Optimization method of metal product production intelligent control system based on PLC technology

Abstract

The application discloses an optimization method of a metal product production intelligent control system based on PLC technology, and relates to the technical field of industrial production intelligent control, which comprises the following steps: reading original production operation logs of a PLC controller, removing invalid pulse data through structured cleaning, and generating a production process execution record set in a unified format; obtaining a process flow diagram, a reference parameter band and a fault propagation link model through process path tracing, steady-state interval extraction and root cause correlation analysis, and generating a comprehensive operation characteristic matrix by synergistically mapping the three; and constructing a control logic adaptive network for dynamic optimization, and forming an optimized control strategy set containing control instruction sequences and logic weight coefficients. The application realizes multi-dimensional fusion representation of production data and dynamic adaptation of control logic, and optimizes the running state of a metal product production PLC control system.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control technology for industrial production, specifically an optimization method for an intelligent control system for metal product manufacturing based on PLC technology. Background Technology

[0002] Existing intelligent control systems for metal product manufacturing mostly use PLC controllers as the core to perform equipment control. During system operation, they simply store data such as equipment action codes, process parameters, and processing sequences. There is no specific processing mechanism for invalid pulse data generated by signal jitter. The three types of operating data—process parameters, process flow, and fault alarms—are processed by independent acquisition and analysis. The analysis of system operating status relies only on single-dimensional data. The control logic and control strategy adopt a pre-set static configuration mode, which makes it impossible to adjust the control logic and command parameters according to the real-time production status.

[0003] Under the existing PLC control mode, the original operating data sequence contains invalid data interference, the data format cannot be unified and standardized, the analysis process of process flow, process parameters, and fault alarms is fragmented, and it is impossible to form a collaborative relationship of multi-dimensional data. The fault propagation path cannot be correlated with process parameters and process flow, and the health status of the production system cannot be characterized by a quantitative matrix.

[0004] It is necessary to eliminate invalid pulse interference in the original operating data and standardize the data format. It is necessary to coordinate the process flow diagram, benchmark parameter band, and fault propagation link model to generate a comprehensive operating feature matrix. It is necessary to build a control logic adaptation network based on the comprehensive operating feature matrix to achieve dynamic optimization and form an optimized control strategy set containing control command sequence and logic weight coefficient. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, this invention proposes an optimization method for an intelligent control system for metal product manufacturing based on PLC technology, including: The original production operation log containing equipment action codes, process parameter settings, and processing timing trigger points is read from the PLC controller of the intelligent control system for metal product production to form the original operation data sequence. The original operating data sequence is subjected to structured cleaning to remove invalid pulse data caused by signal jitter, and a production process execution record set with a unified format is generated. The production process execution record set includes process flow identifiers, multi-dimensional process parameter time series, and fault alarm event time series. Process path tracing is performed on the process flow identifier to obtain the complete process flow map of the current batch of products. Steady-state interval extraction is performed on the time series of the multi-dimensional process parameters to obtain the baseline parameter band of each process link. Root cause correlation analysis is performed on the time series of the fault alarm events to obtain the fault propagation link model. The process flow diagram, the baseline parameter band, and the fault propagation link model are input into the system state reconstruction device for collaborative mapping to generate a comprehensive operating feature matrix that reflects the current health status of the production system. Based on the comprehensive operational feature matrix, a control logic adaptation network is constructed for dynamic optimization, forming an optimized control strategy set that includes control instruction sequences and logic weight coefficients.

[0006] Furthermore, process path tracing is performed on the process flow identifiers to obtain a complete process flow map of the current batch of products, including: Extract the equipment number, fixture number, and material batch number from the process flow identifier, and construct a three-dimensional index relationship table; Based on the three-dimensional index relationship table, the sensor feedback signals of each workstation in the intelligent control system for metal product production are retrieved to reconstruct the physical movement trajectory of the material between each workstation. The physical movement trajectory is compared with the preset standard process route, and all abnormal flow nodes that deviate from the standard route are marked. By connecting normal flow nodes and abnormal flow nodes with directed edges, a complete process flow graph containing branch paths and return paths is generated.

[0007] Furthermore, steady-state interval extraction is performed on the time series of the multi-dimensional process parameters to obtain the baseline parameter bands for each process step, including: For each parameter dimension in the multi-dimensional process parameter time series, calculate its mean and variance within the sliding window; A dynamic threshold is set, and when the variance within the sliding window is less than the dynamic threshold, it is determined that the parameters within the sliding window are in a steady state. Consecutive adjacent steady-state windows are merged to form a continuous steady-state data segment, and the maximum and minimum values ​​are extracted from the steady-state data segment; The maximum and minimum values ​​of all parameters corresponding to each process step are floated upward and downward by preset engineering margins, respectively, to form the reference parameter band of the process step.

[0008] Furthermore, root cause correlation analysis is performed on the time series of the fault alarm events to obtain a fault propagation link model, including: Traverse the time sequence of the fault alarm events and extract the accompanying signal status at the time of each fault alarm. The accompanying signal status includes the air pressure value, hydraulic pressure value and servo motor current value at that time. Establish a co-occurrence matrix between fault phenomena and accompanying signal states, and statistically analyze the probability of different fault phenomena occurring simultaneously under the same accompanying signal state. Based on the co-occurrence probability, a causal directed graph is constructed between the fault phenomena, and the arrows in the causal directed graph point from the cause to the effect. The transitive closure operation is performed on the causal directed graph to eliminate redundant paths, resulting in a simplified fault propagation link model.

[0009] Furthermore, a co-occurrence matrix between fault phenomena and accompanying signal states is established, and the probability of different fault phenomena occurring simultaneously under the same accompanying signal state is statistically analyzed, including: Extract all types of fault phenomena that have occurred in the historical records from the time series of the fault alarm events, and extract all recorded types of accompanying signal states; Construct a two-dimensional matrix where the row indices represent different fault phenomenon types and the column indices represent different accompanying signal state types. The number of rows in the matrix is ​​the same as the number of fault phenomenon types, and the number of columns in the matrix is ​​the same as the number of accompanying signal state types. Traverse each historical fault record in the time series of the fault alarm events, identify the specific fault phenomenon type and the specific accompanying signal state type recorded at that time in the historical fault record, and find the corresponding "fault phenomenon-accompanying signal state" cell in the two-dimensional matrix. The count value of the corresponding cell is incremented to record the number of times the "fault phenomenon - accompanying signal state" combination occurs together in the historical record; After traversal, for each row of the matrix, the sum of the cumulative counts of all cells in that row is taken as the total frequency of the corresponding fault phenomenon occurring in that row; The cumulative count recorded in each cell of the matrix is ​​divided by the total frequency of the fault phenomenon corresponding to its row. The result is used as the probability value of the fault phenomenon type represented by the cell in the corresponding accompanying signal state, thereby transforming the counting matrix into a probability matrix.

[0010] Furthermore, the process flow diagram, the baseline parameter band, and the fault propagation link model are input into the system state reconstruction device for collaborative mapping to generate a comprehensive operational feature matrix reflecting the current health status of the production system, including: In the system state reconstruction device, each process node in the process flow diagram is mapped to a state variable; Substitute the upper and lower limits of the reference parameter band into the constraint attribute of the state variable; The root cause nodes in the fault propagation link model are mapped to risk factors of the state variables, and the propagation path is mapped to the influence coefficients between the state variables. Iterate through all state variables, combine their current actual values, constraints, risk factors, and influence coefficients into a multi-dimensional vector, and arrange all multi-dimensional vectors in the order of the process to form the comprehensive operation characteristic matrix.

[0011] Furthermore, based on the comprehensive operational feature matrix, a control logic adaptation network is constructed for dynamic optimization, forming an optimized control strategy set containing control command sequences and logic weight coefficients, including: The comprehensive operational feature matrix is ​​expanded into the input feature vector of the control logic adaptation network; A hidden layer is set in the control logic adaptation network, which is used to learn the nonlinear mapping relationship between the current production condition and the complexity of the control logic. The output layer of the control logic adaptation network generates two sets of data: one set is the specific operation sequence for driving the actuator, defined as the control instruction sequence; the other set is the value for adjusting the priority of soft component contacts and coils in the PLC ladder diagram, defined as the logic weight coefficient. The control command sequences and logic weight coefficients output under different operating conditions are archived to form the optimized control strategy set.

[0012] Furthermore, it also includes: Based on the optimized control strategy set, the abstract control logic weight coefficients are transformed into specific ladder diagram logic blocks or structured text fragments through the PLC instruction generation engine, generating an incremental control program package for the intelligent control system for metal product production. The incremental control program package is sent to the online simulation verification environment of the intelligent control system for metal product production for logical consistency verification. After the verification is passed, the verified incremental control program package is loaded into the running memory of the PLC controller. The step of converting abstract control logic weight coefficients into specific ladder diagram logic blocks or structured text fragments using a PLC instruction generation engine, based on the optimized control strategy set, includes: The logical weight coefficients in the optimized control strategy set are analyzed to identify the interlocking protection logic that needs to be given higher priority. In the ladder logic block, by adjusting the position of the normally open contact of the corresponding auxiliary relay and placing it at the beginning of the logic scan, the purpose of strengthening priority can be achieved. For analog closed-loop control loops that require flexible adjustment, they are converted into structured text fragments, and dynamic adjustment formulas for proportional coefficients and integral time are introduced into the structured text fragments. The adjusted ladder logic blocks and the generated structured text fragments are syntactically compiled to generate the incremental control program package that conforms to the IEC61131-3 standard.

[0013] Furthermore, the step of sending the incremental control program package to the online simulation verification environment of the intelligent control system for metal product manufacturing for logical consistency verification includes: Import the incremental control program package into the online simulation verification environment and load the process formula data of the current batch of products; Start the virtual PLC runtime environment and execute the control instruction sequence in the incremental control program package to simulate the actions of the entire production process; The intermediate variable states in the virtual environment are collected synchronously and compared point by point with the expected states in the comprehensive operational feature matrix. If there is a logical conflict or state deadlock in the comparison results, the variable address of the conflict point is extracted and a logical error report is generated. If the comparison results are completely consistent, the logical consistency check is deemed to have passed.

[0014] Furthermore, after successful verification, loading the verified incremental control program package into the PLC controller's running memory includes: Read the PLC controller's memory allocation table to determine the currently unoccupied program storage space address range; The incremental control program package is decomposed into several functional sub-blocks and data blocks, and the byte length of each functional sub-block and data block is calculated respectively; The functional sub-blocks and data blocks are arranged sequentially according to the starting address of the program storage space address segment, and a blank padding area of ​​a specified length is inserted between adjacent blocks to prevent address overlap. The arranged functional sub-blocks, data blocks, and blank filling areas are written into the running memory of the PLC controller via a bus communication protocol, thus completing the loading process.

[0015] Compared with the prior art, the beneficial effects of the present invention are: Process flow identifiers are traced back to the process path, steady-state intervals are extracted from multi-dimensional process parameter time series, and root cause correlation analysis is performed on fault alarm event time series. The process flow map, benchmark parameter band, and fault propagation link model are input into the system state reconstruction device for collaborative mapping to generate a comprehensive operation feature matrix. Multi-dimensional operation data of the production process are correlated and fused. The representation of the system operation status covers the entire dimensions of process flow, process parameters, and fault propagation. The standardized records after invalid pulse data are removed provide a regular data source for state mapping. The system health status is represented in a structured and quantitative manner in matrix form.

[0016] Based on the comprehensive operational feature matrix, a control logic adaptation network is constructed for dynamic optimization, forming an optimized control strategy set that includes control command sequences and logic weight coefficients. The adaptation process of the control logic is linked in real time with the comprehensive operational status of the system. Control commands are output in sequence form, and the configuration of logic weight coefficients is tailored to the dynamic operational characteristics of the production system. The generation of control strategies is free from the constraints of static presets. The optimization process of the adaptation network directly matches the comprehensive operational characteristics of the system, and the components of the control strategy correspond precisely to the operational status of the production system. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the optimization method for the intelligent control system for metal product manufacturing based on PLC technology described in this invention. Figure 2 A flowchart for performing root cause correlation analysis on the time series of fault alarm events; Figure 3 Heat map of inter-process influence coefficients; Figure 4 A comparison chart of the baseline and actual values ​​of process parameters; Figure 5 A heatmap was used to simulate and verify the logical consistency. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1The system reads raw production operation logs from the PLC controller of the intelligent control system for metal product manufacturing. These logs contain equipment action codes, process parameter setpoints, and processing timing trigger points, forming a raw operation data sequence. The raw operation data sequence undergoes structured cleaning to remove invalid pulse data caused by signal jitter, generating a unified set of production process execution records. This set includes process flow identifiers, multi-dimensional process parameter time series, and fault alarm event time series. Process flow identifiers are traced back to obtain a complete process flow map for the current batch of products. Steady-state interval extraction is performed on the multi-dimensional process parameter time series to obtain baseline parameter bands for each process step. Root cause correlation analysis is performed on the fault alarm event time series to obtain a fault propagation link model. The process flow map, baseline parameter bands, and fault propagation link model are input into a system state reconstruction device for collaborative mapping, generating a comprehensive operation feature matrix reflecting the current health status of the production system. Based on the comprehensive operation feature matrix, a control logic adaptation network is constructed for dynamic optimization, forming an optimized control strategy set containing control command sequences and logic weight coefficients.

[0020] In one embodiment of the present invention, see [reference] Figure 2The equipment number, fixture number, and material batch number are extracted from the process flow identifiers to construct a three-dimensional index table. Each record corresponds to a specific moment of process execution. For example, a record "Equipment#M03-Fixture#J12-Batch#LOT2025001" indicates that on equipment M03, fixture J12 is processing material with batch number LOT2025001. Based on the three-dimensional index table, sensor feedback signals deployed at the loading, stamping, welding, and unloading stations in the intelligent control system for metal product manufacturing are retrieved. These signals include the on / off status of photoelectric switches, the trigger time of proximity sensors, and RFID reader reading records. By integrating the timestamps of these signals with the entries in the three-dimensional index table, the physical movement trajectory of material LOT2025001 from "loading-positioning-stamping-welding-cooling-unloading" can be reconstructed. In practice, the reconstructed physical movement trajectory is compared with the preset standard process route "loading-positioning-stamping-welding-unloading". When the physical movement trajectory encounters a "cooling" stop node not defined in the standard route after the "welding" process, the system marks "cooling" as an abnormal flow node deviating from the standard route. It can be understood that the standard process route is defined based on an ideal production model, while the physical movement trajectory reflects the actual equipment scheduling and material flow. By comparing the two, path deviations caused by equipment failure, buffer scheduling, or process adjustments can be identified. The normal flow nodes "loading", "positioning", "stamping", "welding", and "unloading" are connected to the identified abnormal flow node "cooling" through directed edges. The direction of the directed edge represents the actual flow direction of the material, for example, from "welding" to "cooling", and then from "cooling" to "unloading". This generates a complete process flow diagram that includes branch paths and return paths. The "cooling" node and its associated edges constitute a branch path. If the material returns from the "welding" station to the "stamping" station for rework due to quality inspection, a return path from "welding" to "stamping" will be generated.

[0021] In some embodiments, steady-state interval extraction is performed for each parameter dimension in the multi-dimensional process parameter time series. Taking the "welding current" parameter dimension of the welding process as an example, its time series is a series of values ​​recorded according to the sampling period. For the "welding current" time series, the mean and variance within its sliding window are calculated. The length of the sliding window is configurable, for example, set to include 100 consecutive sampling points. A dynamic threshold is set. The dynamic threshold is not a fixed value, but a baseline adaptively calculated based on the long-term statistical variance of the "welding current" parameter throughout the entire historical production batch. When the variance within the sliding window is less than the dynamic threshold, the "welding current" parameter within the sliding window is determined to be in a steady state. It can be understood that variance is an indicator of data volatility. A small variance within the sliding window means that the process parameter is stable within that time segment, which meets the expectation of steady-state production operation. Consecutive adjacent steady-state windows are merged to form a continuous steady-state data segment. For example, if windows with sampling points 101-200 and 201-300 are both determined to be in steady state and are adjacent, they are merged into a single steady-state data segment from sampling point 101 to sampling point 300. The maximum and minimum values ​​of the welding current are extracted from this steady-state data segment; for example, the maximum value is 210.5 amperes and the minimum value is 209.2 amperes. The maximum and minimum values ​​of all parameters corresponding to each process step are floated upwards and downwards by preset engineering margins to form the baseline parameter band for the process step. For example, in the welding step, in addition to "welding current," there are parameters such as "wire feed speed" and "shielding gas flow rate." Their steady-state maximum and minimum values ​​are extracted. The steady-state maximum value of "welding current" (210.5 amps) is floated upwards by 1.5 amps to obtain an upper limit of 212.0 amps, and the steady-state minimum value (209.2 amps) is floated downwards by 1.5 amps to obtain a lower limit of 207.7 amps, thus forming the baseline parameter band for "welding current" [207.7A, 212.0A]. The set of baseline parameter bands for all parameters in the welding step constitutes the baseline parameter band for that process step. In specific implementation, the steady-state interval extraction process can be formalized, and the formula for calculating the sliding window variance is:

[0022] in: This represents the variance of the parameters within the current sliding window. Indicates the size of the sliding window. Indicates the first in the window The parameter values ​​of each sampling point This represents the mean of the parameters within the sliding window. When Less than the dynamic threshold When this window is identified as a steady-state window, consecutive steady-state windows are merged to extract the extreme values ​​of the steady-state data segment.

[0023] In one embodiment of the present invention, the fault alarm event time sequence is traversed to extract the accompanying signal status at the time of each fault alarm. The accompanying signal status includes the hydraulic value, air pressure value and servo motor current value at that time. For example, a record is "Fault code: ALM-101, description: mold clamping timeout, accompanying signal: hydraulic = 8.5MPa, air pressure = 0.6MPa, current = 12.3A".

[0024] In practical implementation, a co-occurrence matrix between fault phenomena and accompanying signal states is established, and the probability of different fault phenomena occurring simultaneously under the same accompanying signal state is statistically analyzed. All fault phenomenon types that have occurred in historical records are extracted from the fault alarm event time series. For example, three fault phenomenon types are extracted: "Mold clamping timeout (ALM-101)," "Servo axis positioning deviation (ALM-202)," and "Low air pressure alarm (ALM-303)." All recorded accompanying signal state types are also extracted, such as "Hydraulic < 9.0 MPa," "Hydraulic >= 9.0 MPa," "Air pressure < 0.65 MPa," "Air pressure >= 0.65 MPa," "Current < 10.0 A," and "Current >= 10.0 A." A two-dimensional matrix is ​​constructed, where the row index represents different fault phenomenon types, and the column index represents different accompanying signal state types. The number of rows in the matrix is ​​the same as the number of fault phenomenon types, and the number of columns in the matrix is ​​the same as the number of accompanying signal state types. The process iterates through each historical fault record in the time series of fault alarm events, identifying the specific fault phenomenon type and the corresponding accompanying signal state type recorded at that time, and finding the corresponding "fault phenomenon-accompanying signal state" cell in a two-dimensional matrix. The count value of the corresponding cell is accumulated, recording the number of times the "fault phenomenon-accompanying signal state" combination occurs in the historical records. After the traversal is complete, for each row of the matrix, the sum of the accumulated counts of all cells in the row is taken as the total frequency of the corresponding fault phenomenon in that row. The accumulated count recorded in each cell is divided by the total frequency of the corresponding fault phenomenon in its row, and the result is taken as the probability value of the fault phenomenon type represented by the cell occurring in its corresponding accompanying signal state, thus transforming the counting matrix into a probability matrix.

[0025] Based on the co-occurrence probabilities, a causal directed graph is constructed between fault phenomena, with arrows pointing from cause to effect. The causal strength between fault phenomena can be quantified using conditional probability; the probability of fault phenomenon B occurring given that fault phenomenon A has already occurred can be calculated. In some embodiments, a causal strength metric is defined. The causal strength metric pointing from fault phenomenon A to fault phenomenon B is... The calculation method is as follows:

[0026] in: This represents the causal strength measure from fault phenomenon A to fault phenomenon B. This represents the probability of failure B occurring given that failure A has already occurred. This represents the marginal probability of fault phenomenon B occurring. If If the value exceeds a preset causal threshold, a directed edge is created in the causal directed graph from node A to node B. For example, the calculated edge "hydraulic pressure < 9.0 MPa" points to "mold clamping timeout (ALM-101)". If the value is high and exceeds the threshold, an edge is created in the causal directed graph from "hydraulic pressure < 9.0 MPa" to "mold clamping timeout (ALM-101)".

[0027] In one embodiment of the present invention, in the system state reconstruction device, each process node in the process flow diagram is mapped to a state variable. For example, the process flow diagram contains five nodes: "loading", "stamping", "welding", "cooling", and "unloading", which are mapped to state variables S1, S2, S3, S4, and S5, respectively. The upper and lower limits of the reference parameter band are substituted into the constraint attribute of the state variable. Taking the state variable S3 "welding" as an example, its constraint attribute is set to "lower limit of welding current reference parameter band 207.7A, upper limit of welding current reference parameter band 212.0A; lower limit of wire feeding speed reference parameter band 8.0m / min, upper limit of wire feeding speed reference parameter band 9.5m / min". In the fault propagation link model, the root cause node is mapped to a risk factor of the state variable, and the propagation path is mapped to the influence coefficient between the state variables. For example, in the fault propagation link model, "air pressure < 0.65MPa" is identified as the root cause node of "abnormal welding spatter", and "abnormal welding spatter" may lead to "substandard weld quality". In the system state reconstruction device, "air pressure < 0.65MPa" is mapped to a risk factor of the state variable S3 "welding", and the propagation path strength from "abnormal welding spatter" to "substandard weld quality" is mapped to the influence coefficient of the state variable S3 "welding" on the state variable S5 "material cutting" (assuming that "material cutting" includes quality inspection). Iterate through all state variables, combining their current actual values, constraints, risk factors, and influence coefficients into a multidimensional vector. For example, the multidimensional vector of state variable S3 is represented as ["Welding", 210.5, 8.5, 207.7, 212.0, 8.0, 9.5, Risk Factor: {"Low Gas Pressure": 0.8}, Influence Coefficient: {"For S5": 0.6}], where "Welding" is the state name, 210.5 is the current actual value of the welding current, 8.5 is the current actual value of the wire feed speed, 207.7 and 212.0 are the lower and upper limits of the welding current reference parameter, 8.0 and 9.5 are the lower and upper limits of the wire feed speed reference parameter, the risk factor list includes "Low Gas Pressure" with a confidence level of 0.8, and the influence coefficient list includes the influence coefficient of 0.6 on state variable S5. All multidimensional vectors are arranged in the order of the operation process to form a comprehensive operational characteristic matrix, with each row of the matrix corresponding to a multidimensional vector of a state variable.

[0028] In practical implementation, a control logic adaptation network is constructed based on the comprehensive operational feature matrix for dynamic optimization, forming an optimized control strategy set containing control command sequences and logic weight coefficients. The comprehensive operational feature matrix is ​​expanded into an input feature vector for the control logic adaptation network. The input feature vector is a flattened concatenation of the actual values ​​of all state variables, constraint boundaries, risk factor confidence levels, and influence coefficients in the comprehensive operational feature matrix. Hidden layers are set in the control logic adaptation network. These hidden layers are used to learn the nonlinear mapping relationship between the current production conditions and the complexity of the control logic. The hidden layers can contain multiple fully connected layers and employ nonlinear activation functions. The output layer of the control logic adapter network generates two sets of data. One set is the specific operation sequence used to drive the actuator, defined as a control instruction sequence. For example, it outputs a sequence containing instructions such as "start the cooling pump," "adjust the welding current to 209A," "delay for 2 seconds," and "start the wire feeder." The other set is the numerical value used to adjust the priority of soft component contacts and coils in the PLC ladder diagram, defined as a logic weight coefficient. For example, it outputs a set of values ​​[1.2, 0.8, 1.5], corresponding to the priority adjustment coefficients of "emergency stop interlock logic," "air pressure protection logic," and "temperature control logic" in the ladder diagram, respectively. In some embodiments, the hidden layer calculation of the control logic adapter network can be described as the activation value calculation process of the hidden layer nodes. Activation value of each node Calculated in the following way:

[0029] in: Indicates the hidden layer number 1 The activation value of each node. This represents a non-linear activation function, such as the ReLU function. Represents the input feature vector Dimensions Represents the input feature vector The One portion, Indicates the number of input layers. The node is connected to the hidden layer. The connection weight of each node Indicates the hidden layer number 1 The bias term of each node. The control logic adaptation network learns the weights through training. and bias Parameters, etc. It can be understood that the hidden layer extracts deep features representing the complexity of the current production condition by performing a nonlinear transformation on the input feature vector. The output layer generates a control command sequence and logical weight coefficients based on the activation values ​​of the hidden layer. The control command sequence can be a one-hot encoded sequence, and the logical weight coefficients can be a real number vector. The output control command sequences and logical weight coefficients under different operating conditions are archived to form an optimized control strategy set.

[0030] See Figure 3 This is a heatmap of inter-process influence coefficients, focusing on the system state reconstruction stage of the intelligent control system for metal product manufacturing. Its core purpose is to visualize the strength of influence relationships between various process nodes. It intuitively presents the coupling relationships of the entire process, clearly identifying welding (S3) as the core source of influence, providing a quantitative basis for system state reconstruction and control logic optimization. The welding (S3) process is under key control, with the reference bands for parameters such as welding current and wire feed speed narrowed to reduce the impact of their fluctuations on subsequent cooling and blanking stages. For high-influence coefficient (0.6-0.7) links, a welding-blanking process linkage monitoring system is established, triggering buffer adjustment strategies in advance when risk factors occur in the welding process. For medium- and low-influence links such as stamping-welding and cooling-blanking, standardized process parameters are used to reduce the transmission of deviations between processes.

[0031] In one embodiment of the present invention, based on an optimized control strategy set, the abstract control logic weight coefficients are transformed into specific ladder diagram logic blocks or structured text fragments through a PLC instruction generation engine, generating an incremental control program package for an intelligent control system for metal product manufacturing. The logic weight coefficients in the optimized control strategy set are analyzed to identify interlocking protection logic that requires increased priority. For example, if the logic weight coefficient vector output by the optimized control strategy set is [1.2, 0.8, 1.5], the third coefficient 1.5 corresponds to "temperature control logic," and a value greater than 1.0 indicates that the priority of this logic needs to be increased. In the ladder diagram logic block, the normally open contact position of the corresponding auxiliary relay is adjusted to place it at the beginning of the logic scan to achieve the purpose of increasing priority. It can be understood that in the scanning execution mechanism of the PLC ladder diagram, logic at the beginning is scanned and evaluated first. For the auxiliary relay M205 associated with the "temperature control logic", the PLC instruction generation engine locates the normally open contact of M205 in its original ladder diagram network and moves this contact from its original network branch to a more forward position in the main control logic network, such as moving it to the beginning of the main control contactor coil control branch, to ensure that the temperature condition is checked first before any main equipment operation. In some embodiments, refer to Table 1 for the logic weight coefficients that need to be parsed and their corresponding relationships.

[0032] Table 1: Examples of Logical Weight Coefficients and Corresponding Control Logic Modules

[0033] For analog closed-loop control loops requiring flexible adjustment, they are converted into structured text snippets, and dynamic adjustment formulas for the proportional gain and integral time are introduced into these snippets. Taking closed-loop control of welding current as an example, the optimized control strategy set may indicate the need for dynamic adjustment of the proportional gain under specific operating conditions. In practice, the PID control logic, originally implemented as function blocks in a ladder diagram, is converted into program snippets described in Structured Text (ST) language. Dynamic adjustment formulas for the proportional gain Kp and integral time Ti are introduced into these structured text snippets, and these formulas can be updated based on real-time state variables. For example, a dynamic adjustment formula based on state variable deviation can be designed as follows:

[0034] in: This indicates the current scaling factor. The reference value representing the proportionality coefficient. This indicates that the sensitivity coefficient is being adjusted. This represents the real-time deviation between the setpoint and the process variable. This is understandable. The values ​​of the state variables in the comprehensive operating characteristic matrix can be compared with the set values ​​of the baseline parameter band. The integral time Ti can also be dynamically adjusted using a similar linear or nonlinear function. The PLC instruction generation engine performs syntax compilation on the adjusted ladder logic blocks and the generated structured text fragments to generate an incremental control package conforming to the IEC61131-3 standard. The incremental control package includes the modified ladder network, the newly added structured text program organization unit (POU), and the interface declarations of both.

[0035] See Figure 4 This is a comparison chart of the baseline and actual values ​​of process parameters. It displays the lower limit, actual value, and upper limit of five core process parameters in the production of metal products, corresponding to the comparison between the baseline parameter extraction and actual operating conditions during the system state reconstruction phase. The actual values ​​of all parameters fall within the baseline parameter range formed by the lower and upper limits, indicating that the current production conditions meet the preset process requirements and there is no risk of parameter exceeding limits. The temperature baseline range is the widest, with the actual value of 720 closer to the upper limit, requiring attention to the potential risk of thermal deformation or changes in material properties due to further temperature increases. The pressure baseline range is the narrowest, with the actual value of 8 in the middle of the range, indicating the highest steady-state control accuracy and best process stability for this parameter. The actual values ​​of speed, current, and voltage are all biased towards the middle of their ranges, demonstrating good steady-state control effects; the actual temperature value is biased towards the upper limit, making it a key focus for subsequent control optimization.

[0036] In one embodiment of the present invention, the process of sending the incremental control package to the online simulation verification environment of the intelligent control system for metal product manufacturing for logical consistency verification includes several steps. The incremental control package is imported into the online simulation verification environment, and the process formula data for the current batch of products is loaded, for example, the processing parameters of the workpiece with batch number LOT2025033 are loaded, including a welding current setting of 210 amperes, a wire feed speed setting of 9.0 meters per minute, and a welding time of 1.5 seconds. The virtual PLC runtime environment is started. The virtual PLC runtime environment is a software simulator running on an industrial computer that can simulate the instruction execution, I / O scanning, and logic operation cycles of a real PLC. The control instruction sequence in the incremental control package is executed to simulate the actions of the entire production process, such as simulating the extension of the loading cylinder, clamping of the fixture, positioning of the servo motor, starting of the welding power supply, and gripping of the unloading robot. The intermediate variable states in the virtual environment are synchronously collected. These intermediate variable states include the real-time values ​​of the internal auxiliary relay M area and the data register D area. These values ​​are then compared point-by-point with the expected states in the comprehensive operational feature matrix. The expected states in the comprehensive operational feature matrix are derived from the previously analyzed steady-state parameters and process logic. It can be understood that the intermediate variable states in the virtual environment reflect the actual logical results of the incremental control program package during simulation execution, while the expected states in the comprehensive operational feature matrix represent the logical states that the system should have under healthy operating conditions. If logical conflicts or state deadlocks are found in the comparison results, the variable addresses of the conflict points are extracted, and a logic error report is generated. For example, if the comparison finds that the incremental control program package causes the internal flag M1023 to fail to reset in the loop, triggering a state deadlock, the online simulation verification environment extracts the variable address M1023 and records its state change timing, generating a logic error report containing a conflict description, variable address, and timing information. If the comparison results are completely consistent, the logical consistency check is deemed to have passed. For example, the execution results of all key steps in the virtual environment, such as the setting time of the "welding completed" flag M200 and the fluctuation curve of the actual value of welding current D150, are consistent with the expected state in the comprehensive operation feature matrix.

[0037] In some embodiments, after successful verification, the verified incremental control program package is loaded into the PLC controller's runtime memory. The PLC controller's runtime memory allocation table is read to determine the currently unoccupied program storage address ranges. This table, provided by the PLC's programming software or system maintenance tools, marks used and free memory blocks. For example, the PLC controller's runtime memory allocation table might show that address range 0x1000 to 0x1FFF is occupied by the main program, and address range 0x2000 to 0x2FFF is free. The incremental control program package is then broken down into several functional sub-blocks and data blocks. Functional sub-blocks are ladder diagram networks or structured text program organization units (POUs) that implement different control functions, while data blocks contain the global variables and initial values ​​defined within them. The byte length of each functional sub-block and data block is calculated separately; for example, the functional sub-block "PID_Temp_Control" is calculated to be 512 bytes long, and the data block "Global_Vars" is calculated to be 256 bytes long. Functional sub-blocks and data blocks are arranged sequentially according to the starting address of the program's memory space address segment. For example, starting from the starting address of the free address segment at 0x2000, the functional sub-block "PID_Temp_Control" (occupying addresses 0x2000-0x21FF), the functional sub-block "Safety_Interlock" (occupying addresses 0x2200-0x23FF), and the data block "Global_Vars" (occupying addresses 0x2400-0x24FF) are stored sequentially. A blank padding area of ​​a specified length is inserted between adjacent blocks to prevent address overlap. The content of the blank padding area is usually all zeros or no operation instructions. In practical implementation, address allocation needs to ensure that there is no overlap between blocks or between a block and an already occupied memory area, which requires calculating the starting address offset. The starting address of the next functional sub-block or data block... The calculation method is as follows:

[0038] in: This indicates the starting address of the next functional sub-block or data block in runtime memory. This indicates the starting address of the current functional sub-block or data block. Indicates the byte length of the current functional sub-block or data block. Indicates the length of the specified blank padding area.

[0039] See Figure 5This is a heatmap comparing the consistency of simulation verification logic, used to quantitatively evaluate the degree of matching between the incremental control package and the expected state of the comprehensive operating characteristic matrix under different production batches. The consistency coefficients of all batches are ≥0.97, indicating that the incremental control package has high logic stability in multi-batch simulations and no serious conflicts or deadlocks have occurred. In batch 3, the consistency of M200 (welding completed) is 0.97, the lowest value in the entire graph, requiring investigation of the timing logic of the welding completed flag. In batch 2, the consistency of M300 (material unloading) is 0.97, requiring verification of whether the material unloading trigger condition matches the expectation. For points M200 and M300, supplementary virtual environment variable timing logs should be added to locate the specific time points of inconsistency. For points with large fluctuations between batches, the timing thresholds of the control logic should be optimized to improve cross-batch stability.

[0040] The above embodiments are only used to illustrate the technical methods 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 methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. An optimization method for an intelligent control system for metal product manufacturing based on PLC technology, characterized in that, The method includes: The original production operation log containing equipment action codes, process parameter settings, and processing timing trigger points is read from the PLC controller of the intelligent control system for metal product production to form the original operation data sequence. The original operating data sequence is subjected to structured cleaning to remove invalid pulse data caused by signal jitter, and a production process execution record set with a unified format is generated. The production process execution record set includes process flow identifiers, multi-dimensional process parameter time series, and fault alarm event time series. Process path tracing is performed on the process flow identifier to obtain the complete process flow map of the current batch of products. Steady-state interval extraction is performed on the time series of the multi-dimensional process parameters to obtain the baseline parameter band of each process link. Root cause correlation analysis is performed on the time series of the fault alarm events to obtain the fault propagation link model. The process flow diagram, the baseline parameter band, and the fault propagation link model are input into the system state reconstruction device for collaborative mapping to generate a comprehensive operating feature matrix that reflects the current health status of the production system. Based on the comprehensive operational feature matrix, a control logic adaptation network is constructed for dynamic optimization, forming an optimized control strategy set containing control command sequences and logic weight coefficients, including: The comprehensive operational feature matrix is ​​expanded into the input feature vector of the control logic adaptation network; A hidden layer is set in the control logic adaptation network, which is used to learn the nonlinear mapping relationship between the current production condition and the complexity of the control logic. The output layer of the control logic adaptation network generates two sets of data: one set is the specific operation sequence for driving the actuator, defined as the control instruction sequence; the other set is the value for adjusting the priority of soft component contacts and coils in the PLC ladder diagram, defined as the logic weight coefficient. The control command sequences and logical weight coefficients output under different operating conditions are archived to form the optimized control strategy set. The root cause correlation analysis performed on the time series of the fault alarm events to obtain the fault propagation link model includes: Traverse the time sequence of the fault alarm events and extract the accompanying signal status at the time of each fault alarm. The accompanying signal status includes the air pressure value, hydraulic pressure value and servo motor current value at that time. Establish a co-occurrence matrix between fault phenomena and accompanying signal states, and statistically analyze the probability of different fault phenomena occurring simultaneously under the same accompanying signal state, i.e., the co-occurrence probability. Based on the co-occurrence probability, a causal directed graph is constructed between the fault phenomena, and the arrows in the causal directed graph point from the cause to the effect. The transitive closure operation is performed on the causal directed graph to eliminate redundant paths, resulting in a simplified fault propagation link model.

2. The optimization method for the intelligent control system for metal product manufacturing based on PLC technology according to claim 1, characterized in that, By performing process path tracing on the process flow identifier, a complete process flow diagram of the current batch of products is obtained, including: Extract the equipment number, fixture number, and material batch number from the process flow identifier, and construct a three-dimensional index relationship table; Based on the three-dimensional index relationship table, the sensor feedback signals of each workstation in the intelligent control system for metal product production are retrieved to reconstruct the physical movement trajectory of the material between each workstation. The physical movement trajectory is compared with the preset standard process route, and all abnormal flow nodes that deviate from the standard route are marked. By connecting normal flow nodes and abnormal flow nodes with directed edges, a complete process flow graph containing branch paths and return paths is generated.

3. The optimization method for the intelligent control system for metal product manufacturing based on PLC technology according to claim 2, characterized in that, Steady-state interval extraction is performed on the time series of the multi-dimensional process parameters to obtain the baseline parameter bands for each process step, including: For each parameter dimension in the multi-dimensional process parameter time series, calculate its mean and variance within the sliding window; A dynamic threshold is set, and when the variance within the sliding window is less than the dynamic threshold, it is determined that the parameters within the sliding window are in a steady state. Consecutive adjacent steady-state windows are merged to form a continuous steady-state data segment, and the maximum and minimum values ​​are extracted from the steady-state data segment; The maximum and minimum values ​​of all parameters corresponding to each process step are floated upward and downward by preset engineering margins, respectively, to form the reference parameter band of the process step.

4. The optimization method for the intelligent control system for metal product manufacturing based on PLC technology according to claim 3, characterized in that, Establish a co-occurrence matrix between fault phenomena and accompanying signal states, and statistically analyze the probability of different fault phenomena occurring simultaneously under the same accompanying signal state, including: Extract all types of fault phenomena that have occurred in the historical records from the time series of the fault alarm events, and extract all recorded types of accompanying signal states; Construct a two-dimensional matrix where the row indices represent different fault phenomenon types and the column indices represent different accompanying signal state types. The number of rows in the matrix is ​​the same as the number of fault phenomenon types, and the number of columns in the matrix is ​​the same as the number of accompanying signal state types. Traverse each historical fault record in the time series of the fault alarm events, identify the specific fault phenomenon type and the specific accompanying signal state type recorded at that time in the historical fault record, and find the corresponding "fault phenomenon-accompanying signal state" cell in the two-dimensional matrix; The count value of the corresponding cell is incremented to record the number of times the "fault phenomenon - accompanying signal state" combination occurs together in the historical record; After traversal, for each row of the matrix, the sum of the cumulative counts of all cells in that row is taken as the total frequency of the corresponding fault phenomenon occurring in that row; The cumulative count recorded in each cell of the matrix is ​​divided by the total frequency of the fault phenomenon corresponding to its row. The result is used as the probability value of the fault phenomenon type represented by the cell in the corresponding accompanying signal state, thereby transforming the counting matrix into a probability matrix.

5. The optimization method for the intelligent control system for metal product manufacturing based on PLC technology according to claim 4, characterized in that, The process flow diagram, the baseline parameter band, and the fault propagation link model are input into the system state reconstruction device for collaborative mapping to generate a comprehensive operational feature matrix reflecting the current health status of the production system, including: In the system state reconstruction device, each process node in the process flow diagram is mapped to a state variable; Substitute the upper and lower limits of the reference parameter band into the constraint attribute of the state variable; The root cause nodes in the fault propagation link model are mapped to risk factors of the state variables, and the propagation path is mapped to the influence coefficients between the state variables. Traverse all state variables, combine the current actual values, constraints, risk factors and influence coefficients of all state variables into a multi-dimensional vector, and arrange all multi-dimensional vectors in the order of the process to form the comprehensive operation characteristic matrix.

6. The optimization method for the intelligent control system for metal product manufacturing based on PLC technology according to claim 5, characterized in that, Also includes: Based on the optimized control strategy set, the abstract control logic weight coefficients are transformed into specific ladder diagram logic blocks or structured text fragments through the PLC instruction generation engine, generating an incremental control program package for the intelligent control system for metal product production. The incremental control program package is sent to the online simulation verification environment of the intelligent control system for metal product production for logical consistency verification. After the verification is passed, the verified incremental control program package is loaded into the running memory of the PLC controller. The step of converting abstract control logic weight coefficients into specific ladder diagram logic blocks or structured text fragments using a PLC instruction generation engine, based on the optimized control strategy set, includes: The logical weight coefficients in the optimized control strategy set are analyzed to identify the interlocking protection logic that needs to be given higher priority. In the ladder logic block, by adjusting the position of the normally open contact of the corresponding auxiliary relay and placing it at the beginning of the logic scan, the purpose of strengthening priority can be achieved. For analog closed-loop control loops that require flexible adjustment, they are converted into structured text fragments, and dynamic adjustment formulas for proportional coefficients and integral time are introduced into the structured text fragments. The adjusted ladder logic blocks and the generated structured text fragments are syntactically compiled to generate the incremental control program package that conforms to the IEC61131-3 standard.

7. The optimization method for the intelligent control system for metal product manufacturing based on PLC technology according to claim 6, characterized in that, The step of sending the incremental control program package to the online simulation verification environment of the intelligent control system for metal product manufacturing for logical consistency verification includes: Import the incremental control program package into the online simulation verification environment and load the process formula data of the current batch of products; Start the virtual PLC runtime environment and execute the control instruction sequence in the incremental control program package to simulate the actions of the entire production process; The intermediate variable states in the virtual environment are collected synchronously and compared point by point with the expected states in the comprehensive operational feature matrix. If there is a logical conflict or state deadlock in the comparison results, the variable address of the conflict point is extracted and a logical error report is generated. If the comparison results are completely consistent, the logical consistency check is deemed to have passed.

8. The optimization method for the intelligent control system for metal product manufacturing based on PLC technology according to claim 7, characterized in that, After successful verification, the incremental control program package that has passed verification is loaded into the running memory of the PLC controller, including: Read the PLC controller's memory allocation table to determine the currently unoccupied program storage space address range; The incremental control program package is decomposed into several functional sub-blocks and data blocks, and the byte length of each functional sub-block and data block is calculated respectively; The functional sub-blocks and data blocks are arranged sequentially according to the starting address of the program storage space address segment, and a blank padding area of ​​a specified length is inserted between adjacent blocks to prevent address overlap. The arranged functional sub-blocks, data blocks, and blank filling areas are written into the running memory of the PLC controller via a bus communication protocol, thus completing the loading process.

Images