Intelligent flow scheduling method for high-performance oil way of sheet metal pipe network

Abstract

The invention relates to the technical field of data management, in particular to an intelligent flow scheduling method for a high-performance oil way of a metal plate pipe network. The specific implementation process comprises the following steps: establishing a distributed intelligent executor and a resource optimization model based on a sheet metal pipe network comprising a schedulable unit and production management information comprising a historical performance baseline and a production plan; the distributed intelligent executor obtains and evaluates the operation data of the schedulable unit and generates real-time state data; combining the real-time state data and the historical performance baseline to analyze the drift trend and generate performance degradation data; based on the production plan and the performance degradation data, quantifying the comprehensive operation cost as a target function to calculate and generate a resource allocation strategy and an operation instruction; and the distributed intelligent executor combines the operation instruction, the real-time state data and the performance degradation data to execute dynamic compensation management to generate a scheduling instruction. According to the method, resource allocation is optimized by evaluating the real-time and long-term states of the equipment, so that risk avoidance and cost dynamic optimization are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of data management, and in particular to a flow intelligent scheduling method for a high-performance oil circuit in a sheet metal pipe network. Background Art

[0002] In sheet metal processing production operations management, the operational efficiency of production equipment is key to achieving on-time delivery and cost control. Oil system flow scheduling impacts production efficiency, machining accuracy, and equipment lifespan. Ensuring cost-effective and forward-looking flow scheduling is a prerequisite for achieving lean production and intelligent operations management.

[0003] Existing operations management and scheduling methods present several challenges. For one thing, traditional scheduling methods rely on pre-set models based on fixed parameters. These methods are unable to dynamically adjust resource allocation when oil parameters fluctuate during production, resulting in slow system response and energy waste. Furthermore, applied data-driven intelligent algorithms are typically static strategies that lack the ability to adapt online to long-term changes in equipment performance, making them unable to meet the continuous optimization needs of resource scheduling and risk avoidance in industrial scenarios.

[0004] In summary, existing technologies are unable to cope with both short-term operational changes and long-term performance changes, resulting in low overall efficiency. To address this issue, a flow intelligent scheduling method for high-performance oil circuits in sheet metal pipe networks is proposed. Summary of the Invention

[0005] The present invention aims to provide a method for intelligent flow scheduling of high-performance oil circuits in sheet metal pipe networks for intelligent data management. To address the problems of the prior art, the present invention first establishes a distributed intelligent executor and a resource optimization model based on a sheet metal pipe network containing schedulable units and production management information. The production management information includes historical performance baselines and production plans. Next, the present invention obtains the operational data of the schedulable units and generates real-time status data based on the operational data through the distributed intelligent executor. The current output state is calculated for preset test instructions, and the drift trend of the current output state from the historical performance baseline is analyzed to generate performance degradation data. Secondly, based on the production plan and the performance degradation data, the comprehensive operating cost is quantified and generated. Using the comprehensive operating cost as the objective function, the resource optimization model calculates and generates a resource allocation strategy. Job instructions are then generated based on the resource allocation strategy. Finally, the distributed intelligent executor combines the job instructions, real-time status data, and performance degradation data to perform dynamic compensation management to generate scheduling instructions. The schedulable units dynamically track the job instructions based on the scheduling instructions. By predicting the load requirements of future production tasks and the evolution trend of their own performance status, the present invention achieves a collaborative scheduling effect of proactive risk avoidance and dynamic cost optimization.

[0006] To achieve the above object, the present application provides the following technical solutions: A flow intelligent scheduling method of a high-performance oil circuit of a sheet metal pipe network, comprising: Based on the sheet metal pipe network containing schedulable units and production management information, a distributed intelligent executor and a resource optimization model are established; the production management information contains historical performance baselines and production plans; Obtain operation data of the schedulable units, the distributed intelligent executor evaluates the operation data to generate real-time state data; calculate the current output state according to a preset test instruction, analyze the drift trend of the current output state deviating from the historical performance baseline, and generate performance degradation data of the schedulable units; Based on the production plan and the performance degradation data, the comprehensive operation cost is quantitatively generated, the comprehensive operation cost is taken as a target function, the resource optimization model is calculated to generate a resource allocation strategy, and a job instruction is generated based on the resource allocation strategy; The distributed intelligent executor combines the job instruction, real-time state data and performance degradation data to execute dynamic compensation management to generate a scheduling instruction; the schedulable unit dynamically tracks the job instruction based on the scheduling instruction.

[0007] Preferably, the distributed intelligent executor contains a state evaluation unit and a performance trend prediction unit; the state evaluation unit contains an information network model, receives the operation data to generate real-time state data, and the information network model contains a fluid state model; the performance trend prediction unit contains an online learning algorithm, analyzes the output state of the preset test job instruction and the historical performance baseline to generate performance degradation data; based on the job instruction and by fusing the real-time state data and the performance degradation data, the dynamic compensation management is executed to calculate the final control output.

[0008] Preferably, the resource optimization model is constructed based on a demand prediction model and an optimization algorithm, and is realized through a data integration, optimization decision and instruction generation stage; the data integration stage receives and integrates the production plan and the performance degradation data to generate a global decision view containing pipe network state and future task load; the optimization decision stage executes the forward-looking optimization solution based on the global decision view to generate the resource allocation strategy, and the resource allocation strategy contains the running role and the work task of the schedulable unit; the instruction generation stage parses the running role and the work task into a job instruction containing timing and physical parameters.

[0009] Preferably, the historical performance baseline is a mapping relationship between the baseline job and the required resource consumption when the schedulable unit is in a reference state.

[0010] Preferably, the process of generating real-time state data comprises: taking the operation data as input of the information network model, the operation data comprising temperature data and pressure data, the information network model performing nonlinear transformation on the temperature data and pressure data to generate preliminary inference values of real-time state data; the real-time state data comprising oil viscosity and density; calculating reference values of real-time state data through a fluid state model, the fluid state model comprising a viscosity-temperature relationship model and a density-temperature-pressure relationship model, comparing the preliminary inference values with the reference values of real-time state data and calculating differences, determining weights based on the differences, and obtaining final state data by weighted summation of the preliminary inference values and the reference values of real-time state data.

[0011] Preferably, the process of generating performance degradation data comprises: calculating a sequence signal and obtaining a required current output state based on a preset test instruction; comparing the current output state with a historical performance baseline to generate a real-time deviation value sequence; filtering out random noise of the real-time deviation value sequence to obtain a smooth curve, and extracting a dynamic drift trend signal by applying linear regression trend analysis to the smooth curve; extracting feature parameters to generate a feature vector based on the dynamic drift trend signal, the feature parameters comprising average bias, response delay, overshoot and oscillation amplitude, and drift rate; normalizing and weightedly summing the feature parameters by using a mapping model in the online learning algorithm to generate a current performance degradation index and a future degradation rate, matching the feature parameters with a preset fault rule library to identify a potential fault signature, calculating a predicted remaining useful life based on the current performance degradation index, the future degradation rate, and a preset failure threshold; and integrating the current performance degradation index, the future degradation rate, the potential fault signature, and the predicted remaining useful life into performance degradation data output.

[0012] Preferably, the process of generating a resource allocation strategy comprises: parsing the production plan to generate specific process steps, obtaining a hydraulic demand curve of the specific process steps, generating a time-series load demand sequence and a predicted total energy consumption demand based on the hydraulic demand curve; combining the predicted total energy consumption demand and the performance degradation data to quantitatively evaluate performance states, operation energy efficiency, and fault risks of the schedulable units, determining resource loss costs, energy costs, and risk costs, and weightedly summing the resource loss costs, the energy costs, and the risk costs to generate a quantitative comprehensive operation cost; taking satisfaction of the time-series load demand sequence by the total output of the schedulable units as a constraint to construct an objective function, the objective function taking minimization of the comprehensive operation cost as an optimization target, and solving the objective function to generate the resource allocation strategy.

[0013] Preferably, the execution of dynamic compensation management to generate scheduling instructions includes: parsing the job instructions to obtain the target trajectory, inputting the target trajectory into a simplified dynamic model to generate an initial control output sequence, wherein the simplified dynamic model includes the state parameters and response characteristics of the schedulable unit in the reference state; based on the deviation between the real-time state data and the reference state system data, performing feedforward compensation to obtain a control output sequence after feedforward compensation; utilizing the potential fault signature of the performance degradation data to perform adaptive feedback correction on the control output sequence after feedforward compensation, calculating the final control output sequence, and generating scheduling instructions.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention utilizes an online learning algorithm and a preset fault rule library to perform in-depth quantification and pattern matching on characteristic parameters extracted based on the deviation between the current output state of the schedulable unit and the historical performance baseline. In this way, the current real-time operating status of the equipment can be perceived, and long-term performance evolution trends, potential failure modes, and remaining service life can be predicted. Performance degradation data is generated, achieving a transition from passive response to active prediction, effectively avoiding unplanned downtime caused by equipment aging, and improving the predictability and reliability of equipment management.

[0015] 2. The present invention establishes a resource optimization model and uses an optimization solver to perform forward-looking collaborative optimization of the global decision-making view that includes production plans and performance degradation data. At the same time, the comprehensive operating costs including resource loss, energy consumption and potential failure risks are quantified, and the quantified results are used as the objective function for global optimization. This operation takes the equipment status as the core of production scheduling decisions, proactively balances the relationship between short-term production tasks and long-term equipment health, and preferentially allocates high-precision or heavy-load tasks to units with better health status. At the same time, while meeting the constraints of the overall production plan, it achieves dynamic minimization of global operating costs.

[0016] 3. The present invention uses a dynamic compensation management mechanism within a distributed intelligent executor to perform closed-loop regulation including feedforward compensation and adaptive feedback correction on job instructions that combine real-time status data and performance degradation data, forming a control system with dual protection. Feedforward compensation offsets disturbances caused by short-term status data changes in real time, and adaptive feedback correction continuously addresses the impact of long-term performance degradation. This ensures that job instructions can be accurately executed under the dual challenges of changing working conditions and continuous equipment performance degradation, guarantees the long-term stability of processing accuracy and product quality, and enhances the robustness of the entire production system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Fig. 1 This is a flow chart of a method for intelligent flow scheduling of high-performance oil circuits in sheet metal pipe networks proposed in an embodiment of the present invention; Fig. 2 A resource scheduling flowchart for the embodiments of the present application is provided. Fig. 3 A smart decision flowchart for the embodiments of the present application is provided. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0019] Please refer to Figs. 1 to 3 The present application provides a flow intelligent scheduling method for a high-performance oil circuit of a sheet metal pipe network, and the technical solutions are as follows: A flow intelligent scheduling method for a high-performance oil circuit of a sheet metal pipe network, comprising the following steps: Based on a sheet metal pipe network comprising schedulable units and production management information, a distributed intelligent executor and a resource optimization model are established; the production management information comprises historical performance baselines and production plans; Obtain running data of the schedulable units, and the distributed intelligent executor evaluates the running data to generate real-time state data; calculate a current output state according to a preset test instruction, analyze a drift trend of the current output state deviating from the historical performance baselines, and generate performance degradation data of the schedulable units; Based on the production plans and the performance degradation data, quantitatively generate comprehensive operation costs, take the comprehensive operation costs as a target function, and the resource optimization model calculates and generates a resource allocation strategy based on the resource allocation strategy to generate job instructions; The distributed intelligent executor combines the job instructions, real-time state data and performance degradation data to execute dynamic compensation management and generate scheduling instructions; and the schedulable units dynamically track the job instructions based on the scheduling instructions.

[0020] Embodiment one The present embodiment provides a specific application of a flow intelligent scheduling method for a high-performance oil circuit of a sheet metal pipe network, and a typical application scenario is that enterprise A introduces a flow intelligent scheduling method to reduce the energy consumption of a hydraulic system and avoid unplanned downtime caused by equipment aging.

[0021] Reference Fig. 1 The method comprises the following steps: S1, based on a sheet metal pipe network comprising schedulable units and production management information, a distributed intelligent executor and a resource optimization model are established; the production management information comprises historical performance baselines and production plans; S2, obtaining the running data of the schedulable unit, the distributed intelligent executor evaluates the running data to generate real-time state data; S3, calculating the current output state for the preset test instruction, analyzing the drift trend of the current output state deviating from the historical performance baseline to generate performance degradation data of the schedulable unit; S4, based on the production plan and the performance degradation data, quantifying the comprehensive operation cost, taking the comprehensive operation cost as the objective function, the resource optimization model calculates to generate a resource allocation strategy, and generates a job instruction based on the resource allocation strategy; S5, the distributed intelligent executor combines the job instruction, real-time state data and performance degradation data to execute dynamic compensation management to generate a scheduling instruction; the schedulable unit dynamically tracks the job instruction based on the scheduling instruction.

[0022] Further, based on the sheet metal pipe network containing the schedulable unit and the production management information, a resource optimization model and a distributed intelligent executor are established; the production management information includes historical performance baseline and production plan, corresponding to the above S1 step, the specific process includes: Analyzing the physical sheet metal pipe network, identifying the core value unit and taking the core value unit as an independent schedulable unit in the system, such as a hydraulic press and a high-performance oil circuit station; giving the schedulable unit a unique identifier and entering the physical parameters of the schedulable unit into the system, the physical parameters include maximum pressure, rated flow, cylinder size; providing the defined schedulable unit with an edge computing device, such as an industrial PC and a high-performance PLC; deploying a distributed intelligent executor based on the edge computing device, the deployment process specifically instantiates a software architecture containing a state evaluation unit and a performance trend prediction unit; the state evaluation unit contains an information network model, which is constructed based on a physical information neural network and contains a fluid state model; for example, the information network model uses a fully connected feedforward neural network containing 3 hidden layers, the input layer receives temperature and pressure data, uses ReLU as the activation function, and the output layer outputs the preliminary inference value of oil viscosity and density.

[0023] The loss function of the information network model is obtained by summing a data loss and a physical constraint, the data loss being a mean square error of the preliminary inference value and a reference value obtained by consulting a specification; the physical constraint is specifically that the viscosity and density values inferred by the information network model are compared with theoretical values calculated by a fluid state model according to temperature and pressure data to obtain absolute values of viscosity residual error and density deviation, and the physical constraint is obtained by weighted summing the absolute values of the viscosity residual error and the density deviation, the weights being set based on the importance of the actual working conditions, for example, for tasks that require precise pressure retention, pressure stability is crucial, and the weight of the density deviation affected by the pressure can be set to 0.6, and the weight of the viscosity deviation can be set to 0.4; for auxiliary tasks that do not require high requirements, the weights can be set to 0.5.

[0024] The performance trend prediction unit is configured with an online learning algorithm containing a mapping model, for example, a recursive least squares algorithm based on online gradient descent, which updates the mapping model in real time according to newly acquired performance characteristics; the distributed intelligent executor is bound to the corresponding schedulable unit to form a minimum functional closed loop. In this way, the distributed intelligent executor can infer oil liquid state data in real time and adaptively update the internal information network model, ensuring high-precision execution of control instructions under varying working conditions, improving the state perception and adaptive ability of the system, and ensuring the accuracy and efficiency of production scheduling.

[0025] Through a standard data interface, the production management system is connected to a locally deployed resource optimization model server; a digital file is established for the schedulable unit, and a historical performance baseline is generated, the process of generating the historical performance baseline including performing a preset benchmark operation based on the benchmark state of the schedulable unit and recording the mapping relationship between the preset benchmark operation and the required resource consumption, storing the mapping relationship to form the historical performance baseline, for example, recording the benchmark control output required by the servo valve when a new hydraulic machine executes operation instructions covering a commonly used working condition range. The historical performance baseline is established by the control output of the new device performing the benchmark operation, providing a data basis for calculating the performance degradation index and predicting the remaining service life, converting the device aging process into a quantitative specific value, improving the accuracy of state evaluation, and ensuring the reliability of resource scheduling.

[0026] A resource optimization model is constructed based on a demand prediction model and an optimization algorithm, and the resource optimization model is configured in a serial processing mode, including a data integration stage, an optimization decision stage and an instruction generation stage; the data integration stage receives production plans of a production management system and real-time performance degradation data of a distributed intelligent execution body through a configured data interface, fuses the system production plans and the real-time performance degradation data to generate a global decision view containing a current state of a pipe network and future task load; the optimization decision stage includes an optimization solver, which uses intelligent optimization algorithms such as genetic algorithms and simulated annealing algorithms, takes the time sequence load demand specified by the production plans as a constraint, and comprehensively optimizes and solves a target function taking the operating cost as the target function, to generate a resource allocation strategy containing a schedulable unit operation role and a work task; the comprehensive operating cost is based on the performance degradation data and the predicted energy consumption demand, and the resource loss cost, the energy cost and the failure risk cost are quantified and weighted to obtain a sum; the instruction generation stage parses the resource allocation strategy generated by the optimization decision into job instructions containing time sequence and physical parameters, which can be directly executed by the distributed intelligent execution body. By constructing the resource optimization model, the performance degradation, the energy consumption and the risk are quantified as the comprehensive operating cost and used as the optimization target, so that the scheduling decision can prospectively balance the short-term production and the long-term equipment health, and the scientificity and the economy of the decision are improved.

[0027] Further, operation data of the schedulable unit are acquired, and the distributed intelligent execution body evaluates the operation data to generate real-time state data, and the specific process corresponding to the S2 step includes: The state evaluation unit of the distributed intelligent executive body collects sensor data installed at the key nodes of the schedulable unit in real time to obtain operation data, including real-time temperature and pressure data; the real-time temperature and pressure data are taken as inputs of an information network model in the state evaluation unit to infer real-time state data, including oil viscosity and density; the inference process includes: taking the real-time temperature and pressure data as input data of the information network model, the information network model processes the input data through nonlinear transformation to generate a preliminary inference value of the real-time state data; a fluid state model is used to calculate a theoretical reference value of the real-time state data based on the real-time temperature and pressure data, the fluid state model includes a viscosity-temperature relationship model and a density-temperature-pressure relationship model, the viscosity-temperature relationship model converts the nonlinear viscosity-temperature relationship into a linear relationship through double logarithmic mathematical transformation of viscosity and absolute temperature; the density-temperature-pressure relationship model first calculates the base density under normal pressure through a polynomial function about temperature, and then adds the compression effect of high pressure through a correction term containing a natural logarithm, the oil product characteristic constant in the model is pre-calibrated by consulting the product specification book, and the oil product characteristics include reference viscosity and density; for example, hydraulic machine A1 uses hydraulic oil a to perform a continuous stamping task, when the oil temperature rises to 65°C and the pressure reaches 20 MPa, the calculation process of the viscosity-temperature relationship model is as follows: the system first consults the specification book of hydraulic oil a to obtain the calibration data of hydraulic oil a at two key points of "40°C, viscosity 46 cSt" and "100°C, viscosity 6.8 cSt", and connects the two key points into a straight line in the logarithmic coordinate system through double logarithmic transformation, thereby defining a linear viscosity-temperature relationship; when the real-time temperature measured by the sensor is 65°C, the system can obtain the corresponding theoretical viscosity value of 23.5 cSt through the straight line. The calculation process of the density-temperature-pressure relationship model is as follows: first, the system consults the specification book of hydraulic oil a to obtain the standard density of hydraulic oil a at 15°C, which is 877 kg / m³, and calculates the base density at "65°C, normal pressure" through a second-order polynomial function, which is about 858 kg / m³; then, the system calculates the density increase value caused by pressure through a correction term containing a natural logarithm based on the current pressure, which is about 12.5 kg / m³. Finally, the base density and the density increase value are added to obtain the final theoretical density under the current working condition.

[0028] The difference between the preliminary inference value and the theoretical reference value is calculated, and a weight is dynamically allocated according to the difference, for example, a trust threshold is set based on the theoretical reference value, for example, the trust threshold is set to 5% of the theoretical reference value, if the difference is within the trust threshold, it is considered that the information network model and the fluid state model are reliable under the current working condition, and the information network model can be given a weight of 0.8; if the difference exceeds the trust threshold, it is considered that the model prediction deviates from the physical constraints, and the weight of the information network model starts to decrease from 0.8, and linearly decreases to 0.2 in the interval from 5% to 15% of the difference, for example, when the difference is 10%, the weight can take the intermediate value 0.5, and if the difference exceeds 15%, the weight is fixed at the minimum value 0.2; the sum of the preliminary inference value and the reference value generates real-time state data. The real-time state data is generated by fusing the network inference and the fluid state model verification, which realizes accurate perception of the real-time state data and effectively improves the risk resistance, providing a high-precision and high-robustness data basis for accurate control and optimal scheduling.

[0029] Further, the current output state is calculated according to the preset test instruction, the drift trend of the current output state deviating from the historical performance baseline is analyzed, and the performance degradation data of the schedulable unit is generated. Corresponding to the above S3 step, the specific process includes: In the non-production gap of the schedulable unit, the performance trend prediction unit of the distributed intelligent executor automatically executes the standardized test instruction, for example, in the waiting stage after a hydraulic machine of enterprise A completes the stamping task, the instruction of "driving a certain hydraulic cylinder to linearly climb from 0MPa to 20MPa in 500ms" is executed; the system collects the actual pressure output state, compares the actual pressure output state with the stored historical performance baseline in real time, and generates a real-time deviation value sequence reflecting the current performance deviation; the real-time deviation value sequence removes random noise through a low-pass filtering algorithm to generate a smooth deviation curve; the trend of the smooth deviation curve is analyzed through a linear regression algorithm to extract a dynamic drift trend signal; based on the dynamic drift trend signal, a characteristic parameter is quantified, and the characteristic parameter includes an average bias, a response delay, an overshoot and an oscillation amplitude, and a drift rate, for example, the average bias is that the current pressure output is 0.15MPa lower than the historical performance baseline on average, indicating that there may be internal leakage and pump efficiency decline; the response delay is that the time point of reaching 90% of the target pressure is delayed by 60ms compared with the baseline, indicating that the system response performance has decreased; the overshoot and the oscillation amplitude increase from 1% to 3% of the baseline, and the residual oscillation amplitude after stabilization increases, indicating that it is related to the deterioration of the dynamic characteristics of the control valve; the drift rate is quantified according to historical diagnosis data, and it is considered that the average bias deteriorates at a rate of 0.01MPa per week.

[0030] The online learning algorithm in the performance trend prediction unit utilizes a mapping model to normalize and weightedly sum the characteristic parameters to generate a quantitative index; the mapping model is a multivariate linear model, the current performance degradation index is obtained by weightedly summing the normalized characteristic parameters, and the weighted weight is updated in real time by a recursive least square algorithm based on online gradient descent; the specific updating process is to preset the initial weight coefficient as an equal value, the system obtains the verification result of the device state by associating a specific maintenance event, takes the verification result as the target index, for example, when the schedulable unit triggers an unscheduled maintenance due to performance problems, the system automatically marks the target index corresponding to the performance characteristic vector collected before maintenance as 0.9 to provide a supervised learning sample for the algorithm; the newly collected performance characteristic vector is taken as the input of the recursive least square algorithm, and the error between the minimum performance degradation index and the target index is taken as the optimization target to iteratively adjust and update the weight coefficient of the multivariate linear model.

[0031] The current performance degradation index is obtained, for example, the current performance degradation index is 0.58 calculated by the comprehensive characteristic parameters, the performance degradation index is set to 0 for complete effectiveness and 1 for failure; secondly, based on the drift rate, the future degradation rate is predicted to be an index increase of 0.015 per week; at the same time, the characteristic parameters are matched with a preset fault rule library, the preset fault rule library includes identifying typical failure modes of the hydraulic system by using failure mode and effects analysis method; for the failure mode, the correlation between the failure mode and the data characteristic parameters is collected by analyzing the historical maintenance records; based on these correlation, logical rules readable by machines are refined and solidified to form a rule library. For example, the rule entries can be: Rule ID01: if the average bias is greater than 0.1 MPa and the response delay is greater than 50 ms, the potential fault signature is "increased risk of main valve internal leakage"; Rule ID02: if the overshoot is greater than 5% and the steady-state oscillation amplitude is greater than 1.5%, the potential fault signature is "control loop parameter drift and servo valve dynamic characteristic degradation"; Rule ID03: if the response delay does not change significantly, but the overshoot and oscillation amplitude continue to increase, the potential fault signature is "control system PID parameter mismatch".

[0032] Then, based on the current degradation index, the future degradation rate and the preset failure threshold, the predicted remaining useful life is calculated; finally, the current performance degradation index, the future degradation rate, the potential fault signature and the predicted remaining useful life are integrated into structured performance degradation data. By analyzing the deviation of the current performance from the historical baseline, the performance degradation data is generated, realizing the transition from passive fault response to active health prediction, improving the predictability and reliability of device management, and providing quantitative basis for risk assessment and cost optimization decision.

[0033] Further, in the process of generating performance degradation data, a dynamic correction mechanism for predicting remaining useful life is introduced; the system matches the extracted performance feature vector with the pre-set fault rule library, identifies the fault signature and calls the adjustment coefficient associated with the fault signature, corrects the preliminary predicted future degradation rate, and recalculates the predicted remaining useful life based on the corrected future degradation rate. Through the dynamic correction mechanism for predicting remaining useful life, the prediction of remaining useful life is transformed from general trend extrapolation to precise calculation based on specific physical failure modes, improving the accuracy and reliability of the prediction of remaining useful life, providing reliable decision basis for preventive maintenance, and reducing the resource loss and failure risk cost in the comprehensive operation cost.

[0034] Further, based on the production plan and the performance degradation data, the comprehensive operation cost is quantified, the resource optimization model takes the comprehensive operation cost as the objective function, calculates the resource allocation strategy, and generates the job instruction based on the resource allocation strategy. Corresponding to the above S4 step, the specific process includes: The resource optimization model receives the production plan of the production management system of enterprise A, for example, 5000 sheet metal parts of type A1 in the next 24 hours; based on the process file, the hydraulic demand curve required for stamping A1 parts is obtained, the time series load demand sequence and the predicted total energy consumption are generated; combined with the performance degradation data, the cost of the task performed by the schedulable unit is quantified in advance. For example, cost analysis is performed on hydraulic machine A2 with a performance degradation index of 0.58, a potential fault signature of "increased risk of main pump or main valve internal leakage", and a predicted remaining useful life of 21 weeks.

[0035] The comprehensive operation cost is obtained by weighted summation of resource consumption cost, energy cost and risk cost, and the weight coefficients are based on task setting, for example, under the normal production task, the weight coefficients are respectively set to 0.3, 0.5 and 0.2, when the task priority or accuracy requirement changes, the weight of the risk cost is increased to 0.7, and the other weights are normalized; the resource consumption cost is proportional to the square of the performance degradation index, the performance degradation index is given a penalty weight in a nonlinear amplification manner, for example, the penalty weight of the healthy device A1 with a performance degradation index of 0.05 can be 0.0025, and the penalty weight of the degraded device A2 with a performance degradation index of 0.58 can be 0.34; the energy cost is calculated based on the predicted energy consumption power, and the predicted energy consumption power is composed of the baseline power and the additional loss caused by performance degradation; the risk cost is obtained by adding the device cost and the potential quality unqualified cost, the device cost is inversely proportional to the predicted remaining service life, specifically, the replacement cost of the device is divided by the predicted remaining service life to obtain the device cost; the potential quality unqualified cost is quantified by a risk multiplier, and the risk multiplier is dynamically adjusted according to the task accuracy and the device state, for example, when A2 performs a high-precision task, based on the fault signature of "increased risk of main pump or main valve leakage", the system can set the risk multiplier to 5.0; and if it is an ordinary auxiliary task that is not sensitive to accuracy, the risk multiplier can be set to 1.0.

[0036] Further, in the process of quantifying the comprehensive operation cost, a weight dynamic self-adaptive mechanism is introduced; the resource optimization model automatically obtains the real-time attributes of the task, and dynamically adjusts the weight coefficients according to the real-time attributes, and takes the comprehensive operation cost function after adjusting the weight coefficients as a new optimization target for solving. Through the weight dynamic self-adaptive mechanism, the scheduling decision is real-time aligned with the business priority, and the optimal decision is made through cost optimal calculation when dealing with sudden tasks, and the flexibility and intelligence of the decision are improved, and the production scheduling is changed from a static process of passive execution of predetermined plan to a dynamic value creation process of active real-time response to market changes and internal state.

[0037] Based on the cost quantification results, an optimization problem is constructed. The objective function is the total integrated operating cost of all dispatchable units. The constraints are that the combined output of all dispatchable units meets the sequential load requirements of the production plan and that the final total output equals the planned output. By solving this objective function, the model generates an optimal resource allocation strategy. For example, the optimal strategy might be to assign 4,500 units of work to the hydraulic press in good condition and the remaining 500 units to the degraded hydraulic press A2, instructing it to operate at a lower load and automatically generating a high-priority maintenance work order to inspect hydraulic press A2 within one week. The model interprets this resource allocation strategy as a job instruction with precise timing and physical parameters, such as a target pressure of 20 MPa and a hold time of 2 seconds. By quantifying equipment performance degradation and failure risk as integrated operating costs and minimizing this integrated operating cost as the optimization objective, the model proactively balances short-term production tasks with long-term equipment health, achieving optimal control of global operating costs and risks.

[0038] Furthermore, the distributed intelligent executor combines the job instructions, real-time status data, and performance degradation data to perform dynamic compensation management to generate scheduling instructions; the schedulable unit dynamically tracks the job instructions based on the scheduling instructions, corresponding to the above step S5. The specific process includes: The dispatchable unit of the distributed intelligent executive receives and parses the job instructions to obtain the target trajectory, which includes the time curves of pressure and flow. The target trajectory is input into a simplified dynamic model to generate an initial control output sequence of the ideal state. For example, a second-order model describing the dynamics of hydraulic cylinder pressure provides a reasonable feedforward open-loop control variable for the control system. The dispatchable unit obtains real-time status data, for example, the current oil temperature is 55°C, and the viscosity has decreased by 30% compared to the baseline state of 40°C. Based on the deviation between the current real-time state data and the reference state, the system calculates the feedforward compensation adjustment amount. Specifically, the system extracts the current oil viscosity and density from the real-time state data and calculates the deviation between the oil viscosity, density and the reference state data; based on the preset compensation rules, the system processes the deviation to generate a specific control signal adjustment amount. The compensation rule can be stipulated as follows: when the system senses that the real-time oil viscosity has decreased relative to the reference state, the intensity of the initial control instruction will be reduced by a preset proportional coefficient. The magnitude of the proportional coefficient is linearly positively correlated with the magnitude of the detected viscosity decrease. For example, the proportional coefficient is preset based on experimental calibration and can be preset to reduce the intensity of the initial control instruction by 2% when the oil viscosity decreases by 10%. Therefore, if the system detects that the current oil viscosity has decreased by 30%, the intensity of the initial control instruction will be reduced by 6%, thereby offsetting the pressure overshoot that may be caused by the change in oil state in advance; the control signal adjustment amount is superimposed on the initial control output sequence to obtain the control output sequence after feedforward compensation.

[0039] With the potential fault signature, the control output sequence after feedforward compensation is adaptively feedback corrected, which is manifested as real-time adjustment of PID controller parameters; for example, the potential fault signature is "increased risk of internal leakage of main pump or main valve", the controller increases the integral gain and reduces the differential gain through online algorithm, and the adjustment amplitude is proportional to the current performance degradation index; for example, when the performance degradation index increases from 0 to 1, the integral gain is increased by a maximum of 50% based on the benchmark value, and the differential gain is reduced by a maximum of 30%; when the system detects that the current performance degradation index is 0.58, the integral gain will be increased by about 29% based on the benchmark value, and the differential gain will be reduced by about 17.4%.

[0040] When responding to and executing the scheduling instructions, the distributed intelligent executor generates the sequence after feedforward compensation and adaptive feedback correction as the final control output sequence; and sends the final control output sequence to the physical actuator of the hydraulic press A2 to drive the completion of the stamping work. By combining feedforward compensation and adaptive feedback correction, it is ensured that the work instructions are still executed with high precision and high stability in the face of the dual challenges of variable working conditions and continuous equipment aging, thereby enhancing the robustness of the system and reducing the processing scrap rate and energy loss caused by control misalignment.

[0041] Further, in the adaptive feedback correction in dynamic compensation management, a control strategy switching mechanism based on fault mode is introduced; the system queries and matches from a preset control strategy library according to the identified potential fault signature, automatically selects and switches to the optimal control strategy of the potential fault signature, and the control strategy library includes adaptive PID algorithm, model predictive control and robust control; the matching and switching are based on preset rules, for example, when the fault signature is "control loop parameter drift" or "servo valve dynamic characteristic degradation", the system switches to the adaptive PID algorithm; when the confidence of the fault signature is higher than a threshold and is "increased risk of internal leakage of main pump or main valve", the system switches to the model predictive control; when the system faces increasing sensor noise or unmodeled external disturbance, it can switch to robust control. Through the control strategy switching mechanism, specific physical property changes caused by equipment aging are corrected by targeted control algorithms, ensuring that the system executes the work instructions with high precision and high stability, reducing the processing scrap rate and energy loss caused by control misalignment, enhancing the adaptive ability and overall robustness of the production system to long-term performance degradation, and ensuring the long-term stability of product quality.

[0042] The present application generates a resource allocation strategy based on the optimal risk and cost by evaluating the equipment performance degradation state and quantifying the comprehensive operating cost, thereby improving the economy and foresight of production scheduling decisions. Meanwhile, through dynamic compensation and closed-loop adjustment of the distributed intelligent executor, accurate execution of the work instructions is realized, thereby improving the robustness and reliability of the system under variable working conditions.

[0043] Example Two The application of the flow intelligent scheduling method of a high-performance oil circuit of sheet metal pipe network to sheet metal processing workshop B with a central high-performance oil circuit station is described in the example of the present application. Referring to Fig. 2 , the central high-performance oil circuit station simultaneously provides power for hydraulic machines B1, B2 and B3. The production plans of the three hydraulic machines are different, and the demand for hydraulic resources, including flow and pressure, is coupled and in competition.

[0044] The core of the central high-performance oil circuit station is a variable pump group with a rated pressure of 25 MPa and a maximum output flow of 400 L / min. Within a 60-second production cycle, hydraulic machine B1 performs a high-precision forming task: fast descent for the first 5 seconds, requiring a flow of 200 L / min and a pressure of 5 MPa; 5 to 15 seconds of stamping and pressure maintenance, requiring a flow of 20 L / min and a pressure of 20 MPa, with an accuracy requirement of ±0.2 MPa; 15 to 20 seconds of fast return, requiring a flow of 180 L / min and a pressure of 4 MPa; Hydraulic machine B2 performs a regular stamping task: fast descent for 10 to 14 seconds, requiring a flow of 150 L / min and a pressure of 5 MPa; 14 to 25 seconds of stamping, requiring a flow of 30 L / min and a pressure of 18 MPa; 25 to 29 seconds of fast return, requiring a flow of 140 L / min and a pressure of 4 MPa; Hydraulic machine B3 performs standby and auxiliary tasks: 0 to 60 seconds of standby, requiring a flow of 5 L / min and a pressure of 2 MPa.

[0045] The system has established a resource optimization model and corresponding distributed intelligent execution body containing the central oil circuit station and the three hydraulic machines B1, B2 and B3. The maximum output flow and pressure of the central oil circuit station are entered into the system as key parameters; the resource optimization model receives the production plans of hydraulic machines B1, B2 and B3, analyzes the process steps, and generates three independent time sequence load demand sequences; the model superimposes the three time sequence load demand sequences based on the time axis vector to generate a global total load prediction curve; the model identifies potential resource competition conflicts in the 10th to 14th seconds by analyzing the global total load prediction; the system obtains the performance degradation data of each unit to determine that hydraulic machines B1, B2 and B3 are in a healthy state.

[0046] An optimization problem is constructed, which includes an objective function of "minimizing the sum of total energy consumption and potential quality risk cost of central oil circuit stations" and constraint conditions that the instantaneous output flow and pressure of the central high-performance oil circuit station meet the global total load prediction curve, and the pressure fluctuation of B1 in the pressure maintaining stage is less than ±0.2 MPa. By quantifying the energy consumption and quality risk as cost and taking them as the optimization target, the system can actively avoid resource conflicts while ensuring high-precision tasks, find the most economical scheduling scheme under the condition of meeting multiple complex constraints, and improve the utilization efficiency of resources After the optimization solver calculates, a resource allocation strategy for different devices is generated, for example, instructing the central oil circuit station to increase the system pressure from 20 MPa to 20.5 MPa at the 9th second, improving the response preparation level of the pump as a feedforward compensation for the flow demand of B2; a new operation instruction is generated: delaying the fast downlink start time of B2, smoothing the downlink speed curve, reducing the peak flow, and prolonging the downlink time; keeping the original operation instruction of hydraulic machine B1 unchanged, and the intelligent execution body enters the high-precision closed-loop regulation mode. Through the resource optimization model, the mutually coupled production plans are converted into coordinated operation instructions, solving the resource competition and conflict problem caused by time sequence overlap among multiple devices, improving the resource allocation efficiency in a complex coupled scene, and enhancing the coordination and stability of the production system.

[0047] Each distributed intelligent execution body receives and executes the above operation instruction; the central oil circuit station increases the pressure in advance according to the instruction; and the execution body of B2 controls the servo valve according to the corrected time and speed curve.

[0048] Embodiment Three In the embodiment of the application, the flow intelligent scheduling method of the high-performance oil circuit of sheet metal pipe network is applied to a sheet metal processing workshop C having hydraulic machines C1 and C2 with different performances.

[0049] In the initial state, the performance degradation index of hydraulic machine C1 is 0.05, which is in good condition and is executing a production task of "standard part A" with a batch of 500 pieces, a single piece consuming 30 seconds, and a task priority of "regular"; the performance degradation index of hydraulic machine C2 is 0.28, which has a potential fault signature of "increased risk of main valve internal leakage", a response delay greater than the baseline, and is executing a production task of "auxiliary part B" with a batch of 1000 pieces, a single piece consuming 25 seconds, and a task priority of "regular".

[0050] Reference is made to Fig. 3, the production management system sends an urgent insert order through the data interface: produce 10 high-precision parts X, with extremely high precision requirements and a delivery time of half an hour. The resource optimization model receives the urgent insert order from the production management system and parses it into specific process steps containing delivery time limit and process precision requirements; at the same time, the resource optimization model obtains the latest performance degradation data of C1 and C2, and confirms that C1 meets the process precision requirements of part X, and C2 does not meet the process precision requirements. After identifying the "urgent" and "high precision" attributes, the system adjusts the weight of the comprehensive operation cost function in real time; for part X task, the weight of risk cost is increased. The resource optimization model takes the updated cost function as the objective function, simulates multiple possible rescheduling schemes, and quantifies the comprehensive operation cost: Simulation scheme 1: place the urgent task X in the task queue of hydraulic machine C1, and wait for the hydraulic machine C1 in good condition to complete the current task before starting to produce part X. The comprehensive operation cost of this scheme includes current operation cost and delay risk cost; the current operation cost includes the regular resource consumption and energy cost generated by C1 and C2 executing tasks; the delay risk cost is specifically that the urgent task X requires delivery within half an hour, and waiting for C1 to complete the current task will cause the urgent task X to be delayed, and the system quantifies the delay as a delay risk cost. This scheme guarantees the continuity of regular tasks, but cannot meet the key time limit constraints of high-priority tasks, resulting in an increase in risk cost.

[0051] Simulation scheme 2: assign the urgent task X to hydraulic machine C2 with performance degradation data of 0.28. The comprehensive operation cost of this scheme includes current operation cost and quality risk cost; the current operation cost is the basic cost of system operation; the quality risk cost is specifically: the system obtains the performance degradation index and potential failure signature of hydraulic machine C2 through performance degradation data, judges that C2 performance state does not meet the high precision requirements of part X, and C2 production will cause the increase of waste rate and quality hidden danger, the system quantifies the quality hidden danger as high quality risk cost. This scheme meets the time requirement, but cannot meet the product quality requirement, resulting in an increase in risk cost.

[0052] Simulation scheme 3: interrupt the current task of hydraulic machine C1 in good condition, execute the urgent task X first, and then resume the production task of "standard part A". The comprehensive operation cost of this scheme includes current operation cost, interruption and mold changing cost, and task A delay cost; the current operation cost is the basic cost of system operation; the interruption and mold changing cost is the production downtime and resource consumption cost caused by interrupting production, changing mold and restarting equipment. The task A delay cost is the cost generated by postponing the original plan of standard part A, and the priority of standard part A task is "regular", the delay cost is lower than the delay risk cost of urgent task X.

[0053] The comprehensive operation cost of simulation scheme 3 includes the newly added controllable interruption and die change cost, and avoids the uncontrollable delay risk cost and quality risk cost in simulation schemes 1 and 2; through calculation, the comprehensive operation cost of scheme 3 is lower than the first two schemes, the system determines the optimal scheme, and generates a new resource allocation strategy: C1 safely suspends the production of the current "standard part A", records the breakpoint information; C1 executes the production task of high-precision part X; the interrupted "standard part A" task state is updated to "paused", and is re-arranged in the production queue after C1 completes task X. By dynamically adjusting the weight of the risk item in the comprehensive operation cost, the task attribute is quantified as an economic indicator, so that the system can make the optimal trade-off among multiple potential schemes through cost calculation when responding to emergencies, and the flexibility and intelligence of the scheduling decision are improved.

[0054] After receiving the instruction, the distributed intelligent executor of hydraulic machine C1 suspends the current task and guides the automation system to complete the die change within 5 minutes; hydraulic machine C1 completes the production of 10 qualified parts X within the time limit, and then automatically resumes the production of standard part A.

[0055] Although embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent flow scheduling of high-performance oil circuits in sheet metal pipe networks, characterized in that: include: Based on the sheet metal pipe network and production management information including schedulable units, a distributed intelligent execution body and resource optimization model are established; The production management information includes historical performance baselines and production plans; Acquiring operation data of the dispatchable unit, wherein the distributed intelligent executive body evaluates the operation data to generate real-time status data; Calculate the current output state for the preset test instructions, analyze the drift trend of the current output state from the historical performance baseline, and generate performance degradation data of the schedulable unit; Based on the production plan and the performance degradation data, a comprehensive operating cost is quantified and generated, the comprehensive operating cost is used as an objective function, the resource optimization model calculates and generates a resource allocation strategy, and a job instruction is generated based on the resource allocation strategy; The distributed intelligent executor combines the job instructions, real-time status data and performance degradation data to perform dynamic compensation management to generate scheduling instructions; the schedulable unit dynamically tracks the job instructions based on the scheduling instructions.

2. The method for intelligent flow scheduling of high-performance oil circuits in sheet metal pipe networks according to claim 1, characterized in that: The distributed intelligent executor includes a state evaluation unit and a performance trend prediction unit; the state evaluation unit includes an information network model, which receives the operation data and generates real-time state data, and the information network model includes a fluid state model; the performance trend prediction unit includes an online learning algorithm, which analyzes the output state and historical performance baseline of the preset test job instructions and generates performance degradation data; based on the job instructions and fusing the real-time state data and the performance degradation data, the dynamic compensation management is executed, the final control output is calculated, and the scheduling instructions are generated.

3. The method for intelligent flow scheduling of high-performance oil circuits in sheet metal pipe networks according to claim 1, characterized in that: The resource optimization model is constructed based on the demand forecast model and optimization algorithm, and is implemented through data integration, optimization decision and instruction generation stages. The data integration stage receives and integrates the production plan and the performance degradation data to generate a global decision view that includes the pipeline network status and future task load. In the optimization decision stage, based on the global decision view, a forward-looking optimization solution is performed to generate the resource allocation strategy, wherein the resource allocation strategy includes the operation roles and work tasks of the schedulable units; The instruction generation stage parses the running roles and work tasks into job instructions containing timing and physical parameters.

4. The method for intelligent flow scheduling of high-performance oil circuits in sheet metal pipe networks according to claim 1, characterized in that: The historical performance baseline is a mapping relationship between the benchmark operation and the required resource consumption when the schedulable unit is in a benchmark state.

5. The method for intelligent flow scheduling of high-performance oil circuits in sheet metal pipe networks according to claim 1, characterized in that: The process of generating real-time status data includes: taking the operating data as the input of the information network model, the operating data including temperature data and pressure data, and the information network model performing nonlinear transformation on the temperature data and pressure data to generate a preliminary inference value of the real-time status data; the real-time status data includes oil viscosity and density; calculating the real-time status data reference value through a fluid state model, the fluid state model including a viscosity-temperature relationship model and a density-temperature-pressure relationship model; comparing the preliminary inference value and the real-time status data reference value and calculating the difference, determining a weight based on the difference, and obtaining the final status data by weighted summation of the preliminary inference value and the real-time status data reference value.

6. The method for intelligent flow scheduling of high-performance oil circuits in sheet metal pipe networks according to claim 2, characterized in that: The process of generating performance degradation data includes: calculating a sequence signal and obtaining a required current output state based on a preset test instruction; comparing the current output state with a historical performance baseline to generate a real-time deviation value sequence; filtering out random noise from the real-time deviation value sequence to obtain a smooth curve, and extracting a dynamic drift trend signal by applying linear regression trend analysis to the smooth curve; extracting characteristic parameters based on the dynamic drift trend signal to generate a characteristic vector, wherein the characteristic parameters include average bias, response delay, overshoot and oscillation amplitude, and drift rate; utilizing a mapping model in the online learning algorithm to normalize the characteristic parameters and perform weighted summation to generate a current performance degradation index and a future degradation rate, matching the characteristic parameters with a preset fault rule library to identify potential fault signatures, and calculating a predicted remaining service life based on the current performance degradation index, the future degradation rate, and a preset failure threshold; integrating the current performance degradation index, the future degradation rate, the potential fault signature, and the predicted remaining service life into performance degradation data output.

7. The method for intelligent flow scheduling of high-performance oil circuits in sheet metal pipe networks according to claim 1, characterized in that: The process of generating a resource allocation strategy includes: parsing the production plan to generate specific process steps, obtaining the hydraulic demand curve of the specific process steps, and generating a time-series load demand sequence and a predicted total energy consumption demand based on the hydraulic demand curve; combining the predicted total energy consumption demand and the performance degradation data, quantitatively evaluating the performance status, operating energy efficiency and failure risk of the dispatchable unit, determining the resource loss cost, energy cost and risk cost, and weighted summing the resource loss cost, energy cost and risk cost to generate a quantitative comprehensive operating cost; taking the total output of the dispatchable unit to meet the time-series load demand sequence as a constraint, constructing an objective function, and the objective function takes minimizing the comprehensive operating cost as the optimization goal, and solving the objective function to generate the resource allocation strategy.

8. The method for intelligent flow scheduling of high-performance oil circuits in sheet metal pipe networks according to claim 1, characterized in that: The execution of dynamic compensation management to generate scheduling instructions includes: parsing the job instructions to obtain a target trajectory, inputting the target trajectory into a simplified dynamic model to generate an initial control output sequence, wherein the simplified dynamic model includes state parameters and response characteristics of the schedulable unit in a reference state; performing feedforward compensation based on the deviation between the real-time state data and the reference state system data to obtain a feedforward compensated control output sequence; utilizing the potential fault signature of the performance degradation data to perform adaptive feedback correction on the feedforward compensated control output sequence, calculating a final control output sequence, and generating scheduling instructions.

Images