Rock wool board production control method and control system

Abstract

The invention relates to a rock wool board production control method and control system. The method comprises the following steps: firstly, constructing a static production steady state domain for defining an ideal state based on historical superior product process parameters; and multi-dimensional process parameters are fused in real time, and equipment and environment constraints are combined to construct a dynamic production control domain representing a current safe operation boundary. The real-time production state points of the current working condition are combined and positioned in the two areas, and the production process is divided into the optimal state, the drifting state, the risk state and the degradation state. And aiming at different states, the system automatically generates optimized regression, emergency avoidance or forced regression instructions to guide production parameter adjustment, so that accurate regulation and control of the production process are realized, and the superior product rate and the production stability are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of production control, and in particular to a rock wool board production control method and control system. BACKGROUND

[0002] Rock wool board, as a kind of inorganic fiber material with excellent performance, has been widely used in the fields of construction, industry, shipbuilding, etc. due to its excellent thermal insulation, heat insulation, fireproof and sound absorption performance. The production process of rock wool board is a complex multivariate and strongly coupled continuous industrial process, and its core links involve multiple processes such as high-temperature melting, high-speed centrifugation, glue spraying, cotton collecting, curing, cutting, etc. The process parameters in these processes, such as furnace temperature, centrifuge speed, glue application flow, curing furnace temperature, and conveyor belt speed, etc. are complexly and nonlinearly related to each other, and together determine the key quality indicators of the final product such as density, strength, and thermal conductivity coefficient.

[0003] In the existing rock wool board production control practice, the mode of combining single parameter set value control with manual experience is generally adopted. Specifically, the production line usually sets a fixed target value or a narrow fluctuation range for each key process parameter, and tries to maintain its stability through independent feedback control loop. When production abnormalities or product quality decline occur, experienced operators are mainly relied on for observation, judgment, and manual trial adjustment of one or more parameters. This control method has significant defects: first, it ignores the synergistic effect between process parameters, and cannot grasp the overall running state of the production process from a global perspective. When multiple parameters deviate slightly at the same time, the single control loop cannot determine whether this combined effect has caused the production state to deviate from the optimal region, thereby causing the product quality to slowly "drift". Secondly, the traditional safety control logic is usually based on fixed and static parameter limit thresholds, which cannot adapt to dynamic risk factors such as raw material batch fluctuations, key equipment wear, and environmental temperature and humidity changes. This makes the production process very vulnerable when facing these potential disturbances, lacking of predictive risk avoidance ability, and easily leading to production interruption and even safety accidents. Finally, the lag and subjectivity of manual intervention make the control adjustment not timely and accurate enough, making it difficult to achieve optimal response to complex working conditions.

[0004] Therefore, there is an urgent need in the industry for a new production control method that can overcome the above-mentioned defects. This method needs to be able to comprehensively evaluate the overall impact of multiple process parameters, and construct a global description of the ideal production state; at the same time, it also needs to be able to real-time perceive and quantify the dynamic risks in the production process, and form adaptive safety operating boundaries; and on this basis, intelligently diagnose the current production state, automatically generate precise and coordinated multi-parameter control instructions, so as to realize the stable, efficient and safe operation of the production process, and comprehensively improve the product quality and production efficiency of rock wool board. Summary of the Invention

[0005] To address the shortcomings of existing technologies, embodiments of the present invention provide a method and control system for controlling the production of rock wool boards.

[0006] The rock wool board production control method of the present invention includes the following steps: Historical process parameters corresponding to historical high-quality batches are extracted in advance, and the historical process parameters are nonlinearly fused and mapped to generate a set of historical fusion feature vectors. Then, a static production steady-state domain for defining the ideal production state is constructed based on the overall distribution of the set of historical fusion feature vectors. Multiple rock wool process parameters during the production process are acquired in real time. All the acquired rock wool process parameters are nonlinearly fused and mapped to generate a real-time fused feature vector that characterizes the overall operating condition of the current production process. A dynamic production control domain is constructed to define the safe operating boundary of the current operating condition. The real-time fused feature vector is determined as the real-time production state point in the low-dimensional latent space of the autoencoder neural network. The real-time production state point is synchronously projected and located in the static production steady-state domain and the dynamic production control domain to obtain the combined positional relationship between the real-time production state point and the static production steady-state domain and the dynamic production control domain. Then, the production state region of the real-time production state point is determined according to the combined positional relationship. The production state region includes: the optimal state region, the drift state region, the risk state region, and the deterioration state region. When the real-time production status point is in the drift state zone, an optimization regression instruction is generated and executed; When the real-time production status point is located in the risk status zone, an emergency avoidance instruction is generated and executed. When the real-time production status point is in the deterioration state zone, a forced regression instruction is generated and executed.

[0007] According to a preferred embodiment, performing a nonlinear fusion mapping on all real-time acquired rock wool process parameters to generate a real-time fusion feature vector includes: Multiple rock wool process parameters acquired in real time are combined into an input vector; the rock wool process parameters include furnace zone temperature, centrifuge speed, blower air pressure, adhesive application flow rate, curing oven temperature, and conveyor belt speed; The encoder in an autoencoder neural network is used to perform nonlinear compression on the input vector to generate real-time fused feature vectors in the low-dimensional latent space of the autoencoder neural network.

[0008] According to a preferred embodiment, constructing a static production steady-state domain based on the overall distribution of a set of historical fusion feature vectors includes: A set of historical fusion feature vectors is identified as a set of historical optimal state points in the low-dimensional latent space of the autoencoder neural network and aggregated into a core state point cloud. Calculate the geometric center of the core state point cloud to determine the geometric center as the ideal process centroid representing the theoretically optimal working condition; A robustness metric is calculated based on the statistical distribution characteristics of all historical best state points in the core state point cloud relative to the ideal process centroid. In the low-dimensional potential space, a closed geometry is constructed with the ideal process centroid as the center and based on the robustness metric. The internal space region defined by the closed geometry is determined as the static production steady-state domain.

[0009] According to a preferred embodiment, constructing a dynamic production control domain includes: Within the low-dimensional potential space, a maximum operational envelope is predefined, which is mapped from the physical limits of the equipment and the process safety constraints. A set of process constraint factors characterizing the potential risks of the current production process are acquired in real time; the process constraint factors include raw material batch fluctuations, equipment wear status, or abnormal environmental parameters. The set of process constraint factors is input into a pre-trained risk quantification model to output a set of boundary convergence vectors; the direction of the boundary convergence vectors represents the direction of boundary contraction; the magnitude of the boundary convergence vectors represents the magnitude of boundary contraction. The maximum operational envelope is contracted according to the set of boundary convergence vectors to generate a dynamic closed surface, and the internal spatial region defined by the dynamic closed surface is determined as the dynamic production control domain.

[0010] According to a preferred embodiment, determining the production status region of a real-time production status point based on a combination of positional relationships includes: When the real-time production status point is simultaneously located within the static production steady-state domain and the dynamic production control domain, it is determined that the real-time production status point is located in the optimal state zone. When the real-time production status point is located inside the dynamic production control domain but outside the static production steady-state domain, it is determined that the real-time production status point is located in the drift state region. When the real-time production status point is located inside the static production steady-state domain but outside the dynamic production control domain, it is determined that the real-time production status point is located in the risk state zone. When the real-time production status point is located outside both the static production steady-state domain and the dynamic production control domain, it is determined that the real-time production status point is located in the deterioration state zone.

[0011] According to a preferred embodiment, when the real-time production state point is located in the drift state region, generating and executing the optimization regression instruction includes: In the low-dimensional potential space, the ideal process centroid of the static production steady-state domain is determined as the optimization regression target point; Calculate an optimized regression vector pointing from the real-time production status point to the optimized regression target point; The decoder in the autoencoder neural network is used to inversely map the optimized regression vector into a set of process parameter adjustment values ​​in the process parameter space; Generate and execute an optimized regression instruction that includes the process parameter adjustment values.

[0012] According to a preferred embodiment, when the real-time production status point is located in a risky state zone, generating and executing an emergency avoidance instruction includes: In the low-dimensional potential space, on the boundary of the dynamic production control domain, a boundary point that is closest to the real-time production state point in Euclidean distance is determined as the emergency avoidance target point. Calculate an emergency avoidance vector pointing from the real-time production status point to the emergency avoidance target point; The decoder in the autoencoder neural network is used to inversely map the emergency avoidance vector into a set of process parameter adjustment values ​​in the process parameter space; Generate and execute an emergency avoidance instruction containing the process parameter adjustment values.

[0013] According to a preferred embodiment, when the real-time production status point is located in the deterioration state zone, generating and executing a forced regression instruction includes: In the low-dimensional potential space, a pre-defined safety state benchmark point located within the dynamic production control domain is determined as the forced regression target point; Calculate a forced regression vector pointing from the real-time production status point to the forced regression target point; The decoder in the autoencoder neural network is used to inversely map the forced regression vector into a set of process parameter adjustment values ​​in the process parameter space; Generate and execute a forced regression instruction containing the process parameter adjustment values.

[0014] According to a preferred embodiment, the optimized regression instruction aims to adjust the real-time production state point to within the static production steady-state domain; the emergency avoidance instruction aims to adjust the real-time production state point to within the dynamic production control domain; and the forced regression instruction aims to adjust the real-time production state point to within the dynamic production control domain.

[0015] The rock wool board production control system of the present invention includes: The static domain construction module is used to pre-extract historical process parameters corresponding to historical high-quality batches, perform nonlinear fusion mapping on the historical process parameters to generate a set of historical fusion feature vectors, and construct a static production steady-state domain for defining the ideal production state based on the overall distribution of the set of historical fusion feature vectors. The dynamic domain construction module is used to acquire various rock wool process parameters in the production process in real time, and to perform nonlinear fusion mapping on all the acquired rock wool process parameters to generate a real-time fusion feature vector that characterizes the overall operating condition of the current production process, and to construct a dynamic production control domain that defines the safe operating boundary of the current operating condition. The state determination module is used to determine the real-time production state point in the low-dimensional latent space of the autoencoder neural network by the real-time fused feature vector, and to synchronously project and locate the real-time production state point in the static production steady-state domain and the dynamic production control domain to obtain their combined positional relationship, and to determine the production state region to which the real-time production state point belongs based on the combined positional relationship. The production state region includes: the optimal state region, the drift state region, the risk state region, and the deterioration state region. Control command generation module, used for When it is determined that the real-time production state point is in the drift state region, an optimized regression instruction is generated and executed to adjust the real-time production state point to the interior of the static production steady-state domain. When it is determined that the real-time production status point is located in the risk state zone, an emergency avoidance instruction is generated and executed to adjust the real-time production status point to the inside of the dynamic production control domain. When it is determined that the real-time production status point is in the deterioration state zone, a forced regression instruction is generated and executed to adjust the real-time production status point to the dynamic production control domain.

[0016] The present invention has the following beneficial effects: 1. Improve product quality stability and yield. This application establishes a quantitative benchmark for ideal production by constructing a static steady-state domain based on historical yield data. When production deviates from this target, the system can identify it as a drift state and initiate optimization commands to flexibly adjust process parameters back to the optimal range. This effectively suppresses process drift, ensuring that production always revolves around best practices, thereby significantly improving the yield and batch-to-batch quality consistency.

[0017] 2. Enhanced safety and robustness of the production process. This application's unique dynamic production control domain can dynamically adjust the safe operating boundary based on real-time risk factors such as equipment wear and raw material anomalies. Once the production state point exceeds this boundary, the system determines it as a risky or deteriorating state and triggers emergency avoidance or forced return commands. This mechanism acts like an intelligent safety barrier, effectively anticipating and avoiding production risks, greatly enhancing the system's anti-interference capabilities.

[0018] 3. Achieving intelligent and precise process control. This application utilizes neural networks to fuse multi-dimensional process parameters into low-dimensional state points that characterize the overall operating condition, simplifying monitoring. Based on this, the system clearly divides complex operating conditions into four states and matches precise multi-parameter collaborative adjustment commands, realizing the transformation from passive adjustment of single parameters to proactive global optimization, significantly improving the automation and intelligence level of production control. Attached Figure Description

[0019] Figure 1 A structural block diagram of a rock wool board production control system provided as an exemplary embodiment; Figure 2 A flowchart of a rock wool board production control method provided as an exemplary embodiment. Detailed Implementation

[0020] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0021] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0022] It should be understood that although the terms first, second, third, etc., may be used in this invention to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of this invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0023] See Figure 1 The rock wool board production control system of the present invention may include: The static domain construction module is used to pre-extract historical process parameters corresponding to historical high-quality batches, perform nonlinear fusion mapping on the historical process parameters to generate a set of historical fusion feature vectors, and construct a static production steady-state domain for defining the ideal production state based on the overall distribution of the set of historical fusion feature vectors. The dynamic domain construction module is used to acquire various rock wool process parameters in the production process in real time, and to perform nonlinear fusion mapping on all the acquired rock wool process parameters to generate a real-time fusion feature vector that characterizes the overall operating condition of the current production process, and to construct a dynamic production control domain that defines the safe operating boundary of the current operating condition. The state determination module is used to determine the real-time production state point in the low-dimensional latent space of the autoencoder neural network by the real-time fused feature vector, and to synchronously project and locate the real-time production state point in the static production steady-state domain and the dynamic production control domain to obtain their combined positional relationship, and to determine the production state region to which the real-time production state point belongs based on the combined positional relationship. The production state region includes: the optimal state region, the drift state region, the risk state region, and the deterioration state region. Control command generation module, used for When it is determined that the real-time production state point is in the drift state region, an optimized regression instruction is generated and executed to adjust the real-time production state point to the interior of the static production steady-state domain. When it is determined that the real-time production status point is located in the risk state zone, an emergency avoidance instruction is generated and executed to adjust the real-time production status point to the inside of the dynamic production control domain. When it is determined that the real-time production status point is in the deterioration state zone, a forced regression instruction is generated and executed to adjust the real-time production status point to the dynamic production control domain.

[0024] See Figure 2 The rock wool board production control method of the present invention includes: S1. Pre-extract historical process parameters corresponding to historical high-quality batches, and perform nonlinear fusion mapping on the historical process parameters to generate a set of historical fusion feature vectors. Then, construct a static production steady-state domain to define the ideal production state based on the overall distribution of a set of historical fusion feature vectors.

[0025] Optionally, the historical superior batches include multiple superior batches, and the historical process parameters include combinations of rock wool process parameters from multiple superior batches during the production of rock wool boards. When we perform a "nonlinear fusion mapping" on the combination of rock wool process parameters from a specific superior batch at a specific point in time, we output a "historical fusion feature vector." This historical fusion feature vector represents a "historical optimal state point" in the low-dimensional latent space.

[0026] In a preferred embodiment, constructing a static production steady-state domain based on the overall distribution of a set of historical fusion feature vectors includes: A set of historical fusion feature vectors is identified as a set of historical optimal state points in the low-dimensional latent space of the autoencoder neural network and aggregated into a core state point cloud. Calculate the geometric center of the core state point cloud to determine the geometric center as the ideal process centroid representing the theoretically optimal working condition; A robustness metric is calculated based on the statistical distribution characteristics of all historical best state points in the core state point cloud relative to the ideal process centroid. In the low-dimensional potential space, a closed geometry is constructed with the ideal process centroid as the center and based on the robustness metric. The internal space region defined by the closed geometry is determined as the static production steady-state domain.

[0027] Optionally, the mean vector of the core state point cloud is calculated and determined as the ideal process centroid representing the theoretically optimal working condition.

[0028] Optionally, the specific process for calculating the robustness metric based on the statistical distribution characteristics of all historical best state points in the core state point cloud relative to the ideal process centroid is as follows: calculate the covariance matrix of the core state point cloud, which describes the shape of the core state point cloud in the low-dimensional latent space, and calculate the Mahalanobis distance from each historical best state point to the ideal process centroid. The robustness metric is a distance threshold (for example, taking the average of the Mahalanobis distances of all historical points plus 3 times the standard deviation, or taking a certain confidence threshold of the chi-square distribution).

[0029] Optionally, the specific process of constructing a closed geometry based on the robustness measure is as follows: using the calculated "ideal process centroid" as the center, a hyperellipsoid is constructed using the covariance matrix and the robustness measure.

[0030] Optionally, the static production steady-state domain is directly used to define the ideal production state. This region represents the theoretically most perfect operating mode of the rock wool board production process, that is, the range of process parameter combinations that can continuously and stably produce high-quality products.

[0031] S2. Real-time acquisition of various rock wool process parameters during the production process, nonlinear fusion mapping of all real-time acquired rock wool process parameters to generate a real-time fusion feature vector to characterize the overall operating condition of the current production process, and construction of a dynamic production control domain to define the safe operating boundary of the current operating condition.

[0032] Optionally, the real-time fusion feature vector represents the overall operating condition of the current production process. It is the product of nonlinear fusion mapping of various rock wool process parameters acquired in real time (including furnace zone temperature, centrifuge speed, blower air pressure, adhesive application flow rate, curing oven temperature, and conveyor belt speed). It is not a single isolated physical parameter, but a digital feature that comprehensively reflects the overall operating status of the production line at this moment.

[0033] Optionally, the rock wool process parameters include furnace zone temperature, centrifuge speed, blower air pressure, adhesive application flow rate, curing oven temperature, and conveyor belt speed.

[0034] Preferably, performing nonlinear fusion mapping on all real-time acquired rock wool process parameters to generate a real-time fusion feature vector includes: Multiple rock wool process parameters acquired in real time are combined into an input vector; The encoder in an autoencoder neural network is used to perform nonlinear compression on the input vector to generate real-time fused feature vectors in the low-dimensional latent space of the autoencoder neural network.

[0035] In a preferred embodiment, constructing a dynamic production control domain includes: Within the low-dimensional potential space, a maximum operational envelope is predefined, which is mapped from the physical limits of the equipment and the process safety constraints. A set of process constraint factors characterizing the potential risks of the current production process are acquired in real time; the process constraint factors include raw material batch fluctuations, equipment wear status, or abnormal environmental parameters. The set of process constraint factors is input into a pre-trained risk quantification model to output a set of boundary convergence vectors; the direction of the boundary convergence vectors represents the direction of boundary contraction; the magnitude of the boundary convergence vectors represents the magnitude of boundary contraction. The maximum operational envelope is contracted according to the set of boundary convergence vectors to generate a dynamic closed surface, and the internal spatial region defined by the dynamic closed surface is determined as the dynamic production control domain.

[0036] In this embodiment, the risk quantification model can employ a backpropagation (BP) neural network model. The input layer of this model receives normalized process constraint factors, and the output layer outputs the magnitude and direction angle of the boundary convergence vector. The risk quantification model is trained using supervised learning with boundary trial data from historical production to establish a nonlinear mapping relationship between risk factors and the shrinkage of the safety boundary.

[0037] It should be understood that the risk quantification model is not limited to neural networks, but can also be implemented using support vector regression, random forest regression or fuzzy logic inference systems, as long as the corresponding boundary adjustment parameters can be quantitatively calculated based on the input risk factors.

[0038] In this embodiment, in order to determine the initial maximum boundary of the dynamic production control domain, the system first needs to construct a "maximum operational envelope" in a low-dimensional potential space. The maximum operational envelope represents the limit range within which the production equipment is physically allowed to operate without fundamental safety incidents (such as explosions, fires, or equipment damage).

[0039] Optionally, the dynamic production control domain is the safe operating boundary under the current working conditions. Unlike the static steady-state production domain, which focuses on "quality," the dynamic domain focuses on "feasibility" and "safety." It defines the parameter limits within which the production process can operate safely without accidents under the current equipment status, raw material conditions, and environmental factors.

[0040] S3. The real-time fused feature vector is determined as the real-time production state point in the low-dimensional latent space of the autoencoder neural network. The real-time production state point is synchronously projected and located in the static production steady-state domain and the dynamic production control domain to obtain the combined positional relationship between the real-time production state point and the static production steady-state domain and the dynamic production control domain. Then, the production state region of the real-time production state point is determined according to the combined positional relationship.

[0041] Optionally, the real-time production status point represents the overall operating condition of the current production process. It is not a direct reflection of a single process parameter, but rather the result of a nonlinear fusion mapping of multiple rock wool process parameters, such as furnace temperature, centrifuge speed, blower pressure, adhesive application flow rate, curing oven temperature, and conveyor belt speed. It comprehensively characterizes the overall operating status of the production line at the current moment, representing the specific coordinate position of the "real-time fused feature vector" in the "low-dimensional latent space" of the autoencoder neural network. Through compression by the autoencoder, it transforms complex high-dimensional physical parameters into a point that can be geometrically located in mathematical space.

[0042] Optionally, the production status areas include: the optimal status area, the drift status area, the risk status area, and the deterioration status area.

[0043] Preferably, determining the production status area of ​​the real-time production status point based on the combined positional relationship includes: When the real-time production status point is simultaneously located within the static production steady-state domain and the dynamic production control domain, it is determined that the real-time production status point is located in the optimal state zone. When the real-time production status point is located inside the dynamic production control domain but outside the static production steady-state domain, it is determined that the real-time production status point is located in the drift state region. When the real-time production status point is located inside the static production steady-state domain but outside the dynamic production control domain, it is determined that the real-time production status point is located in the risk state zone. When the real-time production status point is located outside both the static production steady-state domain and the dynamic production control domain, it is determined that the real-time production status point is located in the deterioration state zone.

[0044] The optimal state zone refers to the spatial region where the real-time production state point simultaneously lies within both the static steady-state production domain and the dynamic production control domain (i.e., the intersection of the two domains). This region represents the current production process being in an ideal condition that is both "high-quality and safe." On the one hand, the combination of process parameters highly conforms to the statistical characteristics of historical high-quality batches, indicating that the currently produced rock wool boards are of excellent quality; on the other hand, the current operating state is completely within the safety envelope of the equipment's physical limits and environmental safety constraints, the equipment operates smoothly, and the risk is extremely low. At this time, the system usually does not intervene, or only performs minor maintenance controls to maintain the continuity of the current state.

[0045] The drift state zone refers to the spatial region where the real-time production state point is located within the dynamic production control domain but outside the static steady-state production domain. This region indicates that although the production process is within safe operating limits and has not exceeded equipment limits, the combined effect of process parameters has deviated from the historical optimal standard. This is usually caused by "chronic drift" due to slight fluctuations in raw material batches, slow changes in ambient temperature and humidity, or cumulative errors in the control system. While being in this zone will not cause temporary downtime, continued operation will lead to a decrease in product yield and a decline in product consistency. At this point, an "optimization regression instruction" is triggered, flexibly adjusting parameters with the ideal process centroid as the target, aiming to improve quality.

[0046] The risk zone refers to the spatial region where the real-time production state point is located within the static steady-state production domain but outside the dynamic production control domain. This is a special area with high concealment and danger. From a quality perspective, the current process parameters may still seem to be within the range of historically optimal products (within the static domain); however, due to the existence of special "process constraint factors" at the current moment (such as severe centrifuge wear or excessively high raw material impurities), the safety boundary (dynamic domain) has significantly contracted. This causes parameters that were historically considered "good" to become "dangerous" parameters under the current adverse equipment or raw material conditions. For example, historically, a speed of 4000 rpm was an optimal parameter, but currently, due to severe bearing wear, the dynamic domain boundary has contracted to 3500 rpm. If 4000 rpm is maintained at this point, although it can produce excellent products, it is extremely easy to cause equipment failure. At this time, an "emergency avoidance instruction" is triggered, prioritizing the safety boundary and sacrificing some theoretical optimality in exchange for system safety.

[0047] The deterioration state zone refers to the spatial region where the real-time production state point simultaneously lies outside both the static steady-state production domain and the dynamic production control domain. This zone indicates that the production process is on the verge of "out of control," failing to meet product quality requirements and exceeding current safety operating limits. This typically signifies a severe sudden disturbance, equipment failure, or human error. Operating in this state not only generates a large amount of scrap but also greatly increases the probability of equipment damage, fire, or unplanned production line shutdown. At this point, a "forced regression command" is triggered, forcefully pulling the state point back to an absolutely safe baseline. If necessary, interlocking shutdown protection can be activated.

[0048] S4. When the real-time production status point is in the drift state region, generate and execute an optimization regression instruction. The optimization regression instruction aims to adjust the real-time production status point to the static production steady-state domain.

[0049] Preferably, when the real-time production state point is located in the drift state region, generating and executing the optimization regression instruction includes: In the low-dimensional potential space, the ideal process centroid of the static production steady-state domain is determined as the optimization regression target point; Calculate an optimized regression vector pointing from the real-time production status point to the optimized regression target point; The decoder in the autoencoder neural network is used to inversely map the optimized regression vector into a set of process parameter adjustment values ​​in the process parameter space; Generate and execute an optimized regression instruction that includes the process parameter adjustment values.

[0050] Optimized regression instructions are applicable to the drift state region. At this point, production is within a safe range, but quality exhibits slight fluctuations or deviates from the optimal trend.

[0051] Execution Process: In the low-dimensional latent space, the difference vector from the current "real-time production state point" to the "ideal process centroid" is calculated, i.e., the optimization regression vector. This vector is input into the decoder, which uses its nonlinear generation capability to transform it into a set of fine-tuning increments in the original process parameter space (e.g., furnace temperature -2℃, conveyor belt speed +0.1m / s). This instruction typically manifests as a series of smooth, minute parameter adjustments, designed to eliminate accumulated errors and flexibly "pull" the production process back to its historical optimal state, thereby maximizing the yield of high-quality products.

[0052] S5. When the real-time production status point is located in the risk state zone, an emergency avoidance instruction is generated and executed. The emergency avoidance instruction aims to adjust the real-time production status point to the dynamic production control domain.

[0053] Preferably, when the real-time production status point is located in the risk status zone, generating and executing emergency avoidance instructions includes: In the low-dimensional potential space, on the boundary of the dynamic production control domain, a boundary point that is closest to the real-time production state point in Euclidean distance is determined as the emergency avoidance target point. Calculate an emergency avoidance vector pointing from the real-time production status point to the emergency avoidance target point; The decoder in the autoencoder neural network is used to inversely map the emergency avoidance vector into a set of process parameter adjustment values ​​in the process parameter space; Generate and execute an emergency avoidance instruction containing the process parameter adjustment values.

[0054] Emergency avoidance commands apply to high-risk zones. While the combination of process parameters theoretically produces high-quality products, equipment wear or sudden environmental changes have triggered the current, contracted safety boundary. The system doesn't care whether it returns to the center; it simply searches for the closest Euclidean distance boundary point within the dynamic production control domain near that point as the target point.

[0055] Execution process: Calculate the emergency avoidance vector from the current point to the nearest boundary point. This vector represents the shortest path to escape the risk. After decoding, a set of targeted avoidance parameters is generated. This instruction typically manifests as a rapid correction of one or more specific parameters (e.g., if excessive centrifuge vibration is detected and the risk model shrinks the speed boundary, the instruction will prioritize a significant reduction in centrifuge speed while keeping other parameters as constant as possible). Its purpose is to quickly bring the production state back to the safety envelope with minimal process disturbance, preventing accidents from occurring.

[0056] S6. When the real-time production status point is in the deterioration state zone, a forced regression instruction is generated and executed. The forced regression instruction aims to adjust the real-time production status point to the dynamic production control domain.

[0057] Preferably, when the real-time production status point is in the deterioration state zone, generating and executing the forced regression instruction includes: In the low-dimensional potential space, a pre-defined safety state benchmark point located within the dynamic production control domain is determined as the forced regression target point; Calculate a forced regression vector pointing from the real-time production status point to the forced regression target point; The decoder in the autoencoder neural network is used to inversely map the forced regression vector into a set of process parameter adjustment values ​​in the process parameter space; Generate and execute a forced regression instruction containing the process parameter adjustment values.

[0058] Forced regression instructions are applicable to the deterioration state region. At this point, the production process is neither safe nor can quality be guaranteed, and it is on the verge of serious loss of control. The system no longer relies on the current boundary calculations, but directly locks a pre-set safety state benchmark point deep within the dynamic production control domain (i.e., the absolute safety zone) (usually corresponding to a conservative process mode with low load and high stability).

[0059] Execution process: Calculate the forced regression vector from the current point to the safety baseline. This is a large-scale vector, which, after decoding, generates a set of reset parameters. This instruction typically manifests as a strong intervention and significant pullback of all process parameters. The system forces the production line into a degraded but absolutely safe "conservative operating mode," prioritizing equipment and personnel safety, and then waiting for operator intervention or system re-optimization after stabilization.

[0060] This application establishes a quantitative benchmark for ideal production by constructing a static steady-state domain based on historical high-quality product data. When production deviates from this benchmark, the system can identify it as a drift state and initiate optimization commands to flexibly adjust process parameters back to the optimal range. This effectively suppresses process drift, ensuring that production always revolves around best practices, thereby significantly improving the yield rate of finished products and the consistency of quality between batches.

[0061] This application's unique dynamic production control domain can dynamically adjust the safe operating boundary based on real-time risk factors such as equipment wear and raw material anomalies. Once the production status point exceeds this boundary, the system determines it to be a risky or deteriorating state and triggers an emergency avoidance or forced return command. This mechanism acts like an intelligent safety barrier, effectively anticipating and avoiding production risks, and greatly enhancing the system's anti-interference capability.

[0062] This application utilizes neural networks to fuse multidimensional process parameters into low-dimensional state points that characterize the overall operating condition, simplifying monitoring. Based on this, the system clearly divides complex operating conditions into four states and matches them with precise multi-parameter collaborative adjustment commands, realizing a shift from passive single-parameter adjustment to proactive global optimization, significantly improving the automation and intelligence level of production control.

[0063] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0064] The present invention discloses a non-transitory computer-readable storage medium storing computer instructions, which, when executed by a processor, cause the processor to perform the above-described method.

[0065] Those skilled in the art will understand that all or part of the steps in the above methods can be implemented by a program instructing related hardware (e.g., processor, FPGA, ASIC, etc.), and the program can be stored in a readable storage medium, such as a read-only memory, a disk, or an optical disk. All or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Accordingly, each module in the above embodiments can be implemented in hardware, such as by using integrated circuits to implement its corresponding function, or it can be implemented as a software functional module, such as by a processor executing a program / instruction stored in memory to implement its corresponding function. The embodiments of the present invention are not limited to any particular combination of hardware and software.

[0066] Furthermore, the functional units in the various embodiments of this document can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0067] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this article, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this article. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of controlling the production of rock wool slabs, characterized in that, The method comprises the following steps: extracting historical process parameters corresponding to historical good batches in advance, and performing nonlinear fusion mapping on the historical process parameters to generate a set of historical fusion feature vectors, and constructing a static production steady state domain for defining an ideal production state according to the overall distribution of the set of historical fusion feature vectors; real-time acquisition of a plurality of rock wool process parameters in the production process, nonlinear fusion mapping of all rock wool process parameters acquired in real time to generate a real-time fusion feature vector for representing the overall operation condition of the current production process, and constructing a dynamic production control domain for defining the current operating condition safety boundary; determining the real-time fusion feature vector as a real-time production state point in the low-dimensional latent space of the autoencoder neural network, and synchronously projecting the real-time production state point in the static production steady state domain and the dynamic production control domain to obtain the combined positional relationship of the real-time production state point with the static production steady state domain and the dynamic production control domain, and then determining the production state region of the real-time production state point according to the combined positional relationship; the production state region includes an optimal state region, a drift state region, a risk state region and a deterioration state region; when the real-time production state point is located in the drift state region, an optimization regression instruction is generated and executed; when the real-time production state point is located in the risk state region, an emergency avoidance instruction is generated and executed; when the real-time production state point is located in the deterioration state region, a forced regression instruction is generated and executed.

2. The method of claim 1, wherein, nonlinear fusion mapping of all rock wool process parameters acquired in real time to generate a real-time fusion feature vector includes: combining a plurality of rock wool process parameters acquired in real time into an input vector; the rock wool process parameters include furnace zone temperature, centrifuge speed, air blower pressure, sizing flow, curing oven temperature and conveyor belt speed; nonlinear compression of the input vector by using an encoder in the autoencoder neural network to generate a real-time fusion feature vector in the low-dimensional latent space of the autoencoder neural network.

3. The method of claim 2, wherein, constructing a static production steady state domain based on the overall distribution of a set of historical fusion feature vectors includes: determining a set of historical optimal state points as a core state point cloud in the low-dimensional latent space of the autoencoder neural network, and aggregating them into a core state point cloud; calculating the geometric center of the core state point cloud to determine the ideal process centroid representing the theoretical optimal operating condition as the geometric center; calculating a robustness measure based on the statistical distribution characteristics of all historical optimal state points in the core state point cloud relative to the ideal process centroid; in the low-dimensional latent space, centering on the ideal process centroid, and constructing a closed geometric body according to the robustness measure, determining the internal space region bounded by the closed geometric body as the static production steady state domain.

4. The method of claim 3, wherein, constructing a dynamic production control domain includes: predefining a maximum operating envelope in the low-dimensional latent space, which is mapped from the device physical limit and the process safety constraint; real-time acquisition of a set of process constraint factors representing the potential risks of the current production process; the process constraint factors include raw material batch fluctuation, equipment wear state or environmental parameter anomaly; inputting the set of process constraint factors into a pre-trained risk quantification model to output a set of boundary convergence vectors; a direction of the boundary convergence vector represents a direction of boundary contraction; a modulus of the boundary convergence vector represents a magnitude of boundary contraction; performing boundary contraction on the maximum operation envelope according to the set of boundary convergence vectors to generate a dynamic closed surface, and determining an internal space region bounded by the dynamic closed surface as a dynamic production control domain.

5. The method of claim 4, wherein, determining a production state region of the real-time production state point according to the combined position relationship includes: when the real-time production state point is simultaneously inside the static production steady state domain and the dynamic production control domain, determining that the real-time production state point is in an optimal state region; when the real-time production state point is inside the dynamic production control domain but outside the static production steady state domain, determining that the real-time production state point is in a drift state region; when the real-time production state point is inside the static production steady state domain but outside the dynamic production control domain, determining that the real-time production state point is in a risk state region; when the real-time production state point is simultaneously outside the static production steady state domain and the dynamic production control domain, determining that the real-time production state point is in a deterioration state region.

6. The method of claim 5, wherein, when the real-time production state point is in the drift state region, generating and executing an optimization regression instruction includes: in the low-dimensional latent space, determining an ideal process centroid of the static production steady state domain as an optimization regression target point; calculating an optimization regression vector pointing from the real-time production state point to the optimization regression target point; applying a decoder in the autoencoder neural network to reversely map the optimization regression vector into a set of process parameter adjustment values in the process parameter space; generating and executing an optimization regression instruction containing the process parameter adjustment values.

7. The method of claim 6, wherein, when the real-time production state point is in the risk state region, generating and executing an emergency avoidance instruction includes: in the low-dimensional latent space, determining a boundary point on a boundary of the dynamic production control domain that is closest to the real-time production state point as an emergency avoidance target point; calculating an emergency avoidance vector pointing from the real-time production state point to the emergency avoidance target point; applying a decoder in the autoencoder neural network to reversely map the emergency avoidance vector into a set of process parameter adjustment values in the process parameter space; generating and executing an emergency avoidance instruction containing the process parameter adjustment values.

8. The method of claim 7, wherein, when the real-time production state point is in the deterioration state region, generating and executing a forced regression instruction includes: in the low-dimensional latent space, determining a pre-set safe state reference point inside the dynamic production control domain as a forced regression target point; calculating a forced regression vector pointing from the real-time production state point to the forced regression target point; applying a decoder in the autoencoder neural network to reversely map the forced regression vector into a set of process parameter adjustment values in the process parameter space; generating and executing a forced regression instruction containing the process parameter adjustment values.

9. The method of claim 8, wherein, The optimization regression instruction aims to adjust the real-time production state point to the inside of the static production steady state domain; the emergency avoidance instruction aims to adjust the real-time production state point to the inside of the dynamic production control domain; and the forced regression instruction aims to adjust the real-time production state point to the inside of the dynamic production control domain.

10. A rock wool slab production control system characterized by, It comprises: a static domain construction module configured to extract historical process parameters corresponding to historical good batches in advance, perform nonlinear fusion mapping on the historical process parameters to generate a set of historical fusion feature vectors, and construct a static production steady state domain for defining ideal production states according to the overall distribution of the set of historical fusion feature vectors; a dynamic domain construction module configured to acquire a plurality of rock wool process parameters in a production process in real time, perform nonlinear fusion mapping on all the rock wool process parameters acquired in real time to generate a real-time fusion feature vector for representing the overall operation condition of the current production process, and construct a dynamic production control domain for defining the safe operation boundary of the current operation condition; a state determination module configured to determine a real-time production state point of the real-time fusion feature vector in a low-dimensional latent space of an autoencoder neural network, project and locate the real-time production state point in the static production steady state domain and the dynamic production control domain to obtain a combined positional relationship, and determine a production state region to which the real-time production state point belongs according to the combined positional relationship, the production state region comprising an optimal state region, a drift state region, a risk state region and a deterioration state region; a control instruction generation module configured to generate and execute an optimization regression instruction aiming to adjust the real-time production state point to the inside of the static production steady state domain when it is determined that the real-time production state point is located in the drift state region; generate and execute an emergency avoidance instruction aiming to adjust the real-time production state point to the inside of the dynamic production control domain when it is determined that the real-time production state point is located in the risk state region; and generate and execute a forced regression instruction aiming to adjust the real-time production state point to the inside of the dynamic production control domain when it is determined that the real-time production state point is located in the deterioration state region.

Images