Energy Consumption Prediction and Control Method for Injection Molding Process of Air Deflector Based on Particle Swarm Optimization Algorithm

Abstract

This invention relates to the field of intelligent manufacturing and industrial process control technology in injection molding, specifically a method for predicting and controlling energy consumption in the injection molding process of a guide vane based on particle swarm optimization (PSO) algorithm. The method includes: a process data synchronous acquisition step: acquiring process state time-series data, energy consumption time-series data, environmental state data, and process event parameters; a theoretical benchmark construction step: establishing a theoretical energy consumption reference model and generating an ideal energy consumption curve; a loss correction and residual analysis step: generating a simulated energy consumption curve, constructing a real residual time-series and a theoretical residual time-series, and segmenting them; and a prediction and control optimization step: using PSO algorithm to identify loss parameters and predict the energy consumption of the next cycle, further optimizing control parameters under quality threshold constraints, and issuing control commands to achieve interpretable prediction and closed-loop control of the guide vane injection molding process.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and industrial process control technology for injection molding, specifically a method for predicting and controlling energy consumption in the injection molding process of air guide plates based on particle swarm optimization algorithm. Background Technology

[0002] Currently, the energy consumption prediction and control mechanism for the air guide plate injection molding process is usually based on historical energy consumption averages, fixed process parameters, or total power monitoring. When there are fluctuations in ambient temperature, changes in raw material rheology, or mechanical wear during the operation of the injection molding equipment, the existing solutions cannot accurately identify the actual dissipation at each stage of the molding cycle, nor can they optimize the control parameters in a targeted manner while ensuring molding quality, thus reducing the accuracy of energy consumption prediction and the effectiveness of online control. Summary of the Invention

[0003] The purpose of this invention is to provide an energy consumption prediction and control method for the injection molding process of air guide plates based on particle swarm optimization, addressing the following technical problems: Existing energy consumption prediction and control mechanisms for the injection molding process of air guide plates struggle to accurately identify the actual energy dissipation at each stage of the molding cycle when faced with environmental temperature fluctuations, raw material rheological changes, or mechanical wear, and are insufficient in optimizing control parameters to ensure molding quality. There is an urgent need for a method that can decompose and identify on-site energy consumption, dynamically optimize control parameters under quality constraints, and achieve interpretable prediction and closed-loop optimization. The purpose of this invention can be achieved through the following technical solutions:

[0004] A method for predicting and controlling energy consumption in the injection molding process of a wind deflector based on particle swarm optimization includes:

[0005] S100: Acquire the process status time sequence data, energy consumption time sequence data, environmental status data and process event parameters of the injection molding equipment during the air guide plate molding cycle, and synchronize them according to a unified timestamp or a unified periodic trigger signal to construct a process data set.

[0006] S200. Based on the process data set, construct a theoretical energy consumption reference model determined by preset raw material thermophysical property parameters, preset mold filling parameters and preset molding stage time allocation rules, and generate an ideal energy consumption curve.

[0007] S300. Based on the ideal energy consumption curve and the preset loss parameter vector, the simulated energy consumption curve is obtained by correction.

[0008] S400. Calculate the total energy consumption time series based on the energy consumption time series data, align the total energy consumption time series with the ideal energy consumption curve and the simulated energy consumption curve respectively, generate the actual residual time series and the theoretical residual time series, and segment them according to the forming stage to obtain the segmented residual sequence.

[0009] S500. Taking the minimum dynamic time warping distance between the actual residual time series and the theoretical residual time series as the optimization objective, the particle swarm optimization algorithm is used to optimize the preset loss parameter vector, and the predicted energy consumption curve and / or predicted total energy consumption value of the next forming cycle are output according to the parameters after optimization convergence.

[0010] S600: Extract the set parameters of the injection molding equipment to construct a set of control parameters to be optimized. Based on the predicted energy consumption curve and / or the predicted total energy consumption value, the preset molding quality threshold and the preset quality prediction model, the particle swarm optimization algorithm is used to optimize the set of control parameters to be optimized, obtain the target control parameters, generate control commands and send them to the controller of the injection molding equipment.

[0011] Optionally, the process state timing data includes multi-segment barrel temperature timing, injection pressure timing, injection speed timing, and screw torque timing;

[0012] The process event parameters include pressure holding switching time and / or pressure holding switching location parameters;

[0013] The energy consumption time series data includes the active power time series of the servo motor and the active power time series of the heating coil, which are collected by the power acquisition unit;

[0014] The environmental status data includes the temperature difference between the inlet and outlet water of the mold temperature controller and the ambient temperature of the workshop, which are collected by the environmental sensing unit.

[0015] The process data set also includes the corresponding current waveform used to correct the power acquisition results.

[0016] Optionally, a theoretical energy consumption reference model is constructed based on the process data set, determined by preset raw material thermophysical properties, preset mold filling parameters, and preset molding stage time allocation rules, to generate an ideal energy consumption curve, including:

[0017] S201. Based on the specific heat parameters and melting temperature range of the raw material corresponding to the air guide plate, calculate the theoretical heating energy consumption of the raw material from the initial temperature to the molten state, wherein the specific heat parameters and the melting temperature range belong to the preset raw material thermal property parameters.

[0018] S202. Based on the preset mold filling parameters, which include at least cavity geometry parameters, flow channel resistance parameters, melt viscosity parameters, and mold filling pressure parameters, calculate the theoretical flow energy consumption of the melt filling process.

[0019] S203. Based on the preset duration or actual timing nodes of the plasticizing stage, molding stage, holding pressure stage, and cooling stage as the preset molding stage time allocation rules, the theoretical heating energy consumption and the theoretical flow energy consumption are allocated over time to generate the ideal energy consumption curve.

[0020] Optionally, the preset loss parameter vector includes a thermal loss coefficient, a mechanical loss coefficient, and a viscous loss coefficient; the simulated energy consumption curve is obtained by correcting the ideal energy consumption curve and the preset loss parameter vector, including:

[0021] S301. Establish the heat loss coefficient corresponding to environmental heat dissipation;

[0022] S302. Establish the mechanical loss coefficients corresponding to screw mechanical wear and transmission friction;

[0023] S303. Establish the viscosity loss coefficient corresponding to the non-Newtonian rheological fluctuations of the raw materials.

[0024] S304. Based on the heat loss coefficient, the mechanical loss coefficient and the viscosity loss coefficient, the ideal energy consumption curve is corrected to generate the simulated energy consumption curve.

[0025] Optionally, the total energy consumption time series is calculated based on the energy consumption time series data. The total energy consumption time series is then aligned with the ideal energy consumption curve and the simulated energy consumption curve to generate a real residual time series and a theoretical residual time series. These are then segmented according to the forming stage to obtain a segmented residual sequence, including:

[0026] S401. The servo motor active power time sequence and heating coil active power time sequence in the energy consumption time sequence data are resampled and summed according to a unified timestamp to obtain the total energy consumption time sequence. Then, the total energy consumption time sequence is aligned with the ideal energy consumption curve to calculate the actual residual time sequence.

[0027] S402. Align the simulated energy consumption curve with the ideal energy consumption curve, and calculate the theoretical residual timing.

[0028] S403. Based on the process state timing data and the process event parameters, the actual residual timing and the theoretical residual timing are periodically segmented into the plasticizing stage, the mold filling stage, the pressure holding stage, and the cooling stage to form the segmented residual sequence.

[0029] Optionally, a particle swarm optimization algorithm is used to optimize the preset loss parameter vector, and the dynamic time warping distance between the actual residual time series and the theoretical residual time series is calculated based on the segmented residual sequence. The predicted energy consumption curve and / or predicted total energy consumption value for the next molding cycle are then output, including:

[0030] S501. Encode the preset loss parameter vector into particle positions;

[0031] S502. Calculate the total dynamic time warping distance and the dynamic time warping distance of each stage between the actual residual time series and the theoretical residual time series based on the segmented residual sequence.

[0032] S503. Construct a fitness function with the goal of minimizing the weighted sum of the total dynamic time warp distance and / or the dynamic time warp distance of each stage based on a preset weight, and update the particle velocity and particle position.

[0033] S504. After reaching the maximum number of iterations and / or the fitness change of multiple consecutive iterations is less than the preset convergence threshold, output the target loss parameter vector and the predicted energy consumption curve and / or the predicted total energy consumption value of the next molding cycle.

[0034] Optionally, based on the predicted energy consumption curve and / or predicted total energy consumption value, a preset molding quality threshold, and a preset quality prediction model, a particle swarm optimization algorithm is used to optimize the set of control parameters to be optimized, obtain the target control parameters, generate control commands, and send them to the controller of the injection molding equipment, including:

[0035] S601, The barrel temperature setpoint, injection speed setpoint, holding pressure setpoint, and holding pressure switching point setpoint are constructed into the set of control parameters to be optimized;

[0036] S602. Based on the target loss parameter vector, calculate the candidate energy consumption results corresponding to each candidate control parameter in the set of control parameters to be optimized;

[0037] S603. Based on the preset quality prediction model constructed using historical process setting parameters and corresponding offline product quality inspection datasets, calculate the candidate quality indicators corresponding to each candidate control parameter.

[0038] S604. Based on the preset molding quality threshold, the candidate quality indicators are constrained and determined, and the candidate energy consumption results that meet the constraint determination are screened.

[0039] S605. Select the target control parameter from the control parameters that meet the preset molding quality threshold, and generate the control command.

[0040] Optionally, the candidate quality indicators are constrained based on the preset molding quality threshold, and the candidate energy consumption results that meet the constraints are filtered, including:

[0041] S6041. When the candidate quality index is lower than the preset molding quality threshold, the corresponding control parameter is determined to be an invalid parameter and is removed.

[0042] S6042. When the candidate quality index is greater than or equal to the preset molding quality threshold, retain the corresponding control parameter and record the corresponding candidate energy consumption result.

[0043] S6043. Select the control parameter with the minimum predicted total energy consumption from the retained control parameters as the target control parameter.

[0044] Optionally, it also includes:

[0045] S700: Obtain the new molding cycle process data set after the controller executes the control command;

[0046] S800. Update the actual residual time series based on the new forming cycle process data set, and calculate the dynamic time warping distance between it and the theoretical residual time series, or verify the similarity value obtained by converting the dynamic time warping distance based on the preset mapping function.

[0047] S900: When the dynamic time warping distance is higher than the preset distance threshold, or the similarity value is lower than the preset similarity threshold, S200 to S600 are re-executed; when the dynamic time warping distance is not higher than the preset distance threshold, and the similarity value is equal to or higher than the preset similarity threshold, the current target loss parameter vector and target control parameters are maintained.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. Existing technologies often rely on historical averages or general total power monitoring, which are difficult to cope with dynamic changes in the environment or equipment. This method extracts multi-dimensional process and environmental data, constructs a theoretical energy consumption reference model based on raw material thermophysical parameters and molding parameters to generate an ideal energy consumption curve, and further combines multi-dimensional loss parameter vectors to correct the simulated energy consumption curve. This mechanism scientifically decomposes the highly coupled on-site energy consumption problem into a physical ideal baseline and multi-dimensional real loss identification, which can accurately identify and locate the specific dissipation sources at each stage of molding, effectively solving the problems of low prediction accuracy and inability to track local anomalies in traditional methods.

[0050] 2. During injection molding, scaling or peak misalignment often occurs in the molding stage duration, making conventional point-to-point data comparison prone to misjudgment. This method not only segments the total energy consumption time sequence according to molding stages such as plasticizing, filling, holding pressure, and cooling to generate actual and theoretical residuals, but also introduces dynamic time warping distance combined with particle swarm optimization to evaluate and optimize the loss parameter vector. This innovative design greatly improves the algorithm's tolerance to periodic stretching and time sequence drift, ensuring that even under complex working conditions with rheological fluctuations or signal delays, it can still reliably fit the predicted energy consumption trajectory for the next cycle.

[0051] 3. Traditional energy-saving strategies are prone to causing quality defects in thin-walled parts, such as insufficient adhesive and warping, when reducing energy consumption. In the secondary optimization stage of control parameters, this method deeply integrates the predicted energy consumption results with the preset molding quality threshold and the quality prediction model. When iterating the control parameter set through the particle swarm optimization algorithm, the system strictly implements gating screening: control parameters below the quality threshold are judged as invalid parameters and eliminated, and only the combination with the minimum predicted total energy consumption is selected from the remaining qualified items to generate control instructions. This constraint architecture with quality as the priority condition avoids energy consumption optimization that deviates from the product yield standard.

[0052] 4. To address the issue that on-site conditions continue to drift over time after control commands are issued, this method adds a review and update mechanism for the new forming cycle. By acquiring the process data of the new cycle after the execution of the command in real time, the actual residual timing is recalculated and the dynamic time normalization distance or similarity value is compared again. Once the on-site dissipation characteristics are found to deviate from the preset threshold, the system can immediately and automatically trigger the previous round of modeling and optimization process to reconstruct the model. This adaptive closed-loop mechanism improves the self-correction capability of the control system and ensures the adaptive energy-saving effect of the production line in long-term operation. Attached Figure Description

[0053] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0055] Example 1:

[0056] Please see Figure 1 A method for predicting and controlling energy consumption in the injection molding process of a wind deflector based on particle swarm optimization includes:

[0057] S100: Acquire the process status time sequence data, energy consumption time sequence data, environmental status data and process event parameters of the injection molding equipment during the air guide plate molding cycle, and synchronize them according to a unified timestamp or a unified periodic trigger signal to construct a process data set.

[0058] S200: Based on the process data set, construct a theoretical energy consumption reference model determined by preset raw material thermophysical parameters, preset mold filling parameters and preset molding stage time allocation rules, and generate an ideal energy consumption curve.

[0059] S300, based on the ideal energy consumption curve and the preset loss parameter vector, the simulated energy consumption curve is obtained by correction;

[0060] S400. Calculate the total energy consumption time series based on the energy consumption time series data, align the total energy consumption time series with the ideal energy consumption curve and the simulated energy consumption curve respectively, generate the actual residual time series and the theoretical residual time series, and divide them into segments according to the forming stage to obtain the segmented residual sequence.

[0061] S500 takes minimizing the dynamic time warping distance between the actual residual time series and the theoretical residual time series as the optimization objective, uses the particle swarm optimization algorithm to optimize the preset loss parameter vector, and outputs the predicted energy consumption curve and / or predicted total energy consumption value for the next molding cycle based on the parameters after optimization convergence.

[0062] S600: Extract the set parameters of the injection molding equipment to construct a set of control parameters to be optimized. Based on the predicted energy consumption curve and / or the predicted total energy consumption value, the preset molding quality threshold and the preset quality prediction model, the particle swarm optimization algorithm is used to optimize the set of control parameters to be optimized, obtain the target control parameters, generate control commands and send them to the controller of the injection molding equipment.

[0063] This embodiment provides an injection molding energy consumption prediction and control mechanism for a continuous production line of automotive air conditioning air guide plates. Specifically, the scenario is set as follows: a servo injection molding machine continuously produces the same type of air guide plate, with a production cycle of approximately 45 seconds per mold. Since the air guide plate is a thin-walled, large-area part, reducing the heat or deviating from the preset lower limit of holding pressure may cause insufficient glue or warping. On the other hand, increasing the temperature and pressure beyond the preset upper limit will lead to long-term high-load operation of the heating coil and servo system. Therefore, this embodiment does not directly use historical energy consumption for average calculation, but first establishes an energy consumption baseline under ideal conditions, and then superimposes the actual on-site dissipation in a parameterized manner. The particle swarm optimization algorithm is used to simultaneously complete energy consumption prediction and control parameter optimization.

[0064] Specifically, in S100, four types of data are collected around a complete molding cycle and uniformly placed into the process data set. The so-called uniformity does not mean that the same sampling frequency must be used. They can be sampled at their original frequencies first, and then synchronized through a unified timestamp or cycle start and end signal. For example, the barrel temperature is updated every 100 milliseconds, the injection pressure is updated every 10 milliseconds, the power is updated every 20 milliseconds, and the ambient temperature is updated every 2 seconds. Then, with the start of mold closing in this cycle as the zero point and the end of mold opening and ejection as the end point, all data are mapped to the same time axis.

[0065] For ease of understanding, let's assume a scaled-down example cycle is divided into 8 sampling points, with the servo motor power sequence being [2,4,7,6,5,3,2,1] and the heating coil power sequence being [5,5,4,4,3,3,2,2]. Each sampling point can form a set of synchronous records. For example, the 3rd point corresponds to time t3, barrel zone 2 temperature 243℃, injection pressure 82MPa, servo power 7kW, heating power 4kW, workshop temperature 21℃, and V / P switching marked as not triggered. This forms the process data set for subsequent modeling.

[0066] In S200, a theoretical energy consumption reference model is established using the raw material type, cavity information, and molding stage information in the process data set, and an ideal energy consumption curve is output. The ideal energy consumption does not refer to the parameters that must be achieved under actual working conditions, but rather represents the minimum theoretical energy benchmark required to complete the molding of the air guide plate under the condition of excluding additional heat dissipation, mechanical aging, and rheological fluctuation interference. If the mass of the raw material in a single mold is 100 grams, it is theoretically necessary to heat the raw material from 25°C to 240°C. The melt needs to be propelled through the gate, runner, and cavity, which requires a flow energy component. Then, it is distributed on the time axis according to the four stages of plasticizing, filling, holding pressure, and cooling. Taking 8 sampling points as an example, the ideal energy consumption curve after converting the pure theoretical total energy can be written as [3,4,6,5,4,2,1,1], where the first 2 points mainly fall in the plasticizing stage, the middle 3 points reflect the filling and holding pressure stages, and the last 3 points reflect the cooling stage.

[0067] In S300, the ideal energy consumption curve is combined with a preset loss parameter vector to generate a simulated energy consumption curve. Here, the loss parameter vector is a set of multi-dimensional adjustment variables describing the actual dissipation on site, including at least three directions: heat loss, mechanical loss, and viscosity loss. If the initial heat loss coefficient is set to 0.15, the mechanical loss coefficient to 0.10, and the viscosity loss coefficient to 0.08, then the ideal curve [3,4,6,5,4,2,1,1] can be corrected to [3.5,4.7,7.2,6.1,4.9,2.5,1.3,1.2]. The simulated curve obtained in this way is not the actual measured value on site, but the energy consumption trajectory that the equipment should exhibit under a certain set of loss assumptions.

[0068] In S400, the total energy consumption time series is obtained from the energy consumption time series data, and then compared with the ideal curve and the simulation curve to construct two types of residuals. Continuing with the example data above, if the measured total energy consumption time series is obtained by summing the servo and heating power [7,9,11,10,8,6,4,3], then the actual residual can be obtained by subtracting the ideal energy consumption from the measured total energy consumption [4,5,5,5,4,4,3,2]; the theoretical residual can be obtained by subtracting the ideal energy consumption from the simulation energy consumption [0.5,0.7,1.2,1.1,0.9,0.5,0.3,0.2]. Since the two are of different magnitudes, normalization or scaling can be further performed during implementation. The focus of the comparison is on the waveform change trend rather than the absolute value.

[0069] The residual is then divided into several stages based on the process event nodes; for example, points 1 to 2 are the plasticizing stage, points 3 to 4 are the mold filling stage, point 5 is the pressure holding stage, and points 6 to 8 are the cooling stage; thus, the above segmentation operation can not only achieve quantitative comparison of the total deviation, but also further analyze the molding stage where the deviation is mainly concentrated.

[0070] In S500, each particle in the particle swarm optimization algorithm represents a set of candidate loss parameters. For each set of candidate values, the corresponding simulation curve and theoretical residual are first generated, and then the dynamic time warping distance between the actual residual and the theoretical residual is calculated. If the stage distance obtained under a certain set of parameters is the smallest, it means that this set of parameters is closer to the actual loss state of the current equipment. For ease of understanding, the coordinates of the three particles can be set as P1=0.10,0.08,0.06, P2=0.18,0.12,0.09, and P3=0.30,0.05,0.15. After several rounds of iteration, if the total distance of P2 is the smallest, the target loss parameter vector corresponding to P2 is output, and the predicted energy consumption curve for the next cycle is generated accordingly, for example, the prediction is [7.2,9.1,10.8,9.9,8.1,5.8,4.1,3.0]. At the same time, the predicted total energy consumption value can be obtained by integration.

[0071] In the S600, after identifying the actual loss environment, the particle swarm optimization algorithm is used to perform a second-level optimization of the control parameters. At this time, the particles no longer represent loss coefficients, but rather control variables such as barrel temperature, injection speed, holding pressure, and V / P switching point. Each set of candidate control variables is first processed by the energy consumption prediction model to obtain candidate energy consumption, and then by the quality prediction model to determine whether it meets the molding quality threshold. If the quality constraint is met, it participates in the energy consumption minimization comparison; if it is not met, it is directly eliminated. Finally, the optimal control parameters are formed into control instructions, written into the injection molding machine controller, and executed in the next molding cycle.

[0072] As an anomaly handling mechanism, when a sampling is missing, an event node is lost, or the power waveform is interrupted in a certain period, a data integrity check can be performed first. If the missing ratio is lower than a preset threshold, such as less than 5%, it can be supplemented by neighbor interpolation or the average value of the same stage in the previous period. If the missing ratio exceeds the threshold, the period is only used for equipment monitoring and does not participate in parameter updates. The target loss parameters and target control parameters of the previous round are used. For example, when the duration of a certain stage output by the theoretical model deviates from the actual event duration by more than a preset time deviation threshold, the actual event duration can be used first to reconstruct the time axis to avoid the subsequent distance calculation being skewed.

[0073] After the start of the winter night shift, the ambient temperature in the workshop dropped from 26℃ to 19℃, and the temperature difference between the inlet and outlet water of the mold temperature controller also widened, resulting in an increased load on the heating coil. If the barrel setting temperature is directly reduced based solely on the increase in total energy consumption, weld lines and localized missing glue may easily appear on the air guide plate. In this embodiment, the increase in energy consumption is first identified by comparing the ideal benchmark with the double residual to determine that this increase is mainly due to environmental heat dissipation. Then, the holding pressure and injection speed are finely adjusted under the premise of meeting the molding quality threshold, instead of simply lowering the temperature based on a single variable, thereby obtaining a more stable low-energy consumption control result.

[0074] The purpose of this step is to break down the complex on-site energy consumption problem into three executable steps: ideal baseline, loss identification, and control optimization under quality constraints, so as to achieve interpretable prediction and closed-loop control of the air guide plate injection molding process.

[0075] The process status timing data includes multi-segment barrel temperature timing, injection pressure timing, injection speed timing, and screw torque timing;

[0076] Process event parameters include pressure holding switching time and / or pressure holding switching location parameters;

[0077] The energy consumption time-series data includes the active power time-series data of the servo motor and the active power time-series data of the heating coil, which are collected by the power acquisition unit;

[0078] Environmental status data includes the temperature difference between the inlet and outlet water of the mold temperature controller and the ambient temperature of the workshop, collected by the environmental sensing unit;

[0079] The process data set also includes the corresponding current waveforms used to correct the power acquisition results.

[0080] This embodiment provides a data organization mechanism for refining the definition of on-site data collection objects. Specifically, although the above basic process can form a process data set, if the data granularity of the collection fields is lower than the preset granularity requirement, such as only recording the total temperature, total pressure, or total power, it is easy to cover up local anomalies in the injection molding of the air guide plate. Especially in the production of thin-walled parts, the energy consumption and quality risks caused by the low temperature at the front end of the barrel and the high temperature at the back end are different. Therefore, this embodiment further refines the collection dimensions.

[0081] Specifically, the multi-stage barrel temperature can be divided into three, four, or five zones according to the actual injection molding machine configuration. Taking four zones as an example, a temperature time sequence T1, T2, T3, and T4 can be formed with respect to time t. The injection pressure time sequence reflects the melt propulsion resistance, the injection speed time sequence reflects the screw forward speed, and the screw torque time sequence reflects the mechanical load during the plasticizing stage. If, in a certain cycle, a certain sampling point from T1 to T4 is 210℃, 225℃, 238℃, and 242℃ respectively, and the screw torque increases while the injection speed does not increase, it is more likely to indicate an increase in the plasticizing resistance of the raw material, rather than a simple change in ambient temperature.

[0082] The process event parameters include at least one of the following: the pressure holding switching time and the pressure holding switching position. The former is used to mark the moment when the mold filling process enters the pressure holding process on the time axis, and the latter is used to mark the switching threshold on the displacement axis or screw position axis. For example, in a scaled-down scenario, there are a total of 8 sampling points in one cycle. If the pressure holding switching is triggered at the 4th point, then the 1st to 3rd points are more biased towards the mold filling process, and the pressure holding and cooling analysis begins after the 4th point. If the switching position is recorded as 12 mm at the same time, the stage boundary can still be stably identified by the position parameter even when the actual duration of different cycles varies slightly.

[0083] The energy consumption time series data is preferably collected using dual-channel acquisition of servo motor active power and heating coil active power. The reason for this is that the energy-saving strategies corresponding to mechanical drive power and heat input power are different. If the increase in total power consumption mainly occurs in the servo channel, it is more likely to be related to injection speed, pressure, and screw resistance. If it mainly occurs in the heating coil channel, it is more likely to be related to material temperature setting, heat preservation performance, and environmental heat dissipation.

[0084] In addition to the ambient temperature of the workshop, the environmental status data also includes the temperature difference between the inlet and outlet water of the mold temperature controller; the latter can indirectly reflect the efficiency of the mold's heat exchange; for example, an inlet water temperature of 28℃ and an outlet water temperature of 34℃ correspond to a temperature difference of 6℃, while another shift has an inlet water temperature of 28℃ and an outlet water temperature of 31℃, with a difference of only 3℃, indicating that the heat exchange status is different; if only the room temperature is considered and the mold temperature circulation is ignored, the theoretical model may misjudge the heat dissipation of the mold as a decrease in the efficiency of the equipment itself;

[0085] Furthermore, the process data set also retains the corresponding current waveform to correct the power acquisition results. This is because when the field power module experiences harmonic interference or transient peak values ​​exceeding the preset peak threshold, relying solely on low-frequency active power records may result in deviations. During implementation, phase verification can be performed on the current waveform and voltage sampling. If a difference greater than the preset deviation threshold is found between the power value and the integral estimate of the current waveform within a certain sampling window, such as a deviation exceeding 10%, the corrected power value is used to replace the original value, or a confidence level marker is added to the window for subsequent model weighting.

[0086] As an anomaly handling mechanism, if a certain type of signal is temporarily unavailable, such as an injection molding machine without the corresponding sensor being unable to output screw torque timing, it can degenerate into estimating the torque trend using the motor current waveform and reducing the weight of this dimension in the model; if the mold temperature controller data access fails, at least the workshop ambient temperature should be retained and the search range of the heat loss coefficient should be appropriately widened to compensate for the uncertainty caused by insufficient environmental observation.

[0087] On the same air guide plate production line, the first and second production periods use the same tools and raw materials, but the heat exchange efficiency of the mold temperature controller decreases in the second production period, causing the temperature difference between the inlet and outlet water to increase from 4°C to 7°C. If only the total power consumption is recorded, the system can only see the increase in heating power. However, in this embodiment, since multiple temperature and power channels, ambient temperature and water temperature difference are recorded simultaneously, the system can identify that this increase is more biased towards the heat dissipation side, thereby avoiding the problem being mistakenly attributed to improper injection speed setting.

[0088] The purpose of this step is to break down and observe different energy sources through finer-grained data organization, thereby providing an interpretable input basis for subsequent loss identification and control optimization.

[0089] Based on the process data set, a theoretical energy consumption reference model is constructed, determined by preset raw material thermophysical properties, preset mold filling parameters, and preset molding stage time allocation rules, generating an ideal energy consumption curve, including:

[0090] S201. Based on the specific heat parameters and melting temperature range of the raw material corresponding to the air guide plate, calculate the theoretical heating energy consumption of the raw material from the initial temperature to the molten state. The specific heat parameters and melting temperature range belong to the preset thermal property parameters of the raw material.

[0091] S202. Based on preset filling parameters including at least cavity geometry parameters, flow channel resistance parameters, melt viscosity parameters and filling pressure parameters, calculate the theoretical flow energy consumption of the melt filling process.

[0092] S203. Based on the preset duration or actual timing nodes of the plasticizing stage, molding stage, holding pressure stage and cooling stage as preset molding stage time allocation rules, allocate the theoretical heating energy consumption and theoretical flow energy consumption over time to generate an ideal energy consumption curve.

[0093] This embodiment provides a specific construction step for an ideal energy consumption curve; specifically, simply collecting data is insufficient to explain which energy consumptions are unavoidable molding-required energies; without a theoretical benchmark with clear physical meaning, subsequent residuals would lose their reference; therefore, this embodiment breaks down ideal energy consumption into two parts: theoretical heating energy consumption and theoretical flow energy consumption, and then maps them to the time axis according to the molding stage;

[0094] Specifically, in S201, the theoretical heating energy consumption is first calculated based on the thermal properties of the raw material. Taking the PC / ABS alloy commonly used in air guide plates as an example, the system pre-stores the specific heat parameter range, melting temperature range, and recommended processing range of the raw material. For ease of explanation, it is assumed that the mass of the raw material in a single mold is 0.10 kg, the initial temperature is 25℃, and the target melting equivalent temperature is 240℃. Then, the theoretical heating energy can be simplified as mass × average specific heat × temperature rise. In implementation, it is not required to disclose complex thermodynamic derivations; a theoretical heating amount Qh can be directly calculated by looking up the table parameters. If the average specific heat of different batches of raw materials is found to be slightly different, the corresponding parameter set can be automatically switched according to the raw material batch number.

[0095] In S202, the theoretical flow energy consumption is calculated based on the filling parameters. Because the length-to-diameter ratio of the air guide plate is greater than the preset length-to-diameter ratio threshold and the wall thickness is less than the preset wall thickness threshold, the resistance of the melt in the runner, gate, and cavity changes drastically and is greater than the preset resistance change threshold. Therefore, the flow energy consumption cannot be simply estimated based on the average injection pressure. In practice, the cavity geometry parameters, runner resistance parameters, melt viscosity parameters, and filling pressure parameters can be input into a single filling calculation module. For example, different resistance coefficients can be assigned to the main runner, branch runner, and thin-walled cavity of the air guide plate, and the theoretical flow rate Qf can be obtained from the expected filling pressure and melt propulsion volume. To reduce the computational complexity, the complex cavity can be discretized into several segments, such as the inlet segment, the middle spreading segment, and the end feeding segment, and then calculated separately and summed.

[0096] In S203, Qh and Qf are allocated to the time axis according to the molding stage to obtain the ideal energy consumption curve. If a preset duration rule is adopted, for example, plasticizing is set to 15 seconds, filling is set to 2 seconds, holding pressure is set to 8 seconds, and cooling is set to 20 seconds. If there are already precise event nodes on site, the actual nodes are used first. Continuing with the example of the above 8 sampling points, assuming that plasticizing accounts for the first 2 points, filling is set for the 3rd point, holding pressure is set for the 4th to 5th points, and cooling is set for the 6th to 8th points. If the theoretical heating energy accounts for 70% of the total theoretical energy and the theoretical flow energy accounts for 30%, then most of the heating energy can be allocated to the first half of plasticizing and holding pressure, and the flow energy can be concentrated at the filling point, thus forming an ideal curve similar to [2.5, 3.0, 5.5, 4.0, 3.0, 1.5, 1.0, 0.8].

[0097] Furthermore, to avoid the ideal curve being too sharp or not conforming to the output characteristics of the equipment, a smoothing process can be added after time allocation; for example, a sliding average can be used for adjacent points so that the theoretical peak value of the filling stage transitions slightly between adjacent points, so as to be closer to the response inertia of the actual drive and heating system.

[0098] As an anomaly handling mechanism, if the raw material batch number is unknown or the thermophysical parameters are missing, the default parameter set of the raw material family can be called first, and the cycle can be marked as having low parameter confidence. If the cavity geometry parameters are not yet digitized, the equivalent resistance parameters generated by the mold flow analysis can be used instead. If the difference between the preset duration and the actual timing node is greater than the preset time difference threshold, for example, the actual filling time is only 1.2 seconds while the preset time is 2 seconds, the actual node is used to reconstruct the ideal curve first, so as to prevent the theoretical reference model from deviating from the actual cycle time of the equipment.

[0099] During the initial continuous operation phase of the air guide plate molding, the process engineer has determined the PC / ABS raw materials, air guide plate wall thickness distribution, and V / P switching position. In this embodiment, these known parameters are used to construct the theoretical minimum energy consumption profile for each mold. In this way, even if the room temperature drops, the mold temperature fluctuates, and the screw wear gradually appears during the subsequent night shift, the system can still project all anomalies onto the same unified benchmark for comparison.

[0100] The purpose of this step is to establish an interpretable minimum theoretical energy consumption reference based on the thermal properties of the raw materials and the mold filling mechanism, so as to achieve a unified starting point for subsequent loss parameter injection and residual analysis.

[0101] The preset loss parameter vector includes the thermal loss coefficient, mechanical loss coefficient, and viscous loss coefficient; the simulated energy consumption curve is obtained based on the ideal energy consumption curve and the preset loss parameter vector, including:

[0102] S301. Establish the heat loss coefficient corresponding to environmental heat dissipation;

[0103] S302. Establish the mechanical loss coefficients corresponding to screw mechanical wear and transmission friction;

[0104] S303. Establish the viscosity loss coefficient corresponding to the non-Newtonian rheological fluctuations of the raw materials.

[0105] S304. Based on the heat loss coefficient, mechanical loss coefficient and viscosity loss coefficient, the ideal energy consumption curve is corrected to generate the simulated energy consumption curve.

[0106] This embodiment provides a correction mechanism that parameterizes unavoidable field losses into the theoretical baseline. Specifically, while using only an ideal energy consumption curve can provide the lowest theoretical boundary, it does not couple with actual physical dissipation characteristics, making it difficult to accurately fit the real dissipation trajectory under actual operating conditions. Especially during the mass production of air guide plates, environmental heat dissipation, screw wear, and raw material rheological fluctuations often coexist. If all these factors are mixed into a single total correction, subsequent control decisions will lack specificity. Therefore, this embodiment decomposes losses into three dimensions: thermal loss, mechanical loss, and viscosity loss.

[0107] Specifically, a heat loss coefficient is established in S301. This coefficient mainly describes the effect of environmental heat dissipation and additional losses from the insulation structure on energy consumption. Its effect is usually more pronounced during plasticizing, holding pressure, and the early cooling stage. When the ambient temperature in the workshop drops from 26°C to 20°C, the heat loss coefficient can be increased from the original 0.10 to 0.18, indicating that the part of the ideal curve related to heat retention needs to be compensated by 18%. In practice, this coefficient can also be affected by the temperature difference between the inlet and outlet water of the mold temperature controller. The greater the temperature difference, the stronger the heat dissipation, and the coefficient can be increased accordingly.

[0108] In S302, a mechanical loss coefficient is established. This coefficient mainly describes the mechanical additional work caused by screw wear, transmission friction, bearing resistance, etc. Unlike heat loss, it usually affects the screw rotation load during the plasticizing stage and the servo output during the injection propulsion stage more. For example, if the motor current is found to increase overall at the same injection speed after three months of continuous operation, the mechanical loss coefficient can be increased from 0.06 to 0.11. To enhance interpretability, the screw torque timing and servo active power can be compared in the same stage. If both increase synchronously, it further supports the judgment that mechanical loss is increasing.

[0109] A viscosity loss coefficient is established in S303. This coefficient is used to describe the additional energy consumption caused by non-Newtonian rheological fluctuations of the raw material, which is particularly significant when the raw material batch changes, the proportion of recycled material changes, or the degree of dryness changes. For the same PC / ABS raw material, if the moisture content of a batch of raw material is slightly higher, it will exhibit different shear viscosity responses during the high-speed mold filling stage. At this time, the viscosity loss coefficient can be adjusted from 0.05 to 0.09. This coefficient has a significant impact on the peak value of mold filling and the waveform shape near the V / P switching.

[0110] In S304, the ideal curve is piecewise corrected using the three coefficients mentioned above to generate a simulation curve. A single, fixed mathematical form is not required; a phased weighted correction based on physical influence weights can be used. The specific calculation flow rules are as follows: the system pre-sets different loss activation influence matrices for the four stages of plasticizing, molding, holding, and cooling. For example, the molding stage is severely affected by mechanical and viscosity fluctuations, so its corresponding mechanical and viscosity activation weights are set to high values. The loss activation influence matrix is ​​specifically a diagonal matrix or weight sequence vector whose dimensions match the number of sampling points on the time axis. The weight distribution values ​​within it are set a priori by the relevant technical personnel based on known thermodynamic and rheological experience of different plastic raw materials at each molding stage, or obtained through offline inversion calibration of equipment benchmark test data under standard operating conditions.

[0111] During the correction, the additional energy compensation value is obtained by multiplying the ideal energy consumption value of each stage by the sum of the products of the loss coefficient and its activation weight in the corresponding stage, and then superimposing it on the ideal curve.

[0112] If the ideal curve is [3,4,6,5,4,2,1,1], then the first two points have higher thermal-related activation weights, the third to fifth points have higher mechanical and viscosity-related activation weights, and the sixth to eighth points have thermal-related activation weights again. Assuming a heat loss coefficient of 0.2, a mechanical loss coefficient of 0.1, and a viscosity loss coefficient of 0.15, after the above-mentioned staged weighting calculation, we can obtain [3.6,4.8,7.5,6.1,4.8,2.4,1.2,1.2]. It can be seen that the peak value does not increase evenly, but is more prominent before and after mold filling and pressure holding. This clearly maps the temporal action mechanism of each physical loss dimension to the simulated energy consumption curve.

[0113] As an anomaly handling mechanism, if it is difficult to identify the three types of losses simultaneously in a certain period, such as when the raw material batch has just changed and the workshop temperature is fluctuating beyond the preset fluctuation threshold, one type of coefficient can be fixed to the value of the previous stable period, and only the other two types of coefficients can be searched to prevent excessive mutual compensation between parameters; for example, if the corrected simulation curve is lower than the ideal curve at a certain stage, it indicates that the coefficient direction is abnormal, which can trigger boundary constraints and judge the particle as invalid to avoid the physical unreasonable situation of negative dissipation value;

[0114] After the air guide plate production entered the night shift, the decrease in ambient temperature increased heat loss; at the same time, since the equipment had been producing continuously for a week without stopping, the temperature rise of the screw and transmission components caused changes in friction characteristics; in addition, a new batch of raw materials was replaced that night, and the peak filling value was slightly higher than that of the day shift; in this embodiment, these three types of effects are encoded into the simulation curves, so that the system no longer only has a single feature of increased total energy consumption, but clearly identifies the specific sources of dissipation;

[0115] The purpose of this mechanism is to decompose the complex and mixed field dissipation into searchable, comparable, and interpretable parameter vectors, thereby achieving a controllable approximation of the simulated energy consumption curve to the real process.

[0116] The total energy consumption time series is calculated based on the energy consumption time series data. This total energy consumption time series is then aligned with the ideal energy consumption curve and the simulated energy consumption curve to generate the actual residual time series and the theoretical residual time series. Furthermore, segmented residual sequences are obtained according to the forming stage, including:

[0117] S401. Resample and sum the servo motor active power time sequence and heating coil active power time sequence in the energy consumption time sequence data according to a unified timestamp to obtain the total energy consumption time sequence. Then align the total energy consumption time sequence with the ideal energy consumption curve and calculate the actual residual time sequence.

[0118] S402. Align the simulated energy consumption curve with the ideal energy consumption curve and calculate the theoretical residual timing.

[0119] S403. Based on the process state timing data and process event parameters, the actual residual timing and theoretical residual timing are divided into periodic segments for the plasticizing stage, the mold filling stage, the pressure holding stage, and the cooling stage, forming a segmented residual sequence.

[0120] This embodiment provides a dual residual construction and segment alignment mechanism. Specifically, in the aforementioned steps, the measured power, ideal curve, and simulation curve have been obtained. However, if the total value of the three curves is directly compared, phenomena such as stage misalignment, peak advance, or lag can easily be ignored. The typical characteristics of the air guide plate injection molding process are usually not reflected in the total difference being greater than the preset energy consumption difference threshold, but rather in the change of the shape of the energy consumption peak at a certain molding stage. Therefore, this embodiment projects the problem onto the same reference benchmark through two paths: actual residual and theoretical residual.

[0121] Specifically, in S401, the active power timing sequence of the servo motor and the active power timing sequence of the heating coil are first resampled and summed according to a unified timestamp. The original sampling point of the servo power is once every 20 milliseconds, and the heating power is once every 100 milliseconds, so it can be uniformly resampled to every 100 milliseconds. If the average servo power is 6kW and the heating power is 4kW within a certain time window, then the total power obtained at that time is 10kW. After being multiplied by the corresponding time window step size, the time node-based variable is obtained. Total energy consumption time series Then, the total energy consumption time sequence Compared with the ideal energy consumption curve Alignment and subtraction at time points yield the actual residual time series. ; where, the operator This indicates a point-by-point element-by-element subtraction operation performed on two sets of time-series data at a unified timestamp node;

[0122] In S402, simulate the energy consumption curve. Compared with the ideal energy consumption curve Alignment was performed to calculate the theoretical residual time series. For example, in the embodiment with the above 8 sampling points, if the total energy consumption time series is [7,9,11,10,8,6,4,3] and the ideal energy consumption curve is [3,4,6,5,4,2,1,1], then the actual residual time series is [4,5,5,5,4,4,3,2]; if the simulated energy consumption curve is

[0123] Given [3.5, 4.7, 7.2, 6.1, 4.9, 2.5, 1.3, 1.2], the theoretical residual time series is:

[0124] [0.5,0.7,1.2,1.1,0.9,0.5,0.3,0.2]; In implementation, both can be further normalized in amplitude or scaled relative to each other, so that subsequent distance calculations focus more on waveform topology rather than absolute order of magnitude;

[0125] In S403, the process is segmented according to the process status and event nodes. If the screw rotation from the start to the stop is detected as the plasticizing stage, the injection from the start to the V / P switch is the mold filling stage, the V / P switch to the end of the holding pressure stage is the holding pressure stage, and the holding pressure end to the mold opening stage is the cooling stage, then the two residual curves can be synchronously divided into four stage segments. For example, when there are 8 points, it can be divided into plasticizing [points 1-2], mold filling [point 3], holding pressure [points 4-5], and cooling [points 6-8]. The segmented residual sequence formed in this way is not only beneficial for calculating the total distance later, but also allows for staged evaluation of whether the plasticizing is unstable or the mold filling peak is abnormal.

[0126] Furthermore, if the actual molding cycle length fluctuates slightly, for example, the total mold time changes from 45 seconds to 47 seconds, then before segmentation, local stretching or compression alignment can be performed according to the event nodes, rather than simply scaling the entire cycle to the same length; this can avoid the cooling stage becoming longer, causing the peak value in the mold filling stage to be incorrectly diluted.

[0127] As an anomaly handling mechanism, if the V / P switching moment is lost but the screw position sequence is available, the corresponding moment can be deduced from the pressure holding switching position; if both the moment and position are abnormal, the pressure slope abrupt change point or the injection speed sudden drop point can be used as the alternative boundary; if the number of effective sampling points in a certain stage is less than the preset number threshold, for example, if there is only one abnormal point left in the filling stage due to sampling failure, then this stage can be marked as low confidence, and its impact can be reduced in subsequent distance weighting to prevent single-point failure from dragging down the whole wheel optimization.

[0128] During the mass production of air guide plates, it was found that the total energy consumption increased by only 3% compared to the day shift on a certain night shift. This may not seem significant in terms of the total amount, but this embodiment found through residual segmentation that: the residual amplitude change in the plasticizing section was less than the preset judgment threshold; the residual peak waveform width in the molding section increased and exceeded the preset widening threshold; the amplitude in the holding pressure section increased but remained within the first tolerance range; and the difference in the corresponding residual values ​​in the cooling section was less than the preset alignment tolerance. This indicates that the problem is more likely to be caused by changes in raw material flowability or injection resistance, rather than environmental heat dissipation. Therefore, subsequent optimization can prioritize adjusting the speed and switching point, rather than adjusting the cooling time based on a single variable.

[0129] The purpose of this step is to transform the total energy consumption difference into a phased, aligned, and comparable residual sequence, thereby enabling more sensitive identification of dissipation features.

[0130] The particle swarm optimization algorithm is used to optimize the preset loss parameter vector. Based on the piecewise residual sequence, the dynamic time warping distance between the actual residual time series and the theoretical residual time series is calculated. The predicted energy consumption curve and / or predicted total energy consumption value for the next molding cycle are output, including:

[0131] S501. Encode the preset loss parameter vector into particle positions;

[0132] S502. Calculate the total dynamic time warping distance and the dynamic time warping distance of each stage between the actual residual time series and the theoretical residual time series based on the piecewise residual sequence.

[0133] S503. Construct a fitness function with the objective of minimizing the weighted sum of the total dynamic time warp distance and / or the dynamic time warp distance of each stage based on preset weights, and update the particle velocity and particle position.

[0134] S504. After reaching the maximum number of iterations and / or the fitness change of multiple consecutive iterations is less than the preset convergence threshold, output the target loss parameter vector and the predicted energy consumption curve and / or the predicted total energy consumption value for the next molding cycle.

[0135] This embodiment provides a loss parameter identification and prediction mechanism based on particle swarm optimization and dynamic time warping distance. Specifically, relying on manual experience to adjust the three coefficients of heat loss, mechanical loss, and viscosity loss makes it difficult to obtain the optimal parameter combination under various disturbance coupling conditions. On the other hand, if only the point-to-point Euclidean distance of the residual sequence is compared, it is easy to misjudge cases where the peak position is slightly delayed but the shape is similar as the distance index being greater than the preset distance difference threshold. Therefore, this embodiment combines the particle swarm optimization algorithm with dynamic time warping distance to adapt to the common staged time stretching and contraction in the injection molding cycle.

[0136] Specifically, in S501, the loss parameter vector is encoded as particle position; the three-dimensional coordinates of each particle correspond to the thermal loss coefficient, mechanical loss coefficient, and viscous loss coefficient, respectively; for example, the particle coordinates are P1=0.12,0.06,0.05, P2=0.18,0.09,0.07, P3=0.22,0.11,0.10; at the same time, a velocity vector is set for each particle to control the search step size in the next round; the search boundary can be set according to process experience, such as the thermal loss coefficient range of 0 to 0.40, the mechanical loss coefficient range of 0 to 0.25, and the viscous loss coefficient range of 0 to 0.20;

[0137] In S502, a simulation curve and theoretical residual are generated for each particle, and then compared with the actual residual in segments using dynamic time warping. To make the implicit calculation process of dynamic time warping explicit, the specific structured calculation logic is as follows: For any forming stage segment, a cost matrix is ​​constructed with the actual residual time series segment as the row and the theoretical residual time series segment as the column. The matrix elements represent the absolute error of the difference at the corresponding time point. Dynamic programming is used to search for an alignment path with the minimum cumulative error from the upper left corner to the lower right corner in the cost matrix. The minimum cumulative error of this path is the dynamic time warping distance of the corresponding stage.

[0138] For ease of explanation, it is assumed that the distance values ​​for the four stages of plasticizing, mold filling, pressure holding, and cooling are calculated separately. , , , The total distance is If the distances for each segment corresponding to P1 are [1.2, 2.5, 1.1, 0.8], and the total distance is 5.6; for P2, the distances are [0.9, 1.4, 0.8, 0.7], and the total distance is 3.8; and for P3, the distances are [1.5, 1.6, 1.2, 0.9], and the total distance is 5.2, then it can be preliminarily seen that P2 is closer to the actual dissipation state on site. The significance of dynamic time warping is that if the actual peak value of the mold filling is one sampling point later than the simulated peak value, but the overall peak shape is similar, a smaller distance can still be obtained, and it will not be over-amplified as in a strict point-to-point comparison.

[0139] In S503, a fitness function is constructed, and the velocity and position are updated based on the individual and swarm optima of particles. To ensure the uniqueness and consistency of mathematical notation and terminology, fitness evaluation no longer mixes old and new notations, but instead uses fitness values ​​calculated based on the aforementioned distance components. The weighted formula is:

[0140]

[0141] in, The total dynamic time-normalized distance. , , , These are the dynamic time warping distances corresponding to the plasticizing stage, the mold filling stage, the pressure holding stage, and the cooling stage, respectively. This is the total distance weighting coefficient. to These are the distance weighting coefficients for the corresponding stages; for example, under a specific setting, this formula can be expressed as:

[0142]

[0143] The weights are not fixed; if defects in the air guide plate are more likely to occur at the end of the mold filling process, the corresponding weights during the mold filling and holding stages can be increased. and The weighting index; after the particle is updated, if the new position goes out of bounds, boundary clipping or bounce processing is performed to make it fall back into the legal range;

[0144] In S504, the search stops when the maximum number of iterations is reached, or when the fitness change is below the convergence threshold for several consecutive rounds, and the target loss parameter vector is output. For example, after 12 iterations, the population optimal converges to (0.17, 0.10, 0.08). The predicted energy consumption curve for the next forming cycle is then reconstructed using this set of parameters, and the predicted total energy consumption value can be obtained by integration or summation. For example, the predicted total energy consumption is 67.5 kJ. If the system updates using a sliding window, the optimal parameters of the most recent N cycles can be smoothed to avoid large fluctuations in the control quantity due to accidental anomalies in a certain mode.

[0145] As an anomaly handling mechanism, if the fitness difference of each particle is less than the preset proximity threshold after the first few rounds of searching, it indicates that the current data is not good at distinguishing the three types of loss. At this time, a conservative mode can be enabled, that is, the adjustment range of the control side is reduced, and only the prediction results are updated without immediately modifying the process parameters whose range is greater than the preset parameter adjustment threshold. If the dynamic time warping path is abnormally distorted, for example, a large number of repeated matchings are caused by noise in a certain stage, the stage can be smoothed in advance or a maximum distortion window can be set to prevent the distance index from being distorted. If the search does not converge for a long time, the number of particles can be increased, some particles can be reset, or the parameter center of the previous stable batch can be re-initialized.

[0146] In the production of air guide plates, the peak filling time of a certain period shifts back by about 0.15 seconds relative to the overall day shift, but the peak amplitude fluctuation is less than the preset amplitude judgment threshold. If ordinary point-to-point comparison is used, the system will mistakenly think that the viscosity loss has increased and exceeded the preset warning line. However, this embodiment identifies that the problem is mainly a time alignment issue through dynamic time warping. Finally, the particle swarm converges to the result that the coefficient corresponding to heat loss increases positively, the mechanical loss is stable, and the coefficient corresponding to viscosity loss increases within the allowable tolerance range, making the energy consumption prediction of the next mold more consistent with the actual situation on site.

[0147] The purpose of this mechanism is to improve the tolerance of similarity metrics to stage drift through dynamic time warping, and to achieve efficient search of loss parameters by using particle swarm optimization, thereby enabling reliable energy consumption prediction for the next cycle.

[0148] Based on the predicted energy consumption curve and / or predicted total energy consumption value, the preset molding quality threshold, and the preset quality prediction model, the particle swarm optimization algorithm is used to optimize the set of control parameters to be optimized, obtain the target control parameters, generate control commands, and send them to the controller of the injection molding equipment, including:

[0149] S601, construct the barrel temperature setpoint, injection speed setpoint, holding pressure setpoint, and holding pressure switching point setpoint as a set of control parameters to be optimized;

[0150] S602. Based on the target loss parameter vector, calculate the candidate energy consumption results corresponding to each candidate control parameter in the set of control parameters to be optimized.

[0151] S603. Based on the preset quality prediction model constructed using historical process setting parameters and corresponding offline product quality inspection datasets, calculate the candidate quality indicators corresponding to each candidate control parameter.

[0152] S604. Based on the preset molding quality threshold, the candidate quality indicators are constrained and judged, and the candidate energy consumption results that meet the constraint judgment are screened.

[0153] S605. Select the target control parameter from the control parameters that meet the preset molding quality threshold, and generate control instructions.

[0154] This embodiment provides a mechanism for secondary optimization of control parameters under quality constraints. Specifically, although the aforementioned steps have obtained relatively accurate loss parameters and energy consumption predictions for the next cycle, this only completes the identification of the actual dissipation state of the current equipment. If the parameter with the lowest predicted energy consumption is directly used as the control target, it is easy for energy consumption to decrease while the forming quality is compromised. The air guide plate is sensitive to appearance, size, and warping. Therefore, this embodiment introduces a quality prediction model and a quality threshold to limit the optimal energy consumption under the premise that the quality first meets the standard.

[0155] Specifically, in S601, the barrel temperature setpoint is... Injection speed setting value Pressure holding pressure setting value and pressure holding switching point setting value The set of control parameters to be optimized is constructed; it should be noted that, considering the multi-stage control characteristics of modern injection molding machines, the barrel temperature setpoint... Injection speed setting value Pressure holding pressure setting value It can be a single scalar set for global settings or a multi-dimensional vector set for multi-level control settings; in order to avoid the particle swarm optimization dimension being too high and causing the computing power to be difficult to converge, and also for the convenience of the text description of this embodiment, the following specific examples are mainly based on the setting values ​​of a single scalar or representative key segments.

[0156] Using the aforementioned test example, in order to strictly distinguish it from the time-series function variables of the multi-segment barrel temperature in S200, such as T1, to ensure symbol uniqueness, and at the same time to strictly map it with the weighted terms, a particle can be represented as a parameter quadruple containing the above-mentioned set of control parameters. Three candidate particles can be set:

[0157]

[0158]

[0159]

[0160] In S602, based on the identified target loss parameter vector, each set of candidate control parameters is substituted into the prediction model to obtain candidate energy consumption results. This means that under the same equipment dissipation background, different process settings will change the mold filling resistance, pressure holding and feeding, and heating load, thus generating different candidate energy consumption curves and total energy consumption values. If the current target loss parameter is (0.17, 0.10, 0.08), then... , , The corresponding predicted total energy consumptions are 67.2 kJ, 65.9 kJ, and 64.8 kJ, respectively.

[0161] In S603, a preset quality prediction model is used to calculate candidate quality indicators for each group of candidate control parameters. The quality indicator can be a single score or a probability of compliance obtained by combining factors such as warpage risk, short-shot risk, and weight stability. A quality score between 0 and 1 can be output. It is 0.95. It is 0.91. The score is 0.72; the higher the score, the greater the possibility of meeting the requirements for the formation of the air guide plate; in order to avoid the preset quality prediction model becoming an untraceable implicit mapping mechanism, it can be implemented by pre-collecting offline quality inspection datasets containing candidate control quantities for each historical cycle and corresponding module products, such as warpage, weighing data and short shot records from coordinate measuring machines, and constructing them using supervised learning algorithms such as support vector regression.

[0162] The specific construction process includes: extracting known barrel setting temperature, injection setting speed, holding pressure setting pressure, holding pressure switching point, and the actual loss parameter vector identified through calculation within the historical complete molding cycle as the input feature vector of the model; using the actual warpage, actual weight deviation, and short shot records obtained from offline quality inspection of the corresponding batch of products as the output label of the model; and using the historical mapping dataset constructed above to train the support vector regression model offline and determine the model hyperparameters, so that the model can deduce quantitative predicted defect physical quantities in real time based on the input candidate control parameters when optimizing online.

[0163] The calculation process is as follows: First, the input candidate control parameter set and the current target loss parameter vector are mapped to the estimated melt stress and volume shrinkage rate characteristics. The specific physical quantity of the predicted defect is output, such as the predicted warpage of 0.5 mm. According to the tolerance limit in the product specification, such as the maximum allowable warpage of 1.0 mm, the normalized quality score is calculated by dividing the difference between the upper limit of the tolerance and the predicted defect quantity by the upper limit of the tolerance. When there are multiple defect risks, the weighted minimum value of the normalized scores is taken as the comprehensive candidate quality index to ensure that the molding bottom line is not breached.

[0164] In S604, constraint determination is performed based on a preset molding quality threshold; assuming the threshold is set to 0.90, then... and It can proceed to the next round of energy consumption screening, and Although it had the lowest energy consumption, it was eliminated because its quality score was only 0.72. The remaining candidates were compared in terms of energy consumption, and the group with the lowest energy consumption that met the quality constraints was selected.

[0165] In S605, the target control parameters are determined and control commands are generated; taking the above example, Simultaneously meeting the quality threshold and having the lowest energy consumption among qualified candidates, it is selected as the target control parameter; the control instructions can be further converted into a setting format recognizable by the programmable logic controller or injection molding machine controller, such as writing the target temperature of the barrel in zones one to four, the injection speed in zones one to three, the holding pressure in zone one, the position switching register address, etc., which will take effect at the start of the next molding cycle or the next batch.

[0166] As an anomaly handling mechanism, if all candidate control parameters fail to meet the quality threshold, the low-energy consumption scheme will not be selected. Instead, it will automatically revert to the control parameters of the previous stable cycle, or narrow the search range around the center of the previous stable parameter to re-optimize. If a candidate parameter has a quality score slightly lower than the threshold, but the difference between it and the threshold is less than the preset tolerance threshold, it can be marked as a boundary candidate and used only for offline analysis, without entering the online distribution process. If the controller does not support writing all parameters at once, it can be distributed step by step in the order of updating temperature first, followed by speed and pressure holding, and the device will be checked to ensure that it is in a safe window that allows parameter modification before distribution.

[0167] During production periods with lower ambient temperatures, the system found that due to the cooling of the environment, appropriately increasing the injection speed and slightly reducing the holding pressure was more effective in reducing energy consumption than simply lowering the barrel temperature. The quality prediction model also determined that if the material temperature dropped too much, the risk of missing glue at the thin-walled end would increase rapidly. Therefore, the final choice was a combination of medium material temperature, higher injection speed, and moderately reduced holding pressure, rather than the combination with the lowest theoretical energy consumption but unstable quality.

[0168] The purpose of this mechanism is to incorporate energy consumption optimization into the quality constraint framework in order to achieve stable energy consumption control in the production of air guide plates, rather than simply reducing energy consumption without considering quality constraints.

[0169] Candidate quality indicators are constrained and determined based on a preset molding quality threshold, and candidate energy consumption results that meet the constraints are then screened, including:

[0170] S6041. When a candidate quality index is lower than the preset molding quality threshold, the corresponding control parameter is determined to be an invalid parameter and is removed.

[0171] S6042. When a candidate quality index is greater than or equal to a preset molding quality threshold, the corresponding control parameters are retained and the corresponding candidate energy consumption results are recorded.

[0172] S6043. Select the control parameter with the minimum predicted total energy consumption from the retained control parameters as the target control parameter.

[0173] This embodiment provides a candidate control parameter screening mechanism under quality gating. Specifically, the above-mentioned secondary optimization has incorporated quality prediction into the decision-making process. However, if it is not clearly specified when to eliminate, when to retain, and when to select the minimum energy consumption, inconsistencies in screening criteria between different batches are likely to occur during online control. Therefore, this embodiment further refines the constraint judgment process so that the controller can execute clear and definite decisions.

[0174] Specifically, in S6041, when a candidate quality index is lower than the preset molding quality threshold, the control parameters are directly determined to be invalid and removed. The determination and removal operation indicates that the corresponding control parameters will no longer enter the subsequent energy consumption ranking and optimization process. Continuing with the aforementioned sandbox example, if the threshold is 0.90, and the candidate scores are 0.95, 0.91, and 0.72, then the control parameter corresponding to the score of 0.72, although predicting the lowest total energy consumption, cannot be retained. This processing can prevent the system from exceeding the molding threshold in order to save energy.

[0175] In S6042, when a candidate quality index is greater than or equal to a threshold, the control parameters of that group are retained and the corresponding candidate energy consumption results are recorded. This can be achieved using a retention table. The retention table records two rows of data: the first row is C1, corresponding to a quality of 0.95 and an energy consumption of 67.2 kJ; the second row is C2, corresponding to a quality of 0.91 and an energy consumption of 65.9 kJ. At this time, the system will no longer include the previously rejected C3 in the subsequent calculation process because it does not qualify to enter the production instruction set.

[0176] In S6043, the control parameter with the lowest predicted total energy consumption is selected from the retention table as the target control parameter. As can be seen from the above data, C2 is the lowest energy consumption item in the qualified set, so it is selected. If multiple candidate control parameters have the same predicted total energy consumption, the quality indicators can be compared, and the scheme with a larger quality margin can be retained first. If the quality indicators are also the same, the scheme with the smaller difference from the current equipment setting can be selected first to reduce parameter jumps.

[0177] Furthermore, to improve online stability, a dual-threshold mechanism can be set. For example, the preset forming quality threshold includes an entry quality threshold and an early warning quality threshold, with the entry quality threshold being 0.90 and the early warning quality threshold being 0.93. When the candidate quality is between 0.90 and 0.93, although it can be retained, it will be marked as critically qualified. If there is another set of control parameters with slightly higher energy consumption but significantly more stable quality, the system can select the more stable set according to the preset strategy. This can reduce the risk of batch fluctuations when the air guide plate is sensitive to boundary conditions.

[0178] As an exception handling mechanism, if the reserved table is empty, it means that the current search has not met the target. At this time, no new low energy consumption command will be issued. Instead, it will automatically trigger re-optimization, widen the search range and roll back, or restore to the previous stable parameter group. If the reserved table has only one set of control parameters, and the energy consumption of this set is not low but the quality is the most stable, it can still be issued as the target control parameter. If the difference between the lowest energy consumption scheme in the reserved table and the current operating parameters is greater than the preset parameter difference threshold, such as the material temperature is to be reduced by more than 15°C, the ramp transition strategy can be enabled to gradually approach the target value in multiple cycles.

[0179] In the mass production of air guide plates, a search revealed that one set of ultra-low energy consumption solutions would result in insufficient filling of the terminal area, with a quality prediction value of only 0.86; another set of solutions had slightly higher energy consumption of 1.1kJ, but the quality prediction value reached 0.94; in this embodiment, according to the gating rules, the corresponding control parameters will be determined to be invalid parameters and the former will be eliminated, while the latter will be retained and finally issued, so that the energy-saving results are based on stable qualified products.

[0180] The purpose of this mechanism is to ensure the consistency and feasibility of control decisions by clearly prioritizing quality over energy consumption, thereby achieving industrial availability of the online optimization process.

[0181] Example 2:

[0182] Also includes:

[0183] S700: Obtain the set of process data for the new molding cycle after the controller executes the control command;

[0184] S800: Update the actual residual time series based on the new forming cycle process data set, and calculate the dynamic time warping distance between it and the theoretical residual time series, or verify the similarity value obtained by converting the dynamic time warping distance based on the preset mapping function.

[0185] S900: When the dynamic time warp distance is higher than the preset distance threshold, or the similarity value is lower than the preset similarity threshold, re-execute S200 to S600; when the dynamic time warp distance is not higher than the preset distance threshold, and the similarity value is equal to or higher than the preset similarity threshold, maintain the current target loss parameter vector and target control parameters.

[0186] This embodiment provides a closed-loop mechanism for review and re-optimization after control is issued. Specifically, the preceding steps have completed prediction and issuance, but the injection molding site is constantly changing. If the control command is fixed for a long time after it is issued, the original optimal parameters may quickly become invalid as the ambient temperature, mold temperature heat exchange, raw material batches, and even equipment friction state continue to change. Therefore, this embodiment continues to collect new cycle data after control execution, reviews the actual effect after issuance, and decides whether to restart the whole process optimization.

[0187] Specifically, in the S700, after the controller executes the new settings, the system acquires the process data set of the new molding cycle. The acquired content is consistent with the above, including process status, energy consumption, environmental status and event parameters. This ensures that the data structure of the previous and subsequent cycles is consistent, which is convenient for direct comparison. If the new control parameters have reduced the holding pressure from 62MPa to 60MPa, the system will fully record the actual power and quality-related process characteristics after the adjustment in the next mold.

[0188] In S800, the actual residuals are recalculated based on the new cycle data and compared with the current theoretical residuals through dynamic time warping. To facilitate rapid decision-making, the distance value can be used directly during implementation, or it can be converted into a similarity value through a preset mapping function. For example, a smaller distance can be mapped to a higher similarity. Suppose that the total dynamic time warped distance obtained in a certain new cycle is 3.2, corresponding to a similarity of 0.93; the distance in another cycle is 6.8, with a similarity of only 0.71. The former indicates that the current model is still relatively consistent with the field, while the latter indicates that the environmental or equipment state has deviated significantly.

[0189] In S900, a review decision is made based on a threshold. If the dynamic time warp distance is higher than the preset distance threshold, or the similarity value is lower than the preset similarity threshold, the entire process from theoretical model construction to control parameter optimization is re-executed. Otherwise, the current target loss parameter vector and target control parameters are maintained. If the distance threshold is set to 4.5 and the similarity threshold is set to 0.85, the current settings are maintained when the distance is 3.2 and the similarity is 0.93. When the distance is 6.8 and the similarity is 0.71, a recalculation is triggered immediately. This judgment can be triggered in a single cycle or confirmed in two to three consecutive cycles to reduce false triggering caused by occasional noise.

[0190] Furthermore, the mapping function can adopt a monotonically decreasing relationship, so that the greater the distance, the lower the similarity; in implementation, it is not required to be fixed in a certain specific form, as long as the similarity after mapping can be used for stable comparison; for the production site, using similarity values ​​is easier for process personnel to understand, for example, the current model matching degree of 92% can be displayed on the monitoring interface;

[0191] As an anomaly handling mechanism, if the data collection for a new cycle is incomplete, resulting in insufficient confidence in the distance calculation, then that cycle is only used as a reference and will not be immediately re-optimized, but will wait for the next complete cycle for review. In practice, the system can first calculate the segmented dynamic time warp distance of each forming stage, and then convert it into segmented similarity by a preset nonlinear monotonically decreasing mapping function and perform weighted fusion. This nonlinear mapping mechanism may cause the total dynamic time warp distance and the comprehensive weighted similarity to not be absolutely equivalent.

[0192] Therefore, if the distance threshold and similarity threshold give contradictory judgments, for example, if the distance is greater than the preset single-segment distance threshold due to a minor stage, such as the end of the cooling period, resulting in the total dynamic time regularization distance being slightly higher than the preset distance threshold, but the high matching in the core stage, such as the filling stage, makes the weighted comprehensive similarity equal to or higher than the preset similarity threshold, then a more conservative strategy can be adopted. For example, first enter the observation state, and then continuously review for one cycle before making a decision. If the deviation between the product quality inspection result and the prediction after the control command is executed is greater than the preset quality verification threshold, the modeling and optimization steps can be directly triggered to be re-executed regardless of the distance.

[0193] In the latter half of the night during the production of the air guide plate, although the model matching degree of the first two molds remained above 0.90, the third mold, due to the fluctuation of the mold temperature controller, caused the residual of the cooling stage to deviate significantly, resulting in a drop in similarity to 0.78. Based on this, the system automatically re-executed the theoretical baseline update, loss parameter identification, and control parameter search, readjusted the cooling-related settings and pressure holding stage parameters, and restored the production line to a new stable and low-energy consumption state.

[0194] The purpose of this mechanism is to enable the control system to have the ability to self-verify and continuously correct after the system is issued, thereby achieving closed-loop stable optimization of the air guide plate injection molding process in long-term operation.

[0195] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting and controlling energy consumption in the injection molding process of a guide vane based on particle swarm optimization algorithm, characterized in that, include: S100: Acquire the process status time sequence data, energy consumption time sequence data, environmental status data and process event parameters of the injection molding equipment during the air guide plate molding cycle, and synchronize them according to a unified timestamp or a unified periodic trigger signal to construct a process data set. S200. Based on the process data set, construct a theoretical energy consumption reference model determined by preset raw material thermophysical property parameters, preset mold filling parameters and preset molding stage time allocation rules, and generate an ideal energy consumption curve. S300. Based on the ideal energy consumption curve and the preset loss parameter vector, the simulated energy consumption curve is obtained by correction. S400. Calculate the total energy consumption time series based on the energy consumption time series data, align the total energy consumption time series with the ideal energy consumption curve and the simulated energy consumption curve respectively, generate the actual residual time series and the theoretical residual time series, and segment them according to the forming stage to obtain the segmented residual sequence. S500. Taking the minimum dynamic time warping distance between the actual residual time series and the theoretical residual time series as the optimization objective, the particle swarm optimization algorithm is used to optimize the preset loss parameter vector, and the predicted energy consumption curve and / or predicted total energy consumption value of the next forming cycle are output according to the parameters after optimization convergence. S600: Extract the set parameters of the injection molding equipment to construct a set of control parameters to be optimized. Based on the predicted energy consumption curve and / or the predicted total energy consumption value, the preset molding quality threshold and the preset quality prediction model, the particle swarm optimization algorithm is used to optimize the set of control parameters to be optimized, obtain the target control parameters, generate control commands and send them to the controller of the injection molding equipment.

2. The method for energy consumption prediction and control in the injection molding process of a guide vane based on particle swarm optimization algorithm as described in claim 1, characterized in that, The process status timing data includes multi-segment barrel temperature timing, injection pressure timing, injection speed timing, and screw torque timing; The process event parameters include pressure holding switching time and / or pressure holding switching location parameters; The energy consumption time series data includes the active power time series of the servo motor and the active power time series of the heating coil, which are collected by the power acquisition unit; The environmental status data includes the temperature difference between the inlet and outlet water of the mold temperature controller and the ambient temperature of the workshop, which are collected by the environmental sensing unit. The process data set also includes the corresponding current waveform used to correct the power acquisition results.

3. The method for energy consumption prediction and control in the injection molding process of a guide vane based on particle swarm optimization algorithm according to claim 1, characterized in that, Based on the process data set, a theoretical energy consumption reference model is constructed, determined by preset raw material thermophysical properties, preset mold filling parameters, and preset molding stage time allocation rules, to generate an ideal energy consumption curve, including: S201. Based on the specific heat parameters and melting temperature range of the raw material corresponding to the air guide plate, calculate the theoretical heating energy consumption of the raw material from the initial temperature to the molten state, wherein the specific heat parameters and the melting temperature range belong to the preset raw material thermal property parameters. S202. Based on the preset mold filling parameters, which include at least cavity geometry parameters, flow channel resistance parameters, melt viscosity parameters, and mold filling pressure parameters, calculate the theoretical flow energy consumption of the melt filling process. S203. Based on the preset duration or actual timing nodes of the plasticizing stage, molding stage, holding pressure stage, and cooling stage as the preset molding stage time allocation rules, the theoretical heating energy consumption and the theoretical flow energy consumption are allocated over time to generate the ideal energy consumption curve.

4. The method for energy consumption prediction and control in the injection molding process of a guide vane based on particle swarm optimization algorithm according to claim 3, characterized in that, The preset loss parameter vector includes the heat loss coefficient, mechanical loss coefficient, and viscosity loss coefficient; The simulated energy consumption curve is obtained based on the ideal energy consumption curve and the preset loss parameter vector correction, including: S301. Establish the heat loss coefficient corresponding to environmental heat dissipation; S302. Establish the mechanical loss coefficients corresponding to screw mechanical wear and transmission friction; S303. Establish the viscosity loss coefficient corresponding to the non-Newtonian rheological fluctuations of the raw materials. S304. Based on the heat loss coefficient, the mechanical loss coefficient and the viscosity loss coefficient, the ideal energy consumption curve is corrected to generate the simulated energy consumption curve.

5. The method for energy consumption prediction and control in the injection molding process of a guide vane based on particle swarm optimization algorithm according to claim 1, characterized in that, The total energy consumption time series is calculated based on the energy consumption time series data. This total energy consumption time series is then aligned with the ideal energy consumption curve and the simulated energy consumption curve to generate the actual residual time series and the theoretical residual time series. Finally, segmented residual sequences are obtained according to the forming stage, including: S401. The servo motor active power time sequence and heating coil active power time sequence in the energy consumption time sequence data are resampled and summed according to a unified timestamp to obtain the total energy consumption time sequence. Then, the total energy consumption time sequence is aligned with the ideal energy consumption curve to calculate the actual residual time sequence. S402. Align the simulated energy consumption curve with the ideal energy consumption curve, and calculate the theoretical residual timing. S403. Based on the process state timing data and the process event parameters, the actual residual timing and the theoretical residual timing are periodically segmented into the plasticizing stage, the mold filling stage, the pressure holding stage, and the cooling stage to form the segmented residual sequence.

6. The method for energy consumption prediction and control in the injection molding process of a guide vane based on particle swarm optimization algorithm according to claim 5, characterized in that, The particle swarm optimization algorithm is used to optimize the preset loss parameter vector. Based on the segmented residual sequence, the dynamic time warping distance between the actual residual time series and the theoretical residual time series is calculated. The predicted energy consumption curve and / or predicted total energy consumption value for the next molding cycle are output, including: S501. Encode the preset loss parameter vector into particle positions; S502. Calculate the total dynamic time warping distance and the dynamic time warping distance of each stage between the actual residual time series and the theoretical residual time series based on the segmented residual sequence. S503. Construct a fitness function with the goal of minimizing the weighted sum of the total dynamic time warp distance and / or the dynamic time warp distance of each stage based on a preset weight, and update the particle velocity and particle position. S504. After reaching the maximum number of iterations and / or the fitness change of multiple consecutive iterations is less than the preset convergence threshold, output the target loss parameter vector and the predicted energy consumption curve and / or the predicted total energy consumption value of the next molding cycle.

7. The method for energy consumption prediction and control in the injection molding process of a guide vane based on particle swarm optimization algorithm according to claim 6, characterized in that, Based on the predicted energy consumption curve and / or predicted total energy consumption value, the preset molding quality threshold, and the preset quality prediction model, a particle swarm optimization algorithm is used to optimize the set of control parameters to be optimized, obtain the target control parameters, generate control commands, and send them to the controller of the injection molding equipment, including: S601, The barrel temperature setpoint, injection speed setpoint, holding pressure setpoint, and holding pressure switching point setpoint are constructed into the set of control parameters to be optimized; S602. Based on the target loss parameter vector, calculate the candidate energy consumption results corresponding to each candidate control parameter in the set of control parameters to be optimized; S603. Based on the preset quality prediction model constructed using historical process setting parameters and corresponding offline product quality inspection datasets, calculate the candidate quality indicators corresponding to each candidate control parameter. S604. Based on the preset molding quality threshold, the candidate quality indicators are constrained and determined, and the candidate energy consumption results that meet the constraint determination are screened. S605. Select the target control parameter from the control parameters that meet the preset molding quality threshold, and generate the control command.

8. The method for energy consumption prediction and control in the injection molding process of a guide vane based on particle swarm optimization algorithm according to claim 7, characterized in that, Based on the preset molding quality threshold, the candidate quality indicators are constrained and determined, and the candidate energy consumption results that meet the constraints are screened, including: S6041. When the candidate quality index is lower than the preset molding quality threshold, the corresponding control parameter is determined to be an invalid parameter and is removed. S6042. When the candidate quality index is greater than or equal to the preset molding quality threshold, retain the corresponding control parameter and record the corresponding candidate energy consumption result. S6043. Select the control parameter with the minimum predicted total energy consumption from the retained control parameters as the target control parameter.

9. The method for energy consumption prediction and control in the injection molding process of a guide vane based on particle swarm optimization algorithm according to claim 1, characterized in that, Also includes: S700: Obtain the new molding cycle process data set after the controller executes the control command; S800. Update the actual residual time series based on the new forming cycle process data set, and calculate the dynamic time warping distance between it and the theoretical residual time series, or verify the similarity value obtained by converting the dynamic time warping distance based on the preset mapping function. S900: When the dynamic time warping distance is higher than the preset distance threshold, or the similarity value is lower than the preset similarity threshold, S200 to S600 are re-executed; when the dynamic time warping distance is not higher than the preset distance threshold, and the similarity value is equal to or higher than the preset similarity threshold, the current target loss parameter vector and target control parameters are maintained.

Images