Intelligent monitoring method for casting production based on internet of things

Abstract

The application provides a foundry production intelligent monitoring method based on the Internet of Things, comprising: if the local difference of the casting liquid viscosity exceeds the preset bubble generation critical viscosity, the bubble generation probability and the bubble size distribution range are determined; if the local difference of the casting liquid viscosity does not exceed the preset bubble generation critical viscosity, the current gating opening timing and the flow distribution are maintained; the multi-gating opening timing is optimized according to the bubble generation probability and the bubble size distribution range, the opening sequence and the gating opening timing deviation of each gating are determined, and the optimized opening timing scheme is generated; the flow distribution proportion of each gating is dynamically adjusted according to the optimized opening timing scheme, the flow distribution initial scheme is generated; the casting liquid flow state feedback data after the execution of the flow distribution initial scheme is obtained, the updated flow distribution scheme is generated; and the foundry filling integrity and the internal stress distribution caused by the pressure fluctuation of the stagnation area are calculated according to the updated flow distribution scheme, so that the internal defect prediction result is obtained.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an intelligent monitoring method for casting production based on the Internet of Things. Background Technology

[0002] Casting production is a core area of ​​manufacturing, directly impacting the quality and efficiency of industries such as machinery, aerospace, and automobiles. The introduction of intelligent monitoring technology has greatly improved production precision and stability. However, existing methods often struggle to adapt to dynamically changing process conditions when dealing with molten casting flow control under complex mold structures, leading to unstable casting quality. In particular, when mold geometry or molten casting viscosity changes, traditional monitoring systems struggle to analyze and adjust process parameters accurately in real time, often resulting in delayed response or coarse control that affects casting forming. In V-process casting, flow distribution within a multi-gating system is a critical technical aspect. Changes in mold geometry directly affect the flow path and velocity distribution of the molten casting. If the flow ratio and opening sequence of each gating point cannot be dynamically adjusted, uneven filling in localized areas may occur. For example, in complex molds, bubbles may form in certain areas due to molten casting stagnation, leading to internal defects, or even stress concentration due to unbalanced filling rates. This uneven flow distribution stems not only from the complexity of the mold structure but is also closely related to the dynamic changes in the molten casting flow characteristics. Fluctuations in flow characteristics exacerbate nonlinear changes in flow distribution, rendering traditional fixed distribution strategies ineffective and hindering the achievement of uniform filling and defect suppression. Therefore, a key challenge in improving casting quality lies in how to utilize IoT technology to analyze mold geometry and molten casting flow characteristics in real time, and dynamically adjust the flow distribution ratio and opening sequence of multi-gate systems to balance filling rates and reduce bubble formation and stress concentration. Summary of the Invention

[0003] This invention provides an intelligent monitoring method for casting production based on the Internet of Things, mainly including:

[0004] Preprocess the mold geometry data and molten casting viscosity distribution data, determine the temperature gradient and local viscosity difference of the stagnant region based on the extracted molten casting flow parameters, analyze the geometric constraints of the stagnant region of the mold structure, and obtain the local pressure gradient of the gate in each gate region.

[0005] If the local difference in the viscosity of the molten casting exceeds the preset critical viscosity for bubble formation, the probability of bubble formation and the range of bubble size distribution are determined. If the local difference in the viscosity of the molten casting does not exceed the preset critical viscosity for bubble formation, the current gate opening sequence and flow distribution are maintained.

[0006] The opening sequence of multiple gates is optimized based on the probability of bubble generation and the distribution range of bubble size. The opening order of each gate and the gate opening sequence deviation are determined, and an optimized opening sequence scheme is generated.

[0007] The flow distribution ratio of each gate is dynamically adjusted according to the optimized start-up timing scheme to generate an initial flow distribution scheme.

[0008] Obtain the molten casting flow feedback data after executing the initial flow allocation scheme, and generate an updated flow allocation scheme;

[0009] Based on the updated flow distribution scheme, the internal stress distribution caused by the casting filling integrity and pressure fluctuations in the stagnant region is calculated, and the internal defect prediction results are obtained.

[0010] If the bubble size distribution range in the internal defect prediction results exceeds the preset threshold, the gate opening timing deviation and flow distribution ratio are iteratively adjusted to obtain optimized casting parameters. If the bubble size distribution range does not exceed the preset threshold, the casting parameters are confirmed and the final casting scheme is output.

[0011] Furthermore, the preprocessed mold geometry data and molten casting viscosity distribution data, based on the extracted molten casting flow parameters, determine the temperature gradient of the stagnant region and the local viscosity differences of the molten casting, and analyze the geometric constraints of the stagnant region of the mold structure to obtain the local pressure gradient of each gating region, including:

[0012] Data on the internal temperature field and molten casting flow velocity are collected to generate a temperature distribution matrix and a velocity distribution matrix. The temperature change rate of adjacent regions is calculated based on the temperature distribution matrix to generate a temperature gradient in the stagnant region. The location coordinates of the stagnant region are identified based on the velocity distribution matrix. Based on the location coordinates of the stagnant region, the channel cross-sectional area and turning angle are extracted from the 3D model of the mold to generate a geometric feature vector of the mold. The molten casting flow velocity data and the temperature distribution matrix are used to generate the molten casting flow parameters. Based on the geometric feature vector of the mold and the molten casting flow parameters, a feature matrix is ​​constructed. The feature matrix is ​​then processed using a convolutional neural network to generate the local pressure gradient of each gating region.

[0013] Furthermore, the step of constructing a feature matrix based on the casting geometric feature vector and the casting flow parameters, and processing the feature matrix using a convolutional neural network to generate the local pressure gradient of each gating region includes:

[0014] Based on the channel cross-sectional area and turning angle in the casting geometric feature vector, a cross-sectional area change rate and a turning angle resistance factor are generated; based on the viscosity value in the casting flow parameters, a viscosity gradient component is generated; based on the cross-sectional area change rate, the turning angle resistance factor, and the viscosity gradient component, the feature matrix is ​​constructed; the feature matrix is ​​convolved and pooled using a convolutional neural network to generate a flow resistance coefficient; based on the flow resistance coefficient and the casting inlet pressure, the local pressure gradient of each gating region is generated.

[0015] Furthermore, the step of optimizing the multi-gate opening sequence based on the bubble generation probability and bubble size distribution range, and determining the opening order of each gate and the gate opening sequence deviation, includes:

[0016] Based on the bubble generation probability and the bubble size distribution range, the expected bubble volume at each gate location is calculated; based on the expected bubble volume, an optimization objective function is constructed; the gate opening sequence is encoded using a genetic algorithm to generate a timing scheme population; based on the timing scheme population, the trajectory of the casting front end position during the casting filling process is calculated; based on the casting front end position trajectory and the bubble generation probability, the spatial superposition volume is calculated to generate a fitness value; based on the fitness value, the timing scheme population is iteratively updated to generate the opening sequence of each gate; based on the opening sequence of each gate, the opening time interval between adjacent gates is calculated to generate the opening sequence deviation.

[0017] Furthermore, the step of calculating the opening time interval between adjacent gates based on the opening sequence of each gate, and generating the opening timing deviation, includes:

[0018] Based on the opening sequence of each gate, calculate the opening time interval between adjacent gates; based on the opening time interval, analyze the velocity vector angle when the molten casting merges; based on the velocity vector angle, adjust the opening time of the gates; based on the adjusted time and the total filling time, generate the opening timing deviation.

[0019] Furthermore, the step of dynamically adjusting the flow distribution ratio of each gate according to the optimized start-up timing scheme to generate an initial flow distribution scheme includes:

[0020] Based on the optimized start-up sequence scheme, extract the time series data of the molten casting velocity in this area, determine the amplitude of the molten casting velocity pulsation, and solve for the basic flow rate value of each gate; based on the basic flow rate value of each gate and the time interval in the start-up sequence, calculate the total flow rate of the gates that are opened simultaneously, and determine the initial flow rate distribution scheme.

[0021] Furthermore, the step of obtaining the molten casting flow feedback data after executing the initial flow allocation scheme and generating an updated flow allocation scheme includes:

[0022] Based on the molten casting flow feedback data, the amplitude of molten casting velocity pulsation is calculated; based on the difference between the amplitude of molten casting velocity pulsation and the advance distance of the molten casting front, the filling rate balance state is determined; based on the filling rate balance state, a reinforcement learning algorithm is used to adjust the flow distribution ratio of each gate, and the updated flow distribution scheme is generated.

[0023] Furthermore, the calculation of the internal stress distribution caused by the casting filling integrity and pressure fluctuations in the stagnant region based on the updated flow distribution scheme, to obtain the internal defect prediction results, includes:

[0024] Based on the updated flow distribution scheme, a three-dimensional finite element model of the casting is constructed; based on the three-dimensional finite element model of the casting, the flow velocity and pressure distribution of the molten casting are calculated, and filling rate distribution data is generated; based on the filling rate distribution data and the temperature gradient of the stagnant region, the comprehensive stress value is calculated; based on the comprehensive stress value and the temperature change rate of the molten casting solidification process, an internal defect prediction result containing the bubble generation probability and the bubble size distribution range is generated.

[0025] Furthermore, the step of generating an internal defect prediction result, including the bubble formation probability and the bubble size distribution range, based on the comprehensive stress value and the temperature change rate during the solidification process of the casting liquid, includes:

[0026] Based on the comprehensive stress value, the number of units exceeding the critical nucleation stress threshold is counted to generate the bubble generation probability; based on the difference between the comprehensive stress value and the critical nucleation stress threshold and the temperature change rate during the solidification process of the casting liquid, the bubble diameter range is estimated to generate the bubble size distribution range.

[0027] Furthermore, if the bubble size distribution range in the internal defect prediction results exceeds a preset threshold, the gate opening timing deviation and flow distribution ratio are iteratively adjusted to obtain optimized casting parameters, including:

[0028] If the maximum value of the bubble size distribution range in the internal defect prediction results exceeds the preset size threshold, the current gate opening timing deviation value and flow distribution ratio data are extracted. Based on the location information of the bubble size exceeding the standard area, the opening time and flow distribution ratio of the corresponding gate are adjusted to obtain the optimized casting parameters.

[0029] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0030] This invention discloses an intelligent monitoring method for casting production based on the Internet of Things (IoT). Addressing the issues of bubble formation and uneven filling caused by differences in molten casting viscosity and stagnant regions during casting, the method acquires real-time data on mold geometry and molten casting viscosity using IoT sensors. After data cleaning and feature extraction, feature vectors and flow parameters are generated. A convolutional neural network is then used to analyze the geometric constraints of stagnant regions and calculate the gate pressure gradient. If the viscosity difference exceeds the limit, fluid dynamics simulation is used to assess turbulence intensity and pressure fluctuations, predicting the probability of bubble formation and size distribution. If the difference is within the limit, the gate timing and flow rate are maintained. A genetic algorithm is used to optimize the gate opening timing, and linear programming is used to dynamically adjust the flow rate distribution. Combined with real-time turbulence intensity and flow velocity fluctuations, an initial scheme is generated and executed via IoT control. If feedback indicates uneven filling, reinforcement learning is used to re-optimize the flow rate distribution. Finally, finite element analysis is used to predict defects and iteratively optimize casting parameters to ensure that the bubble size is controlled within a threshold. This invention significantly reduces bubble defects in castings and improves filling uniformity and casting quality. Attached Figure Description

[0031] Figure 1 This is a flowchart of an intelligent monitoring method for casting production based on the Internet of Things according to the present invention. Detailed Implementation

[0032] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0033] like Figure 1 This embodiment of an intelligent monitoring method for casting production based on the Internet of Things may specifically include:

[0034] S101. Preprocess the geometric data of the mold and the viscosity distribution data of the molten casting. Based on the extracted molten casting flow parameters, determine the temperature gradient of the stagnant region and the local difference in molten casting viscosity. Analyze the geometric constraints of the stagnant region of the mold structure to obtain the local pressure gradient of each gate region.

[0035] IoT sensors are used to collect temperature field data and molten casting flow velocity data inside the mold. Noise filtering and outlier removal are applied to the raw data to obtain temperature and velocity distribution matrices for each region of the mold. The temperature change rate of adjacent regions is calculated based on the temperature distribution matrix, and regions with velocities below a preset threshold are identified based on the velocity distribution matrix, thus determining the coordinates of stagnant flow areas. For the coordinates of these stagnant flow areas, geometric parameters, including channel cross-sectional area, turning angle, and wall roughness, are extracted from the 3D model of the mold. The viscosity value at each point within the stagnant flow area is calculated using the relationship between molten casting viscosity and temperature: μ = μ0exp(E / RT), where μ0 is the reference viscosity, E is the activation energy, R is the gas constant, and T is the temperature. This yields a viscosity distribution map of the stagnant flow area. The viscosity gradient vector is determined based on the ratio of the viscosity difference to the distance between adjacent points in the viscosity distribution map. Based on the channel cross-sectional area, turning angle, wall roughness, and viscosity gradient vector of the stagnant region, a feature matrix is ​​constructed, containing the cross-sectional area change rate, turning resistance factor, and viscosity gradient component. The turning resistance factor is the ratio of the turning angle to 90 degrees. A convolutional neural network is used to perform convolution operations and pooling on the feature matrix to output the flow resistance coefficient for each gate region. Based on the flow resistance coefficient and the casting inlet pressure, the pressure loss formula ΔP = ρv is used. 2 ξ / 2 is used to calculate the pressure loss at the gate, where ρ is the density of the molten casting, v is the flow velocity, and ξ is the flow resistance coefficient. If the pressure loss at a gate exceeds the preset pressure threshold, the gate opening parameter is reduced and the pressure loss is recalculated. The local pressure gradient of each gate area is determined based on the ratio of the adjusted pressure loss to the gate spacing.

[0036] Specifically, in the casting production process, the placement of IoT sensors needs to fully consider the complex environment inside the mold.

[0037] Specifically, temperature sensors typically employ thermocouples, distributed in a grid pattern at key locations in the mold, such as near the gate, at cavity corners, and in thick-walled areas. Each sensor collects temperature data at a fixed frequency. Flow velocity sensors utilize the ultrasonic Doppler principle, measuring the flow velocity of the molten casting by emitting and receiving ultrasonic signals. The raw data collected by these sensors often contains noise from electromagnetic interference and outliers caused by equipment malfunctions. Therefore, a sliding window midpoint filtering method is needed to remove noise, while outliers are identified and eliminated using the 3σ criterion. The temperature distribution matrix is ​​constructed based on the spatial location and measurements of the sensors, using interpolation algorithms to convert discrete measurement point data into a continuous temperature field.

[0038] In one possible implementation, the rate of temperature change between adjacent regions is obtained by calculating the temperature gradient, which helps identify the direction and speed of heat transfer. The velocity distribution matrix reflects the flow state of the molten casting at different locations. When the flow velocity in a certain region remains below a preset threshold, it indicates that there may be flow stagnation in that region, which is the basis for determining stagnant regions. The formation of stagnant regions is often closely related to the geometry of the mold.

[0039] It should be noted that abrupt changes in the channel cross-sectional area lead to increased flow resistance; the larger the turning angle, the more severe the energy loss when the molten metal changes flow direction; and the wall roughness directly affects the formation and development of the boundary layer. The relationship between molten metal viscosity and temperature, μ = μ0exp(E / RT), describes the exponential effect of temperature on viscosity, where the reference viscosity μ0 is the viscosity value at standard temperature, and the activation energy E reflects the sensitivity of viscosity to temperature changes. By calculating the viscosity values ​​at each point in the stagnant region, a viscosity distribution map can be generated. Then, the viscosity gradient vector can be obtained by the ratio of the viscosity difference between adjacent points to their distance; this vector indicates the direction of the most drastic viscosity change. Convolutional neural networks have unique advantages in processing these features.

[0040] Preferably, the input layer of the network receives a feature matrix containing the rate of change of cross-sectional area, the corner drag factor, and the viscosity gradient component. The corner drag factor is defined as the ratio of the actual corner angle to 90 degrees, reflecting the degree to which the corner impedes the flow. Convolutional layers extract local features through multiple filters, while pooling layers reduce data dimensionality and retain key information. The final output layer generates the flow resistance coefficient for each gate region. This coefficient comprehensively reflects the influence of geometric constraints and viscosity distribution on the flow. Pressure loss is calculated using ΔP = ρv. 2 The formula ξ / 2 is used, where the molten casting density ρ and flow velocity v are measured values, and the flow resistance coefficient ξ is output by a neural network. When the pressure loss exceeds a preset threshold, it indicates that the flow at that gate is severely obstructed, requiring a reduction in the gate opening parameter to decrease the flow velocity and thus reduce pressure loss. The adjusted pressure distribution is more uniform, which helps improve casting quality. The local pressure gradient of each gate is determined by the ratio of pressure loss to gate spacing. This gradient reflects the rate of change of pressure in space, providing an important basis for subsequent optimization of the gating system.

[0041] S102. If the local difference in the viscosity of the molten casting exceeds the preset critical viscosity for bubble generation, then determine the probability of bubble generation and the range of bubble size distribution. If the local difference in the viscosity of the molten casting does not exceed the preset critical viscosity for bubble generation, then maintain the current gate opening sequence and flow distribution.

[0042] The viscosity values ​​of the molten casting at different locations within the stagnant region are obtained, and the viscosity difference between adjacent measuring points is calculated to obtain the local viscosity difference value of the molten casting. If the local viscosity difference value exceeds the preset critical viscosity for bubble formation, the temperature gradient data and viscosity distribution data of the stagnant region are extracted to construct a fluid parameter matrix containing temperature, viscosity, and velocity fields. Based on the temperature, viscosity, and velocity field data in the fluid parameter matrix, a fluid dynamics simulation model based on the Navier-Stokes equations is used for numerical solution. By dividing the computational domain into control volumes, the laws of mass and momentum conservation are applied to each control volume to calculate the velocity fluctuation intensity at the boundary between the stagnant and normal flow regions, thus obtaining the turbulence intensity at the stagnant boundary. Simultaneously, the pressure correction value is calculated using velocity divergence to obtain the pressure fluctuation amplitude in the stagnant region. Based on the turbulence intensity at the stagnant boundary and the pressure fluctuation amplitude in the stagnant region, the bubble nucleation rate J = A × exp(-B / ΔP) is calculated. 2 The formula is: A is the frequency factor, B is the energy barrier parameter related to the surface tension of the molten casting, and ΔP is the pressure fluctuation amplitude. The probability of bubble formation is determined, and the bubble size distribution range is obtained based on the relationship between turbulence intensity and bubble diameter: d = C × (I^0.5) × t, where C is the growth coefficient, I is the turbulence intensity, and t is the growth time. If the local difference in molten casting viscosity does not exceed the preset critical viscosity for bubble formation, the current opening sequence of each gate remains unchanged, the existing distribution ratio of molten casting flow among each gate is maintained, and the original casting process parameters are continued.

[0043] Specifically, measuring the local differences in the viscosity of the molten casting involves several key steps.

[0044] Specifically, viscosity sensors arranged within the stagnant flow region acquire real-time viscosity values ​​using rotational or vibrational measurement principles. These sensors are typically distributed at key locations with a 5-centimeter spacing. The viscosity difference between adjacent measuring points reflects the non-uniformity of the molten casting flow. When this difference exceeds a preset critical viscosity for bubble formation, it indicates a significant flow anomaly in a localized area. This critical viscosity value is usually determined based on the characteristics of the molten casting material and historical production data; for aluminum alloy casting, the critical viscosity difference is generally set at 0.8 Pa·s. The construction of the fluid parameter matrix forms the basis for subsequent simulation calculations.

[0045] In one possible implementation, the temperature field data originates from real-time measurements by a thermocouple array, with the temperature value at each measuring point corresponding to its spatial coordinates to form a three-dimensional temperature distribution. The viscosity field uses an interpolation algorithm to extend the discrete viscosity measurements into a continuous distribution. The velocity field acquires the three-dimensional velocity components at each point using an ultrasonic Doppler velocimeter. These three fields are spatially coupled and jointly determine the flow state of the casting liquid. The Navier-Stokes equations describe the fundamental laws of fluid motion, including the continuity equation and the momentum equation.

[0046] It should be noted that in the numerical solution process, the computational domain is divided into numerous hexahedral control volumes, each representing a computational unit. For each control volume, mass conservation requires that the inflow and outflow masses be equal, while momentum conservation considers the balance of pressure, viscous forces, and volume forces. The velocity fluctuation intensity is obtained by calculating the difference between the instantaneous velocity and the time-averaged velocity; this parameter directly reflects the degree of flow instability. The stagnation boundary is the interface between the stagnation region and the normal flow region, and the turbulence intensity at this location has a significant impact on bubble formation. The calculation of pressure fluctuation amplitude is based on velocity divergence; when the fluid is incompressible, the velocity divergence should be zero, and any deviation will result in pressure correction. This pressure fluctuation provides the energy basis for bubble nucleation. The bubble nucleation rate formula is J=A×exp(-B / ΔP). 2 In the formula, frequency factor A represents the frequency of nucleation events that may occur per unit time, which is usually related to temperature and casting properties; energy barrier parameter B is directly related to surface tension. The greater the surface tension, the higher the energy required to form a new interface, and the more difficult nucleation. The square of the pressure fluctuation amplitude ΔP appears in the denominator, indicating that pressure fluctuation has a significant promoting effect on nucleation. The evolution of bubble size is affected by turbulence intensity. The formula d=C×(I^0.5)×t shows that the bubble diameter is proportional to the square root of the turbulence intensity. The growth coefficient C integrates the effects of casting properties and environmental conditions. Turbulence accelerates bubble growth by enhancing the mass transfer process. By tracking the bubble size at different times, a complete size distribution range can be obtained, which is of great significance for assessing casting quality defects. When the local viscosity difference does not exceed the critical value, it indicates that the flow state is relatively stable and no significant bubble defects will occur. At this time, maintaining the existing pouring parameters, including the opening sequence of each gate and the flow distribution ratio, can ensure the continuity and stability of the casting process and avoid new flow disturbances caused by parameter adjustments.

[0047] S103. Optimize the opening sequence of multiple gates based on the probability of bubble generation and the distribution range of bubble size, determine the opening order of each gate and the gate opening sequence deviation, and generate an optimized opening sequence scheme.

[0048] Based on the bubble generation probability and bubble size distribution range, the expected bubble volume is obtained by multiplying the bubble generation probability at each gate location by the corresponding size. A function with the sum of expected bubble volumes as the optimization objective is constructed. A genetic algorithm is used to encode the gate opening sequence, with each gene representing the opening time of a gate. New timing scheme populations are generated through gene recombination and mutation operations. For each individual in the timing scheme population, the molten metal filling process is calculated based on its encoded gate opening time. The trajectory of the molten metal front position over time is obtained through flow simulation. The area traversed by the molten metal front is spatially superimposed with the area where the bubble generation probability is greater than a preset threshold. The overlapping volume is calculated as the fitness value of the timing scheme. The timing scheme with the smaller fitness value is selected to enter the next generation population. After a preset number of iterations, the opening order of each gate is determined. Based on the determined gate opening sequence, the opening time interval between the i-th gate and the (i+1)-th gate is calculated. The velocity vector angle and pressure difference when the two streams of molten casting merge are analyzed through flow simulation. If the angle exceeds a preset angle threshold or the pressure difference exceeds a preset pressure threshold, the gate opening time is adjusted. The difference between the adjusted opening time and the time obtained by dividing the total filling time equally is the gate opening sequence deviation value. An optimized opening sequence scheme containing the specific opening time of each gate and the timing deviation value is generated.

[0049] Specifically, the calculation of the expected value of the bubble volume integrates principles of probability theory and geometry.

[0050] Specifically, when the probability of bubble generation at a certain gate location is 0.7, and the predicted average bubble diameter is 2 mm, the volume of a single bubble is approximately 4.2 cubic millimeters. Therefore, the expected bubble volume at that location is 2.94 cubic millimeters. By performing similar calculations for all gate locations and summing the results, the total expected bubble volume for the entire mold is obtained. The smaller this total value, the fewer bubble defects the casting scheme produces, and the better the casting quality. Genetic algorithms demonstrate unique advantages in optimizing gate opening sequence.

[0051] In one possible implementation, assume the mold has six gates. Each gene locus is represented by an integer from 0 to 100, indicating the relative opening time of that gate, where 0 represents the earliest opening and 100 represents the latest opening. A complete chromosome might be [0, 25, 15, 60, 80, 45], representing that the first gate opens first, followed by the 3rd, 2nd, 6th, 4th, and 5th gates in sequence. Gene recombination generates new individuals by exchanging partial gene segments from two superior individuals, while mutation randomly changes the value of a gene locus, increasing population diversity. Tracking the position of the molten metal front is crucial for evaluating the filling quality.

[0052] It should be noted that the flow simulation is based on the mass and momentum conservation equations, and uses numerical methods to calculate the position distribution of the molten casting at each time step. The molten casting front is the interface between the molten casting and air, and its position advances continuously over time. When multiple gates open in a specific sequence, multiple molten casting fronts are formed, and the way these fronts converge directly affects bubble generation. The calculation of the fitness value reflects the quantification of the optimization objective. When the molten casting front passes through a region with a high probability of bubble generation, that region is more prone to bubble defects. By calculating the spatial overlap volume between the molten casting front trajectory and the high-probability region, the merits of the timing scheme can be evaluated. The smaller the overlap volume, the more the molten casting avoids the dangerous region, and the better the fitness value. After multiple generations of evolution, the timing scheme represented by the individual with the optimal fitness value is selected. The determination of the gate opening time interval involves fluid dynamics analysis.

[0053] For example, when the first stream of molten casting flows at a speed of 1.5 m / s and the second stream flows in from the opposite direction at a speed of 1.2 m / s, the angle between their velocity vectors reflects the intensity of the confluence. The larger the angle, the stronger the turbulence, and the easier it is for gas to be entrained and form bubbles. Simultaneously, the pressure difference between the two streams of molten casting is also an important indicator; an excessively large pressure difference can lead to negative pressure in the low-pressure area, promoting bubble formation. The introduction of timing deviation values ​​provides a flexible adjustment mechanism. Theoretically, if the total filling time is 12 seconds, and the six gates open evenly, the interval between each gate should be 2 seconds. However, after actual optimization, some gates may need to open earlier or later to avoid unfavorable flow interference.

[0054] For example, the third gate was originally scheduled to open at the fourth second, but to avoid violent collision with the molten casting from the second gate, it was adjusted to open at the fourth and fifth second. This 0.5-second difference is the timing deviation value for that gate.

[0055] S104. Dynamically adjust the flow distribution ratio of each gate according to the optimized start-up timing scheme to generate an initial flow distribution scheme.

[0056] Based on the opening time of each gate in the optimized opening sequence scheme, velocity distribution data of each gate cross-section is acquired through real-time sensors. The ratio of the velocity standard deviation to the average velocity is calculated to obtain the turbulence intensity of the gate cross-section. Simultaneously, the area where the probability of bubble generation is greater than a preset probability threshold is determined as the target monitoring area. Time series data of the casting velocity in this area are extracted, and the root mean square value of the velocity difference between adjacent time moments is calculated to determine the casting velocity pulsation amplitude. Based on the turbulence intensity of the gate cross-section and the casting velocity pulsation amplitude, the upper limit of the flow rate of gates with turbulence intensity exceeding a preset intensity threshold is set to be reduced. Flow rate allocation constraints are constructed, including total flow rate conservation constraints and maximum flow rate limit constraints for each gate. The objective function min∑(Ti×Qi) is established using linear programming, where Ti is the turbulence intensity of the i-th gate and Qi is the flow rate of the i-th gate. The basic flow rate value of each gate is obtained by solving the simplex method. Based on the basic flow rate value of each gate and the time interval in the opening sequence, the total flow rate of the gates opened at the same time is calculated. If the total flow rate exceeds the maximum flow rate that the mold can withstand in a certain period, the excess ratio is calculated and the flow rate of each gate in that period is reduced synchronously according to the ratio. The correspondence between the adjusted flow rate value and the gate number constitutes the initial flow rate distribution scheme.

[0057] Specifically, the measurement of turbulence intensity at the gate section is based on turbulence theory in fluid mechanics.

[0058] Specifically, real-time sensors employ hot-wire anemometers or laser Doppler velocimeters to collect instantaneous velocities at different locations on the gating section at high frequency. By calculating the standard deviation of the velocity at the same measuring point within a time window, the degree of velocity fluctuation can be quantified. When the ratio of the standard deviation to the average velocity reaches 0.3, it indicates the presence of significant turbulence at that location. This turbulence increases the chance of mixing between the molten casting and air, raising the risk of bubble formation. The target monitoring area is determined based on the spatial distribution of the bubble formation probability.

[0059] In one possible implementation, an area is marked as requiring close monitoring when the probability of bubble formation exceeds a threshold of 0.6. Within these areas, velocity sensors are densely arranged to collect data at a sampling frequency of 100 Hz. By analyzing velocity data from consecutive time intervals, the root mean square value of these differences is calculated as the velocity pulsation amplitude, reflecting the degree of flow instability. The application of linear programming in flow optimization demonstrates the powerful capabilities of mathematical optimization.

[0060] It should be noted that the objective function min∑(Ti×Qi) physically means minimizing turbulence energy, where the product of turbulence intensity Ti and flow rate Qi represents the turbulence energy generated by that gate. Constraints ensure physical feasibility: the total flow rate conservation constraint guarantees that the sum of the flow rates of all gates equals the total supply of molten casting; the maximum flow rate limit constraint prevents excessive flow rates from a single gate, which could lead to jetting. The simplex method gradually approximates the optimal solution by moving between vertices of the feasible region. The dynamic adjustment mechanism of the flow rate upper limit enhances the flexibility of control. When the turbulence intensity of a gate exceeds a preset threshold of 0.4, its maximum allowable flow rate decreases linearly.

[0061] For example, a gate with a maximum flow rate of 10 liters per second can have its maximum flow rate limit adjusted to 8 liters per second when the turbulence intensity reaches 0.5. This adjustment avoids excessive flow rate under high turbulence conditions, which exacerbates bubble formation. Coordinating the flow rates of simultaneously open gates is a key challenge in multi-gate systems. In actual casting processes, depending on the optimized timing scheme, multiple gates may operate simultaneously.

[0062] For example, during the time interval from the 3rd to the 5th second, three gates open simultaneously, with base flow rates of 8, 6, and 7 liters per second, respectively, totaling 21 liters per second. If the maximum flow rate the mold can withstand is 18 liters per second, the excess is 21 / 18 = 1.17. In this case, the flow rates of the three gates need to be simultaneously reduced to 6.8, 5.1, and 6.0 liters per second, maintaining the original relative proportions. The formation of the initial flow distribution scheme marks a significant milestone in the optimization process. This scheme includes the flow rate values ​​of each gate at different times, forming a three-dimensional data table of time, gate, and flow rate.

[0063] S105. Obtain the molten metal flow feedback data after executing the initial flow allocation scheme, and generate an updated flow allocation scheme.

[0064] The IoT control module converts the flow rate values ​​in the initial flow distribution scheme into valve opening values ​​for each gate actuator. The conversion relationship is: opening value = flow rate value / flow coefficient, where the flow coefficient is determined by the valve characteristic curve. The actuator adjusts the gate valve position according to the opening value, obtaining the actual flow rate value, casting front position, and velocity distribution data for each gate after execution, thus forming casting flow feedback data. Based on the velocity distribution data in the casting flow feedback data, the ratio of the time standard deviation to the average value of the velocity at each measuring point is calculated to obtain the casting velocity pulsation amplitude. By comparing the difference in the advance distance at the casting front position in different areas with a preset distance threshold, it is determined whether the filling rate is balanced. If the difference in the advance distance does not exceed the preset distance threshold and the velocity pulsation amplitude does not exceed the preset pulsation threshold, the filling rate is considered balanced, and the current flow distribution scheme continues to be executed. If the difference in the advance distance exceeds the preset distance threshold, a reinforcement learning algorithm is used to construct an adjustment mechanism. The actual flow rate value and the difference in the advance distance at each gate are used as state inputs, the flow adjustment amplitude is defined as the action space, and the reward value R = -(D 2 +P×G), where D is the difference in advance distance, P is the penalty coefficient, and G is the local pressure gradient of the gate. The action value function is iteratively updated through the Q learning algorithm, the flow adjustment range that maximizes the cumulative reward is selected, and the flow value of each gate is corrected according to the adjustment range to generate the updated flow allocation scheme.

[0065] Specifically, after receiving the flow distribution plan, the IoT control module needs to convert the abstract flow rate value into a control signal that the actuator can understand. The flow coefficient is a key parameter in this conversion, reflecting the non-linear relationship between valve opening and actual flow rate. For butterfly valves, the flow coefficient exhibits approximately linear changes when the opening is between 30 and 60 degrees, but shows significant non-linear characteristics when approaching full open or full closed. By consulting a pre-calibrated valve characteristic curve, the precise opening value corresponding to a specific flow rate can be determined. The actuator's response characteristics directly affect control accuracy.

[0066] In one possible implementation, the electric actuator drives the valve stem to rotate via a stepper motor, with each pulse signal corresponding to a rotation angle of 0.1 degrees. When the control module issues an opening adjustment command, the actuator not only completes the action but also receives feedback of the actual opening value via a position encoder. Simultaneously, an electromagnetic flowmeter installed downstream of the gate measures the molten casting flow rate in real time, and an ultrasonic sensor array monitors the position of the molten casting leading edge. These data collectively constitute complete flow feedback information. The calculation of the molten casting flow velocity pulsation amplitude reveals the flow stability characteristics.

[0067] It should be noted that within the sampling time window, each measuring point may collect hundreds of instantaneous flow velocity values. By calculating the standard deviation of these values, the degree of flow velocity fluctuation can be quantified. When the ratio of the standard deviation to the average flow velocity exceeds 0.25, it indicates a significant flow instability at that location. This instability often foreshadows potential filling defects. The difference in advance distance is a direct indicator for assessing filling uniformity.

[0068] For example, at a certain moment, the molten metal front in the left region advances by 120 mm, while the right region only advances by 95 mm, a difference of 25 mm. If the preset distance threshold is 20 mm, then uneven filling is determined. This unevenness leads to uneven stress distribution within the casting, affecting the final quality. Reinforcement learning demonstrates adaptive optimization capabilities in flow rate adjustment. The Q-learning algorithm guides decision-making by constructing a state-action value table. The state space contains the current flow rate value and filling unevenness of each gate, while the action space is defined as discrete increments or decrements in flow rate, such as increasing by 10%, keeping it unchanged, or decreasing by 10%. The reward function R = -(D 2 The design of +P×G) embodies the idea of ​​multi-objective optimization, where D 2 The algorithm penalizes uneven filling, and the P×G term limits the turbulence risk caused by excessive pressure gradients. The algorithm iteratively learns to gradually optimize the control strategy. In the initial stage, the algorithm may try various flow combinations, accumulating experience through trial and error. As the number of learning iterations increases, the Q-value table gradually converges, and the algorithm can quickly identify the optimal action.

[0069] For example, when a lag in filling is detected in a certain area, the algorithm increases the flow rate of the corresponding gate while appropriately reducing the flow rate of the area that is filling faster, thus achieving a dynamic balance.

[0070] S106. Calculate the internal stress distribution caused by the filling integrity of the casting and the pressure fluctuation in the stagnant area based on the updated flow distribution scheme, and obtain the prediction results of internal defects.

[0071] Based on the updated flow distribution scheme and the opening sequence of each gate, a three-dimensional finite element model of the casting is constructed. The mold geometry is divided into tetrahedral mesh elements, and corresponding material properties and boundary conditions are assigned to each element node. The flow velocity and pressure distribution of the molten casting within the mold are calculated by solving the continuity equation and momentum equation. The filling rate is determined based on the volume ratio of the molten casting within each element, and the filling rate distribution data at each time point is obtained. Based on the filling rate distribution data, areas with filling rates below a preset filling threshold are identified as incompletely filled areas. Simultaneously, pressure variation data over time is extracted from the stagnant areas as pressure fluctuation input. The solidification shrinkage stress is calculated using the thermoelastic stress calculation formula. By superimposing the mechanical stress generated by pressure fluctuations with the solidification shrinkage stress, the comprehensive stress value at each location inside the casting is obtained, and the location coordinates and peak stress magnitude of the stress concentration area are determined. Based on the comprehensive stress value and the temperature change rate during the solidification process of the casting liquid, the tendency of bubble formation is calculated by judging whether the stress value at each location exceeds the critical nucleation stress threshold of the material. The bubble formation probability is obtained by statistically analyzing the ratio of the number of units exceeding the threshold to the total number of units. Based on the difference between the stress value and the critical value and the duration of action, the bubble diameter range is estimated through the bubble growth kinetic relationship, generating an internal defect prediction result that includes the bubble size distribution range and the bubble formation probability.

[0072] Specifically, the construction of the finite element model is the foundation of casting simulation.

[0073] Specifically, the 3D geometric model needs to be transformed into discrete computational units through mesh generation. Tetrahedral meshes are widely used due to their good adaptability to complex geometries. During mesh generation, thin-walled regions of the mold require denser meshes to capture detailed flow characteristics, while thicker areas can use relatively sparse meshes to save computational resources. Each mesh unit is assigned material properties, including physical parameters such as density, viscosity, and thermal conductivity. The relationship between these parameters and temperature is defined through table lookup or function fitting. The continuity equation and momentum equation are the fundamental equations describing fluid motion.

[0074] In one possible implementation, the continuity equation ensures mass conservation, meaning the mass flowing into a control volume equals the mass flowing out plus the change in internal mass. The momentum equation describes the accelerated motion of the fluid under the influence of pressure gradient, viscous force, and gravity. These partial differential equations are discretized using the finite element method, transforming the continuity problem into a system of algebraic equations for solution. The solution process employs an iterative algorithm until convergence to a predetermined accuracy. The calculation of the filling ratio reveals the distribution of the molten casting within the mold.

[0075] It should be noted that each grid cell may be in three states: completely filled with molten casting, partially filled, or completely empty. The filling rate is defined as the ratio of the volume of molten casting within the cell to the total volume of the cell. A filling rate of 1 indicates complete filling, 0 indicates emptiness, and a value in between indicates that the molten casting front is passing through the cell. By statistically analyzing the filling rates of all cells, a filling process diagram of the entire casting can be obtained, visually displaying the location and size of unfilled areas. The impact of pressure fluctuations on casting quality cannot be ignored. In stagnant regions, due to poor flow, pressure will fluctuate periodically or randomly. This fluctuation is transmitted through the molten casting to the solidified casting shell, generating alternating stress. Thermoelastic stress calculations need to consider the material's coefficient of thermal expansion and elastic modulus. When the casting cools from a high temperature, uneven shrinkage in different parts will generate internal stress. The superposition of mechanical and thermal stresses forms a complex stress field, and stress concentration often occurs at geometric abrupt changes or at thickness boundaries. The judgment of bubble formation is based on the material's nucleation theory. The critical nucleation stress threshold is an inherent property of the material, related to the balance of surface energy and volume energy. When local stress exceeds this threshold, tiny bubble nuclei can overcome surface tension and grow. The greater the difference between the stress value and the critical value, the stronger the driving force for bubble formation, and the larger the resulting bubble size. The duration of stress affects bubble growth dynamics; longer periods of high stress allow for sufficient bubble growth, while short stress peaks may only produce tiny bubbles. The generation of internal defect prediction results integrates multi-faceted computational data. Through statistical analysis, a probability distribution map of bubble formation in different regions can be obtained, with high-probability areas often corresponding to stress concentrations and filling difficulties. The prediction of bubble size distribution range provides a quantitative basis for subsequent quality control, enabling engineers to optimize process parameters and reduce casting defects.

[0076] S107. If the bubble size distribution range in the internal defect prediction result exceeds the preset threshold, then iteratively adjust the gate opening timing deviation and flow distribution ratio to obtain the optimized casting parameters. If the bubble size distribution range does not exceed the preset threshold, then confirm the casting parameters and output the final casting scheme.

[0077] If the maximum value of the bubble size distribution range in the internal defect prediction results exceeds the preset size threshold, the current gate opening timing deviation value and flow distribution ratio data are extracted. Based on the location information of the bubble size exceeding the standard area, the opening time of the corresponding gate is adjusted according to the excess ratio. The adjustment amount is the excess ratio multiplied by the reference time interval. At the same time, the flow distribution ratio of the relevant gate in that area is reduced according to the difference between the bubble size and the threshold. Based on the adjusted gate opening timing and flow distribution ratio, the surface tension data of the casting liquid at different temperatures is obtained, and the rate of change of surface tension with temperature is calculated. Simultaneously, the temperature gradient distribution of the stagnant area is extracted, and the surface tension change rate is multiplied by the temperature gradient to obtain the surface tension spatial change rate. When this rate exceeds the preset rate threshold, the upper limit value of the flow rate of the corresponding gate is reduced inversely proportionally, resulting in a corrected casting parameter set including timing deviation, flow distribution ratio, and upper limit flow rate. If the maximum value of the bubble size distribution range in the internal defect prediction results does not exceed the preset size threshold, the corrected casting parameter set or the initial parameter set is confirmed as the final parameters. The gate number, opening time, flow rate value, and execution time information are compiled, and the final casting scheme is output.

[0078] Specifically, the mechanism for judging whether the bubble size exceeds the standard is a key link in quality control.

[0079] Specifically, when the predicted maximum bubble diameter reaches 3 mm, while the preset size threshold is 2 mm, the exceedance rate is 50%. This exceedance rate directly reflects the degree of deviation of the current process parameters. The baseline time interval is usually set as the total pouring time divided by the number of gates. Assuming a total pouring time of 12 seconds and 6 gates, the baseline interval is 2 seconds. When the bubble count in the area corresponding to a certain gate exceeds the limit by 50%, the opening time of that gate needs to be adjusted by 1 second. Opening it earlier can change the flow field distribution in that area and reduce the risk of bubble formation. The adjustment of the flow distribution ratio follows the compensation principle.

[0080] In one possible implementation, the difference between the bubble size and the threshold directly affects the flow rate adjustment. When the bubble size exceeds the threshold by 1 mm, the flow rate at the corresponding gate needs to be reduced proportionally. This reduction is not a simple linear relationship but takes into account fluid dynamics. Reducing the flow rate can decrease the turbulence intensity in this region, thereby inhibiting further bubble growth. Simultaneously, the flow rate at adjacent gates needs to be appropriately increased to compensate for the overall flow rate balance. Surface tension plays a dual role in the casting process.

[0081] It should be noted that surface tension can both stabilize the bubble interface and affect the flow characteristics of the casting liquid. When the casting liquid temperature drops from 1200K to 1000K, the surface tension may increase from 0.8 N / m to 1.0 N / m, a change rate of 0.001 N / m per Kelvin. When this rate of change is multiplied by the temperature gradient, the resulting spatial rate of change reflects the spatial non-uniformity of surface tension. A high spatial rate of change means that the casting liquid at different locations has different flow resistances. The coupling effect of the temperature gradient and surface tension affects the setting of the upper limit of flow rate. When the spatial rate of change of surface tension exceeds 0.05 N / m... 2 This indicates significant flow instability in the region. In this case, the upper limit of the flow rate at the corresponding gate needs to be reduced inversely.

[0082] For example, a gating gate with an original flow rate limit of 10 liters / second may experience a change in flow rate when the space change rate reaches 0.1 N / m. 2 At that time, the upper limit of the flow rate was adjusted to 5 liters / second. This adjustment prevented excessive flow rate from causing drastic disturbances in the high gradient region. The formation of the revised casting parameter set reflects comprehensive optimization of multiple factors. Each parameter is not isolated; timing deviation affects the filling sequence, flow rate ratio determines the filling speed, and the upper limit of the flow rate restricts the maximum flow velocity. These three factors work together to form a parameter system that is both mutually restrictive and complementary. Through iterative adjustments, the optimal solution is gradually approached. The output of the final casting scheme marks the completion of the optimization process. The gate numbers included in the scheme provide clear execution targets, and the opening time is accurate to the 0.1-second level, ensuring the accuracy of timing control. The flow rate value is in liters / second, facilitating precise control of the actuator. The execution duration defines the working cycle of each gate, avoiding over-pouring or under-pouring.

[0083] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for intelligent monitoring of casting production based on the Internet of Things, characterized in that, The method includes: Preprocess the mold geometry data and molten casting viscosity distribution data, determine the temperature gradient and local viscosity difference of the stagnant region based on the extracted molten casting flow parameters, analyze the geometric constraints of the stagnant region of the mold structure, and obtain the local pressure gradient of the gate in each gate region. If the local difference in the viscosity of the molten casting exceeds the preset critical viscosity for bubble formation, the probability of bubble formation and the range of bubble size distribution are determined. If the local difference in the viscosity of the molten casting does not exceed the preset critical viscosity for bubble formation, the current gate opening sequence and flow distribution are maintained. The opening sequence of multiple gates is optimized based on the probability of bubble generation and the distribution range of bubble size. The opening order of each gate and the gate opening sequence deviation are determined, and an optimized opening sequence scheme is generated. The flow distribution ratio of each gate is dynamically adjusted according to the optimized start-up timing scheme to generate an initial flow distribution scheme. Obtain the molten casting flow feedback data after executing the initial flow allocation scheme, and generate an updated flow allocation scheme; Based on the updated flow distribution scheme, the internal stress distribution caused by the casting filling integrity and pressure fluctuations in the stagnant region is calculated, and the internal defect prediction results are obtained. If the bubble size distribution range in the internal defect prediction results exceeds the preset threshold, the gate opening timing deviation and flow distribution ratio are iteratively adjusted to obtain optimized casting parameters. If the bubble size distribution range does not exceed the preset threshold, the casting parameters are confirmed and the final casting scheme is output.

2. The intelligent monitoring method for casting production based on the Internet of Things according to claim 1, characterized in that, The preprocessed mold geometry data and molten casting viscosity distribution data are used to determine the temperature gradient and local viscosity differences in the stagnant region based on the extracted molten casting flow parameters. The geometric constraints of the stagnant region in the mold structure are analyzed to obtain the local pressure gradient of each gating region, including: Data on the internal temperature field and molten casting flow velocity are collected to generate a temperature distribution matrix and a velocity distribution matrix. The temperature change rate of adjacent regions is calculated based on the temperature distribution matrix to generate a temperature gradient in the stagnant region. The location coordinates of the stagnant region are identified based on the velocity distribution matrix. Based on the location coordinates of the stagnant region, the channel cross-sectional area and turning angle are extracted from the 3D model of the mold to generate a geometric feature vector of the mold. The molten casting flow velocity data and the temperature distribution matrix are used to generate the molten casting flow parameters. Based on the geometric feature vector of the mold and the molten casting flow parameters, a feature matrix is ​​constructed. The feature matrix is ​​then processed using a convolutional neural network to generate the local pressure gradient of each gating region.

3. The intelligent monitoring method for casting production based on the Internet of Things according to claim 2, characterized in that, The step of constructing a feature matrix based on the casting geometric feature vector and the casting flow parameters, and processing the feature matrix using a convolutional neural network to generate the local pressure gradient of each gating region includes: Based on the channel cross-sectional area and turning angle in the casting geometric feature vector, a cross-sectional area change rate and a turning angle resistance factor are generated; based on the viscosity value in the casting flow parameters, a viscosity gradient component is generated; based on the cross-sectional area change rate, the turning angle resistance factor, and the viscosity gradient component, the feature matrix is ​​constructed; the feature matrix is ​​convolved and pooled using a convolutional neural network to generate a flow resistance coefficient; based on the flow resistance coefficient and the casting inlet pressure, the local pressure gradient of each gating region is generated.

4. The intelligent monitoring method for casting production based on the Internet of Things according to claim 1, characterized in that, The step of optimizing the multi-gate opening sequence based on the bubble generation probability and bubble size distribution range, and determining the opening order and opening sequence deviation of each gate, includes: calculating the expected bubble volume at each gate position based on the bubble generation probability and bubble size distribution range; constructing an optimization objective function based on the expected bubble volume; encoding the gate opening sequence using a genetic algorithm to generate a sequence scheme population; calculating the casting front-end position trajectory during the casting filling process based on the sequence scheme population; calculating the spatial superposition volume based on the casting front-end position trajectory and the bubble generation probability, and generating a fitness value; iteratively updating the sequence scheme population based on the fitness value to generate the opening order of each gate; and calculating the opening time interval between adjacent gates based on the opening order of each gate, generating the opening sequence deviation.

5. The intelligent monitoring method for casting production based on the Internet of Things according to claim 4, characterized in that, The step of calculating the opening time interval between adjacent gates based on the opening sequence of each gate, and generating the opening timing deviation, includes: calculating the opening time interval between adjacent gates based on the opening sequence of each gate; analyzing the velocity vector angle when the two streams of casting at adjacent gates converge based on the opening time interval; adjusting the opening time of the gate with the later opening sequence among the adjacent gates based on the velocity vector angle to obtain the adjusted time; and generating the opening timing deviation based on the adjusted time and the total filling time.

6. The intelligent monitoring method for casting production based on the Internet of Things according to claim 1, characterized in that, The step of dynamically adjusting the flow distribution ratio of each gate according to the optimized opening sequence scheme to generate an initial flow distribution scheme includes: extracting time series data of the casting velocity in the region according to the optimized opening sequence scheme, determining the casting velocity pulsation amplitude, and solving the basic flow value of each gate; calculating the total flow of gates that are opened simultaneously according to the basic flow value of each gate and the time interval in the opening sequence, and determining the initial flow distribution scheme.

7. The intelligent monitoring method for casting production based on the Internet of Things according to claim 1, characterized in that, The step of obtaining the molten casting flow status feedback data after executing the initial flow allocation scheme and generating an updated flow allocation scheme includes: Based on the molten casting flow feedback data, the amplitude of molten casting velocity pulsation is calculated; based on the difference between the amplitude of molten casting velocity pulsation and the advance distance of the molten casting front, the filling rate balance state is determined; based on the filling rate balance state, a reinforcement learning algorithm is used to adjust the flow distribution ratio of each gate, and the updated flow distribution scheme is generated.

8. The intelligent monitoring method for casting production based on the Internet of Things according to claim 1, characterized in that, The calculation of the internal stress distribution caused by the casting filling integrity and pressure fluctuations in the stagnant region based on the updated flow distribution scheme, to obtain the internal defect prediction results, includes: Based on the updated flow distribution scheme, a three-dimensional finite element model of the casting is constructed; based on the three-dimensional finite element model of the casting, the flow velocity and pressure distribution of the molten casting are calculated, and filling rate distribution data is generated; based on the filling rate distribution data and the temperature gradient of the stagnant region, the comprehensive stress value is calculated; based on the comprehensive stress value and the temperature change rate of the molten casting solidification process, an internal defect prediction result containing the bubble generation probability and the bubble size distribution range is generated.

9. The intelligent monitoring method for casting production based on the Internet of Things according to claim 8, characterized in that, The process of generating internal defect prediction results, including the bubble formation probability and the bubble size distribution range, based on the comprehensive stress value and the temperature change rate during the solidification process of the casting liquid, includes: Based on the comprehensive stress value, the number of units exceeding the critical nucleation stress threshold is counted to generate the bubble generation probability; based on the difference between the comprehensive stress value and the critical nucleation stress threshold and the temperature change rate during the solidification process of the casting liquid, the bubble diameter range is estimated to generate the bubble size distribution range.

10. The intelligent monitoring method for casting production based on the Internet of Things according to claim 1, characterized in that, If the bubble size distribution range in the internal defect prediction results exceeds a preset threshold, the gate opening timing deviation and flow distribution ratio are iteratively adjusted to obtain optimized casting parameters, including: If the maximum value of the bubble size distribution range in the internal defect prediction results exceeds the preset size threshold, the current gate opening timing deviation value and flow distribution ratio data are extracted. Based on the location information of the bubble size exceeding the standard area, the opening time and flow distribution ratio of the corresponding gate are adjusted to obtain the optimized casting parameters.

Images