Magnetorheological semi-active suspension switching control system and method based on image recognition

Abstract

This invention discloses a magnetorheological semi-active suspension switching control system and method based on image recognition. The method includes: using a monocular camera to capture real-time images of the road surface ahead of the vehicle; identifying obstacles using a pre-trained deep learning model; the controller matching control parameters adapted to the current vehicle speed and obstacle type using a control parameter switching rule table obtained through simulation, based on real-time vehicle speed data collected by a vehicle speed sensor and the obstacle identification results; the controller calculating the desired current required by the magnetorheological damper based on the control parameters and control algorithm, using real-time collected vehicle dynamic data; and the current source outputting the actual current to the magnetorheological damper according to the calculated desired current, thereby changing the damping force by adjusting the current input to the magnetorheological damper. This invention enables real-time identification of obstacles ahead and, through optimized control strategies, improves the comfort and stability of the suspension system under complex road conditions, especially significantly improving vibration reduction performance when passing typical obstacles.

Description

Technical Field

[0001] This invention belongs to the field of vehicle control technology, specifically a magnetorheological semi-active suspension switching control system and method based on image recognition. Background Technology

[0002] With the rapid development of the automotive industry, vehicle suspension systems have evolved from traditional passive suspensions to semi-active and active suspension systems. Passive suspension systems absorb vibrations from the road surface solely through springs and dampers, lacking adjustability and resulting in poor comfort and stability under complex road conditions. Semi-active suspension systems, on the other hand, adjust the damping force of the dampers to adapt to different road conditions, thereby improving comfort and stability. Compared to passive suspension, semi-active suspension can more flexibly handle different road conditions, enhancing the vehicle's dynamic performance. Active suspension further adjusts the vehicle's suspension mechanical characteristics through an active control system, achieving more precise control. Nevertheless, semi-active suspension remains the most common choice in the market today due to its lower implementation cost and control complexity, especially in mid-to-high-end models.

[0003] Magnetorheological dampers (MR dampers), as a core component of semi-active suspension systems, have been widely used in recent years. MR dampers utilize the properties of magnetorheological fluids, adjusting the viscosity of the fluid by changing the strength of an applied magnetic field, thereby controlling the damping force. This principle allows MR dampers to rapidly adjust the damping force within milliseconds, achieving a fast response to vehicle vibrations. Compared to traditional hydraulic dampers, MR dampers offer better adjustability and response speed, and can provide different damping effects under various operating conditions, adapting to diverse and complex road conditions. Due to their superior performance and controllability, MR dampers are widely used in the suspension systems of automobiles, railways, and other transportation vehicles.

[0004] Research on suspension control algorithms is one of the key technologies in semi-active suspension systems. Control algorithms are used to adjust the working state of the dampers based on signal inputs (such as vehicle speed, acceleration, etc.) and the vehicle's dynamic response, optimizing the vehicle's comfort and handling under different road conditions. Common suspension control algorithms include methods based on fuzzy logic control, adaptive control, robust control, and optimal control. In addition, in recent years, model predictive control (MPC), as a forward-looking control method, has been increasingly applied to suspension systems. MPC can adjust the damper parameters in advance based on future road conditions, thereby achieving more precise control. However, the implementation of suspension control algorithms still faces certain challenges, especially in balancing multiple objectives (such as comfort and handling stability). For example, a method for comfort control of an unmanned vehicle integrating speed planning and pre-aiming semi-active suspension disclosed in Chinese Patent (Application No.: CN202010080594.7) only controls and optimizes comfort, neglecting the vehicle's handling stability.

[0005] Despite significant advancements in suspension control algorithms, the utilization of road surface information in suspension systems remains limited in practical applications. Most current suspension systems rely on real-time dynamic signals such as vehicle speed and acceleration for control, neglecting the influence of road conditions ahead. With the development of sensor technologies like LiDAR and cameras, acquiring road surface information has become more feasible. Nevertheless, effectively integrating this information into the suspension control system, especially dynamically adjusting suspension parameters under various road conditions, remains a challenging research problem. Existing solutions mostly rely solely on the vehicle's current dynamic response, lacking forward-looking information processing capabilities, resulting in the suspension system failing to fully realize its potential performance. For example, a semi-active suspension control method and system disclosed in Chinese Patent (Application No.: CN202310079638.8) can determine a fuzzy adjustment strategy and adjust control parameters based on the performance parameters of the semi-active suspension and random road disturbance information. However, it still requires the vehicle to travel to the corresponding road surface before adjusting the control parameters, failing to achieve pre-adjustment of the control algorithm, resulting in poor performance. Summary of the Invention

[0006] The purpose of this invention is to address the problems existing in the prior art by providing a magnetorheological semi-active suspension switching control system based on image recognition.

[0007] The technical solution to achieve the purpose of this invention is as follows: On the one hand, a magnetorheological semi-active suspension switching control system based on image recognition is provided. The system includes a monocular camera, a vehicle speed sensor, an acceleration sensor, a controller, a current source, and several magnetorheological dampers.

[0008] The monocular camera is used to capture real-time images of the road surface in front of the vehicle.

[0009] The vehicle speed sensor is used to collect vehicle speed data in real time;

[0010] The sensor module is used to collect vehicle dynamic data in real time;

[0011] The controller is used to identify obstacles based on a deep learning image recognition model and road surface images; it is also used to match control parameters adapted to the current vehicle speed and obstacle type based on real-time vehicle speed data collected by the vehicle speed sensor, combined with obstacle recognition results, and a control parameter switching rule table obtained through simulation in advance; it is also used to calculate the desired current based on the control algorithm, control parameters, and vehicle dynamic data collected by the sensor module, which reflects the magnitude of the damping force required by the magnetorheological damper.

[0012] The current source is used to output an actual current to the magnetorheological damper based on the calculation results output by the controller.

[0013] The magnetorheological damper adjusts its current input through a current source to change its damping force, thereby reducing or enhancing vehicle vibration.

[0014] Furthermore, the process of establishing the deep learning image recognition model includes:

[0015] Step 1.1, Data Acquisition and Preprocessing: A large number of road surface images are acquired using a monocular camera and image enhancement processing is performed. The acquired road surface images include road surface images under different lighting, weather and road conditions, as well as road surface images containing various types of typical obstacles. All images are labeled, including the category of the obstacles.

[0016] Step 1.2: Construct a dataset from the road surface images collected in Step 1.1, and divide the dataset into a training set, a validation set, and a test set;

[0017] Step 1.3, Model Training and Optimization: The convolutional neural network model is trained using the training set through the backpropagation algorithm. During the training process, the cross-entropy loss function is used to classify each type of obstacle, and the model weights are continuously adjusted through the optimization algorithm to minimize the error and improve the recognition accuracy.

[0018] Step 1.4, Model Validation and Tuning: After the model training is completed, performance evaluation is performed on the validation set and test set. Based on the evaluation results, the model is further tuned.

[0019] Further, the step 1.2, which involves constructing a dataset from the road surface images acquired in step 1.1, specifically involves performing transformations on the road surface images acquired in step 1.1, including at least rotation, scaling, and translation, to increase the amount of image data. The dataset is then constructed from the original images and the transformed images.

[0020] Furthermore, in the training process of step 1.3, data augmentation techniques and regularization methods are used to prevent overfitting.

[0021] Furthermore, in step 1.4, the model's recognition ability is evaluated by calculating at least the accuracy, recall, and F1 score of the model under different obstacle types.

[0022] Furthermore, the control parameters include, but are not limited to, the ceiling damping coefficients of different obstacles at different vehicle speeds, and the control algorithm includes, but is not limited to, the ceiling control algorithm.

[0023] Furthermore, the process of establishing the control parameter switching rule table specifically includes:

[0024] Step 2.1: Determine the current control objective based on different driving conditions;

[0025] Step 2.2: Using a simulation model based on vehicle dynamic response, the optimal ceiling damping coefficient under different control objectives is determined by an optimization algorithm at different vehicle speeds and obstacle types.

[0026] Step 2.3: Based on the results of Steps 2.1 and 2.2, a rule table is formed, which lists the optimal ceiling damping coefficients under different driving conditions.

[0027] Furthermore, the control objectives mentioned in step 2.1 include: prioritizing comfort when driving at low speeds; prioritizing handling stability when driving at high speeds; and the custom settings for low and high speeds.

[0028] Furthermore, the control objective described in step 2.1 is dynamically adjusted based on the driver's needs or the automatic detection system.

[0029] On the other hand, a magnetorheological semi-active suspension switching control method based on image recognition is provided, the method comprising:

[0030] Step 1: Use a monocular camera to capture real-time images of the road surface in front of the vehicle, and use a pre-trained deep learning model to identify obstacles.

[0031] Step 2: The controller uses the real-time vehicle speed data collected by the vehicle speed sensor and the obstacle recognition results to match the control parameters that are suitable for the current vehicle speed and obstacle type using the control parameter switching rule table obtained through simulation in advance.

[0032] Step 3: The controller transmits the matched control parameters to the control algorithm, which calculates the desired current required by the magnetorheological damper based on the real-time collected vehicle dynamic data.

[0033] Step 4: The current source outputs the actual current to the magnetorheological damper according to the calculated expected current. By adjusting the current input of the magnetorheological damper, its damping force is changed to reduce or enhance the vibration of the vehicle.

[0034] Compared with the prior art, the significant advantages of this invention are:

[0035] (1) The present invention can identify obstacles in front in real time through a monocular camera and dynamically switch control parameters based on the identification results to optimize the performance of the suspension system, especially when passing through typical obstacles, it can significantly improve comfort and stability.

[0036] (2) Using deep learning algorithms for obstacle recognition can improve the accuracy and real-time performance of recognition, and at a lower cost.

[0037] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the system composition of the present invention.

[0039] Figure 2 This is a flowchart for obstacle recognition and control parameter switching. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0041] Combination Figure 1 The present invention provides a magnetorheological semi-active suspension switching control system based on image recognition. The system includes a monocular camera, a vehicle speed sensor, an acceleration sensor, a controller, a current source, and several magnetorheological dampers.

[0042] The monocular camera is used to capture real-time images of the road surface in front of the vehicle.

[0043] The vehicle speed sensor is used to collect vehicle speed data in real time;

[0044] The sensor module (including but not limited to the vehicle acceleration sensor) is used to collect vehicle dynamic data in real time;

[0045] The controller is used to identify obstacles based on a deep learning image recognition model and road surface images; it is also used to match control parameters adapted to the current vehicle speed and obstacle type based on real-time vehicle speed data collected by the vehicle speed sensor, combined with obstacle identification results, and a control parameter switching rule table obtained through prior simulation; it is also used to calculate the desired current based on the control algorithm, control parameters, and vehicle dynamic data collected by the sensor module. This current reflects the required damping force of the magnetorheological damper, thereby affecting the response of the suspension system (whenever the system identifies a specific obstacle and combines it with the current vehicle speed, the controller will match the most suitable roof damping coefficient from the rule table. The matched control parameters will be passed to the control algorithm and the response of the suspension system will be adjusted to meet the current comfort or handling stability goals).

[0046] The current source is used to output an actual current to the magnetorheological damper based on the calculation results output by the controller.

[0047] The magnetorheological damper, by adjusting its current input through a current source, changes its damping force, thereby affecting the vibration response of the vehicle suspension system and providing optimal comfort and stability.

[0048] Furthermore, in one embodiment, the process of establishing the deep learning image recognition model includes:

[0049] Step 1.1, Data Acquisition and Preprocessing: A large number of road surface images are acquired using a monocular camera and image enhancement processing is performed. The acquired road surface images include road surface images under different lighting, weather and road conditions, as well as road surface images containing various types of typical obstacles. All images are labeled, including the category of the obstacles.

[0050] Typical obstacles here include speed bumps, puddles, and potholes. The categories of obstacles include speed bumps and puddles.

[0051] Step 1.2: Construct a dataset from the road surface images collected in Step 1.1, and divide the dataset into a training set, a validation set, and a test set;

[0052] Here, the training set is used for model learning, the validation set is used to tune the model's hyperparameters, and the test set is used to evaluate the model's accuracy.

[0053] Step 1.3, Model Training and Optimization: The convolutional neural network model is trained using the training set through the backpropagation algorithm. During the training process, the cross-entropy loss function is used to classify each type of obstacle, and the model weights are continuously adjusted through the optimization algorithm to minimize the error and improve the recognition accuracy.

[0054] Step 1.4, Model Validation and Tuning: After the model training is completed, performance evaluation is performed on the validation set and test set. Based on the evaluation results, the model is further tuned to improve its recognition accuracy and real-time performance in practical applications.

[0055] Preferably, in some embodiments, the step 1.2 of constructing a dataset from the road surface images acquired in step 1.1 specifically involves performing transformations on the road surface images acquired in step 1.1, including at least rotation, scaling, and translation, to increase the amount of image data, and constructing a dataset from the original image and the transformed image.

[0056] Preferably, in some embodiments, during the training process of step 1.3, data augmentation techniques and regularization methods are used to prevent overfitting and ensure the model's generalization ability in complex environments.

[0057] Preferably, in some embodiments, step 1.4 evaluates the model's recognition ability by at least calculating the model's accuracy, recall, and F1 score under different obstacle types.

[0058] Furthermore, in one embodiment, the control parameters include, but are not limited to, the ceiling damping coefficients of different obstacles at different vehicle speeds, and the control algorithm includes, but is not limited to, a ceiling control algorithm.

[0059] Furthermore, in one embodiment, the process of establishing the control parameter switching rule table specifically includes:

[0060] Step 2.1: Determine the current control objective based on different driving conditions;

[0061] Step 2.2: Using a simulation model based on vehicle dynamic response, the optimal ceiling damping coefficient under different control objectives is determined by an optimization algorithm at different vehicle speeds and obstacle types.

[0062] Step 2.3: Based on the results of Steps 2.1 and 2.2, a rule table is formed, which lists the optimal ceiling damping coefficients under different driving conditions.

[0063] Preferably, in some embodiments, the control objectives in step 2.1 include: prioritizing comfort (reducing passenger vibration) when driving at low speeds; prioritizing handling stability (improving vehicle stability) when driving at high speeds; and the low-speed and high-speed custom settings.

[0064] Preferably, in some embodiments, the control objective described in step 2.1 is dynamically adjusted based on the driver's needs or an automatic detection system.

[0065] Preferably, in some embodiments, the optimization algorithm in step 2.2 includes, but is not limited to, genetic algorithm and particle swarm optimization algorithm.

[0066] Combination Figure 2 This paper presents a magnetorheological semi-active suspension switching control method based on image recognition, specifically including:

[0067] Step 1: Use a monocular camera to capture real-time images of the road surface in front of the vehicle, and use a pre-trained deep learning model to identify obstacles.

[0068] Step 2: The controller uses the real-time vehicle speed data collected by the vehicle speed sensor and the obstacle recognition results to match the control parameters that are suitable for the current vehicle speed and obstacle type using the control parameter switching rule table obtained through simulation in advance.

[0069] Step 3: The controller transmits the matched control parameters to the control algorithm, which calculates the desired current required by the magnetorheological damper based on the real-time collected vehicle dynamic data.

[0070] Step 4: The current source outputs the actual current to the magnetorheological damper according to the calculated expected current. By adjusting the current input of the magnetorheological damper, its damping force is changed to reduce or enhance the vibration of the vehicle.

[0071] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention without departing from its spirit and scope should be included within the protection scope of the present invention.

Claims

1. A magnetorheological semi-active suspension switching control system based on image recognition, characterized in that, The system includes a monocular camera, a vehicle speed sensor, an acceleration sensor, a controller, a current source, and several magnetorheological dampers. The monocular camera is used to capture real-time images of the road surface in front of the vehicle. The vehicle speed sensor is used to collect vehicle speed data in real time; The acceleration sensor is used to collect vehicle dynamic data in real time; The controller is used to identify obstacles based on a deep learning image recognition model and a road surface image; It is also used to match control parameters that are suitable for the current vehicle speed and obstacle type based on real-time vehicle speed data collected by the vehicle speed sensor and obstacle recognition results, through a control parameter switching rule table obtained in advance through simulation. It is also used to calculate the desired current based on the control algorithm, control parameters and vehicle dynamic data collected by the acceleration sensor. This current reflects the magnitude of the damping force required by the magnetorheological damper. The current source is used to output an actual current to the magnetorheological damper based on the calculation results output by the controller. The magnetorheological damper adjusts its current input through a current source to change its damping force, thereby reducing or enhancing vehicle vibration. The process of establishing the deep learning image recognition model includes: Step 1.1, Data Acquisition and Preprocessing: A large number of road surface images are acquired using a monocular camera and image enhancement processing is performed. The acquired road surface images include road surface images under different lighting, weather and road conditions, as well as road surface images containing various types of typical obstacles. All images are labeled, including the category of the obstacles. Step 1.2: Construct a dataset from the road surface images collected in Step 1.1, and divide the dataset into a training set, a validation set, and a test set; Step 1.3, Model Training and Optimization: The convolutional neural network model is trained using the training set through the backpropagation algorithm. During the training process, the cross-entropy loss function is used to classify each type of obstacle, and the model weights are continuously adjusted through the optimization algorithm to minimize the error and improve the recognition accuracy. Step 1.4, Model Validation and Tuning: After the model training is completed, performance evaluation is performed on the validation set and test set, and the model is further tuned based on the evaluation results; The control parameters include the canopy damping coefficients of different obstacles at different vehicle speeds, and the control algorithm includes a canopy control algorithm. The process of establishing the control parameter switching rule table specifically includes: Step 2.1: Determine the current control objective based on different driving conditions; the control objective is dynamically adjusted based on the driver's needs or the automatic detection system. Step 2.2: Using a simulation model based on vehicle dynamic response, the optimal ceiling damping coefficient under different control objectives is determined by an optimization algorithm at different vehicle speeds and obstacle types. Step 2.3: Based on the results of Steps 2.1 and 2.2, a rule table is formed, which lists the optimal ceiling damping coefficients under different driving conditions; The control objectives mentioned in step 2.1 include: prioritizing comfort when driving at low speeds; prioritizing handling stability when driving at high speeds; and custom settings for low and high speeds.

2. The magnetorheological semi-active suspension switching control system based on image recognition according to claim 1, characterized in that, Step 1.2, which involves constructing a dataset from the road surface images acquired in step 1.1, specifically involves performing transformations on the road surface images acquired in step 1.1, including at least rotation, scaling, and translation, to increase the amount of image data. The dataset is then constructed from the original images and the transformed images.

3. The magnetorheological semi-active suspension switching control system based on image recognition according to claim 1, characterized in that, In step 1.3, data augmentation and regularization methods are used during training to prevent overfitting.

4. The magnetorheological semi-active suspension switching control system based on image recognition according to claim 1, characterized in that, In step 1.4, the model's recognition ability is evaluated by calculating at least the accuracy, recall, and F1 score of the model under different obstacle types.

5. The magnetorheological semi-active suspension switching control system based on image recognition according to claim 1, characterized in that, The optimization algorithms in step 2.2 include genetic algorithms and particle swarm optimization algorithms.

6. A magnetorheological semi-active suspension switching control method based on image recognition according to any one of claims 1 to 5, characterized in that, The method includes: Step 1: Use a monocular camera to capture real-time images of the road surface in front of the vehicle, and use a pre-trained deep learning model to identify obstacles. Step 2: The controller uses the real-time vehicle speed data collected by the vehicle speed sensor and the obstacle recognition results to match the control parameters that are suitable for the current vehicle speed and obstacle type using the control parameter switching rule table obtained through simulation in advance. Step 3: The controller transmits the matched control parameters to the control algorithm, which calculates the desired current required by the magnetorheological damper based on the real-time collected vehicle dynamic data. Step 4: The current source outputs the actual current to the magnetorheological damper according to the calculated expected current. By adjusting the current input of the magnetorheological damper, its damping force is changed to reduce or enhance the vibration of the vehicle.

Images