A fluid valve control system for automotive thermal management

Abstract

The application discloses a fluid valve control system suitable for automobile thermal management, relates to the technical field of phase change thermal management, and aims to solve the technical problem that the cooling liquid phase change process cannot be recognized and actively and accurately controlled in real time in the field of automobile thermal management, and comprises a multi-physical field sensing module configured to acquire fusion sensing data reflecting the thermal state of an engine cylinder head target region and the physical state of the cooling liquid; a phase change dynamics identification module connected to the multi-physical field sensing module and configured to identify the phase change stage of the cooling liquid in the target region in real time based on the fusion sensing data through a time series deep learning model; and an intelligent control decision module connected to the phase change dynamics identification module and configured to acquire engine working condition parameters and pre-stored phase change response characteristics of the cooling liquid, and to determine an intelligent control decision based on the phase change stage, the working condition parameters and the phase change response characteristics. The application has the advantage that the phase change heat transfer can be actively optimized through intelligent sensing and collaborative intervention.

Description

Technical Field

[0001] This invention relates to the field of phase change thermal management technology, and more specifically, to a fluid valve control system applicable to automotive thermal management. Background Technology

[0002] The automotive thermal management system has a critical impact on engine efficiency, emissions, and reliability. As internal combustion engines evolve towards higher efficiency and lower emissions, their heat load is increasing and becoming more unevenly distributed. Especially under transient high-load conditions, localized areas are prone to excessively high heat flux density and sudden increases in coolant temperature. In such situations, if the coolant undergoes a phase change (such as boiling), its latent heat absorption capacity can significantly enhance heat exchange efficiency. However, if the phase change process is not properly controlled, it can easily lead to unstable flow, localized overheating, or bubble accumulation, thereby reducing cooling effectiveness or even causing thermal damage.

[0003] Traditional cooling systems mostly employ mechanical or simple electronically controlled thermostatic valves to achieve macroscopic temperature control by adjusting coolant flow. However, they lack the ability to perceive and actively intervene in the microscopic phase change state of the coolant in real time. While some existing technologies have attempted to introduce temperature-feedback-based valve control or indirectly determine the gas-liquid state through pressure sensors, they still struggle to accurately and in real-time identify the phase change stage of the coolant (such as nucleation, explosion, and two-phase transport). Furthermore, they cannot actively stimulate or modulate the phase change process, resulting in sluggish response and coarse control when dealing with transient high heat flux, making it difficult to balance heat dissipation efficiency and system stability. Therefore, we propose a fluid valve control system suitable for automotive thermal management. Summary of the Invention

[0004] The purpose of this invention is to provide a fluid valve control system applicable to automotive thermal management, so as to solve the technical problem in the field of automotive thermal management that it is impossible to identify and actively and accurately control the phase change process of coolant in real time.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a fluid valve control system applicable to automotive thermal management, comprising:

[0006] A multi-physics sensing module is configured to acquire fused sensing data reflecting the thermal state and coolant physical state of the target area of ​​the engine cylinder head;

[0007] The phase change dynamics identification module is connected to the multiphysics sensing module. It is configured to identify the phase change stage of the coolant in the target area in real time based on fused sensing data and a time-series deep learning model.

[0008] The intelligent control decision module is connected to the phase change dynamics identification module and is configured to: acquire engine operating parameters and pre-stored coolant phase change response characteristics; based on the phase change stage, operating parameters and phase change response characteristics, generate a set of control instructions for exciting or modulating the target phase change process through rolling optimization of the model predictive control framework; the set of control instructions includes fluid valve control instructions and waveform excitation instructions.

[0009] An actuator array module, connected to the intelligent control decision module and the engine cooling circuit, is configured to receive and execute the control command set. The actuator array module includes a controllable fluid valve driven by a fluid valve control command and a wideband programmable waveform exciter driven by a waveform excitation command. The controllable fluid valve changes the flow rate and flow path of the coolant flowing through the target area by adjusting its opening degree. The waveform exciter triggers or modulates the phase change of the coolant by applying an external physical field excitation of a specific time-frequency domain waveform. The two work together to achieve the target phase change process.

[0010] Preferably, the multiphysics sensing module includes:

[0011] A high-frequency ultrasonic sensing unit is configured to emit ultrasonic waves into the coolant flowing through the target area and receive the echoes in order to obtain ultrasonic characteristic signals that reflect the microstructure state and phase change process in the coolant.

[0012] A distributed temperature and flow field sensing unit is configured to acquire information on the local temperature distribution and flow velocity distribution of the coolant in the adjacent flow channel of the target area.

[0013] A vapor volume fraction sensing unit is configured to non-invasively measure the vapor volume fraction in a gas-liquid two-phase mixture within a flow channel.

[0014] Preferably, in the phase transition dynamics identification module, the temporal deep learning model models and identifies the temporal correlation of phase transition features through the following state update equation:

[0015] ;

[0016] ;

[0017] ;

[0018] ;

[0019] ;

[0020] ;

[0021] in, To indicate The input feature vector at time step; for The input gate activation vector at time t; for The forget gate activation vector at time step; for The output gate activation vector at time 1; for The candidate cell state vector at time t; for The cell state vector at time t; for The cell state vector at time t; for The hidden state vector at time step; , , , These are the input feature vectors. The weight matrix for the input gate, forget gate, output gate, and candidate cell states; , , , The hidden state of the previous moment The weight matrix for the input gate, forget gate, output gate, and candidate cell states; , , , These are the bias vectors for the input gate, forget gate, output gate, and candidate cell state, respectively. It is the sigmoid activation function.

[0022] Preferably, the phase change response characteristics pre-stored in the intelligent control decision module are digital maps of the tendency, intensity, and latent heat release characteristics of the coolant to undergo phase change under different physical excitation conditions.

[0023] The digitized map is mathematically represented by the following functional relationship, and a normalized response vector is output:

[0024] ;

[0025] in, This is a set of physical property parameters for the coolant; This is the current system operating point; The external excitation condition vector; This is the mapping function for the phase transition response characteristic spectrum; The normalized response vector; For the normalized nucleation tendency component; The normalized latent heat release intensity component; This is the normalized bubble growth dynamic component.

[0026] Preferably, the time-domain excitation signal corresponding to the waveform excitation command is synthesized using the following parameterized formula:

[0027] ;

[0028] in, In time The instantaneous value of the excitation signal at time 1; In time The amplitude envelope function value at time t; In time The instantaneous frequency function value at time t; In time The additional phase modulation function value at time; It is the integral variable.

[0029] Preferably, the controllable fluid valve is a variable geometry electronic thermostatic valve;

[0030] The intelligent control decision module achieves coordinated optimization of waveform excitation commands and fluid valve control commands within the model predictive control framework by minimizing the following objective function:

[0031] ;

[0032] The constraints are:

[0033] ;

[0034] ;

[0035] ;

[0036] in, The value of the objective function; In time Control input vector for waveform exciter; In time Control inputs for controllable fluid valves; To predict the length of the time domain; For the predicted future moment The normalized actual response vector predicted by the internal model; For the predicted future moment The expected normalized response vector; The weight matrix for responding to tracking errors; This refers to the penalty weighting coefficient for the control input size of the waveform exciter; This refers to the penalty weighting coefficient for changes in valve control input; The input is a normalized waveform excitation control input; For normalized valve control input; Let be the norm of the vector; An internal state vector describing the dynamics of the system; The dynamic equations describing the evolution of the system state; For measurable external disturbances; , These are the lower and upper limits of the allowable input for the waveform exciter control; , These are the lower and upper limits of the allowable input for valve control.

[0037] Preferably, it also includes a fault-tolerant control module, which is configured to monitor the signal health status of each sensing unit in the multi-physics sensing module, and when it is determined that any sensing unit has failed, reconstruct the missing normalized sensing information based on the data of the remaining healthy sensing units through a state observer based on a Kalman filter.

[0038] The state prediction and update equations for the state observer are as follows:

[0039] Prediction steps:

[0040] ;

[0041] ;

[0042] Update steps:

[0043] ;

[0044] ;

[0045] ;

[0046] in, In order to obtain Before the time observation, for The prior estimate vector of the system state at time t; From Time's up The state transition matrix at time t; for The posterior estimate vector of the system state at time t; for Control input matrix at any given time; for The normalized control input vector applied at each time step; The prior estimate is the error covariance matrix; for The posterior estimation error covariance matrix at time t; State transition matrix The transpose of the matrix; for The process noise covariance matrix at time step; for The posterior estimation error covariance matrix at time t; for Kalman gain matrix at time step; for The observation matrix at each time point; Observation matrix The transpose of the matrix; for The measurement noise covariance matrix at time point; After fusing prediction and observation information, the obtained The posterior estimate vector of the system state at time 1; for The normalized observation vector actually obtained from the remaining health sensors at any given moment; To and An identity matrix with consistent dimensions.

[0047] Preferably, the sensing signals and engine operating parameters of the multi-physics sensing module are transmitted to the phase change dynamics identification module and the intelligent control decision module via the vehicle CAN bus.

[0048] Preferably, it also includes a self-learning optimization module, which is connected to the multi-physics sensing module and the intelligent control decision module. Its configuration is to update the internal model parameters and phase change response characteristics of the intelligent control decision module online based on the actual feedback of the control effect from the fused sensing data.

[0049] Preferably, the self-learning optimization module iteratively updates the phase transition response characteristic spectrum using the following parameter learning formula:

[0050] ;

[0051] in, In the first In the next iteration, the graph mapping function The set of all adjustable parameters internally; The learning rate; For loss function For parameter set The gradient operator; The loss function; It is the measured normalized response vector calculated or estimated based on data from the multiphysics sensing module.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] 1. This invention utilizes a multi-physics sensing module to collect fused data reflecting the microstructure and state of the coolant in real time, and uses a temporal deep learning model to dynamically identify its phase transition stage. This overcomes the limitations of traditional methods that rely solely on single parameters such as temperature or pressure and cannot capture the dynamic temporal sequence of phase transitions. Based on this, the intelligent control decision module, using the phase transition stage, operating parameters, and pre-stored phase transition response characteristics, performs rolling optimization through a model predictive control framework, simultaneously generating fluid valve control commands and waveform excitation commands. Controllable valves in the actuator array regulate flow rate and flow path, while broadband exciters apply external physical field excitations of specific waveforms. Together, they can actively stimulate or modulate the phase transition process in the target region. This design, for the first time, combines real-time identification of the phase transition state with the collaborative intervention of multiple actuators, achieving a leap from "passively responding to macroscopic temperature" to "actively regulating microscopic phase transitions." This significantly improves the system's heat dissipation accuracy and response speed under transient high heat flux conditions, effectively solving the problem of precise control of the phase transition process.

[0054] 2. This invention further introduces a self-learning optimization module, which can update the internal model parameters and phase change response characteristic spectrum in the intelligent control decision module online based on feedback from actual control effects and multi-physics sensing data. By continuously comparing the difference between the predicted response and the actual measurement, and using the gradient descent algorithm to dynamically correct the model, the system can gradually adapt to different engine operating conditions, coolant properties, and performance changes during long-term use. This capability allows the control system to not only rely on the prior knowledge of initial calibration, but also continuously accumulate experience and self-improve in actual operation, thereby maintaining the accuracy and robustness of phase change control over the long term. This design further solves the problem that performance may degrade with operating condition drift when relying on a fixed model, realizing continuous optimization and personalized adaptation of the system.

[0055] 3. This invention also includes a fault-tolerant control module, which continuously monitors the signal health status of each sensing unit. Upon detecting a sensor failure, it reconstructs the missing sensing information based on data from the remaining healthy sensors using a state observer and automatically switches to a degraded control strategy. This mechanism ensures that even with the failure of some sensing units, the system can still maintain basic phase change identification and control functions based on the reconstructed information, preventing system-wide failure due to a single sensor malfunction. This further enhances the reliability and practicality of this invention in complex real-world vehicle environments, resolves the safety hazard of the system potentially losing its phase change regulation capability entirely when sensors malfunction, and ensures the continuous and stable operation of the thermal management function. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the overall system architecture of the present invention.

[0057] Figure 2 This is a schematic diagram of the architecture of the multiphysics sensing module of the present invention.

[0058] Figure 3 This is a schematic diagram of the core control process of the present invention.

[0059] Figure 4 This is a schematic diagram of the self-learning optimization architecture of the present invention.

[0060] Figure 5 This is a schematic diagram of the fault-tolerant control architecture of the present invention.

[0061] Figure 6 This is a schematic diagram of the data flow of the complete system of the present invention. Detailed Implementation

[0062] To facilitate understanding of the technical solution of the present invention by those skilled in the art, the technical solution of the present invention will now be further described in conjunction with the accompanying drawings.

[0063] Example 1, such as Figures 1-6 As shown, the present invention provides a fluid valve control system applicable to automotive thermal management, comprising:

[0064] A multi-physics sensing module is configured to acquire fused sensing data reflecting the thermal state and coolant physical state of the target area of ​​the engine cylinder head;

[0065] The phase change dynamics identification module is connected to the multiphysics sensing module and is configured to identify the phase change stage of the coolant in the target area in real time based on fused sensing data.

[0066] The intelligent control decision module is connected to the phase change dynamics identification module. Its configuration is as follows: to acquire engine operating parameters and pre-stored coolant phase change response characteristics; based on the phase change stage, operating parameters and phase change response characteristics, to generate a set of control instructions for stimulating or modulating the target phase change process through rolling optimization of the model predictive control framework; the set of control instructions includes fluid valve control instructions.

[0067] The actuator array module is connected to the intelligent control decision module and the engine cooling circuit. It is configured to receive and execute a set of control commands. The actuator array module includes controllable fluid valves driven by fluid valve control commands. By adjusting the opening of the controllable fluid valves, the flow rate and flow path of the coolant flowing through the target area are changed to achieve the target phase change process in a coordinated manner.

[0068] The self-learning optimization module is connected to the multi-physics sensing module and the intelligent control decision module. It is configured to update the internal model parameters and phase change response characteristics of the intelligent control decision module online based on the actual feedback of the control effect from the fused sensing data.

[0069] In an embodiment of the present invention, the multiphysics sensing module includes:

[0070] A high-frequency ultrasonic sensing unit is configured to emit ultrasonic waves into the coolant flowing through the target area and receive the echoes in order to obtain ultrasonic characteristic signals that reflect the microstructure state and phase change process in the coolant.

[0071] A distributed temperature and flow field sensing unit is configured to acquire information on the local temperature distribution and flow velocity distribution of the coolant in the adjacent flow channel of the target area.

[0072] A vapor volume fraction sensing unit configured to non-invasively measure the vapor volume fraction in a gas-liquid two-phase mixture within a flow channel.

[0073] More specifically, the high-frequency ultrasonic sensing unit employs a piezoelectric transducer with a frequency of 1-10 MHz, installed at a specific location on the cylinder head water jacket wall, using an oblique incidence method to cover the high heat flux density region with the sound beam. The distributed temperature and flow field sensing unit can be implemented using a combination of a miniature thermocouple array embedded in the flow channel wall and an ultrasonic Doppler velocimetry probe. The steam volume fraction sensing unit can employ a sensor based on the microwave resonance principle, inverting the steam fraction by measuring the change in the dielectric constant of the medium within the flow channel.

[0074] In an embodiment of the present invention, the phase transition dynamics identification module is specifically configured as follows:

[0075] Based on ultrasonic characteristic signals, temperature and flow velocity distribution information, and steam volume fraction, the phase change process of the coolant is identified in real time as the nucleation incubation period, the phase change outbreak period, and the two-phase flow migration period through a time-series deep learning model.

[0076] Temporal deep learning models model and identify the temporal correlation of phase transition features through the following state update equation:

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] ;

[0082] ;

[0083] in, To indicate The input feature vector at time t is composed of all normalized fused sensor data (such as ultrasonic features, temperature, flow rate, and steam fraction) at the current time, and its dimension corresponds to the number of sensor features.

[0084] for The input gate activation vector at time step 1, with each element having a value between 0 and 1, determines how much new information in the corresponding dimension is allowed to be stored in the cell state.

[0085] for The forget gate activation vector at time step, with each element having a value between 0 and 1, determines how much of the cell state information from the previous time step is retained in the corresponding dimension;

[0086] for The output gate activation vector at time step 1, where each element has a value between 0 and 1, determines how much of the current cell state information in the corresponding dimension is output to the hidden state.

[0087] for The candidate cell state vector at time step 1 is obtained by combining the current input and the hidden state from the previous time step. The activation function is generated and contains new information that may be stored in the cell state.

[0088] for The cell state vector at time step is the core memory unit of the network, which is obtained by adding the cell state of the previous time step controlled by the forget gate and the current candidate state controlled by the input gate.

[0089] for The cell state vector at time step 1 is the historical cell state passed from the previous time step to the current time step, carrying... The chronological characteristics of memory prior to a given moment are the target of the forgetting gate;

[0090] for The hidden state vector at time step is the information output at the current time step, which is output after the current cell state is regulated by the output gate;

[0091] , , , These are the input feature vectors. The weight matrix for the input gate, forget gate, output gate, and candidate cell states;

[0092] , , , The hidden state of the previous moment The weight matrix for the input gate, forget gate, output gate, and candidate cell states;

[0093] , , , These are the bias vectors for the input gate, forget gate, output gate, and candidate cell state, respectively.

[0094] The sigmoid activation function maps the input to the (0,1) interval, simulating the "opening and closing" of the gate;

[0095] This is an element-wise multiplication operation for vectors;

[0096] The temporal deep learning model is a long short-term memory network. Input feature vector. The construction method is as follows: The echo signal attenuation coefficient acquired by the high-frequency ultrasonic sensing unit, the temperature and flow velocity at at least three points acquired by the distributed temperature and flow field sensing unit, and the value acquired by the steam volume fraction sensing unit are all subtracted from their calibrated normal operating condition mean values ​​and divided by the standard deviation. After standardization, these are then assembled in a fixed order. This network is trained using historical sensor data sequences containing multiple sets of data under different engine loads and cooling conditions. The data sequences have been labeled with corresponding phase transition stage tags using expert knowledge or synchronous high-speed camera data.

[0097] This formula describes the single-timestep forward propagation process of a Long Short-Term Memory (LSTM) network. The logic is that information in the cell state is selectively updated and output through three control structures: the input gate, the forget gate, and the output gate. The input gate determines how much of the current input information needs to be saved to the cell state, the forget gate determines how much of the cell state from the previous time step needs to be retained or forgotten, and the output gate determines how much information needs to be output to the hidden state based on the current cell state. The cell state, as the network's "memory," is passed along the timeline, thereby enabling the capture and modeling of the temporal dynamics of the phase transition process.

[0098] In an embodiment of the present invention, the phase change response characteristics pre-stored in the intelligent control decision module are digital maps of the tendency, intensity and latent heat release characteristics of the coolant to undergo phase change under different physical excitation conditions.

[0099] The digital map is mathematically represented by the following functional relationship, and a normalized response vector is output:

[0100] ;

[0101] in, It is a set of physical property parameters of the coolant, which is a vector or parameter set containing various parameters describing microstructure characteristics (such as size, concentration, shell properties) and basic hydrothermal properties;

[0102] The current system operating point is a vector that typically contains at least two key state variables: system pressure and mainstream coolant temperature.

[0103] This is the external excitation condition vector, whose elements define the physical field characteristics applied by the actuator array, such as the amplitude and frequency range of the waveform excitation;

[0104] It is a mapping function for the phase transition response characteristic spectrum, and its specific form can be a neural network, a multinomial regression model or a lookup table, which is obtained through calibration and training of a large amount of experimental data;

[0105] The normalized response vector is a mapping function. The output;

[0106] The normalized nucleation tendency component is a dimensionless scalar; the larger the value, the stronger the tendency of the coolant to undergo phase change nucleation under the current conditions.

[0107] The normalized latent heat release intensity component is a dimensionless scalar. The larger the value, the greater the potential rate at which a unit volume of coolant absorbs or releases latent heat during the phase change process.

[0108] The normalized bubble growth dynamic component is a dimensionless scalar; the larger the value, the more intense the bubble growth or movement after the phase transition.

[0109] The transpose symbol for a vector;

[0110] This formula defines a static mapping relationship from input conditions to the coolant's phase change behavior response. The logic is that, taking the inherent physical properties of the coolant, the system's current operating point, and externally applied excitation conditions as inputs, a pre-established nonlinear mapping function directly predicts the standardized phase change behavior of the coolant under these combined conditions. This graph abstracts the complex physicochemical process into a computable data model, serving as a priori knowledge base for the control system.

[0111] In embodiments of the present invention, the control instruction set generated by the intelligent control decision module includes waveform excitation instructions;

[0112] The actuator array module includes a wideband programmable waveform exciter configured to receive and execute waveform excitation commands to apply an external physical field excitation of a specific time-frequency domain waveform to the coolant in the target area.

[0113] The time-domain excitation signal corresponding to the waveform excitation command is synthesized using the following parameterized formula:

[0114] ;

[0115] in, In time The instantaneous value of the excitation signal at a given moment can be in the physical form of voltage, pressure, or displacement.

[0116] In time The amplitude envelope function value at a given time determines the strength or energy level of the signal at different times;

[0117] In time The instantaneous frequency function value at a given moment, measured in Hertz, determines how the signal frequency changes over time.

[0118] In time The additional phase modulation function value at each moment, in radians, is used to further adjust the phase characteristics of the waveform;

[0119] Let be the integral variable, representing the time from the initial time 0 to the current time. The time between;

[0120] More specifically, the broadband programmable waveform exciter is a piezoelectric ceramic stacked actuator, which is integrated into the water jacket wall of the engine cylinder head via threaded fastening or adhesive bonding, directly facing the high heat flux density region. Excitation signal The driving voltage generates a physical field excitation of high-frequency mechanical vibration perpendicular to the wall, which is transmitted to the adjacent coolant through the wall. The exciter is driven by a dedicated power amplifier that receives digital waveform parameters from the intelligent control decision module and converts them into an analog voltage signal. ;

[0121] This formula describes how to synthesize a modulated waveform whose frequency and amplitude both vary with time. The logic is as follows: by defining the amplitude envelope function, instantaneous frequency function, and phase modulation function respectively, the instantaneous frequency is converted into a phase increment using integration, ultimately synthesizing an excitation signal that varies continuously in the time domain. This signal can be converted into mechanical vibration or pressure waves acting on the coolant through actuators such as piezoelectric actuators.

[0122] In an embodiment of the present invention, the controllable fluid valve is a variable geometry electronic thermostatic valve;

[0123] The intelligent control decision module is configured to generate mutually coordinated waveform excitation commands and fluid valve control commands based on the identified phase transition stage.

[0124] Synergy is achieved by minimizing the following objective function with uniform dimensions within the model predictive control framework:

[0125] ;

[0126] The constraints are:

[0127] ;

[0128] ;

[0129] ;

[0130] in, The value of the objective function is the scalar performance metric that needs to be minimized.

[0131] In time For the control input vector of the waveform exciter, its elements are the modulation function. , , The parameter set;

[0132] In time For controllable fluid valves, the control input is usually a target opening command;

[0133] The length of the predicted time domain, i.e., the range of future time that the controller predicts forward;

[0134] For the predicted future moment The normalized actual response vector predicted by the internal model;

[0135] For the predicted future moment The expected normalized response vector is determined by higher-level thermal management objectives;

[0136] The weight matrix for responding to tracking errors is typically a positive definite diagonal matrix, and the size of its diagonal elements determines the corresponding response components. , , The relative importance of tracking accuracy;

[0137] This is a penalty weighting coefficient applied to the magnitude of the control input to the waveform exciter, used to limit the consumption of control energy or the aggressiveness of execution;

[0138] This is a penalty weighting coefficient for changes in valve control input, used to smooth valve action and reduce wear and hydraulic shock;

[0139] The input is a normalized waveform excitation control input;

[0140] For normalized valve control input;

[0141] The norm of a vector is usually the 2-norm (Euclidean norm), and its square is the sum of the squares of the vector's elements.

[0142] An internal state vector describing the dynamics of the system;

[0143] The dynamic equations (state-space model) that describe the state evolution of the system.

[0144] For measurable external disturbances;

[0145] , These are the lower and upper limits of the allowable input for the waveform exciter control;

[0146] , The lower and upper limits of the allowable input for valve control;

[0147] Specifically, the internal model in the model predictive control framework A one-dimensional lumped parameter model is employed. The target cooling channel and adjacent metal are divided into five continuous control volumes. Based on the energy conservation equation, a state equation regarding temperature and bubble quantity is established for each control volume. The convective heat transfer coefficient and flow resistance coefficient in the model are obtained through preheating calibration on an engine test bench. The normalized control input... and The calculation method is as follows:

[0148] ;

[0149] ;

[0150] in, This is the maximum drive voltage of the exciter. and These are the minimum and maximum limits for valve opening.

[0151] This formula describes the rolling optimization objective function of a model predictive controller. The logic is that, in each control cycle, the controller seeks a sequence of optimal control inputs (waveform parameters and valve opening) within a finite future prediction time domain, such that the predicted system response trajectory is as close as possible to the desired trajectory, while the control action itself is not too drastic. This is a constrained multi-objective optimization problem; solving this problem yields the optimal control command to be executed at the current moment.

[0152] In an embodiment of the present invention, the self-learning optimization module is specifically configured as follows:

[0153] By comparing the predicted values ​​of the system state by the model predictive control framework with the actual feedback values ​​of the multiphysics sensing module, the heat transfer model parameters and flow resistance model parameters involved in the intelligent control decision module are dynamically corrected using an online parameter estimation algorithm.

[0154] The iterative update of the phase transition response characteristic spectrum is achieved through the following parameter learning formula based on the normalized response:

[0155] ;

[0156] in, In the first At the next iteration (or learning cycle), the graph mapping function The set of all adjustable parameters internally;

[0157] The learning rate is a positive scalar that controls the step size of each parameter update.

[0158] For loss function For parameter set The gradient operator, the result of which is... Vectors of the same dimension indicate the direction in which each parameter should be adjusted to minimize the loss as quickly as possible;

[0159] The loss function is a scalar function used to quantize the measured normalized response vector. Compared with the current predicted value of the map The difference between them, commonly in the form of mean squared error;

[0160] This refers to the measured normalized response vector calculated or estimated based on data from the multi-physics sensing module.

[0161] This formula describes the online parameter learning rule based on the gradient descent principle. The logic is that during system operation, the difference (loss function) between the graph model's predicted response and the actual sensor measurement is continuously compared. By calculating the gradient of this difference with respect to the model parameters, and adjusting the parameters in the opposite direction of the gradient with a certain step size (learning rate), the model's predictions continuously approach reality, achieving model adaptation and refinement.

[0162] In embodiments of the present invention, a fault-tolerant control module is further included, which is configured as follows:

[0163] Monitor the signal health status of each sensing unit in the multiphysics sensing module;

[0164] When any sensor unit is determined to be faulty, the missing normalized sensing information is reconstructed based on the data from the remaining healthy sensor units using a predefined redundancy estimation algorithm, and then the corresponding degraded control strategy is switched.

[0165] The redundancy estimation algorithm employs a state observer based on a Kalman filter. Its system state vector is composed of normalized physical quantities, and the state prediction and update equations are as follows:

[0166] Prediction steps:

[0167] ;

[0168] ;

[0169] Update steps:

[0170] ;

[0171] ;

[0172] ;

[0173] in, In order to obtain Before the time observation, for The prior estimate vector (predicted value) of the system state at time t.

[0174] From Time's up The state transition matrix at time t describes how the system state evolves on its own;

[0175] for The posterior estimate vector of the system state at time t (the optimal estimate of the previous time step).

[0176] for The control input matrix at each time step describes the control input. How to influence state changes;

[0177] for The normalized control input vector applied at each time step;

[0178] The prior state estimate is quantified by the prior estimation error covariance matrix. Uncertainty;

[0179] for The posterior estimation error covariance matrix at time t;

[0180] State transition matrix The transpose of a matrix is ​​used to achieve dimension matching in matrix multiplication, which is a standard operation in linear algebra.

[0181] for The process noise covariance matrix at time step 1 models the uncertainties not considered in the state transition model.

[0182] for The posterior estimation error covariance matrix at time 1 is quantified. Time-optimal state estimation The uncertainty is one of the core outputs of the Kalman filter update step;

[0183] for The Kalman gain matrix at time step 1 determines the weights that the observations should be assigned in the update step.

[0184] for The observation matrix at time t describes the system state. The relationship with expected observations;

[0185] Observation matrix The transpose of the matrix is ​​used for matrix dimension matching in solving the Kalman gain;

[0186] for The measurement noise covariance matrix at time points models the noise level in sensor observations;

[0187] After fusing prediction and observation information, the obtained The posterior estimate vector of the system state at time 1 (optimal estimate);

[0188] for The normalized observation vector actually obtained from the remaining health sensors at any given moment;

[0189] To and A dimensionally consistent identity matrix;

[0190] Preferably, the method for monitoring the health status of the signals includes: determining whether the signal value of any sensor continuously exceeds its preset physical extreme value range for more than 100 milliseconds; or calculating the saturated vapor pressure based on the current readings of the pressure sensor and temperature sensor, and if the calculated saturation state is seriously contradictory to the reading of the vapor volume fraction sensor and continues for more than 200 milliseconds, then the health status of the vapor volume fraction sensor is determined to be abnormal. When a sensor is determined to be faulty, the observation matrix... The elements of the corresponding row in the matrix are set to zero, and the measurement noise covariance matrix is... The corresponding diagonal element is set to a very large value so that the information of the faulty sensor is ignored in the state update step;

[0191] This formula describes the two core steps of the standard Kalman filter algorithm: prediction and update. The prediction step uses the system model and, based on the state estimate and control input from the previous time step, predicts the current state and uncertainty. The update step, after obtaining new actual observation data, calculates the Kalman gain to balance the model prediction and the observed values, thus fusing the information from both to obtain the optimal current state estimate and updating the uncertainty of the estimate.

[0192] In an embodiment of the present invention, the system is integrated into the vehicle engine controller, and the sensing signals of the multi-physics sensing module and the engine operating parameters are transmitted to the phase change dynamics identification module and the intelligent control decision module through the vehicle CAN bus.

[0193] Example 2: A fluid valve control method applicable to automotive thermal management, comprising the following steps:

[0194] S1. Real-time acquisition of fused sensor data and engine operating parameters;

[0195] S2. Based on fused sensor data, identify the phase change stage of the coolant in the target area of ​​the engine cylinder head in real time;

[0196] S3. Based on the phase change stage, operating parameters and pre-stored coolant phase change response characteristics, the model predictive control is used for rolling optimization to generate a set of control instructions for stimulating or modulating the target phase change process. The set of control instructions includes fluid valve control instructions.

[0197] S4. Execute the control command set to coordinate the achievement of the target phase change process by adjusting the opening of the controllable fluid valve;

[0198] S5. Based on the feedback from the sensor data obtained after control, update the internal model parameters and phase change response characteristics online.

[0199] The embodiments disclosed in this invention are preferred embodiments, but are not limited thereto. Those skilled in the art can easily understand the spirit of this invention based on the above embodiments and make different extensions and variations, but as long as they do not depart from the spirit of this invention, they are all within the protection scope of this invention.

Claims

1. A fluid valve control system suitable for automotive thermal management, characterized in that, include: A multi-physics sensing module is configured to acquire fused sensing data reflecting the thermal state and coolant physical state of the target area of ​​the engine cylinder head; The phase change dynamics identification module is connected to the multiphysics sensing module. It is configured to identify the phase change stage of the coolant in the target area in real time based on fused sensing data and a time-series deep learning model. The intelligent control decision module is connected to the phase change dynamics identification module, and its configuration is as follows: to acquire engine operating parameters and pre-stored coolant phase change response characteristics; Based on the phase transition stage, operating parameters, and phase transition response characteristics, a set of control instructions for exciting or modulating the target phase transition process is generated through rolling optimization of the model predictive control framework; the set of control instructions includes fluid valve control instructions and waveform excitation instructions. An actuator array module, connected to the intelligent control decision module and the engine cooling circuit, is configured to receive and execute the control command set. The actuator array module includes a controllable fluid valve driven by a fluid valve control command and a wideband programmable waveform exciter driven by a waveform excitation command. The controllable fluid valve changes the coolant flow rate and flow path through the target area by adjusting its opening degree, and the waveform exciter triggers or modulates the phase change of the coolant by applying an external physical field excitation of a specific time-frequency domain waveform. The two work together to achieve the target phase change process. It also includes a fault-tolerant control module, which is configured to monitor the signal health status of each sensing unit in the multi-physics sensing module, and when it is determined that any sensing unit has failed, it reconstructs the missing normalized sensing information based on the data of the remaining healthy sensing units through a state observer based on a Kalman filter. The state prediction and update equations for the state observer are as follows: Prediction steps: ； ； Update steps: ； ； ； in, In order to obtain Before the time observation, for The prior estimate vector of the system state at time t; From Time's up The state transition matrix at time t; for The posterior estimate vector of the system state at time t; for Control input matrix at any given time; for The normalized control input vector applied at each time step; The prior estimate is the error covariance matrix; for The posterior estimation error covariance matrix at time t; State transition matrix The transpose of the matrix; for The process noise covariance matrix at time step; for The posterior estimation error covariance matrix at time t; for Kalman gain matrix at time step; for The observation matrix at each time point; Observation matrix The transpose of the matrix; for The measurement noise covariance matrix at time point; After fusing prediction and observation information, the obtained The posterior estimate vector of the system state at time 1; for The normalized observation vector actually obtained from the remaining health sensors at any given moment; To and An identity matrix with consistent dimensions.

2. A fluid valve control system for automotive thermal management according to claim 1, characterized in that, The multiphysics sensing module includes: A high-frequency ultrasonic sensing unit is configured to emit ultrasonic waves into the coolant flowing through the target area and receive the echoes in order to obtain ultrasonic characteristic signals that reflect the microstructure state and phase change process in the coolant. A distributed temperature and flow field sensing unit is configured to acquire information on the local temperature distribution and flow velocity distribution of the coolant in the adjacent flow channel of the target area. A vapor volume fraction sensing unit is configured to non-invasively measure the vapor volume fraction in a gas-liquid two-phase mixture within a flow channel.

3. A fluid valve control system for automotive thermal management according to claim 2, characterized in that, In the phase transition dynamics identification module, the temporal deep learning model models and identifies the temporal correlation of phase transition features through the following state update equation: ； ； ； ； ； ； in, To indicate The input feature vector at time step; for The input gate activation vector at time t; for The forget gate activation vector at time step; for The output gate activation vector at time 1; for The candidate cell state vector at time t; for The cell state vector at time t; for The cell state vector at time t; for The hidden state vector at time step; , , , These are the input feature vectors. The weight matrix for the input gate, forget gate, output gate, and candidate cell states; , , , The hidden state of the previous moment The weight matrix for the input gate, forget gate, output gate, and candidate cell states; , , , These are the bias vectors for the input gate, forget gate, output gate, and candidate cell state, respectively. It is the sigmoid activation function.

4. A fluid valve control system for automotive thermal management according to claim 1, characterized in that, The phase change response characteristics pre-stored in the intelligent control decision module are digital maps of the tendency, intensity, and latent heat release characteristics of the coolant to undergo phase changes under different physical excitation conditions. The digitized map is mathematically represented by the following functional relationship, and a normalized response vector is output: ； in, This is a set of physical property parameters for the coolant; This is the current system operating point; The external excitation condition vector; This is the mapping function for the phase transition response characteristic spectrum; The normalized response vector; For the normalized nucleation tendency component; The normalized latent heat release intensity component; This is the normalized bubble growth dynamic component.

5. A fluid valve control system for automotive thermal management according to claim 4, characterized in that, The time-domain excitation signal corresponding to the waveform excitation command is synthesized using the following parameterized formula: ； in, In time The instantaneous value of the excitation signal at time 1; In time The amplitude envelope function value at time t; In time The instantaneous frequency function value at time t; In time The additional phase modulation function value at time; It is the integral variable.

6. A fluid valve control system for automotive thermal management according to claim 5, characterized in that, The controllable fluid valve is a variable geometry electronic thermostatic valve; The intelligent control decision module achieves coordinated optimization of waveform excitation commands and fluid valve control commands within the model predictive control framework by minimizing the following objective function: ； The constraints are: ； ； ； in, The value of the objective function; In time Control input vector for waveform exciter; In time Control inputs for controllable fluid valves; To predict the length of the time domain; For the predicted future moment The normalized actual response vector predicted by the internal model; For the predicted future moment The expected normalized response vector; The weight matrix for responding to tracking errors; This refers to the penalty weighting coefficient for the control input size of the waveform exciter; This refers to the penalty weighting coefficient for changes in valve control input; The input is a normalized waveform excitation control input; For normalized valve control input; Let be the norm of the vector; An internal state vector describing the dynamics of the system; The dynamic equations describing the evolution of the system state; For measurable external disturbances; , These are the lower and upper limits of the allowable input for the waveform exciter control; , These are the lower and upper limits of the allowable input for valve control.

7. A fluid valve control system for automotive thermal management according to claim 1, characterized in that, The sensing signals and engine operating parameters of the multi-physics sensing module are transmitted to the phase change dynamics identification module and the intelligent control decision module via the vehicle CAN bus.

8. A fluid valve control system for automotive thermal management according to claim 4, characterized in that, It also includes a self-learning optimization module, which is connected to the multi-physics sensing module and the intelligent control decision module. Its configuration is to update the internal model parameters and phase change response characteristics of the intelligent control decision module online based on the actual feedback of the control effect from the fused sensing data.

9. A fluid valve control system for automotive thermal management according to claim 8, characterized in that, The self-learning optimization module iteratively updates the phase transition response characteristic spectrum using the following parameter learning formula: ； in, In the first In the next iteration, the graph mapping function The set of all adjustable parameters internally; The learning rate; For loss function For parameter set The gradient operator; The loss function; It is the measured normalized response vector calculated or estimated based on data from the multiphysics sensing module.

Images