Edge computing monitoring system for rotomolding equipment of a storage tank

Abstract

This invention relates to the field of edge computing technology, specifically to an edge computing monitoring system for rotational molding equipment for storage bins. The system includes a data access terminal, edge computing nodes, a state estimation program, and a control output terminal. The data access terminal obtains process formula numbers, stage switching signals, heating output status, temperature process values, rotation speed, and cooling execution status from the existing controller of the rotational molding equipment. The edge computing nodes generate a process state sequence according to time windows, identifying the heating, melting, homogenization, and cooling stages, and obtain a molding state estimate based on a thermal inertia state model that includes heat accumulation, heat dissipation attenuation, stage coupling, and boundary inheritance. The control output terminal outputs stage hold or stage switching correction commands based on the deviation between the molding state estimate and the target state trajectory. This system reflects the thermal inertia lag state of the rotational molding process locally on the equipment side, reducing molding state deviations caused by fixed time and single-point temperature judgments.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and more specifically to an edge computing monitoring system for rotomolding equipment for storage bins. Background Technology

[0002] Rotational molding equipment for storage bins is typically used for molding hollow bin-shaped products. Its control process generally includes stages such as mold rotation after material feeding, heating, material melting, homogenization, cooling and setting, and demolding. Existing equipment often uses programmable logic controllers (PLCs) or industrial controllers to execute fixed process programs. Operators pre-enter the process formula, which typically includes parameters such as heating temperature, heating time, holding time, rotation speed, cooling start timing, and cooling duration. During operation, the controller outputs heating, rotation, and cooling control signals according to the set sequence and reads the temperature process value, execution status, and stage switching signals to determine if the current process is within the set range. For monitoring, conventional solutions primarily rely on temperature upper and lower limits, running time elapsed, actuator start / stop status, and alarm contact status as the basis for judgment. When the temperature reaches the set value or a certain stage's timing ends, the controller enters the next process stage; when the temperature, speed, or execution status exceeds the preset range, the system triggers an alarm or suspends operation.

[0003] In similar existing technologies, some rotational molding equipment is equipped with a local monitoring terminal or upper-level monitoring software to centrally display temperature curves, heating output, rotation speed, and cooling status uploaded by the controller, and save batch data as historical records. This approach allows for traceability of equipment operation and assists operators in identifying the source of anomalies. Some solutions also upload the collected data to a server, where curve comparison, batch statistics, or process parameter management are performed, and recommended parameters or alarm information are returned to the field control terminal. A common characteristic of these technologies is that process judgment still relies primarily on preset time and single-point temperature conditions, with weak correlation between the monitoring logic and the continuous state of material heating, melting, and cooling during rotational molding. Because the mold continuously rotates during the rotational molding process of storage tanks, the temperature data collected typically only reflects the thermal state at the collection location or equipment control point, and cannot directly represent the degree of material spreading, melting, and homogenization on the mold's inner wall. Server-side analysis is also limited by communication links, data integrity, and real-time requirements, making it difficult to serve as a direct basis for stage switching.

[0004] The main technical problem with existing technologies is that fixed process procedures and threshold-based monitoring fail to address the thermal inertia lag during the rotational molding of storage tanks, leading to discrepancies between stage switching judgments and the actual molding state. Specifically, there is a time delay between heating output, temperature process values, and the material's melting state; reaching the set temperature does not necessarily indicate that the material inside the tank has been uniformly melted. Similarly, there is a lag between the cooling execution state and the tank's shaping degree; reaching the cooling time does not necessarily indicate that the molding state has stabilized. When raw material batches, ambient temperature, equipment thermal response, and rotational load change, the temperature curve under the same formula will deviate from the molding process. Existing solutions typically cannot infer the actual state of the heating, melting, homogenization, and cooling stages locally on the equipment based on continuous process data; they can only mechanically switch or trigger alarms after preset conditions are met. Therefore, monitoring results easily lag behind the molding process, and there is a lack of calculable state basis for stage maintenance or stage switching. Summary of the Invention

[0005] The purpose of this invention is to provide an edge computing monitoring system for rotomolding equipment for storage bins, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An edge computing monitoring system for a rotomolding equipment for storage bins includes a data access terminal, an edge computing node, a state estimation program, and a control output terminal. The data access terminal communicates with the existing controller of the rotomolding equipment for acquiring process formula numbers, stage switching signals, heating output status, temperature process values, rotation speed, and cooling execution status. The edge computing node runs the state estimation program, generating a process state sequence based on the temperature process values, heating output status, rotation speed, and cooling execution status according to time windows. It then identifies the heating, melting, homogenization, and cooling stages based on the process state sequence and obtains a molding state estimate based on a thermal inertia state model. The control output terminal outputs a stage hold or stage switching correction command based on the molding state estimate.

[0007] Preferably, the edge computing node performs stage synchronization processing on the process state sequence, configures a stage label for each time window based on the stage switching signal, and generates a feature group including temperature rise slope, heating duty cycle duration, rotational speed deviation, cooling start / stop duration, and stage cumulative duration within the same time window; when the stage switching signal is inconsistent with the temperature rise slope or the cooling start / stop duration, the edge computing node retains the original stage label and generates a candidate stage label; when there is missing data in the same time window, the edge computing node generates a missing data placeholder based on the sampling timestamp and appends the sampling completeness identifier to the corresponding feature group.

[0008] Preferably, the state estimation program establishes a candidate set of stage boundaries based on the changing direction of the feature group in the continuous time window, the delay relationship between the heating output state and the temperature response, the stability of the rotation speed deviation, and the correspondence between the cooling execution state and the cooling trend. The state estimation program filters the candidate set of stage boundaries according to the continuity of adjacent time windows, the order of stages, the consistency of process formula numbers, and the minimum stage retention constraint, and writes the filtered stage boundaries into the process state sequence to form a segmented sequence of heating, melting, homogenization, and cooling stages.

[0009] Preferably, the thermal inertia state model includes a heat accumulation term, a heat dissipation attenuation term, a stage coupling term, and a boundary inheritance term. The edge computing node updates the heat accumulation term based on the heating duty cycle and the temperature rise slope during the heating stage, updates the stage coupling term based on the fluctuation range of the temperature process value and the rotational speed deviation during the cooling stage, updates the heat dissipation attenuation term based on the cooling start-stop duration and the cooling slope generated by the temperature process value during the cooling stage, and inherits the end state of the previous stage by the boundary inheritance term at the boundary of adjacent stages, combining them to form the forming state estimate.

[0010] Preferably, before the control output terminal outputs the stage hold or stage switching correction instruction, the edge computing node matches the molding state estimate with the target state trajectory corresponding to the process formula number to obtain the stage offset; when the stage offset corresponds to insufficient heating, excessive heat equalization, or incomplete cooling, the edge computing node generates heating extension, heat preservation shortening, cooling extension, or switching prevention instructions according to the stage sequence constraints, and writes the instructions into the pending instruction queue; the instructions in the pending instruction queue are limited to the execution range allowed by the existing controller of the storage tank rotational molding equipment.

[0011] Preferably, the edge computing node sets up a local baseline update process. The local baseline update process takes the heat accumulation item, heat dissipation attenuation item, stage coupling item, boundary inheritance item, and stage duration of consecutive batches under the same process recipe number as input, and removes batch records containing alarm shutdown, manual forced switching, or communication interruption markers to form an individual thermal response baseline for the device. When the next batch starts, the edge computing node writes the individual thermal response baseline of the device into the local correction area of ​​the thermal inertia state model and the target state trajectory.

[0012] Preferably, after generating the stage offset, the edge computing node performs a credibility determination. The credibility determination forms a credibility vector by considering the consistency of the screening of the stage boundary candidate set, the sampling integrity identifier, the continuity of the rotation speed deviation, the validity of the cooling start-stop duration, and the applicability of the device individual thermal response baseline. When any dimension of the credibility vector does not meet the preset credibility condition, the control output terminal only outputs the stage hold instruction and the abnormal flag, does not output the stage early switch instruction, and writes the corresponding time window index into the auxiliary record of the pending instruction queue.

[0013] Preferably, the edge computing node is configured with a shadow simulation process. The shadow simulation process, without writing to the existing controller of the storage tank rotational molding equipment, uses the process state sequence of the current batch, the individual thermal response baseline of the equipment, the confidence vector, and the target state trajectory as inputs to calculate the original formula stage trajectory and the corrected stage trajectory in parallel. The control output terminal outputs the stage switching correction instruction in the pending instruction queue to the existing controller of the storage tank rotational molding equipment only when the corrected stage trajectory satisfies the stage sequence constraint, the execution range constraint, and the confidence determination.

[0014] Preferably, the edge computing node performs version verification on the corrected stage trajectory formed by the shadow inference process. The version verification writes the process recipe number, the individual thermal response baseline of the equipment, the parameter version of the thermal inertia state model, the confidence vector, and the queue of instructions to be executed into the candidate correction record. When a new candidate correction record and the most recent valid candidate correction record produce a switching correction in opposite directions at the same stage, the control output terminal pauses the output of the stage switching correction instruction, writes the stage hold instruction into the queue of instructions to be executed, and retains the rollback mark of the new candidate correction record.

[0015] Preferably, the edge computing node is configured to run offline. In response to a communication interruption with the cloud or upper-level monitoring terminal, the offline running process locks the current parameter version of the thermal inertia state model, the individual thermal response baseline of the device, and the most recent valid candidate correction record. It continues to generate stage hold or stage switch correction instructions according to the process state sequence, the confidence vector, and the queue of instructions to be executed. After communication is restored, the edge computing node only uploads a batch summary containing stage boundaries, forming state estimates, candidate correction record indexes, and correction instruction indexes.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention establishes an edge computing node on the side of the rotomolding equipment for storage tanks. This node reads the process formula number, stage switching signal, heating output status, temperature process value, rotational speed, and cooling execution status, enabling the local generation of a process state sequence organized by time windows. The state estimation program segments the heating, melting, homogenization, and cooling stages based on the temperature rise slope, heating duty cycle duration, rotational speed deviation, cooling start / stop duration, and stage cumulative duration. Then, a thermal inertia state model including heat accumulation, heat dissipation attenuation, stage coupling, and boundary inheritance terms is used to obtain the molding state estimate, ensuring that stage judgment no longer relies solely on single-point temperature or fixed timing. After matching the molding state estimate with the target state trajectory, a stage offset is generated. The control output terminal outputs a stage hold or stage switching correction command accordingly, ensuring that heating extension, heat preservation shortening, cooling extension, or switching prevention all have a continuous process data basis, mitigating molding state misjudgments caused by thermal inertia lag.

[0017] 2. This invention, through stage synchronization processing, stage boundary candidate screening, local baseline updating, credibility determination, and shadow inference, enables the edge-side monitoring process to constrain data gaps, stage signal inconsistencies, and individual device thermal response differences. Gap placeholders and sampling integrity identifiers are used to preserve the location of abnormal data, preventing missing data from being directly used for state estimation. Individual device thermal response baselines are formed from consecutive batch records under the same process recipe number, excluding alarm shutdowns, manual forced switching, and communication interruption records, allowing the thermal inertia state model to be locally corrected according to the actual thermal response changes of individual devices. Credibility vectors and shadow inference are used to restrict correction instructions that do not meet stage sequence constraints, execution range constraints, or credibility conditions from entering the controller. During communication interruptions, edge computing nodes can still lock the current model version, individual device thermal response baselines, and the most recent valid candidate correction records to continue running, and upload batch summaries after communication is restored, maintaining the continuity of the monitoring and correction process. Attached Figure Description

[0018] Figure 1 This is an overview diagram of the overall system data flow and control flow of the present invention; Figure 2 This is a flowchart of the stage synchronization processing and stage boundary identification of the present invention; Figure 3 This is a flowchart of the thermal inertia state estimation and stage correction output of the present invention; Figure 4 This is a flowchart illustrating the device individual thermal response baseline update, reliability determination, and shadow simulation process of the present invention. Figure 5 This is a flowchart of the candidate correction record versioning verification and offline operation process of the present invention. Detailed Implementation

[0019] refer to Figure 1 In one embodiment, an edge computing monitoring system for a piggy bank rotational molding equipment is deployed locally on the existing control link of the equipment. It consists of a data access terminal, an edge computing node, a state estimation program, and a control output terminal. The data access terminal communicates with the existing controller of the piggy bank rotational molding equipment, reading process formula numbers, stage switching signals, heating output status, temperature process values, rotation speed, and cooling execution status. The data access terminal does not change the control logic of the existing controller; it only mirrors the data generated or collected by the existing controller and writes data from different sources to the input buffer of the edge computing node according to a unified sampling timestamp. The edge computing node runs the state estimation program, dividing the process data in the input buffer into time windows. For each time window, it extracts state variables related to the rotational molding process, generating a process state sequence. The control output terminal communicates with the process stage control entry point or the upper-level control logic of the existing controller to receive stage hold instructions or stage switching correction instructions generated by the edge computing node. Before outputting, the control output terminal converts the instruction format to ensure that the output content corresponds to control items such as stage hold, heating extension, cooling extension, or switching prevention that can be recognized by the existing controller, thus avoiding unconstrained continuous control quantities output from the edge side.

[0020] In this embodiment, the process state sequence is constructed in the following form: ; in, Indicates the first A process state vector for each time window A numerical identifier representing the process formula number. This indicates that a phase switching signal has been provided by an existing controller. This indicates the average duty cycle of the heating output state within the current time window. This represents the average value of the temperature process within the current time window. This represents the average rotational speed within the current time window. This indicates the average activation level of the cooling execution status within the current time window. This indicates the duration of the current process stage. This indicates the proportion of valid sampling points to the total number of sampling points within the current time window. (Operator) This indicates vector concatenation, where each component is written to the same record location in a fixed order. For example, when a certain time window corresponds to recipe number 3, stage switching signal 1, average heating output duty cycle 0.82, average temperature process value 168.6, average rotational speed 7.5, cooling activation value 0, stage duration 420, and sampling integrity 0.96, the following is obtained: The closer the average duty cycle of heating output is to 1, the higher the continuity of heating execution within the current time window; a cooling activation rate of 0 indicates that the current time window is not in a cooling execution state; and a sampling integrity rate of 0.96 indicates that only a small number of sampling points are missing.

[0021] Table 1 describes the fields in the process status sequence, which are used to define the meaning of the fields in edge computing node access and data organization.

[0022] After the edge computing nodes form a process state sequence, the state estimation program identifies the heating, melting, homogenization, and cooling stages. Instead of relying solely on existing controller stage switching signals, the program incorporates these signals along with the temperature process value, heating output status, rotational speed, and cooling execution status into the state estimation input. The state estimation program jointly analyzes the temperature rise trend, heating persistence, rotational speed stability, and cooling activation status within a continuous time window to obtain segmented results for calculating the thermal inertia state. The thermal inertia state model calculates heat accumulation, heat dissipation attenuation, and stage inheritance relationships within each stage to obtain a molding state estimate. The control output outputs a stage hold or stage switch correction command based on the offset between the molding state estimate and the target state trajectory. Therefore, the storage tank rotational molding equipment can maintain the current stage when the temperature reaches the set value but the molding state estimate has not yet reached the target state; and can switch stages according to the correction command when the molding state estimate has reached the target state and meets the control constraints. The advantage of this embodiment is that the local side of the rotational molding equipment can form a molding state estimate based on continuous process data, and the stage control does not rely solely on fixed time and single-point temperature, thus constraining the stage judgment deviation caused by thermal inertia lag.

[0023] refer to Figure 2In one embodiment, the edge computing node performs stage synchronization processing on the process state sequence. This process uses the stage switching signal output by the existing controller as the initial stage label, assigning each time window to the corresponding process stage. Specifically, when the stage switching signal remains unchanged within a continuous time window, the edge computing node writes the same stage label into the corresponding time window. When the stage switching signal changes, the edge computing node records the time window index of the stage switch and checks whether the temperature rise slope, cooling start / stop duration, and rotational speed deviation are consistent with the characteristics of the new stage within several consecutive time windows at the beginning of the new stage. For the heating stage, the temperature rise slope and heating duty cycle duration should have a unidirectional response; for the cooling stage, the cooling start / stop duration and cooling trend should correspond; for the homogenization stage, the temperature fluctuation amplitude should be lower than that of the heating stage, and the rotational speed deviation should remain within a determinable stable range. If the stage switching signal has changed, but the temperature rise slope or cooling start / stop duration is inconsistent with the new stage, the edge computing node retains the original stage label and generates candidate stage labels. The original stage label is used to record the control state given by the existing controller, and the candidate stage label is used by the thermal inertia state model to determine the actual forming stage, so as to avoid directly overwriting the field control record when there is a lag between the controller stage signal and the thermal response.

[0024] In this embodiment, the edge computing node generates a feature group for each time window. The feature group includes the temperature rise slope, heating duty cycle duration, rotational speed deviation, cooling start / stop duration, and cumulative stage duration. The temperature rise slope is obtained by the difference between the average temperature values ​​of adjacent time windows; the heating duty cycle duration is obtained by accumulating the average heating output duty cycle of the current time window and adjacent preceding time windows; the rotational speed deviation is obtained by the difference between the average rotational speed of the current time window and the set rotational speed corresponding to the current recipe; the cooling start / stop duration is obtained by the activation length of the cooling execution state in consecutive time windows; and the cumulative stage duration is calculated by converting the stage label duration window number. If there is missing data in the same time window, the edge computing node does not directly interpolate and overwrite the missing values. Instead, it writes a missing data placeholder in the corresponding field and appends a sampling integrity identifier to the feature group. The missing data placeholder participates in subsequent reliability determination, and the sampling integrity identifier participates in stage boundary filtering. Therefore, the location of missing data remains traceable during subsequent processing.

[0025] In a preferred embodiment, the state estimation procedure identifies a set of candidate stage boundaries using boundary scores. The boundary scores are calculated as follows: ; in, Indicates the first Stage boundary scoring for each time window Indicates the first The temperature rise slope of each time window, Indicates the first Average duty cycle of heating output for each time window Indicates the first To the The arithmetic mean of the average duty cycle of heating output within a time window. This indicates the cooling response intensity obtained from both the cooling activation level and the temperature decrease trend. Indicator function representing changes in stage switching signals; when and The value is 1 when they are different, and 0 when they are the same. , , , The boundary score weights are non-negative and sum to 1. This represents absolute value operations. For example, if... , , , , No. A time window satisfies , , And if the stage switching signal changes, then This value is used to compare with the candidate boundary conditions under the same formulation. The higher the value, the more likely the current time window is to be the stage boundary.

[0026] Table 2 explains the stage boundary screening constraints, which are used to explain the criteria for determining whether the candidate set of stage boundaries enters the segment sequence.

[0027] When the state estimation program filters the candidate set of stage boundaries, it uses the continuity of adjacent time windows, the order of stages, the consistency of process recipe numbers, and the minimum stage retention constraint as common conditions. If a candidate boundary is generated only by a temperature abrupt change in a single time window, but there is no change in heating output or cooling response in adjacent time windows, it is treated as an isolated disturbance. If the candidate stage label causes the cooling stage to revert to the heating stage, the candidate boundary is not written into the segment sequence. If the process recipe numbers on both sides of the boundary are different, the current batch segmentation is terminated, and a new batch sequence is established. The stage boundaries after filtering are written into the process state sequence, forming a segment sequence of heating, melting, homogenization, and cooling stages. The advantage of this embodiment is that the stage segmentation simultaneously retains the existing controller stage state and the edge-side candidate stage state, which can form a calculable segmentation basis when the control signal and thermal response are inconsistent, avoiding mistaking short-term sampling fluctuations or single-point temperature jumps for true stage boundaries.

[0028] refer to Figure 3 In one embodiment, the thermal inertia state model consists of a heat accumulation term, a heat dissipation attenuation term, a stage coupling term, and a boundary inheritance term. Edge computing nodes update the heat accumulation term during the heating phase based on the heating duty cycle and temperature rise slope; during the homogenization phase, they update the stage coupling term based on the fluctuation amplitude of the temperature process value and the rotational speed deviation; during the cooling phase, they update the heat dissipation attenuation term based on the cooling start / stop duration and the cooling slope generated by the temperature process value; and at the boundary of adjacent stages, the boundary inheritance term inherits the end state of the previous stage. This model does not require mold structure parameters or additional detection mechanisms as necessary conditions, but rather uses process data provided by the existing controller for state inference. The heat accumulation term represents the effective heating degree of the material and mold system under continuous heating output; the heat dissipation attenuation term represents the state decline caused by cooling execution; the stage coupling term represents the influence of rotational state on the continuity of heat distribution; and the boundary inheritance term prevents the thermal state of the previous stage from being reset to zero during stage switching.

[0029] In this embodiment, the molding state estimate is calculated in the following form: ; in, Indicates the first The estimated form of the shape state within a time window This represents the estimated state of formation in the previous time window. Indicates the stage type corresponding to the current segment sequence. This represents the boundary inheritance coefficient at the current stage. This represents the cumulative heat term. This indicates the heat dissipation attenuation term. Indicates stage coupling terms, This represents the heat transfer coefficient at the current stage. This represents the heat dissipation conversion coefficient at the current stage. This represents the rotational coupling coefficient at the current stage. Indicates the length of the time window. This represents the normalized deviation of the rotational speed. (Operator) Represents state superposition, operators This indicates a status deduction. This represents absolute value operations. For example, within a certain time window of a heating phase, if... , , , , , , , , ,but , , ,get The calculation results indicate that within the current time window, the preceding thermal state is preserved according to the stage inheritance coefficient, and the current thermal input and rotational coupling contributions are superimposed.

[0030] Furthermore, when the state estimation program identifies the boundary between the heating stage and the melting stage, the edge computing nodes are not reset. Instead, it is the end of the previous stage. pass Inherited to the next stage. If the stage switching signal has already entered the soaking stage, but... If the value is still lower than the corresponding value of the target state trajectory in the same stage, the edge computing node will record the stage offset as insufficient heating or incomplete melting. If cooling execution is already enabled, but... If the temperature process value is small and the cooling slope corresponding to it does not show a continuous change, the edge computing node will maintain the cooling state as incomplete. The coefficients of each stage of the thermal inertia state model are stored in the edge-side formula model area, and the edge computing node retrieves the corresponding coefficients according to the process formula number; when the local baseline is updated, the local correction area in the formula model area participates in the next batch of calculations. The advantage of this embodiment is that the molding state estimate remains continuous between stages, which can reflect the lag relationship of heating, homogenization and cooling processes, and avoids treating the rotational molding process as unrelated discrete time periods when switching stages.

[0031] In one embodiment, after obtaining the molding state estimate, the edge computing node matches the molding state estimate with the target state trajectory corresponding to the process recipe number. The target state trajectory is stored in the edge computing node's local recipe library and indexed according to the process recipe number, stage type, and cumulative stage duration. The target state trajectory is not a direct copy of the temperature setpoint curve, but is formed by the thermal inertia state estimate under normal batch conditions, representing the state range that a certain recipe should reach in each stage. Based on the process recipe number and segmented stage of the current time window, the edge computing node reads the corresponding target value from the target state trajectory and calculates the stage offset of the current molding state estimate relative to the target value. When the stage offset corresponds to insufficient heating, the edge computing node generates a heating extension command; when the stage offset corresponds to excessive heat homogenization, the edge computing node generates a heat preservation shortening command; when the stage offset corresponds to incomplete cooling, the edge computing node generates a cooling extension command; when the stage switching signal has met the existing controller conditions but the edge-side state does not meet the switching conditions, the edge computing node generates a switching prevention command.

[0032] In this embodiment, the pending instruction queue stores correction instructions that have not yet been output by the control output or are awaiting verification. Each record in the instruction queue includes a batch index, stage index, instruction type, state estimate index, stage offset, execution range verification result, confidence index, candidate correction record index, and rollback flag. Before writing to the pending instruction queue, the edge computing node reads the execution range allowed by the existing controller and restricts the correction instructions to discrete control items that the existing controller can accept, such as stage holding, stage switching, extending the current stage, or preventing switching. If the existing controller is currently in a state of manual forced control, alarm shutdown, or recipe switching, the pending instruction queue only retains candidate records and does not release them to the control output. If there are multiple correction instructions in the same batch, stage, and direction in the instruction queue, the edge computing node retains the record with the newer time window index and marks the older record as overwritten, avoiding the repeated output of multiple instructions with the same direction in the same stage.

[0033] Table 3 describes the fields in the pending instruction queue, which are used to explain the recorded content from the generation to the output of the edge-side correction instruction.

[0034] Specifically, when the temperature process value is close to the set temperature and the existing controller is preparing to enter the homogenization stage, if the estimated molding state is lower than the value corresponding to the target state trajectory, the edge computing node does not directly change the heating output ratio. Instead, it writes a stage hold or heating extension instruction to the queue of instructions to be executed. After reading the queue record, the control output terminal converts the instruction into a stage hold command that the existing controller can recognize. For the cooling stage, if the cooling start / stop duration has reached the formula-set duration, but the state corresponding to the heat dissipation attenuation term has not yet reached the cooling completion interval, the edge computing node writes a cooling extension instruction. This method allows the edge computing node to correct the stage behavior instead of directly controlling the actuator by bypassing the existing controller. The advantage of this embodiment is that the offset between the estimated molding state and the target state trajectory can be converted into a constrained stage correction instruction. The correction content is limited to the range acceptable to the existing control system, and the edge-side state judgment is consistent with the field control safety boundary.

[0035] refer to Figure 4In one embodiment, the edge computing node sets up a local baseline update process. This process takes the accumulated heat, heat dissipation attenuation, stage coupling, boundary inheritance, and stage duration of consecutive batches under the same process recipe number as input to form an individual thermal response baseline for the device. After each batch, the edge computing node reads the stage segment sequence, molding state estimate, correction instruction index, and operational anomaly flags from the batch summary. If the batch record contains alarm shutdown, manual forced switching, or communication interruption flags, the edge computing node excludes the corresponding batch from the baseline update input. If the batch record does not contain the above flags and the sampling completeness meets the baseline update conditions, the edge computing node writes the thermal response characteristics of that batch into the baseline candidate cache. The baseline candidate cache is grouped according to the process recipe number to avoid data mixing between different storage tank sizes or different rotational molding processes.

[0036] In this embodiment, the individual thermal response baseline of the device is updated in the following manner: in, Indicates the previous process formula number under the same process formula A baseline vector of individual device thermal response formed from available batches, Indicates the first Each batch of thermal response feature vectors includes thermal accumulation characteristics, heat dissipation attenuation characteristics, load stability characteristics, and stage duration characteristics. Indicates the first The weight of each batch participating in baseline updates, Indicates the first The batch number indicates whether no alarm shutdown occurred; if no alarm occurred, the value is 1; if an alarm occurred, the value is 0. Indicates the first Whether a batch did not experience a manual forced switchover; use 1 if no switchover occurred, and 0 if a switchover occurred. Indicates the first The batch indicates whether no communication interruption occurred; a value of 1 indicates no interruption, and a value of 0 indicates an interruption. Operators This represents summing over batch indices, while fractions represent weighted averages. For example, if the thermal response feature vectors of three batches are... , , The third batch involved a forced manual switchover. , , ,get The calculation results indicate that the baseline of the individual thermal response of the equipment is formed from the batches that do not include anomaly markers.

[0037] Furthermore, the individual thermal response baseline of the equipment is written into the local correction area of ​​the thermal inertia state model and the target state trajectory. During writing, the edge computing node retains the original formula model area and does not directly overwrite the factory formula or the basic model distributed from the cloud. When calculating the thermal inertia state model, the individual thermal response baseline of the equipment in the local correction area is read first; when there is no available baseline with the same process formula number in the local correction area, the edge computing node uses the basic model parameters. The local correction area of ​​the target state trajectory corrects the time index of the target state at each stage according to the individual thermal response baseline of the equipment, so that the same formula maintains a longer heating or homogenization state on equipment with a slower thermal response, and avoids unfounded extension of stages on equipment with a faster thermal response. If a subsequent batch experiences an alarm shutdown or manual forced switchover, the edge computing node does not use that batch to update the baseline, and retains the abnormal batch index in the baseline record. The advantage of this embodiment is that the edge side can form a local thermal response baseline based on the continuous batch operation data of a single equipment, the thermal inertia state model and the target state trajectory can reflect the individual differences of the equipment, and abnormal batches are excluded from baseline updates.

[0038] In one embodiment, after generating the stage offset, the edge computing node performs a credibility determination. This determination forms a credibility vector by considering the consistency of the stage boundary candidate set, the sampling integrity indicator, the continuity of the rotational speed deviation, the validity of the cooling start-stop duration, and the applicability of the individual device's thermal response baseline. The consistency of the stage boundary candidate set is determined by the persistence of the boundary candidates within adjacent time windows and the stage sequence; the sampling integrity indicator follows the process state sequence. The continuity of rotational speed deviation is determined by the variation range of rotational speed deviation between adjacent time windows; the effectiveness of cooling start / stop duration is determined by the correspondence between cooling execution status and temperature decrease trend; the applicable marker for the individual equipment thermal response baseline is determined by whether the current process formula number is consistent with the formula number recorded on the baseline, and whether the baseline is formed from available batches. Confidence determination is not used to directly change the estimated molding state, but rather to determine whether a correction instruction can enter the output path.

[0039] In this embodiment, the confidence level is calculated in the following form: in, Indicates the first The credibility judgment value for each time window. Indicates consistency in stage boundary screening. Indicates the completeness of sampling. This indicates the continuity of the rotational speed deviation. Indicates the validity of the cooling start / stop duration. Indicates the applicable markings for the individual thermal response baseline of the equipment; , , , , This represents the credibility weight, with non-negative values ​​and a total sum of 1. (Symbol) This indicates a weighted summation, and multiplication indicates the contribution of each dimension's value to the decision value. For example, if... , , , , ,and , , , , ,but The calculation result indicates that the correction instruction corresponding to the current time window has a high degree of data support. If any dimension does not meet the preset confidence conditions, such as the sampling integrity indicator being too low or the individual thermal response baseline of the device being inapplicable, the control output terminal will only output the stage hold instruction and the abnormal flag, without outputting the stage early switch instruction, and will write the corresponding time window index into the supplementary record of the pending instruction queue.

[0040] Furthermore, the edge computing node sets up a shadow simulation process, which runs without being written into the existing controller of the rotomolding equipment. The inputs include the process state sequence of the current batch, the individual thermal response baseline of the equipment, the confidence vector, and the target state trajectory. The shadow simulation process calculates the original formula stage trajectory and the corrected stage trajectory in parallel. The original formula stage trajectory is calculated according to the stage switching sequence of the current formula on the existing controller, while the corrected stage trajectory is calculated according to the stage correction content in the pending instruction queue. Both trajectories obtain molding state estimates through the same thermal inertia state model, differing only in the stage index and stage hold time. The edge computing node compares the deviation of the two trajectories from the target state trajectory and checks whether the corrected stage trajectory meets the stage sequence constraints, execution range constraints, and confidence judgment. Only when all the above constraints are met does the control output send the stage switching correction instructions from the pending instruction queue to the existing controller.

[0041] The trajectory difference in shadow projection is calculated in the following form: in, This represents the difference between the corrected stage trajectory and the original formula stage trajectory within the same projection interval. and These represent the start and end time window indices of the projection interval, respectively. This indicates the corrected stage trajectory at the [number]th [stage]. The estimated form of the shape state within a time window This indicates the original formula's stage trajectory at the [number]th [stage]. The estimated form of the shape state within a time window This indicates the corrected stage trajectory at the [number]th [stage]. Stage index for each time window This indicates the original formula's stage trajectory at the [number]th [stage]. Stage index for each time window Represents the weight of the stage index difference. Operator This indicates summing over time windows within the projection interval. This represents absolute value operations. For example, if the deduction interval contains 3 time windows, then within the 3 time windows... The values ​​were 0.5, 0.4, and 0.3 respectively, and the differences in the stage indexes were 0, 1, and 1 respectively. ,but This value is used to record the magnitude of change in the corrected trajectory relative to the original formula trajectory, and participates in the output determination together with the execution range constraint. The advantage of this embodiment is that the reliability determination restricts the correction instructions driven by low-quality data from entering the output path, and the shadow simulation compares the original formula trajectory and the corrected trajectory without writing to the controller, so that the stage correction has a traceable calculation basis.

[0042] refer to Figure 5 In one embodiment, the edge computing node performs version verification on the corrected stage trajectory formed during the shadow simulation process. Version verification writes the process recipe number, individual equipment thermal response baseline, parameter version of the thermal inertia state model, confidence vector, and queue of pending instructions into a candidate correction record. The candidate correction record is indexed according to batch and stage indices, and retains the generation time window, state estimate index, stage offset, shadow simulation difference, and output state. If a new candidate correction record and the most recent valid candidate correction record produce a switching correction in opposite directions at the same stage, the control output pauses the output of the stage switching correction instruction, writes the stage hold instruction into the queue of pending instructions, and writes a rollback flag into the new candidate correction record. Opposite-direction switching corrections include the previous record indicating an early entry into the next stage, while the new record indicates an extension of the current stage; or the previous record indicating a shortened homogenization stage, while the new record indicates continued homogenization. Through version verification, the edge computing node can identify reverse corrections caused by data fluctuations or model version differences within the same stage, preventing correction instructions from changing repeatedly in a short period.

[0043] In this embodiment, candidate correction records are not directly output as control commands. Before outputting, the edge computing node reads the parameter version and the individual thermal response baseline version of the candidate correction record. If the current thermal inertia state model has been updated and the candidate correction record originates from an older model version, the record is changed to a pending review state. If the current process recipe number differs from the recipe number in the candidate correction record, the record does not participate in the current batch output. If any dimension of the confidence vector fails in the current time window, the record is only retained in the batch summary. For records with rollback flags, the control output can only output stage hold instructions, not stage early switch instructions. Thus, version verification binds the model version, process recipe, confidence level, and instruction queue to the same record, enabling the location of the state basis for the correction instruction during subsequent batch reviews.

[0044] Furthermore, the edge computing nodes are configured for offline operation, responding to communication interruptions with the cloud or upper-level monitoring. During communication interruption, the edge computing nodes lock the current parameter version of the thermal inertia state model, the individual device thermal response baseline, and the most recently valid candidate correction record, refraining from accepting new remote models or remote recipe changes. The edge computing nodes continue to generate stage-holding or stage-switching correction instructions according to the local process state sequence, confidence vector, and pending instruction queue. If sampling integrity decreases or confidence judgment fails during communication interruption, the control output only outputs stage-holding instructions and anomaly flags. After communication is restored, the edge computing nodes do not upload the complete original curves but instead upload batch summaries. These batch summaries include stage boundaries, molding state estimates, candidate correction record indices, and correction instruction indices. The batch summaries are indexed against the locally stored original process state sequences. When the remote end needs to trace back, it requests the corresponding record through the batch index, avoiding a large amount of original data consuming the link during communication recovery.

[0045] In a preferred embodiment, the offline operation process also performs a consistency comparison of the model versions before and after the communication interruption. If the model version returned by the cloud or the upper-level monitoring terminal after communication is restored is different from the model version locked during the offline period, the edge computing node first marks the candidate correction record formed during the offline period as the offline version record, and then writes the new model into the verification area. The new model participates in shadow simulation in subsequent batches and does not directly replace the current executable model. When the corrected stage trajectory formed by the new model is consistent with the trajectory direction of the current executable model, and the credibility judgment meets the conditions, the edge computing node writes the new model version into the current parameter version area. If the new model and the offline version record generate a reverse switching correction in the same stage, the control output continues to output according to the stage hold instruction and retains the rollback mark. The advantage of this embodiment is that the version verification and offline operation process enable the edge computing node to maintain certain processing rules in scenarios of communication anomalies, model updates, and instruction rollbacks. There is a correspondence between the stage correction record, the model version, and the batch summary, and the on-site monitoring process will not be interrupted due to changes in the remote connection status.

Claims

1. An edge computing monitoring system for rotational molding equipment for storage bins, characterized in that, This includes data access terminals, edge computing nodes, state estimation programs, and control output terminals; The data access terminal communicates with the existing controller of the storage tank rotational molding equipment to obtain the process formula number, stage switching signal, heating output status, temperature process value, rotation speed and cooling execution status; The edge computing node runs the state estimation program to generate a process state sequence by dividing the temperature process value, the heating output state, the rotation speed and the cooling execution state into time windows, and identifies the heating, melting, homogenization and cooling stages according to the process state sequence, and obtains the molding state estimate based on the thermal inertia state model. The control output terminal outputs a stage hold or stage switching correction command based on the molding state estimate.

2. The system according to claim 1, characterized in that, The edge computing node performs stage synchronization processing on the process state sequence, configures stage labels for each time window according to the stage switching signal, and generates a feature group including temperature rise slope, heating duty duration, rotation speed deviation, cooling start / stop duration and stage cumulative duration within the same time window. When the stage switching signal is inconsistent with the temperature rise slope or the cooling start / stop duration, the edge computing node retains the original stage label and generates candidate stage labels; When there is missing data in the same time window, the edge computing node generates a missing data placeholder based on the sampling timestamp and attaches the sampling integrity identifier to the corresponding feature group.

3. The system according to claim 2, characterized in that, The state estimation program establishes a candidate set of stage boundaries based on the changing direction of the feature group in the continuous time window, the delay relationship between the heating output state and the temperature response, the stability of the rotation speed deviation, and the correspondence between the cooling execution state and the cooling trend. The state estimation program filters the candidate set of stage boundaries according to the continuity of adjacent time windows, the order of stages, the consistency of process formula numbers, and the minimum stage preservation constraint, and writes the filtered stage boundaries into the process state sequence to form a segmented sequence of heating, melting, homogenization and cooling stages.

4. The system according to claim 3, characterized in that, The thermal inertial state model includes a heat accumulation term, a heat dissipation attenuation term, a stage coupling term, and a boundary inheritance term; During the heating phase, the edge computing node updates the heat accumulation term based on the heating duty cycle and the temperature rise slope. During the homogenization phase, it updates the stage coupling term based on the fluctuation range of the temperature process value and the rotation speed deviation. During the cooling phase, it updates the heat dissipation attenuation term based on the cooling start-stop duration and the cooling slope generated by the temperature process value. At the boundary of adjacent stages, the boundary inheritance term inherits the end state of the previous stage and combines them to form the forming state estimate.

5. The system according to claim 4, characterized in that, Before the control output terminal outputs the stage hold or stage switching correction command, the edge computing node matches the molding state estimate with the target state trajectory corresponding to the process formula number to obtain the stage offset. When the stage offset corresponds to insufficient heating, excessive heat equalization, or incomplete cooling, the edge computing node generates heating extension, heat preservation shortening, cooling extension, or switching blocking instructions according to the stage sequence constraints, and writes the instructions into the pending instruction queue. The instructions in the queue of instructions to be executed are limited to the execution range allowed by the existing controller of the storage bucket rotational molding equipment.

6. The system according to claim 5, characterized in that, The edge computing node sets up a local baseline update process. The local baseline update process takes the heat accumulation item, heat dissipation attenuation item, stage coupling item, boundary inheritance item and stage duration of consecutive batches under the same process recipe number as input. After removing batch records containing alarm shutdown, manual forced switching or communication interruption markers, an individual thermal response baseline for the equipment is formed. When the edge computing node starts up in the next batch, it writes the individual thermal response baseline of the device into the local correction area of ​​the thermal inertial state model and the target state trajectory.

7. The system according to claim 6, characterized in that, After generating the stage offset, the edge computing node performs a credibility determination. The credibility determination forms a credibility vector by considering the consistency of the screening of the stage boundary candidate set, the sampling integrity identifier, the continuity of the rotation speed deviation, the validity of the cooling start-stop duration, and the applicability of the device individual thermal response baseline. When any dimension of the credibility vector does not meet the preset credibility condition, the control output terminal only outputs the stage hold instruction and the abnormal flag, does not output the stage early switch instruction, and writes the corresponding time window index into the auxiliary record of the instruction queue to be executed.

8. The system according to claim 7, characterized in that, The edge computing node is configured with a shadow simulation process. Without writing into the existing controller of the storage tank rotational molding equipment, the shadow simulation process uses the process state sequence of the current batch, the individual thermal response baseline of the equipment, the confidence vector, and the target state trajectory as inputs to calculate the original formula stage trajectory and the corrected stage trajectory in parallel. The control output terminal outputs the stage switching correction instruction in the pending instruction queue to the existing controller of the storage bucket rotational molding equipment only when the corrected stage trajectory satisfies the stage sequence constraint, the execution range constraint, and the credibility determination.

9. The system according to claim 8, characterized in that, The edge computing node performs version verification on the corrected stage trajectory formed by the shadow inference process. The version verification writes the process recipe number, the individual thermal response baseline of the equipment, the parameter version of the thermal inertia state model, the confidence vector, and the queue of instructions to be executed into the candidate correction record. When a new candidate correction record and the most recent valid candidate correction record produce a switching correction in opposite directions at the same stage, the control output terminal pauses the output of the stage switching correction instruction, writes the stage hold instruction into the pending instruction queue, and retains the rollback mark of the new candidate correction record.

10. The system according to claim 9, characterized in that, The edge computing node is set to run offline. In response to the interruption of communication with the cloud or the upper monitoring terminal, the offline running process locks the current parameter version of the thermal inertial state model, the individual thermal response baseline of the device and the most recent valid candidate correction record, and continues to generate stage hold or stage switch correction instructions according to the process state sequence, the confidence vector and the queue of instructions to be executed. After communication is restored, the edge computing node only uploads a batch summary containing stage boundaries, forming state estimates, candidate correction record indexes, and correction instruction indexes.

Images