Digital valve flow control system and optimal control method thereof

Abstract

The application belongs to the field of digital valve flow control, and discloses a digital valve flow control system and an optimal control method thereof, which comprises N high-speed on-off two-position two-way digital valves, each digital valve is connected in parallel between the same oil supply pipeline and a load loop, the flow area of each digital valve is configured according to a binary weighted proportion, the flow area of the i-th digital valve, the inlets of the digital valves are jointly connected to a system pressure oil circuit, and the outlets are jointly connected to a load flow output end; the scheme realizes valve group life balance, nonlinear adaptive compensation, disturbance suppression and full-range high-precision flow regulation through a binary parallel digital valve group, flow distribution optimization, RBF nonlinear compensation, extended state disturbance observation, model predictive control and sectional variable frequency module serial collaborative architecture, and the comprehensive performance is far superior to that of a traditional scheme in which each module is optimized alone.

Description

Technical Field

[0001] This invention relates to the field of digital valve flow control, and more particularly to a digital valve flow control system and its optimal control method. Background Technology

[0002] High-speed switching digital valves are increasingly widely used in hydraulic flow control due to their advantages such as simple structure, low cost, strong resistance to oil contamination, and fast response speed. Compared with traditional proportional valves and servo valves, high-speed switching valves only have two working states: open and closed. They achieve digital regulation of output flow by driving the valve core to switch at high frequency through PWM (pulse width modulation) signals. This retains the flexibility of digital control while reducing the system's requirements for oil cleanliness.

[0003] Existing high-speed two-position two-way digital valve parallel flow control systems mostly employ binary weighted diameter valve groups paired with independent PWM drives. However, traditional control schemes simply allocate the target total flow without simultaneously considering the instantaneous flow balance of each valve and the balance of cumulative switching action counts. This easily leads to problems such as long-term overload of a single valve, large differences in valve wear, and inconsistent overall valve group lifespan. Furthermore, digital valves themselves exhibit strong nonlinear characteristics caused by dead zone, saturation, and pressure difference changes. Conventional control lacks an online adaptive nonlinearity compensation mechanism, and predictive control models are prone to drift with operating conditions, resulting in flow tracking deviations. Moreover, external disturbances such as oil supply pressure and load fluctuations cannot be observed in real time, and the duty cycle solution range cannot be constrained, making control saturation and flow oscillation overshoot highly likely. In addition, traditional systems with PWM... With a fixed carrier frequency, it is impossible to dynamically adapt the frequency according to the dead zone, linear zone, and saturation zone of the duty cycle. It is difficult to achieve both full-range flow regulation accuracy and pulsation suppression effect. Furthermore, the functional modules of flow distribution, nonlinear compensation, disturbance observation, and model predictive control are independent of each other and lack a unified serial information flow collaborative architecture. The optimization effects of each module cancel each other out, making it impossible to simultaneously improve valve group life, control accuracy, disturbance rejection capability, and full-range regulation quality. Summary of the Invention

[0004] The present invention aims to provide a digital valve flow control system and its optimal control method to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A digital valve flow control system includes: N high-speed switching two-position two-way digital valves. Each digital valve is connected in parallel between the same oil supply line and the load circuit; the nominal area of ​​each digital valve is configured according to a binary weighted ratio, i.e., the... The diameter area of ​​a digital valve ,in The flow area of ​​the minimum diameter valve, The inlets of each digital valve are connected to the system pressure oil circuit, and the outlets are connected to the load flow output terminal; each digital valve is independently configured with a PWM drive signal. The system pressure sensor is installed on the inlet side of the valve group; the total flow sensor is installed on the outlet side of the valve group; the branch flow sensor or differential pressure sensor is installed on the outlet side of each digital valve. The controller receives signals collected by various sensors. The controller contains a flow distribution optimizer, an RBF neural network nonlinear compensator, an extended state observer, and a model prediction controller. The flow distribution optimizer receives the target total flow command and the real-time differential pressure signal of each digital valve. With the optimization objectives of equalizing the instantaneous flow of each valve and equalizing the cumulative number of switching actions of each valve, it solves the flow distribution command of each valve and sends the flow distribution command to the RBF neural network nonlinear compensator and the model predictive controller respectively. The RBF neural network nonlinear compensator receives the flow distribution command and the real-time differential pressure signal, outputs the duty cycle compensation amount of each valve, and inputs the compensation amount into the model predictive controller in an online update manner to update the state transition matrix and input matrix of the model predictive controller. The expansion state observer receives the actual total flow rate of the system and the actual duty cycle of each valve, outputs the estimated total disturbance of the system, and inputs the estimated disturbance value into the model prediction controller in a real-time feedback manner to dynamically adjust the upper and lower limits of the duty cycle optimization solution interval of each valve. The model predictive controller receives the state transition matrix and input matrix updated by the RBF neural network nonlinear compensator, the duty cycle optimization solution interval adjusted by the extended state observer, and the flow allocation instructions for each valve issued by the flow allocation optimizer, and solves the optimal duty cycle control sequence for each digital valve. Each digital valve's PWM drive signal is generated independently according to its optimal duty cycle, and drives the corresponding digital valve.

[0006] Preferably, the flow allocation optimizer is equipped with a trigger threshold judgment mechanism: when the cumulative number of switching actions of any digital valve exceeds the product of the average number of actions of all digital valves and a preset coefficient, the optimization weight of the target for balancing the cumulative number of switching actions of each valve is automatically increased, so that subsequent flow allocation instructions will preferentially call the digital valves with fewer actions.

[0007] Preferably, the controller further includes a segmented frequency adaptive adjustment module. The segmented frequency adaptive adjustment module receives the optimal duty cycle of each valve output by the model prediction controller and determines the PWM carrier frequency of the corresponding digital valve according to the interval category in which the duty cycle is located: a first frequency is used when it is in the dead zone or saturation zone, and a second frequency higher than the first frequency is used when it is in the linear zone and the current flow tracking error is less than a preset threshold.

[0008] Preferably, the minimum diameter valve is configured such that the PWM duty cycle corresponding to the valve under the minimum operating flow condition of the system is not less than 30% of its linear adjustment range.

[0009] A method for optimal flow control of digital valves is also disclosed, applied to a high-speed switching two-position two-way digital valve flow control system. The system comprises N digital valves connected in parallel and configured with their pipe areas according to a binary weighted ratio. The method includes: A step PWM signal is applied to each digital valve, and the opening delay time, opening motion time, closing delay time and closing motion time of each valve are extracted to establish a piecewise nonlinear mapping relationship between the PWM duty cycle of each valve and the output flow rate. After entering closed-loop control, perform the following steps: Step 1: Receive the target total flow command, calculate the differential pressure correction coefficient of each valve based on the real-time differential pressure of each digital valve, and solve the flow distribution command of each valve with the instantaneous flow balance of each valve and the balance of the cumulative number of switching actions of each valve as multi-objective functions; wherein, the weight of the cumulative number of switching actions balance objective in the first control cycle is zero, and from the second control cycle onwards, this weight is updated in real time according to the cumulative action count statistics of the previous cycles; Step 2: Using the flow distribution command and real-time differential pressure signal of each valve as input to the RBF neural network, output the duty cycle compensation of each valve, and input the compensation amount into the model predictive controller in an online update manner to update the state transition matrix and input matrix of the model predictive controller; Step 3: Using the actual total flow rate of the system and the actual duty cycle of each valve as inputs to the extended state observer, output the estimated total disturbance of the system. Input the estimated disturbance value into the model predictive controller in real time feedback mode to dynamically adjust the upper and lower limits of the duty cycle optimization solution interval of each valve. Step 4: The model predictive controller uses the updated state transition matrix and input matrix, the adjusted duty cycle optimization solution interval, and the flow distribution commands of each valve output in Step 1 as a basis to solve for the optimal duty cycle control sequence of each digital valve. Step 5: Based on the range category of the optimal duty cycle of each valve output in Step 4, determine the PWM carrier frequency of each digital valve, generate an independent PWM drive signal for each valve, and apply it to the electromagnetic coil of each digital valve. Step 6: In the next control cycle, return to Step 1.

[0010] Preferably, in step one, the weight update method for the balance target of the cumulative number of switching actions of each valve is as follows: the cumulative number of switching actions of each digital valve is counted in real time, and when the cumulative number of actions of any digital valve exceeds the product of the average number of actions of all valves and a preset coefficient, the weight of the life balance target in the multi-objective function is increased.

[0011] Preferably, in step three, the duty cycle optimization solution interval of each valve is dynamically adjusted according to the disturbance estimate as follows: when the absolute value of the disturbance estimate increases, the upper limit of the duty cycle of each valve decreases and the lower limit increases, causing the optimization solution interval to shrink; when the absolute value of the disturbance estimate decreases, the upper limit of the duty cycle of each valve increases and the lower limit decreases, causing the optimization solution interval to expand.

[0012] Preferably, in step five, the PWM carrier frequency of each digital valve is determined as follows: when the optimal duty cycle of a valve is in the dead zone or saturation zone, a first frequency is used; when it is in the linear zone and the current flow tracking error is less than a preset threshold, a second frequency higher than the first frequency is used; when it is in the nonlinear transition zone or the tracking error is not less than a preset threshold, a third frequency between the first frequency and the second frequency is used.

[0013] Preferably, the piecewise nonlinear mapping relationship includes four intervals: dead zone, nonlinear transition zone, linear zone, and saturation zone; the linear zone is the interval where the output flow rate and PWM duty cycle have an approximately linear relationship; the dead zone is the interval where the output flow rate is zero when the PWM duty cycle is less than the activation threshold; the saturation zone is the interval where the output flow rate reaches its maximum value when the PWM duty cycle is greater than the saturation threshold and no longer increases with the duty cycle.

[0014] Preferably, the orifice diameter of each digital valve is according to The proportional configuration; the combined flow range of each digital valve covers the range from the rated flow of a single minimum diameter valve to the total flow when all valves are fully open; under low flow conditions, only the minimum diameter digital valve is opened, and under high flow conditions, the total flow is distributed according to the principle of equalizing the instantaneous flow of each valve.

[0015] The beneficial effects of this technical solution compared to existing technologies are as follows: (1) This scheme adopts a unidirectional information flow collaborative architecture with serial coupling of flow distribution optimizer, RBF neural network nonlinear compensator, model predictive controller and extended state observer, to simultaneously realize multi-objective flow balance, nonlinear online compensation, disturbance adaptive constraint and segmented frequency conversion control, which is different from the defects of the independent optimization effects of each module in the existing technology. On the one hand, by balancing the instantaneous load and switching frequency of each valve through dual-objective flow distribution, the loss gap of each valve is greatly reduced and the overall service life of the valve group is extended. On the other hand, relying on the RBF online correction prediction model and the extended state observer to dynamically constrain the duty cycle solution range, the valve nonlinearity and external operating condition disturbance are adaptively offset. In addition, the segmented adaptive PWM frequency matching of different adjustment range characteristics is combined to simultaneously improve the full range flow control accuracy, anti-disturbance capability and output stability of the system, and realize the systematic synchronous optimization of four indicators: valve group life, control accuracy, anti-disturbance performance and full range regulation quality. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of the present invention; Detailed Implementation The present invention will now be described in further detail with reference to the accompanying drawings and embodiments: like Figure 1 As shown, the present invention provides a digital valve flow control system, including N high-speed switching two-position two-way digital valves 1, a system pressure sensor 2, a total flow sensor 3, a branch flow sensor 4 or a differential pressure sensor 5, and a controller 6.

[0017] In this embodiment, N is 4, meaning there are 4 digital valves, designated as first digital valve 11, second digital valve 12, third digital valve 13, and fourth digital valve 14. The four digital valves are connected in parallel between the same oil supply line and the load circuit. The inlet of each digital valve is connected to the system pressure oil circuit P port, and the outlet is connected to the load flow output terminal L port.

[0018] The flow areas of the four digital valves are configured according to a binary weighted ratio of 1:2:4:8. Specifically, if the flow area of ​​the smallest diameter valve 11 is A0, then the flow area of ​​the second digital valve 12 is 2A0, the flow area of ​​the third digital valve 13 is 4A0, and the flow area of ​​the fourth digital valve 14 is 8A0. By different combinations of the open and closed states of each valve, 24-1=15 discrete flow levels can be achieved. Based on this, combined with the continuous adjustment of the PWM duty cycle of each valve, fine continuous flow control can be achieved between each level. For example, when the target flow rate is 3A0 equivalent, the first digital valve 11 and the second digital valve 12 can be opened, while the third digital valve 13 and the fourth digital valve 14 remain closed to achieve the basic flow level of 3A0. Then, by adjusting the PWM duty cycle of the first digital valve 11 or the second digital valve 12, continuous flow output between 3A0 and 4A0 can be achieved.

[0019] In this embodiment, each digital valve 1 is a normally closed two-position two-way high-speed switching solenoid valve, and the rated flow rate of the minimum diameter valve 11 is 5L / min. Each digital valve is independently configured with a PWM drive signal, and the controller 6 outputs four independent PWM signals to drive the solenoid coils of the four digital valves respectively.

[0020] System pressure sensor 2 is installed on the oil supply line on the inlet side of the valve group to collect the system oil supply pressure in real time. The total flow sensor 3 is installed on the load circuit on the outlet side of the valve group to collect the total flow rate output by the valve group in real time. Branch flow sensors 4 are installed on the outlet side of each digital valve 1 to collect the branch flow of each valve in real time. As an alternative, differential pressure sensors 5 can be installed on the outlet side of each digital valve. By measuring the actual differential pressure at each valve port and combining it with the PWM duty cycle of each valve, the controller 6 can calculate the branch flow of each valve in real time.

[0021] Controller 6 employs an industrial-grade embedded controller with a built-in processor. Controller 6 communicates with the host computer via CAN bus or Ethernet to receive the target total flow command; it acquires signals from various sensors through an analog input interface; and it outputs drive signals for each digital valve through a PWM output interface. Internally, controller 6 includes, in the form of software modules, a flow distribution optimizer 601, an RBF neural network nonlinear compensator 602, an expansion state observer 603, a model predictive controller 604, and a segmented frequency adaptive adjustment module 605.

[0022] In another embodiment, an initialization identification step is performed before the system engages closed-loop control.

[0023] The controller 6 sequentially applies step PWM signals to each digital valve 1 using its step signal generation function. Taking the first digital valve 11 as an example, a step voltage signal with a 100% duty cycle is applied, while the coil current feedback signal and the downstream pressure response signal of the valve are acquired at a high sampling rate (sampling frequency not less than 10kHz). Based on the acquired current and pressure curves, the controller 6 extracts four inherent time parameters of the valve: opening delay time. Start your exercise time Closing delay time Turn off exercise time .

[0024] Specifically, the opening delay time is the time from the rising edge of the step signal until the first abrupt change in the coil current (which corresponds to the change in inductance caused by the change in the air gap when the valve core begins to move). The time from the point of sudden current change to the point where the current reaches the second point of sudden change (which corresponds to the point where the valve core reaches its maximum displacement) is the opening time. Similarly, the closing delay time is the time from the falling edge of the step signal until the downstream pressure begins to drop significantly. The time from the onset of pressure drop to the point where pressure stabilizes is the shutdown time. .

[0025] In this embodiment, the four time parameters of the first digital valve 11 obtained by actual measurement are as follows: The second digital valve 12, the third digital valve 13, and the fourth digital valve 14 have the same structure, and their time parameters are basically the same as those of the first digital valve 11. In practical applications, they can be measured separately or the same set of parameters can be used.

[0026] After parameter identification is completed, the PWM duty cycle of each valve is established based on the above four time parameters. With output flow A piecewise nonlinear mapping relationship exists between the intervals, which includes four intervals: when At this time, the valve is in the dead zone, and the output flow rate is... ,in , The PWM signal period; when At this time, the valve is in the nonlinear transition region, and the output flow rate increases nonlinearly with the duty cycle; when When the valve is in the linear region, the output flow rate has an approximately linear relationship with the duty cycle. ; when At this time, the valve is in the saturation region, and the output flow rate reaches its maximum value. And it no longer increases with the duty cycle.

[0027] In actual engineering, the nonlinear transition region may present as a single nonlinear interval or as two nonlinear sub-intervals (i.e., the rising edge transition sub-interval and the falling edge transition sub-interval) depending on the specific parameters of each valve. This is a natural morphological difference of the same nonlinear transition region under different valve parameters, and all of them fall within the protection scope of the four-segment piecewise nonlinear mapping relationship described in this invention.

[0028] The piecewise nonlinear mapping relationship is stored in controller 6 in the form of a lookup table for real-time determination of duty cycle interval category in subsequent steps.

[0029] In another embodiment, after the system enters closed-loop control, the flow allocation optimizer 601 of the controller 6 performs the following flow allocation optimization steps.

[0030] Assuming the current target total flow instruction is The maximum output flow rates of the four digital valves are as follows: , , , The system pressure sensor 2 measures the current oil supply pressure. The real-time differential pressure measured by the branch flow sensor 4 or differential pressure sensor 5 on the outlet side of each valve is as follows: , , , .

[0031] First, the flow distribution optimizer 601 calculates the differential pressure correction coefficient based on the real-time differential pressure of each valve. : in The rated differential pressure for each valve is 1 MPa in this embodiment. This correction factor is based on the throttling flow characteristics; when the actual differential pressure of a valve is lower than the rated differential pressure, the correction factor is greater than 1, indicating that the valve can output a larger flow rate under the same duty cycle, and its flow distribution weight should be appropriately increased; the opposite is also true.

[0032] Then, the flow allocation optimizer 601 constructs a multi-objective optimization function: in The instantaneous flow balance target for each valve is expressed as the sum of squares of the differences between the ratio of the flow allocated to each valve to its maximum flow (i.e., the instantaneous flow rate of each valve) and the average ratio. The target for balancing the cumulative number of switching actions of each valve is expressed as the sum of squares of the ratios of the cumulative number of actions of each valve to the average number of actions. and These are the weighting coefficients for the two objectives. The instantaneous flow balancing objective ensures that under any operating condition, each valve shares the total flow proportionally to its diameter, avoiding any valve being overloaded instantaneously.

[0033] Under constraints and Solve the above optimization problem and output the optimal flow distribution command for each valve. .

[0034] It should be noted that during the first control cycle, the cumulative number of switching actions of each valve is zero. The item has no practical significance, therefore it is set. The optimization objective is solely instantaneous flow balance. From the second control cycle onwards... The system updates in real time based on the cumulative number of actions from previous cycles. Specifically, controller 6 counts the cumulative number of opening and closing actions of each valve in real time. When the cumulative number of actuations of any valve More than the average number of valve actions With preset coefficient (In this embodiment) When taking the product of 1.5), that is... The threshold judgment mechanism 6011 is triggered, and the threshold is automatically increased. The value of determines whether subsequent flow distribution commands prioritize the use of digital valves with fewer operations for regulation.

[0035] For example, after the system has been running for a period of time, the first digital valve 11 accumulates a certain number of actuations due to its frequent participation in small flow regulation. The cumulative number of actuations for the other three valves were as follows: Second-rate, Second-rate, Number of times, average number of movements Next. At this time. This triggers the lifespan balancing mechanism. Increasing the value from the initial 0.3 to 0.6 causes the flow allocation optimizer to prioritize the allocation of flow to the second, third, and fourth digital valves in subsequent cycles, reducing the frequency of the first digital valve being called.

[0036] In another embodiment, the flow allocation optimizer 601 outputs flow allocation commands for each valve. The signals are transmitted to the RBF neural network nonlinear compensator 602 and the model prediction controller 604, respectively.

[0037] The RBF neural network nonlinear compensator 602 uses flow distribution commands from each valve. and real-time differential pressure As the input feature vector, the actual output flow rate of each valve As the network output target, the network parameters are continuously optimized through online training to output the predicted compensation amount of duty cycle for each valve. .

[0038] In this embodiment, the RBF neural network adopts a three-layer structure: the input layer contains 2 neurons (corresponding to...) and The hidden layer contains 7 neurons, and the output layer contains 1 neuron (corresponding to...). The activation function for the hidden layer is a Gaussian radial basis function. in For the input vector, Let be the center vector of the Gaussian function of the i-th hidden layer neuron. The basis width vector. The center of the Gaussian function. The basal width vector is initialized offline using the K-means clustering algorithm. Take 1.5 times the average distance between all cluster centers. The network training algorithm described below is merely an exemplary implementation of this invention, and the invention is not limited thereto.

[0039] The output of the network output layer is: in These are the network weights from the hidden layer to the output layer.

[0040] The network training uses gradient descent to update the weights online. Center of Gaussian function and base width vector Define the loss function: in This represents the traffic output value predicted by the RBF neural network. Based on the gradient of the loss function with respect to each parameter, the network parameters are adjusted online according to the following update rules: in , , These represent the learning rates for each parameter, which are set to 0.01, 0.005, and 0.005 respectively in this embodiment. The parameter updates of the RBF neural network are performed in each control cycle to ensure that the network output can track the changes in the nonlinear characteristics of the digital valve in real time.

[0041] The duty cycle prediction compensation amount output by the RBF neural network nonlinear compensator 602 The compensation is input to the model predictive controller 604 via online updates during each control cycle. This compensation is not simply added to the output of the MPC, but is directly used to correct the predictive model parameters of the MPC—that is, to obtain the state transition matrix of the MPC. and input matrix The system updates in real time based on the received compensation amount during each control cycle. in and In this embodiment, the learning rate for model updates is set to 0.1 and 0.1 respectively. In engineering implementation, the partial derivatives can be obtained using a difference approximation method, without the need for analytical expressions. Through this online model correction mechanism, the MPC prediction model can track the nonlinear characteristic drift of the digital valve caused by factors such as pressure difference and oil temperature changes in real time, thus maintaining the model accuracy.

[0042] In another embodiment, the extended state observer 603 uses the actual total system flow rate. and the actual duty cycle of each valve Input: Estimated total system disturbance. Output: Estimated total system disturbance. .

[0043] First, a state-space model of the digital valve flow control system is established. The weighted sum of the flow rates of each valve is used as the system output, and the duty cycle command of each valve is used as the system input. Considering the uncertainty of system modeling and external load fluctuations, the total disturbance is extended into new state variables, and an extended state observer is constructed.

[0044] As a specific implementation, this embodiment employs a fourth-order extended state observer. The observer order and parameter tuning methods described below are merely examples; those skilled in the art can select extended state observers of different orders based on the actual order of the controlled system. The total system disturbance is extended into a fourth-order state variable. The observation equations for the fourth-order extended state observer are as follows: in For observation error, , , System status , , The observed values, This is the estimated total system disturbance. , is the observer gain coefficient.

[0045] The observer gain coefficient is determined using the quadruple pole placement method. Let the desired bandwidth of the observer be... (In this embodiment, we take) If the frequency is approximately four times the open-loop cutoff frequency of the system, then: , , , The total system disturbance estimate output by the extended state observer 603 The disturbance estimate is input to the model predictive controller 604 in real-time feedback during each control cycle. This disturbance estimate is not simply added to the output of the MPC as a feedforward compensation term, but is used to dynamically adjust the upper and lower limits of the duty cycle optimization solution interval for each valve. in and These are the initially set upper and lower limits of the duty cycle (95% in this embodiment) and the duty cycle lower limit (5% in this embodiment), respectively. This is the disturbance sensitivity coefficient (0.3 in this embodiment).

[0046] When the system suffers a large load disturbance As the disturbance increases, the upper limit of the duty cycle decreases and the lower limit increases, dynamically shrinking the optimization solution range and ensuring that the optimal solution of MPC is always within the controllable range of the valve, thus avoiding control saturation. When the disturbance decreases, the upper limit of the duty cycle increases and the lower limit decreases, dynamically expanding the optimization solution range and fully releasing the optimal adjustment capability of MPC.

[0047] In another embodiment, the model prediction controller 604 receives the updated state transition matrix via the RBF neural network nonlinear compensator 602. and input matrix The duty cycle optimization solution interval adjusted by the extended state observer 603 and the flow allocation instructions for each valve issued by the flow allocation optimizer 601. Solve a quadratic programming problem with dynamic constraints in the prediction time domain N, and output the optimal duty cycle control sequence for each digital valve.

[0048] In this embodiment, the prediction time domain N is 10, and the control time domain M is 3. The model predictive controller 604 first establishes a discrete state-space prediction model for each valve based on the updated state transition matrix and input matrix: in Let i be the state vector of the i-th valve at time k. The duty cycle control input for the i-th valve at time k is... Let be the flow output of the i-th valve at time k.

[0049] Then, the model prediction controller 604 constructs the following performance metrics: The first item is the traffic tracking error penalty. The first term is the tracking error weight matrix; the second term is the control input penalty term. To control the weight matrix.

[0050] In dynamic constraints Solve the above quadratic programming problem to obtain the optimal duty cycle control sequence for each valve. The model predictive controller 604 outputs only the first element of the control sequence as the duty cycle command for the current cycle to the segmented frequency adaptive adjustment module 605. In the next control cycle, it re-solves the problem based on the new feedback value to achieve rolling time-domain closed-loop optimal control.

[0051] In another embodiment, the segmented frequency adaptive adjustment module 605 receives the optimal duty cycle of each valve output by the model prediction controller 604. Based on the interval category of the duty cycle and the magnitude of the current flow tracking error, the PWM carrier frequency of each digital valve 1 is determined.

[0052] The determination of the interval category is based on the piecewise nonlinear mapping relationship established in the initialization identification step. This piecewise nonlinear mapping relationship includes four intervals: dead zone, nonlinear transition zone, linear zone, and saturation zone. Frequency selection is strictly performed according to the following three-branch logic: First branch: When Located in a dead zone or saturation region Regardless of the magnitude of the tracking error, the valve always uses the first frequency. Driven. In this embodiment, the first frequency is 100Hz (period is 10ms). Under low-frequency drive, the duration of each cycle of the PWM signal is relatively large, the duty cycle range corresponding to the dead zone and saturation zone is relatively small, the linear controllable range is expanded, which helps to overcome the adverse effects of the dead zone and saturation zone on flow control.

[0053] Second branch: When In the linear region And the current traffic tracking error Less than the preset threshold (In this embodiment) When the rated flow rate is 3% (i.e., 0.15 L / min), the valve uses the second frequency. Driven by a high frequency, the second frequency is set to 500Hz (with a period of 2ms). Under high-frequency drive, the duration of each cycle of the PWM signal is shorter, and the amplitude of the flow pulsation after the valve is significantly reduced, which is beneficial to improving the accuracy of steady-state flow control.

[0054] Third branch: When In the nonlinear transition region Or it is in the linear region but the current flow tracking error Not less than the preset threshold At that time, the valve uses the third frequency. Driven. In this embodiment, the third frequency is 200Hz (with a period of 5ms), which is between the first frequency and the second frequency, in order to balance the requirements of linear controllable range and flow pulsation suppression.

[0055] It should be noted that the PWM carrier frequency of each digital valve is determined independently. For example, within a certain control cycle, the optimal duty cycle of the first digital valve 11 is in the linear region and the tracking error is small, so it is driven by a high frequency of 500Hz; while the optimal duty cycle of the second digital valve 12 is in the dead zone, so it is driven by a low frequency of 100Hz. Each valve is configured with its own PWM frequency, and they do not affect each other.

[0056] The segmented frequency adaptive adjustment module 605 combines the optimal duty cycle of each valve with the corresponding PWM carrier frequency, and the controller 6 generates an independent PWM drive signal for each valve, which is then applied to the electromagnetic coil of each digital valve 1.

[0057] In another embodiment, the complete execution flow of the digital valve flow optimal control method of the present invention is described.

[0058] Assuming the system has completed initialization and identification, the orifice ratio of the four digital valves (11 to 14) is 1:2:4:8, and the system is currently in closed-loop control.

[0059] Controller 6 receives the target total flow command from the host computer. The flow distribution optimizer 601 reads the current oil supply pressure measured by the system pressure sensor 2. And the real-time differential pressure measured by the differential pressure sensor 5 on the outlet side of each valve. , , , After calculating the differential pressure correction coefficient for each valve, and taking the balance of instantaneous flow rate and cumulative switching frequency of each valve as multiple objectives, the flow distribution command for each valve is obtained by solving the problem. At this point, the cumulative number of actuations for each valve is as follows: The lifetime balancing threshold was not triggered. Keep it at 0.3.

[0060] RBF neural network nonlinear compensator 602 receives and As input, after forward propagation calculation, the output is the duty cycle prediction compensation amount of the first digital valve. (That is, a 3% duty cycle compensation). This compensation is used to update the state transition matrix of the first digital valve in the model predictive controller 604. and input matrix The same applies to the other three valves.

[0061] The expanded state observer 603 reads the actual total system flow. and the actual duty cycle of each valve , 0、 , After the observer performs calculations, it outputs the estimated total disturbance of the current system. (A positive disturbance indicates that the actual flow rate is lower than expected). This disturbance estimate is used to dynamically adjust the duty cycle optimization solution range for each valve. Taking the first digital valve as an example, the initial duty cycle upper limit is 95% and the lower limit is 5%, which is adjusted to an upper limit of 92% and a lower limit of 8%.

[0062] Model predictive controller 604 with updated and Adjusted duty cycle range and traffic allocation instructions Based on this, in the prediction time domain Solve the quadratic programming problem internally and output the optimal duty cycle control sequence for the first digital valve. Take the first element As the duty cycle instruction for the current cycle.

[0063] Segmented Frequency Adaptive Adjustment Module 605 Determination It is within the linear region of the first digital valve (the linear region of this valve is 0.15~0.75), and the current flow tracking error is... Less than the threshold Therefore, the PWM carrier frequency of the first digital valve is determined to be 500Hz. Controller 6 combines the duty cycle of 28% with the frequency of 500Hz to generate and output the PWM drive signal for the first digital valve.

[0064] The above process is repeated in the next control cycle (the control cycle is 10ms in this embodiment).

[0065] In another embodiment, to more clearly illustrate the advantages of the present invention over a wide adjustment range, typical control processes for low-flow and high-flow conditions are given below.

[0066] Low-flow operating conditions: Assuming a target total flow At this point, the flow distribution optimizer 601, based on the flow distribution capacity of each valve and the current differential pressure, prioritizes distributing the flow to the smallest diameter valve 11. Specifically, the solution yields... At this point, only the first digital valve 11 participates in regulation, while the other three digital valves remain closed. The duty cycle of the first digital valve 11 at 2 L / min is approximately 40%, far exceeding the lower limit of its linear regulation range (approximately 15%). This valve operates within an optimal opening range, fundamentally avoiding the risks of cavitation and erosion at small openings. Simultaneously, since the other three large-diameter valves are closed, the system experiences no additional throttling losses.

[0067] High-traffic scenario: Assuming a target total traffic volume At this point, the flow distribution optimizer 601 takes the instantaneous flow balance of each valve as the main optimization objective, and solves the problem to obtain... The instantaneous flow rates (actual flow / rated flow) of each valve were 64%, 64%, 64.5%, and 56.25%, respectively, achieving a basic balance in flow distribution. All four valves participated in regulation simultaneously, with each operating within its optimal opening range, avoiding the significant throttling losses associated with a single valve under high flow conditions. Compared to a single-valve full-range throttling scheme, this embodiment reduces throttling energy consumption by approximately 40% under high flow conditions.

[0068] In another embodiment, the functional modules of the controller 6 are implemented as software programs and run on an embedded real-time operating system. The numerical solution algorithms described below are all exemplary implementation schemes.

[0069] The multi-objective optimization solution of the flow allocation optimizer 601 adopts the sequential quadratic programming (SQP) algorithm, which is called once in each control cycle and the solution time does not exceed 2ms.

[0070] The RBF neural network nonlinear compensator 602 is trained online using the gradient descent method, performing a complete parameter update once in each control cycle to ensure that the network output tracks the changes in the nonlinear characteristics of the system in real time, with a computation time of no more than 1ms.

[0071] The extended state observer 603 uses the fourth-order Runge-Kutta method for discretization and is executed once in each control cycle, with a computation time of no more than 0.5ms.

[0072] The quadratic programming solution of the Model Predictive Controller 604 adopts the effective set method, which is executed once in each control cycle, and the solution time does not exceed 3ms.

[0073] The segmented frequency adaptive adjustment module 605 performs a lookup comparison operation once per control cycle, with a calculation time of no more than 0.1ms.

[0074] The total calculation time of the entire control algorithm in a single cycle does not exceed 6.6ms, and the control cycle is set to 10ms, leaving sufficient margin to meet the requirements of real-time control.

[0075] The controller 6 communicates with the host computer via a CAN bus at a rate of 500kbps. The controller 6 sends system status data (including actual flow rate, duty cycle, cumulative number of actions, and current disturbance estimate of each valve) to the host computer every 10ms, and simultaneously receives target flow rate commands and parameter adjustment commands from the host computer.

[0076] The above descriptions are merely embodiments of the present invention, and common knowledge such as specific technical solutions and / or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A digital valve flow control system, characterized in that: include: N high-speed switching two-position two-way digital valves (1). Each digital valve (1) is connected in parallel between the same oil supply line and the load circuit; the bore area of ​​each digital valve (1) is configured according to a binary weighted ratio, i.e., the first... One digital valve (1) i ) duct area ,in For the flow area of ​​the minimum diameter valve (11), The inlets of each digital valve (1) are connected to the system pressure oil circuit, and the outlets are connected to the load flow output terminal. Each digital valve (1) is configured with an independent PWM drive signal. The system pressure sensor (2) is installed on the inlet side of the valve group; the total flow sensor (3) is installed on the outlet side of the valve group; the branch flow sensor (4) or differential pressure sensor (5) is installed on the outlet side of each digital valve (1); The controller (6) receives signals collected by each sensor. The controller (6) is equipped with a flow distribution optimizer (601), an RBF neural network nonlinear compensator (602), an expansion state observer (603), and a model prediction controller (604). The flow distribution optimizer (601) receives the target total flow command and the real-time differential pressure signal of each digital valve (1). With the instantaneous flow balance of each valve and the balance of the cumulative number of switching actions of each valve as the optimization objectives, it solves the flow distribution command of each valve and sends the flow distribution command to the RBF neural network nonlinear compensator (602) and the model predictive controller (604) respectively. The RBF neural network nonlinear compensator (602) receives the flow distribution command and the real-time differential pressure signal, outputs the duty cycle compensation amount of each valve, and inputs the compensation amount into the model predictive controller (604) in an online update manner to update the state transition matrix and input matrix of the model predictive controller (604). The expanded state observer (603) receives the actual total flow of the system and the actual duty cycle of each valve, outputs the estimated total disturbance of the system, and inputs the estimated disturbance into the model prediction controller (604) in a real-time feedback manner to dynamically adjust the upper and lower limits of the solution interval for the duty cycle optimization of each valve; The model predictive controller (604) receives the state transition matrix and input matrix updated by the RBF neural network nonlinear compensator (602), the duty cycle optimization solution interval adjusted by the extended state observer (603), and the flow distribution instructions for each valve issued by the flow distribution optimizer (601), and solves the optimal duty cycle control sequence for each digital valve (1). The PWM drive signal of each digital valve (1) is generated independently according to its optimal duty cycle, and drives the corresponding digital valve (1) respectively.

2. The digital valve flow control system as described in claim 1, characterized in that, The flow allocation optimizer (601) is equipped with a trigger threshold judgment mechanism (6011): when any digital valve (1 i When the cumulative number of switching actions of a valve exceeds the product of the average number of actions of all digital valves (1) and a preset coefficient, the optimization weight of the target for balancing the cumulative number of switching actions of each valve is automatically increased, so that subsequent flow distribution commands will preferentially call the digital valves (1) with fewer actions.

3. The digital valve flow control system as described in claim 1, characterized in that, The controller (6) is further provided with a segmented frequency adaptive adjustment module (605). The segmented frequency adaptive adjustment module (605) receives the optimal duty cycle of each valve output by the model prediction controller (604) and determines the PWM carrier frequency of the corresponding digital valve (1) according to the interval category in which the duty cycle is located: the first frequency is used when it is in the dead zone or saturation zone, and the second frequency higher than the first frequency is used when it is in the linear zone and the current flow tracking error is less than the preset threshold.

4. The digital valve flow control system as described in claim 1, characterized in that, The minimum diameter valve (11) is configured such that the PWM duty cycle corresponding to the valve under the minimum operating flow condition of the system is not less than 30% of its linear adjustment range.

5. A method for optimal flow control of a digital valve, characterized in that: A flow control system for a high-speed switching two-position two-way digital valve is applied, wherein the system comprises N digital valves connected in parallel and configured with their diameter areas in a binary weighted ratio (1). The method includes: Apply a step PWM signal to each digital valve (1), extract the opening delay time, opening motion time, closing delay time and closing motion time of each valve, and establish a piecewise nonlinear mapping relationship between the PWM duty cycle and the output flow of each valve; After entering closed-loop control, perform the following steps: Step 1: Receive the target total flow command, calculate the pressure difference correction coefficient of each valve based on the real-time pressure difference of each digital valve (1), and solve the flow distribution command of each valve with the instantaneous flow balance of each valve and the balance of the cumulative number of switching actions of each valve as the multi-objective function; wherein, the weight of the cumulative number of switching actions balance target in the first control cycle is zero, and from the second control cycle onwards, this weight is updated in real time according to the cumulative number of actions statistics of the previous cycles; Step 2: Using the flow distribution command and real-time differential pressure signal of each valve as input to the RBF neural network, output the duty cycle compensation of each valve, and input the compensation amount into the model predictive controller in an online update manner to update the state transition matrix and input matrix of the model predictive controller; Step 3: Using the actual total flow rate of the system and the actual duty cycle of each valve as inputs to the extended state observer, output the estimated total disturbance of the system. Input the estimated disturbance value into the model predictive controller in real time feedback mode to dynamically adjust the upper and lower limits of the duty cycle optimization solution interval of each valve. Step 4: The model predictive controller uses the updated state transition matrix and input matrix, the adjusted duty cycle optimization solution interval, and the valve flow distribution instructions output in Step 1 as the basis to solve the optimal duty cycle control sequence of each digital valve (1). Step 5: Based on the range category of the optimal duty cycle of each valve output in Step 4, determine the PWM carrier frequency of each digital valve (1), generate an independent PWM drive signal for each valve, and apply it to the electromagnetic coil of each digital valve (1). Step 6: In the next control cycle, return to Step 1.

6. The digital valve flow optimal control method as described in claim 5, characterized in that, In step one, the weight update method for the target of balancing the cumulative switching action count of each valve is as follows: The cumulative switching action count of each digital valve (1) is counted in real time. When any digital valve (1)... i When the cumulative number of actions of a valve exceeds the product of the average number of actions of all valves and a preset coefficient, the weight of the life balance objective in the multi-objective function is increased.

7. The digital valve flow optimal control method according to claim 5, characterized in that, In step three, the duty cycle optimization solution range of each valve is dynamically adjusted according to the disturbance estimate as follows: when the absolute value of the disturbance estimate increases, the upper limit of the duty cycle of each valve decreases and the lower limit increases, causing the optimization solution range to shrink; when the absolute value of the disturbance estimate decreases, the upper limit of the duty cycle of each valve increases and the lower limit decreases, causing the optimization solution range to expand.

8. The digital valve flow optimal control method according to claim 5, characterized in that, In step five, the PWM carrier frequency of each digital valve (1) is determined as follows: when the optimal duty cycle of a valve is in the dead zone or saturation zone, the first frequency is used; when it is in the linear zone and the current flow tracking error is less than the preset threshold, the second frequency higher than the first frequency is used; when it is in the nonlinear transition zone or the tracking error is not less than the preset threshold, the third frequency between the first frequency and the second frequency is used.

9. The digital valve flow optimal control method according to claim 5, characterized in that, The piecewise nonlinear mapping relationship includes four regions: dead zone, nonlinear transition zone, linear zone, and saturation zone. The linear zone is the region where the output flow rate and PWM duty cycle have an approximately linear relationship. The dead zone is the region where the output flow rate is zero when the PWM duty cycle is less than the activation threshold. The saturation zone is the region where the output flow rate reaches its maximum value when the PWM duty cycle is greater than the saturation threshold and no longer increases with the duty cycle.

10. The digital valve flow control system according to any one of claims 1 to 4 or the digital valve flow optimal control method according to any one of claims 5 to 9, characterized in that, The orifice diameter of each digital valve (1) is according to The proportional configuration; the combined flow range of each digital valve (1) covers the total flow from the rated flow of a single minimum diameter valve (11) to the total flow when all valves are fully open; under low flow conditions, only the minimum diameter digital valve (11) is opened, and under high flow conditions, the total flow is distributed according to the principle of equalizing the instantaneous flow of each valve.

Images