Upper-wearing high-temperature-resistant hard-seal ball valve and self-adaptive dynamic control system thereof

Abstract

This invention relates to the field of industrial automation and control technology, and discloses a "top-mounted high-temperature resistant hard-seal ball valve and its adaptive dynamic control system." This system aims to solve the technical problem of excessive wear or thermal jamming risk on the valve sealing surface caused by using a fixed drive torque. Its features include: a multi-point distributed sensing subsystem, a valve dynamic control unit, and an intelligent electric actuator. The valve dynamic control unit incorporates a hybrid torque prediction model based on physical information and data-driven principles. It predicts the optimal torque in real time using sensor data and drives the actuator to dynamically respond, and possesses online self-learning capabilities. This invention enables accurate prediction and dynamic adjustment of valve operating torque, effectively avoiding overstress damage, improving system adaptability and generalization ability, and extending valve service life.

Description

Technical Field

[0001] This invention belongs to the field of industrial automation and control technology, specifically relating to a top-mounted high-temperature resistant hard-seal ball valve and its adaptive dynamic control system. Background Technology

[0002] In modern industrial automation control systems, valves, as key opening, closing, and regulating components in fluid transport pipeline systems, are crucial to the stability of the entire process flow due to their reliability, safety, and service life. Especially in harsh conditions such as high temperature and high pressure or media containing solid particles in petrochemical, energy, and metallurgical industries, top-entry high-temperature hard-seal ball valves have gained widespread application due to their compact structure, excellent sealing performance, and ability to be maintained online. The core technical feature of this type of valve lies in its metal-to-metal hard-seal structure. Reliable line or surface contact is formed through the strong compression between the ball and the valve seat sealing surface, achieving effective isolation of the medium in high-temperature environments.

[0003] To ensure reliable closure and sealing of valves under all possible operating conditions, especially at design extreme temperatures, traditional control strategies generally employ a fixed torque output mode based on a "worst-case" design. Specifically, the output torque of the valve's actuator, whether electric or pneumatic, is typically set to a fixed value with a large safety margin before leaving the factory or during field commissioning. This setting is based on theoretical calculations or experimental measurements of the total torque required to overcome medium pressure, packing friction, and the maximum frictional torque between the ball and seat at the highest operating temperature and maximum operating pressure differential. Correspondingly, the control system simply issues "open" or "close" position commands to the actuator, which then completes the action with its preset constant torque, confirming the completion of the operation through mechanical limit switches or torque switches at the end of its stroke. In a specific historical period, this design philosophy was advantageous due to its simple logic, low implementation cost, and ability to ensure valve closure in most cases, meeting the core safety requirements of industrial production.

[0004] However, as related technologies place more refined demands on equipment operating efficiency, service life, and intelligence levels, the inherent limitations of the aforementioned control philosophy—sacrificing redundancy for universal reliability—are becoming increasingly apparent. The root cause lies in a profound and dynamic contradiction between this control strategy and the inherent physical characteristics of high-temperature hard-seal ball valves. The core of this contradiction is that the actual driving torque required for valve operation is a dynamic variable that changes drastically with temperature, while traditional control systems provide static, fixed driving capabilities. Specifically, the metal materials constituting the ball and seat exhibit significant thermal expansion and contraction. When the valve is started from room temperature or operates under low-temperature conditions, the material contracts, resulting in relatively small preload and contact pressure between the ball and seat, and consequently, a lower required frictional torque. Under these circumstances, the excessive torque output by the actuator, set based on high-temperature conditions, will generate far greater contact stress on the sealing surface than necessary. This will not only drastically accelerate abrasive wear and scratches on the sealing surfaces, significantly shortening the valve's service life and maintenance cycle, but also cause significant waste of driving energy and unnecessary impact on the transmission gear system. Conversely, when the system experiences abnormal overheating, or during the heating phase of certain processes, if the valve body temperature exceeds the reference temperature point set for torque, excessive material expansion will cause a sudden increase in the normal pressure and frictional torque between the ball and the valve seat. At this time, the previously seemingly sufficient preset torque may become insufficient to overcome the sharply increased operating resistance, leading to frequent activation of the actuator torque protection, failure to drive the valve to the predetermined position, and even, in extreme cases, thermal jamming where the ball and valve seat completely "lock up," posing a serious production safety hazard.

[0005] This open-loop control method based on fixed torque essentially treats the valve as a black box with constant physical properties, completely ignoring the dynamic influence of temperature, a core variable, on its internal mechanical state. As a result, the valve's operating state constantly oscillates between the two undesirable extremes of "excessive wear" and "risk of jamming" across a wide temperature range, failing to achieve an ideal balance that considers reliability, economy, and long lifespan.

[0006] Therefore, how to break through the inherent bottleneck of the traditional fixed torque control mode and establish an intelligent control method that can accurately respond to and even predict the actual torque demand of the valve under different thermal conditions, and accordingly make adaptive dynamic adjustments to the driving force, so as to achieve full-condition optimization of valve operation, and minimize the wear of the sealing surface, reduce energy consumption and avoid the risk of overheating and jamming while ensuring sealing reliability, has become a key challenge and a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide an adaptive dynamic control system for a top-mounted high-temperature hard-seal ball valve, in order to solve the technical problems in the prior art that lead to excessive wear of the valve sealing surface or the risk of thermal jamming under different temperature conditions due to the use of a fixed drive torque.

[0008] To achieve the aforementioned objectives, the present invention provides an adaptive dynamic control system for a top-mounted high-temperature resistant hard-seal ball valve. This system is systematically constructed to achieve accurate prediction and closed-loop dynamic adjustment of the valve's operating torque. The system includes: a multi-point distributed sensing subsystem, a valve dynamic control unit serving as the core of system computation and decision-making, and an intelligent electric actuator with dynamic torque response capabilities. The multi-point distributed sensing subsystem functions to acquire key parameters characterizing the valve's current physical state in real-time and with high precision. The valve dynamic control unit internally incorporates a hybrid torque prediction model based on physical information and data-driven principles, and establishes a bidirectional data communication connection with the multi-point distributed sensing subsystem and the intelligent electric actuator. This unit receives external operating commands and, based on real-time data provided by the sensing subsystem, calculates the optimal torque value required to execute the command using the prediction model. The intelligent electric actuator differs from traditional actuators in that it not only receives position control commands but also receives and accurately executes dynamic torque commands issued by the valve dynamic control unit, thereby completing the valve's opening or closing action with the most suitable force.

[0009] Furthermore, the multi-point distributed sensing subsystem consists of a set of precisely configured temperature sensors, pressure sensors, and position sensors. Specifically, the temperature sensors employ three Class A precision armored platinum resistance temperature sensors, with platinum as the resistive element material and a nominal resistance of 1000 ohms at 0 degrees Celsius, encapsulated within a 3 mm diameter stainless steel protective tube. The first temperature sensor is tightly attached to a preset temperature measurement point on the outer wall of the valve seat upstream of the valve body using a highly thermally conductive metal adhesive or mechanical clamping. The second temperature sensor is similarly positioned at the corresponding temperature measurement point on the outer wall of the valve seat downstream of the valve body. The third temperature sensor is installed at a specific location on the valve cover or middle flange. This distributed layout aims to accurately capture the non-uniform temperature field generated by the valve due to media flow and environmental heat exchange. By weighted averaging of the multi-point temperature data, an effective temperature value that more accurately reflects the thermal expansion state of the valve ball and valve seat is obtained. The pressure sensor employs a piezoresistive pressure transmitter, installed at the pressure taps on the upstream and downstream pipelines of the valve. Its range covers the highest operating pressure designed for the process, outputting a 4-mA standard current signal or a digital bus signal for real-time monitoring of the pressure difference (ΔP) across the valve. This pressure difference is a key input variable for calculating the frictional torque caused by the thrust of the medium. The position sensor is a multi-turn absolute encoder integrated into the output shaft of the intelligent electric actuator, capable of continuously providing precise angular data of the valve from fully closed to fully open, with a resolution better than 0.01 degrees.

[0010] As the core of this invention, the valve dynamic control unit is an embedded industrial computing platform specifically designed for this system. At the hardware architecture level, the main controller of the valve dynamic control unit adopts a 64-bit quad-core industrial-grade system-on-a-chip based on the ARM Cortex-A72 architecture, with a main frequency of no less than 1.5GHz to ensure the computational performance of complex algorithms. This industrial-grade system-on-a-chip integrates a 4GB LPDDR4 dynamic random access memory to support the operation of the real-time operating system and applications; and a 32GB eMMC non-volatile flash memory for storing the operating system, applications, torque prediction model, and long-term operating data. Crucially, the circuit board of the valve dynamic control unit integrates a dedicated neural processing unit with an integer arithmetic capability of no less than 4 TOPS, specifically responsible for hardware acceleration of the parallel computation of the torque prediction model, thereby controlling the inference time of a single torque prediction to within 5 milliseconds, meeting the real-time requirements of industrial control. The valve dynamic control unit also integrates a communication interface supporting the industrial Ethernet protocol.

[0011] Furthermore, the valve dynamic control unit internally stores and runs a physical information neural network torque prediction model that has been pre-trained offline and possesses online self-learning capabilities. This model is a deep feedforward neural network, whose structure includes an input layer, six hidden layers, and an output layer.

[0012] The input layer of the model receives a state vector consisting of five key physical quantities, specifically including: the weighted average valve body temperature calculated from multi-point temperature sensor data. The valve inlet and outlet pressure difference ΔP measured by the pressure sensor; the valve current angular position provided by the position sensor. The target angular position issued by the upper-level control system and internal variables characterizing the current wear state of the valve. The model has six hidden layers, each containing eight neurons, using a modified linear unit (MRU) activation function. The output layer is a single neuron that outputs a scalar value, i.e., the predicted value from the previous input. arrive Optimal instantaneous drive torque required for motion .

[0013] As a key technical feature of this invention, the construction and training process of the physical information neural network torque prediction model integrates physical laws and data-driven methods. In the offline training phase, a digital twin model of the target valve is first established using high-fidelity finite element analysis software. This digital twin model precisely defines the geometric dimensions and material properties of core components such as the valve ball and valve seat, including their linear thermal expansion coefficient, elastic modulus, Poisson's ratio, and the temperature-dependent friction coefficient between the sealing surfaces based on Stellite hard alloy. By applying temperature loads (-℃ to +6℃) and pressure loads (0 to 1.1 times the design pressure) covering the entire design operating range of the valve to this digital twin model, thousands of transient simulations of thermo-mechanical coupling are performed, thereby generating a model containing (… ,ΔP,θ, A large-scale, high-precision dataset of correspondences. Subsequently, when training a physical information neural network model, its loss function... Defined as data fitting loss With physical constraint loss Weighted sum: .in, This is the mean square error between the model-predicted torque and the actual torque in the simulation dataset. It is a regularization term that directly encodes the known physical equation into the loss function, which describes the relationship between the total driving torque τ and the individual torque components: Among them, the frictional torque of the sealing surface It is a function of temperature T and contact normal force N, and the contact normal force N itself is the result of the combined effect of material thermal expansion and medium pressure. This is achieved by minimizing [the contact normal force] during training. This forces the output of the neural network not only to fit the simulation data, but also to physically follow the basic laws of thermoelasticity and contact friction, thereby greatly improving the model's generalization ability and prediction accuracy under unseen working conditions.

[0014] As another key technical feature of this invention, the valve dynamic control unit also integrates an online self-learning and model optimization module to enable the system to adapt to performance changes throughout the valve's entire lifecycle. The core function of this module is to continuously and non-invasively collect real-world operational data during actual valve operation and use this data to periodically fine-tune the pre-trained physical information neural network model. Specifically, the valve dynamic control unit issues torque commands to the intelligent electric actuator... At the same time, the q-axis current component of the permanent magnet synchronous motor will be read back in real time from the internal driver of the actuator. Because under the field-oriented control strategy, the electromagnetic torque of the motor has a strictly linear relationship with the q-axis current ( ,in (where the motor torque constant is), therefore It can be accurately converted into the actual output torque of the motor. .Should After deducting the fixed friction losses of the internal transmission system of the actuator, the actual driving torque acting on the valve stem is obtained. The online self-learning module records each successful operation ( ,ΔP,θ, This data is stored as a new, high-quality data point in the non-volatile memory of the valve dynamic control unit. The data is accumulated until a preset number of data points are reached, or when the predicted torque is detected. Compared with actual torque When the average deviation continuously exceeds a certain threshold, this module will be automatically triggered. After triggering, it will utilize this newly acquired real data and employ a transfer learning strategy to incrementally train the last few hidden layers of the existing physical information neural network model. This process enables the model to effectively capture the valve torque characteristic drift caused by factors such as long-term wear of the sealing surface, packing aging, and changes in media characteristics, thereby dynamically updating the wear state variables. And maintain the long-term accuracy of torque prediction.

[0015] Furthermore, the intelligent electric actuator is a highly integrated unit that combines power, transmission, drive, and local control. Its power source is a high-efficiency, high-torque-density permanent magnet synchronous motor. The transmission mechanism is a multi-stage planetary gear reducer to achieve a large speed ratio and high torque output. Its core lies in its internal servo driver, which is based on Texas Instruments' TMS3F2837xD series dual-core C00 microcontroller and runs a sophisticated field-oriented control algorithm. This driver, through its communication interface, can receive composite commands from the valve's dynamic control unit, which include not only the target position... It also includes dynamic torque limit values ​​calculated by a physical information neural network model. When performing valve opening or closing actions, the servo drive uses a three-loop closed-loop control system: a position loop as the outer loop, a speed loop as the middle loop, and a current loop as the inner loop. Its unique feature lies in the q-axis current command of the current loop. The upper limit is no longer a fixed factory setting, but is dynamically set based on the received torque command. Converted This means that the maximum output torque of the actuator at any given time is precisely limited to a "just enough" level by the valve dynamic control unit, thereby minimizing overstress damage to the valve sealing surface while ensuring the completion of the action.

[0016] To ensure the real-time, deterministic, and synchronous data interaction between the valve dynamic control unit, the multi-point sensing subsystem, and the intelligent electric actuator, all units within the system are connected via an industrial Ethernet bus. The bus uses the EtherCAT protocol. In this network topology, the valve dynamic control unit is configured as the sole EtherCAT master, responsible for initiating all communication cycles and managing the distributed clock synchronization of the entire network. The PT1000 temperature sensor, piezoresistive pressure transmitter, and intelligent electric actuator all have built-in interface chips supporting the EtherCAT slave protocol, connecting to the network as slave devices. Through EtherCAT's precise clock synchronization mechanism, the valve dynamic control unit can synchronously acquire snapshots of physical quantities from all sensors at the same moment within each communication cycle (set to 1 millisecond) and issue precisely timed control commands to the actuator. This high-speed bus architecture based on hardware synchronization completely eliminates control errors caused by communication delays and data asynchrony in traditional control systems, providing a solid foundation for achieving high dynamic response torque closed-loop control.

[0017] This invention also provides an adaptive dynamic control method for a top-mounted high-temperature resistant hard-seal ball valve based on the above-mentioned system. After system power-on initialization, this method is primarily executed by the valve dynamic control unit. First, the valve dynamic control unit, acting as an EtherCAT master, scans the bus and establishes communication connections with all slave devices (sensors and actuators), synchronizes the network clock, and loads the physical information neural network torque prediction model stored in its internal flash memory to the neural processing unit acceleration unit. In standby mode, the valve dynamic control unit continuously polls each sensor at a frequency of 10 Hz, monitoring the valve's temperature field and pressure difference across its terminals in real time, and updating the system's current state. When the valve dynamic control unit's control port receives an operation command from the upper-level distributed control system or the local human-machine interface, the control flow is triggered. The valve dynamic control unit immediately reads the latest weighted average valve body temperature. The pressure difference ΔP, combined with the current position in the command. With the target location and internally maintained wear state variables Together, these constitute a five-dimensional input vector. This input vector is fed into the physical information neural network model running in the neural processing unit for a forward inference calculation, which calculates the optimal torque prediction value required to complete this journey within milliseconds. Subsequently, the valve dynamic control unit uses this torque prediction value. With the target location The instruction is encapsulated into an EtherCAT process data object and sent to the intelligent electric actuator in the next communication cycle. Upon receiving the instruction, the actuator's servo driver immediately... The current loop is set to its dynamic upper limit, and the motor is started to move towards the target position. Throughout the movement, the actuator continuously adjusts its internal q-axis actual current. The high-precision encoder position data is fed back to the valve dynamic control unit in real time via the EtherCAT bus. The valve dynamic control unit then compares this feedback data with the corresponding operating parameters (…). The data (ΔP) is recorded together with the data and used as the raw data for the online self-learning module. After the operation is completed, the system returns to standby monitoring state. In the background of the system operation, the online self-learning module of the valve dynamic control unit uses accumulated real operating data to iteratively optimize the physical information neural network model according to preset trigger conditions, thereby realizing the continuous self-improvement of system performance and precise adaptation to individual valve differences and aging processes.

[0018] This invention discloses a top-mounted high-temperature resistant hard-seal ball valve, including the above-mentioned adaptive dynamic control system for the top-mounted high-temperature resistant hard-seal ball valve.

[0019] Furthermore, it also includes a valve body, in which valve seats are installed, a ball is pressed between two valve seats, a valve cover for fixing the ball is installed on the valve body, and a valve stem that cooperates with the ball is provided in the middle of the valve cover;

[0020] It also includes: two sets of sealing kits, one set of sealing kits is located at the mounting point between the valve body and the valve seat in the sealing kit, the mounting point between the valve body and the valve seat is connected to the sealing kit by a cylindrical helical compression spring, and the other set of sealing kits is located between the valve stem and the valve cover, the sealing kit includes a copper sleeve and a gasket, and a heat insulation sleeve that mates with the copper sleeve is also provided between the valve stem and the valve cover;

[0021] The valve cover is provided with heat dissipation fins, the end of the valve stem is fitted with a turbine housing, and packing is provided at the connection between the valve stem and the valve cover. Attached Figure Description

[0022] Figure 1 This is a block diagram of the overall structure of the system of the present invention. Figure 2 This is a schematic diagram of the multi-point distributed sensing subsystem layout on the valve body in this invention. Figure 3 This is a hardware architecture block diagram of the valve dynamic control unit in this invention. Figure 4 This is a schematic diagram of the physical information neural network torque prediction model in this invention. Figure 5 This is a flowchart illustrating the adaptive dynamic control method provided by the present invention.

[0023] Figure 6 This is a cross-sectional view of the top-mounted high-temperature resistant hard-seal ball valve of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0025] Reference Figure 1 This invention discloses an adaptive dynamic control system for a top-loading high-temperature resistant hard-seal ball valve. The system aims to provide intelligent and precise torque control for top-loading hard-seal ball valves operating under high-temperature and high-pressure differential conditions. The entire system is constructed as a closed-loop control architecture at both the logical and physical levels. Its core components include a multi-point distributed sensing subsystem for real-time sensing of the valve's physical state, a valve dynamic control unit serving as the system's decision-making center, and an intelligent electric actuator capable of executing dynamic torque commands. These three core subsystems achieve high-speed, synchronous data interaction via an industrial Ethernet bus, forming an organic whole that works together to control the top-loading high-temperature resistant hard-seal ball valve, achieving adaptive dynamic control throughout its entire lifecycle and across all operating conditions.

[0026] Specifically, the function of the multi-point distributed sensing subsystem is to comprehensively and in real-time measure the key physical parameters affecting the valve's operating torque. In a specific embodiment, referring to... Figure 2 This sensing subsystem is deployed at a critical thermodynamic node of the ball valve. It consists of a set of temperature sensors, pressure sensors, and position sensors. The placement of the temperature sensors takes into full account the non-uniform temperature field generated when the valve is traversed by a high-temperature medium. The first temperature sensor is installed on the outer wall of the valve seat on the upstream side of the valve body, the second temperature sensor is installed on the corresponding area on the outer wall of the valve seat on the downstream side of the valve body, and the third temperature sensor is installed on the upper structure of the valve, such as the valve cover or the middle flange. This distributed layout aims to capture the actual temperature of various valve components, especially the thermally expanding components that directly affect the sealing clearance, under the combined effects of medium heat transfer and environmental heat exchange.

[0027] In a preferred embodiment, the first, second, and third temperature sensors are all Class A precision armored platinum resistance temperature sensors. Their core sensing element is a PT1000 type platinum resistance thermometer, meaning its nominal resistance is 1000 ohms at 0 degrees Celsius. The sensor is encapsulated in a 3mm diameter protective sleeve made of 316L stainless steel to resist vibration and corrosion in industrial environments. To ensure accuracy and response speed, the sensor probe is physically bonded to the temperature sensing points on the valve body, valve seat, and valve cover using a highly thermally conductive metal-based adhesive or a specially designed mechanical clamping device. The collected multi-point temperature data is transmitted to the valve dynamic control unit via an industrial Ethernet bus. The dynamic control unit's internal algorithm performs weighted averaging to obtain an effective temperature value that comprehensively reflects the thermal expansion state of the sealing surfaces. .

[0028] The pressure sensors are industrial-grade piezoresistive pressure transmitters, configured to cover 1.1 to 1.2 times the valve's design operating pressure to ensure accurate measurement even under maximum differential pressure conditions. They are installed on process piping pressure taps immediately upstream and downstream of the valve to continuously monitor the pressure before and after the valve, converting the pressure signal into a standard 4-mA current signal or directly outputting a digital signal via a module with an integrated EtherCAT interface. The valve dynamic control unit receives these two pressure values ​​and calculates the real-time inlet and outlet pressure difference ΔP, a key input parameter for calculating the frictional torque generated by the media thrust on the valve ball.

[0029] The position sensor is a high-precision absolute encoder integrated into the output shaft of the intelligent electric actuator. In a specific embodiment, this encoder is a multi-turn absolute encoder with bit resolution, capable of providing continuous angular position information of the valve from 0 degrees (fully closed) to 90 degrees (fully open), with a theoretical resolution better than 0.01 degrees. This encoder can retain its position after a power outage, eliminating the need for a zero-return operation upon power restoration. Its high-precision position data is not only used for conventional position control but also serves as a crucial input to torque prediction models, accurately describing the instantaneous state of the valve during its opening or closing stroke.

[0030] As the core of the technical solution of this invention, the valve dynamic control unit is an embedded computing platform specifically designed for this system, possessing powerful computing capabilities and industrial-grade reliability. (Refer to...) Figure 3 Its hardware architecture has been specifically designed. The main controller uses a 64-bit quad-core industrial-grade system-on-a-chip based on the ARM Cortex-A72 architecture, with its operating frequency set at no less than 1.5GHz, ensuring sufficient computational margin for running the real-time operating system and complex torque prediction algorithms. To support the high-speed operation of the main controller, the circuit board has at least 4GB of LPDDR4 dynamic random access memory for application runtime, and a 32GB eMMC non-volatile flash memory for storing the embedded Linux operating system, applications, the core torque prediction model, and long-term storage of historical valve operation data.

[0031] Crucially, the valve dynamic control unit integrates a dedicated neural processing unit within its hardware architecture. This neural processing unit is a hardware accelerator designed for efficient neural network inference calculations, with an integer processing capability of at least 4 TOPS. The introduction of this neural processing unit allows the forward inference process of the torque prediction model, which previously required significant CPU cycles on the main controller, to be completely offloaded to this dedicated hardware for parallel processing. Through this hardware acceleration, the calculation time for a single torque prediction is strictly controlled to within 5 milliseconds, a performance metric essential for meeting the stringent real-time requirements of industrial control. Furthermore, the valve dynamic control unit integrates a communication interface 25 supporting the industrial Ethernet protocol; in this embodiment, this interface is configured to support EtherCAT master station functionality. The entire valve dynamic control unit is encapsulated in a fanless aluminum alloy heat sink housing suitable for DIN rail mounting within an industrial control cabinet, providing excellent electromagnetic compatibility and environmental adaptability.

[0032] Furthermore, within the valve dynamic control unit, a physical information neural network torque prediction model, serving as the basis for intelligent system decision-making, is embedded and operates. This model is key to achieving adaptive control in this invention. (Refer to...) Figure 4 The model is structurally a deep feedforward neural network. Its input layer is designed to receive a state vector consisting of five key physical quantities. Specifically, this vector includes: the weighted average effective valve body temperature calculated from multi-point temperature sensor data. The valve inlet and outlet pressure difference ΔP is measured by the pressure sensor, and the valve current angular position is provided by the position sensor. The target angular position is issued by the upper-level control system or set within the valve dynamic control unit. And internal variables characterizing the current cumulative wear and aging state of the valve. .

[0033] Following the input layer, this physical information neural network model contains six hidden layers, each consisting of eight neurons. Each neuron uses a modified linear unit (MRU) as its activation function, widely used in deep learning models due to its computational simplicity and effectiveness in mitigating the vanishing gradient problem. After nonlinear transformations and feature extraction through the six hidden layers, the model's output layer is a single neuron that directly outputs a scalar value without an activation function. This output value represents the model's response to the current state, indicating how to drive the valve from its current position. Move to the target position The predicted value of the required optimal instantaneous drive torque is denoted as .

[0034] As a key technical feature of this invention, the construction and training process of the physical information neural network torque prediction model deeply integrates first principles of physics and data-driven methods, thereby ensuring its prediction accuracy and generalization ability. In the offline training phase, a high-fidelity digital twin model of the target ball valve is first established using professional finite element analysis software. This model accurately reproduces the geometric structure of core components such as the valve ball, valve seat, and valve stem, and assigns them detailed material properties. These properties include not only conventional elastic modulus and Poisson's ratio, but also, in particular, the linear thermal expansion coefficient that varies with temperature and the coefficient of friction of the sealing surface material at different temperatures.

[0035] Based on this digital twin model, large-scale thermo-mechanical coupled transient simulations were conducted. The simulation conditions covered the entire operating range of the valve design; for example, temperature loads varied from a low temperature of -℃ to an ultra-high temperature of +6℃, and pressure loads varied from 0MPa to 1.1 times the design pressure. By simulating the valve opening and closing processes under thousands of different temperature and pressure differential combinations, a large-scale and accurate training dataset was generated, in which each data point contains (…). ,ΔP,θ, This is a set of correspondences, namely, the precise torque required to complete a small displacement under specific temperature, pressure difference, and position.

[0036] When training a physical information neural network model using this dataset, its loss function Innovatively designed as data fitting loss With physical constraint loss The weighted sum, i.e. .in, This is a standard loss term, using mean squared error (MSE) to measure the model's predicted torque. Compared with the actual torque in the simulation dataset The differences between them. This is the regularization term, which directly encodes the known physical equation describing the valve torque into the loss function. The physical equation is: This equation shows that the total driving torque τ is the sum of four torques: the friction torque of the sealing surface. Packing friction torque bearing friction torque and hydrodynamic torque Among them, the most critical and drastically changing is the frictional torque of the sealing surface. The contact pressure N is a complex function of temperature T and the contact normal force N; the contact normal force N itself is the result of the combined effects of material thermal expansion (related to temperature T) and medium pressure (related to pressure difference ΔP). During training, the optimizer aims to minimize the total loss. This means that the neural network is not only driven to fit existing simulation data points ( The function of (the physical law), and its output must also mathematically comply with the constraints of the aforementioned physical laws (the physical law). The weighting factor λ (e.g., set to 0.1) is used to balance the relative importance of these two factors. This training method enables the model to learn the underlying physical laws behind the data, thereby greatly improving its prediction accuracy and physical interpretability under conditions not seen in the training dataset.

[0037] As another key technical feature of this invention, to address the performance drift of valves caused by wear, aging, and other factors during long-term operation, the valve dynamic control unit also integrates an online self-learning and model optimization module. This module endows the system with adaptive capabilities throughout its entire lifecycle. Its core function is to continuously and non-invasively collect real-world operational data during the actual operation of the valve, and use this data to periodically fine-tune the pre-trained physical information neural network model online.

[0038] The specific implementation method is as follows: When the valve dynamic control unit sends a torque command to the intelligent electric actuator... Then, it reads back in real time the actual q-axis current component of its drive motor (usually a permanent magnet synchronous motor) from the servo driver inside the actuator via the EtherCAT bus. Under the field-oriented control strategy, the electromagnetic torque of the motor... With q-axis current There is a strict linear relationship between them: ,in This is the motor's torque constant, a parameter precisely calibrated at the motor's factory. Therefore, by reading... The valve dynamic control unit can very accurately calculate the actual output torque of the motor. After deducting the fixed friction loss of the internal transmission system (such as the gearbox) of the actuator (which can be pre-calibrated and stored in a lookup table), the true driving torque acting on the valve stem can be obtained. .

[0039] The online self-learning module records each successful valve operation, forming a record containing ( ,ΔP,θ, The module collects new, high-quality real data points and stores them in the non-volatile flash memory of the valve dynamic control unit. The module's trigger conditions are preset to two types: one is data accumulation triggering, such as when the number of newly collected real data points reaches a preset number (e.g., 1000); the other is performance deviation triggering, such as when the model's predicted torque is detected. Compared with actual torque When the average relative error exceeds a certain threshold (e.g., 5%) in consecutive operations.

[0040] Once triggered, the online self-learning module will initiate the model optimization process. It will utilize the newly collected real-world data and employ transfer learning strategies to incrementally train the existing physical information neural network model. Specifically, to preserve the fundamental physical laws already learned by the model and improve training efficiency, this process "freezes" the weights of the first few hidden layers, only fine-tuning the weights of the last few hidden layers and the output layer. This process effectively allows the model to capture valve torque characteristic drift caused by factors such as long-term wear and tear of the sealing surface, aging of the valve stem packing, scaling of the medium, or changes in its characteristics. Simultaneously, this process dynamically updates the internal wear state variables. This allows the valve's health status to be quantitatively reflected, thus taking wear effects into account in subsequent predictions and maintaining the long-term accuracy of torque prediction.

[0041] Furthermore, the intelligent electric actuator is a highly integrated mechatronic unit. Its power source employs a high-efficiency, high-torque-density permanent magnet synchronous motor to provide sufficient power within a compact size. Its transmission mechanism is a multi-stage planetary gear reducer with a high reduction ratio, such as up to 10:1, used to convert the motor's high speed and low torque into the low speed and high torque required to drive the valve stem. Its control core is an internally integrated servo driver, whose main control chip uses Texas Instruments' TMS3F2837xD series dual-core C00 microcontroller. This chip possesses powerful floating-point arithmetic capabilities and rich control peripherals, specifically designed for running sophisticated motor control algorithms, such as field-oriented control.

[0042] This servo drive, through its integrated EtherCAT slave communication interface, is able to receive complex commands from the valve dynamic control unit. These commands include not only the target position required by traditional actuators. More importantly, it also includes dynamic torque limit values ​​calculated in real time by a physical information neural network model. When performing valve opening or closing actions, the control algorithm inside the servo drive forms a three-loop closed-loop control system with the position loop as the outermost loop, the speed loop as the middle loop, and the current loop as the innermost loop. Its fundamental difference from traditional actuators lies in the q-axis current command of the current loop. The upper limit value is no longer a fixed mechanical or electronic torque switch value set during factory commissioning, but is dynamically and in real-time set based on the received torque command. Through the motor torque constant The converted current limit value This means that at any moment during the operation of the actuator, its maximum output torque is precisely limited by the valve dynamic control unit to a level that is "just enough" to complete the current task. This ensures that the valve can operate reliably under various operating conditions and fundamentally avoids unnecessary overstress impact and accelerated wear on the hard alloy sealing surface caused by applying torque far exceeding the actual requirement.

[0043] To ensure the real-time, deterministic, and synchronous nature of massive data interaction between the valve dynamic control unit, the multi-point sensing subsystem, and the intelligent electric actuator, all units within the system are connected via an industrial Ethernet bus using the EtherCAT protocol. In this network topology, the valve dynamic control unit is configured as the sole EtherCAT master, responsible for initiating and managing all network communication cycles. The PT1000 temperature sensor, piezoresistive pressure transmitter, and intelligent electric actuator all have built-in dedicated interface chips supporting the EtherCAT slave protocol (e.g., Beckhoff's ET00 or ET10 series ASICs) to access the bus as slave devices.

[0044] The valve dynamic control unit utilizes the distributed clock mechanism built into the EtherCAT protocol to synchronize the clocks of all slave stations in the network with nanosecond-level precision. Based on this, the valve dynamic control unit operates with extremely short communication cycles (e.g., set to 1 millisecond). Within each communication cycle, the valve dynamic control unit synchronously acquires snapshots of physical quantities measured by all sensors at exactly the same moment and precisely and timely sends control messages containing torque and position commands to the actuator. This high-speed bus architecture based on hardware clock synchronization fundamentally eliminates the control errors and time lags caused by software polling, communication protocol stack delays, and data asynchrony in traditional control systems, providing a solid and reliable communication foundation for achieving high dynamic response torque closed-loop control.

[0045] This invention also provides an adaptive dynamic control method for a top-mounted high-temperature resistant hard-seal ball valve based on the above system. (Refer to...) Figure 5 The specific execution flow of this method is as follows: During the system power-on initialization phase, the valve dynamic control unit, acting as the EtherCAT master, first performs a bus scan to identify all connected slave devices in the network and establish communication connections with them. Subsequently, it starts a distributed clock synchronization program to unify the system time of the entire network. Simultaneously, the valve dynamic control unit loads its physical information neural network torque prediction model, which is stored in non-volatile flash memory, into the memory of the neural processing unit, preparing for hardware-accelerated computation.

[0046] In standby mode, the valve dynamic control unit continuously polls the multi-point distributed sensing subsystem at a low frequency (e.g., 10 Hz) to acquire and update the effective temperature of the valve body in real time. The static physical environment of the valve is monitored by the pressure difference ΔP between the two ends.

[0047] When the control port of the valve dynamic control unit (which can be an industrial Ethernet interface or a local human-machine interface) receives an operation command from the superior distributed control system or the operator (for example, the command is "open the valve from the current 0-degree fully closed position to the 90-degree fully open position"), the control process is officially triggered.

[0048] Upon triggering, the valve dynamic control unit immediately reads the latest weighted average effective valve body temperature. The inlet and outlet pressure difference ΔP, combined with the current position included in the operating instructions. =0° and target position =90°, and the current valve wear state variable read from its own memory. Together, they form a five-dimensional input vector.

[0049] The input vector is immediately fed into the physical information neural network model running in the neural processing unit for a forward inference calculation. Thanks to the hardware acceleration capabilities of the neural processing unit, the model calculates the optimal torque prediction value required to complete this valve opening stroke in less than 5 milliseconds. .

[0050] Subsequently, the valve dynamic control unit uses this torque prediction value With the target location =90° encapsulation into an EtherCAT process data object, which is then sent to the intelligent electric actuator via the industrial Ethernet bus within the next 1 millisecond communication cycle.

[0051] Upon receiving the command, the servo drive of the actuator immediately parses out the torque limit value. This is then set as the dynamic upper limit of its internal current loop. Subsequently, the driver controls the permanent magnet synchronous motor to start, driving the valve to the target position of 90 degrees. Throughout this movement, the actuator continuously applies its internal q-axis actual current. The high-precision encoder position data is encapsulated in the PDO and fed back to the valve dynamic control unit in real time via the EtherCAT bus.

[0052] The valve dynamic control unit records and processes this feedback data in the background, converting each set of valid operating parameters ( ,ΔP) and the corresponding actual torque It is stored together with the position θ data as the raw data material for the online self-learning module.

[0053] Once the valve reaches the target position and the operation is complete, the system returns to standby monitoring mode. Meanwhile, in the background of the system operation, the online self-learning module of the valve dynamic control unit is activated at appropriate times based on preset trigger conditions (accumulated data or performance deviation). It uses accumulated real-world operating data to iteratively optimize the physical information neural network model, thereby achieving continuous self-improvement of system performance and precise adaptation to individual valve differences and aging processes.

[0054] Please refer to Figure 6 The present invention also provides an upper-mounted high-temperature resistant hard-seal ball valve and its adaptive dynamic control system, including a valve body 1, a valve seat 2 installed inside the valve body 1, a ball 3 pressing between two valve seats 2, a valve cover 4 for fixing the ball 3 installed on the valve body 1, and a valve stem 5 that cooperates with the ball 3 in the middle of the valve cover 4.

[0055] It also includes: two sets of sealing kits 6, one set of sealing kits 6 is located at the mounting point between the valve body 1 and the valve seat 2 in the sealing kit, the mounting point between the valve body 1 and the valve seat 2 is connected to the sealing kit 6 by a cylindrical helical compression spring 7, the other set of sealing kits 6 is located between the valve stem 5 and the valve cover 4, the sealing kit 6 includes a copper sleeve and a gasket, and a heat insulation sleeve 8 that cooperates with the copper sleeve is also provided between the valve stem 5 and the valve cover 4;

[0056] The valve cover 4 is provided with heat sink 9, the end of the valve stem 5 is engaged with the turbine box 10, and the connection between the valve stem 5 and the valve cover 4 is provided with packing.

[0057] In this invention, the valve stem 5 passes through the valve cover 4 and connects to the top of the ball 3. The connection between the valve stem 5 and the valve cover 4 is filled with packing to ensure sealing performance. The valve stem 5 is designed to be longer and is equipped with a heat insulation sleeve 8 to effectively prevent the high temperature of the medium from affecting the sealing of the valve stem 5.

[0058] The top-mounted design allows the ball 3 to be installed from the top of the valve body 1. The ball 3 is fixed by the installation of the valve cover 4 and the valve body 1. During maintenance, only the valve cover 3 needs to be opened and the ball 3 can be lifted out for repair and replacement of sealing rings and other accessories. This design is particularly suitable for ball valves in underground pipelines, which greatly improves maintenance efficiency and safety.

[0059] Example: This example describes the application of the present invention in a specific industrial scenario. The application is a 4-inch 2500LB class top-entry high-temperature hard-seal ball valve used in a superheated steam pipeline of a thermal power plant, designed to shut off at ANSI Class VI. The valve body is made of A182F91 stainless steel, and the valve ball and seat are made of A182F91 stainless steel with Stellite 6 hard alloy sealing surfaces welded onto them. The valve dynamic control unit in this system uses NXP's i.MX8MPlus industrial-grade system-on-a-chip as its main controller, which integrates a neural processing unit with 2.3 TOPS of computing power. The intelligent electric actuator uses a 7W permanent magnet synchronous motor with a 10:1 planetary gear reducer, and its internal servo driver is based on TI's TMS3F28379D microcontroller. Scenario setting: The valve needs to perform a shut-off operation under typical harsh operating conditions. At this time, the medium in the pipeline is superheated steam at 430℃, the upstream pressure of the valve is 9.0 MPa, and the downstream pressure is 1.0 MPa, meaning the pressure difference ΔP is 8.0 MPa. The operating procedure is as follows:

[0060] The valve dynamic control unit receives a "close" command from the DCS. The valve position is now... =90°, target position =0°.

[0061] The multi-point distributed sensing subsystem reports data in real time, and the valve dynamic control unit calculates the weighted average effective temperature. =5℃, pressure difference ΔP=8.0MPa. Wear state variables maintained internally by the valve dynamic control unit. Since the valve has only been in operation for a short time, its value is close to the initial value of 1.0.

[0062] The valve dynamic control unit feeds the input vector [5,8.0,90,0,1.0] into the physical information neural network model running in the neural processing unit.

[0063] After approximately 3 milliseconds, the physical information neural network model outputs the predicted torque. =18 Nm. This value is the precise torque required by the model to overcome the thrust generated by the 8.0 MPa pressure difference under the current high temperature and low friction coefficient, with an additional safety margin, and is derived from physical laws and data learning.

[0064] The valve dynamic control unit immediately sends the target position of 0° and torque limit of 18Nm to the actuator via the EtherCAT bus.

[0065] The actuator smoothly drives the valve to close with a torque not exceeding 18 Nm. During the closing process, the valve dynamic control unit continuously records the actual output torque fed back by the actuator as 13330 Nm, ultimately achieving a reliable seal at the 0° position. No torque surges occurred throughout the entire process.

[0066] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive dynamic control system for a top-mounted high-temperature resistant hard-seal ball valve, characterized in that, The system includes: A multi-point distributed sensing subsystem is configured to collect the physical state parameters of the ball valve in real time. A valve dynamic control unit establishes a data communication connection with the multi-point distributed sensing subsystem. The valve dynamic control unit is configured to receive the physical state parameters from the multi-point distributed sensing subsystem and calculate the target drive torque value required to perform valve operation based on the physical state parameters. And an intelligent electric actuator, which establishes a data communication connection with the valve dynamic control unit and is mechanically connected to the ball valve, the intelligent electric actuator being configured to receive and execute the target drive torque value issued by the valve dynamic control unit to apply a dynamically adjusted drive torque to the ball valve; The multi-point distributed sensing subsystem includes: A temperature sensor is installed at a preset temperature measurement position on the ball valve to measure the temperature of the ball valve; A pressure sensor, installed in the upstream or downstream pipeline of the ball valve, is used to measure the medium pressure that the ball valve is subjected to; And a position sensor, which is integrated into the intelligent electric actuator, for providing the current angular position of the ball valve; The valve dynamic control unit internally stores and runs a physical information neural network torque prediction model. The model is a deep feedforward neural network, and its structure includes an input layer, a hidden layer, and an output layer. The input layer of the physical information neural network torque prediction model receives a state vector consisting of five physical quantities, the state vector including: The model includes the weighted average valve body temperature calculated from the temperature sensor data, the valve inlet and outlet pressure difference measured by the pressure sensor, the current angular position of the valve provided by the position sensor, the target angular position issued by the upper control system, and internal variables characterizing the valve wear state; the output layer of the model outputs a scalar value, which is the predicted optimal driving torque required to complete the movement from the current angular position to the target angular position; The physical information neural network torque prediction model is generated through offline training; its offline training process includes: A digital twin model of the ball valve was established using finite element analysis software. This model defined the geometric dimensions, material properties, linear thermal expansion coefficients with temperature variation, and temperature-dependent friction coefficients of the ball and seat. A thermo-mechanical coupling simulation was performed on the digital twin model by applying temperature and pressure loads covering the entire design operating range, generating a dataset containing the correspondence between temperature, pressure difference, angular position, and required torque. When training the physical information neural network torque prediction model, a total loss function was used. This total loss function was defined as the weighted sum of data fitting loss and physical constraint loss. The physical constraint loss encoded the physical relationship equations between the total driving torque and the sealing surface friction torque, packing friction torque, bearing friction torque, and hydrodynamic torque into the loss function.

2. The adaptive dynamic control system for the top-mounted high-temperature resistant hard-seal ball valve according to claim 1, characterized in that, The temperature sensor employs three armored platinum resistance temperature sensors. The first temperature sensor is tightly fitted to the outer wall of the valve seat on the upstream side of the valve body, the second temperature sensor is tightly fitted to the outer wall of the valve seat on the downstream side of the valve body, and the third temperature sensor is installed on the valve cover or middle flange of the ball valve. The valve dynamic control unit is configured to perform weighted average processing on the multi-point temperature data collected by the three temperature sensors to obtain an effective temperature value.

3. The adaptive dynamic control system for the top-mounted high-temperature resistant hard-seal ball valve according to claim 1, characterized in that, The pressure sensor is a piezoresistive pressure transmitter, which is installed at the pressure taps of the pipeline upstream and downstream of the ball valve to monitor the pressure before and after the valve in real time; the valve dynamic control unit is configured to calculate the pressure difference between the valve inlet and outlet based on the measured values ​​of the pressure sensor.

4. The adaptive dynamic control system for the top-mounted high-temperature resistant hard-seal ball valve according to claim 1, characterized in that, The valve dynamic control unit also integrates an online self-learning and model optimization module. This module is configured to continuously collect real operating data during the actual operation of the valve and use the real operating data to periodically fine-tune the pre-trained physical information neural network torque prediction model.

5. A top-mounted high-temperature resistant hard-seal ball valve, characterized in that, The adaptive dynamic control system includes the top-mounted high-temperature resistant hard-seal ball valve as described in any one of claims 1-4.

6. The top-mounted high-temperature hard-seal ball valve according to claim 5, comprising a valve body (1), a valve seat (2) installed inside the valve body (1), a ball (3) pressing between two valve seats (2), a valve cover (4) for fixing the ball (3) installed on the valve body (1), and a valve stem (5) cooperating with the ball (3) provided in the middle of the valve cover (4); characterized in that, It also includes: two sets of sealing kits (6), one set of sealing kits (6) is located at the mounting point between the valve body (1) and the valve seat (2), the mounting point between the valve body (1) and the valve seat (2) is connected to the sealing kit (6) by a cylindrical helical compression spring (7), the other set of sealing kits (6) is located between the valve stem (5) and the valve cover (4), the sealing kit (6) includes a copper sleeve and a gasket, and a heat insulation sleeve (8) that cooperates with the copper sleeve is also provided between the valve stem (5) and the valve cover (4); the valve cover (4) is provided with heat sink (9), the end of the valve stem (5) cooperates with the worm gear box (10), and the connection between the valve stem (5) and the valve cover (4) is provided with packing.

Images