System and method for setting granular mixture pressing process parameters based on Bayesian optimization

Abstract

The invention provides a granular mixture pressing process parameter setting system and method based on Bayesian optimization, and relates to the technical field of powder metallurgy and granular mixture forming equipment. The system comprises a pressing execution subsystem used for executing physical pressing action on granular mixtures and collecting multi-dimensional state data in the pressing process in real time; the multi-source sensing subsystem is used for receiving and preprocessing the multi-dimensional state data; the lower control subsystem is used for performing real-time closed-loop control on the physical suppression action executed by the suppression execution subsystem according to the received process parameter setting instruction; the upper decision optimization subsystem is used for generating predicted molding density according to the received and stored multi-dimensional state data; and with prediction of the molding density as an optimization target, a Bayesian optimization algorithm is adopted to update a process parameter set value, and a process parameter setting instruction is generated. The problems that in the prior art, pressing parameter adjustment depends on manual trial and error, precision is low, and efficiency is poor are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of powder metallurgy and granular material forming equipment, and in particular to a system and method for setting process parameters for granular material pressing based on Bayesian optimization. Background Technology

[0002] The quality of granular material compression molding is highly dependent on the precise setting of process parameters, with compression pressure and holding time being the most critical factors determining the molding density. Existing compression equipment control systems typically employ a traditional "open-loop setting" mode, where operators pre-input a set of fixed process parameters (such as pressure and holding time) onto the Human-Machine Interface (HMI) based on experience, and the PLC (Programmable Logic Controller) mechanically executes these instructions. However, in actual production, uncontrollable disturbances such as raw material batch fluctuations, mold wear, and changes in environmental temperature and humidity often cause fluctuations in the density of products produced under fixed parameters. To ensure product quality, the current common practice is an offline manual intervention mode of "trial compression-measurement-adjustment." This approach has significant lag and relies excessively on human experience, resulting in high trial-and-error costs. Although some machine learning-based quality prediction algorithms (such as the Physical Information Neural Network PINN) have emerged in academia, most of these algorithms run on offline computers and fail to achieve deep physical integration with the press hardware control system, thus failing to form a closed-loop optimization hardware system capable of automatic sensing, automatic decision-making, and automatic instruction issuance. Therefore, there is an urgent need for an intelligent device that can integrate a high-precision quality prediction model and an efficient parameter optimization algorithm into the control loop, so as to realize the automated optimization and adaptive adjustment of pressing process parameters. Summary of the Invention

[0003] To address the shortcomings of the existing technologies, this invention proposes a Bayesian optimization-based system and method for setting process parameters for granular material pressing by integrating multi-source state perception, physical information neural network density prediction, and Bayesian online optimization technology. This aims to solve the problems of low accuracy and poor efficiency in the existing technologies, where pressing parameter adjustment relies on manual trial and error.

[0004] On the one hand, this invention proposes a Bayesian optimization-based system for setting process parameters of granular material pressing, the system comprising:

[0005] The compression execution subsystem is used to perform physical compression actions on the granules and to collect raw signals of multi-dimensional state data in real time during the compression process.

[0006] The multi-source sensing subsystem is communicatively connected to the suppression execution subsystem. It is used to receive the raw signals of multi-dimensional state data and preprocess them to obtain standardized multi-dimensional state data.

[0007] The lower-level control subsystem is electrically connected to the pressing execution subsystem and the multi-source sensing subsystem, respectively. It is used to set instructions according to the received process parameters and to perform real-time closed-loop control of the physical pressing action executed by the pressing execution subsystem based on the standardized multi-dimensional state data.

[0008] The upper-level decision optimization subsystem is communicatively connected to the multi-source sensing subsystem and the lower-level control subsystem, respectively, for receiving and storing standardized multi-dimensional state data; based on the received multi-dimensional state data, it generates a predicted forming density using a pre-constructed granular forming density prediction model; and using the predicted forming density as the optimization target, it calculates the granular forming process parameters for the next pressing cycle using a Bayesian optimization algorithm, and generates a process parameter setting command to be sent to the lower-level control subsystem.

[0009] Furthermore, the multidimensional state data includes: real-time forming pressure, real-time vertical displacement of the moving crossbeam, and mold temperature during the pressing process.

[0010] Furthermore, the pressing execution subsystem includes: a frame, a hydraulic master cylinder, a movable crossbeam, a guide mechanism, a mold worktable, a pressure sensor, a displacement sensor, a temperature sensor, and a hydraulic system;

[0011] The hydraulic master cylinder is fixedly mounted on the top of the frame; the guide mechanism is vertically mounted on the frame; the movable crossbeam is slidably mounted on the guide mechanism and rigidly connected to the piston rod of the hydraulic master cylinder; the hydraulic system is connected to the hydraulic master cylinder and is used to receive hydraulic execution commands and drive the hydraulic master cylinder so that the movable crossbeam moves vertically reciprocating along the guide mechanism under the drive of the hydraulic master cylinder.

[0012] The mold workbench is fixedly installed below the machine frame and is positioned opposite to the movable crossbeam. It is used to install the lower mold of the forming mold, and the bottom of the movable crossbeam is used to install the upper mold of the forming mold.

[0013] The pressure sensor is installed between the movable crossbeam and the piston rod of the hydraulic master cylinder or on the force transmission path of the molding die, and is used to collect the raw signal of the real-time molding pressure.

[0014] The displacement sensor is mounted on the frame or movable crossbeam and is used to collect the raw signal of the real-time vertical displacement of the movable crossbeam.

[0015] The temperature sensor is installed inside or on the surface of the molding die to collect the raw temperature signal of the die during the pressing process.

[0016] Furthermore, the multi-source sensing subsystem includes: a PLC acquisition module, a temperature acquisition module, and a data processing module;

[0017] The PLC acquisition module is used to receive the original signal of the real-time molding pressure and the original signal of the real-time vertical displacement.

[0018] The temperature acquisition module is used to receive the raw signal of the mold temperature;

[0019] The data processing module is used to synchronize the original signals of real-time molding pressure, real-time vertical displacement, and mold temperature in time, and send the synchronized real-time molding pressure signal and real-time vertical displacement signal to the lower-level control subsystem, and send the synchronized mold temperature signal to the upper-level decision optimization subsystem.

[0020] Furthermore, the lower-level control subsystem includes: a programmable logic controller (PLC);

[0021] The lower-level control subsystem includes a programmable logic controller (PLC), which compares the process parameter setting value in the received process parameter setting instruction with the real-time forming pressure signal and the real-time vertical displacement signal, generates a pressure closed-loop control instruction based on the comparison result, and sends the pressure closed-loop control instruction to the hydraulic system to control the opening and closing of the hydraulic valve group in the hydraulic system, thereby driving the hydraulic master cylinder to realize real-time closed-loop control of the pressing force.

[0022] The process parameter settings include: target pressure setting. Pressure holding time setting value Pressurization speed setting value and molding quality setting value .

[0023] Furthermore, the communication interface of the programmable logic controller (PLC) is an Ethernet interface, and the PLC also has a holding register inside for storing process parameter settings.

[0024] Furthermore, the higher-level decision optimization subsystem includes: a quality prediction module and a Bayesian optimization module;

[0025] The quality prediction module is used to generate a predicted forming density based on the received multidimensional state data and the current process parameter settings, using a granular forming density prediction model based on physical information neural network, and send it to the Bayesian optimization module.

[0026] The Bayesian optimization module is used to generate the process parameter setting values ​​for the next pressing cycle based on the preset target molding density, the predicted molding density, and the mold temperature signal, with the process parameter setting values ​​as the optimization target, using the Bayesian optimization algorithm, and then generating the process parameter setting instructions for the next pressing cycle and writing them into the holding register.

[0027] Furthermore, the granular material pressing process parameter adaptive setting system also includes a power distribution cabinet, which provides power and circuit protection for the pressing execution subsystem, the multi-source sensing subsystem, and the lower-level control subsystem.

[0028] On the other hand, this invention proposes a method for setting process parameters for granular material pressing based on Bayesian optimization, which includes the following steps:

[0029] For the compression process of any granular material, set the initial process parameter settings for the first compression cycle and execute it, and collect multidimensional state data during the compression cycle.

[0030] Starting from the second suppression cycle, for each current suppression cycle, perform the following procedure:

[0031] Based on the multidimensional state data from the previous pressing cycle, a predicted forming density is generated using a granular forming density prediction model based on a physical information neural network.

[0032] Based on the process parameter settings and predicted molding density in the previous pressing cycle, the process parameter settings for the current pressing cycle are generated using a Bayesian optimization algorithm.

[0033] Based on the process parameter settings during the current pressing cycle, execute the current pressing cycle and collect multi-dimensional status data during the current pressing cycle.

[0034] Repeat the pressing cycle several times until the entire granular material pressing process is completed, thereby achieving adaptive optimization of the pressing process parameters.

[0035] Furthermore, the specific content of generating the process parameter settings for the current pressing cycle using a Bayesian optimization algorithm based on the process parameter settings and predicted molding density from the previous pressing cycle is as follows:

[0036] Based on the predicted molding density and the preset target molding density, with the objective function being to minimize the error between the predicted molding density and the target molding density, a Bayesian algorithm is used to iteratively optimize the process parameter settings in the previous pressing cycle, and the following process is executed in each iteration:

[0037] New sample points are generated by combining the process parameter settings during the current pressing cycle with the predicted molding density. And add to the historical sample set ;in This indicates the predicted molding density;

[0038] Based on the updated historical sample set A probabilistic proxy model between process parameters and molding density is constructed using Gaussian process regression algorithm, which is used to generate the expected value and prediction variance of molding density corresponding to any set value of process parameters.

[0039] The parameter space for setting process parameter settings is used. For any point in the parameter space, the expected value and variance of the molding density prediction for that point are generated using the probabilistic proxy model.

[0040] Construct a collection function, using the expected value and variance of the molding density prediction at that point as inputs to the collection function, and define the output of the collection function as the recommendation score for that point;

[0041] The process parameter setting value corresponding to the point with the highest recommended score in the parameter space is used as the process parameter setting value for the current pressing cycle.

[0042] The beneficial effects of adopting the above technical solution are as follows:

[0043] This invention innovatively constructs a dual-channel data acquisition architecture, transmitting pressure and displacement signals to the PLC and temperature signals directly to the host computer via serial port. The host computer is equipped with a quality prediction module and a Bayesian optimization module, forming a closed-loop control circuit. This system can use the PINN model to evaluate molding quality in real time and automatically calculate the optimal pressing pressure and holding time using a Bayesian algorithm. It directly corrects PLC parameters via Ethernet, realizing a shift from experience-based trial and error to intelligent optimization, significantly improving the consistency and production efficiency of granular material molding. Specific analysis is as follows:

[0044] (1) This system integrates the Bayesian optimization algorithm into the press control system. It can use the PINN model as a proxy model to perform low-cost virtual optimization and only sends the optimal result to the PLC for execution, realizing the efficient mode of "offline calculation and online control" and significantly reducing the trial and error cost.

[0045] (2) An innovative dual-channel acquisition architecture of "PLC + serial port" was designed, which not only ensures the real-time control stability of high-frequency mechanical signals such as pressure and displacement, but also realizes high-precision independent acquisition of thermodynamic signals such as temperature, providing rich data input for physical information neural networks.

[0046] (3) The lower-level PLC continues to maintain high-reliability traditional logic control (PID or logic switch), while the complex algorithm calculation is handled by the upper-level computer, and the two are decoupled through Ethernet. This architecture not only ensures the safety of the equipment, but also endows the equipment with self-evolving intelligent characteristics, making it easy to upgrade old equipment. Attached Figure Description

[0047] Figure 1This is a structural diagram of the granular material pressing process parameter setting system based on Bayesian optimization in this embodiment;

[0048] Figure 2 This is a schematic diagram of the internal logic and data flow of the higher-level decision optimization subsystem in this embodiment;

[0049] Figure 3 This is a flowchart of the method for setting process parameters for granular material pressing based on Bayesian optimization in this embodiment;

[0050] Among them: 1-Hydraulic master cylinder, 2-Modible crossbeam, 3-Guide mechanism, 4-Mold worktable, 5-Pressure sensor, 6-Displacement sensor, 7-Hydraulic system, 8-Power distribution cabinet, 9-Host PC terminal. Detailed Implementation

[0051] To facilitate understanding of this application, specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. The following embodiments are illustrative of the invention but are not intended to limit its scope. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0052] Example 1:

[0053] This embodiment presents a Bayesian optimization-based system for setting process parameters for granular material pressing, such as... Figure 1 As shown, the system includes: a suppression execution subsystem, a multi-source sensing subsystem, a lower-level control subsystem, and a higher-level decision optimization subsystem.

[0054] The compression execution subsystem is used to perform physical compression of the granules and to collect raw signals of multidimensional state data in real time during the compression process.

[0055] The multidimensional state data includes: real-time forming pressure, real-time vertical displacement of the moving crossbeam, and mold temperature during the pressing process.

[0056] The pressing execution subsystem includes: a frame, a hydraulic master cylinder 1, a movable crossbeam 2, a guide mechanism 3, a mold worktable 4, a pressure sensor 5, a displacement sensor 6, a temperature sensor, and a hydraulic system 7.

[0057] The hydraulic master cylinder 1 is fixedly installed on the top of the frame; the guide mechanism 3 is vertically arranged on the frame; the movable crossbeam 2 is slidably installed on the guide mechanism 3 and rigidly connected to the piston rod of the hydraulic master cylinder 1; the hydraulic system 7 is connected to the hydraulic master cylinder 1 and is used to receive hydraulic execution commands and drive the hydraulic master cylinder 1 so that the movable crossbeam 2 moves vertically reciprocating along the guide mechanism 3 under the drive of the hydraulic master cylinder 1.

[0058] A pressure transmitter is also installed on the main oil inlet of the hydraulic master cylinder 1 or the output end of the hydraulic system 7 to monitor the system oil pressure of the hydraulic system in real time.

[0059] like Figure 1 As shown, the hardware architecture design of this embodiment adopts a four-column hydraulic press. The hydraulic master cylinder is installed on the upper crossbeam of the frame, and the piston rod of the hydraulic master cylinder 1 is kept extending vertically downward to drive the movable crossbeam to move up and down, pressing the granular material (such as pharmaceutical powder) in the forming mold. The guiding mechanism can adopt guide columns or precision guide rails, which are symmetrically installed on the frame columns, and the movable crossbeam is slidably installed on the guiding mechanism to achieve vertical reciprocating motion.

[0060] The mold workbench 4 is fixedly installed below the machine frame and is positioned opposite to the movable crossbeam 2. It is used to install the lower mold of the forming mold, and the bottom of the movable crossbeam 2 is used to install the upper mold of the forming mold.

[0061] The pressure sensor 5 is installed between the movable crossbeam 2 and the piston rod of the hydraulic master cylinder 1 or on the force transmission path of the molding die, and is used to collect the raw signal of the real-time molding pressure.

[0062] The displacement sensor 6 is mounted on the frame or movable crossbeam 2 and is used to collect the raw signal of the real-time vertical displacement of the movable crossbeam 2.

[0063] In this embodiment, the displacement sensor 6 is disposed on the side of the frame or movable crossbeam.

[0064] The temperature sensor is installed inside or on the surface of the molding die to collect the raw temperature signal of the die during the pressing process.

[0065] The multi-source sensing subsystem is communicatively connected to the suppression execution subsystem and is used to receive the raw signals of multi-dimensional state data and preprocess them to obtain standardized multi-dimensional state data.

[0066] The multi-source sensing subsystem includes: a PLC acquisition module, a temperature acquisition module, and a data processing module.

[0067] The PLC acquisition module is used to receive the raw signal of the real-time molding pressure and the raw signal of the real-time vertical displacement.

[0068] The temperature acquisition module is used to receive the raw signal of the mold temperature.

[0069] In this embodiment, the multi-source sensing subsystem adopts a dual-path heterogeneous data acquisition architecture, including an independent PLC acquisition module and a temperature acquisition module. For the PLC acquisition module, a pressure transmitter (for detecting system oil pressure) is installed in the hydraulic main cylinder oil circuit, and a pressure sensor (for detecting actual pressing force) is installed below the forming mold. The acquired signals are connected to the analog module within the PLC acquisition module via an analog input port. A magnetostrictive displacement sensor or a pull-rope displacement sensor is installed on the side of the movable crossbeam, and its signals are also connected to the PLC. That is, the input terminals of the multi-source sensing subsystem are electrically (or communicatively) connected to the pressure sensor, displacement sensor, and temperature acquisition group, respectively, thereby acquiring the real-time forming pressure and real-time vertical displacement through the PLC acquisition module configured within the multi-source sensing subsystem. For the temperature acquisition module, in order to accurately capture the thermal effect during the pressing process (which is crucial for the prediction of the PINN model), thermocouples are pre-embedded in the mold cavity wall.

[0070] The data processing module is used to synchronize the original signals of real-time molding pressure, real-time vertical displacement, and mold temperature in time, and send the synchronized real-time molding pressure signal and real-time vertical displacement signal to the lower-level control subsystem, and send the synchronized mold temperature signal to the upper-level decision optimization subsystem.

[0071] In this embodiment, the synchronized real-time molding pressure signal and real-time vertical displacement signal are connected to the analog input port of the lower-level control subsystem. Since temperature changes are relatively slow and high resolution is required, the synchronized mold temperature signal is connected directly to the host computer's USB port (i.e., COM port) via an RS485 to USB converter, without occupying PLC resources, thus forming a second data acquisition path independent of the lower-level control subsystem.

[0072] The lower-level control subsystem is electrically connected to the pressing execution subsystem and the multi-source sensing subsystem, respectively. It is used to set instructions according to the received process parameters and to perform real-time closed-loop control of the physical pressing action executed by the pressing execution subsystem based on the standardized multi-dimensional state data.

[0073] The lower-level control subsystem includes a programmable logic controller (PLC) for comparing the process parameter setting values ​​in the process parameter setting instruction with the real-time forming pressure signal and the real-time vertical displacement signal, generating a pressure closed-loop control instruction based on the comparison result, and sending the pressure closed-loop control instruction to the hydraulic system to control the opening and closing of the hydraulic valve group in the hydraulic system, thereby driving the hydraulic master cylinder to achieve real-time closed-loop control of the pressing force.

[0074] The process parameter settings include: target pressure setting. Pressure holding time setting value Pressurization speed setting value and molding quality setting value .

[0075] The communication interface of the programmable logic controller (PLC) is an Ethernet interface, and the PLC also has a holding register inside for storing process parameter settings.

[0076] In this embodiment, the core controller of the lower-level control subsystem is a Siemens S7-1200 series PLC. The PLC controls the opening and closing of the hydraulic valve group through the I / O module to realize the action cycle of "fast down-slow pressure-pressure holding-pressure release-return". The PLC has an internal holding register to store the process parameter settings.

[0077] The upper-level decision optimization subsystem is communicatively connected to the multi-source sensing subsystem and the lower-level control subsystem, respectively, for receiving and storing standardized multi-dimensional state data; based on the received multi-dimensional state data, it generates a predicted forming density using a pre-constructed granular forming density prediction model; and using the predicted forming density as the optimization target, it calculates the granular forming process parameters for the next pressing cycle using a Bayesian optimization algorithm, and generates a process parameter setting command to be sent to the lower-level control subsystem.

[0078] The upper-level decision optimization subsystem and the lower-level control subsystem exchange data bidirectionally via the TCP / IP protocol.

[0079] The higher-level decision optimization subsystem includes a quality prediction module and a Bayesian optimization module.

[0080] The quality prediction module is used to generate a predicted forming density based on the received multidimensional state data and the current process parameter settings, using a granular forming density prediction model based on physical information neural networks, and then send the predicted forming density to the Bayesian optimization module.

[0081] The Bayesian optimization module is used to generate the process parameter setting values ​​for the next pressing cycle based on the preset target molding density, the predicted molding density, and the mold temperature signal, with the process parameter setting values ​​as the optimization target, using the Bayesian optimization algorithm, and then generating the process parameter setting instructions for the next pressing cycle and writing them into the holding register.

[0082] The optimization variable of the Bayesian optimization module is the process parameter setting value.

[0083] In this embodiment, as Figure 2As shown, for the upper-level decision-making and communication design of the upper-level decision-making optimization subsystem, the upper-level decision-making optimization subsystem adopts a human-machine interface industrial control computer, and the upper-level PC 9 is directly connected to the lower-level PLC via a network cable, communicating based on the S7 protocol or Modbus-TCP protocol. The upper-level decision-making optimization subsystem PC runs intelligent control software developed based on C#, which integrates a quality prediction module and a Bayesian optimization module. The quality prediction module is used to predict the forming density of the product based on the current process parameters and sensor data; the Bayesian optimization module is configured with a preset target density value, and is used to calculate the recommended process parameters for the next batch based on the feedback from the quality prediction module and the Bayesian optimization algorithm, and write the recommended process parameters into the setpoint register of the lower-level control subsystem through the communication interface.

[0084] The granular material pressing process parameter adaptive setting system also includes a power distribution cabinet 8, which provides power and circuit protection for the pressing execution subsystem, the multi-source sensing subsystem, and the lower-level control subsystem.

[0085] Example 2:

[0086] This embodiment presents a method for setting process parameters for granular material compression based on Bayesian optimization. This method is implemented using the Bayesian optimization-based granular material compression process parameter setting system described in Embodiment 1. This system employs a "batch-to-batch adaptive setting" working mode, such as... Figure 3 As shown, the method includes the following steps:

[0087] For the compression process of any granular material, set the initial process parameter settings for the first compression cycle and execute it, and collect multidimensional state data during the compression cycle.

[0088] In this embodiment, after a compression cycle is completed, the upper-level decision optimization subsystem reads the actual pressure curve and displacement curve obtained from the lower-level control subsystem through Ethernet, and reads the temperature curve through the serial port.

[0089] Starting from the second suppression cycle, for each current suppression cycle, perform the following procedure:

[0090] Based on the multidimensional state data from the previous pressing cycle, a predicted forming density is generated using a granular material forming density prediction model based on a physical information neural network.

[0091] In this embodiment, the upper-level decision optimization subsystem inputs the collected multidimensional data into the internally encapsulated quality prediction module. This module utilizes an embedded granular material forming density prediction model built based on physical laws (monotonicity constraints) and the deep neural network PINN to calculate the predicted forming density of the pressed product. This step is equivalent to completing a virtual measurement without damaging the product or using a densitometer. Specifically, the granular forming density prediction model based on physical information neural networks is implemented using a lightweight fully connected neural network MLP, including an input layer, hidden layers, and an output layer; and based on a configurable composite physical loss function framework, a composite physical loss function consisting of a data fitting term and an optional physical constraint term is constructed.

[0092] Based on the process parameter settings and predicted molding density from the previous pressing cycle, a Bayesian optimization algorithm is used to generate the process parameter settings for the current pressing cycle.

[0093] In this embodiment, the predicted molding density is read using a Bayesian optimization module. and compare it with the preset target molding density. By comparing the results, we can construct the objective function, which is to minimize the molding density error. Based on Gaussian Process Regression (GP), the Bayesian optimization algorithm updates the surrogate model using historical data and searches the parameter space using an acquisition function to calculate the recommended combination of process parameters most likely to achieve the target density for the next batch. In this embodiment, the main optimization variable is the pressing pressure setpoint. Pressure holding time setting value Molding quality setting value It should be noted that, in embodiments with servo control capabilities, the pressurization speed... It can also be used as an optimization variable.

[0094] The specific content of generating the process parameter settings for the current pressing cycle using a Bayesian optimization algorithm based on the process parameter settings and predicted molding density from the previous pressing cycle is as follows:

[0095] Based on the predicted molding density and the preset target molding density, with the objective function being to minimize the error between the predicted molding density and the target molding density, a Bayesian algorithm is used to iteratively optimize the process parameter settings in the previous pressing cycle, and the following process is executed in each iteration:

[0096] New sample points are generated by combining the process parameter settings during the current pressing cycle with the predicted molding density. And add to the historical sample set .

[0097] In this embodiment, the process parameter settings and predicted molding density corresponding to each pressing cycle are treated as a sample point and saved to the historical sample set. In this process, real-time accumulation of operating condition data is achieved.

[0098] Based on the updated historical sample set A probabilistic proxy model between process parameters and molding density is constructed using a Gaussian process regression algorithm. This model is used to generate the expected value and variance of the molding density prediction for any set value of process parameters.

[0099] In this embodiment, based on the updated sample set This study utilizes the Gaussian Process Regression (GP) algorithm to model the nonlinear mapping relationship between process parameters and molding density, constructing a probabilistic surrogate model between the two. This model can output the expected value (mean, representing the prediction result) and the prediction variance (variance, representing the degree of uncertainty) of the molding density under any given parameters, thereby quantifying the probability distribution characteristics within the parameter space.

[0100] The parameter space for setting process parameter values ​​is used. For any point in the parameter space, the expected value and variance of the molding density prediction for that point are generated using the probabilistic surrogate model.

[0101] Construct a data acquisition function, using the expected value and variance of the molding density prediction at that point as inputs, and define the output of the data acquisition function as the recommended score for that point.

[0102] The acquisition function employs a confidence upper bound strategy, combining the expected value and variance of the shape density prediction through weighted summation to obtain the recommendation score.

[0103] In this embodiment, an Upper Confidence Bound (UCB) strategy is employed. This strategy uses a weighted summation method, combining "predicted expected value" for exploration (a refined search within known high-optimal regions) and "predicted variance" for exploration (attempted search within unknown, high-uncertainty regions). The parameter combination with the highest score is selected as the process parameter setpoint for the next experiment. Alternatively, this embodiment can also employ an Expected Improvement (EI) strategy to construct the acquisition function.

[0104] The process parameter setting value corresponding to the point with the highest recommended score in the parameter space is used as the process parameter setting value for the current pressing cycle.

[0105] In this embodiment, within a defined parameter space (e.g., pressure 10 MPa-50 MPa, time 5 s-30 s), the system searches for process parameter settings that maximize the acquired function value. This set of process parameter settings represents the optimal process parameters that the system believes are most likely to bring the next batch close to the target molding density.

[0106] Based on the process parameter settings during the current pressing cycle, execute the current pressing cycle and collect multi-dimensional status data during the current pressing cycle.

[0107] In this embodiment, the upper-level decision optimization subsystem directly writes the new process parameter setpoints calculated by the Bayesian module into the designated DB block address of the lower-level control subsystem PLC via Ethernet (e.g., DB1.DBD236 corresponds to pressure, and DB1.DBW34 corresponds to time). When the lower-level control subsystem PLC executes the compression task of the next pharmacy column, it directly calls the new process parameter setpoints in the register to control the compression execution subsystem.

[0108] Repeat the pressing cycle several times until the entire granular material pressing process is completed, thereby achieving adaptive optimization of the pressing process parameters.

[0109] Through the above process, this embodiment realizes the automatic iteration and self-evolution of the pressing process parameters, and can make the product density quickly converge to the target value without manual intervention, effectively overcoming the problems of poor accuracy and low efficiency of manual parameter adjustment in traditional open-loop control.

[0110] Finally, 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 them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the present invention.

Claims

1. A system for setting process parameters for granular material pressing based on Bayesian optimization, characterized in that, The system includes: The compression execution subsystem is used to perform physical compression actions on the granules and to collect raw signals of multi-dimensional state data in real time during the compression process. The multi-source sensing subsystem is communicatively connected to the suppression execution subsystem. It is used to receive the raw signals of multi-dimensional state data and preprocess them to obtain standardized multi-dimensional state data. The lower-level control subsystem is electrically connected to the pressing execution subsystem and the multi-source sensing subsystem, respectively. It is used to set instructions according to the received process parameters and to perform real-time closed-loop control of the physical pressing action executed by the pressing execution subsystem based on the standardized multi-dimensional state data. The upper-level decision optimization subsystem is communicatively connected to the multi-source sensing subsystem and the lower-level control subsystem, respectively, for receiving and storing standardized multi-dimensional state data; based on the received multi-dimensional state data, it generates a predicted forming density using a pre-constructed granular forming density prediction model; and using the predicted forming density as the optimization target, it calculates the granular forming process parameters for the next pressing cycle using a Bayesian optimization algorithm, and generates a process parameter setting command to be sent to the lower-level control subsystem.

2. The granular material pressing process parameter setting system based on Bayesian optimization according to claim 1, characterized in that, The multidimensional state data includes: real-time forming pressure, real-time vertical displacement of the moving crossbeam, and mold temperature during the pressing process.

3. The granular material pressing process parameter setting system based on Bayesian optimization according to claim 2, characterized in that, The pressing execution subsystem includes: a frame, a hydraulic master cylinder, a movable crossbeam, a guide mechanism, a mold worktable, a pressure sensor, a displacement sensor, a temperature sensor, and a hydraulic system; The hydraulic master cylinder is fixedly mounted on the top of the frame; the guide mechanism is vertically mounted on the frame; the movable crossbeam is slidably mounted on the guide mechanism and rigidly connected to the piston rod of the hydraulic master cylinder; the hydraulic system is connected to the hydraulic master cylinder and is used to receive hydraulic execution commands and drive the hydraulic master cylinder so that the movable crossbeam moves vertically reciprocating along the guide mechanism under the drive of the hydraulic master cylinder. The mold workbench is fixedly installed below the machine frame and is positioned opposite to the movable crossbeam. It is used to install the lower mold of the forming mold, and the bottom of the movable crossbeam is used to install the upper mold of the forming mold. The pressure sensor is installed between the movable crossbeam and the piston rod of the hydraulic master cylinder or on the force transmission path of the molding die, and is used to collect the raw signal of the real-time molding pressure. The displacement sensor is mounted on the frame or movable crossbeam and is used to collect the raw signal of the real-time vertical displacement of the movable crossbeam. The temperature sensor is installed inside or on the surface of the molding die to collect the raw temperature signal of the die during the pressing process.

4. The granular material pressing process parameter setting system based on Bayesian optimization according to claim 3, characterized in that, The multi-source sensing subsystem includes: a PLC acquisition module, a temperature acquisition module, and a data processing module; The PLC acquisition module is used to receive the original signal of the real-time molding pressure and the original signal of the real-time vertical displacement. The temperature acquisition module is used to receive the raw signal of the mold temperature; The data processing module is used to synchronize the original signals of real-time molding pressure, real-time vertical displacement, and mold temperature in time, and send the synchronized real-time molding pressure signal and real-time vertical displacement signal to the lower-level control subsystem, and send the synchronized mold temperature signal to the upper-level decision optimization subsystem.

5. The granular material pressing process parameter setting system based on Bayesian optimization according to claim 4, characterized in that, The lower-level control subsystem includes: a programmable logic controller (PLC); The lower-level control subsystem includes a programmable logic controller (PLC), which compares the process parameter setting value in the received process parameter setting instruction with the real-time forming pressure signal and the real-time vertical displacement signal, generates a pressure closed-loop control instruction based on the comparison result, and sends the pressure closed-loop control instruction to the hydraulic system to control the opening and closing of the hydraulic valve group in the hydraulic system, thereby driving the hydraulic master cylinder to realize real-time closed-loop control of the pressing force. The process parameter settings include: target pressure setting. Pressure holding time setting value Pressurization speed setting value and molding quality setting value .

6. The granular material pressing process parameter setting system based on Bayesian optimization according to claim 5, characterized in that, The communication interface of the programmable logic controller (PLC) is an Ethernet interface, and the PLC also has a holding register inside for storing process parameter settings.

7. The granular material pressing process parameter setting system based on Bayesian optimization according to claim 6, characterized in that, The higher-level decision optimization subsystem includes: a quality prediction module and a Bayesian optimization module; The quality prediction module is used to generate a predicted forming density based on the received multidimensional state data and the current process parameter settings, using a granular forming density prediction model based on physical information neural network, and send it to the Bayesian optimization module. The Bayesian optimization module is used to generate the process parameter setting values ​​for the next pressing cycle based on the preset target molding density, the predicted molding density, and the mold temperature signal, with the process parameter setting values ​​as the optimization target, using the Bayesian optimization algorithm, and then generating the process parameter setting instructions for the next pressing cycle and writing them into the holding register.

8. The granular material pressing process parameter setting system based on Bayesian optimization according to claim 1, characterized in that, The granular material pressing process parameter adaptive setting system also includes a power distribution cabinet, which provides power and circuit protection for the pressing execution subsystem, the multi-source sensing subsystem, and the lower-level control subsystem.

9. A method for setting process parameters for granular material pressing based on Bayesian optimization, implemented using the granular material pressing process parameter setting system based on Bayesian optimization as described in any one of claims 1-8, characterized in that... This method includes the following steps: For the compression process of any granular material, set the initial process parameter settings for the first compression cycle and execute it, and collect multidimensional state data during the compression cycle. Starting from the second suppression cycle, for each current suppression cycle, perform the following procedure: Based on the multidimensional state data from the previous pressing cycle, a predicted forming density is generated using a granular forming density prediction model based on a physical information neural network. Based on the process parameter settings and predicted molding density in the previous pressing cycle, the process parameter settings for the current pressing cycle are generated using a Bayesian optimization algorithm. Based on the process parameter settings during the current pressing cycle, execute the current pressing cycle and collect multi-dimensional status data during the current pressing cycle. Repeat the pressing cycle several times until the entire granular material pressing process is completed, thereby achieving adaptive optimization of the pressing process parameters.

10. The method for setting process parameters for granular material pressing based on Bayesian optimization as described in claim 9, characterized in that, The specific content of generating the process parameter settings for the current pressing cycle using a Bayesian optimization algorithm based on the process parameter settings and predicted molding density from the previous pressing cycle is as follows: Based on the predicted molding density and the preset target molding density, with the objective function being to minimize the error between the predicted molding density and the target molding density, a Bayesian algorithm is used to iteratively optimize the process parameter settings in the previous pressing cycle, and the following process is executed in each iteration: New sample points are generated by combining the process parameter settings during the current pressing cycle with the predicted molding density. And add to the historical sample set ;in This indicates the predicted molding density; Based on the updated historical sample set A probabilistic proxy model between process parameters and molding density is constructed using Gaussian process regression algorithm, which is used to generate the expected value and prediction variance of molding density corresponding to any set value of process parameters. The parameter space for setting process parameter settings is used. For any point in the parameter space, the expected value and variance of the molding density prediction for that point are generated using the probabilistic proxy model. Construct a data acquisition function, using the expected value and variance of the molding density prediction at that point as inputs to the data acquisition function, and define the output of the data acquisition function as the recommended score for that point; The process parameter setting value corresponding to the point with the highest recommended score in the parameter space is used as the process parameter setting value in the current pressing cycle.

Images