Intelligent control system and method for automatic production of salt sculptures

Abstract

This invention relates to the field of automated production control technology for salt sculptures, specifically, to an intelligent control system and method for automated salt sculpture production. The system includes a raw material pretreatment module, a preform forming and initial state calibration module, a drying state spatiotemporal observation module, an adaptive predictive control module, and a demolding and quality closed-loop module. First, salt powder, adhesive, and water are mixed to generate a homogeneous mixture, which is then injected into a mold to form a salt sculpture preform. The surface temperature and humidity of the preform, as well as the overall moisture content, are collected. Based on extended Kalman filtering, the internal pressure and surface densification degree are estimated, generating a risk spatiotemporal distribution map. The risk distribution is converted into a potential field, and a spray path is planned along the negative gradient direction. Spraying is executed according to the risk index, and heating power is adjusted in zones to obtain a dried and mature salt sculpture. Finally, demolding detection is performed, and feedback parameters are generated for subsequent control. This invention achieves precise control of the salt sculpture drying process through closed-loop control of state observation and path planning.

Description

Technical Field

[0001] This invention relates to the field of automated production control technology for salt carving, and more specifically, to an intelligent control system and method for automated salt carving production. Background Technology

[0002] Salt sculptures are handicrafts made primarily from edible salt, through processes such as mixing, shaping, drying, and curing. To achieve mass production, automated production lines are gradually replacing manual labor, with the drying and curing stage being the core process determining the final quality of the salt sculpture.

[0003] In the salt sculpture production process, salt powder, adhesive, and water are mixed in a certain proportion and then filled into a mold. The resulting semi-finished product is called the salt sculpture blank. This blank needs to be heated and dried to evaporate the moisture and solidify the adhesive in order to obtain a finished salt sculpture with sufficient strength.

[0004] In the actual drying process, the moisture on the surface of the embryo evaporates rapidly after being heated, and the salt particles and adhesives together form a relatively dense surface layer; the internal moisture vaporizes and expands but cannot penetrate the hard shell to escape, resulting in the accumulation of internal pressure and the generation of hidden defects such as micropores and microcracks.

[0005] To address the aforementioned issues, existing technologies typically employ extended heating times or increased temperatures to continue drying. While this approach does allow for further evaporation of surface moisture, alleviating the dampness, the accelerated water loss from the surface leads to increased density. This, in turn, hinders the smooth expulsion of internal moisture, resulting in increased internal pressure buildup, raising the risk of surface cracking, or causing existing microcracks to expand further.

[0006] Therefore, there is an urgent need to provide an intelligent control method that can suppress premature densification of the surface layer and avoid self-hindering effects. Summary of the Invention

[0007] One objective of this invention is to provide an intelligent control system for automated salt sculpture production. This system collects real-time data on the surface temperature and humidity distribution and overall moisture content of the salt sculpture blank. It then uses an extended Kalman filter to recursively estimate the pressure accumulation level and surface densification state within the blank. Based on the predicted risk level, it dynamically plans the spray path and heating power distribution of the moving nozzles, thereby solving the problems mentioned in the background art. During the heating and drying process, the surface of the salt sculpture blank becomes too dense too early, which prevents internal moisture from escaping, leading to micro-cracks or even bursting.

[0008] To achieve the above objectives, the intelligent control system includes a raw material pretreatment module and a preform forming and initial state calibration module. The raw material pretreatment module is used to mix salt powder, adhesive and water to generate a homogeneous mixture. The preform forming and initial state calibration module is used to inject the homogeneous mixture into a mold to generate a salt sculpture preform. The system is characterized by further including a drying state spatiotemporal observation module and an adaptive predictive control module. The dry state spatiotemporal observation module takes the salt sculpture blank as the observation object, collects its surface temperature field matrix, surface humidity field matrix and overall average moisture content, and recursively estimates the pressure value and surface densification degree value of each grid point inside the blank by substituting the extended Kalman filter into the state transition equation and observation equation. The self-hindrance risk index, which characterizes the risk of internal pressure buildup in the embryo, is calculated based on the pressure value and the degree of surface densification, and then predicted forward to generate a risk spatiotemporal distribution map and an intervention priority queue. The adaptive predictive control module converts the risk spatiotemporal distribution map into a continuous risk potential field function and calculates its gradient field. Starting from the local maximum point of risk, it generates a descent path of the moving nozzle along the negative gradient direction. The movable micro-mist nozzle moves along the descent path and sprays according to the self-obstruction risk index of the current location, while adjusting the heating power of the corresponding location according to the self-obstruction risk index of each area. When the self-obstruction risk index of all grid points is below the safety threshold and the overall average moisture content meets the standard, dry and mature salt embryos are output.

[0009] In the aforementioned technical solution, during the drying process of the salt-carved embryo, there is a strong nonlinear coupling relationship between surface densification and internal pressure accumulation, and this relationship dynamically changes with spatial location and time. Relying solely on the overall average moisture content as feedback fails to identify early signs of pressure buildup in localized areas, leading to control decisions lagging behind defect formation. Therefore, the spatiotemporal observation module for the drying state introduces an extended Kalman filter to map the measurable surface temperature and humidity distribution to the unmeasurable internal state space, enabling recursive estimation of the densification degree and pressure value at each grid point. Without this state estimation step, subsequent spraying or heating adjustments would lack spatial resolution and fail to differentiate the varying needs of different areas. The adaptive predictive control module constructs a risk potential field based on the estimated risk distribution and plans an intervention path along its negative gradient direction. The necessity of this design lies in the fact that the gradient direction of the risk potential field represents the direction of the fastest decrease in local risk; moving the nozzle along this direction ensures maximum pressure relief per unit spray volume. Conversely, if a fixed path scanning or equidistant traversal is used, the nozzle may stay too short in high-risk areas while over-spraying in low-risk areas, resulting in localized over-wetting or incomplete pressure relief.

[0010] Based on this, the state transition equation used by the extended Kalman filter of the dry state spatiotemporal observation module is that the partial derivative of internal pressure with respect to time is equal to the first model coefficient multiplied by the Laplace operator of the surface temperature field, plus the second model coefficient multiplied by the difference between the overall average moisture content and the equilibrium moisture content, and then minus the third model coefficient multiplied by the product of the surface densification degree and the difference between internal pressure and environmental pressure. The observation equation is a vector composed of surface temperature measurement, surface humidity measurement, and overall average moisture content measurement. It is equal to the observation matrix multiplied by the state vector plus the measurement noise.

[0011] In another technical solution, the model coefficients used in the extended Kalman filter are calibrated as follows: a standard sample is prepared and a sensor is pre-embedded. During the drying process, the measured pressure, temperature, humidity and moisture content are recorded. The sum of squared errors between the measured values ​​and the calculated values ​​is minimized, and the coefficients are obtained by using a nonlinear least squares fitting algorithm.

[0012] In this technical solution, the construction of the state transition equation and the observation equation, along with the calibration method for the model coefficients, constitute the core computable framework of the extended Kalman filter. The design of the state transition equation is based on the physical mechanism of the salt carving drying process: the Laplace operator of the surface temperature field reflects the spatial non-uniformity of the temperature distribution; the larger the temperature gradient, the more intense the thermal stress-driven moisture migration and the faster the internal pressure rises; the difference between the overall average moisture content and the equilibrium moisture content represents the overall moisture removal driving force of the embryo; and the product of the surface densification degree and the difference between the internal pressure and the ambient pressure quantifies the hindering effect of the dense hard shell on moisture removal. Without any of these, the state estimation cannot accurately describe the multi-factor coupling relationship in the actual drying process; the design of the observation equation solves the mapping problem between measurable and non-measurable quantities, enabling the Kalman filter to use sensor data to correct the prediction results. The calibration method for the model coefficients uses measured data from standard samples and nonlinear least squares fitting. This is necessary because the drying kinetic parameters of the salt-starch-water system cannot be derived from theory and must be determined through experimental data inversion. Without this calibration, the state transition equation will have a systematic deviation due to coefficient mismatch, causing the internal pressure estimate to deviate from the true value, which in turn will render subsequent risk assessment and intervention decisions ineffective.

[0013] The second objective of this invention is to provide an intelligent control method for automated salt carving production, comprising the following steps: S1. Mix salt powder, adhesive and water, collect stirring torque during mixing and calculate mixing uniformity index to generate homogeneous mixture; S2. Inject the homogeneous mixture into the mold cavity, collect the injection pressure curve and the mold cavity pressure distribution, and generate a salt sculpture blank with an initial density distribution matrix. S3. Using the salt sculpture blank as the observation object, its surface temperature field matrix, surface humidity field matrix and overall average moisture content are collected respectively. Based on the extended Kalman filter, the pressure value and surface densification degree value of each grid point inside the blank are recursively estimated by substituting into the state transition equation and the observation equation. The self-hindrance risk index, which characterizes the risk of internal pressure buildup in the embryo, is calculated based on the pressure value and the degree of surface densification, and then predicted forward to generate a risk spatiotemporal distribution map and an intervention priority queue. S4. Convert the risk spatiotemporal distribution map into a continuous risk potential field function and calculate its gradient field. Generate the descent path of the moving nozzle along the negative gradient direction, starting from the local maximum point of risk. The movable micro-mist nozzle moves along the descent path and sprays according to the self-obstruction risk index of the current location, while adjusting the heating power of the corresponding location according to the self-obstruction risk index of each area. When the self-obstruction risk index of all grid points is below the safety threshold and the overall average moisture content meets the standard, dry and mature salt embryos are output.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention introduces extended Kalman filtering into the salt carving drying process. By constructing a state transition equation that includes temperature gradient driving force, moisture content difference and surface densification resistance term, measurable surface temperature and humidity data are converted into real-time estimates of internal pressure accumulation and densification process, enabling the system to identify the spatial distribution of self-hindrance risk before crack formation.

[0015] 2. This invention employs a path planning method based on the risk potential field gradient, which moves the spray action along the direction of the fastest local risk reduction, and dynamically adjusts the spray parameters according to the location risk value and the remaining path length. Combined with the differentiated allocation of zone heating power, it achieves spatial resolution control of the drying process. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall structure of the intelligent control system for automated salt carving production according to the present invention; Figure 2 This is a flowchart illustrating the structure of the dry state spatiotemporal observation module of the present invention. Figure 3 This is a flowchart illustrating the structure of the adaptive predictive control module of the present invention. Figure 4 This is an overall flowchart of the intelligent control method for automated salt carving production according to the present invention.

[0017] The meanings of the labels in the diagram are as follows: 100. Raw material pretreatment module; 200. Preform forming and initial state calibration module; 300. Spatiotemporal observation module for drying state; 400. Adaptive predictive control module; 500. Demolding and quality closed-loop module. 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] Currently, there is a problem with the premature densification of the surface layer of salt carving blanks during the heating and drying process, which prevents internal moisture from escaping, leading to microcracks or even bursting. One objective of this invention is to provide an intelligent control system for automated salt carving production. See [link to relevant documentation]. Figure 1 As shown, it includes a raw material pretreatment module 100, a preform forming and initial state calibration module 200, a drying state spatiotemporal observation module 300, an adaptive predictive control module 400, and a demolding and quality closed-loop module 500.

[0020] This system collects the temperature and humidity distribution on the surface of the embryo in real time through a sensor array, calculates the internal pressure field and the surface densification degree field using an extended Kalman filter, and dynamically executes a graded intervention strategy based on the predicted risk level to achieve closed-loop adaptive control of the balance between surface densification and internal moisture removal.

[0021] The raw material pretreatment module 100 includes a loss-in-weight weighing unit and a dual planetary mixing unit. This module first receives the feeding instruction for this production task from the upper control system. This instruction includes the target ratio parameters of salt powder, adhesive and water, as well as the total batch weight. Subsequently, the loss-in-weight weighing unit performs the weighing according to the instruction.

[0022] The loss-in-weight weighing unit consists of a storage bin, a screw feeder, and a high-precision load cell. When material falls from the storage bin through the screw feeder, the load cell measures the rate of change of the total weight of the storage bin and its contents in real time. The controller calculates the actual discharge flow rate based on the weight reduction per unit time and compares it with the target flow rate. It then adjusts the speed of the screw feeder using a proportional-integral-derivative algorithm to ensure that the actual discharge rate follows the target value.

[0023] For example, when the target ratio is salt powder as the majority, adhesive as a minor portion, and water as a small proportion, the loss-in-weight weighing unit weighs the three materials independently in sequence. During the feeding process of each material, the controller continuously adjusts the feeding speed until the cumulative feeding amount reaches the target value. After weighing, each material is temporarily stored in the weighing hopper, waiting to be sent to the mixing container.

[0024] After weighing is completed, the loss-in-weight weighing unit opens the pneumatic valve at the bottom of the weighing hopper, allowing the salt powder, adhesive, and water to fall sequentially or simultaneously into the mixing container of the double planetary mixing unit.

[0025] The dual planetary mixing unit consists of a mixing container, two planetary mixing blades and a wall scraper. The mixing blades rotate on their own axis while revolving around the central axis, creating a complex flow field that causes the material to be continuously sheared, stretched and folded within the mixing container.

[0026] For example, after the salt powder and adhesive are first added to the mixing container, the agitator starts to run at a low speed to initially mix the two powders. Water is then added in a spray form to prevent localized over-wetting and clumping. The scraper continuously scrapes off the material adhering to the inner wall of the container and reintroduces it into the mixing zone, ensuring that all materials participate in the mixing process.

[0027] The dual planetary mixer unit has a torque sensor mounted on the mixing shaft. This sensor detects the torsional torque on the mixing bearing in real time using a magnetoelectric method and converts the torque signal into an analog voltage output to a data acquisition card. The data acquisition card continuously records the waveform of the torque changing over time at a fixed sampling frequency, forming a torque time-domain signal, denoted as . .

[0028] The raw material preprocessing module 100 performs a Fast Fourier Transform on the torque waveform within each fixed-duration time window, converting the time-domain signal into a frequency-domain signal to obtain the power spectral density distribution P(f) of the torque signal. The calculation formula is as follows: ; In the formula, P(f) represents the power spectral density at frequency f, with the unit being the square of the torque unit per hertz, reflecting the energy distribution of the torque signal at different frequency components; t0 is the start time of the current analysis time window, in seconds; Δt is the length of the time window, in seconds, representing the duration of the signal captured in each Fourier transform. is the kernel function for the Fourier transform, where j is the imaginary unit and f is the frequency variable in Hertz; Integration This indicates that the product of the torque signal and the kernel function is accumulated and summed within a time window; This means taking the modulus of a complex number, squaring the modulus, and obtaining the power value of that frequency component.

[0029] Furthermore, the dominant frequency f is extracted from the power spectrum. max That is, the frequency at which the power spectral density reaches its maximum value: ; Then the main frequency f max The corresponding energy value is P(f) max Simultaneously calculate the total energy P across the entire frequency band. total : ; Next, we define the mixing uniformity index M. mix The ratio of the main frequency energy to the total energy across the entire frequency band: ; For example, in the initial stage of mixing, the material distribution is extremely uneven, and the stirring paddle periodically cuts through the clumps. The torque waveform exhibits large low-frequency fluctuations. At this time, the low-frequency component dominates the power spectrum, and the frequency corresponding to the dominant frequency is low, but its energy proportion is not high. Therefore, M mix The value is relatively small.

[0030] As mixing proceeds, the materials gradually become homogeneous, the torque waveform tends to stabilize, and the dominant frequency energy in the power spectrum gradually shifts to the mid-to-high frequency range, with its energy proportion increasing. mix The value gradually increases and eventually stabilizes near a high value. The raw material pretreatment module 100 repeats the above calculation every time window, tracking the change trajectory M of the mixing uniformity index in real time. mix (n), where n is the sequence number of the time window.

[0031] Specifically, the raw material pretreatment module 100 sets a fluctuation threshold and a duration threshold for the mixing uniformity index. When the mixing uniformity index calculated within multiple consecutive time windows is greater than the preset uniformity target value, and the absolute value of the index difference between adjacent windows is less than the fluctuation threshold, it is determined that the mixing has reached a uniform state.

[0032] For example, after mixing for a certain period of time, if the mixing uniformity index for three consecutive time windows is higher than the target value, and the maximum difference between these three indices is less than the fluctuation threshold, the module outputs a mixing end signal and stops the agitator. Subsequently, the raw material pretreatment module 100 opens the discharge valve at the bottom of the mixing container to discharge all the uniformly mixed material. This material is now called a homogeneous mixture, with a stable distribution ratio of salt powder, binder, and water, and no local agglomeration or uneven wetting. The raw material pretreatment module 100 packages the homogeneous mixture and transmits it, along with the batch identification information, to the preform forming and initial state calibration module 200.

[0033] The preform forming and initial state calibration module 200 includes a servo screw injection unit, a mold cavity pressure distribution array unit, and a mold cavity body. This module first receives homogeneous mixture and batch identification information from the raw material pretreatment module 100, and simultaneously receives mold cavity parameters provided by mold installers or an automatic mold changing device. These parameters include the geometric contour and volume of the mold cavity. The homogeneous mixture is injected into the mold cavity to form the salt sculpture preform. Simultaneously, the pressure changes and pressure distribution within the mold cavity during the injection process are collected, specifically including the following steps: Ⅰ: After completing the injection molding preparation, the preform forming and initial state calibration module 200 first performs injection molding process control and pressure curve acquisition. The servo screw injection unit consists of a servo motor, reducer, screw, and barrel. This unit feeds the homogeneous mixture into the barrel, the servo motor drives the screw to rotate, and the screw's spiral groove pushes the material forward, injecting it into the cavity of the mold body through the nozzle.

[0034] During the injection molding process, the real-time feedback channel of the servo driver outputs the current torque current value of the motor. This torque current value is linearly related to the thrust output by the screw, and after calibration, it is converted into the injection molding pressure value. The data acquisition card continuously records the waveform of injection molding pressure changing over time at a fixed sampling frequency, forming an injection molding pressure curve. .

[0035] The preform forming and initial state calibration module 200 calculates the first derivative of the injection molding pressure curve. The changes in the derivative are monitored. In the initial stage of injection molding, the mold cavity is in a hollow state, the material flow resistance is small, the injection pressure rises slowly, and the first derivative is a small positive value.

[0036] As the material gradually fills the mold cavity, pressure begins to accumulate, and the injection pressure curve shows an inflection point, with the first derivative changing from positive to negative or suddenly dropping from a high positive value. This module identifies the peak position of the first derivative and assigns the corresponding time t to this peak point. peak The moment at which the injection molding ends is determined, and the peak injection pressure at this point is recorded. .

[0037] For example, in the injection molding process of a mold with a complex internal cavity, the pressure curve maintains a gentle upward trend when the material just touches the farthest edge of the cavity. Once the cavity is completely filled, the screw continues to push material, causing the pressure to rise sharply. The first derivative of the pressure curve rapidly increases to a peak value. Subsequently, as the material compression approaches saturation, the rate of pressure increase slows down, and the first derivative begins to decline. This module utilizes this peak value of the derivative to accurately determine the injection molding endpoint.

[0038] II: Acquisition of mold cavity pressure distribution and density calculation. The mold cavity pressure distribution array unit consists of multiple miniature pressure sensors embedded in the bottom and sidewalls of the mold cavity body. These sensors are arranged in a two-dimensional array with a fixed number of rows and columns. Each sensor corresponds to a detection point, with coordinates (u, v), where u is the row number and v is the column number. During the injection molding process and the pressure holding stage after injection molding, this array unit acquires the pressure value p at each point in real time. uv .

[0039] At the end of injection molding time t peak This module extracts the pressure values ​​from all points to form an initial pressure distribution matrix. Calculate the mean of this matrix. and standard deviation : ; ; In the formula, U is the total number of rows of pressure sensors; V represents the total number of columns.

[0040] Furthermore, define the initial compactness distribution matrix of the embryo. The density index d at each point represents the degree of deviation between the pressure value at each point and the average value across the entire field. uv The calculation is as follows: ; in d is a very small positive number to avoid the case where the denominator is zero. uv A positive value indicates that the pressure at that point is higher than the average level, suggesting that the material is more densely packed at that location; d uv A negative value indicates that the pressure at that point is below the average level, suggesting that the filling is loose or that there are local voids.

[0041] For example, in a complex mold with deep cavities and bosses, the bottom of the deep cavity is often not filled densely due to the long material flow path and high resistance. uv Negative values ​​may occur; however, in the area near the gate, the material is directly pushed by the screw, resulting in dense filling, d uv The result is a positive value. This distribution matrix records the differences in the initial compaction state at different locations of the embryo.

[0042] The preform forming and initial state calibration module 200 will determine the injection molding end time t. peak Peak injection pressure P peak Initial pressure distribution matrix and the initial density distribution matrix of the embryo The initial state labels of the combined salt sculpture blanks are used as the initial state tags for this batch. These tags, together with the already formed blanks within the mold cavity, constitute salt sculpture blanks with initial state information. This module transmits the salt sculpture blanks, along with their initial state tags, to the drying state spatiotemporal observation module 300 for use in subsequent drying processes to distinguish between initial density differences and stress changes generated during the drying process.

[0043] The dry state spatiotemporal observation module 300 includes an infrared thermal imaging array for acquiring the temperature field distribution on the surface of the embryo, a laser humidity reflectometer array for acquiring the humidity field distribution on the surface of the embryo, a microwave humidity sensor for detecting the overall average moisture content of the embryo, a servo slide for mounting the laser humidity reflectometer array to achieve region-by-region scanning, a data acquisition and processing unit for synchronously acquiring and preprocessing all sensor signals, and an extended Kalman filter for estimating the unmeasurable internal state from the measurable signals.

[0044] like Figure 2 As shown, the drying state spatiotemporal observation module 300 first receives the salt sculpture blank and its initial state tag from the blank forming and initial state calibration module 200, and then collects the temperature and humidity distribution and overall moisture content of the blank surface in real time through a sensor array deployed in the drying chamber. Specifically, it includes the following steps: Ⅰ: After completing the data acquisition preparation, the spatiotemporal observation module 300 for the drying state deploys the sensor array and acquires spatiotemporal data. The infrared thermal imaging array is installed on the inner side wall of the drying chamber, consisting of multiple infrared thermal imaging sensors arranged in a fixed number of rows and columns, with each sensor corresponding to a detection field of view.

[0045] The servo slide, equipped with the infrared thermal imaging array, can move horizontally and vertically, allowing the array to scan the entire surface of the salt sculpture blank area by area. A laser humidity reflectometer array is also mounted on the servo slide, moving synchronously with the infrared thermal imaging array. A microwave humidity sensor is fixedly mounted on the top of the drying chamber, with its transmitting antenna pointing downwards towards the salt sculpture blank.

[0046] During the drying process, the servo slide moves along a preset scanning path, and the infrared thermal imaging array collects the surface temperature distribution within the corresponding field of view at each stationary position. After stitching, a complete surface temperature field matrix is ​​formed. , where x and y are the spatial coordinates of the embryo surface, and t is time.

[0047] The laser humidity reflectometer array emits a laser beam at each stationary position to illuminate the surface of the substrate, receives the reflected light signal, and infers the surface moisture content based on the attenuation of the reflected light intensity. After being stitched together, a complete surface humidity field matrix is ​​formed. .

[0048] For example, a large salt sculpture blank is divided into 9 scanning areas. The servo slide first moves to the upper left corner area, where the infrared thermal imaging array and the laser humidity reflectometer array simultaneously collect temperature and humidity data. Then, the slide moves sequentially to the upper center, upper right, middle left, and center areas until it covers the entire surface of the blank. After each scanning cycle, the system obtains a complete set of surface temperature field matrices and surface humidity field matrices.

[0049] A microwave humidity sensor continuously emits low-power microwaves at a fixed frequency. These microwaves penetrate the salt sculpture and are received by a receiving antenna on the other side. Water molecules inside the salt sculpture absorb microwave energy; the higher the water content, the greater the microwave attenuation. The sensor measures the ratio of transmitted power to received power to calculate the microwave attenuation. The average moisture content of the entire embryo can be deduced by using a pre-calibrated decay-moisture content curve. .

[0050] All the above sensors maintain a consistent sampling frequency, and the data acquisition and processing unit records all data in a time-synchronized manner, forming a three-dimensional spatiotemporal data stream that changes over time.

[0051] II: Constructing a Dry State Observer. The dry state spatiotemporal observation module 300 incorporates a dry state observer based on an extended Kalman filter, used to estimate the indirect internal pressure field from the measurable surface temperature field, surface humidity field, and overall average moisture content. and the degree of surface densification field .

[0052] First, define the state vector at a spatial location (x, y) on the surface of the embryo at time t as: ; The first two components T in the state vector s and H s For quantities that can be directly measured, the latter two components P int and D s For the immeasurable quantity to be estimated.

[0053] Furthermore, a state transition equation is constructed using the following formula to describe the physical evolution of the drying process: ; In the formula, Let be the partial derivative of internal pressure with respect to time, representing the rate of change of internal pressure with time at a fixed spatial location, in Pascals per second. The Laplace operator for the surface temperature field matrix is ​​calculated as follows: ,in This represents the second-order partial derivative of surface temperature with respect to spatial coordinate x. This represents the second-order partial derivative with respect to the spatial coordinate y. The Laplace operator reflects the degree of curvature or non-uniformity of temperature distribution in space; the larger the temperature gradient, the larger this value.

[0054] symbol The symbol for partial derivatives is used. When it appears before the time variable t, it indicates the partial derivative with respect to time. When it appears before the spatial coordinates x or y, it indicates the partial derivative with respect to space. W avg The overall average moisture content; W eq The equilibrium moisture content is determined by the current ambient temperature and humidity. The difference between the two (W) avg -W eq The difference represents the driving force for the overall dehumidification of the embryo. The larger the difference, the stronger the tendency for internal moisture to migrate to the surface. P amb For ambient atmospheric pressure, (P) int -P amb The pressure difference between the internal pressure and the ambient pressure represents the driving force for water vapor to escape. D s This represents the degree of surface densification. A larger value indicates a denser surface crust, which stronger impedes the expulsion of water vapor. Therefore, this term appears in the equation with a negative sign, representing D. s The larger P is int The greater the restriction on the rate of increase, the faster the internal pressure will accumulate when the surface is completely dense; α1, α2, and α3 are model coefficients. The calibration method is as follows: A standard sample with the same formula as the salt sculpture to be produced is prepared. The sample is placed in a temperature- and humidity-controlled drying oven. A miniature pressure sensor is embedded inside the sample, and a thermocouple and humidity probe are attached to its surface. During the drying process, the measured pressure, surface temperature, surface humidity, and ambient temperature and humidity are recorded at fixed time intervals. The overall mass change of the sample is also recorded to calculate the average moisture content. The collected data is substituted into the state transition equation, and a nonlinear least squares fitting algorithm is used to iteratively optimize and solve for the optimal estimates of α1, α2, and α3, with the objective of minimizing the sum of squared errors between the measured pressure values ​​and the calculated values.

[0055] Surface densification degree field Its own evolutionary pattern is described by the following equation: ; In the formula, This represents the rate of change of surface humidity over time. A negative value indicates that the surface is losing water. The faster the water loss rate, the faster the surface densification. T is a sigmoid function used to describe the accelerating effect of temperature on the densification rate. ref Using the reference temperature, β2 controls the temperature sensitivity; when the surface temperature is lower than the reference temperature, the function value is close to zero, and when the surface temperature is higher than the reference temperature, the function value approaches 1. D eq To balance the degree of densification, it represents the upper limit of densification that the surface layer can ultimately achieve under given environmental conditions; β1, β2, and β3 are model coefficients, and their calibration method is the same as that of α1, α2, and α3. The experimental data of the same set of standard samples are used. The measured surface humidity change rate, surface temperature, and surface densification degree estimates are substituted into the equation. The goal is to minimize the sum of squared errors between the calculated values ​​and the measured values. The optimal estimates of the β1, β2, and β3 phases are solved by a nonlinear least squares fitting algorithm.

[0056] After constructing the state transition equations and designing the prediction stage of the extended Kalman filter, in order to use the actual measurement data acquired by the sensor to correct the prediction results, it is necessary to establish a mathematical relationship between the sensor measurements and the state vector. The system's observation equations describe the relationship between the sensor measurements and the state variables, and are constructed as follows: ; In the formula, and These are the actual measured values ​​of the infrared thermal imaging array and the laser humidity reflectometer array, which are directly equal to the corresponding components in the state vector, so the corresponding position in the observation matrix is ​​1; Overall average moisture content measurement value With internal pressure field P int and the degree of surface densification field D s There is a correlation between them, represented by coefficients c1 and c2. This is because the higher the internal pressure and the denser the surface layer, the slower the overall moisture dissipation of the embryo, and the slower the average moisture content decreases. c1 and c2 are observation coefficients, obtained by linear regression fitting of the measured average moisture content of the standard sample with the corresponding internal pressure and surface densification degree.

[0057] It should be noted that the extended Kalman filter operates according to the standard prediction-update two-stage recursive process.

[0058] Before starting the filter, the initial state estimate and initial covariance matrix of the filter are set. The specific setting method is as follows: In the initial state estimation, the surface temperature and surface humidity components are taken as the first measurement values ​​when the preform is inserted into the cavity, the internal pressure component is taken as the ambient atmospheric pressure, and the surface densification degree component is taken as zero. The initial covariance matrix is ​​set as a diagonal matrix. The diagonal elements of the temperature and humidity components are taken as the square of the sensor calibration error, the diagonal elements of the internal pressure component are taken as an empirical small value, and the diagonal elements of the densification degree component are taken as a large value.

[0059] In addition, the process noise covariance matrix Q and the measurement noise covariance matrix R need to be set, and the specific setting method is as follows: Q is calibrated through static experiments: the standard salt sculpture blank is placed in a constant temperature and humidity environment, sensor data is continuously collected, and the zero mean fluctuation variance of the measured values ​​of each state component is calculated as the diagonal element of Q. The measurement noise covariance matrix R is directly taken from the variance of the output data of each sensor under static conditions.

[0060] In subsequent recursions, Q and R are treated as constant matrices and substituted into... and use.

[0061] In the prediction phase, the filter predicts the current state value and covariance matrix based on the state estimate and state transition equation from the previous time step. In the update phase, the filter calculates the Kalman gain based on the current sensor measurements and observation equation, and corrects the prediction result using the measurements to obtain the optimal state estimate for the current time step.

[0062] Specifically, for each grid point (x, y), the filter independently performs the following recursive process. Let... The state at time k is predicted based on the estimate at time k-1. This is the optimal estimate after correction of the measurement value at time k. The prediction equation is: ; In the formula, This is the discretized form of the state transition equation. The prediction covariance matrix is: ; In the formula, Let be the Jacobian matrix of the state transition function with respect to the state; Let be the process noise covariance matrix.

[0063] The update equation is: ; In the formula, The Kalman gain matrix; This is the observation matrix, i.e., the coefficient matrix in the observation equation above; To measure the noise covariance matrix; This is the actual measurement vector at the current moment; I is the identity matrix.

[0064] Through the above recursive calculation, the filter outputs the estimated internal pressure value for each grid point at each time step. and estimate the degree of surface densification .

[0065] III: Calculate the self-hindrance risk index and predict future states. First, based on the estimated values ​​output by the observer, define the current self-hindrance risk index for each grid point using the following formula. : ; In the formula, The internal pressure safety threshold is determined by the tensile strength of the preform material. When the internal pressure exceeds this value, the preform is at risk of cracking. For reference to the degree of densification, used for normalization item; The above formula means that the risk index is directly proportional to the internal pressure. At the same time, the degree of surface densification acts as an amplification factor. When the degree of surface densification is high, even if the internal pressure has not yet reached the threshold, the risk index will be amplified, reflecting the early warning logic that the surface is blocked and the internal pressure is about to build up.

[0066] The dry state spatiotemporal observation module 300 further utilizes the state transition equation, using the estimated state at the current moment as the initial condition, to predict the future. The state after time t. The prediction calculation uses a numerical integration method, integrating the state transition equation from the current time t to... The predicted internal pressure value was obtained. and the predicted surface densification value Then, substituting the values ​​into the risk index formula, we obtain the predicted risk index. .

[0067] The dry state spatiotemporal observation module 300 can simultaneously calculate multiple risk indices with different prediction durations, such as predicting risk values ​​for the next 30 seconds, 60 seconds, and 90 seconds, forming a set of prediction sequences.

[0068] For example, in a certain local area, the current internal pressure has not yet reached the safe threshold, but the surface densification degree is already very high, and the temperature gradient remains large. In this case, the state transition equation predicts that the internal pressure will rapidly rise and exceed the threshold in the next 60 seconds, and the predicted risk index will rise to the warning level accordingly. Based on this, the module determines that there is an impending risk of pressure buildup in this area, requiring early intervention.

[0069] IV: Generate a spatiotemporal risk distribution map and construct an intervention priority queue. Specifically, the dry state spatiotemporal observation module 300 comprehensively evaluates the current risk index of all grid points and the predicted risk index for each prediction duration.

[0070] For each grid point, the maximum value between the current risk index and the predicted risk indices for all prediction durations is taken as the comprehensive risk value for that point. This module fills the comprehensive risk value into the grid corresponding to the sensor array according to the spatial coordinates, generating a pseudo-color spatiotemporal risk distribution map. The risk values ​​in the map correspond to a gradient from cool to warm colors as they increase from low to high, intuitively displaying the current and future risk levels of various regions on the surface of the embryo.

[0071] The dry-state spatiotemporal observation module 300 then sorts all grid points from highest to lowest based on their comprehensive risk value, forming an intervention priority queue. The grid point with the highest comprehensive risk value is placed at the front of the queue, indicating that without intervention, this location will be the first to experience pressure buildup or cracking in the current or short term.

[0072] For multiple grid points with similar overall risk values ​​and adjacent spatial locations, this module merges them into a continuous high-risk area and records the boundary coordinates set and the average risk value within the area. For example, in a protruding part of a salt sculpture, if the overall risk values ​​of multiple adjacent grid points in this area are significantly higher than those in other areas, these grid points are merged into a high-risk area; while in flat parts of the sculpture, the overall risk values ​​are generally lower and they are not prioritized for intervention.

[0073] Furthermore, the dry state spatiotemporal observation module 300 packages the intervention priority queue, the boundary coordinate set of each high-risk area, and the regional average risk value, and transmits them together with the salt sculpture embryo to the adaptive prediction control module 400.

[0074] like Figure 3 As shown, the adaptive predictive control module 400 includes an independently controllable array of partitioned infrared heating tubes, a movable micro-mist nozzle, a cold air auxiliary unit, a path planner, a power distributor, and an online learning submodule. This module receives data from the dry state spatiotemporal observation module 300, including the salt sculpture embryo, a risk spatiotemporal distribution map, an intervention priority queue, a set of boundary coordinates of high-risk areas, and the average risk value of the area. Based on the spatial distribution of risk, it dynamically plans the motion trajectory of the movable nozzle and the power allocation of the partitioned heating tubes, specifically including the following steps: Ⅰ: Upon receiving the risk spatiotemporal distribution map, this module first constructs a risk gradient field, converting discrete grid risk values ​​into a continuous risk potential field function. Here, x and y are the planar coordinates of the embryo surface. For the comprehensive risk value at discrete grid points, this module uses a bilinear interpolation method to construct a continuously differentiable risk potential field between grid points. Based on this, the gradient vector of the risk potential field is calculated. : ; In the formula, and These are the partial derivatives of the risk value in the x and y directions, respectively.

[0075] The gradient vector points in the direction of the fastest increase in risk value, and in the opposite direction. This indicates the direction in which the risk value decreases the fastest. This module uses the gradient field as the basis for subsequent path planning and power allocation.

[0076] For example, in a risk spatiotemporal distribution map, if the risk value of a certain local area decreases from the center outwards, then the gradient vector in that area points outwards from the center and the negative gradient direction points towards the center, indicating that the center is the point with the highest risk and needs to be dealt with first.

[0077] II: Path planning for the moving nozzle. This module controls the movable micro-mist nozzle to move above the surface of the substrate in the negative direction of the risk gradient. The core of the path planner is a continuous path generation algorithm based on gradient descent, with the following specific steps: First, the path planner identifies all local maxima in the risk potential field, i.e., those that satisfy... Furthermore, the Hessian matrix is ​​negative definite at these points. Each local maximum point corresponds to a risk peak region. The path planner starts from each local maximum point and generates a descent path along the negative gradient direction until it reaches a region where the risk value is below the mild threshold.

[0078] Each path consists of a series of consecutive spatial points, and the step size between adjacent points is determined by the gradient magnitude at the current point. Specifically, this step uses an inverse linear relationship: let the gradient magnitude at the current point be g, then the step size between adjacent points is calculated using the following formula: ; In the formula, The preset base step size is taken within the range of the surface resolution of the embryo, typically 1-5 mm. ε is a very small positive number, with a value at least one order of magnitude smaller than the magnitude of all possible gradients, for example, 10. -6 The purpose is to prevent the denominator from being too small when the gradient magnitude is close to zero, which would lead to an excessively large step size, while also avoiding a denominator of zero.

[0079] In addition, to avoid unreasonable paths due to excessively large or small step sizes, an upper limit on the step size is further limited. and lower limit ,Right now: ; In this way, the larger the gradient magnitude and the smaller the step size, the slower and more precise the spraying will be in areas where the risk changes drastically. The smaller the gradient magnitude, the larger the step size, and the faster the smooth region is traversed.

[0080] For example, if there are two risk peaks on the surface of the embryo, the path planner generates two descent paths starting from these two points. The first path meanders downwards along the negative gradient direction from the peaks, passing through the high-risk region before entering the safe region; the second path is similar. The two paths may intersect in some areas, merging into a single path after the intersection.

[0081] Once multiple descent paths are generated, the path planner sorts them according to their average risk value, with paths having higher average risk values ​​having higher priority. The movable micro-nozzle then traverses each path sequentially according to priority, moving from the starting point to the ending point on each path.

[0082] III: Determine dynamic spray parameters. As the movable micro-nozzle moves along the planned path, the adaptive predictive control module 400 calculates spray parameters in real time based on the risk value, gradient modulus, and path curvature of the nozzle's current position. Spray parameters include spray flow rate and spray duration.

[0083] spray flow Risk value relative to current location and gradient magnitude The relationship is: ; In the formula, Basic traffic; The spray flow rate proportionality coefficient was calibrated as follows: A standard sample with the same formula as the salt sculpture to be produced was prepared. Multiple humidity and pressure sensors were placed on the sample surface, and the sample was placed in an experimental drying oven. During the drying process, the sample surface was sprayed with different spray flow rates, and the changes in surface humidity, internal pressure, and surface densification were recorded before and after spraying. The collected data were then substituted into the spray flow rate formula, aiming to maximize the sum of the increase in surface humidity and the decrease in internal pressure. A single-variable optimization method, such as the golden section search, was used to solve for the optimal solution. The value is stored in the control system after calibration and can be recalibrated offline according to different formulations or environmental conditions.

[0084] The meaning of this formula is that areas with higher risk values ​​and more drastic risk changes require a larger spray flow rate to quickly soften the surface.

[0085] Spray duration Remaining path length from current location to the end of the path And related to the current risk value: ; In the formula, Based on duration; The threshold is mild. This represents the total length of the current path. δ is a small constant to prevent division by zero.

[0086] The meaning of this formula is: the farther away from the endpoint and the higher the current risk value, the longer the spray duration, because the nozzle has a long enough path to gradually reduce the risk; when approaching the endpoint, even if the risk is still high, the spray time will be shortened to avoid excessive spraying in local areas.

[0087] For example, if the nozzle is located near the start of a path and the remaining path is long, and the current risk value is in the severe range, then the spray flow rate is large and the duration is long; when the nozzle moves to a position near the end point and the remaining path is short, even if the risk value is still moderate, the spray duration will automatically shorten.

[0088] IV: While the nozzle moves and sprays, this module dynamically distributes power to the zoned infrared heating tube array via a power distributor. The goal of power distribution is to reduce heating power in high-risk areas to prevent further densification, maintain normal heating in low-risk areas to promote drying, and coordinate with the spraying action.

[0089] The power distributor first divides the surface of the preform into several power control zones corresponding to the heating tube array. For each power control zone, the module calculates the heating power coefficient η of that zone based on the risk value of that zone at the current moment, whether that zone is being sprayed, and the distance between that zone and the current position of the nozzle.

[0090] The formula for calculating the heating power coefficient is: ; In the formula, The base power factor is set to 1 here; This represents the average risk value for the power control zone. The higher the coefficient, the smaller the power consumption, and the greater the power reduction. This represents the maximum possible risk value.

[0091] As a spray interference factor, when the area is being sprayed, A value less than 1 is chosen to further reduce the heating power, even to zero, to avoid rapid secondary densification of the surface layer caused by heating during spraying; after a period of time has elapsed since the spraying ended... Gradually recover to 1; This is a distance factor, related to the distance from the current position of the nozzle to the area. The closer the distance, the better. The smaller the value, the more the heating power should be temporarily reduced in the area near the nozzle to match the localized high humidity environment created by the spray; the farther the distance, the better. The closer it is to 1.

[0092] The power distributor adjusts the actual output power of each zone heating element based on the calculated power coefficient through pulse width modulation or analog voltage regulation.

[0093] For example, the nozzle is spraying water above a high-risk area, and the power control zone for that area... Very high, It is the minimum value. It's also very small; the calculated power coefficient is close to zero, meaning the heating element in this area hardly generates any heat. Meanwhile, in another low-risk area far from the nozzle... Very low, =1, The power coefficient is close to 1, indicating that the heating element in this area is heating normally.

[0094] The adaptive predictive control module 400 determines whether there are areas with extremely high risk values ​​and large gradient magnitudes based on the risk gradient field. These areas indicate that local pressure buildup is very severe, and simple spraying may not be enough to quickly release the pressure. Cooling is required first to cause the surface to shrink and crack. Therefore, the cold air auxiliary unit is only activated in the above-mentioned high-risk areas when rapid cooling is required.

[0095] The triggering condition is: the overall risk value of a certain area is higher than the severe threshold, and the gradient modulus of that area is greater than the preset gradient threshold. When the condition is met, the adaptive predictive control module 400 activates the cold air auxiliary unit to blow room temperature or low temperature air into the area until the risk value of the area drops below the moderate threshold or the gradient modulus decreases below the threshold.

[0096] When multiple high-risk areas exist on the surface of the substrate and are far apart, the adaptive predictive control module 400 can simultaneously control multiple movable micro-mist nozzles. Each nozzle is assigned an independent descent path. The path planner uses a virtual repulsive force mechanism when generating the path to prevent different nozzles from having paths that are too close, leading to duplicate spraying or mutual interference. Specifically, a repulsive potential field generated by the current positions of other nozzles is superimposed on the risk potential field, and the magnitude of the repulsive force is inversely proportional to the distance between the nozzles. When each nozzle calculates its own descent path, it is simultaneously affected by the negative direction of the original risk gradient and the repulsive force from other nozzles, thus naturally avoiding the areas being processed by other nozzles.

[0097] After each drying batch is completed, the online learning submodule obtains the integral value of demolding resistance and the proportion of surface defect area, and constructs a comprehensive cost function, which is obtained by weighted summation of the proportion of surface defect area and the integral value of demolding resistance.

[0098] Specifically, this step uses the coordinate descent method to adjust and optimize these key control parameters, which include the base spray flow rate, proportional coefficient, mild threshold, moderate threshold, and severe threshold.

[0099] It should be noted that the specific steps of the coordinate descent method are as follows: Arrange all parameters to be optimized in a fixed order. Take the first parameter, keeping the other parameters unchanged, and try increasing and decreasing the step size by a small step size based on the current value, obtaining two new values. Use these two new values ​​in the subsequent two batches of production to calculate the corresponding cost function values. Then select the value that makes the cost function smaller as the new setting value for the parameter. Then process the next parameter in order, keeping the other parameters unchanged and performing single-parameter optimization. After traversing all parameters in one round, if the sum of the decrease in the cost function of all parameters in this round is less than the preset stopping threshold, the optimization is considered to have converged, and the adjustment ends; otherwise, the next round of the loop begins.

[0100] The adaptive predictive control module 400 monitors the risk value of all grid points and the overall average moisture content in real time. Drying is considered complete when the following two conditions are met: the risk value of all grid points is lower than the mild threshold; the overall average moisture content is lower than the target final moisture content, and the absolute value of the rate of change of moisture content within multiple consecutive time windows is less than the set value.

[0101] After the determination is completed, the adaptive predictive control module 400 outputs the dried mature salt embryo and the complete control log of this drying process, including the trajectory point sequence of each planned path, the actual spray flow rate and duration at each location, the power change curve of each power control zone, the start and stop records of the cold air auxiliary unit, and the evolution data of the risk value over time, and transmits it to the demolding and quality closed-loop module 500.

[0102] The demolding and quality closed-loop module 500 receives the dried mature salt pellets and their complete control logs from the adaptive predictive control module 400, and is used to complete the demolding operation, quality inspection, and closed-loop feedback of production data for the dried mature salt pellets. This module includes a lifting demolding mechanism, a demolding resistance sensor, a machine vision inspection unit, and a quality database.

[0103] At the start of the demolding process, the demolding and quality closed-loop module 500 sends the dried, mature salt ingot along with the mold into the demolding station. The lifting demolding mechanism consists of a servo motor, ejector pin, and top plate, with the top plate aligned with the ejector hole at the bottom of the mold. After demolding is initiated, the servo motor drives the ejector pin upward, and the top plate pushes the bottom of the salt ingot, causing it to gradually detach from the mold cavity. During the upward movement of the ejector pin, a demolding resistance sensor installed between the ejector pin and the top plate collects the axial force on the ejector pin in real time, generating a curve showing the change in demolding resistance as the ejector pin displacement changes.

[0104] The typical shape of the demolding resistance curve is as follows: In the initial stage, the bottom of the salt blank disengages from the mold, and the resistance is relatively small; as the ejector pin continues to rise, the largest cross-section of the salt blank experiences sliding friction with the mold sidewall, and the resistance gradually increases and reaches a peak; when the largest cross-section of the salt blank completely disengages from the mold, the resistance rapidly decreases to near zero. This module records the peak demolding resistance and the ejector pin displacement from the start of demolding to the appearance of the resistance peak. It also calculates the integral value of the demolding resistance curve from the start to the end point throughout the entire demolding process; this integral value reflects the overall difficulty of demolding.

[0105] For example, a well-dried salt blank has a smooth surface, stable dimensions, and moderate resistance when demolding, with the resistance curve showing a smooth single-peak shape; while a salt blank with internal micro-cracks or residual stickiness on the surface may show abnormal fluctuations in resistance or excessively high peaks when demolding, indicating that there is a problem with the drying quality of this batch.

[0106] After the salt ingot is completely demolded, the demolding and quality closed-loop module 500 conveys it to the machine vision inspection station. The machine vision inspection unit consists of a high-resolution industrial camera, a ring light source, and an image processing computer. The ring light source is installed around the camera lens to provide uniform shadowless illumination, clearly revealing the surface features of the salt ingot. The camera captures full-surface images of the salt ingot from multiple angles, including top view, bottom view, and multiple side views.

[0107] The image processing computer sequentially performs image preprocessing, target region segmentation, and defect identification on the acquired images. Preprocessing includes grayscale conversion, filtering and noise reduction, and contrast enhancement; in the segmentation stage, a threshold segmentation algorithm is used to extract the contour region of the salt embryo, separating the background from the target; within the target region, an edge detection algorithm is used to identify the contours of defects such as surface cracks, pores, missing corners, and deformation.

[0108] For each identified defect, the demolding and quality closed-loop module 500 calculates its area, length, width, and the proportion of the defect area to the total surface area of ​​the salt blank. The ratio of the total defect area to the total surface area of ​​the salt blank is defined as the surface defect area percentage. For example, a high-quality salt blank has a surface defect area percentage close to zero; while a salt blank with multiple microcracks may have a surface defect area percentage of one percent or higher. This module also records the type and location distribution of defects, forming a defect distribution map for subsequent quality traceability.

[0109] The demolding and quality closed-loop module 500 associates and stores the integral value of demolding resistance, peak value of demolding resistance, proportion of surface defect area, and defect distribution map obtained from each demolding process with the complete control log from the adaptive predictive control module 400 to form a quality database. Each quality record corresponds to a batch of salt embryos and includes the control parameters used in the production process of that batch, risk evolution data during the drying process, and the final drying quality indicators.

[0110] The demolding and quality closed-loop module 500 performs statistical analysis on historical data in the quality database. For each set of control parameters, it calculates the average demolding resistance integral value and the average surface defect area ratio of the corresponding batch. By comparing the quality indicators under different parameter combinations, it identifies the control parameter combination with the best performance.

[0111] Meanwhile, the demolding and quality closed-loop module 500 analyzes the correlation between various control parameters and quality indicators, and identifies several key parameters that have the most significant impact on drying quality, such as the spray flow rate ratio coefficient, the threshold of each risk level, and the power reduction coefficient.

[0112] For example, by analyzing historical data, it may be found that when the severity threshold is set too high, the proportion of surface defect area increases significantly, indicating that the severity threshold needs to be appropriately lowered; when the spray flow rate ratio coefficient is too low, the integral value of demolding resistance is generally too high, indicating that insufficient spray volume leads to a certain degree of densification on the surface, and the spray flow rate needs to be increased.

[0113] Based on the above statistical analysis results, the demolding and quality closed-loop module 500 generates feedback adjustment instructions, calculating the correction direction and approximate magnitude of each parameter with the dual objectives of minimizing the average surface defect area ratio and the average demolding resistance integral value. The feedback adjustment instructions are sent to the online learning step of the adaptive predictive control module 400 for updating control parameters in subsequent batch production. Simultaneously, the demolding and quality closed-loop module 500 marks salt blanks with a surface defect area ratio below the acceptable threshold as finished salt blanks for output, and sends severely defective salt blanks with a surface defect area ratio above the acceptable threshold to the scrap recycling channel.

[0114] like Figure 4 As shown, the second objective of this invention is to provide an intelligent control method for automated salt carving production, comprising the following steps: S1. Mix salt powder, adhesive and water, collect stirring torque during mixing and calculate mixing uniformity index to generate homogeneous mixture; S2. Inject the homogeneous mixture into the mold cavity, collect the injection pressure curve and the mold cavity pressure distribution, and generate a salt sculpture blank with an initial density distribution matrix. S3. Using the salt sculpture blank as the observation object, its surface temperature field matrix, surface humidity field matrix and overall average moisture content are collected respectively. Based on the extended Kalman filter, the pressure value and surface densification degree value of each grid point inside the blank are recursively estimated by substituting into the state transition equation and the observation equation. The self-hindrance risk index, which characterizes the risk of internal pressure buildup in the embryo, is calculated based on the pressure value and the degree of surface densification, and then predicted forward to generate a risk spatiotemporal distribution map and an intervention priority queue. S4. Convert the risk spatiotemporal distribution map into a continuous risk potential field function and calculate its gradient field. Generate the descent path of the moving nozzle along the negative gradient direction, starting from the local maximum point of risk. The movable micro-mist nozzle moves along the descent path and sprays according to the self-obstruction risk index of the current location, while adjusting the heating power of the corresponding location according to the self-obstruction risk index of each area. When the self-obstruction risk index of all grid points is below the safety threshold and the overall average moisture content meets the standard, dry and mature salt embryos are output.

[0115] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent control system for automated salt sculpture production, comprising a raw material pretreatment module (100) and a blank forming and initial state calibration module (200), wherein the raw material pretreatment module (100) is used to mix salt powder, adhesive and water to generate a homogeneous mixture, and the blank forming and initial state calibration module (200) is used to inject the homogeneous mixture into a mold to generate a salt sculpture blank, characterized in that, It also includes a dry state spatiotemporal observation module (300) and an adaptive predictive control module (400); The dry state spatiotemporal observation module (300) takes the salt sculpture blank as the observation object, collects its surface temperature field matrix, surface humidity field matrix and overall average moisture content respectively, and recursively estimates the pressure value and surface densification degree value of each grid point inside the blank by substituting the extended Kalman filter into the state transition equation and observation equation. The self-hindrance risk index, which characterizes the risk of internal pressure buildup in the embryo, is calculated based on the pressure value and the degree of surface densification, and then predicted forward to generate a risk spatiotemporal distribution map and an intervention priority queue. The adaptive predictive control module (400) converts the risk spatiotemporal distribution map into a continuous risk potential field function and calculates its gradient field. Starting from the local maximum point of risk, it generates a descent path of the moving nozzle along the negative gradient direction. The movable micro-mist nozzle moves along the descent path and sprays according to the self-obstruction risk index of the current location, while adjusting the heating power of the corresponding location according to the self-obstruction risk index of each area. When the self-obstruction risk index of all grid points is below the safety threshold and the overall average moisture content meets the standard, dry and mature salt embryos are output.

2. The intelligent control system for automated salt carving production according to claim 1, characterized in that: The raw material pretreatment module (100) collects the stirring torque during the mixing of salt powder, adhesive and water, and calculates the mixing uniformity index by extracting the main frequency energy ratio of the torque spectrum through fast Fourier transform. When the mixing uniformity index is continuously stable, the mixing is determined to be complete.

3. The intelligent control system for automated salt carving production according to claim 1, characterized in that: The preform forming and initial state calibration module (200) collects the injection pressure curve during the injection molding process, determines the injection molding end time by identifying the peak point of the first derivative of the curve, and collects the pressure value at each point at the injection molding end time through the mold cavity pressure distribution array to form the initial pressure distribution matrix of the salt sculpture preform.

4. The intelligent control system for automated salt carving production according to claim 3, characterized in that: The embryo forming and initial state calibration module (200) calculates the density index of each point according to the initial pressure distribution matrix, and generates the initial density distribution matrix of the salt sculpture embryo.

5. The intelligent control system for automated salt carving production according to claim 1, characterized in that: The dry state spatiotemporal observation module (300) acquires the surface temperature field matrix through an infrared thermal imaging array, acquires the surface humidity field matrix through a laser humidity reflectometer array mounted on a servo slide, and acquires the overall average moisture content through a microwave humidity sensor fixed to the top of the drying chamber.

6. The intelligent control system for automated salt carving production according to claim 1, characterized in that: The state transition equation used by the extended Kalman filter of the dry state spatiotemporal observation module (300) is that the partial derivative of internal pressure with respect to time is equal to the first model coefficient multiplied by the Laplace operator of the surface temperature field, plus the second model coefficient multiplied by the difference between the overall average moisture content and the equilibrium moisture content, and then minus the third model coefficient multiplied by the product of the surface densification degree and the difference between internal pressure and environmental pressure. The observation equation is a vector composed of surface temperature measurement, surface humidity measurement, and overall average moisture content measurement. This vector is equal to the observation matrix multiplied by the state vector plus the measurement noise.

7. The intelligent control system for automated salt carving production according to claim 1, characterized in that: The model coefficients used in the extended Kalman filter are calibrated as follows: a standard sample is prepared and a sensor is pre-embedded. During the drying process, the measured pressure, temperature, humidity and moisture content are recorded. The sum of squared errors between the measured values ​​and the calculated values ​​is minimized, and the coefficients are obtained by using a nonlinear least squares fitting algorithm.

8. The intelligent control system for automated salt carving production according to claim 1, characterized in that: The adaptive predictive control module (400) activates cold air-assisted cooling when the self-impedance risk index exceeds the severe threshold, causing micro-cracks to form on the surface hard shell to release internal pressure.

9. The intelligent control system for automated salt carving production according to claim 1, characterized in that: The adaptive predictive control module (400) transmits the dried mature salt embryo to the demolding and quality closed-loop module (500). The demolding and quality closed-loop module (500) demolds the dried mature salt embryo and collects the demolding resistance. After detecting surface defects, it generates feedback parameters and transmits them to the adaptive predictive control module (400).

10. An intelligent control method using the intelligent control system for automated salt carving production according to any one of claims 1-9, characterized in that, Includes the following steps: S1. Mix salt powder, adhesive and water, collect stirring torque during mixing and calculate mixing uniformity index to generate homogeneous mixture; S2. Inject the homogeneous mixture into the mold cavity, collect the injection pressure curve and the mold cavity pressure distribution, and generate a salt sculpture blank with an initial density distribution matrix. S3. Using the salt sculpture blank as the observation object, its surface temperature field matrix, surface humidity field matrix and overall average moisture content are collected respectively. Based on the extended Kalman filter, the pressure value and surface densification degree value of each grid point inside the blank are recursively estimated by substituting into the state transition equation and the observation equation. The self-hindrance risk index, which characterizes the risk of internal pressure buildup in the embryo, is calculated based on the pressure value and the degree of surface densification, and then predicted forward to generate a risk spatiotemporal distribution map and an intervention priority queue. S4. Convert the risk spatiotemporal distribution map into a continuous risk potential field function and calculate its gradient field. Generate the descent path of the moving nozzle along the negative gradient direction, starting from the local maximum point of the risk. The movable micro-mist nozzle moves along the descent path and sprays according to the self-obstruction risk index of the current location, while adjusting the heating power of the corresponding location according to the self-obstruction risk index of each area. When the self-obstruction risk index of all grid points is below the safety threshold and the overall average moisture content meets the standard, dry and mature salt embryos are output.

Images