Milk filling dynamic switching system based on digital twinning

Abstract

The present application relates to the technical field of food filling automation control, in particular to a milk filling dynamic switching system based on digital twinning, comprising: a topology mapping and wave front deduction step: an edge computing gateway analyzes pipe network process data to construct a topology wave front directed graph, generates a virtual wave front, dynamically updates edge weights and deduces predicted arrival time, and generates an asynchronous time offset vector; a real state detection step: a state detection sensor detects the real arrival time of a real miscible zone passing through a target node; a closed-loop correction step: an edge computing gateway calculates time residuals and uses a filtering algorithm to correct fluid property parameters through back propagation; an asynchronous switching execution step: an execution controller receives the asynchronous time offset vector, controls an end execution mechanism to execute asynchronous switching actions according to absolute time stamps; the present application reduces high-dimensional fluid evolution to graph network wave front tracking, reduces the redundant waiting time of pipe network flushing, and improves the interception rate of qualified dairy products.

Description

Technical Field

[0001] This invention relates to the field of automated control technology for food filling, specifically a dynamic switching system for milk filling based on digital twins. Background Technology

[0002] Fluid simulation and synchronous rigid control technology is a traditional method for media switching in continuous dairy product filling production lines. It can solve the three-dimensional physical flow field through partial differential equations and can set a fixed rinsing waiting time to synchronously control the opening and closing of all filling valves, which has high reliability in preventing cross-contamination and process stability.

[0003] With the development of industrial automation technology, pipeline media switching has gradually evolved from simple timed flushing to complex centralized control combined with fluid dynamics calculations. However, when performing high-frequency dynamic tracking of complex and elongated mixed-phase zone boundaries, the traditional partial differential equation solution is still limited by the computing power of the edge industrial control computer, and the synchronous rigid action logic cannot achieve spatial matching of the timing of each node, resulting in excessively long redundant waiting time for pipeline flushing, which leads to low retention rate of qualified dairy products and low material utilization rate. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic switching system for milk filling based on digital twins, and to solve the following technical problems: By avoiding the computational bottleneck of traditional partial differential equation solving, it can dynamically track the boundary of the mixed-phase region and overcome concept drift under the limited computing power of edge computing. Under the premise of ensuring that materials do not cross-contaminate, it reduces the redundant waiting time of pipeline flushing and improves the retention rate of qualified dairy products.

[0005] The objective of this invention can be achieved through the following technical solutions: A dynamic switching system for milk filling based on digital twins, comprising: The edge computing gateway is configured to: parse preset pipeline process data to construct a topological wavefront directed graph and bind physical addresses to graph nodes; generate a virtual wavefront at the source node of the topological wavefront directed graph when receiving a medium switching command; collect fluid state data and drive equipment operation data in real time, dynamically update the edge weights of the topological wavefront directed graph based on the fluid state data and drive equipment operation data, deduce the predicted arrival time of the wavefront in the mixed-phase region in the topological wavefront directed graph; and generate an asynchronous time offset vector based on the predicted arrival time and send it to the execution controller. A state detection sensor is installed in the pipeline network and configured to detect the actual arrival time of the real mixed region when it passes through the target node and upload the actual arrival time to the edge computing gateway. The edge computing gateway is also configured to: receive the actual arrival time, calculate the time residual between the actual arrival time and the predicted arrival time, and use a filtering algorithm to backpropagate the time residual to correct the fluid property parameters of the entire network. The execution controller is configured to receive the asynchronous time offset vector sent by the edge computing gateway and control the end-effector to perform asynchronous switching actions according to the absolute timestamp corresponding to the asynchronous time offset vector.

[0006] Preferably, the edge computing gateway includes: The spatial mapping module is configured to acquire preset pipeline flow data and map the three-dimensional fluid pipeline space into a topological wavefront directed graph. The address binding module is configured to bind the physical valve addresses in the execution controller to the corresponding nodes in the topological wavefront directed graph; In the topological wavefront directed graph, the nodes represent the shunt actuators or end actuators. The edges of a topological wavefront directed graph represent fluid transport channels.

[0007] Preferably, the edge computing gateway also includes: The weight update module is configured to define the edge weight as the dynamic convection delay time. The weight update module is also configured to: obtain the preset pipe length, calculate the real-time flow rate based on the driving equipment operation data and the preset pump displacement coefficient, obtain the fluid kinematic viscosity based on the fluid state data and the preset fluid property table; divide the preset pipe length by the actual flow velocity determined by the real-time flow rate and the fluid kinematic viscosity, and calculate and update the dynamic convection delay time.

[0008] Preferably, the edge computing gateway also includes: The wavefront evolution module is configured to generate a virtual wavefront at the source node when receiving a medium switching command; The wavefront evolution module is also configured to: use state space matrix multiplication, combined with the current wavefront state matrix and control input matrix, to calculate in real time the predicted arrival time of each node in the virtual wavefront arrival topology directed graph.

[0009] Preferably, the edge computing gateway also includes: The residual calculation module is configured to calculate the time residual between the actual arrival time and the predicted arrival time. The parameter correction module is configured to determine the relationship between the absolute value of the time residual and a preset residual threshold. If the absolute value of the time residual is greater than the preset residual threshold, the time residual is backpropagated using an extended Kalman filter to correct the viscosity coefficient parameter in the fluid property parameters in real time. If the absolute value of the time residual is equal to the preset residual threshold, keep the viscosity coefficient parameter unchanged and output a boundary warning signal; If the absolute value of the time residual is less than the preset residual threshold, keep the viscosity coefficient parameter unchanged and output a normal operation flag.

[0010] Preferably, the edge computing gateway also includes: The control generation module is configured to calculate the cut-off time point corresponding to each end effector based on the predicted arrival time, and combine all the cut-off time points to generate an asynchronous time offset vector. The instruction issuing module is configured to issue asynchronous time offset vectors to the execution controller; The execution controller is configured to trigger the opening and closing actions of the corresponding end effector at different absolute timestamps based on the asynchronous time offset vector, so as to cut the wavefront of the mixed region.

[0011] Preferably, the edge computing gateway also includes: The fatigue diagnosis module is configured to calculate the time residuals of specific nodes within a preset historical period. The fatigue diagnosis module is also configured to calculate the mean and variance of the time residuals, and to determine the relationship between the mean and a preset residual mean threshold, and between the variance and a preset variance threshold. If the mean is greater than the preset residual mean threshold and the fluctuation variance is less than the preset variance threshold, the physical component corresponding to the specific node is determined to be in an abnormal physical state, and equipment maintenance work order information is generated. If the variance of the fluctuation is greater than or equal to the preset variance threshold, it is determined to be fluid fluctuation interference, and a fluid fluctuation data label is generated.

[0012] Preferably, the fluid state data includes milk temperature data; the drive equipment operation data includes real-time speed data of the variable frequency feed pump; the execution controller is a programmable logic controller; the end effector is a pneumatic reversing valve array; and the state detection sensor is a high-frequency conductivity sensor located at the bifurcation of the main pipeline.

[0013] Preferably, the system is limited to continuous flow switching control of incompressible fluids in a closed pipeline network, and the control is achieved by changing the duty cycle or absolute timestamp of the actuator in the programmable logic controller.

[0014] Preferably, the specific determination condition for detecting the true arrival time using the high-frequency conductivity sensor is as follows: perform a first-order difference operation on the received conductivity time series stream to obtain the real-time rate of change; when the absolute value of the real-time rate of change crosses the preset threshold for the first time, and the exceeding state continues for a set digital de-jittering filter window period, the determination time that satisfies this timing condition is locked as the true arrival time.

[0015] The beneficial effects of this invention are: 1. This invention constructs a topological wavefront directed graph by parsing pipeline process data through an edge computing gateway, mapping the complex three-dimensional fluid pipeline space into a one-dimensional directed network, and using state space matrix multiplication to deduce the virtual wavefront position in real time. This dimension reduction mapping mechanism avoids the computing power bottleneck of solving traditional partial differential equations, significantly compresses the computational complexity, ensures that the hard real-time constraints of the underlying control are met under the limited computing power of edge devices, and realizes high-frequency dynamic tracking of the boundary of the mixed phase region. 2. This system generates an asynchronous time offset vector based on the predicted arrival time, and controls the end effector to perform switching actions at different absolute timestamps. This strategy breaks the logic of synchronous rigid action of all filling valves in the traditional production line. It implements point-by-point timing matching for the elongated mixed zone, ensuring that each filling head opens at the critical moment when pure new material arrives. This reduces redundant waiting time while preventing cross-contamination, and improves the retention rate of qualified dairy products and the utilization rate of materials. 3. This invention utilizes a state detection sensor to collect the actual arrival time and calculates the residual between it and the predicted time. The time residual is backpropagated through an extended Kalman filter to correct the fluid viscosity coefficient parameter in real time. This closed-loop self-calibration mechanism gives the system adaptive evolution capability, effectively overcoming the concept drift problem caused by thermal hysteresis, raw milk batch differences, or equipment wear, and ensuring the robustness and high fidelity of the prediction model throughout its entire life cycle. 4. The system collects fluid state and drive equipment operation data in real time, and defines the edge weight as the dynamic convection delay time determined by the real-time flow rate and fluid kinematic viscosity. This dynamic weight update mechanism can sensitively capture the nonlinear velocity decay caused by the fluctuation of residual temperature during pipeline cleaning or the change of pump speed without performing intensive dynamic equation solving. It internalizes physical constraints into lightweight graph properties, improving the adaptability of the system in wavefront prediction under time-varying conditions. 5. This invention uses a fatigue diagnosis module to statistically analyze the mean and variance of time residuals, and determines the mechanical fatigue aging or fluid fluctuation interference of the physical actuator based on statistical characteristics. This diagnostic logic does not require additional acoustic or vibration sensors, and can extract underlying mechanical state characteristics from fluid evolution deviations, realizing cross-domain mapping from fluid characteristics to mechanical health status, and providing low-cost data support for predictive maintenance of production line equipment. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a schematic diagram of the logic flow of a dynamic switching system for milk filling based on digital twins, provided in an embodiment of the present invention. Detailed Implementation

[0018] 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.

[0019] Please see Figure 1 As shown, the dynamic switching system for milk filling based on digital twins includes: an edge computing gateway configured to: parse preset pipeline process data to construct a topological wavefront directed graph and bind physical addresses to graph nodes; generate a virtual wavefront at the source node of the topological wavefront directed graph when receiving a medium switching command; collect fluid state data and drive equipment operation data in real time, dynamically update the edge weights of the topological wavefront directed graph based on the fluid state data and drive equipment operation data, deduce the predicted arrival time of the wavefront in the mixed-phase region in the topological wavefront directed graph; and generate an asynchronous time offset vector based on the predicted arrival time and send it to the execution controller. A state detection sensor is installed in the pipeline network and configured to detect the actual arrival time of the real mixed zone when it passes through the target node and upload the actual arrival time to the edge computing gateway. The edge computing gateway is also configured to: receive the actual arrival time, calculate the time residual between the actual arrival time and the predicted arrival time, and use a filtering algorithm to backpropagate the time residual to correct the fluid property parameters of the entire network; and execute the controller, configured to: receive the asynchronous time offset vector sent by the edge computing gateway, and control the end actuator to perform asynchronous switching actions according to the absolute timestamp corresponding to the asynchronous time offset vector.

[0020] As the core decision-making hub, the edge computing gateway parses the preset pipeline process data from the engineering database to construct a directed graph of the topological wavefront and binds the physical address to the graph node. This aims to reduce the complexity of the three-dimensional physical flow field to a one-dimensional directed network. This mechanism lays the topological foundation for the subsequent millisecond-level deduction of the system. In response to the received medium switching command, the edge computing gateway immediately constructs a virtual wavefront in place at the source node of the directed graph of the topological wavefront. The edge computing gateway collects fluid state data and drive equipment operation data in real time through the underlying I / O interface. Based on the fluid state data and drive equipment operation data, it dynamically updates the edge weights of the topological wavefront directed graph and infers the predicted arrival time of the wavefront in the mixed region in the topological wavefront directed graph. Based on this, the edge computing gateway generates an asynchronous time offset vector based on the predicted arrival time and sends it to the execution controller to ensure spatial matching of the control timing of each node; at the same time, the state detection sensor set in the pipeline network detects the actual arrival time when the real mixed zone passes through the target node and uploads the actual arrival time to the edge computing gateway. The edge computing gateway receives the actual arrival time, calculates the time residual between the actual arrival time and the predicted arrival time, and uses a filtering algorithm to backpropagate the time residual to correct the fluid property parameters of the entire network, thereby overcoming the concept drift caused by thermal hysteresis or batch switching. The execution controller receives the asynchronous time offset vector sent by the edge computing gateway and controls the end actuator to perform asynchronous switching actions according to the absolute timestamp corresponding to the asynchronous time offset vector, thus completing the closed-loop mapping from fluid dynamics calculation to underlying physical actions. In the context of a continuous filling production line for dairy products, this embodiment avoids the computational bottleneck of solving traditional partial differential equations by reducing the high-dimensional fluid evolution to a graph network wavefront tracking. This cascade mechanism combining dimensionality reduction mapping and closed-loop correction enables the system to dynamically track the elongated boundary of the mixed-phase region under the limited computing power of the edge industrial control computer. Under the premise of ensuring that the materials do not cross-contaminate, it reduces the redundant waiting time for pipeline flushing and improves the retention rate of qualified dairy products.

[0021] In this embodiment, the edge computing gateway includes: a spatial mapping module configured to acquire preset pipeline process data and map the three-dimensional fluid pipeline space into a topological wavefront directed graph; and an address binding module configured to bind the physical valve address in the execution controller to the corresponding node of the topological wavefront directed graph; wherein, the node of the topological wavefront directed graph represents a diversion execution mechanism or an end-effector; and the edge of the topological wavefront directed graph represents a fluid transmission channel.

[0022] The spatial mapping module acquires preset pipeline flow data from pipeline and instrumentation diagram engineering files, and maps the three-dimensional fluid pipeline space into a topological wavefront directed graph. This aims to strip away the nonlinear turbulence details of fluid in complex bends and extract the core skeleton that determines convection time. Specifically, during the mapping operation, the spatial mapping module traverses and parses the component attribute identifiers in the pipeline process data. When a component identified as a diversion actuator or end actuator is detected, its corresponding three-dimensional coordinate instance is extracted as a discrete graph node of the topological wavefront directed graph. Based on the pipeline connection network table, the depth-first search algorithm is used to track the fluid flow direction, and the pipe segments that connect adjacent components in physical space are transformed into directed edges connecting the corresponding graph nodes. The direction of the edges is strictly limited by the initial pressure gradient vector of the medium supply source. The address binding module binds the physical valve addresses in the actuator controller to the corresponding nodes in the topological wavefront directed graph to establish an addressing mapping relationship between the digital virtual space and the physical actuators. Specifically, the address binding module parses the underlying device network configuration file, reads the media access control address or network protocol port number corresponding to each physical actuator, and uses the internal index ID of the graph node as the access key and the corresponding physical address as the storage value to construct a hash addressing dictionary table, thereby unambiguously associating the abstract graph nodes with the physical output channels of the underlying programmable logic controller. In this graph, the nodes of the topological wavefront directed graph represent flow splitting actuators or end effectors, forming decision-making bifurcation points for the fluid path; the edges of the topological wavefront directed graph represent fluid transmission channels, carrying the physical constraints of fluid convection. In the scenario of a continuous filling production line for dairy products, the spatial mapping and address binding mechanism constructed in this embodiment transforms the intricate aseptic pipeline network into a discrete mathematical graph structure. This design enables the node states generated by the upper-level algorithm to be seamlessly and unambiguously converted into port level signals of the underlying programmable logic controller, verifying the robustness of information flow in the bidirectional transmission process between the virtual and physical worlds, and providing a solid addressing foundation for the coordinated operation of multiple filling head arrays.

[0023] In this embodiment, the edge computing gateway further includes: a weight update module, configured to define the edge weight as a dynamic convection delay time; the weight update module is further configured to: obtain a preset pipe length, calculate the real-time flow rate based on the driving device operation data and the preset pump displacement coefficient, obtain the fluid kinematic viscosity based on the fluid state data and the preset fluid property table; divide the preset pipe length by the actual flow rate determined by the real-time flow rate and the fluid kinematic viscosity, and calculate and update the dynamic convection delay time.

[0024] The weight update module defines edge weights as dynamic convection delay time, replacing the static physical distance in traditional graph theory. The module obtains the preset pipe length from the preset pipeline flow data and calculates the real-time flow rate based on the drive equipment operating data and the preset pump displacement coefficient from the equipment nameplate. The specific calculation logic is as follows: the real-time speed of the variable frequency feed pump extracted from the drive equipment operating data is multiplied by the preset pump displacement coefficient, and then multiplied by the preset pump volumetric efficiency to compensate for internal backflow losses under high pressure, thereby obtaining the real-time flow rate. The preset pump volumetric efficiency is determined based on the equipment's factory calibration data or on-site volumetric testing. The weight update module obtains the fluid kinematic viscosity based on fluid state data and a preset fluid property table from the local database, aiming to quantify the nonlinear attenuation effect of fluid internal friction on the overall flow velocity. Specifically, during the mapping operation, the weight update module divides the preset pipe length by the actual flow velocity determined by the real-time flow rate and fluid kinematic viscosity to calculate and update the dynamic convection delay time. The specific calculation logic is implemented through the following formula: in, The dynamic convection delay time is calculated in real time by this module and represents the real-time estimated time required for fluid to flow through this section of the pipeline. The preset pipeline length is directly extracted from the three-dimensional coordinate constraints of the preset pipeline network process data; The actual flow rate is determined by both the real-time flow rate and the fluid's kinematic viscosity. Specifically, in order to overcome the nonlinear deviation between the actual flow field in the pipeline network and the theoretical homogeneous flow, this actual flow velocity... The acquisition logic is as follows: divide the calculated real-time flow rate by the preset pipeline cross-sectional area obtained from the pipeline network process data. The average flow velocity at the foundation cross section is calculated using the following formula: in, The average flow velocity at the base cross section, For real-time traffic, The cross-sectional area of ​​the pipe is preset. Using the average flow velocity at this base cross-section, the preset pipe inner diameter, and the obtained fluid kinematic viscosity, the Reynolds number of the fluid currently inside the pipe is calculated. The calculation formula is as follows: in, The dimensionless Reynolds number, The preset pipe inner diameter, Let the fluid's kinematic viscosity be denoted by Reynolds number. Based on this Reynolds number and a preset absolute pipe wall roughness parameter, the current pipe friction coefficient is calculated using the Körbruck equation. The specific formula is as follows: This equation relates to the friction coefficient. The implicit equations need to be solved numerically. Among them, This is the friction factor along the pipeline. The parameters are preset absolute roughness parameters of the pipe wall; since this equation is implicit, this module uses Newton's iteration method for numerical solution to obtain the parameters. The precise value; Specifically, when using Newton's iterative method to solve the problem, to ensure rapid convergence of the algorithm under the limited computing power of the edge computing gateway, the system uses the result calculated by the Harland explicit approximation equation as the initial guess value for the iteration. Specifically, the calculation expression of the Harland explicit approximation equation is as follows: in, This is the estimated initial friction coefficient, which is obtained by directly using the known preset absolute pipe wall roughness parameters. Preset pipe inner diameter and the currently calculated Reynolds number By performing a one-time algebraic operation, an initial value that approximates the true solution can be obtained without iteration, thereby strictly controlling the number of iterations within a preset safety threshold and avoiding control cycle timeouts due to iteration divergence. Multiplying the average velocity at the base section by the velocity profile correction factor derived from the friction coefficient along the flow path yields the true physical velocity, which accurately accounts for the internal friction effects of laminar or turbulent flow and the hysteresis damping attenuation of the boundary layer. This actual velocity... The calculation formula is: in, This is a velocity profile correction factor used to correct the cross-sectional average velocity to the maximum velocity at the pipe center. Considering the non-uniformity of velocity distribution under turbulent conditions, based on Prandtl's mixing length theory and Niculas' experimental data, in a fully developed turbulent pipe, the ratio of the maximum velocity at the center to the average velocity is related to the Reynolds number and the drag coefficient. Theoretically, it can be described using a power-law formula: in, For empirical indices related to the Reynolds number, such as The relevant function; in order to more directly utilize the drag coefficient calculated above. To reduce intermediate variable transformations, this embodiment preferably uses an empirical approximation formula based on the drag coefficient: This formula, as a correction factor, effectively covers the phenomenon of prominent central velocity caused by pipe wall friction. Since the wavefront propagation velocity in the miscible zone during fluid displacement is mainly dominated by the maximum velocity at the center of the pipe, this correction factor is introduced to adjust the one-dimensional average velocity of the basic cross-section. The physical velocity is modified to characterize the actual propulsion speed at the front of the mixed region, thereby accurately quantifying the nonlinear hysteresis effect of pipe wall friction on the velocity profile. This method of deducing flow velocity step by step from the underlying fluid dynamics characteristics clarifies the mathematical calculation transmission path of fluid dynamics elements to dynamic delay weights. In the context of a continuous filling production line for dairy products, this embodiment internalizes physical constraints such as fluid viscosity and pipe diameter into lightweight graph network edge attributes. This dynamic edge weighting mechanism can sensitively capture flow rate changes caused by batch differences in raw milk or residual temperature fluctuations after pipe cleaning without performing computationally intensive dynamic equation solving, thereby improving the adaptability and accuracy of wavefront prediction under time-varying conditions.

[0025] In this embodiment, the edge computing gateway further includes: a wavefront evolution module, configured to generate a virtual wavefront at the source node when receiving a medium switching command; the wavefront evolution module is also configured to: use state space matrix multiplication, combined with the wavefront state matrix and control input matrix at the current moment, to calculate in real time the predicted arrival time of the virtual wavefront to each node in the directed graph of the topological wavefront.

[0026] Upon receiving a medium switching command, the wavefront evolution module generates a virtual wavefront at the source node in situ, serving as the initial boundary condition for fluid replacement in the digital space. To complete the full-network simulation within an extremely short control cycle, the wavefront evolution module utilizes state-space matrix multiplication, combining the current wavefront state matrix and control input matrix, to calculate in real-time the predicted arrival time of the virtual wavefront at each node in the directed graph of the topological wavefront. This state evolution process is implemented through the following discrete state transition equation: in, For nonlinear step activation functions; subscript and Let these represent the discrete-time step indices of the current and the next adjacent evolution step, respectively, satisfying... The recursive relationship; wavefront state matrix It is a multidimensional column vector. Let the total number of nodes in the directed graph of the topological wavefront be . ,but for 3D column vector, i.e. , Representing the set of real numbers, each element in the vector represents the current state of the corresponding node. The state value indicating whether the moment is covered by the wavefront of the mixed phase region, for example, 0 indicates not arrived and 1 indicates arrived; At the initial moment wavefront state matrix Initialized as follows: except for the element corresponding to the source node, all other node elements are assigned a value of 0, thus defining the initial boundary conditions for iterative updates; the wavefront state matrix for the next time step is calculated iteratively using the formula. Also The column vectors reflect the spatial propagation distribution of the wavefront across the entire network; Furthermore, to calculate the predicted arrival time, after obtaining a series of wavefront state matrix sequences corresponding to discrete time steps through continuous iteration, the wavefront evolution module traverses the state matrix sequence for each node to find the critical discrete time step index when its state value first jumps from 0 to 1; it extracts the cumulative advance time residual of the node in the previous time step, calculates the ratio of it to the dynamic convection delay time corresponding to the node as the substep interpolation coefficient, and performs linear interpolation by combining the critical discrete time step index and the single-step state evolution time period to calculate the continuous predicted arrival time with floating-point precision. Furthermore, to prevent the risk of infinite iterative evolution caused by physical disconnection or abnormal parameters, the wavefront evolution module has a built-in deadlock prevention and circuit breaker mechanism. Specifically, the system pre-calculates and sets a global maximum evolution step threshold. The calculation logic for this threshold is as follows: extract the total length of the longest connected path network from the source node in the topological wavefront directed graph, divide it by the preset minimum safe flow velocity to prevent blockage. The preset minimum safe flow velocity to prevent blockage is derived from the lower limit empirical flow velocity value recommended in the fluid mechanics handbook to prevent suspended matter sedimentation, so as to provide a conservative worst-case assessment benchmark, and then divide it by the single-step state evolution time period. ; During the process of traversing and searching for the critical discrete time step index for each node, if the current number of iterations detected reaches the global maximum evolution step threshold and the state value of the corresponding node has not changed from 0 to 1, the traversal search process for the unreached node is forcibly stopped, its predicted arrival time is marked as the floating-point limit maximum value, and an abnormal disconnection signal of pipeline flow blockage is thrown to the upper-layer diagnostic interface, thereby ensuring that the edge computing gateway still meets the timeout return constraint of real-time computing under any extreme and unexpected working conditions. State transition matrix It is A square matrix of order, where the subscripts are... The index explicitly represents the current time step and is used to quantify the time delay effect of fluid convection propagation from upstream nodes to downstream nodes; in the specific implementation process of each discrete time step, the matrix is... Initialized as an all-zero matrix, the specific mathematical mapping generation rules for its diagonal and off-diagonal elements are defined as follows: The single-step discrete time period of the system evolution is set. ,Should The value of is strictly controlled by the Courant-Friedrich-Riemann stability condition of the pipeline network to prevent node loss during evolution. Specifically, the system traverses the directed edge graph of the topological wavefront, extracts the shortest pipe length corresponding to all directed edges, and extracts the theoretical maximum flow velocity calculated based on the maximum variable frequency feed rate corresponding to the equipment nameplate. The shortest pipe length is divided by the theoretical maximum flow velocity to obtain the minimum convection crossing time. The single-step discrete time period is then calculated. It is set to half of the minimum convection cross-time, thereby ensuring that the wavefront can only advance to the adjacent immediate node at most within a single time step; And maintain a cumulative propagation time variable for each directed edge of the topological wavefront directed graph. When the upstream node When the state first changes to 1, i.e., when it is covered by the wavefront, the cumulative advance time corresponding to that edge is initialized. In each subsequent discrete time step, if the node The state remains 1 and the downstream node If the state is 0, then the execution time accumulation update logic is performed, and its calculation formula is: In the directed graph of the topological wavefront, the upstream node... Pointing to downstream nodes A directed edge, if the current cumulative propagation time of the wavefront on that directed edge... Greater than or equal to the dynamic convection delay time corresponding to that edge ( (Calculated in real-time for this edge by the aforementioned weight update module), then the matrix... The middle corresponds to the first Line number The corresponding off-diagonal elements of the column The value is assigned as 1, and a superscript is used here. A separate time step index is used to avoid confusion with the node row and column indexes; Its explicit mathematical condition expression is: when hour, ;when hour, This indicates that, when the conditions are met, the wavefront state can be propagated across adjacent nodes within this single-step cycle, and once a downstream node is detected... If the state has successfully changed to 1 at the current time step, meaning the wavefront has completed crossing the directed edge, then the cumulative advancement time corresponding to that edge will stop in subsequent evolution cycles. The accumulation operation is performed and locked to avoid redundant computational overhead and maintain the unidirectional irreversible property of the physical state; At the same time, regarding the diagonal elements of the matrix Once the corresponding node is detected If the wavefront has been covered in the previous state matrix sequence, then it is forcibly assigned a value and locked to 1 in order to maintain the unidirectional irreversible physical state characteristics of the incompressible fluid in the continuous replacement process of the closed pipe network. Furthermore, to prevent Boolean states from diverging into floating-point values ​​during continuous matrix multiplication iterations and causing logical distortion, the wavefront evolution module performs internal linear matrix operations each time. Then, a strict step activation limiting process is applied to all vector elements, that is, elements greater than 0 are hard-clamped to 1, and the remaining elements are constrained to zero, thereby ensuring that the evolution matrix always converges within the effective wavefront state space; specifically, this step activation limiting process is equivalent to nesting a nonlinear activation function outside the linear operation result of the state transition equation, that is, the actual complete evolution logic is as described in the aforementioned formula: in, This is a step activation function; considering that the elements of the state transition matrix, state vector, and control input matrix are all non-negative integers, the internal linear operation result vector... any scalar element in It must be a non-negative integer; its specific mathematical expression is defined as: when hour This indicates that the node has been covered by the wavefront or has received an activation control signal; when hour This indicates that the current node has neither upstream wavefront inflow nor received an external activation control signal, aiming to maintain an inactive state uncovered by fluid; this design clearly defines the critical state. The Boolean logic meaning and physical mapping relationship of time eliminate ambiguity in the handling of untriggered states during code implementation; its core function is to filter out redundant state values ​​generated by the convergence and accumulation of multiple paths, i.e. In this case, ensure that the wavefront state variables are strictly confined to the Boolean space. Inside; Control input vector It is the sequence of master control medium switching command signals received from outside the system. Let the number of independent control command channels outside the system be... ,but for 3D column vector, i.e. Correspondingly, control input matrix It is determined based on the physical locations of the diversion valves and pumps in the pre-set pipeline flow data. A constant matrix is ​​used to map external control commands to the wavefront state evolution of each node; Its specific mapping generation rules depend on the hash addressing dictionary table generated by the aforementioned address binding module: During the initialization phase, the system traverses all... The channel identifier of the external control command channel is retrieved from the hash addressing dictionary table. The physical address to which each graph node is bound; If the first The identifier of the external control command channel is the same as the first one. If the physical addressing addresses corresponding to each graph node are matched, it is determined that there is a direct physical connection and driving relationship, and then the matrix is... The Line number The column element is assigned a value of 1, otherwise it is assigned a value of 0. This explicit judgment logic based on data field comparison ensures that the external medium switching command can be read by the program without ambiguity and accurately injected into the specified node of the topological wavefront directed graph. In the scenario of a continuous filling production line for dairy products, this embodiment introduces state-space matrix multiplication to transform the complex three-dimensional fluid spatiotemporal iterative solution into linear algebraic operations for graph network state transitions. This mechanism compresses the computational complexity from exponential to linear, ensuring that under the limited computing power of edge computing nodes, the single operation cycle of wavefront prediction meets the hard real-time constraints of the underlying controller, breaking the dynamic mismatch barrier between high-dimensional fluid simulation and the underlying industrial control system.

[0027] In this embodiment, the edge computing gateway further includes: a residual calculation module configured to calculate the time residual between the actual arrival time and the predicted arrival time; and a parameter correction module configured to determine the relationship between the absolute value of the time residual and a preset residual threshold: if the absolute value of the time residual is greater than the preset residual threshold, the time residual is backpropagated using an extended Kalman filter to correct the viscosity coefficient parameter in the fluid property parameters in real time; if the absolute value of the time residual is equal to the preset residual threshold, the viscosity coefficient parameter is kept unchanged and a boundary warning signal is output; if the absolute value of the time residual is less than the preset residual threshold, the viscosity coefficient parameter is kept unchanged and a normal operation flag is output.

[0028] The residual calculation module calculates the time residual between the actual arrival time and the predicted arrival time, aiming to quantify the deviation between pure theoretical deduction and actual physical evolution; the parameter correction module initiates a cascaded judgment mechanism to evaluate the relationship between the absolute value of the time residual and the preset residual threshold. The logic for setting the preset residual threshold is as follows: during the system debugging phase, historical time residual data is collected during multiple reference fluid switching processes, the standard deviation of the historical residual distribution is calculated, and the preset residual threshold is set to a constant multiple of the standard deviation, such as 3 times the standard deviation, to reliably distinguish between normal measurement noise and substantial system model drift; in response to the absolute value of the time residual being greater than the preset residual threshold, the system determines that the model has deviated from the actual physical state, and uses an extended Kalman filter to backpropagate the time residual and correct the viscosity coefficient parameter in the fluid property parameters in real time; It should be noted that the viscosity coefficient parameter in this embodiment, i.e., the fluid kinematic viscosity parameter, is the same as the fluid kinematic viscosity used to calculate the dynamic convection delay time in the previous embodiment. Both refer to the same physical parameter and are explained uniformly here to avoid confusion. When using the extended Kalman filter for backpropagation correction, the system uses the time residual as the filter's observation information, combines it with the prediction covariance matrix to calculate the Kalman gain, and uses this Kalman gain to map the time-dimensional residual to the fluid property parameter space, adjusting the specific value of the viscosity coefficient parameter in real time. Specifically, within the extended Kalman filter framework, the system treats the viscosity coefficient parameter as an implicit state variable to be optimized, denoted as... ; When the filter is first started, the system must initialize the state variable and the error covariance. Specifically, the theoretical value of the standard kinematic viscosity corresponding to the current measured milk temperature data in the preset fluid property table is used as the initial state variable. Assign values ​​to provide an iterative starting point that approximates real-world operating conditions; At the same time, the initial error covariance matrix The diagonal matrix is ​​set as a preset value, with the diagonal elements set to a large positive number, such as 1.0, based on the initial uncertainty of the pipeline network after cleaning. This gives the initial observation data a higher convergence weight in the first gain calculation. Since the viscosity is assumed to be constant in the short term, its state prediction equation is... The corresponding prediction error covariance is: in, The preset process noise covariance is derived from the viscosity natural fluctuation variance statistically analyzed during long-term operation of the system under constant conditions. It is used to characterize the drift uncertainty of fluid properties under uncontrolled conditions. Arrival time is treated as an explicit observation variable, and the observation Jacobian matrix is ​​obtained by differentiating the predicted arrival time through the extrapolation calculation process. , that is, the local partial derivative of the predicted arrival time of the target node with respect to the viscosity coefficient parameter, is calculated as follows: in, This represents the predicted arrival time of the target node obtained through the network-wide extrapolation using the aforementioned wavefront evolution module, given the viscosity parameters. The preset minimum perturbation step size is set between 0.1% and 1% of the current viscosity reference value to achieve an optimal balance between truncation error and rounding error in numerical differentiation. Because and The relationship between the two is a complex nonlinear mapping, making it difficult to obtain analytical derivatives. Therefore, the central difference method is used for numerical solution. In practical applications, the edge computing gateway approximates the partial derivative by calling the wavefront evolution module twice, thus avoiding the difficulty of analytical differentiation of complex nonlinear velocity profile equations. The Kalman gain is then calculated. Its formula is: in, The preset measurement noise covariance is derived from the statistical variance of the arrival time signal output by the state detection sensor in the static calibration experiment, and is used to quantify the hardware measurement uncertainty of the sensor itself. Using the calculated Kalman gain With time residual As new observation information, a viscosity correction compensation is generated through multiplication. This compensation is then added to the previous viscosity coefficient parameter reference value to complete the state update. The update formula is as follows: The updated error covariance matrix is ​​calculated using the following formula: in, To provide an identity matrix that matches the dimension of the state variables, in this univariate viscosity correction scenario, Specifically, this manifests as a scalar value The first-order identity matrix is ​​used to maintain the dimensional consistency of the error covariance matrix update within the matrix algebra operation framework. Through the rigorous formula derivation above, the system compensates for unmodeled nonlinear thermodynamic and physical disturbances with the optimal estimation theory closed loop. In response to the absolute value of the time residual being equal to the preset residual threshold, the system keeps the viscosity coefficient parameter unchanged and outputs a boundary warning signal to indicate potential process fluctuations. When the absolute value of the time residual is less than the preset residual threshold, the system keeps the viscosity coefficient parameter unchanged and outputs a normal operation flag, indicating that the current digital twin model has high fidelity. In the scenario of a continuous filling production line for dairy products, this embodiment constructs a closed-loop self-calibration mechanism based on a small amount of physical anchor data. By performing residual backpropagation through an extended Kalman filter, the system is endowed with the ability to adaptively evolve, effectively overcoming the concept drift problem caused by mechanical wear of the feed pump or fluctuations in the characteristics of raw milk in different seasons during long-term operation, and ensuring the reliability of the predicted arrival time throughout the entire life cycle of the equipment.

[0029] In this embodiment, the edge computing gateway further includes: a control generation module configured to calculate the cutoff time point corresponding to each end effector based on the predicted arrival time, and combine all cutoff time points to generate an asynchronous time offset vector; an instruction issuing module configured to issue the asynchronous time offset vector to the execution controller; and the execution controller configured to: trigger the opening and closing actions of the corresponding end effectors at different absolute timestamps according to the asynchronous time offset vector, so as to cut the wavefront of the mixed region.

[0030] The control generation module calculates the cutoff time point for each end effector based on the predicted arrival time, aiming to finely manage the spatial distribution differences of asynchronous arrival of wavefronts in the mixed-phase region within complex branch pipe networks. Specifically, when calculating the cutoff time point, the control generation module extracts the corresponding predicted arrival time for each end effector as a theoretical switching benchmark, and uses linear addition and subtraction operations to correct it in the time dimension. The specific calculation formula is as follows: in, This refers to the candidate cutoff time point for the end effector; This is the predicted arrival time for the end effector; The inherent mechanical and electrical action response delay time preset for the end effector is accurately measured by sending a trigger level command to the end effector during the offline debugging phase of the equipment and using a high-speed oscilloscope to record the time difference from the issuance of the command to the actual physical displacement of the valve core. The preset safety flushing buffer time is set based on the requirements for preventing cross-contamination of materials. Its value is determined based on the minimum time margin required to remove residual materials in historical flushing verification experiments. The absolute time corrected by the above formula is used as the candidate cut-off time point for the end effector. Furthermore, to prevent the calculated candidate cutoff time points from being causally reversed on the timeline due to sudden extreme flow rate anomalies (i.e., the calculated cutoff time point is earlier than the current system's absolute timestamp), thus preventing the underlying controller from executing expired instructions, the control generation module performs a certain step each time it completes the process. Forced causal circuit breaker determination after calculation: Extract the high-precision absolute timestamp of the current system. ,judge and Time margin; If the judgment ,in Due to the fixed communication network delay between the preset edge computing gateway and the execution controller, the system determines that the normal operation window of the node has been closed. If a forced switch is performed at this time, some mixed liquid will enter the finished product pipeline. Therefore, the system immediately triggers the emergency safety circuit breaker mechanism: either keeping the current valve state from flipping, or forcibly directing the fluid to the preset waste liquid discharge pipeline, and simultaneously generating the highest priority severe cross-contamination blocking alarm signal, linking the downstream filling machine to perform rejection operation, rather than simply resetting the timestamp to perform a delayed action, thereby ensuring the absolute safety of the finished product quality; After the causal constraint logic is determined, the final absolute time is used as the precise cut-off time point corresponding to the end effector; the control generation module maps and arranges all independently calculated cut-off time points according to the node physical address order, and combines them to generate an asynchronous time offset vector. The instruction issuing module sends the asynchronous time offset vector to the execution controller; the execution controller triggers the opening and closing actions of the corresponding end effector at different absolute timestamps based on the asynchronous time offset vector, so as to accurately cut the wavefront of the mixed region in both spatial and temporal dimensions. In the scenario of a continuous dairy product filling production line, this embodiment breaks the control logic of synchronous rigid action of all filling valves in the traditional production line by generating and executing an asynchronous time offset vector. This asynchronous feedforward control strategy implements point-by-point timing matching for the elongated and unevenly distributed mixed phase zone, so that each independent filling head can be opened at the critical threshold moment when pure new material arrives, thereby reducing the discharge of mixed waste liquid and improving the material utilization rate of the overall pipeline network.

[0031] In this embodiment, the edge computing gateway further includes: a fatigue diagnosis module, configured to statistically analyze the time residuals of specific nodes within a preset historical period; The fatigue diagnosis module is also configured to calculate the mean and variance of the time residuals, and determine the relationship between the mean and the preset residual mean threshold, and the variance and the preset variance threshold: if the mean is greater than the preset residual mean threshold and the variance is less than the preset variance threshold, it is determined that the physical component corresponding to the specific node is in an abnormal physical state, and equipment maintenance work order information is generated. If the variance of the fluctuation is greater than or equal to the preset variance threshold, it is determined to be fluid fluctuation interference, and a fluid fluctuation data label is generated.

[0032] The fatigue diagnosis module statistically analyzes the time residuals of specific nodes within a preset historical period, aiming to extract underlying mechanical state characteristics from fluid evolution deviations. The preset historical period is defined as a sliding time window containing a fixed number of fluid switching batches, such as the past 50 to 100 consecutive filling switching batches, and the time residual is strictly defined as the algebraic difference between the actual arrival time and the predicted arrival time; the fatigue diagnosis module calculates the mean and variance of the time residual. To comprehensively capture both systematic advance and systematic lag anomaly patterns and avoid logical loopholes caused by missed detections due to negative biases, the absolute value of the algebraic mean of the time residuals is specifically used when participating in logical judgments; specifically, the total number of fluid switching batches included in the preset historical period is set to... The corresponding collected time residual sequence is denoted as The mean used for judgment The calculation formula is: Fluctuation variance The calculation formula is: in, The algebraic mean without absolute values. The Bessel correction factor is used to provide an unbiased estimate of the population variance, ensuring accurate quantification of mechanical condition fluctuations in a finite sample batch. Specifically, the preset residual mean threshold and preset variance threshold required for subsequent judgments are set based on the system's historical baseline data: During the healthy operation period of the actuator in its new state, the system records its time residual sequence, and defines the upper limit of the absolute value of the algebraic mean of the residuals during this period as the preset residual mean threshold, and sets the maximum allowable variance of the residual fluctuation during this period as the preset variance threshold; on this basis, the system judges the relationship between the mean, i.e., the absolute value of the algebraic mean, and the preset residual mean threshold, as well as the relationship between the fluctuation variance and the preset variance threshold; In response to the mean being greater than a preset residual mean threshold and the fluctuation variance being less than a preset variance threshold, the system determines that the node has a stable and unidirectional systematic time deviation; further, the system traces the algebraic sign before the absolute value of the mean: if the algebraic sign is positive, that is, the actual arrival time is systematically later than the theoretical prediction time, it is determined that the physical actuator corresponding to the specific node has mechanical fatigue aging, such as the slow action caused by the increased friction of the valve core; If the algebraic sign is negative, meaning the actual arrival time is significantly earlier than the theoretical prediction time, it indicates that the model has a systematic positive bias, such as the actual flow cross-sectional area being reduced and the flow velocity increasing due to pipe scaling, or the valve having internal leakage. Whether it is a systematic lag or a systematic advance, the system determines it as an abnormal mechanical condition and generates equipment maintenance work order information. In response to the fluctuation variance being greater than or equal to the preset variance threshold, the system determines that the residual exhibits irregular high-frequency oscillations, which is judged as fluid fluctuation interference, and generates fluid fluctuation data labels. Furthermore, in response to the mean being less than or equal to the preset residual mean threshold and the fluctuation variance being less than the preset variance threshold, the system determines that the physical actuator corresponding to the specific node is responding normally and the fluid transmission is stable, determines that the equipment is within the healthy baseline range, generates the normal operation log of the equipment and maintains the current control strategy, thus forming a complete closed-loop determination logic for the equipment health status covering bidirectional deviations. In the context of a continuous filling production line for dairy products, this embodiment performs secondary statistical analysis on the time residual of fluid wavefront prediction. This eliminates the need to install expensive acoustic or vibration sensors on the production line, enabling the in-situ construction of a health monitoring mechanism for underlying mechanical equipment such as pneumatic reversing valves. This diagnostic logic, which maps fluid characteristics to mechanical states across domains, provides low-cost data support for predictive maintenance of the production line.

[0033] In this embodiment, the fluid state data includes milk temperature data; the drive equipment operation data includes real-time speed data of the variable frequency feed pump; the execution controller is a programmable logic controller; the end effector is a pneumatic reversing valve array; and the state detection sensor is a high-frequency conductivity sensor located at the bifurcation of the main pipeline.

[0034] Fluid state data includes milk temperature data from patch-type temperature probes, designed to provide a thermodynamic basis for estimating dynamic fluid viscosity; drive equipment operation data includes real-time speed data of variable frequency feed pumps from the inverter feedback interface, serving as the basis for calculating pipeline flow boundary conditions; the execution controller is set as a programmable logic controller to ensure the hard real-time performance and timing accuracy of the underlying asynchronous cut-off action; The end effector is set as a pneumatic directional valve array, which is responsible for performing high-frequency fluid cut-off and reversal; the status detection sensor is set as a high-frequency conductivity sensor at the bifurcation of the main pipeline, which aims to capture the arrival time of the real mixed phase region with extremely low delay by utilizing the conductivity change when different media alternate. In the specific implementation, in order to reliably map the continuous analog conductivity signal collected by the sensor into discrete physical trigger time points, the system performs a first-order difference operation on the received conductivity time series stream to obtain the real-time rate of change, and sets a synchronization judgment condition: when the absolute value of the real-time rate of change crosses the preset limit threshold for the first time, and the overshoot state continues for a set digital de-jitter filtering window period, it is determined that the real mixed phase region boundary has just cut through the sensor profile, and then the judgment time that meets this timing condition is locked as the real arrival time. The threshold value is set by taking twice the maximum baseline conductivity ripple peak value recorded in the no-load calibration test of the production line under clear water and pure milk conditions, so as to ensure the sensitivity of the judgment while being immune to false triggers caused by isolated bubbles or flow pulsation. In the context of a continuous filling production line for dairy products, this embodiment clearly defines the physical hardware anchor point for the implementation of the digital twin algorithm. By deeply binding the abstract topology calculation with specific industrial control equipment, the system is fully feasible in actual high-frequency filling operations. At the same time, the cooperation between the high-frequency conductivity sensor and the programmable logic controller provides high signal-to-noise ratio real observation data for the closed-loop correction of the algorithm model.

[0035] In this embodiment, the system is limited to continuous flow switching control of incompressible fluids in a closed pipeline network, and the control is achieved by changing the duty cycle or absolute timestamp of the actuator in the programmable logic controller.

[0036] The system is limited to continuous flow switching control of incompressible fluids in closed pipe networks, aiming to avoid the drastic density fluctuations caused by compressible media and ensure that the wavefront evolution algorithm based on volume conservation and convection delay can converge stably. The system is limited to control by changing the duty cycle or absolute timestamp of the actuator in the programmable logic controller, thus converging the complex flow field intervention means into discrete time-dimensional switching actions. In the context of a continuous dairy product filling production line, this embodiment clearly defines the physical boundaries and control methods of the system. By excluding unsuitable scenarios such as open spraying or compressible media, this solution focuses on the specific dynamic characteristics of incompressible fluid pipelines. This boundary limitation ensures the prediction accuracy of the dimensionality reduction topology algorithm in this specific industrial automation field and verifies the robustness and applicability of the system under specific process constraints.

[0037] 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 dynamic switching system for milk filling based on digital twins, characterized in that, The system includes: The edge computing gateway is configured to: parse preset pipeline process data to construct a topological wavefront directed graph and bind physical addresses to graph nodes; generate a virtual wavefront at the source node of the topological wavefront directed graph when receiving a medium switching command; collect fluid state data and drive equipment operation data in real time, dynamically update the edge weights of the topological wavefront directed graph based on the fluid state data and the drive equipment operation data, deduce the predicted arrival time of the mixed-phase wavefront in the topological wavefront directed graph; and generate an asynchronous time offset vector based on the predicted arrival time and send it to the execution controller. A state detection sensor is installed in the pipeline network and configured to detect the actual arrival time of a real mixed region when it passes through a target node, and upload the actual arrival time to the edge computing gateway. The edge computing gateway is further configured to: receive the actual arrival time, calculate the time residual between the actual arrival time and the predicted arrival time, and use a filtering algorithm to backpropagate the time residual to correct the fluid property parameters of the entire network; The execution controller is configured to receive the asynchronous time offset vector sent by the edge computing gateway and control the end-effector to perform an asynchronous switching action according to the absolute timestamp corresponding to the asynchronous time offset vector.

2. The milk filling dynamic switching system based on digital twins according to claim 1, characterized in that, The edge computing gateway includes: The spatial mapping module is configured to acquire the preset pipeline process data and map the three-dimensional fluid pipeline space into the topological wavefront directed graph. The address binding module is configured to bind the physical valve address in the execution controller to the corresponding node in the topological wavefront directed graph; The nodes of the topological wavefront directed graph represent splitting actuators or end-effectors. The edges of the topological wavefront directed graph represent fluid transport channels.

3. The milk filling dynamic switching system based on digital twins according to claim 2, characterized in that, The edge computing gateway also includes: The weight update module is configured to define the edge weight as a dynamic convection delay time. The weight update module is further configured to: obtain a preset pipe length; calculate the real-time flow rate based on the driving equipment operating data and the preset pump displacement coefficient; obtain the fluid kinematic viscosity based on the fluid state data and the preset fluid property table; divide the preset pipe length by the actual flow velocity determined by the real-time flow rate and the fluid kinematic viscosity; and calculate and update the dynamic convection delay time.

4. The milk filling dynamic switching system based on digital twins according to claim 3, characterized in that, The edge computing gateway also includes: The wavefront evolution module is configured to generate the virtual wavefront at the source node when the medium switching command is received; The wavefront evolution module is further configured to: use state space matrix multiplication, combined with the wavefront state matrix and control input matrix at the current moment, to calculate in real time the predicted arrival time of the virtual wavefront to each node in the topological wavefront directed graph.

5. The milk filling dynamic switching system based on digital twins according to claim 1, characterized in that, The edge computing gateway also includes: The residual calculation module is configured to calculate the time residual between the actual arrival time and the predicted arrival time; The parameter correction module is configured to determine the relationship between the absolute value of the time residual and a preset residual threshold. If the absolute value of the time residual is greater than the preset residual threshold, the time residual is backpropagated using an extended Kalman filter to correct the viscosity coefficient parameter in the fluid property parameters in real time. If the absolute value of the time residual is equal to the preset residual threshold, the viscosity coefficient parameter remains unchanged and a boundary warning signal is output. If the absolute value of the time residual is less than the preset residual threshold, the viscosity coefficient parameter remains unchanged and a normal operation flag is output.

6. The milk filling dynamic switching system based on digital twins according to claim 1, characterized in that, The edge computing gateway also includes: The control generation module is configured to calculate the cut-off time point corresponding to each of the end effectors based on the predicted arrival time, and combine all the cut-off time points to generate the asynchronous time offset vector. The instruction issuing module is configured to issue the asynchronous time offset vector to the execution controller; The execution controller is configured to trigger the opening and closing actions of the corresponding end effector at different absolute timestamps based on the asynchronous time offset vector, so as to cut the wavefront of the mixed region.

7. The milk filling dynamic switching system based on digital twins according to claim 5, characterized in that, The edge computing gateway also includes: The fatigue diagnosis module is configured to statistically analyze the time residuals of specific nodes within a preset historical period; The fatigue diagnosis module is further configured to calculate the mean and variance of the time residuals, and determine the relationship between the mean and a preset residual mean threshold, and between the variance and a preset variance threshold. If the mean is greater than the preset residual mean threshold and the fluctuation variance is less than the preset variance threshold, the physical component corresponding to the specific node is determined to be in an abnormal physical state, and equipment maintenance work order information is generated. If the fluctuation variance is greater than or equal to the preset variance threshold, it is determined to be fluid fluctuation interference, and a fluid fluctuation data label is generated.

8. The dynamic switching system for milk filling based on digital twins according to any one of claims 1-7, characterized in that, The fluid state data includes milk temperature data; The driving equipment operating data includes the real-time speed data of the variable frequency feed pump; The execution controller is a programmable logic controller; The end effector is a pneumatic directional valve array; The status detection sensor is a high-frequency conductivity sensor installed at the bifurcation point of the main pipeline.

9. The milk filling dynamic switching system based on digital twins according to claim 8, characterized in that, The system is limited to continuous flow switching control of incompressible fluids in closed pipe networks, and the control is achieved by changing the duty cycle or absolute timestamp of the actuator in the programmable logic controller.

10. The milk filling dynamic switching system based on digital twins according to claim 8, characterized in that, The specific determination condition for detecting the true arrival time using the high-frequency conductivity sensor is as follows: perform a first-order difference operation on the received conductivity time series stream to obtain the real-time rate of change. When the absolute value of the real-time rate of change crosses the preset threshold for the first time, and the exceeding state continues for a set digital de-jittering filter window period, the determination time that meets this timing condition is locked as the true arrival time.

Images