Intelligent adaptive gearbox and multi-mode cooperative control system and method thereof

Abstract

The invention discloses an intelligent self-adaptive gearbox system based on a multi-mode fusion control theory and a control method of the intelligent self-adaptive gearbox system, and belongs to the technical field of vehicle transmission systems. The system adopts an innovative four-layer intelligent control architecture, and comprises a basic transmission layer used for hardware interaction; the state sensing layer is used for realizing dynamic balance, multi-physics field efficiency evaluation and working condition identification through wavelet transformation and based on an efficiency pulse spectrogram pre-established by finite element and computational fluid dynamics joint simulation and a support vector machine algorithm; the intelligent decision-making layer is used for generating a globally optimized gear shifting instruction through the cooperation of nonlinear model predictive control, graph theory A search and a time sequence planning algorithm; and the adaptive optimization layer realizes health prediction and strategy persistent evolution through a hierarchical reinforcement learning framework of a hidden Markov model and offline experience playback. Through engineering fusion of a multi-mode theory, the technical problems that an existing gearbox control strategy is rigid and is lack of self-adaption and predictive capacity are solved.

Description

Technical Field

[0001] This invention relates to the field of vehicle transmission system technology, specifically to an intelligent adaptive transmission system and its control method, and in particular to a system architecture and algorithm implementation scheme that improves the overall performance of the transmission by integrating multiple advanced control theories. Background Technology

[0002] Currently, automatic transmission control systems largely rely on pre-calibrated shift pulse patterns and fixed control logic. When faced with complex and ever-changing actual driving conditions, different driver styles, and the complex multi-physics (mechanical, thermal, and fluid) coupling effects within the transmission, these methods reveal three core limitations: First, the perception dimension is shallow, relying on only a few sensors, making it impossible to deeply assess the system's efficiency and health status; second, the decision-making ability is rigid, lacking foresight and global optimization capabilities; and third, the system lacks evolution, unable to adapt to changes or predict faults through continuous learning.

[0003] While advanced algorithms such as model predictive control and machine learning theoretically offer the possibility of solving the above problems, in engineering practice, deeply integrating these heterogeneous algorithms and meeting the stringent requirements of automotive-grade systems under the constraints of real-time performance, reliability, and computing power remains a significant challenge. Currently, no publicly available architecture for a transmission control system can systematically solve the entire chain of problems from "deep perception" to "intelligent decision-making" and then to "continuous evolution." Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent adaptive transmission system and its control method based on multimodal fusion control theory. The core innovation of this solution lies not in proposing new fundamental theoretical formulas, but in creatively constructing a hierarchical, decoupled, and collaborative engineering implementation architecture. This architecture organically integrates various known advanced control algorithms and solves the key problems of their engineering implementation in automotive embedded environments, thereby systematically improving the overall performance of the transmission.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is summarized as follows: This invention proposes a four-layer intelligent control architecture.

[0006] In the basic transmission layer, all-round state perception is achieved through multiple types of distributed sensors (such as fiber optic temperature arrays and MEMS accelerometers).

[0007] At the state perception layer, an innovative strategy of "offline high-fidelity simulation modeling and online efficient table lookup evaluation" is adopted. That is, an efficiency pulse spectrum covering all working conditions is pre-established through joint simulation of finite element and fluid dynamics. In real-time control, the multi-physics coupling efficiency is quickly obtained by table lookup interpolation, which solves the problem of real-time calculation of complex physical models. At the same time, wavelet transform and support vector machine algorithms are combined to handle vibration and working condition classification problems respectively.

[0008] At the intelligent decision-making layer, nonlinear model predictive control (for local fine-grained optimization), graph theory A-search (for global energy management), and time-series planning (for multi-actuator coordination) are combined in parallel to generate the optimal shift decision.

[0009] In the adaptive optimization layer, a hidden Markov model is introduced for health status identification and prediction. A hierarchical reinforcement learning mechanism of "online data collection and offline asynchronous updates" is designed to enable the system to continuously optimize the control strategy throughout its life cycle, while avoiding the computing power bottleneck caused by high-intensity neural network training in the real-time control loop.

[0010] Through the above architecture, this invention constructs a complete technical closed loop from accurate perception to intelligent decision-making and then to self-learning optimization. Beneficial effects

[0011] Compared with the prior art, the present invention can achieve the following beneficial effects through the above technical solution: 1. Improved Shift Smoothness and Dynamic Quality: Through the synergistic effect of precise programming of mechanical actions using a timing-based programming algorithm, optimization of the clutch engagement process using nonlinear model predictive control, and dynamic vibration suppression based on wavelet transform and Lyapunov stability, a multi-layered, closed-loop algorithmic guarantee is provided to fundamentally reduce shift shock and improve ride comfort. Simulation analysis shows that this synergistic mechanism exhibits a significantly better improvement trend in shift smoothness compared to traditional single control strategies.

[0012] 2. Comprehensive Energy Efficiency Optimization of the Transmission System: An innovative strategy of "offline high-fidelity simulation modeling and online efficient table lookup evaluation" is adopted. Through real-time multiphysics efficiency evaluation based on efficiency pulse spectrum diagrams, the system is able to perceive the comprehensive energy loss state of mechanical and hydraulic systems during operation for the first time. Combined with the global energy path planning capability of the graph theory A* search algorithm, the system can proactively select the high-efficiency zone during gear shifting decisions, thereby optimizing the energy flow of the entire vehicle. This method provides a perception and decision-making dimension for improving the comprehensive efficiency of the transmission system that is not available in traditional control systems.

[0013] 3. The system possesses predictive maintenance and continuous evolution capabilities: By identifying and predicting the system's health status through a Hidden Markov Model, it achieves a shift from "post-event maintenance" to "pre-event early warning," laying the technical foundation for implementing predictive maintenance and extending component lifespan. More importantly, by designing a hierarchical reinforcement learning mechanism of "online data collection and offline asynchronous training and updating," reinforcement learning has been successfully introduced into the safety-critical vehicle control system. This enables the control strategy to continuously adapt to changes in driver habits and operating conditions, achieving full lifecycle performance evolution of the system. This mechanism effectively resolves the core contradiction in the engineering implementation of self-learning algorithms in real-time control systems.

[0014] 4. Engineering Feasibility and High Reliability: The entire technical solution is designed with full consideration of the hard constraints of automotive-grade embedded systems. Core algorithms are implemented through mature engineering pathways; for example, the efficiency pulse spectrum diagram (LUT) avoids complex online simulations, embedded code generation tools (such as ACADO) solve the deployment challenges of optimization algorithms, and the offline asynchronous training mechanism separates high-load learning from high-real-time control. Simultaneously, the four-layer architecture achieves functional decoupling, significantly improving the modularity, testability, and maintainability of the system software, ensuring the overall reliability of the system. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the four-layer control architecture of the intelligent adaptive transmission system of the present invention.

[0016] Figure 2 This is a schematic diagram illustrating the implementation process of the "offline simulation-online table lookup" module in the multiphysics coupling analysis of this invention.

[0017] Figure 3 This is a schematic diagram illustrating the principle of multi-algorithm collaborative operation in the intelligent decision-making layer of this invention.

[0018] Figure 4 This is a schematic diagram of the closed loop of health prediction and policy learning in the adaptive optimization layer of this invention.

[0019] Figure 5 This is a flowchart illustrating the overall steps of the control method of the present invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. Those skilled in the art will understand that the following description is intended to illustrate the feasibility of the solutions and is not intended to be the sole limitation of the present invention.

[0021] This system can be upgraded based on existing automatic transmissions (such as AT or DCT). Core hardware includes: 1. Central Processing Unit: Employs an automotive-grade multi-core microcontroller compliant with AUTOSAR standards, such as the Infineon TC397. This chip provides sufficient computing power to run the control algorithm and meets the functional safety ASIL-D level requirements.

[0022] Angular displacement sensor: Employs an optical encoder to measure the input / output shaft speed.

[0023] Vibration sensors: ADI ADXL357 series triaxial MEMS accelerometers are used, and are placed at the measurement points on the housing determined by modal analysis.

[0024] Temperature sensor: A fiber Bragg grating sensor array is used to read the distributed temperature of key points in the gearbox through a MicronOptics demodulator.

[0025] Pressure sensor: The Kistler 4067E series miniature sensor is integrated into the hydraulic control valve block.

[0026] Position sensor: Melexis MLX90316 linear Hall sensor is used to detect the position of the shift fork.

[0027] Actuators: These include a high-frequency proportional solenoid valve for controlling clutch pressure and a DC motor for shifting gears.

[0028] The software architecture follows the AUTOSAR standard, encapsulating the four-layer control logic described in this invention into independent software components, which communicate through RTE to ensure modularity and portability.

[0029] II. Engineering Implementation Path of Core Algorithm Module

[0030] The core of this module is to resolve the contradiction between "high-precision physical models" and "real-time onboard computing." We adopt the following approach: Step A (Offline High-Fidelity Modeling): In a commercial simulation platform (such as ANSYS Workbench), establish a parametric finite element model and a computational fluid dynamics model of the gearbox. Through co-simulation, traverse the typical operating range of the engine (e.g., speed 1000-6000 rpm, torque 100-400 Nm), and calculate the gear transmission mechanical efficiency (η_mech) and hydraulic system loss efficiency (η_hyd) corresponding to each operating point.

[0031] Step B (Generate Efficiency Pulse Graph): Organize the data obtained in Step A into a two-dimensional lookup table (LUT), i.e., the efficiency pulse graph, where the horizontal and vertical axes represent speed and torque, respectively, and the data is (η_mech, η_hyd). Burn this table into the controller's non-volatile memory.

[0032] Step C (Online Efficient Evaluation): During real-time vehicle operation, the controller reads the current actual engine speed and torque, and instantly retrieves the corresponding efficiency value from the efficiency pulse spectrum using a bilinear interpolation algorithm. The total efficiency can be estimated as η_total = η_mechη_hyd. This method transforms the solution of complex differential equations into a single memory access and simple interpolation calculation, perfectly meeting real-time requirements.

[0033] 2.1 Model Simplification and Code Generation: To meet the requirements of real-time optimization calculations, a simplified lumped-parameter nonlinear prediction model for the gearbox needs to be established. This model can be constructed based on physical laws (moment of inertia, clutch slip formula, etc.). Subsequently, ACADO (a toolkit for automatic control and dynamic optimization) is used to automatically generate highly optimized embedded C code from this model and the defined optimization objectives and constraints (such as pressure limits and impact limits).

[0034] 2.2 Deployment and Operation: The generated C code is integrated into the microcontroller's software components. In each control cycle (e.g., 1 ms), based on the current system state (e.g., rotational speed) and the reference trajectory, the code solves for the optimal control sequence within a set prediction time domain (e.g., 20 steps) and outputs the first control input to the actuator. By using specialized code generation tools, the efficiency and determinism of the optimization algorithm under limited computing power are ensured.

[0035] The key to this module is to design a feasible and safe learning mechanism that does not interfere with real-time control.

[0036] 3.1 Network Structure and Data Flow: A hierarchical reinforcement learning framework is designed. The higher-level network determines the target gear based on the operating conditions, while the lower-level network outputs a specific clutch pressure curve based on the target. During system operation, all sensor data, control commands, and immediate rewards calculated by the model (such as those based on impact intensity and friction work) are combined into an "experience tuple" and stored in a fixed-size experience replay buffer.

[0037] 3.2 Offline Asynchronous Training Mechanism: A crucial design feature is that neural network training does not occur within the real-time control loop. The training task is set as a low-priority task and is only initiated when the system is detected to be in a safe, idle state (e.g., the vehicle is stationary and powered on, or connected to a charging station). At this time, the system randomly samples a batch of historical data from the buffer and performs one round of gradient descent updates for the neural network. Alternatively, on vehicles with vehicle-to-cloud communication capabilities, the data in the buffer can be encrypted and uploaded to a cloud server for larger-scale training, and then the updated model parameters can be securely distributed back to the vehicle.

[0038] 3.3 Model Switching and Validation: The updated policy model is tested and validated in a separate area of ​​memory to ensure its performance is stable before safely replacing the old online model. This mechanism completely separates the computationally intensive training process from the real-time-critical control process, providing a practical solution for applying reinforcement learning in safety-critical systems.

[0039] III. Feasibility Analysis of the Plan Those skilled in the art will recognize that the solution described in this invention has a clear implementation path: 1. The hardware it relies on (automotive-grade MCU, various sensors) are all commercially available industrial-grade products.

[0040] 2. The development tools and methods used (AUTOSAR software architecture, ANSYS simulation, ACADO code generation, TensorFlow / PyTorch for offline training) are all standard or mainstream tools in the current automotive R&D field.

[0041] 3. The proposed core engineering strategies (efficiency pulse spectrum lookup table, predictive control code generation, and reinforcement learning offline update) are all designed specifically for the constraints of the vehicle environment, effectively avoiding the challenges of real-time performance, security, and computing power faced by directly porting theoretical algorithms.

[0042] Therefore, although the patent application stage mainly focuses on principle simulation verification, the detailed engineering implementation path provided by this solution is sufficient to convince those skilled in the art of its industrial feasibility and to carry out specific development, calibration and testing work accordingly. Industrial applicability

[0043] The intelligent adaptive transmission system and its control method provided by this invention offer a complete and specific technical solution, closely integrated with automotive industry engineering practices. It provides a practical and feasible architecture and algorithm integration solution for addressing the challenges of intelligent, adaptive, and predictive maintenance in transmission control. This solution can be widely applied to the development and upgrading of automatic transmission control systems for various passenger and commercial vehicles, aligning with the automotive industry's trend towards intelligent and connected vehicles, and possesses clear industrial applicability and broad market prospects.

Claims

1. An intelligent adaptive gearbox system based on multi-modal fusion control theory, characterized by, The application relates to a central processing unit, a distributed sensor array in communication connection with the central processing unit, and an actuator in communication connection with and controlled by the central processing unit. The distributed sensor array comprises a high-precision angular displacement sensor, a multi-axis acceleration sensor, a fiber Bragg grating temperature sensor array, a micro-electro-mechanical system pressure sensor, and a Hall effect position sensor. The actuator comprises a clutch solenoid valve and a gear shifting motor. The central processing unit is configured to perform multivariate nonlinear coupling control, and is specifically used for: (1) collecting data of the sensor array and controlling the actuator; (2) performing high-order information extraction on the collected data, including: realizing dynamic balance control through wavelet transform and adaptive filtering; performing multi-physical field efficiency evaluation based on an efficiency map pre-built through joint simulation of finite elements and computational fluid dynamics and stored locally; and performing working condition mode identification through a support vector machine algorithm; (3) performing optimization decision based on the extracted information, including: performing rolling time domain optimization through nonlinear model predictive control; performing global path planning in a transmission ratio map through a graph theory A* search algorithm; and coordinating actions of multiple actuators through a time sequence planning algorithm; (4) performing system health management and strategy evolution, including: performing health state prediction based on a hidden Markov model; and updating a control strategy through experience replay and offline asynchronous training based on a hierarchical reinforcement learning framework. The high-precision angular displacement sensor is configured to measure rotational speeds of gearbox input and output shafts and a phase difference; the multi-axis acceleration sensor is configured to monitor vibration acceleration of a gearbox body in three-dimensional space; the fiber Bragg grating temperature sensor array is configured to distributely measure temperature distribution of gear meshing areas, bearing positions and lubricating oil channels; the micro-electro-mechanical system pressure sensor is configured to monitor main pressure of a hydraulic control system and working pressure of a clutch in real time; and the Hall effect position sensor is configured to accurately detect displacement positions of gear shifting forks.

2. The intelligent adaptive transmission system of claim 1, wherein, The realization of dynamic balance control through wavelet transform and adaptive filtering comprises: designing independent adaptive filters for each frequency band vibration component obtained through wavelet transform decomposition, and dynamically calculating an optimal distribution of clutch engagement force based on Lyapunov stability theory.

3. The intelligent adaptive transmission system of claim 1, wherein, The multi-physical field efficiency evaluation based on the efficiency map specifically comprises: the central processing unit obtains corresponding mechanical efficiency and hydraulic efficiency evaluation values from the efficiency map through table lookup and interpolation according to real-time collected rotational speed and torque parameters, and calculates comprehensive efficiency.

4. The intelligent adaptive transmission system of claim 1, wherein, The rolling time domain optimization through nonlinear model predictive control specifically comprises: an embedded code generation tool is used to convert an optimization problem containing system dynamics and actuator constraints into a code capable of running on the central processing unit, and the code is solved at each control period.

5. The intelligent adaptive transmission system of claim 1, wherein, The health state prediction based on the hidden Markov model specifically comprises: feature extraction and discretization are performed on sensor signals to form an observation sequence, a most possible system health hidden state sequence is calculated through a Viterbi decoding algorithm, and remaining service life is predicted based on a state transition matrix.

6. The intelligent adaptive transmission system of claim 1, wherein, ​ 7. The intelligent adaptive transmission system of claim 1, wherein, The updating control strategy based on the hierarchical reinforcement learning framework specifically comprises: collecting experience data in system operation and storing in a buffer; when a safety condition is met, starting an asynchronous task to sample data from the buffer, and performing offline training on the high-level policy network and the low-level execution network; after verification, deploying the updated network parameters to the online control system.

8. A vehicle characterized by comprising: An intelligent adaptive gearbox system comprising the gearbox system of any one of claims 1 to 7.

9. A gearbox control method based on a multi-modal fusion control theory, characterized in that, The method comprises the following steps: Step S1: collecting gearbox operation data through the distributed sensor array according to claim 1; Step S2: performing multi-modal perception and fusion processing on the collected data by the central processing unit, specifically including wavelet decomposition of vibration signals and dynamic balance control, multi-physical field efficiency evaluation based on pre-stored efficiency map, and working condition mode recognition based on support vector machine; Step S3: performing intelligent collaborative decision-making based on the perception results by the central processing unit, specifically including rolling optimization based on nonlinear model predictive control, global gear path planning based on graph theory A* search, and multi-actuator action coordination based on time sequence planning; Step S4: executing the control instructions generated by step S3 optimization, and collecting system feedback data after execution; Step S5: performing adaptive optimization by the central processing unit, specifically including health state prediction and life estimation based on hidden Markov model, and hierarchical reinforcement learning strategy updating based on experience replay and offline asynchronous training; Step S6: cyclically executing steps S1 to S5 to realize closed-loop control and continuous optimization.

Images