration exceeds the preset lateral acceleration threshold, it is determined to be cornering braking, and the stability of the curve is further subdivided based on the current road surface adhesion coefficient. If it is not emergency braking, then when the motorcycle lean angle continuously falls below the preset motorcycle lean angle threshold and the lateral acceleration continuously falls below the preset lateral acceleration threshold, it is determined to be straight-line braking.

[0009] As a preferred technical solution, in step S103, the step of inputting the preprocessed multimodal data into a pre-trained braking control machine learning model, wherein the braking control machine learning model generates a corresponding target braking force change curve based on the identified real-time braking scenario, includes: The pre-trained braking control machine learning model is optimized through supervised learning on a training set consisting of historical and simulation data. The loss function used in the braking control machine learning model simultaneously penalizes factors such as excessive braking distance, excessive tire slippage, and vehicle attitude instability. The multimodal data preprocessed in step S101 is integrated with the real-time braking scenario, current braking demand level, and current road adhesion coefficient output in step S102 into a high-dimensional feature vector, which is then input into the pre-trained braking control machine learning model. The braking control machine learning model learns the complex nonlinear mapping between the high-dimensional feature vector and the optimal braking force response through a deep network structure, and generates the target braking force change curve based on the multi-objective optimization knowledge internalized in offline training.

[0010] As a preferred technical solution, in step S104, the braking is performed by controlling the motorcycle drive motor and motorcycle calipers based on the target braking force change curve, collecting braking effect feedback data, and periodically optimizing and training the braking control machine learning model based on historical driving habit data and the collected braking effect feedback data, updating the parameters of the braking control machine learning model, and obtaining the optimized braking control machine learning model, including: The target braking force change curve generated by S103 is analyzed by the PID controller into a braking control command for the real-time current of the motorcycle drive motor. The motorcycle drive motor drives the piston of the motorcycle caliper to generate the target braking force. Braking effect feedback data is collected simultaneously, including actual piston displacement, drive motor current, braking pressure, wheel speed, vehicle deceleration, and vehicle attitude data. The braking control command and real-time braking scenario are combined to form a timestamped historical driving habit data. The newly collected braking effect feedback data and historical driving habit data are periodically merged in the cloud or backend server to form an expanded dataset. The braking control machine learning model is incrementally trained or fine-tuned based on the expanded dataset. The parameters of the braking control machine learning model are updated through the backpropagation algorithm to obtain the optimized braking control machine learning model.

[0011] As a preferred technical solution, in step S105, the method of predicting the attenuation trend of caliper braking efficiency based on the wear state of the brake disc and the hysteresis characteristics of the piston displacement using a preset component degradation model, and compensating for the braking force control amount based on the attenuation trend of caliper braking efficiency to update the optimized braking control machine learning model, thereby obtaining the final braking control machine learning model, includes: By long-term monitoring of the increasing trend of the motorcycle caliper piston displacement reference value under the same braking control command, the wear state of the brake disc is estimated; by analyzing the morphological changes of the piston displacement and drive motor current closed-loop curve in the braking cycle, the hysteresis characteristics of the piston displacement are calculated; the wear state and hysteresis characteristic parameters are input into a preset component degradation model, which calculates and predicts the attenuation trend of the motorcycle caliper braking efficiency; the braking efficiency coefficient is calculated in real time based on the attenuation trend of braking efficiency, and the feedforward compensation braking force control quantity for braking control command is calculated based on the braking efficiency coefficient to offset the performance degradation; the braking efficiency coefficient is used as a new state feature or a dataset simulating different degrees of degradation is used to retrain or fine-tune the optimized braking control machine learning model obtained in step S104, thereby obtaining the final braking control machine learning model that can adapt to component aging.

[0012] As a preferred technical solution, in step S106, the updated target braking force change curve is obtained based on the final braking control machine learning model. Based on the updated target braking force change curve, the piston mechanism of the motorcycle caliper is driven by the motorcycle drive motor. Simultaneously, based on the feedback piston displacement, a closed-loop control algorithm is used to adjust the motorcycle drive motor current in real time to adapt to the target braking force, including: Based on the final braking control machine learning model obtained in step S105, an updated target braking force change curve is generated, and the updated target braking force change curve is converted into a target piston displacement curve through a preset calibration mapping relationship; the piston of the motorcycle caliper is driven by the motorcycle drive motor, and the actual piston displacement is collected in real time as a feedback signal; a closed-loop control algorithm is used to continuously compare the target piston displacement with the actual piston displacement, and the drive current of the motorcycle drive motor is calculated and adjusted in real time according to the displacement error; the motor output torque is controlled by the power driver so that the piston displacement accurately tracks the target piston displacement curve.

[0013] As a preferred technical solution, in step S107, the brake disc temperature is continuously monitored during braking. If the brake disc temperature exceeds a preset brake disc temperature threshold, a heat fade compensation strategy is triggered, dynamically correcting the output braking force, including: During braking, a non-contact infrared temperature sensor continuously monitors the brake disc temperature and compares it in real time with a preset brake disc temperature threshold. When the monitored brake disc temperature exceeds the preset threshold, a brake fade compensation strategy is triggered. This strategy dynamically calculates a brake fade compensation coefficient less than 1 based on real-time temperature and temperature rise rate. The braking control command from the PID controller is divided by the brake fade compensation coefficient to obtain a dynamically increased corrected braking force command. This corrected braking force command is sent to the actuator closed-loop control system, which ultimately adjusts the motorcycle drive motor current to output the compensated braking force. The brake fade compensation strategy remains in effect until the brake disc temperature falls below the preset threshold, at which point it automatically exits the strategy.

[0014] This application provides an intelligent control system for motorcycle calipers, the system comprising: The multi-source data acquisition and preprocessing module is used to acquire raw multimodal data such as vehicle speed, wheel speed, piston displacement, brake disc temperature, motorcycle lean angle and lateral acceleration in real time through multi-source sensors integrated into the motorcycle, and to perform time alignment and filtering on the raw multimodal data to obtain preprocessed multimodal data. The braking scene recognition module is used to calculate the current braking demand level and road adhesion coefficient based on the preprocessed multimodal data, and identify real-time braking scenes by combining the motorcycle tilt angle and lateral acceleration. The real-time braking scenes include straight-line braking, cornering braking or emergency braking. The intelligent braking control module includes a pre-trained braking control machine learning model, which receives the pre-processed multimodal data and generates a corresponding target braking force change curve based on the real-time braking scenario. The module is also used to periodically optimize and train the braking control machine learning model based on braking effect feedback data and historical driving habit data, and to correct the model based on the caliper braking efficiency decay trend predicted by the component degradation model, so as to obtain the final braking control machine learning model and output the updated target braking force change curve. The braking force execution and closed-loop adjustment module is used to drive the piston mechanism of the motorcycle caliper through the motorcycle drive motor to brake based on the updated target braking force change curve, and to adjust the current of the motorcycle drive motor in real time based on the feedback piston displacement using a closed-loop control algorithm to adapt to the target braking force. The heat fade compensation module is used to continuously monitor the brake disc temperature during braking, and when the brake disc temperature exceeds a preset threshold, it triggers a heat fade compensation strategy to dynamically correct the output braking force.

[0015] Compared with the prior art, the beneficial effects of this application are as follows: In steps S101-S102, data is collected through multi-source sensors (such as vehicle speed, wheel speed, piston displacement, disc temperature, and tilt angle), and time alignment and filtering are performed to identify real-time braking scenarios (such as straight-line, curve, or emergency braking). This allows the system to dynamically adapt to different riding conditions (such as changes in curve tilt angle), avoiding improper braking force distribution caused by the single scenario in traditional braking systems, thus improving braking safety and comfort. This application achieves more comprehensive braking decisions by fusing environmental and vehicle state data.

[0016] In steps S103-S104, preprocessed data is input into a pre-trained machine learning model to generate a target braking force curve, and the model parameters are periodically optimized based on historical driving habits and feedback data. The model can learn driver habits (such as braking force preferences), providing a personalized braking experience, and adapting to long-term changes through continuous optimization. This application achieves self-learning and evolution of the system, improving its intelligence level.

[0017] In steps S105-S107, the caliper braking efficiency degradation trend is predicted using a component degradation model, and a thermal fade compensation strategy is triggered based on disc temperature monitoring to dynamically adjust the braking force. This solves the performance degradation problem (such as wear and overheating) of the braking system during long-term use, maintains braking performance through predictive compensation, and extends component life. This application significantly improves the durability and safety of the system.

[0018] In step S106, the piston mechanism is controlled by a drive motor, and the current is adjusted in real time using a closed-loop algorithm based on piston displacement feedback to ensure that the braking force conforms to the target curve. Closed-loop control reduces the impact of external disturbances (such as road surface changes) and improves the real-time performance and accuracy of braking response. This application enhances control robustness through a feedback mechanism. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the steps of an intelligent control method for a motorcycle caliper according to this application. Figure 2 This is a schematic diagram of a motorcycle caliper. Figure 3 An exploded view of a motorcycle caliper; Figure 4 This is a cross-sectional structural diagram of a motorcycle caliper. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in specific embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0021] like Figure 1 As shown, this application provides an intelligent control method for motorcycle calipers.

[0022] The intelligent control method includes the following steps: S101, through multi-source sensors integrated into the motorcycle, collects raw multimodal data in real time, including vehicle speed, wheel speed, piston displacement of motorcycle calipers, brake disc temperature, motorcycle lean angle and lateral acceleration. The raw multimodal data is time-aligned and filtered to obtain preprocessed multimodal data.

[0023] S102 calculates the current braking demand level and road adhesion coefficient based on the preprocessed multimodal data, and identifies real-time braking scenarios by combining time-aligned and filtered motorcycle tilt angle and lateral acceleration. Real-time braking scenarios include straight-line braking, cornering braking, or emergency braking.

[0024] S103, the preprocessed multimodal data is input into the pre-trained braking control machine learning model, and the braking control machine learning model generates the corresponding target braking force change curve based on the identified real-time braking scenario.

[0025] S104 controls the motorcycle drive motor and motorcycle caliper to brake based on the target braking force change curve, collects braking effect feedback data, and periodically optimizes and trains the braking control machine learning model based on historical driving habit data and the collected braking effect feedback data, updates the parameters of the braking control machine learning model, and obtains the optimized braking control machine learning model.

[0026] S105, based on the wear state of the brake disc and the hysteresis characteristics of the piston displacement, predicts the decay trend of caliper braking efficiency through a preset component degradation model, and compensates for the braking force control quantity based on the decay trend of caliper braking efficiency, updates and optimizes the braking control machine learning model, and obtains the final braking control machine learning model.

[0027] S106, based on the final braking control machine learning model, obtains the updated target braking force change curve. Based on the updated target braking force change curve, the piston mechanism of the motorcycle caliper is driven by the motorcycle drive motor. At the same time, based on the feedback piston displacement, the current of the motorcycle drive motor is adjusted in real time using a closed-loop control algorithm to adapt to the target braking force.

[0028] S107 continuously monitors the brake disc temperature during braking. If the brake disc temperature exceeds the preset brake disc temperature threshold, a heat fade compensation strategy is triggered to dynamically correct the output braking force.

[0029] like Figure 2-4As shown in this application, the motorcycle caliper adopts the following structure: it includes a caliper body 1, with a horizontally penetrating piston cap mounting hole on one side; a caliper piston cap 2, a sealing ring 3, and a decorative piece 4 are arranged sequentially from the inside to the outside of the piston cap mounting hole; the caliper piston cap 2 is threadedly connected to the piston cap mounting hole, and the sealing ring 3 is pressed between the piston cap 2 and the inner wall of the piston cap mounting hole; the decorative piece 4 is pressed between the end of the caliper piston cap 2 and a protruding structure inside the piston cap mounting hole. Placing the easily damaged sealing structure inside the caliper fundamentally avoids loosening and damage caused by external accidental contact and impact, while greatly improving the overall appearance and aesthetics of the caliper.

[0030] Furthermore, such as Figure 4 As shown, the inner wall of the piston cap mounting hole includes an insertion section 101 and a connecting section 102 with decreasing inner diameters; a tapered bore section 103 is provided between the connecting section 102 and the insertion section 101; the caliper piston cap 2 is correspondingly provided with a limiting section 201 that is clearance-fitted with the insertion section 101 and an externally threaded section 202 that is threadedly connected to the connecting section 102; an annular groove 203 for accommodating the sealing ring 3 is provided on the externally threaded section 202, and the annular groove 203 corresponds to the position of the tapered bore section 103. In order to achieve a precision structure for reliable sealing, the tapered bore section 103 and the annular groove 203 cooperate to achieve uniform and precise compression of the sealing ring 3, forming a highly reliable seal superior to planar contact. The clearance fit between the limiting section 201 and the insertion section 101 ensures accurate alignment of the piston cap and avoids uneven wear.

[0031] like Figure 2 , 3 As shown in Figure 4, a through hole 105 extending from top to bottom is provided on the caliper body 1. The through hole 105 is used to accommodate the brake disc and provides space for the installation tool to extend into the polygonal countersunk hole 206 at the end of the limiting section 201. The dual function of the through hole 105—accommodating the disc and providing operating space—solves the problem of the inability to install internal parts, ensuring the practicality and feasibility of the structure.

[0032] like Figure 3 , 4 As shown, the bottom of the annular groove 203 is rounded. The rounded bottom design allows for better contact with the sealing ring 3, providing uniform support, preventing damage to the sealing ring 3 due to stress concentration during the tightening process, extending the life of the sealing ring, and further improving sealing reliability.

[0033] like Figure 4As shown, the caliper piston cap 2 has a clamping section 204 at its end; the end face of the clamping section 204 is a centrally protruding arc-shaped end face; the protruding structure is an annular protrusion 104 located at the outer port of the piston cap mounting hole; the decorative piece 4 is clamped between the arc-shaped end face and the annular protrusion 104. The arc-shaped end face can convert the axial force of the threaded connection into concentrated pressure on the elastic decorative piece 4, causing it to deform and generate a continuous reverse preload force, effectively resisting thread loosening caused by vibration, making the overall structure more stable.

[0034] like Figure 2 , 3 As shown in Figure 4, the decorative piece 4 has a honeycomb structure. While ensuring mechanical strength, the honeycomb structure significantly increases the heat dissipation area, effectively promoting the dissipation of braking heat, reducing heat accumulation inside the structure, thereby protecting the seals and hydraulic oil, and improving the reliability of the caliper under high-temperature conditions.

[0035] First, insert the sealing ring 3 into the round bottom annular groove 203 of the caliper piston cover 2; then insert the caliper piston cover 2 into the insertion section 101 of the mounting hole from inside the caliper; subsequently, the installation tool extends through the through hole 105 above the caliper body 1 and engages with the polygonal countersunk hole 206 on the limiting section 201; rotate the tool to make the external thread section 202 of the piston cover 2 mesh with the internal thread section of the mounting hole.

[0036] As the screws tighten, the sealing ring 3 in the annular groove 203 gradually enters the conical hole section 103 and is uniformly compressed to form a reliable seal. At the same time, the arc-shaped end face of the caliper piston cover 2 begins to press the honeycomb decorative piece 4, causing it to undergo elastic deformation, until the decorative piece 4 is firmly pressed between the arc-shaped end face and the annular protrusion 104 at the mounting hole port, forming an effective anti-loosening pre-tightening force.

[0037] The decorative piece 4 is preferably made of stainless steel and has a thickness of 0.5 mm; the cone angle of the tapered hole section 103 is preferably 30°-45°.

[0038] Furthermore, in step S101, using multi-source sensors integrated into the motorcycle, raw multimodal data including vehicle speed, wheel speed, piston displacement of the motorcycle caliper, brake disc temperature, motorcycle lean angle, and lateral acceleration are collected in real time. The raw multimodal data is then time-aligned and filtered to obtain preprocessed multimodal data, including: Vehicle speed and wheel speed signals are acquired via CAN bus. Linear displacement sensors (such as linear variable differential transformers (LVDTs) are used to measure the displacement of the motorcycle caliper pistons in real time, directly mounted on the caliper pistons. Non-contact infrared temperature sensors are used to monitor the temperature of the brake discs (non-contact infrared temperature sensors are oriented to the surface of the brake discs and capture temperature changes with millisecond-level response speed). The motorcycle's lean angle and lateral acceleration are collected by a six-axis IMU integrated into the motorcycle body to obtain raw multimodal data.

[0039] A data fusion buffer is set up in the central processing unit (such as ECU) to temporarily store multi-source data and solve the problem of data loss caused by differences in sensor sampling frequencies.

[0040] The original multimodal data is time-aligned using a combination of hardware clock synchronization and interpolation algorithms. Specifically, a unified hardware clock signal is allocated to all sensors to ensure the consistency of data timestamp references. For asynchronously sampled data (such as high-frequency IMU data and low-frequency CAN bus data), linear interpolation or spline interpolation algorithms are used to resample the data to a unified time axis, eliminating timing deviations.

[0041] Wheel speed is filtered using a low-pass filter (such as a Butterworth filter) to suppress road noise (such as wheel speed vibration caused by bumps). Motorcycle lean angle and lateral acceleration are filtered using a Kalman filter, fusing gyroscope and accelerometer data to suppress dynamic noise and improve attitude estimation accuracy. Piston displacement and brake disc temperature are filtered using a moving average to eliminate instantaneous fluctuations and improve data stability. The final output is preprocessed multimodal data.

[0042] Through the above technical solution, step S101 has the following beneficial effects: 1. Improve data reliability and system decision-making accuracy: Time alignment and filtering eliminate timing conflicts and noise interference from multi-source sensors, making the data more closely reflect the actual physical process. For example, wheel speed filtering can accurately identify the true slip ratio, avoiding false ABS triggering due to road bumps; after Kalman filtering, the tilt angle estimation error of IMU data is reduced to within ±0.5°, providing a key basis for identifying cornering braking scenarios. The preprocessed data is directly input into the machine learning model (step S103), and the high-quality data significantly improves the model's classification accuracy of braking scenarios and the accuracy of braking force curve generation.

[0043] 2. Enhance system real-time performance and robustness: Hardware clock synchronization and interpolation algorithms control data latency to the millisecond level, meeting the real-time response requirements of the braking system (e.g., emergency braking needs to be triggered within 100ms). Moving average filtering avoids control command oscillations caused by momentary sensor anomalies (e.g., occasional drift of the temperature sensor). Combined with closed-loop control (step S106), the high stability of the pre-processed data ensures smooth motor current regulation and reduces braking shock.

[0044] 3. Provide data support for predictive maintenance: Continuous monitoring data of piston displacement and disc temperature, after being filtered by moving average, can be used for long-term trend analysis (such as the component degradation model in step S105) to identify caliper wear or thermal fade risks in advance. The filtered temperature data is directly used for thermal fade compensation strategy (step S107) to avoid premature or delayed compensation due to noise, thereby extending the life of the brake disc.

[0045] Furthermore, in step S102, the current braking demand level and road adhesion coefficient are calculated based on the preprocessed multimodal data. Combined with the time-aligned and filtered motorcycle lean angle and lateral acceleration, real-time braking scenarios are identified. These real-time braking scenarios include straight-line braking, cornering braking, or emergency braking. Based on the preprocessed piston displacement (unit: mm) and piston displacement change rate (unit: mm / s), the current braking demand level (e.g., 0-10, where 0 represents no braking demand and 10 represents maximum braking demand) is calculated using a preset mapping function (such as a piecewise linear function or a neural network model). For example, when the piston displacement increases rapidly and the change rate exceeds a threshold, the mapping function outputs a high level (e.g., 8-10), indicating a strong braking intention.

[0046] Calculate wheel slip ratio using pre-processed wheel speed and vehicle speed: ,in For vehicle speed, This refers to wheel speed.

[0047] Based on the vehicle dynamics model, the wheel-end braking force is calculated in real time. (Estimated via brake pressure sensor or motor torque) and vertical load on tires The current road surface adhesion coefficient μ is iteratively calculated by comparing the ratio of IMU data to the motorcycle's mass estimate. Specifically, the adhesion coefficient is iteratively calculated by minimizing the error of the slip ratio-adhesion coefficient curve using gradient descent or a Kalman filter, as shown in the formula. It is then modified using tire models such as Burckhardt's. It iterates every 20ms to adapt to changes in road surface (such as from dry asphalt to wet and slippery surfaces).

[0048] Define a state machine with priorities, where emergency braking has the highest priority, cornering braking has the second highest priority, and straight-line braking has the lowest priority. The priority order is: emergency braking > cornering braking > straight-line braking.

[0049] When the braking demand level exceeds the preset braking demand level (e.g., level 8), regardless of the motorcycle's lean angle and lateral acceleration values, it will immediately be determined as emergency braking. This ensures safety as the priority and avoids response delays caused by scenario misjudgment.

[0050] If it is not an emergency braking, when the absolute value of the motorcycle lean angle continuously exceeds the preset motorcycle lean angle threshold (e.g., 15°) and the lateral acceleration exceeds the preset lateral acceleration threshold (e.g., 3 m / s²), it is determined to be cornering braking. At the same time, the stability of the corner is further subdivided (e.g., stable corner, critical slip corner) by combining the current road surface adhesion coefficient (e.g., marked as low adhesion corner when μ < 0.3).

[0051] If it is not an emergency braking, then if the motorcycle tilt angle is continuously lower than the preset motorcycle tilt angle threshold (e.g., 5°) and the lateral acceleration is continuously lower than the preset lateral acceleration threshold (e.g., 1 m / s²), it is determined to be straight-line braking.

[0052] Use a finite state machine (FSM) to manage scene transitions and ensure conflict-free transitions (e.g., if an emergency condition is triggered during cornering braking, it immediately escalates to emergency braking). Through the above technical solution, step S103 has the following beneficial effects: 1. Improve the accuracy and real-time performance of braking scene recognition: By collaboratively determining braking demand levels and dynamic parameters (such as tilt angle and acceleration), scene recognition accuracy is improved to over 95% (compared to 70-80% for single-parameter judgment). For example, the emergency braking priority mechanism avoids overlooking sudden risks due to excessive tilt angles in curves. The state machine runs every 10ms, and combined with the iteratively calculated road adhesion coefficient, scene switching can be completed within 100ms to meet emergency braking requirements (e.g., the time difference from detecting danger to triggering braking is less than 150ms). Accurate scene recognition provides reliable input to the machine learning model in step S103, ensuring that the generated target braking force curve (such as the progressive curve for cornering braking) highly matches the scene.

[0053] 2. Enhance the system's adaptability to complex environments: Iterative calculation of the road adhesion coefficient can capture road surface changes in real time (such as from dry to wet surfaces), and prevent wheel lock-up by correcting the braking force distribution. For example, when μ decreases from 0.8 to 0.2, the system automatically reduces the braking force growth rate, improving safety on wet and slippery roads. Duration judgments in the state machine (such as the 500ms duration condition for curve determination) effectively filter instantaneous noise (such as tilting vibrations caused by road bumps), reducing false triggering. Combined with the closed-loop control in step S106, adaptive scene recognition makes the motor current adjustment more in line with actual needs (such as balancing the braking force of the front and rear wheels during cornering braking), improving vehicle stability.

[0054] 3. Provide a decision-making basis for personalized and predictive control: The mapping function for braking demand levels can be optimized based on historical data (e.g., a higher threshold corresponds to an aggressive driving style), providing input for the personalized model training in step S104. The priority triggering mechanism for emergency braking, combined with the road surface adhesion coefficient, can identify high-risk scenarios in advance (e.g., low-adhesion curves), providing data support for component degradation prediction in step S105 (e.g., acceleration caliper wear warning during frequent emergency braking). Scenario-specific data (e.g., curve stability) is used for dynamic adjustment of the thermal fade compensation strategy in step S107, extending component lifespan.

[0055] Furthermore, in step S103, the preprocessed multimodal data is input into the pre-trained braking control machine learning model. The braking control machine learning model generates the corresponding target braking force change curve based on the identified real-time braking scenario, including: The pre-trained braking control machine learning model is optimized by supervising learning on a training set consisting of historical and simulation data. The loss function used in the braking control machine learning model simultaneously penalizes factors such as excessive braking distance, excessive tire slippage, and vehicle attitude instability.

[0056] Specifically, a training set is constructed using historical real-world driving data and simulation platform data, covering typical scenarios such as straight-line braking, cornering braking, and emergency braking, with a data volume exceeding 100,000 sets. Simulation data is generated using dynamics software such as ADAMS / Car, covering different road surface adhesion coefficients (μ=0.1-0.9), tilt angles (0°-40°), and braking intensity conditions.

[0057] Using multimodal data (vehicle speed, wheel speed, tilt angle, etc.) as input and expert-calibrated optimal braking force curves as labels, the model is trained to learn their mapping relationship. Five-fold cross-validation is used to prevent overfitting.

[0058] Define the composite loss function: ; in, Penalizes excessive braking distance (mean square error between the actual braking distance and the actual braking distance). Penalty for tire slippage exceeding a safety threshold (e.g., an exponentially increasing penalty when slippage exceeds 20%). Penalize vehicle attitude instability (based on quadratic terms of lateral acceleration and tilt rate of change). Hyperparameters α, β, and γ are optimized through grid search to balance the weights of each objective.

[0059] The multimodal data preprocessed in step S101 is integrated with the real-time braking scenario, current braking demand level, and current road surface adhesion coefficient output in step S102 into a high-dimensional feature vector, which is then input into the pre-trained braking control machine learning model.

[0060] Specifically, the preprocessed multimodal data (vehicle speed, wheel speed, piston displacement, disc temperature, tilt angle, and lateral acceleration) from step S101 are integrated with the output of step S102 (real-time braking scenario, braking demand level, and road adhesion coefficient) into a 15-dimensional feature vector. Continuous features are Z-score normalized, and categorical features (such as braking scenarios) are encoded using one-hot encoding to ensure model convergence stability.

[0061] The braking control machine learning model learns the complex nonlinear mapping between high-dimensional feature vectors and optimal braking force response through a deep network structure, and generates the target braking force change curve based on the multi-objective optimization knowledge internalized by offline training.

[0062] Specifically, the braking control machine learning model employs a deep fully connected neural network (FCN), comprising an input layer (15-dimensional), three hidden layers (128, 64, and 32 neurons respectively), and an output layer (20-dimensional braking force curve points). The hidden layers use the ReLU activation function, and the output layer uses the Sigmoid function to constrain the output range (0-1, corresponding to standardized braking force). The model learns the complex relationship between features and braking force through the hidden layer neurons. For example, in high-adhesion cornering scenarios, the model outputs a progressive braking force curve to avoid sideslip; in low-adhesion emergency braking, the model outputs a stepped curve to prioritize vehicle stability. During online deployment, the model inference frequency is 100Hz to ensure real-time performance. Optimized using frameworks such as TensorRT, the single inference latency is <1ms.

[0063] Through the above technical solution, step S103 has the following beneficial effects: 1. Multi-objective optimization to improve overall braking performance: Braking distance is directly optimized via a loss function. In emergency braking scenarios on dry surfaces (μ=0.8), the braking distance is reduced by an average of 12% compared to the regular basis control method. A slip ratio penalty term automatically reduces the braking force growth rate on low-adhesion surfaces (μ=0.2), keeping the slip ratio within a safe range of 15% to avoid frequent ABS intervention. A posture instability penalty term ensures balanced braking force distribution during cornering braking, reducing lateral acceleration fluctuations by 30% and improving riding comfort. The optimized braking force curve is directly used in the closed-loop control of step S106, providing an ideal reference trajectory for motor current adjustment.

[0064] 2. Adaptability to complex scenes: The model internalizes multi-parameter coupling effects through deep networks. For example, in high tilt angle + low adhesion scenarios, it automatically generates a braking force curve that is initially gentle and then accelerates to avoid the risk of sideslip; in high braking demand + high-temperature disc brakes, it dynamically adjusts the curve peak value to balance braking performance and thermal management. Based on large-scale simulations and training with real data, the model can adapt to unseen scenarios (such as continuous cornering braking when tilt angles change abruptly), reducing the need for manual parameter tuning. The adaptive curve generation and the scenario recognition in step S102 form a closed loop, enhancing the system's robustness in dynamic environments.

[0065] 3. Provides core drivers for system evolution: The pre-trained model supports online incremental learning (see step S104), continuously optimizing weights through real-time feedback data to adapt to changes in driver habits. The model implicitly learns expert experience (such as force distribution strategies for cornering braking), reducing reliance on precise mathematical models. The model output serves as input for component degradation prediction in step S105, and long-term data can be used to identify anomalies in the braking force curve (such as curve distortion caused by decreased caliper efficiency).

[0066] Furthermore, in step S104, the motorcycle drive motor and motorcycle calipers are controlled to brake based on the target braking force change curve. Braking effect feedback data is collected. Based on historical driving habit data and the collected braking effect feedback data, the braking control machine learning model is periodically optimized and trained, and the parameters of the braking control machine learning model are updated to obtain the optimized braking control machine learning model, which includes: The target braking force change curve generated by S103 is analyzed by the PID controller into a braking control command for the real-time current of the motorcycle drive motor. The motorcycle drive motor drives the piston of the motorcycle caliper to generate the target braking force.

[0067] Specifically, a discrete PID controller is used to parse the target braking force change curve (normalized to the 0-1 range) generated by S103 into real-time current control commands for the motorcycle drive motor. The control algorithm is as follows: ; in The proportionality coefficient is the ratio of the target braking force to the actual braking force (calculated by piston displacement). Integral coefficient Differential coefficients Tuning is performed using the Ziegler-Nichols method. The motorcycle drive motor (typically a brushless DC motor) receives current commands and drives the caliper piston via a ball screw mechanism, generating precise mechanical clamping force. The current control frequency is 1kHz to ensure real-time response. Dual hardware and software current limits are set (e.g., 0-5A range), and the system automatically switches to a safe mode when an anomaly is detected (e.g., a current surge exceeding 20%).

[0068] Simultaneously collect braking effect feedback data, including actual piston displacement, drive motor current, braking pressure, wheel speed, vehicle deceleration, and vehicle attitude data. Combine braking control commands and real-time braking scenarios to form time-stamped historical driving habit data.

[0069] Specifically, the following feedback data is collected synchronously at a frequency of 100Hz: Piston displacement: Measured by an LVDT sensor with an accuracy of ±0.1mm; Drive motor current: acquired via a Hall current sensor; Braking pressure: The caliper hydraulic pressure is measured by a piezoelectric pressure sensor, ranging from 0-20 MPa; Wheel speed: acquired via a magnetoelectric wheel speed sensor; Vehicle deceleration: Measured via the IMU's triaxial accelerometer; Vehicle attitude data: including tilt angle (accuracy ±0.5°) and lateral acceleration.

[0070] The feedback data is bound to the corresponding braking control commands and real-time braking scenarios (from S102), and millisecond-level timestamps are added to form a structured historical driving habit database.

[0071] The newly collected braking effect feedback data is periodically merged with historical driving habit data in the cloud or backend server to form an expanded dataset. The braking control machine learning model is then incrementally trained or fine-tuned based on the expanded dataset. The parameters of the braking control machine learning model are updated through the backpropagation algorithm to obtain an optimized braking control machine learning model.

[0072] Specifically, the following processes are automatically executed monthly on the cloud server: outlier data (such as periods of sensor failure); time warping algorithm is used to augment the sample for critical scenarios (such as low-adhesion emergency braking); newly acquired data is merged with historical data, reaching a total sample size of millions. Elastic Weight Consolidation (EWC) is used for incremental training to avoid catastrophic forgetting. The loss function continues the multi-objective design of S103, but adds an adaptive penalty term for new data distributions.

[0073] Parameter fine-tuning process: The initial learning rate was set to 0.001, and the Adam optimizer was used. Freeze the model's bottom-level parameters (feature extraction layer), and mainly fine-tune the top fully connected layer; The training cycle should not exceed 50 rounds, and early stopping (Patience=10) should be used to prevent overfitting. Model validation was conducted through A / B testing to ensure that the optimized model was not inferior to the original model in terms of braking distance, stability, and other indicators.

[0074] Through the above technical solution, step S104 has the following beneficial effects: 1. Improve braking control precision and stability: PID control keeps the braking force tracking error within ±5% (compared to ±15% for open-loop control), ensuring the actual braking force closely matches the target curve. For example, during cornering braking, vehicle body posture fluctuation is reduced by 25%. Execution deviations are corrected in real-time using feedback data, adapting to disturbances such as component wear and temperature changes. When piston response delay is detected, the PID derivative term compensates in advance, reducing overshoot. This high-precision execution lays the foundation for the S106's closed-loop regulation, ensuring synchronization between motor current regulation and the target curve.

[0075] 2. Achieve a personalized driving experience: Historical data continuously records driver preferences (e.g., aggressive drivers often trigger high braking demands), and the model automatically adjusts the braking force curve shape after optimization. For example, for mild-mannered drivers, a smoother braking force growth curve is generated to improve comfort. By analyzing feedback data from specific scenarios (e.g., curves frequently encountered by a driver), the model's braking strategy accuracy in that scenario is improved by 15%. Personalized model outputs are used for component degradation prediction of the S105, more accurately estimating caliper lifespan under different driving styles.

[0076] 3. Continuous system evolution and improved reliability: Compared to traditional batch training, incremental training shortens the iteration cycle for adapting the model to new environments (such as low-adhesion roads in winter) from 3 months to 2 weeks, while reducing resource consumption by 60%. Abnormal patterns in the feedback data (such as persistently high motor current) can identify potential faults (such as caliper jamming) in advance, triggering warnings through the degradation model in S105. The optimized model more accurately predicts disc temperature trends in thermal attenuation compensation in S107, avoiding overcompensation.

[0077] Furthermore, in step S105, based on the wear state of the brake disc and the hysteresis characteristics of the piston displacement, the attenuation trend of the caliper braking efficiency is predicted through a preset component degradation model. The braking force control quantity is then compensated based on this attenuation trend, and the optimized braking control machine learning model is updated. The final braking control machine learning model includes: The wear condition of the brake disc is estimated by long-term monitoring of the increasing trend of the reference value of the motorcycle caliper piston displacement under the same braking control command.

[0078] Specifically, every 1000 braking cycles, the baseline value of the motorcycle caliper piston displacement (i.e., the average value from the initial piston position to the maximum displacement) under the same braking control command (e.g., a target braking force of 500N) is statistically analyzed. Displacement data is collected using a high-precision linear displacement sensor (LVDT) at a sampling frequency of 1kHz. A linear regression model is used to fit the trend of the piston displacement baseline value over time. When the baseline value continues to increase (e.g., an increase of more than 0.1mm per month), it is determined that the brake disc wear has led to an increase in the clamping clearance. The formula for calculating the wear amount ΔW is: ; in This is the current displacement reference value. The initial calibration value is given, and k is the wear coefficient (calibrated via bench testing). The displacement sensor is calibrated monthly using standard test blocks to eliminate sensor drift errors.

[0079] By analyzing the morphological changes of the piston displacement and drive motor current closed-loop curves during the braking cycle, the hysteresis characteristics of the piston displacement are calculated.

[0080] Specifically, in a single braking cycle, the drive motor current (control input) and piston displacement (output) are recorded synchronously to form a current-displacement closed-loop curve. Data acquisition points include the braking application, holding, and release phases. The hysteresis loop area of ​​the current-displacement curve is calculated using numerical integration to reflect energy loss. The formula is: Where I is the motor current and D is the piston displacement.

[0081] Extract the width (maximum displacement difference) and tilt of the hysteresis loop as hysteresis characteristic indicators. An increase in loop width or a change in tilt indicates increased internal friction in the caliper or aging of the seals. Update the hysteresis characteristic parameters every 100 braking cycles to ensure dynamic tracking.

[0082] The wear state ΔW and hysteresis characteristic parameters (hysteresis area A_hysteresis) are input into a preset component degradation model. The component degradation model calculates and predicts the degradation trend of the motorcycle caliper's braking efficiency. Based on the degradation trend of braking efficiency, the braking efficiency coefficient is calculated in real time. Based on the braking efficiency coefficient, the feedforward compensation braking force control quantity used for braking control commands is calculated to offset the performance degradation.

[0083] Specifically, the component degradation model adopts a physics-based exponential decay model to predict the decay trend of caliper braking efficiency η: ; in The initial efficiency is given by λ, the natural decay coefficient is given by β and γ, and the weight parameters are given by training with historical fault data.

[0084] The real-time braking efficiency coefficient K_eff is defined as the ratio of the current efficiency to the initial efficiency: The compensation mechanism is triggered when K_eff < 0.9.

[0085] By using the braking efficiency coefficient as a new state feature or by using a dataset that simulates different degrees of degradation, the optimized braking control machine learning model obtained in step S104 is retrained or fine-tuned to obtain the final braking control machine learning model that can adapt to component aging.

[0086] Specifically, based on the braking efficiency coefficient K_eff, the feedforward braking force compensation ΔF_comp is calculated: ; Where F_target is the target braking force. The compensation amount is directly added to the PID controller output in step S104 to ensure that the actual braking force does not decrease due to efficiency degradation. The compensation amount is updated every 10ms and works in conjunction with the closed-loop control to avoid overcompensation.

[0087] The braking efficiency coefficient K_eff is added as a new state feature, expanding the high-dimensional feature vector in step S103 (from 15 to 16 dimensions), and inputting it into the optimized braking control machine learning model. A component degradation model is used to simulate braking scenarios with different degrees of degradation (e.g., K_eff decreasing from 1.0 to 0.7), generating an augmented dataset containing degradation features. On a cloud server, the top-level parameters of the model are fine-tuned based on simulated data and real feedback data. An efficiency decay penalty term is added to the loss function. ; Where δ is the decay weight, which is optimized through cross-validation. A / B testing is used to verify the performance of the updated model in the degradation scenario, ensuring that the braking distance deviation does not exceed 5%.

[0088] Through the above technical solution, step S105 has the following beneficial effects: 1. Maintaining stable braking performance: The feedforward compensation mechanism directly offsets the impact of caliper efficiency degradation, keeping braking distance fluctuations within 3% throughout the entire lifespan. For example, when disc wear causes a 20% efficiency decrease, the compensation amount automatically adjusts the braking force to prevent extended braking distances. Real-time monitoring of hysteresis characteristics can detect sudden aging (such as seal failure), with a compensation response time of less than 100ms, reducing the risk of sudden performance drops. The compensation control is seamlessly integrated with the closed-loop adjustment in step S106, ensuring that the motor current command always matches the actual demand.

[0089] 2. Component life prediction and maintenance optimization: The degradation model predicts caliper efficiency below a threshold (e.g., K_eff < 0.8) 1000 km in advance, triggering a maintenance warning and reducing the probability of unexpected failures. For example, when a hysteresis area is detected to increase by 15% per month, the system recommends replacing the seal. Through accurate wear estimation, over-maintenance is avoided (e.g., no need for regular disc replacement), reducing maintenance costs by 25%. Predictive data is used to optimize the thermal degradation compensation strategy in step S107, preventing high temperatures from accelerating component degradation.

[0090] 3. Intelligent upgrade of the model: The addition of an efficiency coefficient feature enables the model to distinguish between the braking strategies of new and aging calipers, improving the accuracy of braking force curve generation in degradation scenarios. Through training on simulated data, the model covers the full range of states from brand new to severely aged, adapting to different riding habits (such as frequent braking accelerating degradation). The updated model converges faster in the feedback optimization step S104, reducing retraining time.

[0091] Furthermore, in step S106, based on the final braking control machine learning model, an updated target braking force change curve is obtained. Based on the updated target braking force change curve, the piston mechanism of the motorcycle caliper is driven by the motorcycle drive motor. Simultaneously, based on the feedback piston displacement, a closed-loop control algorithm is used to adjust the motorcycle drive motor current in real time to adapt to the target braking force, including: Based on the final braking control machine learning model obtained in step S105, an updated target braking force variation curve (20-dimensional data points, time resolution 10ms) is generated. This updated curve is then converted into a target piston displacement curve using a pre-defined calibration mapping relationship (a three-dimensional mapping table of braking pressure, piston displacement, and clamping force). The mapping relationship considers the nonlinear characteristics of the caliper mechanism, including: lever ratio change compensation (speed ratio for different clamping positions); thermal expansion compensation (the effect of disc temperature on effective stroke); and stiffness correction (elastic deformation based on a material fatigue model).

[0092] The piston of the motorcycle caliper is driven by a motorcycle drive motor, and the actual piston displacement is collected in real time as a feedback signal. Specifically, a high-precision magnetic grating displacement sensor with a resolution of 0.01 mm and a sampling frequency of 2 kHz is used to monitor the actual piston displacement in real time. The sensor is directly mounted on the side of the piston rod, avoiding mechanical interference through non-contact measurement.

[0093] A closed-loop control algorithm is employed to continuously compare the target piston displacement with the actual piston displacement, and to calculate and adjust the drive current of the motorcycle drive motor in real time based on the displacement error. Specifically, a fuzzy adaptive PID control algorithm is used, with the core control logic as follows: displacement error e(t) = D_target(t) - D_actual(t); current adjustment ΔI(t) = Kp·e(t) + Ki·∫e(t)dt + Kd·de(t) / dt. The fuzzy logic dynamically adjusts the PID parameters (Kp, Ki, Kd) according to the error magnitude. For large errors, the proportional action is enhanced for rapid tracking; for small errors, the integral action is enhanced to eliminate steady-state error.

[0094] The motor output torque is controlled by a power driver, ensuring the piston displacement accurately tracks the target piston displacement curve. Specifically, a three-phase full-bridge PWM drive circuit is used with a switching frequency of 20kHz and a current loop control bandwidth of 1kHz. Space vector modulation (SVPWM) technology is employed to improve DC voltage utilization. A current loop is nested within the displacement loop, forming a dual closed-loop control: the inner loop (current loop) ensures the motor output torque accurately tracks the current command; the outer loop (displacement loop) ensures the piston position accurately tracks the target curve. Real-time compensation mechanisms include: back EMF compensation: feedforward compensation based on motor speed; friction compensation: nonlinear compensation based on the Stribeck friction model; and temperature drift compensation: correcting drive parameters based on temperature sensor readings.

[0095] Through the above technical solution, step S106 has the following beneficial effects: 1. Significantly improved precision in braking force control: Fuzzy adaptive PID control keeps the piston displacement tracking error within ±0.1mm (compared to ±0.5mm for traditional open-loop control), resulting in improved braking force control accuracy. The dual closed-loop control structure ensures a step response time of <50ms and a settling time of <100ms, enabling precise tracking of rapidly changing curves during emergency braking. A real-time compensation mechanism effectively suppresses external disturbances (such as road impacts and temperature changes), reducing braking force fluctuations under continuous braking conditions.

[0096] 2. System security and reliability are significantly enhanced: By monitoring the displacement-current characteristic curve, system anomalies can be identified in real time: protection strategies are triggered when the displacement change rate is abnormal; an alarm is triggered when the current-displacement relationship deviates from the normal range; and when the displacement sensor fails, the system automatically switches to the current-braking force estimation mode to maintain basic braking function.

[0097] 3. Seamless integration with front-end intelligent decision-making Multi-timescale coordination: Slow-changing stages (model updates): executed periodically (S104 / S105); Fast-changing stages (closed-loop tracking): executed in real time (10ms cycle); Two-way information flow: Forward: Intelligent model decision-making → precise execution; Feedback: Execution results → model optimization.

[0098] End-to-end optimization: Real-time performance data of the closed-loop system provides real labels for front-end model optimization, forming a complete closed loop of "decision-execution-optimization".

[0099] Furthermore, in step S107, the brake disc temperature is continuously monitored during braking. If the brake disc temperature exceeds a preset brake disc temperature threshold, a heat fade compensation strategy is triggered to dynamically correct the output braking force, including: During braking, a non-contact infrared temperature sensor continuously monitors the brake disc temperature (e.g., 300°C, calibrated based on material properties) and compares it in real time with a preset brake disc temperature threshold. When the monitored brake disc temperature exceeds the preset threshold, a heat fade compensation strategy is triggered. This strategy dynamically calculates a heat fade compensation coefficient less than 1 based on the real-time temperature and temperature rise rate.

[0100] Specifically, the compensation strategy calculates the thermal decay compensation coefficient (K_thermal, ranging from 0.1 to 0.9) based on the real-time temperature (T) and the temperature rise rate (dT / dt). The temperature rise rate is calculated using the finite difference method (e.g., the temperature change per 10 ms) to reflect the rate of heat accumulation.

[0101] Use empirical formulas or lookup tables to dynamically determine K_thermal: when the temperature rise rate is high (e.g., dT / dt > 10°C / s), K_thermal takes a smaller value (e.g., 0.5), indicating that stronger compensation is needed; when the temperature is high but the temperature rise rate is low, K_thermal takes a larger value (e.g., 0.8) to avoid overcompensation.

[0102] For example, the empirical formula is: ; where α, β, and γ are calibration parameters.

[0103] The thermal attenuation compensation coefficient is updated every 20ms to ensure real-time response.

[0104] Dividing the braking control command from the PID controller by the thermal fade compensation coefficient yields the dynamically increased corrected braking force command: Since K_thermal < 1, the modified instruction dynamically increases to offset the braking force loss caused by thermal decay.

[0105] This corrected braking force command is sent to the actuator closed-loop control system, which ultimately adjusts the motorcycle drive motor current to output the compensated braking force. The heat fade compensation strategy remains in effect until the brake disc temperature drops below a preset brake disc temperature threshold (e.g., below 290°C and maintained below this threshold for more than 500ms), at which point the heat fade compensation strategy automatically exits. The exit process employs a smooth transition algorithm (e.g., linear fade compensation over 100ms) to prevent sudden changes in braking force.

[0106] Through the above technical solution, step S107 has the following beneficial effects: 1. Braking performance stability maintenance: By dynamically increasing braking force commands, the effect of decreased brake pad friction coefficient at high temperatures is directly compensated, ensuring that the actual braking force remains above 90% of the target value even when the disc temperature reaches 500°C. For example, during continuous downhill braking, braking distance fluctuations are controlled within 5%. Coefficient calculations based on the rate of temperature rise can respond in advance to heat accumulation trends, compensating for response delays of <50ms and preventing sudden drops in braking performance. In conjunction with the S106's closed-loop control, it ensures that the compensated braking force accurately tracks the target curve, improving riding safety.

[0107] 2. Improved system security and lifespan: The compensation strategy prevents the disc temperature from rising indefinitely (e.g., by moderately increasing force output to reduce braking time), reducing the probability of disc thermal cracking and extending component life. Real-world testing shows that the average disc lifespan is increased by 15%. It avoids braking distance increases or failures due to thermal fade, especially in emergency braking scenarios, reducing the risk of accidents. Compensation data is fed back to the component degradation model in the S105, optimizing long-term prediction accuracy.

[0108] 3. Energy efficiency optimization and improved driving experience: Precise compensation prevents drivers from excessively operating the brake lever to counteract brake fade, reducing energy waste. A smooth compensation transition mechanism prevents abrupt changes in braking force, improving braking smoothness and riding comfort. Combined with the S104's personalized model, it adapts to different riding styles (such as frequent high-temperature compensation under aggressive riding conditions).

[0109] This application provides an intelligent control system for motorcycle calipers, the system comprising: The multi-source data acquisition and preprocessing module is used to acquire raw multimodal data such as vehicle speed, wheel speed, piston displacement, brake disc temperature, motorcycle lean angle and lateral acceleration in real time through multi-source sensors integrated into the motorcycle, and to perform time alignment and filtering on the raw multimodal data to obtain preprocessed multimodal data.

[0110] The braking scene recognition module is used to calculate the current braking demand level and road adhesion coefficient based on the preprocessed multimodal data, and to identify real-time braking scenes by combining the motorcycle tilt angle and lateral acceleration. Real-time braking scenes include straight-line braking, cornering braking or emergency braking.

[0111] The intelligent braking control module includes a pre-trained braking control machine learning model, which receives pre-processed multimodal data and generates a corresponding target braking force change curve based on real-time braking scenarios. This module also periodically optimizes and trains the braking control machine learning model based on braking effect feedback data and historical driving habit data. Furthermore, it corrects the model based on the caliper braking efficiency decay trend predicted by the component degradation model, resulting in the final braking control machine learning model, which then outputs the updated target braking force change curve.

[0112] The braking force execution and closed-loop adjustment module is used to brake the motorcycle caliper piston mechanism through the motorcycle drive motor based on the updated target braking force change curve, and to adjust the current of the motorcycle drive motor in real time based on the feedback piston displacement using a closed-loop control algorithm to adapt to the target braking force.

[0113] The thermal fade compensation module is used to continuously monitor the brake disc temperature during braking and trigger the thermal fade compensation strategy when the brake disc temperature exceeds a preset threshold, dynamically correcting the output braking force.

[0114] It should be noted that the terms "first," "second," and similar terms used in this application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, "a" or "one," and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. "A plurality" or "several" indicates at least two. Unless otherwise stated, terms such as "front," "back," "left," "right," "lower," and / or "upper" are for illustrative purposes only and are not limited to a location or spatial orientation. Terms such as "comprising" or "including" indicate that the elements or objects preceding "comprising" encompass the elements or objects listed following "comprising" or "including" and their equivalents, and do not exclude other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0115] The singular forms “a,” “the,” and “the” used in this application specification and appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0116] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method of intelligent control of a motorcycle caliper, characterized in that, The method comprises the following steps: S101, collecting original multi-modal data including vehicle speed, wheel speed, piston displacement of motorcycle caliper, brake disc temperature, motorcycle inclination angle and lateral acceleration in real time through multi-source sensors integrated in the motorcycle, performing time alignment and filtering on the original multi-modal data to obtain preprocessed multi-modal data; S102, calculating the current braking demand level and the road adhesion coefficient according to the preprocessed multi-modal data, combining the time-aligned and filtered motorcycle inclination angle and lateral acceleration to identify the real-time braking scene, and the real-time braking scene includes straight-line braking, corner braking or emergency braking; S103, inputting the preprocessed multi-modal data into a pre-trained braking control machine learning model, and the braking control machine learning model generates a corresponding target braking force change curve based on the identified real-time braking scene; S104, controlling the motorcycle drive motor and the motorcycle caliper to brake based on the target braking force change curve, collecting braking effect feedback data, periodically optimizing and training the braking control machine learning model according to historical driving habit data and the collected braking effect feedback data, updating parameters of the braking control machine learning model, and obtaining an optimized braking control machine learning model; S105, predicting the decay trend of the caliper braking efficiency through a preset component degradation model based on the wear state of the brake disc and the hysteresis characteristics of the piston displacement, compensating the braking force control amount based on the decay trend of the caliper braking efficiency, updating the optimized braking control machine learning model, and obtaining a final braking control machine learning model; S106, obtaining an updated target braking force change curve based on the final braking control machine learning model, driving the piston mechanism of the motorcycle caliper through the motorcycle drive motor based on the updated target braking force change curve, and adjusting the motorcycle drive motor current in real time to adapt to the target braking force by adopting a closed-loop control algorithm based on the feedback piston displacement; S107, continuously monitoring the brake disc temperature during braking, and triggering a thermal decay compensation strategy if the brake disc temperature exceeds a preset brake disc temperature threshold to dynamically correct the output braking force.

2. The intelligent control method of a motorcycle caliper according to claim 1, characterized in that, In step S101, the original multi-modal data including vehicle speed, wheel speed, piston displacement of motorcycle caliper, brake disc temperature, motorcycle inclination angle and lateral acceleration is collected in real time through multi-source sensors integrated in the motorcycle, the original multi-modal data is time-aligned and filtered to obtain preprocessed multi-modal data, and the preprocessed multi-modal data comprises: The speed and wheel speed signals are obtained through the CAN bus, the piston displacement of the motorcycle caliper is measured by using a linear displacement sensor, the brake disc temperature is monitored by using a non-contact infrared temperature sensor, and the motorcycle inclination and lateral acceleration are collected by using a six-axis IMU integrated in the vehicle body, so as to obtain original multi-modal data; a data fusion buffer is set in the central processing unit, and the original multi-modal data are time-aligned by using a hardware clock synchronization combined with an interpolation algorithm; the wheel speed is filtered by using a low-pass filter to suppress road noise, the motorcycle inclination and lateral acceleration are filtered by using a Kalman filter, and the piston displacement and brake disc temperature are filtered by using a sliding average filter, so as to finally output the pretreated multi-modal data.

3. The intelligent control method of a motorcycle caliper according to claim 1, characterized in that, In step S102, the current braking demand level and the road adhesion coefficient are calculated according to the pretreated multi-modal data, the time-aligned and filtered motorcycle inclination and lateral acceleration are combined, and the real-time braking scene is identified, including straight-line braking, corner braking or emergency braking, including: The current braking demand level is calculated by using a preset mapping function based on the pretreated piston displacement and the piston displacement rate of change; the wheel slip rate is calculated by using the pretreated wheel speed and vehicle speed, and the current road adhesion coefficient is iteratively calculated by using the ratio of the real-time back-propagated wheel-end braking force to the tire vertical load based on the vehicle dynamics model; a state machine with priorities is defined, in which emergency braking has the highest priority, corner braking has the second priority, and straight-line braking has the lowest priority; when the braking demand level exceeds a preset braking demand level, no matter the values of the motorcycle inclination and lateral acceleration, emergency braking is immediately determined; if it is not emergency braking, when the absolute value of the motorcycle inclination continuously exceeds a preset motorcycle inclination threshold and the lateral acceleration exceeds a preset lateral acceleration threshold, corner braking is determined, and the stability of the corner is further subdivided in combination with the current road adhesion coefficient; if it is not emergency braking, when the motorcycle inclination continuously falls below the preset motorcycle inclination threshold and the lateral acceleration continuously falls below the preset lateral acceleration threshold, straight-line braking is determined.

4. The intelligent control method of a motorcycle caliper according to claim 1, characterized in that, In step S103, the pretreated multi-modal data are input into a pre-trained braking control machine learning model, and the braking control machine learning model generates a corresponding target braking force change curve based on the identified real-time braking scene, including: The pre-trained brake control machine learning model is optimized by supervised learning on a training set composed of historical data and simulation data, and the loss function used by the brake control machine learning model simultaneously punishes factors such as excessive braking distance, excessive tire slip, and vehicle body attitude instability; the multi-modal data preprocessed in step S101 and the real-time braking scene, current braking demand level, and current road adhesion coefficient output in step S102 are integrated into a high-dimensional feature vector and input into the pre-trained brake control machine learning model; the brake control machine learning model learns the complex nonlinear mapping between the high-dimensional feature vector and the optimal braking force response through a deep network structure, and generates a target braking force change curve based on the multi-objective optimization knowledge internalized through offline training.

5. The intelligent control method of a motorcycle caliper according to claim 1, characterized in that, In step S104, the motorcycle driving motor and the motorcycle caliper are controlled based on the target braking force change curve to collect braking effect feedback data, and the brake control machine learning model is periodically optimized and trained according to historical driving habit data and collected braking effect feedback data, the parameters of the brake control machine learning model are updated, and an optimized brake control machine learning model is obtained, including: The target braking force change curve generated in S103 is analyzed into a real-time current braking control instruction for the motorcycle driving motor through a PID controller, and the motorcycle driving motor drives the piston of the motorcycle caliper to generate a target braking force; braking effect feedback data is synchronously collected, including actual piston displacement, driving motor current, braking pressure, wheel speed, vehicle body deceleration, and vehicle body attitude data, and the braking control instruction and real-time braking scene are combined into time-stamped historical driving habit data; the newly collected braking effect feedback data and the historical driving habit data are merged to form an expanded data set on a cloud or background server, and the brake control machine learning model is incrementally trained or fine-tuned based on the expanded data set, the parameters of the brake control machine learning model are updated through a backpropagation algorithm, and an optimized brake control machine learning model is obtained.

6. The intelligent control method of a motorcycle caliper according to claim 1, characterized in that, In step S105, based on the wear state of the brake disc and the hysteresis characteristics of the piston displacement, the decay trend of the caliper braking efficiency is predicted through a pre-set component degradation model, and the braking force control amount is compensated based on the decay trend of the caliper braking efficiency, the optimized brake control machine learning model is updated, and a final brake control machine learning model is obtained, including: The wear state of the brake disc is estimated by monitoring the increasing trend of the reference value of the piston displacement of the motorcycle caliper under the same brake control instruction over a long period of time; the hysteresis characteristic of the piston displacement is calculated by analyzing the shape change of the piston displacement and the closed-loop current curve of the driving motor in the braking cycle; the wear state and the hysteresis characteristic parameters are input into a preset component degradation model, the component degradation model calculates and predicts the decay trend of the braking efficiency of the motorcycle caliper; the braking efficiency coefficient is calculated in real time based on the decay trend of the braking efficiency, and the feedforward compensation braking force control amount for the brake control instruction is calculated based on the braking efficiency coefficient to offset the performance degradation; the braking efficiency coefficient is used as a new state feature or a data set simulating different degradation levels to retrain or fine-tune the optimized brake control machine learning model obtained in step S104, so as to obtain a final brake control machine learning model that can adapt to component aging.

7. The intelligent control method of a motorcycle caliper according to claim 1, characterized in that, In step S106, the updated target brake force change curve is obtained based on the final brake control machine learning model, and based on the updated target brake force change curve, the piston mechanism of the motorcycle caliper is driven by the motorcycle driving motor, and at the same time, the motorcycle driving motor current is adjusted in real time by using a closed-loop control algorithm based on the feedback piston displacement to adapt to the target brake force, including: Based on the final brake control machine learning model obtained in step S105, an updated target brake force change curve is generated, and the updated target brake force change curve is converted into a target piston displacement curve through a preset calibration mapping relationship; the piston of the motorcycle caliper is driven by the motorcycle driving motor, and at the same time, the actual piston displacement is collected in real time as a feedback signal; a closed-loop control algorithm is used to continuously compare the target piston displacement with the actual piston displacement, and the driving current of the motorcycle driving motor is calculated and adjusted in real time according to the displacement error; the motor output torque is controlled by the power driver to make the piston displacement accurately track the target piston displacement curve.

8. The intelligent control method of a motorcycle caliper according to claim 1, characterized in that, In step S107, the brake disc temperature is continuously monitored during braking, and if the brake disc temperature exceeds the preset brake disc temperature threshold, the heat attenuation compensation strategy is triggered to dynamically correct the output brake force, including: During braking, the brake disc temperature is continuously monitored by a non-contact infrared temperature sensor, and is compared in real time with the preset brake disc temperature threshold; when the monitored brake disc temperature exceeds the preset brake disc temperature threshold, the heat attenuation compensation strategy is triggered, the heat attenuation compensation strategy dynamically calculates a heat attenuation compensation coefficient less than 1 based on the real-time temperature and the temperature rise rate; the brake control instruction from the PID controller is divided by the heat attenuation compensation coefficient to obtain a dynamically increased corrected brake force instruction; the corrected brake force instruction is sent to the actuator closed-loop control system, and the compensated brake force is finally output by adjusting the motorcycle driving motor current; the heat attenuation compensation strategy is continuously effective until the brake disc temperature falls below the preset brake disc temperature threshold, and the heat attenuation compensation strategy is automatically exited.

9. An intelligent control system for a motorcycle caliper, characterized in that, The system comprises: The multi-source data acquisition and preprocessing module is configured to acquire original multi-modal data of vehicle speed, wheel speed, piston displacement, brake disc temperature, motorcycle inclination angle and lateral acceleration in real time through multi-source sensors integrated in the motorcycle, and perform time alignment and filtering on the original multi-modal data to obtain preprocessed multi-modal data. The braking scenario recognition module is configured to calculate a current braking demand level and a road adhesion coefficient based on the preprocessed multi-modal data, and recognize a real-time braking scenario in combination with the motorcycle inclination angle and lateral acceleration, wherein the real-time braking scenario includes straight-line braking, corner braking or emergency braking. The intelligent braking control module includes a pre-trained braking control machine learning model, is configured to receive the preprocessed multi-modal data, and generate a corresponding target braking force change curve based on the real-time braking scenario; the module is further configured to periodically optimize and train the braking control machine learning model based on braking effect feedback data and historical driving habit data, and correct the model based on a caliper braking efficiency decay trend predicted by a component degradation model to obtain a final braking control machine learning model, thereby outputting an updated target braking force change curve. The braking force execution and closed-loop regulation module is configured to drive the piston mechanism of the motorcycle caliper to brake based on the updated target braking force change curve through the motorcycle driving motor, and adjust the current of the motorcycle driving motor in real time to adapt to the target braking force by using a closed-loop control algorithm based on the feedback piston displacement. The heat attenuation compensation module is configured to continuously monitor the brake disc temperature during braking, and trigger a heat attenuation compensation strategy when the brake disc temperature exceeds a preset threshold to dynamically correct the output braking force.

Images