Wind power box casting bottom injection type constant pressure pouring system

Abstract

The present application relates to heavy foundry industry and automation control technical field, specifically to wind power box casting bottom pouring type constant pressure pouring system, including: data acquisition module is used for obtaining real-time total weight data from liquid storage container;Data preprocessing module is used for converting the mass loss rate into the cumulative volume of the injected target forming cavity based on the preset fluid medium density;State estimation module is used for obtaining the pre-constructed height-volume integral function data table representing the height-volume integral mapping relationship, and calculating the real-time estimated flow rate of the fluid medium;Predictive control module is used for obtaining the preset target filling rising velocity, and generating the absolute position instruction for adjusting the target valve;Control data output module is used for packaging the absolute position instruction into standard control data packet, and updating the fluid injection state marker of system database;The present application realizes high-precision, zero-delay real-time estimated flow rate calculation.

Description

Technical Field

[0001] This invention relates to the fields of heavy casting industry and automation control technology, specifically to a bottom-pouring constant pressure casting system for wind turbine housing castings. Background Technology

[0002] Bottom-pouring casting process is often used in the manufacture of large wind turbine housing castings. Its technical essence is to dynamically monitor and control the liquid level and filling flow rate of the fluid medium in the mold cavity. Constant pressure casting control is a common method to ensure the molding quality of complex castings.

[0003] Existing methods for acquiring and controlling pouring liquid level and flow rate mostly rely on direct calculus derivation from time-series data of weighing sensors, or on measuring the actual liquid level using a single sensor. Their actuators often employ traditional single-loop proportional-integral-derivative (PID) algorithms for servo drive. For the direct differential derivation of flow rate, the high-frequency mechanical noise from crane swaying and liquid surface oscillations in industrial environments leads to extreme oscillations in the control output. For methods relying solely on sensors to obtain height, there is significant physical damping delay in areas of abrupt changes in cross-sectional area. Furthermore, traditional fixed cross-sectional area conversion methods produce severe model errors when facing nonlinear abrupt changes in volume ratio. Conventional reactive control is prone to overshoot when facing extreme geometric bottlenecks, potentially causing destructive pressure jets in the ingate.

[0004] Currently, for complex scenarios involving multi-dimensional intersections and sudden increases in thickness in the internal structures of wind turbine gearboxes, it is necessary to acquire a zero-delay reference data stream and a micro-flow field feedback with a high signal-to-noise ratio to assist in pressure regulation monitoring. Therefore, how to eliminate the inherent geometric blind spots and feedback lags under strong mechanical interference and nonlinear fluid dynamics environments, and achieve high-precision joint state estimation and feedforward flexible control of valves, has become a problem to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a bottom-pouring constant-pressure casting system for wind turbine housing castings, solving the following technical problems:

[0006] It eliminates the inherent geometric blind zone and feedback lag of traditional single closed-loop PID level control in nonlinear cavities, and endows the actuator with flexible adjustment capability through spatial feedforward prediction, thereby eliminating the risk of sand erosion at the source and significantly reducing defects to achieve a substantial leap in engineering quality.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A bottom-pouring constant pressure casting system for wind turbine housing castings includes a control terminal, which is communicatively connected to a target servo valve, a data acquisition module, a data preprocessing module, a state estimation module, a predictive control module, and a control data output module.

[0009] The data acquisition module is used to obtain real-time total weight data from the liquid storage container and real-time liquid level data from the buffer injection area.

[0010] The data preprocessing module is used to filter the real-time total weight data to obtain the mass loss rate, and convert the mass loss rate into the cumulative volume injected into the target molding cavity based on the preset fluid medium density.

[0011] The state estimation module is used to obtain a pre-constructed height-volume integral function data table that represents the mapping relationship between height and volume integral. Based on the cumulative volume and the height-volume integral function data table, the theoretical liquid level height is deduced. The extended Kalman filter operator is used to fuse the theoretical liquid level height with the real-time liquid level height data to calculate the real-time estimated flow velocity of the fluid medium.

[0012] The predictive control module is used to obtain the preset target filling and rising speed, extract the cross-sectional area step change characteristics of the target forming cavity based on the height volume integral function data table, and use the model predictive control algorithm to combine the cross-sectional area step change characteristics with the real-time estimated flow velocity to generate an absolute position command for adjusting the target servo valve.

[0013] The control data output module is used to encapsulate the absolute position command into a standard control data packet and update the fluid injection status flag in the system database.

[0014] Preferably, the process by which the data acquisition module obtains the real-time total weight data from the liquid storage container and the real-time liquid level data from the buffer injection area includes:

[0015] The real-time total weight data is obtained by a weighing sensor arranged in the liquid storage container; the real-time liquid level data is obtained by a laser rangefinder arranged in the buffer injection area.

[0016] Preferably, the process by which the data preprocessing module filters the real-time total weight data to obtain the mass loss rate, and converts the mass loss rate into the cumulative volume injected into the target molding cavity based on the preset fluid medium density includes:

[0017] The real-time total weight data is denoised using a moving average filtering algorithm.

[0018] The quality loss rate is calculated by extracting the temporal variation features of the denoised real-time total weight data.

[0019] The cumulative mass is obtained by integrating the mass loss rate over time; the cumulative volume is calculated by dividing the cumulative mass by the preset fluid medium density.

[0020] Preferably, the process by which the state estimation module obtains the pre-constructed height-volume integral function data table representing the mapping relationship between height and volume integral includes:

[0021] Obtain slice data of the three-dimensional computer-aided design model of the target molding cavity;

[0022] Extract the cross-sectional area variation characteristics with height from the slice data of the three-dimensional computer-aided design model;

[0023] The three-dimensional physical shape is reduced and compressed into a one-dimensional spatial state diagram; based on the one-dimensional spatial state diagram, a height-volume integral function data table representing the mapping relationship between height and volume integral is constructed.

[0024] Preferably, the process by which the state estimation module deduces the theoretical liquid level height based on the cumulative volume and the height-volume integral function data table includes: using the cumulative volume as an input variable;

[0025] Perform a reverse lookup match in the height-volume integral function data table; output the theoretical liquid level height inside the target forming cavity corresponding to the cumulative volume.

[0026] Preferably, the process of fusing the theoretical liquid level height with the real-time liquid level height data using the extended Kalman filter operator to calculate the real-time estimated flow velocity of the fluid medium includes:

[0027] Construct a state vector that includes real-time altitude state and transient flow velocity state;

[0028] The cross-sectional area step change feature is extracted as a nonlinear driving term in the state transition matrix; the real-time liquid level height data is set as the observation variable;

[0029] The theoretical liquid level height and the actual flow resistance derived from the observed variables are fused using the covariance matrix; the denoised real-time estimated flow velocity is then output.

[0030] Preferably, the process by which the predictive control module generates the absolute position command for adjusting the target valve includes:

[0031] A constant filling speed target is set based on the preset target filling rise speed;

[0032] The predicted cross-sectional area of ​​the liquid surface within a preset time period is predicted using the height-volume integral function data table; the rate of change of the predicted cross-sectional area is calculated.

[0033] The configuration is as follows: if the rate of change is greater than or equal to a preset area mutation threshold, then at a preset early intervention time before reaching the area mutation region, the absolute position command to reduce the opening of the target servo valve is generated; otherwise, the absolute position command to maintain the current flow regulation state is generated based on the real-time estimated flow velocity.

[0034] Preferably, the fluid medium is high-temperature molten iron; the target forming cavity is the internal cavity of the sand casting mold for a wind turbine gearbox.

[0035] The areas of abrupt changes in area include the planetary carrier support hole area or the flange area.

[0036] The beneficial effects of this invention are:

[0037] 1. The system of the present invention uses a moving average filtering algorithm to denoise the real-time total weight data through a data preprocessing module, and combines the fluid medium density to convert it into cumulative volume. This mechanism effectively eliminates high-frequency mechanical noise interference caused by vehicle shaking in industrial sites, avoids the problem of extreme oscillation of control quantity caused by direct calculus, and provides high signal-to-noise ratio and no physical hysteresis benchmark data for state estimation.

[0038] 2. The state estimation module of this invention reduces and compresses the three-dimensional physical shape of the target forming cavity into a height-volume integral data table, and uses the extended Kalman filter operator to fuse the theoretical height with the real-time observed height. This effectively eliminates the model error caused by the conversion of the fixed cross-sectional area of ​​the nonlinear cavity, solves the physical damping delay problem of a single sensor in the cross-sectional area change zone, and realizes high-precision, zero-delay real-time estimation of flow velocity.

[0039] 3. The predictive control module of this invention adopts a model predictive control algorithm, combined with the characteristics of step change in cross-sectional area, to generate an absolute position command to reduce the target valve opening in advance before the liquid surface reaches the area of ​​abrupt change; this gives the actuator the ability to predict the future and make flexible adjustments, avoiding the overshoot phenomenon caused by conventional reactive control at extreme geometric bottlenecks, and eliminating the risk of destructive pressure jets and sand flushing from the root.

[0040] 4. This invention addresses complex application scenarios involving multidimensional intersections and significant thickness increases, such as wind turbine gearboxes. It transforms nonlinear fluid dynamics control into a spatial state estimation and model prediction problem. By eliminating the feedback lag of traditional liquid level control through cross-domain fusion, it can accurately control the instantaneous fluctuations of high-temperature fluid media approaching abrupt change regions under strong mechanical disturbances, thereby significantly improving the engineering quality of heavy equipment manufacturing. Attached Figure Description

[0041] The invention will now be further described with reference to the accompanying drawings.

[0042] Figure 1This is a control block diagram of the bottom-pouring constant pressure casting system for wind turbine housing castings provided in an embodiment of this application. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figure 1 A bottom-pouring constant pressure casting system for wind turbine housing castings includes a control terminal, which is communicatively connected to a target servo valve, a data acquisition module, a data preprocessing module, a state estimation module, a predictive control module, and a control data output module.

[0045] The data acquisition module is used to obtain real-time total weight data from the liquid storage container and real-time liquid level data from the buffer injection area.

[0046] The data preprocessing module is used to filter the real-time total weight data to obtain the mass loss rate, and convert the mass loss rate into the cumulative volume injected into the target molding cavity based on the preset fluid medium density.

[0047] The state estimation module is used to obtain a pre-constructed height-volume integral function data table that represents the mapping relationship between height and volume integral. Based on the cumulative volume and the height-volume integral function data table, the theoretical liquid level height is deduced. The extended Kalman filter operator is used to fuse the theoretical liquid level height with the real-time liquid level height data to calculate the real-time estimated flow velocity of the fluid medium.

[0048] The predictive control module is used to obtain the preset target filling and rising speed, extract the cross-sectional area step change characteristics of the target forming cavity based on the height volume integral function data table, and use the model predictive control algorithm to combine the cross-sectional area step change characteristics with the real-time estimated flow velocity to generate an absolute position command for adjusting the target servo valve.

[0049] The control data output module is used to encapsulate the absolute position command into a standard control data packet and update the fluid injection status flag in the system database.

[0050] This embodiment provides the basic operating mechanism of the bottom-pouring constant pressure casting system for wind turbine housing castings; specifically, in the bottom-pouring process of large wind turbine housing castings, the control terminal coordinates the data flow of each module; the data acquisition module synchronously acquires the total weight data of the liquid storage container and the liquid level height data of the buffer injection area in real time.

[0051] The data preprocessing module filters the high-frequency total weight data and combines it with the preset fluid medium density to convert it into cumulative volume; the state estimation module introduces a pre-compiled height-volume integral function data table, and combines the above cumulative volume to deduce the theoretical liquid level height without delay, and uses the extended Kalman filter operator to construct a simultaneous equation containing state variables.

[0052] By covariance fusion of theoretical liquid level height and real-time liquid level height data with physical lag, a high-precision real-time estimated flow velocity can be calculated. The predictive control module extracts the step change characteristics of the cross-sectional area of ​​the cavity and uses a model predictive control algorithm to generate absolute position commands for adjusting the target servo valve in advance.

[0053] The control data output module encapsulates and executes the data. If, during the multi-source fusion process, the real-time liquid level height data is found to have severe packet loss or distortion due to smoke and dust obstruction, resulting in a confidence level of zero, the system will temporarily degrade and rely solely on the theoretical liquid level height derived from mass loss to maintain open-loop fault-tolerant control until the laser observation data is restored.

[0054] For example, when casting a ductile iron wind turbine gearbox with a weight of up to 15 tons, the main control system executes the above-mentioned data flow closed loop from sensing to execution in a fixed cycle of 20 milliseconds. The purpose of this mechanism is to transform the complex nonlinear fluid dynamics control into a spatial state estimation problem based on a priori topological graph through isomorphic mapping of cross-domain algorithms, thereby eliminating the geometric blind zone and feedback lag inherent in the nonlinear cavity of traditional single closed-loop PID level control.

[0055] In a preferred embodiment of the present invention, the process by which the data acquisition module obtains real-time total weight data from the storage container and real-time liquid level data from the buffer injection area includes:

[0056] Real-time total weight data is obtained by weighing sensors placed in the liquid storage container; real-time liquid level data is obtained by laser rangefinders placed in the buffer injection area.

[0057] The data preprocessing module filters the real-time total weight data to obtain the mass loss rate, and converts the mass loss rate into the cumulative volume injected into the target molding cavity based on the preset fluid medium density. This process includes:

[0058] The real-time total weight data is denoised using a moving average filtering algorithm; the temporal variation characteristics of the denoised real-time total weight data are extracted to calculate the mass loss rate; the mass loss rate is integrated over time to calculate the cumulative mass; the cumulative mass is divided by the preset fluid medium density to calculate the cumulative volume.

[0059] This embodiment provides a noise reduction mechanism for multi-source sensor data acquisition and mass spatiotemporal conversion. In a real heavy casting industrial site, it is not feasible to directly derive the flow rate from the weight data by differentiation based solely on the basic framework of this embodiment. This is because the high-frequency time-series data of the ladle weighing sensor is often mixed with mechanical high-frequency noise caused by slight shaking of the crane and oscillation of the liquid surface. Direct differentiation would lead to extreme oscillation of the control quantity.

[0060] Therefore, in this embodiment, the data acquisition module obtains the current ladle total weight at a preset sampling frequency through a weighing sensor. Furthermore, a laser rangefinder is used to obtain the real-time liquid level height in the pouring cup at a higher frequency. The data preprocessing module uses a moving average filtering algorithm to process the total time series data. Perform noise reduction and smoothing processing;

[0061] By differentiating it, the temporal variation characteristics are extracted, and the stationary real-time quality loss rate is calculated. ;,Will The cumulative mass is obtained by integrating over time; finally, it is divided by the pre-defined density of molten iron at high temperature. The cumulative volume of the material injected into the target molding cavity is accurately calculated. ;

[0062] If during calculus calculations, it is found that... If a negative value or a positive jump that far exceeds the physical limit occurs, it is determined that the weighing sensor is subjected to transient strong mechanical interference from the crane movement. At this time, the algorithm will isolate the current jump value and use the average of the smooth slope of the previous three cycles for interpolation.

[0063] For example, during the high-flow-rate filling stage before the bottom injection of the wind turbine box, the system continuously records the stable decay curve of the ladle weight. After removing high-frequency vibrations, it accurately deduces how many cubic decimeters of high-temperature molten iron per second have been smoothly pressed into the bottom of the complex sand mold. The purpose of this step is to provide a high signal-to-noise ratio, physical hysteresis-free volume reference data stream for subsequent joint estimation of volume status.

[0064] In a preferred embodiment of the present invention, the process by which the state estimation module obtains a pre-constructed height-volume integral function data table representing the mapping relationship between height and volume integral includes: obtaining slice data of a three-dimensional computer-aided design model of the target forming cavity; extracting the cross-sectional area variation characteristics with height from the slice data of the three-dimensional computer-aided design model; reducing and compressing the three-dimensional physical shape into a one-dimensional spatial state diagram; and constructing a height-volume integral function data table representing the mapping relationship between height and volume integral based on the one-dimensional spatial state diagram.

[0065] The process by which the state estimation module deduces the theoretical liquid level height based on the cumulative volume and height-volume integral function data table includes: taking the cumulative volume as an input variable; performing a reverse lookup and matching in the height-volume integral function data table; and outputting the theoretical liquid level height inside the target forming cavity corresponding to the cumulative volume.

[0066] This embodiment provides a mechanism for dimensionality reduction mapping of complex three-dimensional cavities and inverse deduction of theoretical liquid levels. Based on the cumulative volume obtained by the aforementioned mass conversion, if the traditional fixed geometric cross-sectional area is still used to calculate the liquid level height, serious model errors will occur when facing nonlinear abrupt changes in volume ratio such as the flange or dense bearing seat inside the wind turbine housing.

[0067] Therefore, in this embodiment, the state estimation module acquires the slice data of the three-dimensional computer-aided design model of the target forming cavity during the offline preparation stage, and extracts the extremely critical feature of cross-sectional area changing with height. The algorithm reduces and compresses the massive three-dimensional physical shape into a lightweight one-dimensional spatial state diagram, using a high-capacity integral function search tree, and pre-constructs a system based on the formula:

[0068]

[0069] in, For height is Theoretical volume at time height variable The corresponding cross-sectional area; during online casting, the system directly uses the cumulative volume calculated in real time. As an input variable, a reverse lookup match is performed in the high-volume integral function data table with low computational power consumption;

[0070] Quickly output the theoretical liquid level height that should be inside the target molding cavity at this time. If the resin sand mold undergoes severe irreversible expansion due to high-temperature thermal effects during casting, meaning the actual volume exceeds the CAD model volume, resulting in the theoretical liquid level height obtained from the table being significantly higher than the steady-state range of actual physical laws, the system will trigger a priori model deviation warning and proactively adjust the warning in the next process furnace. Injection thermal expansion empirical compensation coefficient;

[0071] For example, when the system calculates that the cumulative volume has reached a certain threshold, it instantly retrieves the current position of the molten iron at a height of 1.2 meters by looking up a table. This position is clearly marked as the mid-section necking abrupt change zone of the box in the prior state diagram. The purpose of this mechanism is to use spatial isomorphic dimensionality reduction technology to achieve millisecond-level pure liquid level reference prediction in complex nonlinear cavities with extremely low controller computing power overhead.

[0072] In a preferred embodiment of the present invention, the process of fusing theoretical liquid level height and real-time liquid level height data using an extended Kalman filter operator to calculate the real-time estimated flow velocity of the fluid medium includes: constructing a state vector containing real-time height state and transient flow velocity state; extracting the cross-sectional area step change feature as a nonlinear driving term in the state transition matrix; setting the real-time liquid level height data as the observation variable; fusing the theoretical liquid level height and the actual flow resistance derived based on the observation variable using the covariance matrix; and outputting the denoised real-time estimated flow velocity.

[0073] This embodiment provides a two-dimensional joint estimation mechanism for flow velocity and liquid level based on extended Kalman filtering. Although the theoretical height is derived through the aforementioned prior model, it will inevitably accumulate steady-state physical errors over long-term integration. Furthermore, relying solely on a laser sensor to obtain the true height of the pouring cup results in a second-level physical damping delay in the cross-sectional area abrupt change region due to the fluid communication pipe effect.

[0074] To resolve this contradiction, in this embodiment, the state estimation module uses an extended Kalman filter to reconstruct the closed loop; the algorithm constructs a state including the real-time height of the cavity. and transient flow rate state of the ingate state vector :

[0075]

[0076] in, For the first A state vector at discrete moments. For the first Real-time height status of the cavity at discrete moments For the first The transient flow velocity state of the ingate at discrete moments Represents the transpose of a matrix or vector; during prediction updates, a nonlinear state transition equation based on fluid dynamics is constructed:

[0077]

[0078] in, It is a nonlinear state transition function. This is the process noise matrix; specifically, it represents the cross-sectional step change features extracted from the CAD model. Embedding, the discretized state transition equation is derived as follows:

[0079]

[0080]

[0081] in, For the fixed cross-sectional area of ​​the ingate, Sampling time, For fluid acceleration control; the corresponding state transition Jacobian matrix. This is a partial derivative matrix, which contains nonlinear driving terms reflecting the abrupt changes in cavity volume. During the observation update, the real-time liquid level data acquired by the laser rangefinder will be used. Define it as the observed variable and construct the observation equation:

[0082]

[0083] in, For real-time liquid level observations, the observation matrix is... This indicates that only the altitude status was directly observed. To measure the noise matrix, the system uses dynamically iterative Kalman gain and state covariance matrix during the algorithm initialization phase. Set as a diagonal matrix, its velocity variance term is calibrated based on the on-site vehicle vibration amplitude. to The measurement noise matrix The variance term is set according to the static calibration error of the laser rangefinder. ;

[0084] The theoretical liquid level height derived from weight integral is fused with real-time observed height data using optimal confidence level fusion, ultimately outputting a real-time estimated flow velocity that is neither lagging nor drifting. ;

[0085] If the covariance matrix exhibits a non-positive definite divergence tendency during iterative calculation due to strong external abrupt disturbances, leading to the failure of joint estimation, the system will trigger a safety lock, temporarily freezing the Kalman gain update and using a pure feedforward model for a short-term transition.

[0086] For example, when the molten iron is about to overflow the thick flange part inside the box, although the outer level gauge has not yet detected the change in resistance, the EKF fusion algorithm has already sensed it in advance through the nonlinear driving term of the covariance matrix and outputs an extremely smooth and zero-delay estimated flow velocity of the inner gate. The purpose of this step is to use modern state estimation theory to eliminate the inherent sensing delay defects in the region of drastic flow field change, and to provide the most realistic microscopic flow field feedback for subsequent constant pressure regulation.

[0087] In a preferred embodiment of the present invention, the process by which the predictive control module generates an absolute position command for adjusting the target valve includes: setting a constant filling speed target based on a preset target filling rise speed; predicting the predicted cross-sectional area of ​​the liquid surface location within a preset time period using a height-volume integral function data table; and calculating the rate of change of the predicted cross-sectional area.

[0088] The configuration is as follows: if the rate of change is greater than or equal to the preset area mutation threshold, then an absolute position command to reduce the target valve opening is generated at a preset early intervention time before reaching the area mutation region; otherwise, an absolute position command to maintain the current flow regulation state is generated based on the real-time estimated flow velocity.

[0089] This embodiment provides a model-predictive-based nonlinear spatial valve feedforward flexible control mechanism. After obtaining the real-time flow velocity without delay, if the conventional PID algorithm is still used for servo drive, due to the hydrodynamic inertia of the liquid-solid interface of the large casting, the system will still overshoot when facing extreme geometric bottlenecks, causing the target servo valve to oscillate frequently and triggering destructive pressure jets in the ingate.

[0090] Therefore, in this embodiment, the predictive control module abandons reactive control and instead adopts model predictive control based on a set constant filling velocity target; the algorithm utilizes an existing height volumetric integral function lookup table, combines it with the currently fused real-time estimated flow velocity for forward integration, and constructs a prediction range of... Predictive models that accurately predict the future Predicted cross-sectional area of ​​the position where the liquid surface front will arrive within one control cycle. ;

[0091] The core of the controller lies in solving the objective cost function through rolling optimization. Minimize the control sequence; construct a quadratic cost function that includes speed following error and control increment penalty:

[0092]

[0093] in, To predict flow velocity for the model, For a constant filling rate target, These are the error weighting coefficients. This refers to the incremental weighting coefficient for valve control. This represents the change in the opening degree of the servo valve. For the prediction range, For the current discrete control time, To predict the step size index, under actual operating conditions of strong mechanical interference in the wind turbine housing, the error weighting coefficient is... Valve control incremental weighting coefficient ratio Set at to The magnitude range, the prediction range Set to 15 to 25 control cycles;

[0094] Indicates based on the first The state at time t is related to the first Predicting the time, Indicates the first Discrete moments;

[0095] The logic is rigorously configured as follows: if a drastic contraction of the predicted cross-sectional area is predicted in the future and the rate of change is greater than or equal to a preset area mutation threshold, the MPC controller will solve for the minimum value of the aforementioned cost function while satisfying physical constraints. This will allow it to proactively output an optimal absolute position command for smoothly reducing the opening of the target servo valve a preset time before the liquid surface actually touches the area mutation region. ;

[0096] Otherwise, instructions to fine-tune or maintain the current opening are generated based only on the real-time estimated flow rate of the current fusion. If the algorithm predicts that the cross-sectional area will only exhibit small-range high-frequency sawtooth geometric fluctuations in the future time period, these minute changes will be shielded through the built-in control dead zone mechanism to avoid high-frequency oscillations that could damage the mechanical life of the servo cylinder.

[0097] For example, when the predictive control module detects that molten iron must be squeezed into the narrow support bearing hole wall within the next second, the target servo valve begins to slowly press down to cut off the flow 800 milliseconds in advance, smoothly unloading the flow channel pressure and effectively avoiding the transient high pressure burst when the liquid surface reaches the bottleneck. The purpose of this mechanism is to give the actuator the flexible adjustment capability of feedforward prediction through spatial feedforward prediction, which greatly reduces the frequency of actuator operation and eliminates the risk of sand flushing from the root.

[0098] In a preferred embodiment of the present invention, the fluid medium is high-temperature molten iron; the target forming cavity is the internal cavity of the sand casting mold for the wind turbine gearbox; the area of ​​abrupt change includes the planetary carrier support hole area or the flange area of ​​the gearbox.

[0099] This embodiment provides a specific application scenario and boundary constraint mechanism applicable to megawatt-level heavy-duty industrial equipment; with the underlying computing power support of all data preprocessing, Kalman flow field estimation and MPC prediction feedforward, this embodiment completely anchors the system to the most vulnerable real extreme working conditions: the controlled fluid medium specifically refers to high-temperature molten iron with viscosity changing drastically with temperature, and the target forming cavity is specifically defined as the internal cavity of the sand casting mold for wind turbine gearbox with an extremely complex structure;

[0100] The system focuses on hard-coding the area of ​​the planetary support hole of the box body with multi-dimensional spatial intersection and the area of ​​the flange with a sudden increase in thickness inside the casting as the core area of ​​abrupt change; when the pouring operation is performed, once the molten iron front approaches these complex sand core dense areas that are prone to turbulence, the system will schedule the predictive feedforward algorithm with the highest priority to strictly control the instantaneous fluctuation of the flow rate.

[0101] If a sudden, localized minor collapse of the sand core or blockage of the vent hole causes an abnormal surge in back pressure within the cavity while flowing through the flange area, the system will immediately exit the normal speed closed loop and trigger the underlying pressure-holding and flow-limiting safety procedure after monitoring the resistance change exceeding the set envelope. This will force the plug rod position to be locked at the minimum safe opening to prevent high-temperature molten metal from splashing back.

[0102] For example, in the batch manufacturing of 50 ductile iron wind turbine housings, each weighing 15 tons, when molten iron at a high temperature of up to 1350 degrees Celsius flowed through the layers of planetary carrier sand core apertures, the servo control system smoothly and minutely adjusted the stopcock, without triggering any destructive high-pressure jets throughout the process. The purpose of this mechanism is to deeply and safely embed a highly abstract adaptive constant pressure algorithm into the rigid manufacturing environment of heavy-duty wind turbine equipment, thereby reducing the area of ​​sand holes and slag inclusions recorded by under-sprue (UT) inspection near the ingate of the casting from the conventional average value. Significantly decreased to This represents a substantial leap forward in project quality.

[0103] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A bottom-pouring constant-pressure casting system for wind turbine housing castings, characterized in that, The system includes a control terminal, which is communicatively connected to a target servo valve, a data acquisition module, a data preprocessing module, a state estimation module, a predictive control module, and a control data output module. The data acquisition module is used to obtain real-time total weight data from the liquid storage container and real-time liquid level data from the buffer injection area. The data preprocessing module is used to filter the real-time total weight data to obtain the mass loss rate, and convert the mass loss rate into the cumulative volume injected into the target molding cavity based on the preset fluid medium density. The state estimation module is used to obtain a pre-constructed height-volume integral function data table that represents the mapping relationship between height and volume integral. Based on the cumulative volume and the height-volume integral function data table, the theoretical liquid level height is deduced. The extended Kalman filter operator is used to fuse the theoretical liquid level height with the real-time liquid level height data to calculate the real-time estimated flow velocity of the fluid medium. The predictive control module is used to obtain the preset target filling and rising speed, extract the cross-sectional area step change characteristics of the target forming cavity based on the height volume integral function data table, and use the model predictive control algorithm to combine the cross-sectional area step change characteristics with the real-time estimated flow velocity to generate an absolute position command for adjusting the target servo valve. The control data output module is used to encapsulate the absolute position command into a standard control data packet and update the fluid injection status flag in the system database.

2. The bottom-pouring constant-pressure casting system for wind turbine housing castings according to claim 1, characterized in that, The process by which the data acquisition module obtains the real-time total weight data from the liquid storage container and the real-time liquid level data from the buffer injection area includes: The real-time total weight data is obtained by a weighing sensor arranged in the liquid storage container; the real-time liquid level data is obtained by a laser rangefinder arranged in the buffer injection area.

3. The bottom-pouring constant-pressure casting system for wind turbine housing castings according to claim 2, characterized in that, The data preprocessing module filters the real-time total weight data to obtain the mass loss rate, and converts the mass loss rate into the cumulative volume injected into the target molding cavity based on the preset fluid medium density. The real-time total weight data is denoised using a moving average filtering algorithm. The quality loss rate is calculated by extracting the temporal variation features of the denoised real-time total weight data. The cumulative mass is obtained by integrating the mass loss rate over time; the cumulative volume is calculated by dividing the cumulative mass by the preset fluid medium density.

4. The bottom-pouring constant-pressure casting system for wind turbine housing castings according to claim 3, characterized in that, The process by which the state estimation module acquires the pre-constructed height-volume integral function data table representing the mapping relationship between height and volume integral includes: Obtain slice data of the three-dimensional computer-aided design model of the target molding cavity; Extract the cross-sectional area variation characteristics with height from the slice data of the three-dimensional computer-aided design model; The three-dimensional physical shape is reduced and compressed into a one-dimensional spatial state diagram; based on the one-dimensional spatial state diagram, a height-volume integral function data table representing the mapping relationship between height and volume integral is constructed.

5. The bottom-pouring constant-pressure casting system for wind turbine housing castings according to claim 4, characterized in that, The process by which the state estimation module deduces the theoretical liquid level height based on the cumulative volume and the height-volume integral function data table includes: using the cumulative volume as an input variable; Perform a reverse lookup match in the height-volume integral function data table; output the theoretical liquid level height inside the target forming cavity corresponding to the cumulative volume.

6. The bottom-pouring constant-pressure casting system for wind turbine housing castings according to claim 5, characterized in that, The process of fusing the theoretical liquid level height with the real-time liquid level height data using the extended Kalman filter operator to calculate the real-time estimated flow velocity of the fluid medium includes: Construct a state vector that includes real-time altitude state and transient flow velocity state; The cross-sectional area step change feature is extracted as a nonlinear driving term in the state transition matrix; the real-time liquid level height data is set as the observation variable; The theoretical liquid level height and the actual flow resistance derived from the observed variables are fused using the covariance matrix; the denoised real-time estimated flow velocity is then output.

7. The bottom-pouring constant-pressure casting system for wind turbine housing castings according to claim 6, characterized in that, The process by which the predictive control module generates the absolute position command for adjusting the target valve includes: A constant filling speed target is set based on the preset target filling rise speed; The predicted cross-sectional area of ​​the liquid surface within a preset time period is predicted using the height-volume integral function data table; the rate of change of the predicted cross-sectional area is calculated. The configuration is as follows: if the rate of change is greater than or equal to a preset area mutation threshold, then at a preset early intervention time before reaching the area mutation region, the absolute position command to reduce the opening of the target servo valve is generated; otherwise, the absolute position command to maintain the current flow regulation state is generated based on the real-time estimated flow velocity.

8. The bottom-pouring constant-pressure casting system for wind turbine housing castings according to claim 7, characterized in that, The fluid medium is high-temperature molten iron; The target forming cavity is the internal cavity of the sand casting mold for wind turbine gearbox; The areas of abrupt changes in area include the planetary carrier support hole area or the flange area.

Images