Manufacturing execution system for prefabricated direct-buried thermal insulation pipe production line

Abstract

This invention relates to the field of industrial automation and manufacturing execution systems (MES), specifically a MES for a prefabricated direct-buried insulated pipe production line. It includes a data acquisition module, a state quantification module, a strategy decision-making module, and a process execution module. The data acquisition module collects state data from the underlying sensors of the production equipment and data on human intervention behaviors in the underlying control system. The state quantification module performs feature extraction and weighted calculation based on a preset risk assessment logic to derive a system risk state index. The strategy decision-making module compares this index with a danger threshold to generate a reinforcement learning-based extreme process strategy or a smooth degradation process strategy. The process execution module converts the strategy into control commands for the underlying equipment and issues them, achieving dynamic process scheduling and improving continuous operation stability.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and manufacturing execution systems, specifically to a manufacturing execution system for a prefabricated direct-buried insulated pipe production line. Background Technology

[0002] The manufacturing execution system of a prefabricated direct-buried insulated pipe production line refers to a production management system that coordinates and controls the traction, injection, and related processes based on the operating status of the production equipment and feedback from the underlying control system. Current production control methods for prefabricated direct-buried insulated pipes typically include control methods that rely on fixed process parameter tables, local closed-loop control methods based on the underlying control system, and on-site adjustment methods dominated by human experience.

[0003] However, when scheduling production based on existing technologies, on the one hand, there is a lack of coordinated analysis of equipment sensor status, human intervention behavior and environmental disturbance factors, making it difficult to identify the risk of the production line approaching instability or shutdown in a timely manner. On the other hand, in high-cycle continuous production scenarios, existing control methods generally lack a mechanism that can dynamically switch between high-yield operation and risk degradation, which can easily reduce the continuous operation stability of the entire line and the accuracy of process scheduling. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention provides a manufacturing execution system for a prefabricated direct-buried insulated pipe production line. Specifically, the technical solution of the present invention includes:

[0005] The prefabricated direct-buried insulated pipe production line includes production equipment and a bottom-level control system. The production equipment includes a fluid raw material injection device, a traction device, a raw material silo temperature sensor, a foaming gun back pressure detection unit, and a pipe body coaxiality detection unit. The manufacturing execution system includes:

[0006] The data acquisition module is communicatively connected to the production equipment and the underlying control system, and is used to collect the status data of the underlying sensors of the production equipment and the human intervention behavior data of the underlying control system.

[0007] The state quantification module is communicatively connected to the data acquisition module and is used to extract features and perform weighted calculations on the underlying sensor state data and the human intervention behavior data based on a preset risk assessment logic to obtain a system risk state index.

[0008] The strategy decision module, which is communicatively connected to the state quantization module, is used to compare the system risk state index with a preset danger threshold: when the system risk state index is lower than the preset danger threshold, a reinforcement learning extreme process strategy is generated; when the system risk state index is greater than or equal to the preset danger threshold, a smooth degradation process strategy is generated.

[0009] The process execution module is communicatively connected to the strategy decision module. It is used to receive the reinforcement learning extreme process strategy or the smooth degradation process strategy and convert it into underlying equipment control instructions. It also sends the underlying equipment control instructions to the underlying control system to perform dynamic process scheduling of the production equipment.

[0010] Optionally, the human intervention behavior data includes at least manual mode switching frequency data and underlying parameter modification rate data;

[0011] The state quantization module contains a pre-stored real-time state space model.

[0012] The state quantization module is used to input the underlying sensor state data, the manual mode switching frequency data, and the underlying parameter modification rate data as state variables into the real-time state space model for state mapping and deviation calculation, so as to obtain the system risk state index.

[0013] Optionally, the manufacturing execution system further includes a risk prediction module;

[0014] The risk prediction module is communicatively connected to the state quantification module; the risk prediction module is used to receive the system risk state index.

[0015] The risk prediction module is also used to compare the system risk status index with a preset failure benchmark index determined based on historical production statistics and calculate the ratio data as the risk contribution rate data of the production equipment reaching the preset shutdown failure boundary.

[0016] Optionally, the reinforcement learning extreme process strategy is generated based on a deep reinforcement learning algorithm;

[0017] The reinforcement learning extreme process strategy includes extreme traction speed commands and extreme foaming injection pressure commands.

[0018] The process execution module is used to control the traction device of the production equipment using the extreme traction speed command after receiving the reinforcement learning extreme process strategy, and to control the fluid raw material injection device of the production equipment using the extreme foaming injection pressure command.

[0019] Optionally, the smooth degradation process strategy includes a deceleration compensation instruction and a proportional-integral-derivative control logic instruction;

[0020] The process execution module is used to reduce the traction speed of the traction device in the production equipment by using the deceleration compensation command after receiving the smooth degradation process strategy, and to replace the limit traction speed command and the limit foaming injection pressure command by using the proportional-integral-derivative control logic command.

[0021] In the smooth degradation process strategy, the rate of change of process parameters is less than a preset rate of change threshold.

[0022] Optionally, the underlying sensor status data includes environmental vibration interference data, and raw material warehouse temperature data collected by the raw material warehouse temperature sensor where the temperature drop rate is greater than or equal to a preset temperature change rate threshold.

[0023] The data acquisition module is also used to acquire the initial viscosity data of the fluid raw material;

[0024] The state quantization module is also used to determine noise characteristics using the environmental vibration interference data and the initial viscosity data, and to remove data that conforms to the noise characteristics from the underlying sensor state data for data cleaning to filter out data noise.

[0025] Optionally, the preset shutdown failure boundary includes an eccentricity state boundary where the eccentricity is greater than or equal to a preset eccentricity threshold and a foam gun blockage state boundary.

[0026] The risk prediction module is also used to send a forced degradation signal to the strategy decision module when the risk contribution rate data is greater than or equal to a preset downtime risk threshold.

[0027] The strategy decision module is also used to generate the smooth degradation process strategy in response to the forced degradation signal.

[0028] Optionally, the data acquisition module is also used to continuously collect updated human intervention behavior data and updated underlying sensor status data after the underlying device control command is issued;

[0029] The state quantification module is also used to recalculate the current system risk state index based on the updated human intervention behavior data and the updated underlying sensor state data;

[0030] The strategy decision module is also used to regenerate the reinforcement learning extreme process strategy when the current system risk state index is lower than the preset danger threshold and the duration is greater than or equal to the preset recovery time threshold.

[0031] Optionally, the manufacturing execution system further includes an edge computing gateway and a data hub;

[0032] The data acquisition module, the state quantification module, the strategy decision-making module, and the process execution module are all integrated within the data hub.

[0033] The data hub is connected to the production equipment and the underlying control system via the edge computing gateway.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. This invention synchronously acquires underlying sensor status data and human intervention behavior data through a data acquisition module, and uses the real-time state space model in the state quantification module to perform state mapping and deviation calculation to obtain a system risk state index. This mechanism overcomes the shortcomings of existing technologies that rely solely on equipment sensors, and analyzes signs of human intervention and physical state in a coordinated manner. At the same time, by combining the risk prediction module to calculate the ratio of the difference between the system risk state index and the preset failure benchmark index, it can proactively quantify the degree to which production equipment approaches the shutdown failure boundary, significantly improving the timeliness of risk identification.

[0036] 2. This invention achieves a smooth switching between high-yield operation and risk degradation of the production line through a strategy decision-making module and a process execution module. When the system risk state index is lower than a preset danger threshold, the system executes a reinforcement learning extreme process strategy generated by a deep reinforcement learning algorithm to maximize capacity. When the index reaches or exceeds the danger threshold, the system automatically switches to a smooth degradation process strategy, adjusting process parameters smoothly at a rate lower than a preset change rate threshold. Furthermore, after the risk has fallen back and stabilized at a preset recovery time threshold, the system can automatically regenerate and execute the reinforcement learning extreme process strategy, forming a complete dynamic closed loop, effectively improving the accuracy of production scheduling and the continuous operation capability of the entire line.

[0037] 3. This invention addresses the complex environmental disturbances in industrial settings by collecting environmental vibration data and initial viscosity data of fluid raw materials. The state quantification module identifies noise characteristics and performs data cleaning, effectively filtering out data noise caused by environmental fluctuations and ensuring the authenticity of the underlying sensor state data. Simultaneously, when the risk contribution rate data is greater than or equal to a preset downtime risk threshold, the system can implement forced protection by sending a forced degradation signal. The architecture, which integrates data acquisition, state quantification, strategy decision-making, and process execution modules within a data hub and communicates with production equipment and the underlying control system through an edge computing gateway, ensures low-latency response and high-reliability execution under disturbed conditions. Attached Figure Description

[0038] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0040] like Figure 1As shown, the manufacturing execution system of the prefabricated direct-buried insulated pipe production line includes production equipment and a bottom control system. The production equipment includes a fluid raw material injection device, a traction device, a raw material silo temperature sensor, a foaming gun back pressure detection unit, and a pipe coaxiality detection unit. The fluid raw material injection device includes a metering pump, a mixing head, and an injection pressure sensor, a mixing head temperature sensor, and a foaming gun back pressure detection unit installed on it. The traction device includes a traction motor, a frequency converter, a traction speed feedback unit, and a traction motor current detection unit. The production line also includes a pipe coaxiality detection unit for detecting the eccentricity of the insulation layer.

[0041] The underlying control system is configured with underlying control parameters, which include at least heating delay, speed limit, pressure limit, and interlock delay; the manufacturing execution system includes:

[0042] The data acquisition module communicates with the production equipment and the underlying control system to collect the status data of the underlying sensors of the production equipment and the data on human intervention behavior of the underlying control system.

[0043] The state quantification module communicates with the data acquisition module and is used to extract features and perform weighted calculations on the underlying sensor state data and human intervention behavior data based on preset risk assessment logic to obtain the system risk state index.

[0044] The strategy decision module, which communicates with the state quantization module, is used to compare the system risk state index with a preset danger threshold: when the system risk state index is lower than the preset danger threshold, a reinforcement learning limit process strategy is generated; when the system risk state index is greater than or equal to the preset danger threshold, a smooth degradation process strategy is generated.

[0045] The process execution module communicates with the strategy decision module. It receives reinforcement learning extreme process strategies or smooth degradation process strategies and converts them into underlying equipment control commands. It also sends the underlying equipment control commands to the underlying control system to perform dynamic process scheduling of production equipment.

[0046] This embodiment provides a dynamic process scheduling mechanism for a prefabricated direct-buried insulated pipe production line. Specifically, this mechanism is applied to a continuous production line operating under high load conditions, producing a polyurethane insulation layer between the outer sheaths of multi-diameter steel pipes. The production line operates at a high cycle time, with the traction device continuously pulling the pipe forward, while the fluid raw material injection device injects a mixture of polyurethane foaming raw materials into the annular cavity according to the pipe diameter and wall thickness requirements. Since the polyurethane foaming, expansion, and curing processes are simultaneously affected by the activity of the raw materials, ambient temperature, equipment response speed, and human operating habits, simply relying on a fixed process schedule is no longer sufficient to balance output, quality, and continuous operation stability. Therefore, the manufacturing execution system implements dynamic process coordination at the upper level for the lower-level control system.

[0047] Specifically, the data acquisition module is preferably connected to the injection pressure sensor, mixing head temperature sensor, traction speed feedback unit, traction motor current detection unit, tube coaxiality detection unit, foaming gun back pressure detection unit, and the mode switching log and parameter rewriting log in the underlying control system; the status data of the underlying sensors here not only reflects whether the equipment is running, but also whether the process is in the physical stable zone.

[0048] For example, an increase in injection pressure usually corresponds to an increase in the flow resistance of the mixture, fluctuations in traction current may indicate changes in traction resistance or asynchronous upstream feeding, while eccentricity detection results directly correspond to an imbalance in the thickness distribution of the insulation layer; human intervention data reflects the degree of acceptance of the current scheduling results by on-site personnel, because frequent switching to manual mode and frequent modification of programmable logic controller parameters are usually not random events, but rather a proactive compensation behavior by operators in response to equipment abnormalities, process abnormalities, or unreliable scheduling; the status quantification module provides a unified interpretation of the above two types of data based on preset risk assessment logic;

[0049] In this embodiment, it does not simply determine whether a single sensor exceeds its limit, but simultaneously observes whether the equipment state is approaching physical instability and whether the human-machine collaborative state deviates from the preset safety threshold. For example, when the injection pressure rises less than the first preset pressure change threshold but is kept stable by manual intervention, the system can interpret it as a controllable fluctuation. However, if the injection pressure rises, the traction fluctuation increases, and the delay parameter is continuously modified on site, it indicates that the process is approaching the compound risk zone of foam gun blockage or eccentricity.

[0050] The resulting system risk state index represents the parameter margin between the current production line state and the safety boundary where extreme cycle control can continue to be implemented. After receiving the risk state index, the strategy decision module first compares it with the danger threshold and then selects different strategies. When the risk state is lower than the danger threshold, it means that the equipment and personnel are still within the acceptable range. At this time, reinforcement learning extreme process strategy can be adopted to increase the output per unit time to the target value and maintain the raw material quality fluctuation within the preset tolerance range.

[0051] When the risk level reaches or exceeds the danger threshold, the system no longer adheres to the maximum production capacity target, but instead switches to a smoothing and degrading process strategy. Here, smoothing emphasizes the continuity and understandability of process parameter changes, avoiding drastic impacts on the residence state of the mixture in the gun, the traction inertia of the tube, and the operational expectations of on-site operators. The process execution module converts the strategy decision results into underlying equipment control commands and issues them.

[0052] The conversion process may include: mapping the target traction speed given by the upper layer to the frequency setting value of the inverter, mapping the target injection pressure or feed ratio to the metering pump speed setting value, and mapping the switching logic to the mode register, enable bit and limiting parameters in the lower control system. In this way, the upper layer is only responsible for determining the process target, and the lower control system is responsible for the specific closed-loop execution, so as to reduce the architectural coupling and facilitate compatibility with existing production equipment.

[0053] Under abnormal operating conditions, if a critical sensor goes offline for a short period of time, such as an interruption in the eccentricity detection signal or a frozen value in the injection pressure feedback, the data acquisition module should mark that channel as unreliable data. The state quantification module should reduce its reliance on that channel in the current cycle to avoid unnecessary large-scale process adjustments triggered by a single distorted signal. If data on human intervention is missing, such as the underlying control system not transmitting logs, the system should at least calculate the risk state based on available equipment status data and limit the adjustment range of the extreme strategy. If the reinforcement learning strategy generates abnormal targets, such as exceeding the maximum allowable speed of the traction device or exceeding the rated mechanical pressure of the injection pump, the process execution module should perform boundary trimming before issuing the data to ensure equipment safety.

[0054] On a continuous production line for producing DN600 prefabricated direct-buried insulated pipes, to meet the high-cycle operation requirements in low-temperature environments, the system compresses the curing safety time margin within a preset limit range. After production begins, the data acquisition module continuously receives foam gun pressure, traction speed, motor load, and eccentricity detection values. The underlying control system is equipped with a programmable logic controller and a touch screen, among other human-machine interfaces, to record the mode switching and parameter adjustment actions of on-site operators. The status quantification module detects that although equipment fluctuations have increased, manual intervention is still below the preset intervention frequency threshold, thus determining that the system is still suitable for maintaining a high cycle. The strategy decision module then outputs the limit process strategy, which the process execution module converts into traction speed increase commands and injection pressure fine compensation commands.

[0055] As the ambient temperature drops below the preset temperature threshold, if the frequency of manual switching to manual mode increases significantly and the traction load also becomes unstable, the state quantification module will increase the system risk state index, and the strategy decision module will then switch to a smooth degradation process strategy, so that the production line can enter a continuous guarantee state from a high-load production state.

[0056] The purpose of this step or mechanism is to incorporate both the physical state of the equipment and the state of operator intervention into the scheduling criteria of the manufacturing execution layer, so that the system no longer optimizes only around the single objective of production capacity, but prioritizes maintaining the ability of the entire production line to operate without stopping under high-risk continuous production conditions.

[0057] In this embodiment, the human intervention behavior data includes at least manual mode switching frequency data and underlying parameter modification rate data;

[0058] The state quantization module contains a pre-stored real-time state space model;

[0059] The state quantization module is used to input the underlying sensor state data, manual mode switching frequency data, and underlying parameter modification rate data as state variables into the real-time state space model for state mapping and deviation calculation, so as to obtain the system risk state index.

[0060] This embodiment provides a state quantification mechanism for human-machine collaborative instability. Specifically, in the aforementioned high-cycle thermal insulation pipe production scenario, relying solely on equipment sensors cannot fully characterize the system state: most anomalies leading to downtime usually do not originate from physical mechanical failures, but rather from a decline in the operator's reliability judgment of the automatic control logic and the resulting frequent takeover control behaviors. Therefore, the state quantification module further introduces manual mode switching frequency data and underlying parameter modification rate data, and, in conjunction with a real-time state space model, provides a more accurate description of process risks compared to the actual situation on site.

[0061] Specifically, the frequency of manual mode switching reflects the operator's immediate acceptance of the automatic scheduling results. When the work team finds that the manufacturing execution system frequently issues process actions that do not meet experience expectations, they will usually switch to manual mode briefly to maintain the equipment operating parameters within the preset safety range. The modification rate of underlying parameters reflects whether the operator is continuously making implicit corrections to the system by modifying heating delay, speed limit, pressure limit, or interlock delay. After both of these are entered into the real-time state space model along with the underlying sensor status data, a state mapping result that takes into account both the physical process and the manual intervention process can be formed. The state space model here can be understood as placing injection stability, traction stability, molding stability, and personnel takeover tendency in the same observation coordinate system, thereby identifying those states that have not yet caused scrap but have already shown signs of system fragility.

[0062] In terms of specific logical flow and business decomposition, state mapping refers to normalizing heterogeneous physical data such as the fluctuation variance of injection pressure and traction speed within a specific time window and human behavior data such as the number of manual switching and the cumulative change rate of parameter modification items within a set time window, and projecting them into a multi-dimensional observation space to form a multi-dimensional state vector representing the current operating status.

[0063] Deviation calculation involves pre-defining an ideal reference origin or vector in the real-time state space model to represent absolutely stable production without human intervention. The degree of deviation is quantified by calculating the spatial distance, such as Euclidean or Mahalanobis distance, between the current multidimensional state vector and this ideal reference origin. The larger the calculated deviation distance, the more severe the instability tendency of the system at both the physical and human-machine collaboration levels. After being scaled or weighted by a preset ratio, this distance deviation is output as the aforementioned system risk state index.

[0064] In a simplified state mapping example, the state of a certain sampling period can be described as four types of state variables: injection stability S1, traction stability S2, manual switching activity S3, and parameter rewriting activity S4. If the volatility of S1 and S2 is within the preset tolerance range, while S3 and S4 remain at a low level for a long time, the model will map this state as controllable equipment disturbance, and the calculated spatial deviation will be small. If the volatility of S1 and S2 is lower than the preset volatility threshold, but the rate of change of S3 and S4 exceeds the preset growth threshold, the model will map this state as a human-machine collaboration tension zone, and the deviation calculation result will increase accordingly. If S1 and S2 both exceed the normal process tolerance range and the increase of S3 and S4 is also greater than the preset growth threshold, the deviation calculation result will show non-linear growth and be closer to the failure benchmark index. The resulting system risk state index does not rely on a single alarm, but reflects whether the equipment change is sufficient to induce on-site personnel to perform continuous reverse operations.

[0065] Under abnormal operating conditions, if a shift changes a short-term frequency of mode switching is high, but the underlying parameters are not continuously modified and the core process flow remains stable, the status quantification module can identify this segment as a management disturbance rather than a true instability, thus avoiding misjudgment. Conversely, if the mode switching log is not returned, but the parameter modification records continue to increase, and the equipment status also begins to fluctuate periodically, the system can still determine that the risk is rising. If both types of human behavior data are temporarily missing, the equipment-side mapping results are retained, but the upper limit of the system risk status index is raised to a conservative range to prevent insufficient risk assessment of implicit human intervention behaviors.

[0066] On the same high-load, continuously operating production line, after the night shift began, the outside temperature dropped, and the manufacturing execution system issued several fine-tuning instructions for traction that were inconsistent with conventional manual operation experience. At this time, although the injection pressure and traction speed had not yet exceeded the equipment alarm line, the operator had switched the system from automatic to manual multiple times within 10 minutes and modified the foam gun pre-rinse delay and traction acceleration limit. After mapping these behaviors together with the status of the equipment sensors, the status quantification module determined that the current situation was not a simple process fluctuation, but a risk-increasing stage in which the operation team was preparing to take full control of the equipment, thereby raising the system risk status index in advance.

[0067] The purpose of this step or mechanism is to transform the signs of human takeover that are usually hidden in the logs into quantifiable process risk factors, so that the manufacturing execution system can identify unstable trends at the human-machine collaboration level before a jam or serious deviation actually occurs.

[0068] In this embodiment, the manufacturing execution system also includes a risk prediction module;

[0069] The risk prediction module communicates with the state quantification module; the risk prediction module is used to receive the system risk state index.

[0070] The risk prediction module is also used to compare the system risk status index with the preset failure benchmark index determined based on historical production statistics and calculate the ratio data as the risk contribution rate data of the production equipment reaching the preset shutdown failure boundary.

[0071] This embodiment provides a downtime boundary look-ahead prediction mechanism. Specifically, in the aforementioned scheme, the system can already obtain the current system risk status index. However, for continuous insulation pipe production, simply knowing that the current system risk status index is high is insufficient to guide scheduling. This is because the on-site team is more concerned with the difference between the current system risk status index and the boundary at which the entire line must be shut down for cleanup, as well as whether continuing to maintain the current cycle time will cause irreversible failure in a short period of time. Therefore, a risk prediction module is added to further convert the system risk status index into risk contribution rate data.

[0072] Specifically, the failure baseline index corresponds to the pre-shutdown hazardous reference state determined after long-term production statistics and process verification. This reference state is not equivalent to a single alarm value, but is closer to a comprehensive baseline that, when combined with continuous eccentricity, foam gun blockage, traction loss of synchronization, and continuous manual intervention, will significantly increase the probability of the entire line shutdown. The risk prediction module compares the current system risk state index with the failure baseline index and outputs risk contribution rate data to indicate the extent to which the current state is pushing the production line closer to the shutdown boundary.

[0073] Furthermore, the data flow logic for calculating the ratio data is clearly defined: the calculation is not a simple comparison without physical meaning, but rather a quantification of the proportion of the current system risk state index consumed by the downtime tolerance; specifically, the risk prediction module obtains the failure benchmark index representing the downtime threshold, calculates the proportion of the current system risk state index to the failure benchmark index, or calculates the percentage of the current system risk state index to the failure benchmark index.

[0074] For example, the logical formula risk contribution rate = (current system risk status index / failure baseline index) × 100% can be used. When the system risk status index continues to rise with equipment fluctuations and frequent manual intervention, this ratio directly projects the degree to which the current production line is approaching shutdown failure. If the rate of change of the difference ratio is greater than the preset rate threshold and approaches a certain high-risk critical percentage such as 85%, it means that the abnormal risks accumulated in the system have made a significant negative contribution to the overall availability of the machine, and are no longer local disturbances, thus more accurately characterizing the depth of risk evolution.

[0075] In a simplified state simulation example, the failure baseline can be regarded as reference state B, and the current state as observation state A. If A is only slightly close to B, it indicates that the current anomaly is more of a recoverable fluctuation, the calculated difference ratio is low, and the advancement to the shutdown boundary is limited. If the degree of closeness between A and B is significantly increased, it indicates that the current combination of injection resistance, traction instability and manual intervention has a high degree of failure orientation, and the difference ratio is significantly increased. Therefore, the risk contribution rate data is more suitable as an early warning basis for the scheduling layer, because it reflects not a point-like anomaly, but how much consumable margin is left before the entire line is shut down.

[0076] Under abnormal operating conditions, if the production line is in the process of changing specifications, switching molds, or trial operation, some state variables may initially deviate from the steady state. In this case, the risk prediction module should call the failure baseline of the corresponding specification or operating condition to avoid misjudging normal switching conditions as a sign of impending shutdown. If the failure baseline has not yet established a complete statistical sample due to the launch of new products, a conservative transition baseline can be used first, and the authorization scope of the extreme strategy can be limited. If the current system risk state index fluctuates briefly but does not form a continuous trend, the risk prediction module can use a time window to confirm before updating the risk contribution rate, so as to reduce the amplification of the decision chain by occasional disturbances.

[0077] During the continuous production of the heating pipe network fittings, the manufacturing execution system monitored that the system risk status index continued to rise over two hours at night. Although it had not yet triggered the interlocking shutdown of the underlying equipment, the risk prediction module compared it with the historical failure benchmark of this specification of pipe and calculated the difference ratio. It determined that its role in advancing the 48-hour furnace cleaning shutdown boundary had significantly increased. For example, the risk contribution rate had been calculated to reach 70% and was rising rapidly. At this time, the system no longer interpreted the anomaly as a normal fluctuation, but rather as a situation where, if it continued to operate at the current extreme pace, the probability of a serious eccentricity or nozzle blockage failure in the subsequent preset number of products would be greater than the preset alarm threshold.

[0078] The purpose of this step or mechanism is to further transform the current system risk status index into a forward-looking indicator oriented towards the shutdown boundary, so that the upper-level dispatcher can not only see the current anomaly, but also the degree to which the anomaly erodes the continuous operation capability of the entire line.

[0079] In this embodiment, the reinforcement learning extreme process strategy is generated based on a deep reinforcement learning algorithm;

[0080] The reinforcement learning-based extreme process strategy includes extreme traction speed commands and extreme foaming injection pressure commands.

[0081] The process execution module is used to control the traction device of the production equipment by using the limit traction speed command after receiving the reinforcement learning limit process strategy, and to control the fluid raw material injection device of the production equipment by using the limit foaming injection pressure command.

[0082] This embodiment provides an extreme process execution mechanism suitable for high-capacity load operation phases; specifically, in the aforementioned main scenario, when the system risk state is still below the danger threshold, the manufacturing execution system may not be satisfied with the conventional conservative process, but instead call the extreme process strategy generated based on the deep reinforcement learning algorithm to implement linkage control for the traction device and the fluid raw material injection device, so as to compress unnecessary process redundancy and improve the output per unit time.

[0083] In summary, the core contradiction in the continuous production of thermal insulation pipes lies in the following: too low a traction speed will reduce output and cause insufficient utilization of the curing station, while too high a traction speed will shorten the reaction margin of foaming and curing; too low an injection pressure will cause insufficient filling or local voids, while too high an injection pressure will increase the risk of eccentricity and gun blockage; the deep reinforcement learning strategy does not simply fix an optimal value, but dynamically generates limit traction speed commands and limit foaming injection pressure commands within the existing equipment capacity boundary, based on the current pipe diameter, ambient temperature, raw material reactivity and real-time molding feedback.

[0084] Its technical significance lies in exploring process combinations that are closer to the upper limit of equipment within a controllable risk range; in order to avoid the algorithm getting stuck in an unverifiable non-transparent mapping model, this embodiment deconstructs the core components of the deep reinforcement learning algorithm at the business level; considering that traction speed and injection pressure are both continuously adjustable variables, the algorithm preferably adopts a deep deterministic policy gradient or soft actor-commenter network architecture based on a continuous action space.

[0085] First, regarding state-space input: the algorithm extracts pipe diameter specifications, real-time ambient temperature or raw material silo temperature, initial viscosity data returned by bottom-level sensors, and the current system risk state index as environmental state characteristics. Second, regarding action-space output: the algorithm directly outputs the dynamic increment or absolute setpoint of the traction speed, as well as the synchronously coordinated injection pressure adjustment target value, and both of these outputs are rigidly limited within the safe range of the equipment's mechanical rated threshold. Third, regarding the design logic of the reward function: the reward value consists of a positive incentive term for production capacity and a negative penalty term for risk; the highest positive reward is given when the model increases the traction speed in a low-risk state, thereby encouraging the model to maximize the utilization of the equipment's operating margin.

[0086] However, if the model increases the traction speed without constraints or applies mismatched injection pressure, causing the risk state index to approach the preset danger threshold, a step-wise negative penalty with preset weights will be given. Through multiple rounds of state transition and evaluation-based iterations in historical production data and offline simulation environment, the strategy network eventually learns to dynamically calculate the best combination of linkage strategies that balances the filling front and reaction rate under specific ambient temperature and raw material viscosity.

[0087] In a simplified process state simulation example, the current system operating condition can be divided into three discrete states: stable filling zone T1, compact cycle zone T2, and critical approach zone T3. If the state falls into T1, the strategy can appropriately increase the traction speed and simultaneously increase the injection pressure to ensure that cavity filling and traction are synchronized. If the state falls into T2, the strategy may maintain a high traction speed but only fine-tune the pressure to prevent raw material reaction lag. If the state is close to T3, although the overall risk has not yet crossed the danger threshold, the strategy will also reduce large fluctuations and prioritize maintaining continuity. After receiving these targets, the process execution module maps the limit traction speed command to the traction motor speed setting and the limit foaming injection pressure command to the metering pump speed and pressure target, thereby forming the actual equipment actions. Unlike conventional static processes, the limit strategy here emphasizes linkage.

[0088] For example, when the reaction rate slows down due to low temperature, the system may not only execute a single deceleration command, but may also moderately increase the injection pressure and fine-tune the traction rhythm within the allowable range to maintain the relative position of the filling front and the curing front in the tube. The physical basis for doing so is that polyurethane foaming is a time-varying process that is significantly affected by temperature and mixing state. Traction and injection must be coordinated, otherwise it is easy to have insufficient filling at the head, excessive expansion at the tail, or local eccentricity.

[0089] Under abnormal operating conditions, if the extreme speed or pressure output by the deep reinforcement learning strategy approaches the rated upper limit of the equipment, the process execution module should superimpose mechanical limiting, pressure ramp-up slope limitation, and interlock checks to prevent the underlying impact from the excessive intensity of the upper target. If the current product specification is launched for the first time and the training samples are insufficient, the system is allowed to revert to the experience process package or semi-automatic optimization mode for that specification, instead of forcibly using the extreme strategy. If any actuator responds with significant lag, such as a persistent deviation between the traction motor feedback and the target, the process execution module should pause further amplification of the extreme control intensity to avoid multiple actuators losing synchronization and process parameter misalignment due to inconsistent execution.

[0090] During this high-load continuous production operation, the daytime environment is relatively stable, the raw material warehouse temperature is controllable, and the system risk status index remains at a low level. The manufacturing execution system then calls the extreme process strategy, giving different high-cycle traction targets and matching injection pressure compensation targets to the DN400 and DN600 mixed production line sections respectively. After the process execution module sends these targets to the traction device and metering pump, the entire line improves the continuous output capacity without changing the original equipment structure, while maintaining the basic uniformity of the foam layer.

[0091] The purpose of this step or mechanism is to fully utilize the potential capabilities of equipment and processes during the risk-controlled phase, and to achieve high-cycle production through the synergistic optimization of traction and injection, rather than simply increasing a single parameter and having other stages passively catch up.

[0092] In this embodiment, the smooth degradation process strategy includes a deceleration compensation instruction and a proportional-integral-derivative control logic instruction;

[0093] The process execution module is used to reduce the traction speed of the traction device in the production equipment by using the deceleration compensation command after receiving the smooth degradation process strategy, and to replace the limit traction speed command and the limit foaming injection pressure command with the proportional-integral-derivative control logic command.

[0094] In the smooth degradation process strategy, the rate of change of process parameters is less than the preset rate of change threshold.

[0095] This embodiment provides a smooth degradation mechanism for the edge of instability. Specifically, the aforementioned extreme process strategy is suitable for the stage where the risk is under control. However, when the ambient temperature drops sharply, the physical properties of the material fluctuate, or the frequency of manual intervention exceeds the threshold, continuing to adhere to the extreme strategy will make the process actions too violent on site, which may further stimulate operator intervention. Therefore, when the system risk state index reaches or approaches the danger threshold, the system actively abandons the extreme process solution and switches to a smooth degradation process strategy that includes deceleration compensation instructions and proportional-integral-derivative control logic instructions.

[0096] In detail, the essence of the deceleration compensation command is to give polyurethane more time to mix, expand and initially cure in the annular cavity. After the traction speed is reduced, the reaction window between the injection of raw materials and the stabilization of the molding is lengthened, and the problem of uneven filling caused by local viscosity increase is also easier to mitigate. The proportional-integral-derivative control logic replaces the more aggressive linkage control in the limit strategy and adopts a continuous adjustment method that is easier to understand on site. Its significance is not to pursue the highest output, but to make key process quantities such as injection pressure and traction speed change at a slope lower than the preset change rate upper limit, thereby taking into account the mechanical inertia of the equipment, the lag in the reaction of raw materials and the stability of operator cognition.

[0097] In a simplified control state simulation example, the degradation process can be viewed as switching from the high-dynamic control zone D1 to the gradual control zone D2. In D1, the traction and injection targets may be continuously fine-tuned within a short period. After entering D2, the system first issues a deceleration compensation command to move the traction target down in stages, while simultaneously using proportional-integral-derivative control logic commands to take over the injection pressure and traction holding logic. The process parameter change rate mentioned here is less than the preset change rate threshold, which means that each adjustment must meet the gradual change requirements of the equipment and process to avoid generating violent oscillation commands with opposite acceleration directions and amplitudes greater than the set limits in adjacent control cycles.

[0098] From an industrial mechanism perspective, the smooth degradation process strategy differs from a single emergency braking deceleration control. Instead, it brings continuous production back from the critical instability zone to an observable, understandable, and sustainable region. For foaming guns, rapid pressure changes can easily cause local stagnation or impacts in the mixing head and conveying pipe section. For traction systems, rapid speed changes can alter the stress conditions of the insulation layer before it is fully stabilized. For personnel, a smooth transition in cycle time and pressure is closer to traditional experience-based control methods, making it easier to restore trust in the system.

[0099] Under abnormal operating conditions, if the system risk status index suddenly spikes and there are clear signs of impending jamming, such as a continuous and rapid increase in gun back pressure, the overall strategy remains to smooth downgrade, but stronger deceleration or injection suspension may be prioritized within the safety interlock range to avoid substantial equipment damage. If a certain feedback quantity required by the proportional-integral-derivative control logic is temporarily missing, the system can switch to a limiting mode to maintain the current safety target instead of continuing to use the extreme strategy. If the system risk status index does not decrease after downgrading, the manufacturing execution system can further limit the production line cycle time until the minimum stability requirements for continuous operation are met.

[0100] On the same heating network order line, after the cold wave worsened at night, the raw material response was significantly delayed. The high-cycle dynamic parameters previously given by the manufacturing execution system began to cause the actual response trajectory of the equipment to deviate from the control expectation. At this time, the system detected an increase in the frequency of manual switching to manual mode, so it proactively abandoned the extreme traction speed command and extreme injection pressure command, and instead used the deceleration compensation command to gradually reduce the speed of the entire line to a more stable range, and used proportional-integral-derivative control logic to maintain the stability of foam gun pressure and traction. Since the parameter changes were not abrupt but gradual, they were easier for on-site operators to accept, and the probability of manual reverse parameter changes also decreased.

[0101] The purpose of this step or mechanism is to adjust in a continuous, interpretable, and process-inertia-compliant manner when a downgrade is necessary, so as to avoid new equipment shocks and human intervention caused by abrupt changes in control style, thereby improving the continuous operation capability of the entire line during abnormal phases.

[0102] In this embodiment, the underlying sensor status data includes environmental vibration interference data, and raw material warehouse temperature data collected by the raw material warehouse temperature sensor when the temperature drop rate is greater than or equal to a preset temperature change rate threshold.

[0103] The data acquisition module is also used to acquire the initial viscosity data of the fluid raw materials;

[0104] The state quantization module is also used to determine noise characteristics using environmental vibration interference data and initial viscosity data, and to remove data that matches the noise characteristics from the underlying sensor state data for data cleaning to filter out data noise.

[0105] This embodiment provides a noise identification and data cleaning mechanism for environmental-level oscillations. Specifically, in the aforementioned scenario, extreme cold weather does not necessarily directly cause equipment shutdown, but it can alter the flowability of raw materials and the stability of equipment measurements through micro-frequency fluctuations in the power grid and sudden drops in raw material warehouse temperature. If all readings from underlying sensors are still considered equally reliable, the manufacturing execution system may mistakenly interpret measurement noise induced by environmental oscillations as actual process deviations and issue overly drastic adjustment commands. Therefore, it is necessary to incorporate environmental oscillation interference data and initial viscosity data of raw materials into the judgment, clean the data first, and then quantify the risk.

[0106] Specifically, fluctuations in power grid frequency can cause slight response drift in metering pumps, frequency converters, and some analog acquisition modules. This drift may not necessarily indicate distortion in the process itself, but it will change the short-term pattern of feedback data such as pressure, flow rate, and speed. A drop in raw material warehouse temperature will more directly affect the initial viscosity of the foaming raw material, causing changes in injection resistance and mixing characteristics. If, in the early stages of the increase in raw material viscosity, the sensor readings show a high-frequency oscillation pattern with a monotonically increasing baseline, and this pattern occurs simultaneously with a sudden drop in warehouse temperature and an increase in initial viscosity, then some of these fluctuations are manifestations of amplified environmental noise, rather than all of them indicating equipment failure.

[0107] In a simplified data cleaning and simulation example, data for a certain period can be divided into three segments: P1 represents stable period data, P2 represents power grid fluctuation period data, and P3 represents rapid temperature drop period data. If the pressure feedback shows regular, small oscillations in P2, but the traction and molding quality do not deteriorate synchronously, then these oscillations can be identified as power grid-related noise characteristics. If in P3, the initial viscosity increases while the average pressure slightly increases, but the eccentricity does not increase significantly, then the peak data exceeding the rate of change in the process physics can be removed. After cleaning the data accordingly, the state quantification module sends data that better represents the actual equipment state into the risk assessment logic, thereby reducing misadjustments. From an industrial mechanism perspective, data cleaning is not about ignoring adverse signals, but about distinguishing between actual process changes and environmentally amplified measurement disturbances.

[0108] For example, a sudden drop in warehouse temperature can indeed cause raw materials to become more viscous, and the increase in injection pressure is a real phenomenon that the system should not filter out entirely. However, if periodic non-smooth fluctuations that do not conform to the physical inertial response model of the pump and pipeline occur in the same stage, then this part is more likely to be sampling or power fluctuation noise and should be eliminated. This is to avoid the upper-level scheduling from frequently correcting around noise interference, which may create new process instability.

[0109] Under abnormal operating conditions, if environmental oscillation data is missing, such as a temporary interruption of the power grid frequency interface, the system can still perform basic cleaning based on the initial viscosity of the raw material and the consistency of multiple sensors. However, the filtering intensity should be reduced to avoid excessive rejection of real anomalies. If the initial viscosity acquisition fails, the warehouse temperature and the response characteristics of the same batch of raw materials in history can be used as alternative criteria. If there are too few key data points after cleaning, the state quantization module should exit the high-sensitivity adjustment mode and maintain only conservative control to prevent over-judgment based on a small amount of residual data.

[0110] During continuous nighttime production on this line, a cold wave caused a significant drop in the temperature of the raw material warehouse within a short period of time, while the power grid in the plant area experienced subtle fluctuations. The data acquisition module recorded that the injection pressure curve began to show fine spikes, but the eccentricity detection did not deteriorate simultaneously. Combining this with the fact that the initial viscosity of the raw material increased, the system determined that some of the spikes were noise characteristics induced by environmental vibrations. Therefore, it first cleaned the data and then assessed the risk based on the effective data after cleaning. In this way, the manufacturing execution system will not continuously issue conflicting traction and injection adjustment commands due to short-term abnormal noise data.

[0111] The purpose of this step or mechanism is to improve the availability and reliability of underlying data when environmental disturbances occur, so that subsequent risk assessments and process decisions are based on data that is closer to the real physical process.

[0112] In this embodiment, the preset shutdown failure boundary includes the eccentricity state boundary where the eccentricity is greater than or equal to the preset eccentricity threshold and the foam gun blockage state boundary.

[0113] The risk prediction module is also used to send a forced degradation signal to the strategy decision module when the risk contribution rate data is greater than or equal to the preset downtime risk threshold.

[0114] The strategy decision module is also used to generate a smooth degradation process strategy in response to a forced degradation signal.

[0115] This embodiment provides a forced degradation mechanism for clearly defined failure consequences. Specifically, the aforementioned risk prediction module can output the risk contribution rate, but if it only provides a warning, the system may still knowingly increase production despite knowing the risk is too high. For prefabricated direct-buried insulated pipe production lines, the two types of shutdown failure boundaries that have the greatest impact on the continuous operation of the production line are usually severe eccentricity and foam gun blockage. The former will cause the entire batch of products to be scrapped and reduce the accuracy of subsequent assembly, while the latter may directly force the entire line to be shut down for cleaning. Therefore, this embodiment uses these two types of boundaries as the triggering basis for forced degradation.

[0116] Specifically, the boundary of the eccentric state can be determined by an online eccentricity detection device or a multi-point wall thickness detection device. Its technical meaning is that the insulation layer thickness distribution has deviated from the design allowable range, and continuing to maintain the current process will increase the probability of subsequent continuous scrap. The boundary of the foam gun blockage state can be comprehensively characterized by phenomena such as a continuous abnormal increase in gun back pressure, an imbalance between flow and pressure, or a significant lag in injection response. Once the risk prediction module determines that the current risk contribution rate has reached or exceeded the shutdown risk threshold, it no longer regards the state as a stage where production can be further increased through optimization. Instead, it sends a forced degradation signal to the strategy decision module, requiring it to directly generate a smooth degradation process strategy.

[0117] In a simplified boundary simulation example, two boundary markers can be set: E1 represents the eccentric boundary, and E2 represents the bubble gun blockage state boundary. If the system finds that although the current state has not actually touched E1 or E2, the risk contribution rate indicates that it is very likely to approach one of them in the following several cycles, then the forced downgrade signal is set to effective. After receiving this signal, the strategy decision module no longer compares the difference in returns between the extreme strategy and the downgrade strategy, but directly locks the downgrade path. This design avoids high-frequency oscillations near the boundary and reduces unstable operations such as extreme production scheduling in the previous cycle and emergency intervention in the next cycle.

[0118] From a process mechanism perspective, approaching the severe eccentricity boundary indicates that the filling front, traction cycle, and tube posture during the foaming layer formation process have deviated from the preset matching parameter range; approaching the foaming gun blockage boundary indicates that the residence, reaction, or flow resistance change of the mixture in the gun and pipeline has exceeded the safety margin; once these two types of boundaries are truly triggered, the process adjustment cycle required for the system to recover to a stable state is relatively long. Therefore, the system should actively release some process margin before the boundary to maintain the continuous operation of the production line.

[0119] In abnormal operating conditions, if the eccentricity detection device gives a short-term false alarm, but the foam gun status and other product molding indicators are normal, the forced downgrade signal can first enter the short confirmation window to avoid frequent strategy switching due to a single false detection. If the sensor related to the foam gun blockage status boundary conflicts with the conclusion of manual inspection, the system can adopt the principle of strictness over leniency to downgrade first, and then wait for further confirmation. If the risk continues to rise after forced downgrade, it should continue to retreat to a more conservative process level, and if necessary, issue a plan adjustment prompt to the upper-level production management system.

[0120] Under the aforementioned night shift cold wave conditions, the eccentricity of several pipes began to approach the quality control line, and at the same time, the back pressure of the foaming gun increased simultaneously with the increase in the viscosity of the raw materials. Based on this, the risk prediction module judged that if the extreme cycle time continued, severe eccentricity or gun blockage was likely to occur, which would directly push the entire line to the boundary of long-term shutdown and furnace cleaning. Therefore, the system immediately issued a forced degradation signal, and the strategy decision module no longer tried to maintain the extreme strategy, but directly switched to the smooth degradation process package.

[0121] The purpose of this step or mechanism is to set up insurmountable downtime boundary protection logic for the manufacturing execution system, so that it can automatically stop aggressive control tendencies when it approaches a high-cost failure point, and prioritize the protection of the continuous operation capability of the entire line and the recoverability of the equipment.

[0122] In this embodiment, the data acquisition module is also used to continuously collect updated human intervention behavior data and updated underlying sensor status data after the underlying device control command is issued.

[0123] The state quantification module is also used to recalculate the current system risk state index based on updated human intervention behavior data and updated underlying sensor state data;

[0124] The strategy decision module is also used to regenerate the reinforcement learning extreme process strategy when the current system risk state index is lower than the preset danger threshold and the duration is greater than or equal to the preset recovery time threshold.

[0125] This embodiment provides a recovery and return mechanism after an anomaly. Specifically, in the aforementioned production scenario, smooth degradation is not the ultimate goal, but rather to gain a recovery window for the production line. If the system can only degrade but cannot determine when it is safe to return to high-cycle operation, it will remain in an inefficient state for a long time, making it difficult to meet the scheduling requirements of continuous production. Therefore, after the control command is issued, the data acquisition module continues to track the updated human intervention behavior data and the underlying sensor status data, the state quantification module recalculates the current system risk state index, and the strategy decision module regenerates the reinforcement learning extreme process strategy after the recovery conditions are met.

[0126] In detail, the recovery judgment must simultaneously meet two dimensions: first, the equipment must return to the stable process zone, such as the convergence of injection pressure fluctuations, smooth traction load, and recovery of eccentricity indicators; second, personnel intervention must be significantly reduced, such as operators no longer frequently switching to manual mode or frequently modifying programmable logic controller parameters. This is because a stable equipment curve does not mean that the extreme cycle time can be immediately resumed. If the frequency of operator intervention is still higher than the preset threshold, directly resuming the reinforcement learning extreme process strategy may easily lead to the system control logic becoming unstable again. Therefore, a recovery time threshold needs to be set. Only when the current system risk state index remains below the danger threshold for a certain period of time is the reinforcement learning extreme process strategy allowed to be reactivated.

[0127] In a simplified recovery simulation example, the observation process after degradation can be divided into three consecutive stages: R1, R2, and R3. In stage R1, equipment fluctuations converge, but human intervention remains excessive. At this stage, only the risk index is updated, and the extreme strategy is not restored. In stage R2, equipment stabilizes and human intervention decreases. If the duration is still short, degradation operation continues. In stage R3, equipment stabilizes, human intervention is low, and the recovery time threshold is reached. In this stage, the strategy decision module regenerates the extreme process strategy and gradually takes over with limited amplitude, rather than instantly restoring full-intensity control. The significance of the recovery process is that the system must not only maintain the current state without failure but also confirm that sustainable high-cycle operating conditions have been re-established. The recovery time threshold acts as an observation buffer to confirm whether raw materials, equipment, environment, and team coordination have been rebalanced. This avoids premature acceleration when the anomaly has just eased, which could lead to repeated risks.

[0128] In abnormal operating conditions, if the equipment status is stable after downgrading but manual intervention data is still abnormally missing, the system can extend the recovery observation time instead of directly restoring the extreme strategy; if the risk index rises again during the recovery process, the recovery judgment for this round is cancelled and the downgraded process continues to run; if multiple recovery attempts are quickly downgraded again, the system can automatically reduce the limit of the extreme strategy authorization for this shift or this environmental condition to prevent frequent switching back and forth.

[0129] After the ambient temperature rises, the raw material warehouse temperature gradually increases. After a period of deceleration and maintenance by proportional-integral-derivative control logic, the injection pressure curve becomes smooth again, the traction load fluctuation is lower than the preset load fluctuation threshold, and the operator no longer frequently enters manual mode. The data acquisition module continuously collects these updated information, and the status quantification module recalculates the current risk status index, finding that it remains below the danger threshold and exceeds the preset recovery time. Subsequently, the strategy decision-making module regenerates the reinforcement learning extreme process strategy and gradually increases the cycle time, enabling the production line to return to a high-production state with acceptable risk.

[0130] The purpose of this step or mechanism is to establish a complete closed loop for the system that can recover after degradation, so that the manufacturing execution system can proactively adjust in the event of an anomaly and return to high-efficiency operation with a basis after the conditions are restored, rather than staying at a conservative process level for a long time.

[0131] In this embodiment, the manufacturing execution system further includes an edge computing gateway and a data hub;

[0132] The data acquisition module, status quantification module, strategy decision-making module, and process execution module are all integrated within the data hub;

[0133] The data hub communicates with production equipment and the underlying control system through an edge computing gateway.

[0134] This embodiment provides a data hub and edge computing architecture for industrial field deployment. Specifically, if the aforementioned modules are distributed across different industrial control computers, servers, or cloud nodes, although the functionality can be achieved, network latency, inconsistent protocol conversion, or single-point communication interruptions may lead to scheduling delays in extremely cold environments, continuous high-cycle operation, and strong interference conditions. Therefore, integrating the data acquisition module, status quantification module, strategy decision-making module, and process execution module into the data hub and communicating with production equipment and the underlying control system through an edge computing gateway can improve the system's field response capability and deployment stability.

[0135] Specifically, the edge computing gateway is located between the device layer and the manufacturing execution layer. On the one hand, it is responsible for interfacing with heterogeneous interfaces such as programmable logic controllers, frequency converters, sensor buses, metering pump controllers, and eccentricity detection equipment. On the other hand, it is responsible for sending real-time data to the data hub in a unified format. The data hub is responsible for real-time computing and policy distribution tasks. The advantage of this design is that the underlying devices can retain their original control structure without large-scale modifications, while the upper-level risk assessment, anomaly degradation, and recovery logic are centrally operated in the same hub, reducing timing deviations caused by cross-node transmission.

[0136] In a simplified architecture simulation example, the system can be divided into three layers: L1 device layer, L2 gateway layer, and L3 central layer. L1 collects injection pressure, traction speed, chamber temperature, viscosity, and manual logs; L2 completes protocol adaptation and timestamp alignment; L3 internally performs state quantization, risk prediction, strategy decision-making, and process execution mapping sequentially. In this way, a complete control chain can be closed within the same central layer, avoiding the accumulation of delays caused by data collection in one place, evaluation in one place, and decision-making in another. For processes like insulated pipes that are sensitive to cycle continuity, time consistency is particularly important. From an engineering implementation perspective, the edge computing gateway can also serve as an on-site anomaly isolation function.

[0137] For example, if a sensor bus experiences a brief jitter, the gateway can first complete retry, caching, and basic verification before sending qualified data to the central hub. If the external network connection with the upper-level information system is interrupted, the central hub can still maintain a real-time control closed loop locally, without affecting the production line's continued execution of the established strategy. After integrating various modules within the data central hub, it is also convenient to uniformly maintain model versions, process parameter boundaries, and log traceability records. Under abnormal operating conditions, if the edge computing gateway loses communication with a device, the central hub can enter a degraded monitoring mode based on the remaining valid signals and limit the extreme control output. If the data central hub restarts or updates, the edge computing gateway can temporarily maintain the latest version of safe process parameters to ensure that the underlying production equipment does not lose control due to a brief pause in the upper-level system. If multiple production lines are running concurrently on-site, the central hub can isolate tasks according to the production line number to prevent abnormal data from one line from polluting the decision-making of other production lines.

[0138] During the continuous operation of the insulation pipes in this heating project, the data hub is installed in the cabinet near the workshop, and the edge computing gateway is connected to the foam gun controller, traction frequency converter, raw material warehouse temperature sensor, online eccentricity detector and programmable logic controller operation log interface respectively; after the cold wave arrives, even if the upper network in the plant is briefly congested, the risk assessment, smooth degradation and recovery judgment can still be completed locally, ensuring that the control chain of the manufacturing execution system for this critical production line is not broken;

[0139] The purpose of this step or mechanism is to reduce communication uncertainty in the industrial field through an architecture of edge near-end computing and centralized decision-making, so that the aforementioned risk identification, strategy switching and process execution capabilities can be stably implemented in a real continuous production environment.

[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A manufacturing execution system for a prefabricated direct-buried insulated pipe production line, characterized in that, The prefabricated direct-buried insulated pipe production line includes production equipment and a bottom control system. The production equipment includes a fluid raw material injection device, a traction device, a raw material silo temperature sensor, a foaming gun back pressure detection unit, and a pipe body coaxiality detection unit. The manufacturing execution system includes: The data acquisition module is communicatively connected to the production equipment and the underlying control system, and is used to collect the status data of the underlying sensors of the production equipment and the human intervention behavior data of the underlying control system. The state quantification module is communicatively connected to the data acquisition module and is used to extract features and perform weighted calculations on the underlying sensor state data and the human intervention behavior data based on a preset risk assessment logic to obtain a system risk state index. The strategy decision module, which is communicatively connected to the state quantization module, is used to compare the system risk state index with a preset danger threshold: when the system risk state index is lower than the preset danger threshold, a reinforcement learning extreme process strategy is generated; when the system risk state index is greater than or equal to the preset danger threshold, a smooth degradation process strategy is generated. The process execution module is communicatively connected to the strategy decision module. It is used to receive the reinforcement learning extreme process strategy or the smooth degradation process strategy and convert it into underlying equipment control instructions. It also sends the underlying equipment control instructions to the underlying control system to perform dynamic process scheduling of the production equipment.

2. The manufacturing execution system for the prefabricated direct-buried insulated pipe production line as described in claim 1, characterized in that, The data on human intervention behavior includes at least manual mode switching frequency data and underlying parameter modification rate data; The state quantization module contains a pre-stored real-time state space model. The state quantization module is used to input the underlying sensor state data, the manual mode switching frequency data, and the underlying parameter modification rate data as state variables into the real-time state space model for state mapping and deviation calculation, so as to obtain the system risk state index.

3. The manufacturing execution system for the prefabricated direct-buried insulated pipe production line as described in claim 2, characterized in that, The manufacturing execution system also includes a risk prediction module; The risk prediction module is communicatively connected to the state quantification module; the risk prediction module is used to receive the system risk state index. The risk prediction module is also used to compare the system risk status index with a preset failure benchmark index determined based on historical production statistics and calculate the ratio data as the risk contribution rate data of the production equipment reaching the preset shutdown failure boundary.

4. The manufacturing execution system for the prefabricated direct-buried insulated pipe production line as described in claim 1, characterized in that, The reinforcement learning-based extreme process strategy is generated based on a deep reinforcement learning algorithm. The reinforcement learning extreme process strategy includes extreme traction speed commands and extreme foaming injection pressure commands. The process execution module is used to control the traction device of the production equipment using the extreme traction speed command after receiving the reinforcement learning extreme process strategy, and to control the fluid raw material injection device of the production equipment using the extreme foaming injection pressure command.

5. The manufacturing execution system for the prefabricated direct-buried insulated pipe production line as described in claim 4, characterized in that, The smooth degradation process strategy includes deceleration compensation instructions and proportional-integral-derivative control logic instructions. The process execution module is used to reduce the traction speed of the traction device in the production equipment by using the deceleration compensation command after receiving the smooth degradation process strategy, and to replace the limit traction speed command and the limit foaming injection pressure command by using the proportional-integral-derivative control logic command. In the smooth degradation process strategy, the rate of change of process parameters is less than a preset rate of change threshold.

6. The manufacturing execution system for the prefabricated direct-buried insulated pipe production line as described in claim 1, characterized in that, The underlying sensor status data includes environmental vibration interference data, as well as raw material warehouse temperature data collected by the raw material warehouse temperature sensor, where the temperature drop rate is greater than or equal to a preset temperature change rate threshold. The data acquisition module is also used to acquire the initial viscosity data of the fluid raw material; The state quantization module is also used to determine noise characteristics using the environmental vibration interference data and the initial viscosity data, and to remove data that conforms to the noise characteristics from the underlying sensor state data for data cleaning to filter out data noise.

7. The manufacturing execution system for the prefabricated direct-buried insulated pipe production line as described in claim 3, characterized in that, The preset shutdown failure boundary includes the eccentricity state boundary where the eccentricity is greater than or equal to the preset eccentricity threshold and the foam gun blockage state boundary. The risk prediction module is also used to send a forced degradation signal to the strategy decision module when the risk contribution rate data is greater than or equal to a preset downtime risk threshold. The strategy decision module is also used to generate the smooth degradation process strategy in response to the forced degradation signal.

8. The manufacturing execution system for the prefabricated direct-buried insulated pipe production line as described in claim 1, characterized in that, The data acquisition module is also used to continuously collect updated human intervention behavior data and updated underlying sensor status data after the control command of the underlying device is issued; The state quantification module is also used to recalculate the current system risk state index based on the updated human intervention behavior data and the updated underlying sensor state data; The strategy decision module is also used to regenerate the reinforcement learning extreme process strategy when the current system risk state index is lower than the preset danger threshold and the duration is greater than or equal to the preset recovery time threshold.

9. The manufacturing execution system for a prefabricated direct-buried insulated pipe production line as described in any one of claims 1 to 8, characterized in that, The manufacturing execution system also includes an edge computing gateway and a data hub; The data acquisition module, the state quantification module, the strategy decision-making module, and the process execution module are all integrated within the data hub. The data hub is connected to the production equipment and the underlying control system via the edge computing gateway.

Images