An energy consumption intelligent optimization control method and system for an automobile pipe profile cutting production line

Abstract

The application discloses a kind of energy consumption intelligent optimization control method and system of automobile pipe profile cutting production line, comprising: real-time acquisition of the instantaneous current signal of cutting motor and obtains the attribute information of pipe profile being cut;Extract short-time energy feature and short-time mutation intensity feature, input pre-constructed state decision model to identify cutting process, obtain idle stage, cutting-in stage, steady cutting stage and cutting-out stage;According to attribute information, call corresponding cooling parameter group from pre-constructed cooling process database, generate cooling liquid supply instruction in combination with cutting phase and execute, according to nonlinear ramp curve control cooling pump power variation;The end surface quality index of pipe profile being cut is compared with the preset quality threshold interval to obtain the deviation value, to correct the first target flow value and the second target flow value in cooling process database, through cutting micro-process stage online identification, realize cooling liquid supply and heat load accurate matching, effectively reduce cooling liquid circulating system energy consumption.

Description

Technical Field

[0001] This application relates to the field of intelligent energy consumption control in automotive tubular profile cutting production lines, and in particular to an intelligent optimization control method and system for energy consumption in automotive tubular profile cutting production lines. Background Technology

[0002] Automotive tubular profile cutting is one of the key processes in automotive parts manufacturing. It mainly involves the fixed-length cutting of various tubular profiles such as body structure tubing, anti-collision beam tubing, and chassis tubing. During the cutting process, the intense friction between the saw blade and the tubing generates a large amount of cutting heat. If it is not cooled in a timely and effective manner, it will lead to quality problems such as burns on the cut end face, increased burrs, and a sharp decrease in tool life. Therefore, the cutting production line is usually equipped with a coolant circulation system, which uses a cooling pump to deliver coolant to the cutting area for forced cooling and lubrication to ensure cutting quality and process stability.

[0003] Currently, the coolant circulation system in automotive pipe cutting production lines generally adopts a constant-speed, full-flow supply method. This means the coolant pump runs continuously at its rated speed, supplying a constant flow of coolant regardless of the cutting machine's operating status. Some energy-saving solutions implement simple start-stop control of the coolant pump based on the cutting machine's load rate; that is, it stops or reduces speed when unloaded and resumes full-speed supply under load. Other solutions select a fixed coolant pump speed from a preset parameter table based on the material and wall thickness of the pipe to be cut, achieving variable flow supply for different pipe specifications.

[0004] However, the above-mentioned existing technical solutions have obvious technical defects: the constant speed full flow method continuously consumes a lot of electrical energy throughout the production process, and the energy consumption of the cooling pump accounts for more than 60% of the total energy consumption of the auxiliary system, resulting in huge non-value-added energy waste. The fixed speed matching based on the pipe specifications only achieves a static matching of one pipe to one fixed flow, and cannot be finely adjusted according to the actual dynamic heat load of the cutting process. Summary of the Invention

[0005] The purpose of this application is to solve the technical problem that the coolant circulation system of the existing automotive tube profile cutting production line cannot be finely adjusted according to the heat load differences of different micro-process stages within a single cutting cycle, resulting in a large amount of non-value-added energy waste. This application provides an intelligent optimization control method and system for energy consumption of automotive tube profile cutting production line.

[0006] To achieve the above objectives, the present application provides a method and system for intelligent optimization control of energy consumption in an automotive tubular profile cutting production line, which adopts the following technical solution: In the first aspect, this application discloses an intelligent optimization control method for energy consumption in an automotive tubular profile cutting production line, which collects the instantaneous current signal of the cutting motor in real time and obtains the attribute information of the tubular profile being cut, including at least the material code and wall thickness value. The short-time energy characteristics and short-time change intensity characteristics of the instantaneous current signal are extracted. The short-time energy characteristics are used as the load intensity criterion, and the short-time change intensity characteristics are used as the stage switching criterion. The input is a pre-built state decision model. The current cutting process is identified through the pre-built state decision model to obtain the cutting stage. The cutting stage includes the no-load stage, the cutting-in stage, the steady-state cutting stage, and the cutting-out stage. Based on the attribute information of the pipe profile being cut, the corresponding cooling parameter group for the cutting-in stage, the cooling parameter group for the cutting-out stage, and the parameter group for the steady-state cutting stage are called from the pre-built cooling process database. The coolant supply command is generated in combination with the cutting stage. The coolant supply command includes: the first transient high-pressure impact command for the cutting-in stage, the second transient high-pressure impact command for the cutting-out stage, the steady-state maintenance command for the steady-state cutting stage, and the dormancy command for the no-load stage. Among them, the cooling parameter set for the cutting-in stage includes a first target flow rate value, the cooling parameter set for the cutting-out stage includes a second target flow rate value, and the parameter set for the steady-state cutting stage includes a steady-state maintenance flow rate value. The corresponding coolant supply command is executed according to the cutting stage, and the power change of the coolant pump is controlled according to the nonlinear ramp curve. Obtain the end face quality index of the current cut pipe profile, and compare the end face quality index with the preset quality qualified threshold range to obtain the deviation value; The first and second target flow rates corresponding to the current pipe profile attributes in the cooling process database are corrected based on the deviation values.

[0007] Preferably, a state decision model is constructed, including: The original current signal sequences of pipe profiles of different specifications and the corresponding pipe profile attribute information were obtained during the historical cutting process. The short-time energy characteristics and short-time change intensity characteristics of the original current signal sequences were extracted to obtain the feature sample set. Based on the pipe profile attribute information corresponding to each sample in the feature sample set, physical division rules are determined. The physical division rules include no-load energy threshold, mutation threshold and steady-state energy range. Automatically generate stage pseudo-labels for each sample in the feature sample set based on physical partitioning rules; The physical partitioning rules are defined as follows: The continuous interval where the short-term energy characteristic is consistently below the empty energy threshold is marked as the empty stage, and the continuous interval where the short-term energy characteristic jumps from below the empty energy threshold and the short-term mutation intensity characteristic shows a peak exceeding the mutation threshold is marked as the entry stage. The continuous interval after the cut-in phase where the short-term mutation intensity characteristics fall back to a low fluctuation level and the short-term energy characteristics remain within the steady-state energy range is marked as the steady-state cutting phase. The continuous interval in which the short-term mutation intensity characteristic at the end of the steady-state cutting stage spikes again and the short-term energy characteristic then drops sharply to below the empty energy threshold is marked as the cutting-out stage. The initial classifier is trained using a feature sample set with stage pseudo-labels to obtain the initial state decision model; The initial state decision model predicts the feature sample set, selects samples with prediction confidence higher than the preset confidence threshold, performs consistency verification between the predicted labels and the stage pseudo labels generated based on physical partitioning rules, and adds the samples that pass the verification and their predicted labels to the high confidence sample set. The initial state decision model is retrained using a high-confidence sample set. The prediction, consistency verification, and retraining processes are iteratively executed until the performance of the initial state decision model converges, thus obtaining the state decision model.

[0008] Preferably, the short-time energy characteristics and short-time mutation intensity characteristics are obtained through the following methods, specifically including: The expression for the short-time energy characteristics is: In the formula, The normalized short-time energy corresponding to the current signal of the i-th frame is used to characterize the motor load intensity within a single frame time sequence; This represents the number of sampling points for a single frame of the current signal. The instantaneous current value of the motor at the k-th sampling point within the i-th frame; The dimensionless load correction factor represents the material properties of the coupling pipe profile. Pipe profile wall thickness Dynamic value retrieval; The expression for the short-term mutation intensity characteristic is: In the formula, The short-term mutation intensity corresponding to the current signal of the i-th frame is used to characterize the temporal fluctuation and the severity of the switching of the cutting condition. These are the normalized short-time energies of the current frame and the previous frame, respectively. It is a dimensionless abrupt change calibration coefficient, which synchronously couples the material properties of the pipe profile with the dynamic values ​​of the wall thickness.

[0009] Preferred methods for constructing a pre-built cooling process database include: For different combinations of pipe profile properties, a thermal-fluid-structure coupled finite element simulation model of the cutting zone is established. The finite element simulation model uses the first target flow rate value in the cutting stage, the second target flow rate value in the cutting stage, and the steady-state maintenance flow rate value in the steady-state cutting stage as adjustable input variables, and outputs the highest temperature value of the cutting zone. With the constraint that the highest temperature in the cutting zone does not exceed the preset critical burn temperature threshold, and with the optimization objective of minimizing the weighted total flow of the first target flow rate, the second target flow rate, and the steady-state maintained flow rate, a multi-objective optimization problem is constructed. The multi-objective particle swarm optimization algorithm is used to solve the multi-objective optimization problem and obtain the Pareto optimal solution set; In the Pareto optimal solution set, the weighted total flow rate corresponding to each solution is sorted according to the preset energy consumption preference weight. The solution whose weighted total flow rate satisfies the constraint condition is selected. The first target flow rate value, the second target flow rate value, and the steady-state maintenance flow rate value contained in the solution are combined with the corresponding pipe profile attribute information to form a cooling parameter group, which is stored in the cooling process database.

[0010] Preferably, the first target flow rate value and the second target flow rate value corresponding to the current pipe profile attribute in the cooling process database are corrected according to the deviation value, including: Obtain the sequence of historical deviation values ​​recorded from the N consecutive cuts prior to the current cut operation, as well as the first target flow rate value and the second target flow rate value currently stored in the cooling process database, and concatenate them to form the current state vector, where N is a preset integer; The current state vector is input into a pre-trained traffic correction policy network, which outputs a first correction increment for the first target traffic value and a second correction increment for the second target traffic value. The first correction increment and the second correction increment are added to the current first target flow rate value and the second target flow rate value, respectively, to obtain the updated first target flow rate value and the second target flow rate value. The cooling process database is then updated based on the updated first target flow rate value and the second target flow rate value.

[0011] Preferably, before inputting the current state vector into the pre-trained traffic correction policy network, the method further includes: Input the first target flow rate value and the second target flow rate value corresponding to the current pipe profile attribute in the current cooling process database into the pre-trained cooling energy consumption prediction model to obtain the predicted cooling energy consumption value. The predicted cooling energy consumption value is appended to the current state vector to form an enhanced state vector; The enhanced state vector is input into the flow correction policy network, and the flow correction policy network outputs the correction increment. The cooling energy consumption prediction model is a time-series neural network model trained based on historical cutting data. The historical cutting data includes the first target flow rate value, the second target flow rate value, and the corresponding real-time power of the cooling pump recorded in the historical cutting operation.

[0012] Preferably, controlling the power variation of the cooling pump according to a nonlinear ramp curve includes: A pressure-stabilizing control agent is constructed based on a proximal policy optimization algorithm. The pressure-stabilizing control agent is driven by a deep neural network. The state space of the pressure-stabilizing control agent includes the real-time pressure deviation between the actual pressure value and the target pressure value of the coolant supply pipeline, the rate of change of the real-time pressure deviation, and the reference slope of the nonlinear ramp curve at the current moment. The action space of the pressure-stabilizing control agent is the adjustment coefficient of the instantaneous slope of the nonlinear ramp curve. Each time the first transient high-pressure impact command or the second transient high-pressure impact command is executed, the state space parameters at the current moment are input into the pressure stabilization control agent, and the pressure stabilization control agent outputs the slope curvature adjustment coefficient corresponding to the current moment. The slope curvature adjustment coefficient is multiplied by the reference slope to obtain the adjusted instantaneous slope, and the power change rate of the cooling pump is controlled by the adjusted instantaneous slope.

[0013] Preferably, the reward function of the pressure-stabilizing control agent is determined by the following formula: In the formula: The cumulative reward value of the pressure stabilization control agent after the end of a transient high-pressure impact process; The actual pressure value collected at time t; The target pressure value; This refers to the relative pressure deviation; This refers to the total duration of the transient high-pressure impact command; The time required from the start of the transient high-pressure impact command until the actual pressure value first enters and remains within the preset error band of the target pressure value; This is the preset reference adjustment time; This refers to the relative adjustment time; It is a weighted sum of the correction increments of the first target traffic value and the second target traffic value in the most recent output of the network according to the traffic correction policy; For collaborative reward and punishment functions, , For the preset boundary values ​​of the stable interval, when When within the stable range, ; These are the preset non-negative weighting coefficients.

[0014] Preferably, each time the first transient high-voltage impact command or the second transient high-voltage impact command is executed, it includes: When the current command is detected to switch from the first transient high-pressure impact command or the second transient high-pressure impact command to the steady-state maintenance command, the steady-state target pressure value corresponding to the steady-state maintenance command is obtained; The steady-state target pressure value is used as the target pressure value of the pressure stabilization control agent. During the preset discharge control period at the start of the switch, the instantaneous slope of the nonlinear ramp curve is used as the action output of the pressure stabilization control agent. The pressure stabilization control agent outputs the dynamically adjusted ramp curvature adjustment coefficient based on the real-time pressure deviation between the actual pressure value and the steady-state target pressure value. When the real-time pressure deviation indicates that the actual pressure value is higher than the steady-state target pressure value, the slope of the nonlinear slope curve is increased by adjusting the slope curvature to accelerate pressure release. When the actual pressure value approaches the steady-state target pressure value, the slope is decreased by adjusting the slope curvature to smoothly enter the steady state.

[0015] Secondly, this application discloses an intelligent energy consumption optimization control system for an automotive tubular profile cutting production line, which applies the intelligent energy consumption optimization control method for an automotive tubular profile cutting production line as described in the first aspect, including: The data acquisition module is used to acquire the instantaneous current signal of the cutting motor in real time and obtain the attribute information of the currently cut pipe profile. The attribute information includes at least the material code and wall thickness value. The feature extraction and stage identification module is used to extract the short-time energy features and short-time change intensity features of the instantaneous current signal. The short-time energy features are used as the load intensity criterion, and the short-time change intensity features are used as the stage switching criterion. The input is a pre-built state decision model, which identifies the current cutting process and obtains the cutting stage. The cutting stage includes the no-load stage, the cutting-in stage, the steady-state cutting stage, and the cutting-out stage. The coolant supply instruction generation module is used to call the corresponding cooling parameter group for the cutting stage, the cooling parameter group for the cutting exit stage, and the parameter group for the steady-state cutting stage from the pre-built cooling process database based on the attribute information of the pipe profile being cut, and generate a coolant supply instruction in combination with the cutting stage. The coolant supply instruction includes: the first transient high-pressure impact instruction for the cutting stage, the second transient high-pressure impact instruction for the cutting exit stage, the steady-state maintenance instruction for the steady-state cutting stage, and the dormancy instruction for the no-load stage. The execution control module is used to execute the corresponding coolant supply command according to the cutting stage and control the power change of the coolant pump according to the nonlinear ramp curve. The quality feedback correction module is used to obtain the end face quality index of the current cut pipe profile, compare the end face quality index with the preset quality qualified threshold range to obtain the deviation value, and correct the first target flow rate value and the second target flow rate value corresponding to the current pipe profile attribute in the cooling process database according to the deviation value.

[0016] Compared with the prior art, this application provides a method and system for intelligent optimization and control of energy consumption in an automotive tubular profile cutting production line, which has the following beneficial effects: 1. By analyzing the instantaneous current signal of the cutting motor in real time, short-time energy characteristics and short-time change intensity characteristics are extracted. Using a state decision model, a single cutting cycle is identified online as four micro-process stages: no-load stage, cutting-in stage, steady-state cutting stage, and cutting-out stage. Differentiated coolant supply commands are generated according to the actual heat load requirements of each stage. Specifically, a transient high-pressure impact command is triggered in the cutting-in and cutting-out stages for forced cooling, a steady-state maintenance command is executed in the steady-state cutting stage to maintain cooling with a low flow rate, and a hibernation command is executed in the no-load stage to stop ineffective supply. This achieves precise dynamic matching between coolant supply and cutting heat load, effectively reducing the overall energy consumption of the coolant circulation system. 2. By acquiring the end face quality index of the current cut pipe profile, the end face quality index is compared with the preset quality qualification threshold range to obtain the deviation value. Based on the deviation value, the corresponding first target flow rate value and second target flow rate value in the cooling process database are corrected to form a closed-loop feedback mechanism of quality perception-deviation calculation-parameter correction. This enables the coolant supply parameters to be dynamically adjusted according to the actual cutting quality. While pursuing energy saving, it effectively prevents quality defects such as end face burns and excessive burrs caused by insufficient cooling, and maintains the stability of the cutting process and the product qualification rate. 3. The cutting micro-process stages are identified online through a pre-constructed state decision model, and a pressure stabilization control agent is constructed by introducing deep reinforcement learning. When executing transient high-pressure impact commands, the instantaneous slope of the nonlinear ramp curve is dynamically adjusted according to the real-time pressure deviation, taking into account both suppressing pressure overshoot and shortening the adjustment time. At the same time, a cooling energy consumption prediction model is introduced to provide energy consumption perception capability for the flow correction strategy network, so that the coolant supply system can adapt to complex working conditions such as changes in pipe specifications, tool wear state drift, and fluctuations in pipeline characteristics. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall structure of an intelligent energy consumption optimization control method and system for an automotive tube profile cutting production line according to an embodiment of this application.

[0018] Figure 2 This is a flowchart illustrating the steps involved in constructing the state decision model using this method.

[0019] Figure 3 This is a flowchart of the steps that need to be performed before the current state vector is input into the pre-trained traffic correction policy network in this method. Detailed Implementation

[0020] The following is in conjunction with the appendix Figure 1-3 This application will be described in further detail.

[0021] Firstly, this application discloses an intelligent optimization and control method for energy consumption in an automotive tubular profile cutting production line. (Refer to...) Figure 1 A method for intelligent optimization and control of energy consumption in an automotive tubular profile cutting production line includes: S1. Real-time acquisition of the instantaneous current signal of the cutting motor and acquisition of the attribute information of the currently cut pipe profile, including at least the material code and wall thickness value; S2. Extract the short-time energy characteristics and short-time change intensity characteristics of the instantaneous current signal. Use the short-time energy characteristics as the load intensity criterion and the short-time change intensity characteristics as the stage switching criterion. Input the pre-built state decision model and identify the current cutting process through the pre-built state decision model to obtain the cutting stage. The cutting stage includes the no-load stage, the cutting-in stage, the steady-state cutting stage, and the cutting-out stage. Specifically, the acquired instantaneous current signal is first processed by framing, with each frame containing N sampling points, where N is a preset integer that can be set according to the sampling rate and the segmentation period, for example, N can be 128 or 256. Short-time energy features and short-time abrupt change intensity features are then extracted from each frame of signal.

[0022] Short-time energy characteristics are used to characterize the intensity level of the motor load at the current moment.

[0023] In one embodiment, the short-time energy characteristic is obtained by calculating the sum of squares or root mean square of the instantaneous current values ​​at each sampling point within the frame. The larger the short-time energy characteristic, the heavier the current cutting load.

[0024] Short-term mutation intensity features are used to characterize the drastic degree of energy change between adjacent frames, i.e., the transient characteristics of operating condition switching.

[0025] In one embodiment, the short-time mutation intensity feature is obtained by calculating the difference between the short-time energy features of the current frame and the previous frame. The larger the short-time mutation intensity feature, the more drastic the cutting state is, such as when the tool is cutting into or out of the pipe.

[0026] The extracted short-term energy features are used as load strength criteria, and the short-term mutation strength features are used as stage switching criteria, which are then input into the pre-built state decision model.

[0027] The cutting process comprises the following four ordered phases: No-load stage: The cutting motor runs idle, the cutter does not contact the tubular profile, and the current signal is at a low and stable no-load level; Cutting phase: The tool begins to contact and cut into the tubular profile, the current signal jumps from the no-load level, and is accompanied by severe impact characteristics; Steady-state cutting stage: The tool fully cuts into the tubular profile and enters a stable cutting state, with the current signal remaining at a high level and fluctuating little; Cutting stage: The tool is about to complete the cutting and cut out the tube profile. The current signal experiences a transient impact and then drops sharply to the no-load level.

[0028] S3. Based on the attribute information of the pipe profile being cut, call the corresponding cooling parameter group for the cutting-in stage, the cooling parameter group for the cutting-out stage, and the parameter group for the steady-state cutting stage from the pre-built cooling process database, and generate a coolant supply command in combination with the cutting stage. The coolant supply command includes: the first transient high-pressure impact command for the cutting-in stage, the second transient high-pressure impact command for the cutting-out stage, the steady-state maintenance command for the steady-state cutting stage, and the dormancy command for the no-load stage. Among them, the cooling parameter set for the cutting-in stage includes a first target flow rate value, the cooling parameter set for the cutting-out stage includes a second target flow rate value, and the parameter set for the steady-state cutting stage includes a steady-state maintenance flow rate value. Specifically, the pre-built cooling process database is a structured parameter storage and query system. The database is indexed by the pipe profile attribute information and stores various sets of cooling parameters obtained through previous process experiments or simulation optimization.

[0029] The cooling process database stores the following three sets of cooling parameters for each type or combination of pipe profile attribute information: Cooling parameter group for the cutting phase: includes the first target flow rate value, which is the target value for supplying high flow rate coolant required during the cutting phase, to provide sufficient cooling and chip removal capacity at the moment of tool entry; Cooling parameter group for the cutting stage: includes a second target flow rate value, which is the target value for supplying high flow rate coolant required in the cutting stage, to prevent end face burns and burr generation at the moment the tool cuts out; Steady-state cutting stage parameter group: includes steady-state maintenance flow rate values, which are target values ​​for supplying the lower flow rate of coolant required during the steady-state cutting stage to maintain basic cooling and lubrication.

[0030] In actual operation, based on the acquired attribute information, the system queries and retrieves the corresponding cooling parameter sets for the cutting-in stage, cutting-out stage, and steady-state cutting stage from the cooling process database. Then, combined with the identified current cutting stage, the system generates the corresponding coolant supply command. When the initiation phase is identified, a first transient high-pressure impact command is generated, which includes a first target flow value. When the cut-out phase is identified, a second transient high-pressure impact command is generated, which includes a second target flow value. When the steady-state cutting phase is identified, a steady-state maintenance command is generated, which includes the steady-state maintenance flow value. When the system is identified as being in an unloaded state, a hibernation command is generated, which instructs the cooling pump to stop operating or enter a very low-speed micro-circulation mode.

[0031] S4. Execute the corresponding coolant supply command according to the cutting stage, and control the power change of the coolant pump according to the nonlinear ramp curve. Specifically, the coolant supply command generated in step S3 is converted into a control signal that can be executed by the coolant pump inverter.

[0032] For example, switching from a hibernation command to a first transient high-pressure impact command or a second transient high-pressure impact command, or switching from a first transient high-pressure impact command or a second transient high-pressure impact command to a steady-state maintenance command, controls the power change of the cooling pump according to a preset nonlinear ramp curve.

[0033] By employing a nonlinear ramp curve for power variation control, the cooling pump speed can smoothly transition during command switching, avoiding pipeline pressure shocks and water hammer effects caused by sudden changes in flow rate. This protects the pipelines and valves of the coolant circulation system and reduces additional energy losses caused by pressure fluctuations, achieving the dual goals of energy saving and system protection.

[0034] S5. Obtain the end face quality index of the current cut pipe profile, and compare the end face quality index with the preset quality qualified threshold range to obtain the deviation value; Specifically, an online visual inspection unit is set up downstream of the cutting station. The online visual inspection unit includes a high-resolution industrial camera and an image processing module.

[0035] After a single cut is completed, the online visual inspection unit automatically acquires the end face image of the currently cut pipe profile and extracts the end face quality indicators from it using image processing algorithms.

[0036] The quality indicators of the end face can be quantifiable indicators that reflect the quality of the cut end face, such as the height of the end face burrs, the area of ​​the end face heat discoloration zone, and the end face roughness.

[0037] The preset quality acceptance threshold range is a quality standard range set in advance according to process requirements, such as an upper limit of burr height of 0.1mm and an upper limit of heat discoloration area ratio of 5%.

[0038] The extracted end-face quality indicators are compared with the preset quality pass threshold range, and the deviation value is calculated. The deviation value can be the amount by which the end-face quality indicators exceed the upper limit of the quality pass threshold range, or the amount by which they fall below the lower limit, or the amount by which they deviate from the center value of the range.

[0039] S6. Correct the first target flow rate value and the second target flow rate value corresponding to the current pipe profile attribute in the cooling process database according to the deviation value.

[0040] Specifically, the deviation value obtained in step S5 is used as a feedback signal to correct the first target flow rate value in the cutting-in stage cooling parameter group and the second target flow rate value in the cutting-out stage cooling parameter group, which are stored in the cooling process database and correspond to the current pipe profile attribute information.

[0041] In one implementation, the correction strategy is as follows: If the deviation value indicates that the end face quality is worse than the preset quality qualification threshold range, such as the burr height exceeding the upper limit, it means that the current cooling is insufficient, and the first target flow value and / or the second target flow value are increased by a preset step. If the deviation value indicates that the end face quality has been at a relatively good level within the quality qualification threshold range for a long time and the cooling pump power consumption is high, the first target flow rate value and / or the second target flow rate value can be lowered by a preset step size to further reduce cooling energy consumption.

[0042] The corrected first and second target flow rates are stored back in the cooling process database, replacing the original corresponding parameter values. When cutting pipe profiles of the same specification subsequently, the corrected parameter values ​​will be used to generate coolant supply commands.

[0043] Therefore, the parameters in the cooling process database can be adaptively adjusted according to changes in actual production conditions, such as gradual tool wear, changes in ambient temperature, and differences in the hardness of different batches of pipes, forming a continuously optimized closed-loop control process. This ensures that the coolant circulation system always operates in a low-energy-consumption state while meeting quality standards throughout the entire production lifecycle.

[0044] Furthermore, refer to Figure 2 Construct a state decision model, including: S21. Obtain the original current signal sequence of pipe profiles of different specifications and the corresponding pipe profile attribute information during the historical cutting process, extract the short-time energy characteristics and short-time change intensity characteristics of the original current signal sequence, and obtain the feature sample set. The original current signal sequence covers the complete process of the no-load stage, the cutting-in stage, the steady-state cutting stage, and the cutting-out stage.

[0045] Furthermore, short-time energy features and short-time mutation intensity features are extracted from each segment of the original current signal sequence collected, using the same framing method as online identification. The extracted feature sequences are then associated with and stored with the corresponding pipe profile attribute information to form a feature sample set.

[0046] S22. Based on the pipe profile attribute information corresponding to each sample in the feature sample set, determine the physical division rules, which include the no-load energy threshold, the mutation threshold, and the steady-state energy range. Among them, the parameters in the physical partitioning rules are dynamically determined based on the pipe profile attribute information corresponding to each sample in the feature sample set.

[0047] Meanwhile, the no-load energy threshold is determined by the no-load power of the cutting motor at the no-load speed and the cutting characteristic parameters corresponding to the current pipe profile properties; the mutation threshold is determined by the historical statistical distribution of the short-time mutation intensity characteristics of all samples in the feature sample set, for example, taking the upper quartile of the distribution; the steady-state energy range is centered on the steady-state value of the short-time energy characteristic corresponding to the theoretical net cutting power, and is extended upward and downward according to a preset proportional coefficient to form the range boundary.

[0048] S23. Automatically generate stage pseudo-labels for each sample in the feature sample set based on physical partitioning rules; The physical partitioning rules are defined as follows: The continuous interval where the short-term energy characteristic is consistently below the empty energy threshold is marked as the empty stage, and the continuous interval where the short-term energy characteristic jumps from below the empty energy threshold and the short-term mutation intensity characteristic shows a peak exceeding the mutation threshold is marked as the entry stage. The continuous interval after the cut-in phase where the short-term mutation intensity characteristics fall back to a low fluctuation level and the short-term energy characteristics remain within the steady-state energy range is marked as the steady-state cutting phase. The continuous interval in which the short-term mutation intensity characteristic at the end of the steady-state cutting stage spikes again and the short-term energy characteristic then drops sharply to below the empty energy threshold is marked as the cutting-out stage. Therefore, each sample frame in the feature sample set is automatically labeled according to the physical partitioning rules, and each sample frame is assigned a corresponding stage label.

[0049] S24. The initial classifier is trained based on the feature sample set with stage pseudo-labels to obtain the initial state decision model. The initial state decision model has initially acquired the ability to identify the micro-process stages. The initial classifier is trained by using a set of feature samples with stage pseudo-labels as training data, short-time energy features and short-time mutation intensity features as input features, and stage pseudo-labels as the expected output.

[0050] The initial classifier can be a random forest classifier, a support vector machine classifier, or a shallow neural network classifier.

[0051] S25. The initial state decision model predicts the feature sample set, selects samples whose prediction confidence is higher than the preset confidence threshold, performs consistency verification between the predicted label and the stage pseudo label generated based on the physical partitioning rule, and adds the samples that pass the verification and their predicted labels to the high confidence sample set. In the specific implementation process, the initial state decision model is used to predict the original feature sample set. The initial state decision model outputs the classification probability of each sample belonging to each cutting micro-process stage, and the maximum value among them is taken as the prediction confidence of the sample.

[0052] S26. Retrain the initial state decision model with a high-confidence sample set, and iteratively execute the prediction, consistency verification and retraining process until the performance of the initial state decision model converges, thus obtaining the state decision model.

[0053] The high-confidence sample set is used as new training data to retrain the initial state decision model. The prediction, confidence screening, consistency verification, and retraining processes are repeated to form an iterative self-training closed loop.

[0054] When the performance of the initial state decision model converges, for example, when the improvement in classification accuracy of the model on the validation set between two adjacent iterations is less than the preset convergence threshold, or when the number of iterations reaches the preset maximum number of iterations, the iteration can be stopped. The initial state decision model with converged performance is used as the final state decision model. The state decision model integrates prior knowledge of physical partitioning rules and self-training iterative optimization capabilities, and can achieve the identification of micro-process stages without manual annotation.

[0055] Furthermore, the short-time energy characteristics and short-time mutation intensity characteristics are obtained through the following methods, specifically including: Based on the positive physical relationship between instantaneous motor current and cutting load, time-domain statistical normalization is used to eliminate sampling parameter interference, and a pipe property coupling correction coefficient is introduced to achieve unified quantification of current characteristics under multiple cutting conditions. This provides stable and unified feature input for physical rule division, pseudo-label generation, and state decision model training in the cutting stage.

[0056] The instantaneous current of the motor represents the size of the load being cut, and the cumulative average of the square of the current within a single frame time sequence represents the average load intensity at that moment, thus avoiding interference from single-point current jitter.

[0057] The basic standardized short-time energy statistics are as follows: the sum of the squares of the currents at all sampling points within a single frame is then averaged to eliminate the influence of the number of sampling points on the characteristic values, ensuring that the characteristic values ​​are only related to the actual load intensity.

[0058] Based on the absolute value of the difference between short-time energy values ​​of adjacent time-series frames, the dynamic fluctuation amplitude of the load can be accurately characterized, representing the severity of the switching between operating conditions. To unify the abrupt change identification threshold for different pipe materials and eliminate the amplitude offset of abrupt change characteristics caused by differences in pipe specifications, a pipe material correction coefficient is introduced to offset the load reference offset caused by different materials and wall thicknesses. The expression for the short-time energy characteristic is thus obtained as follows: In the formula, The normalized short-time energy corresponding to the current signal of the i-th frame is used to characterize the motor load intensity within a single frame time sequence; The number of sampling points for a single frame of current signal, dimensionless; Let be the instantaneous motor current value at the k-th sampling point within the i-th frame, with dimensions in A; The dimensionless load correction factor represents the material properties of the coupling pipe profile. Pipe profile wall thickness Dynamic value retrieval; After overall calculation Dimensions are unified as .

[0059] The expression for the short-term mutation intensity characteristic is: In the formula, The short-term mutation intensity corresponding to the current signal of the i-th frame is used to characterize the temporal fluctuation and the severity of the switching of the cutting condition. These are the normalized short-time energies of the current frame and the previous frame, respectively. All dimensions are The dimensions of short-term energy characteristics remain consistent; It is a dimensionless abrupt change calibration coefficient, which synchronously couples the material properties of the pipe profile with the dynamic values ​​of the wall thickness.

[0060] Among them, the higher the hardness of the pipe material and the greater the wall thickness, the higher the cutting reference load, and the corresponding... , The values ​​are increased synchronously, and the excessively high original characteristic amplitude is normalized and calibrated so that the no-load threshold, mutation threshold and steady-state energy range of pipes of different specifications have a universal judgment standard.

[0061] Furthermore, the methods for constructing the pre-built cooling process database include: S31. For combinations of different pipe profile properties, establish a thermal-fluid-structure coupled finite element simulation model for the cutting zone. The finite element simulation model uses the first target flow rate value in the cutting-in stage, the second target flow rate value in the cutting-out stage, and the steady-state maintenance flow rate value in the steady-state cutting stage as adjustable input variables, and outputs the highest temperature value in the cutting zone. The finite element simulation model couples the interactions of three physical fields: frictional heat generation and plastic deformation heat generation in the contact area between the tool and the tubular profile, coolant flow and convective heat transfer in the kerf, and the stress-strain response of the tool and tubular profile under cutting force. The finite element simulation model uses the first target flow rate value in the entry stage, the second target flow rate value in the exit stage, and the steady-state maintenance flow rate value in the steady-state cutting stage as adjustable input variables. These three flow rate values ​​serve as boundary conditions for the simulation model, applied to the coolant inlet at the corresponding stage. The finite element simulation model numerically solves the coupled control equations and outputs the highest temperature value in the cutting zone. The mapping relationship can be expressed as: In the formula, The first target flow rate value; The second target flow rate value; To maintain the flow rate in steady state, All units are L / min; This represents the highest temperature value in the cutting zone, in °C. This represents the nonlinear mapping function determined by the finite element simulation model.

[0062] By using finite element simulation models, the cooling effect under different combinations of cooling parameters can be predicted with high accuracy without the need for actual trial cutting, providing reliable simulation samples for subsequent optimization.

[0063] S32. With the constraint that the highest temperature value in the cutting zone does not exceed the preset critical burn temperature threshold, and with the optimization objective of minimizing the weighted total flow rate of the first target flow rate, the second target flow rate, and the steady-state maintained flow rate, a multi-objective optimization problem is constructed. The constraint condition for the multi-objective optimization problem is set as follows: the highest temperature value in the cutting zone does not exceed the preset critical burn temperature threshold.

[0064] The critical burn temperature threshold is a temperature limit predetermined based on the thermophysical properties of the tubular profile material. Exceeding this temperature will lead to unacceptable quality defects such as burns on the cut end face and changes in the metallographic structure of the material.

[0065] The constraint is set as follows: the highest temperature value in the cutting zone does not exceed the preset critical burn temperature threshold, that is: In the formula: The critical burn temperature threshold, in °C, is predetermined based on the thermophysical properties of the tubular material. For example, for Q235 carbon structural steel, the critical burn temperature threshold can be set to 750℃ to 800℃; for 6061 aluminum alloy, the critical burn temperature threshold can be set to 350℃ to 400℃.

[0066] The optimization objective is set as: minimizing the weighted total flow of the first target flow value, the second target flow value, and the steady-state maintenance flow value, expressed as: In the formula: This is a normalized weighted total flow, dimensionless. The preset reference flow rate value is the rated flow rate or the commonly used maximum design flow rate of the cooling pump, in L / min. The preset dimensionless weighting coefficients correspond to the time proportions or energy consumption contribution proportions of the entry phase, exit phase, and steady-state cutting phase in the complete cutting cycle, respectively, satisfying the following conditions: ; Therefore, using reference traffic Using the normalized weighted total flow rate as a benchmark, the degree of consumption relative to the rated liquid supply capacity at each stage is measured, and a dimensionless energy consumption evaluation index is obtained after weighting. The smaller the value, the less coolant is consumed relative to the coolant pump's supply capacity to complete a single cut, resulting in lower energy consumption. Normalization eliminates the impact of differences in the absolute flow rate of different pipe specifications on the weighting settings, ensuring the preset weighting coefficients are universally applicable.

[0067] Therefore, the optimization problem can be formulated as: satisfying... ≤ Under the constraints, find ways to make the normalized weighted total flow... Decreasing flow combinations ( , , ).

[0068] Furthermore, the weighted total flow rate is the sum of the three target flow rate values ​​multiplied by their respective weighting coefficients. These weighting coefficients reflect the proportion of time or energy consumption contribution of each stage's flow rate within the entire cutting cycle. The smaller the weighted total flow rate, the less coolant circulation is required to complete one full cut, and the lower the total energy consumption of the cooling pump.

[0069] It should be noted that this optimization problem is a typical multivariable, constrained single-objective optimization problem. However, since the simulation model is a black-box function, an intelligent optimization algorithm is required to solve it, which will not be detailed here.

[0070] S33. Use the multi-objective particle swarm optimization algorithm to solve the multi-objective optimization problem and obtain the Pareto optimal solution set; In this embodiment, a multi-objective particle swarm optimization algorithm is used to solve the above optimization problem. In the algorithm initialization stage, a set of particles is randomly generated. Each particle represents a set of cooling parameter combinations, namely the first target flow rate value, the second target flow rate value, and the steady-state maintenance flow rate value, and is assigned an initial position and velocity within the value range.

[0071] Taking a pipe profile with material code Q235 and wall thickness of 2mm as an example, the preset reference flow rate is... The flow rate is 15 L / min, with a weighting factor of 15 L / min. Critical burn temperature threshold The Pareto optimal solution set is shown in Table 1 below: Table 1 In each iteration, the highest temperature value of the cutting zone corresponding to each particle is calculated through the finite element simulation model to determine whether the constraint conditions are met, and the weighted total flow rate is calculated as the fitness value. The particle updates its velocity and position based on its own historical best position and global best position. After multiple iterations, the algorithm converges and outputs the Pareto optimal solution set.

[0072] The Pareto optimal solution set is a set of multiple candidate solutions that satisfy the constraints and are optimal between the weighted total flow rate and the highest temperature in the cutting zone, providing a diverse selection space for subsequent parameter screening.

[0073] S34. In the Pareto optimal solution set, the weighted total flow rate corresponding to each solution is sorted according to the preset energy consumption preference weight. The solution whose weighted total flow rate satisfies the constraint condition is selected. The first target flow rate value, the second target flow rate value, and the steady-state maintenance flow rate value contained in the solution are combined with the corresponding pipe profile attribute information as a set of cooling parameters and stored in the cooling process database.

[0074] Following the above process, simulation modeling, optimization, and parameter selection are performed one by one for each combination of attribute information in the pipe profile specification list until the cooling process database covers all common specifications. When the completed cooling process database is running on the production line, the corresponding parameter group is directly called based on the attribute information of the currently being cut pipe profile.

[0075] Through the above construction method, the cooling parameters of each group in the cooling process database have been verified by thermal-fluid-solid coupling analysis and multi-objective optimization in the simulation environment, ensuring that the coolant supply flow rate at each stage is at an optimal level while meeting the cutting quality requirements, thus guaranteeing the energy consumption optimization effect of the coolant circulation system from the source.

[0076] Furthermore, based on the deviation value, the first target flow rate value and the second target flow rate value corresponding to the current pipe profile attribute in the cooling process database are corrected, including: S61. Obtain the sequence of historical deviation values ​​recorded from the N consecutive cuts prior to the current cutting operation, and the first target flow rate value and the second target flow rate value currently stored in the cooling process database, and concatenate them to form the current state vector, where N is a preset integer; Specifically, after each cut, the deviation value is recorded and stored in the sliding window history. When parameter correction is needed, the historical deviation value sequence recorded from the previous N consecutive cuts is extracted, for example, N=5 or N=10. This historical deviation value sequence can reflect the recent evolution trend of cutting quality. Simultaneously, the historical deviation value sequence is concatenated with the current first target flow rate value and the second target flow rate value to form a multi-dimensional vector, which serves as the current state vector. The current state vector describes the current cooling parameter configuration and recent cutting quality performance.

[0077] In one embodiment, if N=5, the historical deviation value sequence contains 5 scalar values. These are then concatenated with the current first target flow value and the second target flow value, resulting in a current state vector with a dimension of 7. The values ​​of each dimension can be normalized before being input into the network to improve the stability and convergence speed of network training.

[0078] S62. Input the current state vector into the pre-trained traffic correction policy network, and the traffic correction policy network outputs the first correction increment for the first target traffic value and the second correction increment for the second target traffic value. Specifically, the flow correction strategy network is a pre-trained deep neural network model used to output the optimal parameter correction direction and magnitude based on the current state. The flow correction strategy network adopts an actor-critic architecture and is trained offline based on a deep deterministic policy gradient algorithm.

[0079] Furthermore, during the training process, a simulation model simulating the real cutting environment is built or historical cutting records are directly utilized.

[0080] If the end face quality index of the next pipe cut after correction falls within the preset quality qualification threshold range, a positive reward will be given. The smaller the absolute value of the first correction increment and the second correction increment, that is, the smaller the flow rate increase, the greater the positive reward. If the quality indicators exceed the threshold range, a negative penalty is imposed. Through extensive trial and error iterations, the policy network gradually learns to output the smallest possible correction increment while maintaining acceptable cutting quality, thereby suppressing unnecessary increases in coolant flow and reducing system energy consumption.

[0081] After training, the policy network is deployed to the edge controller of the cutting line for online inference. During online operation, the current state vector is input into the policy network, and the network outputs a first and second correction increment after forward computation. Since energy-saving preferences have been internalized into the policy during training, its output has the characteristic of suppressing over-modulation oscillations.

[0082] S63. Add the first correction increment and the second correction increment to the current first target flow rate value and the second target flow rate value respectively to obtain the updated first target flow rate value and the second target flow rate value, and update the cooling process database based on the updated first target flow rate value and the second target flow rate value.

[0083] Specifically, let the current first target flow rate before correction be _____. The current second target flow value is The first correction increment of the policy network output is... The second correction increment is The updated first target traffic value is The updated second target traffic value is .

[0084] Furthermore, the first target flow rate value in the cooling parameter group of the entry stage corresponding to the current pipe profile attribute in the cooling process database is updated to... Update the second target flow rate value in the cooling parameter group for the cut-out stage to Furthermore, the updated parameters will automatically take effect when cutting pipe profiles of the same specification in subsequent operations.

[0085] Through the aforementioned deep reinforcement learning-driven adaptive correction mechanism, key parameters in the cooling process database can be continuously optimized during long-term operation, avoiding drift in cooling demand caused by factors such as tool wear and changes in ambient temperature. Furthermore, the correction process itself has an energy-saving bias, which minimizes unnecessary increases in coolant flow while ensuring cutting quality, thereby further reducing the overall energy consumption of the coolant circulation system.

[0086] Furthermore, refer to Figure 3 Before inputting the current state vector into the pre-trained traffic correction policy network, the following steps are also included: S621. Input the first target flow rate value and the second target flow rate value corresponding to the current pipe profile attribute in the current cooling process database into the pre-trained cooling energy consumption prediction model to obtain the predicted cooling energy consumption value. The cooling energy consumption prediction model is a time-series neural network model trained based on historical cutting data. The construction method is as follows: during the historical operation of the cutting production line, the actual first target flow rate value, second target flow rate value, and corresponding real-time power of the cooling pump are recorded in each cutting operation.

[0087] The first and second target flow rates from historical segmentation data are used as input features, and the corresponding real-time power of the cooling pump is used as the output label to conduct supervised training on the temporal neural network model.

[0088] Therefore, the temporal neural network model can use long short-term memory network or gated recurrent unit network. Its advantage lies in its ability to capture the dynamic response characteristics of the impact of flow parameter changes on energy consumption, such as the temporal relationship between power overshoot during the acceleration process of the cooling pump and the steady-state power after stabilization.

[0089] After training, the cooling energy consumption prediction model learned the mapping relationship between the first target flow rate value, the second target flow rate value, and the real-time power of the cooling pump.

[0090] During online operation, the first target flow rate value and the second target flow rate value corresponding to the current pipe profile attributes are read from the cooling process database and input into the cooling energy consumption prediction model to output the predicted cooling energy consumption value. The predicted cooling energy consumption value represents the cooling pump energy expected to be consumed when performing a complete cut under the current cooling parameter configuration.

[0091] S622. Patch the predicted cooling energy consumption value into the current state vector to form an enhanced state vector; Specifically, the current state vector is formed by concatenating the historical deviation value sequence with the current first target flow rate value and the second target flow rate value. In this embodiment, the predicted cooling energy consumption value is added as a new dimension to the end of the current state vector to form an enhanced state vector.

[0092] In one embodiment, if the dimension of the current state vector is 7, then after concatenating the predicted cooling energy consumption value, the dimension of the enhanced state vector is expanded to 8.

[0093] S623. Input the enhanced state vector into the flow correction policy network, and the flow correction policy network outputs the correction increment. The cooling energy consumption prediction model is a time-series neural network model trained based on historical cutting data. The historical cutting data includes the first target flow rate value, the second target flow rate value, and the corresponding real-time power of the cooling pump recorded in the historical cutting operation.

[0094] The structure and training method of the traffic correction policy network are the same as those in steps S621-S623 above, except that its input is replaced by the enhanced state vector instead of the current state vector.

[0095] Because the enhanced state vector contains the predicted cooling energy consumption value, the policy network can perceive the energy consumption level corresponding to the current cooling parameters when performing forward inference.

[0096] During the offline training phase, the reward function of the policy network remains consistent with that described in steps S621-S623 above, that is, it simultaneously considers the quality of the corrected cut and the magnitude of the correction increment.

[0097] Since the enhanced state vector already contains the predicted cooling energy consumption value, the policy network can learn the following energy-saving strategy during training: when the predicted cooling energy consumption value is already low, even if the quality deviation is positive, that is, the cooling is slightly insufficient, a small correction increment should be output to avoid unnecessary energy consumption increase. A larger correction increment is allowed only when the predicted cooling energy consumption is high and the quality deviation is large. Thus, the strategy network can proactively suppress excessive energy consumption growth while ensuring acceptable cutting quality, achieving an intelligent balance between quality constraints and energy-saving goals.

[0098] Furthermore, controlling the power variation of the cooling pump according to a nonlinear ramp curve includes: S41. A pressure smooth control agent is constructed based on the near-end strategy optimization algorithm. The pressure smooth control agent is driven by a deep neural network. The state space of the pressure smooth control agent includes the real-time pressure deviation between the actual pressure value and the target pressure value of the coolant supply pipeline, the rate of change of the real-time pressure deviation, and the reference slope of the nonlinear ramp curve at the current moment. The action space of the pressure smooth control agent is the adjustment coefficient of the instantaneous slope of the nonlinear ramp curve. The three dimensions of the state space are defined as follows: The expression for real-time pressure deviation is: In the formula, The actual pressure value collected at time t by a pressure sensor deployed on the coolant supply line; The target pressure value is determined by a preset flow-pressure mapping relationship based on the target flow value corresponding to the currently executed command, namely the first transient high-pressure impact command or the second transient high-pressure impact command. The unit is Pa or bar, which is used to reflect the difference between the current pressure and the target pressure.

[0099] Rate of change of real-time pressure deviation: The pressure deviation can be approximated by the difference between adjacent time points, which can be used to reflect the evolution trend of the pressure deviation, enabling the agent to predict the pressure trend and adjust the control strategy in advance. The baseline slope of the nonlinear ramp curve at the current moment: , that is, the theoretical derivative of the preset nonlinear ramp curve at the current moment, is used to characterize the preset value of the cooling pump power change rate without considering pressure feedback.

[0100] It should be noted that the above three state variables are all normalized before being input into the policy network, so that the values ​​of each dimension are of similar magnitude, thereby improving the convergence speed and stability of network training.

[0101] The action space is the adjustment coefficient for the instantaneous slope of the nonlinear ramp curve, denoted as . Its physical meaning is a multiplicative adjustment factor based on the baseline slope, and the range of values ​​for the adjustment coefficient can be preset to... A value greater than 1 indicates an acceleration power change based on the reference slope, while a value less than 1 indicates a deceleration power change.

[0102] In the multi-objective optimization solution for each combination of pipe profile properties, in addition to the constraints and optimization objectives, a pressure stability index is added, which is evaluated based on the cumulative reward value of the pressure stability control agent after a transient high-pressure impact process.

[0103] Furthermore, the reward function of the stress-stabilizing control agent is determined by the following formula: In the formula: This refers to the cumulative reward value of the pressure-smoothing control agent after a transient high-pressure impact process. The higher the cumulative reward value, i.e., the closer it is to zero, the smoother the pressure transition process. It is treated as an objective or constraint to be improved during optimization. The actual pressure value collected at time t by a pressure sensor deployed in the coolant supply line, in Pascals or Bars; The target pressure value is determined based on the target flow rate corresponding to the currently executed transient high-pressure impact command, i.e., the first target flow rate value or the second target flow rate value, through a preset flow-pressure mapping relationship. The unit is... Consistent; This refers to the relative pressure deviation; This refers to the total duration of the transient high-voltage impact command, in seconds. It measures the cumulative sum of squares of pressure deviations during the entire instruction execution period, in seconds; It should be noted here that, since the integrand is dimensionless, the integral result has the dimension of seconds, but this term is related to the weighting coefficient. After multiplication, the product together with other terms forms a dimensionless sum. When setting the parameters, the conversion of units needs to be considered, for example, let have Dimensionless, or dimensionless by dividing by a normalized time constant, can be achieved by dividing the integral result by a normalized time constant. In practice, the integral result can be divided by... To achieve complete dimensionlessness, this embodiment uses weighting coefficients to maintain formula simplicity. The normalization of the time dimension has been implicitly included, meaning that the numerical value has been made dimensionless, which will not be elaborated further.

[0104] The time required from the start of the transient high-pressure impact command until the actual pressure value first enters and remains within the preset error band of the target pressure value; This is the preset reference adjustment time; This refers to the relative adjustment time; It is the weighted sum of the correction increments of the first target traffic value and the second target traffic value most recently output by the network according to the traffic correction policy, i.e. , Preset weighting coefficients; For collaborative reward and punishment functions, , For the preset boundary values ​​of the stable interval, when When within the stable range, ; The preset non-negative weighting coefficients are used to balance the relative importance of stress bias penalty, adjustment time penalty, and co-adaptation penalty in the total reward.

[0105] S42. Each time the first transient high-pressure impact command or the second transient high-pressure impact command is executed, the state space parameters at the current moment are input into the pressure stabilization control agent, and the pressure stabilization control agent outputs the slope curvature adjustment coefficient corresponding to the current moment. During online operation, when the coolant supply command switches to either the first transient high-pressure impact command or the second transient high-pressure impact command, the controller collects the actual pressure value at the current moment in each control cycle, calculates the real-time pressure deviation and its rate of change, and obtains the reference slope at the current moment from the preset nonlinear ramp curve, assembling it into a state vector input pressure smoothing control agent. After forward calculation by the strategy network, the ramp curvature adjustment coefficient is output. .

[0106] S43. Multiply the slope curvature adjustment coefficient by the reference slope to obtain the adjusted instantaneous slope, and use the adjusted instantaneous slope to control the power change rate of the cooling pump. The adjusted instantaneous slope is calculated using the following expression: In the formula, The adjusted instantaneous slope; Slope curvature adjustment coefficient; The baseline slope is used.

[0107] The power change rate of the cooling pump is controlled by adjusting the instantaneous slope, that is, by adjusting the output frequency of the cooling pump through the frequency converter, so that the actual power increases or decreases according to the slope.

[0108] When the real-time pressure deviation indicates that the actual pressure is lower than the target pressure, the agent tends to output an adjustment coefficient greater than 1 to increase the instantaneous slope and accelerate the pressure rise so that the pressure approaches the target value as soon as possible. When the real-time pressure deviation indicates that the actual pressure has approached or exceeded the target pressure, the agent outputs an adjustment coefficient less than 1 to slow down the pressure rise rate and suppress pressure overshoot.

[0109] Offline training is conducted in a simulation environment equivalent to the characteristics of a real coolant supply pipeline. The reward function is constructed based on the overshoot of the actual pressure value exceeding the preset pressure fluctuation range and the settling time required to reach the target pressure value. This is used to train the pressure smooth control agent to output a ramp curvature adjustment coefficient that balances the suppression of pressure overshoot and the reduction of settling time.

[0110] The offline training of the pressure stabilization control agent was completed in a simulation environment. It was established based on the dynamic characteristics of the real coolant supply pipeline, including pipeline friction resistance, inertia, capacitance effect, and the flow-pressure-speed characteristics of the coolant pump. It can accurately simulate the dynamic response of pipeline pressure under a given power change rate.

[0111] During training, a target pressure value and an initial actual pressure value are randomly initialized to simulate a pressure rise transition process. In each control cycle, the pressure stabilization control agent outputs an adjustment coefficient based on the current state. The simulation environment updates the cooling pump power and pipeline pressure based on the adjusted instantaneous slope and calculates the reward signal. The reward function is designed to encourage the agent to minimize pressure overshoot during the pressure rise process, while minimizing the settling time required from the start of pressure rise until the actual pressure enters and stabilizes within the preset error band of the target pressure.

[0112] Specifically, the cumulative reward function for a single pressure boosting process can be expressed as a combination of pressure overshoot penalty and settling time penalty. For example, the reward value can be set as the integral of the absolute value of the pressure deviation and the negative weighted sum of the settling time. The pressure stabilization control agent continuously adjusts the deep neural network parameters through repeated interactions with the simulation environment using a proximal policy optimization algorithm, ultimately learning to output a ramp curvature adjustment coefficient that can both suppress pressure overshoot and shorten the settling time.

[0113] After training, the policy network is deployed to the edge controller of the cutting production line for online inference. Through the aforementioned AI-based pressure smoothing control method, the power transition process of the nonlinear ramp curve can be adaptively and dynamically adjusted according to the actual pipeline pressure state, further reducing pipeline energy loss and equipment wear caused by pressure surges, and improving the energy efficiency of the coolant circulation system.

[0114] Furthermore, each time the first transient high-voltage impact command or the second transient high-voltage impact command is executed, it includes: S421. When it is detected that the current command has switched from the first transient high pressure impact command or the second transient high pressure impact command to the steady state maintenance command, the steady state target pressure value corresponding to the steady state maintenance command is obtained. During the cutting process, when the state decision model identifies a change in the current micro-process stage of the cutting, it switches the currently executed coolant supply command from the first transient high-pressure impact command or the second transient high-pressure impact command to the steady-state maintenance command.

[0115] Upon detecting the instruction switching event, the steady-state target pressure value corresponding to the steady-state maintenance instruction is first obtained.

[0116] The steady-state target pressure value is determined by a preset flow-pressure mapping relationship based on the steady-state maintenance flow rate value. This mapping relationship is the same as that used in the pressure boosting stage and can be pre-calibrated using the characteristic curve of the cooling pump and the pipeline resistance characteristics. The steady-state target pressure value is usually significantly lower than the target pressure value corresponding to the transient high-pressure impact command, so the pressure relief process involves a significant pressure drop.

[0117] S422. The steady-state target pressure value is used as the target pressure value of the pressure stabilization control agent. During the preset release control period at the start of the switch, the instantaneous slope of the nonlinear ramp curve is used as the action output of the pressure stabilization control agent. The pressure stabilization control agent outputs the dynamically adjusted ramp curvature adjustment coefficient based on the real-time pressure deviation between the actual pressure value and the steady-state target pressure value. In the specific implementation process, during the preset release control period at the start of the instruction switching, the pressure stabilization control agent is reconfigured with its control parameters.

[0118] Specifically, the target pressure value is updated from the target pressure value corresponding to the original transient high-pressure impact command to the steady-state target pressure value corresponding to the steady-state maintenance command, so that the control target of the intelligent agent changes from maintaining high pressure to transitioning to low pressure.

[0119] Furthermore, the instantaneous slope of the nonlinear ramp curve at the current moment is taken as the action output object of the pressure stabilization control agent, that is, the ramp curvature adjustment coefficient output by the pressure stabilization control agent directly acts on the descending slope.

[0120] The release control period begins at the moment of command switching and continues until the actual pressure value first enters and stabilizes within the preset error band of the steady-state target pressure value. The duration of the release control period is dynamically determined by the pressure stabilization control agent by adjusting the descent slope; the faster the adjustment, the shorter the period.

[0121] Within each control cycle, the pressure stabilization control agent obtains the real-time pressure deviation between the current actual pressure value and the steady-state target pressure value. Combining the rate of change of the real-time pressure deviation and the current benchmark descent slope, it performs forward calculation through the policy network and outputs the dynamically adjusted slope curvature adjustment coefficient.

[0122] S423. When the real-time pressure deviation indicates that the actual pressure value is higher than the steady-state target pressure value, the slope of the nonlinear slope curve is increased by the slope curvature adjustment coefficient to accelerate pressure release. When the actual pressure value approaches the steady-state target pressure value, the slope is reduced by the slope curvature adjustment coefficient to smoothly enter the steady state.

[0123] Specifically, the pressure stabilization control agent's adjustment strategy during the depressurization phase is symmetrical to, but in the opposite direction to, that during the pressure increase phase.

[0124] In the initial stage of pressure relief, the actual pressure value is much higher than the steady-state target pressure value, and the real-time pressure deviation is positive and has a large amplitude. At this time, the agent tends to output a ramp curvature adjustment coefficient greater than 1, increasing the descent slope of the nonlinear ramp curve, so that the power of the cooling pump decreases at a faster speed, thereby accelerating the release of pipeline pressure, shortening the unnecessary duration of high pressure, and reducing the extra energy consumption during high pressure operation.

[0125] Through the adaptive pressure relief process, the pressure stabilization control agent shortens the duration of the high-pressure state while ensuring pipeline safety and eliminating impact, thereby achieving energy-saving optimization of the pressure relief process.

[0126] It should be noted that the pressure stabilization control agent uses the same set of deep neural network parameters in both the pressure boosting and depressurization phases, i.e., the same agent. The only difference is that the target pressure value and the object of the action output are different in the two control scenarios. The purpose is to enable the agent to learn a symmetrical control strategy during the training phase, thereby reducing the implementation complexity and maintenance cost of the system.

[0127] Secondly, embodiments of this application disclose an intelligent energy consumption optimization control system for an automotive tubular profile cutting production line, which applies the intelligent energy consumption optimization control method for an automotive tubular profile cutting production line as described in the first aspect, including: The data acquisition module is used to acquire the instantaneous current signal of the cutting motor in real time and obtain the attribute information of the currently cut pipe profile. The attribute information includes at least the material code and wall thickness value. The hardware components of the data acquisition module include a high-frequency Hall current sensor deployed on the spindle drive unit of the cutting host, a barcode scanning and identification device deployed upstream of the cutting station, or a communication interface with the production line manufacturing execution system.

[0128] The high-frequency Hall current sensor acquires the instantaneous current signal of the cutting motor in real time at a sampling rate of not less than 1kHz. The instantaneous current signal can reflect the load change of the cutting motor in real time during the cutting process.

[0129] The barcode scanning device or communication interface is used to obtain the attribute information of the currently cut pipe profile.

[0130] The data acquisition module sends the acquired instantaneous current signal and attribute information to the feature extraction and stage identification module and the coolant supply command generation module, respectively.

[0131] The feature extraction and stage identification module is used to extract the short-time energy features and short-time change intensity features of the instantaneous current signal. The short-time energy features are used as the load intensity criterion, and the short-time change intensity features are used as the stage switching criterion. The input is a pre-built state decision model, which identifies the current cutting process and obtains the cutting stage. The cutting stage includes the no-load stage, the cutting-in stage, the steady-state cutting stage, and the cutting-out stage. The feature extraction and stage identification module includes a signal framing processing unit, a feature calculation unit, and a pre-trained state decision model.

[0132] The signal framing processing unit divides the input instantaneous current signal into frames, with each frame containing a preset number of sampling points.

[0133] The feature calculation unit calculates short-time energy features and short-time abrupt change intensity features for each frame of signal.

[0134] The state decision model is based on the classification model obtained during the construction process and is deployed in the feature extraction and stage identification module for online inference. The feature extraction and stage identification module sends the identified segmentation stage to the coolant supply command generation module and the execution control module in real time.

[0135] The coolant supply instruction generation module is used to call the corresponding cooling parameter group for the cutting stage, the cooling parameter group for the cutting exit stage, and the parameter group for the steady-state cutting stage from the pre-built cooling process database based on the attribute information of the pipe profile being cut, and generate a coolant supply instruction in combination with the cutting stage. The coolant supply instruction includes: the first transient high-pressure impact instruction for the cutting stage, the second transient high-pressure impact instruction for the cutting exit stage, the steady-state maintenance instruction for the steady-state cutting stage, and the dormancy instruction for the no-load stage. The coolant supply instruction generation module has an embedded cooling process database. The cooling process database is indexed by material code and wall thickness, and stores the cooling parameter groups for the cutting-in stage, the cutting-out stage, and the steady-state cutting stage corresponding to each pipe profile specification.

[0136] The coolant supply command generation module receives the attribute information of the currently cut pipe profile provided by the data acquisition module, queries and calls the corresponding cooling parameter group from the cooling process database, combines the current cutting stage provided by the feature extraction and stage identification module, and sends the coolant supply command to the execution control module. The execution control module is used to execute the corresponding coolant supply command according to the cutting stage and control the power change of the coolant pump according to the nonlinear ramp curve. The execution control module includes an instruction parsing unit, a ramp curve generation unit, and a frequency converter drive unit.

[0137] The instruction parsing unit receives the coolant supply instruction sent by the coolant supply instruction generation module and extracts the instruction type and the corresponding target flow rate value.

[0138] The ramp curve generation unit has preset parameters for nonlinear ramp curves, which can be S-curves or exponential curves, used to generate smooth power transition trajectories during command switching.

[0139] In a preferred embodiment, the execution control module further includes a pressure stabilization control agent, which dynamically adjusts the ramp curvature based on real-time pressure feedback during instruction switching to further suppress pressure overshoot and shorten adjustment time.

[0140] The inverter drive unit converts the adjusted instantaneous slope into a frequency control signal for the cooling pump inverter, which is then sent to the cooling pump inverter via industrial Ethernet or fieldbus to regulate the speed and power of the cooling pump.

[0141] The quality feedback correction module is used to obtain the end face quality index of the current cut pipe profile, compare the end face quality index with the preset quality qualified threshold range to obtain the deviation value, and correct the first target flow rate value and the second target flow rate value corresponding to the current pipe profile attribute in the cooling process database according to the deviation value.

[0142] The quality feedback correction module is communicatively connected to the online visual inspection unit located downstream of the cutting station.

[0143] The online visual inspection unit includes a high-resolution industrial camera and an image processing module. After a single cut, it automatically acquires the end face image of the currently cut tube profile and extracts end face quality indicators from it through image processing algorithms. The end face quality indicators can be end face burr height, end face heat discoloration area area, or end face roughness, etc.

[0144] The quality feedback correction module compares the received end-face quality indicators with the preset quality pass threshold range and calculates the deviation value.

[0145] In one implementation, the quality feedback correction module has an embedded flow correction strategy network for outputting correction increments based on deviation values ​​and updating the first target flow value and the second target flow value in the cooling process database.

[0146] In another preferred embodiment, the quality feedback correction module further includes a cooling energy consumption prediction model for providing energy consumption awareness information when making correction decisions.

[0147] The corrected parameters are stored back into the cooling process database embedded in the coolant supply command generation module, achieving closed-loop optimization of the parameters.

[0148] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for intelligent optimization and control of energy consumption in an automotive tubular profile cutting production line, characterized in that, include: The instantaneous current signal of the cutting motor is collected in real time, and the attribute information of the pipe profile being cut is obtained. The attribute information includes at least the material code and wall thickness value. The short-time energy characteristics and short-time change intensity characteristics of the instantaneous current signal are extracted. The short-time energy characteristics are used as the load intensity criterion, and the short-time change intensity characteristics are used as the stage switching criterion. The input is a pre-built state decision model. The current cutting process is identified through the pre-built state decision model to obtain the cutting stage. The cutting stage includes the no-load stage, the cutting-in stage, the steady-state cutting stage, and the cutting-out stage. Based on the attribute information of the pipe profile being cut, the corresponding cooling parameter group for the cutting-in stage, the cooling parameter group for the cutting-out stage, and the parameter group for the steady-state cutting stage are called from the pre-built cooling process database. The coolant supply command is generated in combination with the cutting stage. The coolant supply command includes: the first transient high-pressure impact command for the cutting-in stage, the second transient high-pressure impact command for the cutting-out stage, the steady-state maintenance command for the steady-state cutting stage, and the dormancy command for the no-load stage. Among them, the cooling parameter set for the cutting-in stage includes a first target flow rate value, the cooling parameter set for the cutting-out stage includes a second target flow rate value, and the parameter set for the steady-state cutting stage includes a steady-state maintenance flow rate value. The corresponding coolant supply command is executed according to the cutting stage, and the power change of the coolant pump is controlled according to the nonlinear ramp curve. Obtain the end face quality index of the current cut pipe profile, and compare the end face quality index with the preset quality qualified threshold range to obtain the deviation value; The first and second target flow rates corresponding to the current pipe profile attributes in the cooling process database are corrected based on the deviation values.

2. The intelligent energy consumption optimization and control method for an automotive tube profile cutting production line according to claim 1, characterized in that, Constructing a state decision model includes: The original current signal sequences of pipe profiles of different specifications and the corresponding pipe profile attribute information were obtained during the historical cutting process. The short-time energy characteristics and short-time change intensity characteristics of the original current signal sequences were extracted to obtain the feature sample set. Based on the pipe profile attribute information corresponding to each sample in the feature sample set, physical division rules are determined. The physical division rules include no-load energy threshold, mutation threshold and steady-state energy range. Automatically generate stage pseudo-labels for each sample in the feature sample set based on physical partitioning rules; The physical partitioning rules are defined as follows: The continuous interval where the short-term energy characteristic is consistently below the empty energy threshold is marked as the empty stage, and the continuous interval where the short-term energy characteristic jumps from below the empty energy threshold and the short-term mutation intensity characteristic shows a peak exceeding the mutation threshold is marked as the entry stage. The continuous interval after the cut-in phase where the short-term mutation intensity characteristics fall back to a low fluctuation level and the short-term energy characteristics remain within the steady-state energy range is marked as the steady-state cutting phase. The continuous interval in which the short-term mutation intensity characteristic at the end of the steady-state cutting stage spikes again and the short-term energy characteristic then drops sharply to below the empty energy threshold is marked as the cutting-out stage. The initial classifier is trained using a feature sample set with stage pseudo-labels to obtain the initial state decision model; The initial state decision model predicts the feature sample set, selects samples with prediction confidence higher than the preset confidence threshold, performs consistency verification between the predicted labels and the stage pseudo labels generated based on physical partitioning rules, and adds the samples that pass the verification and their predicted labels to the high confidence sample set. The initial state decision model is retrained using a high-confidence sample set. The prediction, consistency verification, and retraining processes are iteratively executed until the performance of the initial state decision model converges, thus obtaining the state decision model.

3. The intelligent energy consumption optimization and control method for an automotive tube profile cutting production line according to claim 2, characterized in that, Short-time energy characteristics and short-time mutation intensity characteristics are obtained through the following methods, specifically including: The expression for the short-time energy characteristics is: In the formula, The normalized short-time energy corresponding to the current signal of the i-th frame is used to characterize the motor load intensity within a single frame time sequence; This represents the number of sampling points for a single frame of the current signal. The instantaneous current value of the motor at the k-th sampling point within the i-th frame; The dimensionless load correction factor represents the material properties of the coupling pipe profile. Pipe profile wall thickness Dynamic value retrieval; The expression for the short-term mutation intensity characteristic is: In the formula, The short-term mutation intensity corresponding to the current signal of the i-th frame is used to characterize the temporal fluctuation and the severity of the switching of the cutting condition. These are the normalized short-time energies of the current frame and the previous frame, respectively. It is a dimensionless abrupt change calibration coefficient, which synchronously couples the material properties of the pipe profile with the dynamic values ​​of the wall thickness.

4. The intelligent energy consumption optimization and control method for an automotive tube profile cutting production line according to claim 1, characterized in that, The methods for constructing a pre-built cooling process database include: For different combinations of pipe profile properties, a thermal-fluid-structure coupled finite element simulation model of the cutting zone is established. The finite element simulation model uses the first target flow rate value in the cutting stage, the second target flow rate value in the cutting stage, and the steady-state maintenance flow rate value in the steady-state cutting stage as adjustable input variables, and outputs the highest temperature value of the cutting zone. With the constraint that the highest temperature in the cutting zone does not exceed the preset critical burn temperature threshold, and with the optimization objective of minimizing the weighted total flow of the first target flow rate, the second target flow rate, and the steady-state maintained flow rate, a multi-objective optimization problem is constructed. The multi-objective particle swarm optimization algorithm is used to solve the multi-objective optimization problem and obtain the Pareto optimal solution set; In the Pareto optimal solution set, the weighted total flow rate corresponding to each solution is sorted according to the preset energy consumption preference weight. The solution whose weighted total flow rate satisfies the constraint condition is selected. The first target flow rate value, the second target flow rate value, and the steady-state maintenance flow rate value contained in the solution are combined with the corresponding pipe profile attribute information to form a cooling parameter group, which is stored in the cooling process database.

5. The intelligent energy consumption optimization and control method for an automotive tube profile cutting production line according to claim 1, characterized in that, The first and second target flow rates corresponding to the current pipe profile attributes in the cooling process database are corrected based on the deviation value, including: Obtain the sequence of historical deviation values ​​recorded from the N consecutive cuts prior to the current cut operation, as well as the first target flow rate value and the second target flow rate value currently stored in the cooling process database, and concatenate them to form the current state vector, where N is a preset integer; The current state vector is input into a pre-trained traffic correction policy network, which outputs a first correction increment for the first target traffic value and a second correction increment for the second target traffic value. The first correction increment and the second correction increment are added to the current first target flow rate value and the second target flow rate value, respectively, to obtain the updated first target flow rate value and the second target flow rate value. The cooling process database is then updated based on the updated first target flow rate value and the second target flow rate value.

6. The intelligent energy consumption optimization and control method for an automotive tube profile cutting production line according to claim 5, characterized in that, Before inputting the current state vector into the pre-trained traffic correction policy network, the following steps are also included: Input the first target flow rate value and the second target flow rate value corresponding to the current pipe profile attribute in the current cooling process database into the pre-trained cooling energy consumption prediction model to obtain the predicted cooling energy consumption value. The predicted cooling energy consumption value is appended to the current state vector to form an enhanced state vector; The enhanced state vector is input into the flow correction policy network, and the flow correction policy network outputs the correction increment. The cooling energy consumption prediction model is a time-series neural network model trained based on historical cutting data. The historical cutting data includes the first target flow rate value, the second target flow rate value, and the corresponding real-time power of the cooling pump recorded in the historical cutting operation.

7. The intelligent optimization control method for energy consumption in an automotive tube profile cutting production line according to claim 1, characterized in that, Controlling the power variation of the cooling pump according to a nonlinear ramp curve includes: A pressure-stabilizing control agent is constructed based on a proximal policy optimization algorithm. The pressure-stabilizing control agent is driven by a deep neural network. The state space of the pressure-stabilizing control agent includes the real-time pressure deviation between the actual pressure value and the target pressure value of the coolant supply pipeline, the rate of change of the real-time pressure deviation, and the reference slope of the nonlinear ramp curve at the current moment. The action space of the pressure-stabilizing control agent is the adjustment coefficient of the instantaneous slope of the nonlinear ramp curve. Each time the first transient high-pressure impact command or the second transient high-pressure impact command is executed, the state space parameters at the current moment are input into the pressure stabilization control agent, and the pressure stabilization control agent outputs the slope curvature adjustment coefficient corresponding to the current moment. The slope curvature adjustment coefficient is multiplied by the reference slope to obtain the adjusted instantaneous slope, and the power change rate of the cooling pump is controlled by the adjusted instantaneous slope.

8. The intelligent energy consumption optimization and control method for an automotive tube profile cutting production line according to claim 4, characterized in that, The reward function of a stress-controlled agent is determined by the following formula: In the formula: The cumulative reward value of the pressure stabilization control agent after the end of a transient high-pressure impact process; The actual pressure value collected at time t; The target pressure value; This refers to the relative pressure deviation; This refers to the total duration of the transient high-pressure impact command; The time required from the start of the transient high-pressure impact command until the actual pressure value first enters and remains within the preset error band of the target pressure value; This is the preset reference adjustment time; This refers to the relative adjustment time; It is a weighted sum of the correction increments of the first target traffic value and the second target traffic value in the most recent output of the network according to the traffic correction policy; For collaborative reward and punishment functions, , For the preset boundary values ​​of the stable interval, when When within the stable range, ; These are the preset non-negative weighting coefficients.

9. The intelligent energy consumption optimization and control method for an automotive tube profile cutting production line according to claim 7, characterized in that, Each time the first transient high-voltage impact command or the second transient high-voltage impact command is executed, it includes: When the current command is detected to switch from the first transient high-pressure impact command or the second transient high-pressure impact command to the steady-state maintenance command, the steady-state target pressure value corresponding to the steady-state maintenance command is obtained; The steady-state target pressure value is used as the target pressure value of the pressure stabilization control agent. During the preset discharge control period at the start of the switch, the instantaneous slope of the nonlinear ramp curve is used as the action output of the pressure stabilization control agent. The pressure stabilization control agent outputs the dynamically adjusted ramp curvature adjustment coefficient based on the real-time pressure deviation between the actual pressure value and the steady-state target pressure value. When the real-time pressure deviation indicates that the actual pressure value is higher than the steady-state target pressure value, the slope of the nonlinear slope curve is increased by adjusting the slope curvature to accelerate pressure release. When the actual pressure value approaches the steady-state target pressure value, the slope is decreased by adjusting the slope curvature to smoothly enter the steady state.

10. An intelligent energy consumption optimization control system for an automotive tube profile cutting production line, employing the intelligent energy consumption optimization control method for an automotive tube profile cutting production line as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire the instantaneous current signal of the cutting motor in real time and obtain the attribute information of the currently cut pipe profile. The attribute information includes at least the material code and wall thickness value. The feature extraction and stage identification module is used to extract the short-time energy features and short-time change intensity features of the instantaneous current signal. The short-time energy features are used as the load intensity criterion, and the short-time change intensity features are used as the stage switching criterion. The input is a pre-built state decision model, which identifies the current cutting process and obtains the cutting stage. The cutting stage includes the no-load stage, the cutting-in stage, the steady-state cutting stage, and the cutting-out stage. The coolant supply instruction generation module is used to call the corresponding cooling parameter group for the cutting stage, the cooling parameter group for the cutting exit stage, and the parameter group for the steady-state cutting stage from the pre-built cooling process database based on the attribute information of the pipe profile being cut, and generate a coolant supply instruction in combination with the cutting stage. The coolant supply instruction includes: the first transient high-pressure impact instruction for the cutting stage, the second transient high-pressure impact instruction for the cutting exit stage, the steady-state maintenance instruction for the steady-state cutting stage, and the dormancy instruction for the no-load stage. The execution control module is used to execute the corresponding coolant supply command according to the cutting stage and control the power change of the coolant pump according to the nonlinear ramp curve. The quality feedback correction module is used to obtain the end face quality index of the current cut pipe profile, compare the end face quality index with the preset quality qualified threshold range to obtain the deviation value, and correct the first target flow rate value and the second target flow rate value corresponding to the current pipe profile attribute in the cooling process database according to the deviation value.

Images