Spiral spike process parameter optimization system and method based on image deep learning

Abstract

The present invention discloses a spiral spike process parameter optimization system and method based on image deep learning, which relates to the field of intelligent control technology. The system collects spiral spike target information, real-time image data and a process parameter set, and uses image processing technology to analyze various real-time quality parameters of the spiral spike in the real-time image data; based on the real-time quality parameters and the spiral spike target information, a quality evaluation function is constructed to analyze the quality of the extruded product; based on the quality of the extruded product and the process parameter set, a process adjustment strategy is generated using a deep reinforcement learning model; based on the process adjustment strategy, the process parameters are dynamically adjusted using an automated control system; the automated control system receives image processing feedback information in real time, and iteratively trains the deep reinforcement learning model based on the image processing feedback information. The present invention can adaptively optimize process settings and improve the flexibility of the production process and product consistency.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a system and method for optimizing process parameters of a spiral spike based on image deep learning. Background Art

[0002] The complexity of the multi-layer composite extrusion process of bimetallic composite spiral spikes lies in how to accurately control the process parameters during the extrusion process to ensure good bonding of each layer of metal and consistency of the finished product. Traditional process control methods often rely on experienced operators and static process parameter settings, and it is difficult to respond in real time to changes in material properties and environmental factors that occur during the production process. This inflexible control method can easily lead to unstable product quality, and when faced with changes in materials from different batches, it is impossible to quickly adjust to optimize product quality. In addition, traditional methods lack the ability to deeply analyze real-time data, resulting in process adjustments relying more on workers' experience and trial and error, which is time-consuming and may increase production costs. The shortcomings of existing technologies in terms of automation level and intelligent decision-making support urgently require a more intelligent solution to improve process efficiency and product quality.

[0003] The Chinese patent application with publication number CN116500975A discloses a method, device, CNC machine tool and readable storage medium for process control of a numerical control system. When processing a workpiece, a workpiece image of at least one workpiece is acquired in real time, and the geometric quality information corresponding to the workpiece image is determined, wherein the geometric quality information includes the geometric features and quality information of the workpiece surface; all process parameters for processing the workpiece are determined based on the geometric quality information, and the adjustable parameters in all the process parameters are divided into sets to obtain multiple subsets; each subset is optimized in parallel according to a preset optimization algorithm model to obtain an optimized process parameter combination, and the workpiece is processed according to the optimized process parameter combination. However, the optimization model used in the invention is isolated in time, and cannot adaptively guide the quality of the workpiece at the next moment based on the quality of the workpiece at the previous moment.

[0004] To this end, the present invention proposes a spiral spike process parameter optimization system and method based on image deep learning. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a spiral spike process parameter optimization system and method based on image deep learning, which can adaptively optimize process settings and improve the flexibility of the production process and product consistency.

[0006] To achieve the above objectives, a spiral spike process parameter optimization method based on image deep learning is proposed, which includes the following steps:

[0007] Step 1: Collect spiral spike target information, real-time image data and process parameter set;

[0008] Step 2: Use image processing technology to analyze various real-time quality parameters of the spiral spikes in the real-time image data; construct a quality evaluation function based on the real-time quality parameters and the spiral spike target information;

[0009] Step 3: Based on the real-time quality parameters and spiral spike target information, use the quality evaluation function to analyze the quality of the extruded product;

[0010] Step 4: Generate a process adjustment strategy using a deep reinforcement learning model based on the extrusion product quality and process parameter set;

[0011] Step 5: Based on the process adjustment strategy, the process parameters are dynamically adjusted using the automated control system. The automated control system receives the image processing feedback information in real time and iteratively trains the deep reinforcement learning model based on the image processing feedback information.

[0012] The spiral spike target information is collected in the following manner:

[0013] Collecting the real-time expected values ​​of various quality parameters of the spiral spikes expected to be produced in each unit time in the production plan, and collecting the material properties and surface treatment information of the raw materials to be extruded in the production plan; the real-time expected values ​​of various quality parameters, the material properties and surface treatment information of the raw materials constitute the target information of the spiral spikes;

[0014] The real-time image data is collected in the following manner:

[0015] During the multi-layer composite extrusion production process of the bimetallic composite spiral spike, a high-precision camera installed at the outlet of the extruder to monitor the extruded product in real time is used to collect images of the extruded spiral spikes per unit time as real-time image data;

[0016] The process parameter set is collected in the following manner:

[0017] During the multi-layer composite extrusion production process of the bimetal composite spiral spike, various extrusion parameters of the extruder are collected in real time through the control background of the extruder;

[0018] The method of using image processing technology to analyze various real-time quality parameters of the spiral spike in real-time image data is as follows:

[0019] Performing image preprocessing on the real-time image of the spiral spike in the real-time image data;

[0020] Perform edge detection on the real-time image of the spiral spike to obtain the sideline edge in the real-time image, and then find the edge of the spiral spike by detecting the brightness change position in the real-time image; the brightness change position is determined by setting a brightness change threshold;

[0021] Extract the edge lines of the detected spiral spikes and identify the contours of the spiral spikes through contour detection technology;

[0022] Perform frequency domain analysis on the extracted spiral spike profile to identify the spiral periodicity characteristics, and calculate the real-time pitch of the spiral spike, that is, the period of spiral repetition, by analyzing the peak position in the spectrum;

[0023] The detected contour of the spiral spike is fitted into a circle or an ellipse by using the shape fitting algorithm of the least square method, and the maximum diameter is determined from the fitting result as the diameter measurement value of the spiral spike, and the diameter measurement value is the real-time diameter;

[0024] The minimum circumscribed rectangle of the spiral spike is used to obtain the length of the spiral spike, and the image pixel length is converted into the actual physical length, which is the real-time length, and the image calibration data is used for calibration;

[0025] The real-time pitch, real-time diameter and real-time length of the spiral spike constitute real-time quality parameters;

[0026] The method of constructing the quality evaluation function based on the real-time quality parameters and the spiral spike target information is as follows:

[0027] For any unit of time during the extrusion process:

[0028] Calculate the difference between the real-time value of each quality parameter in the real-time quality parameter and the real-time expected value of the corresponding quality parameter in the spiral spike target information at the corresponding time as the quality parameter difference of the quality parameter;

[0029] Based on the differences of various quality parameters, a quality assessment function is constructed;

[0030] The method of analyzing the quality of extruded products based on real-time quality parameters and spiral spike target information and using a quality evaluation function is as follows:

[0031] Collecting real-time extrusion images of the spiral spike to be monitored during the extrusion process;

[0032] Then, the real-time values ​​of various quality parameters corresponding to the real-time extrusion image are analyzed by image processing technology to form the real-time quality parameters of the spiral spike to be monitored;

[0033] Then, according to the real-time quality parameters of the spiral spike to be monitored and the target information of the spiral spike, the difference of each quality parameter at each unit time is calculated respectively;

[0034] Substitute the difference of each quality parameter into the quality evaluation function, and calculate the quality of the extruded product by the quality evaluation function;

[0035] The method of using the deep reinforcement learning model to generate a process adjustment strategy is:

[0036] Set the deep reinforcement learning model to the Actor-Critic network model;

[0037] Defining a learning environment for a deep reinforcement learning model; the learning environment includes a state space, an action space, and a reward function;

[0038] The state space is defined as a set of current process parameters and product quality indicators;

[0039] The state space includes real-time readings of temperature, pressure, speed, and product quality expressed using a comprehensive quality score calculated by a quality assessment function;

[0040] Defining the action space as a range of adjustable process parameters; the action space includes increasing or decreasing values ​​of temperature, pressure and speed;

[0041] The reward function is used to encourage the deep reinforcement learning model to optimize product quality, and the reward function is constructed based on the implemented actions and the comprehensive quality score;

[0042] Initialize the Actor-Critic network model;

[0043] The Actor-Critic network model is preliminarily trained using the action space, selected actions, and rewards recorded in the past extrusion production process to collect state, action, and reward triplets under different production conditions;

[0044] Initialize the weights of the Actor network and the Critic network;

[0045] In each unit of time, the Actor network selects an action based on the current state, and after executing the action, obtains a new state and reward; the Critic network evaluates the value of the new state and updates the value function;

[0046] Use gradient ascent to update the Actor network to maximize long-term rewards;

[0047] Use the temporal difference error to update the Critic network to improve the accuracy of state assessment;

[0048] Use the Actor network in the trained deep reinforcement learning model to output the action selected from the action space per unit time of the spiral spike to be monitored as a process parameter adjustment strategy;

[0049] The dynamically adjusting the process parameters using the automated control system includes adjusting the corresponding process parameters according to the adjustment values ​​of the process parameters in the process adjustment strategy;

[0050] The image processing feedback information is the difference between various quality parameters after adjustment by the process adjustment strategy and before adjustment.

[0051] This application proposes a spiral spike process parameter optimization system based on image deep learning, including a process data collection module, a product quality assessment module, a strategy generation module and a control feedback module; wherein each module is electrically connected;

[0052] A process data collection module collects spiral spike target information, real-time image data and a set of process parameters, and sends the spiral spike target information, real-time image data and a set of process parameters to a product quality assessment module;

[0053] The product quality assessment module uses image processing technology to analyze various real-time quality parameters of the spiral spikes in the real-time image data; based on the real-time quality parameters and the spiral spike target information, a quality assessment function is constructed, and based on the real-time quality parameters and the spiral spike target information, the quality of the extruded product is analyzed using the quality assessment function, and the quality of the extruded product is sent to the strategy generation module;

[0054] The strategy generation module sends the extrusion product quality to the automation control system, generates a process adjustment strategy based on the extrusion product quality and the process parameter set using a deep reinforcement learning model, and sends the process adjustment strategy to the control feedback module;

[0055] The control feedback module uses an automated control system to dynamically adjust process parameters based on the process adjustment strategy. The automated control system receives image processing feedback information in real time and iteratively trains the deep reinforcement learning model based on the image processing feedback information.

[0056] The present application proposes an electronic device, comprising: a processor and a memory, wherein the memory stores a computer program that can be called by the processor;

[0057] The processor executes the above-mentioned method for optimizing process parameters of spiral spikes based on image deep learning by calling the computer program stored in the memory.

[0058] The present application proposes a computer-readable storage medium having a rewritable computer program stored thereon;

[0059] When the computer program is run on a computer device, the computer device executes the above-mentioned method for optimizing process parameters of spiral spikes based on image deep learning.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] The present invention collects spiral spike target information, real-time image data and a set of process parameters, and uses image processing technology to analyze various real-time quality parameters of the spiral spike in the real-time image data; based on the real-time quality parameters and the spiral spike target information, a quality evaluation function is constructed; based on the real-time quality parameters and the spiral spike target information, the quality of the extruded product is analyzed using the quality evaluation function; based on the quality of the extruded product and the set of process parameters, a process adjustment strategy is generated using a deep reinforcement learning model; based on the process adjustment strategy, the process parameters are dynamically adjusted using an automated control system; the automated control system receives image processing feedback information in real time, and iteratively trains the deep reinforcement learning model based on the image processing feedback information; through a high-precision image processing system, key quality parameters of the extruded product, such as pitch, diameter and length, are monitored in real time, and these data are compared with the target parameters, and then based on the product quality score and the current process parameters, the deep reinforcement learning model dynamically adjusts the process parameters to minimize the quality deviation. Through continuous online learning and feedback adjustment, the process settings can be adaptively optimized to improve the flexibility of the production process and the consistency of the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a flow chart of a method for optimizing process parameters of a spiral spike based on image deep learning in Example 1 of the present invention;

[0063] Figure 2 This is a module connection relationship diagram of the spiral spike process parameter optimization system based on image deep learning in Example 2 of the present invention. DETAILED DESCRIPTION

[0064] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] Example 1

[0066] like Figure 1 As shown, the method for optimizing process parameters of spiral spikes based on image deep learning includes the following steps:

[0067] Step 1: Collect spiral spike target information, real-time image data and process parameter set;

[0068] Step 2: Use image processing technology to analyze various real-time quality parameters of the spiral spikes in the real-time image data; construct a quality evaluation function based on the real-time quality parameters and the spiral spike target information;

[0069] Step 3: Based on the real-time quality parameters and spiral spike target information, use the quality evaluation function to analyze the quality of the extruded product;

[0070] Step 4: Generate a process adjustment strategy using a deep reinforcement learning model based on the extrusion product quality and process parameter set;

[0071] Step 5: Based on the process adjustment strategy, the process parameters are dynamically adjusted using the automated control system. The automated control system receives the image processing feedback information in real time and iteratively trains the deep reinforcement learning model based on the image processing feedback information.

[0072] The spiral spike target information is collected in the following manner:

[0073] Collecting the real-time expected values ​​of various quality parameters of the spiral spikes expected to be produced in each unit time in the production plan, and collecting the material properties and surface treatment information of the raw materials to be extruded in the production plan; the real-time expected values ​​of various quality parameters, the material properties and surface treatment information of the raw materials constitute the target information of the spiral spikes;

[0074] Specifically, the quality parameters include but are not limited to the outer diameter and inner diameter of the spiral spike, the pitch representing the axial length of each spiral turn, and the total length of the screw;

[0075] Furthermore, in each unit time, the real-time expected value of each quality parameter can be obtained by screening out several standard extrusion processes in the historical extrusion process, and recording the real-time value of each quality parameter in each unit time of each standard extrusion process; and then the average value of each quality parameter in each unit time of all standard extrusion processes is counted as the real-time expected value; the standard extrusion process can be an extrusion process manually screened out so that the quality parameters of the final extruded product reach the expected target; it can be understood that by setting the expected target for each unit time in the extrusion process, it is possible to understand in real time whether the current extrusion progress meets the expectations during the actual extrusion, so that further corresponding control measures can be taken;

[0076] The material properties include, but are not limited to, material type indicating the type of metal used in the inner and outer layers, including mechanical properties such as yield strength, tensile strength, and ductility;

[0077] The surface treatment information refers to the surface treatment type and thickness such as galvanizing and oxidation treatment;

[0078] Furthermore, the real-time image data is collected in the following manner:

[0079] During the multi-layer composite extrusion production process of the bimetallic composite spiral spike, a high-precision camera installed at the exit of the extruder to monitor the extruded product in real time is used to collect images of the extruded spiral spikes per unit time as real-time image data;

[0080] Furthermore, the process parameter set is collected in the following manner:

[0081] During the multi-layer composite extrusion production process of the bimetal composite spiral spike, various extrusion parameters of the extruder are collected in real time through the control background of the extruder;

[0082] Specifically, the extrusion parameters include but are not limited to temperature, pressure and extrusion speed;

[0083] Furthermore, the method of using image processing technology to analyze various real-time quality parameters of the spiral spike in the real-time image data is as follows:

[0084] Performing image preprocessing on the real-time image of the spiral spike in the real-time image data;

[0085] Specifically, the image preprocessing includes but is not limited to converting the acquired image into a grayscale image to simplify subsequent processing steps and improve contrast, and applying filtering techniques (such as Gaussian filtering) to remove noise in the image and enhance edge clarity;

[0086] Perform edge detection on the real-time image of the spiral spike to obtain the sideline edge in the real-time image, and then find the edge of the spiral spike by detecting the position where the brightness changes significantly in the real-time image; the brightness change significantly can be determined by setting a brightness change threshold, for example, a brightness change greater than the brightness change threshold is a brightness change significantly;

[0087] Extract the edge lines of the detected spiral spikes and identify the contours of the spiral spikes through contour detection technology;

[0088] Performing frequency domain analysis on the extracted profile of the spiral spike to identify the spiral periodicity feature, and calculating the real-time pitch of the spiral spike, that is, the period of spiral repetition, by analyzing the peak position in the frequency spectrum; the frequency domain analysis can be achieved by using Fourier transform technology;

[0089] The detected contour of the spiral spike is fitted with a circle or an ellipse by using a shape fitting algorithm of the least square method, and a maximum diameter is determined from the fitting result as a diameter measurement value of the spiral spike, and the diameter measurement value is the real-time diameter;

[0090] The length of the spiral spike is estimated using the minimum circumscribed rectangle of the spiral spike's outline, and the image pixel length is converted into the actual physical length, which is the real-time length, and calibrated using the image calibration data;

[0091] The real-time pitch, real-time diameter and real-time length of the spiral spike constitute real-time quality parameters;

[0092] Furthermore, the quality evaluation function is constructed based on the real-time quality parameters and the spiral spike target information in the following manner:

[0093] For any unit of time during the extrusion process:

[0094] Calculate the difference between the real-time value of each quality parameter in the real-time quality parameter and the real-time expected value of the corresponding quality parameter in the spiral spike target information at the corresponding time as the quality parameter difference of the quality parameter;

[0095] Based on the differences of various quality parameters, a quality assessment function is constructed;

[0096] Specifically, the quality assessment function may be constructed as follows:

[0097] The quality parameter difference of the pitch is marked as ;

[0098] The quality parameter difference of the diameter is marked as ;

[0099] The quality parameter difference of the length is marked as ;

[0100] The quality assessment function is labeled ;

[0101] Then the quality assessment function The expression can be:

[0102] ;

[0103] Among them, w1, w2 and w3 are all preset proportional coefficients, or weights determined by analyzing historical data and experimental results using statistical methods such as regression analysis or principal component analysis;

[0104] , and These are the influence functions of pitch deviation, diameter deviation and length deviation on the quality of the finished product after extrusion of spiral spikes;

[0105] Said and It can be a linear function, a nonlinear function or a piecewise function, and the specific form depends on the actual situation of the impact of each parameter on the quality;

[0106] Preferably, the Can be set to:

[0107] ; k1 is the preset pitch deviation proportional coefficient; that is, a linear function is used to describe the effect of pitch deviation on quality, because the larger the deviation, the worse the quality; this function ensures that when the pitch deviation is zero, the score is 1, and when the deviation increases, the score decreases;

[0108] Said Can be set to:

[0109] ; k2 is the preset diameter deviation proportional coefficient; that is, a linear function is used to describe the effect of diameter deviation on quality, because the larger the deviation, the worse the quality; this function also ensures that when the diameter deviation is zero, the score is 1, and when the deviation increases, the score decreases;

[0110] Said Can be set to:

[0111] ; k3 is the preset length deviation proportional coefficient; that is, a linear function is used to describe the effect of length deviation on quality, because the larger the deviation, the worse the quality; this function also ensures that when the length deviation is zero, the score is 1, and when the deviation increases, the score decreases;

[0112] Further preferably, a surface defect scoring function may also be applied to the quality assessment function to indicate the influence of cracks and surface defects that may appear on the surface during the extrusion process of the spiral spike on the quality of the final extruded product;

[0113] Specifically, the surface defect scoring function can be expressed as: ;

[0114] The surface defect scoring function The specific formula form can be set as:

[0115] ; Wherein, k4 is a preset defect score ratio coefficient, S is a score of surface defects, which may be a weighted sum based on the number and severity of defects. Both the number and severity of defects may be evaluated by computer vision technology. For example, the length and number of cracks may be identified by a target recognition algorithm, wherein the length is used as the severity and the number is used as the number of defects. Similarly, the evaluation method of the number and severity may also be applied to scratches on the surface of spiral spikes;

[0116] Therefore, the quality evaluation function can be expressed as , the formula form of the quality assessment function can be set as:

[0117] ;

[0118] Wherein, w4 is the proportional coefficient of the preset surface defect scoring function;

[0119] Furthermore, the method of analyzing the quality of the extruded product based on the real-time quality parameters and the spiral spike target information and using the quality evaluation function is as follows:

[0120] Collecting real-time extrusion images of the spiral spike to be monitored during the extrusion process;

[0121] Then, the real-time values ​​of various quality parameters corresponding to the real-time extrusion image are analyzed by image processing technology to form the real-time quality parameters of the spiral spike to be monitored;

[0122] Then, according to the real-time quality parameters of the spiral spike to be monitored and the target information of the spiral spike, the difference of each quality parameter at each unit time is calculated respectively;

[0123] Substitute the difference of each quality parameter into the quality evaluation function, and calculate the quality of the extruded product by the quality evaluation function;

[0124] Furthermore, the method of using the deep reinforcement learning model to generate the process adjustment strategy is:

[0125] Set the deep reinforcement learning model to the Actor-Critic network model;

[0126] Defining a learning environment for a deep reinforcement learning model; the learning environment includes a state space, an action space, and a reward function;

[0127] The state space is defined as a set of current process parameters and product quality indicators;

[0128] The state space includes real-time readings of process parameters such as temperature, pressure, speed, and the like, and product quality expressed using a comprehensive quality score calculated by a quality assessment function;

[0129] The action space is defined as an adjustable range of process parameters; the action space includes increasing or decreasing the values ​​of process parameters such as temperature, pressure, speed, etc.;

[0130] The reward function is used to encourage the deep reinforcement learning model to optimize product quality, and the reward function is constructed based on the implemented actions and the comprehensive quality score;

[0131] Preferably, the reward function may be the degree to which the quality score is improved or maintained within a desired range, for example:

[0132] ;

[0133] Where R represents the reward function, It represents the change of the comprehensive quality score of the spiral spike before and after the action is implemented, i is the number of each process parameter, represents the adjustment value of the i-th process parameter in the implemented action. It can be understood that The larger the value, the better the effect of the action. The larger the value, the greater the adjustment range of process parameters. The extrusion of spiral spikes is a precision operation. Excessive changes will greatly affect the quality of spiral spikes. is the preset proportional coefficient;

[0134] It should be noted that the comprehensive quality score of the spiral spike after the action is implemented is predicted by the Critic network based on the historical squeeze data, because the Critic network is responsible for the value evaluation of the current state and the action taken. It guides the Actor network to improve the strategy by calculating the value of the current state (or state-action value). The output of the Critic is usually a value function or action value function, which is used to measure the expected cumulative reward after taking a certain action in a certain state, and this cumulative reward reflects the comprehensive quality score of the spiral spike after the action is implemented;

[0135] Initialize the Actor-Critic model;

[0136] Specifically, the Actor network is responsible for selecting actions and outputting process parameter adjustment strategies, so a deep neural network structure is used to approximate the optimal strategy; while the Critic network is used to evaluate the value of the action strategy selected by the Actor and output the state value, so a deep neural network structure is used to approximate the value function;

[0137] The Actor-Critic network model is preliminarily trained using the action space, selected actions, and rewards recorded in the past extrusion production process to collect state, action, and reward triplets under different production conditions;

[0138] Initialize the weights of the Actor network and the Critic network;

[0139] In each unit of time, the Actor network selects an action based on the current state, and after executing the action, obtains a new state and reward; the Critic network evaluates the value of the new state and updates the value function;

[0140] Use gradient ascent to update the Actor network to maximize long-term rewards;

[0141] Use the temporal difference error to update the Critic network to improve the accuracy of state assessment;

[0142] Use the Actor network in the trained deep reinforcement learning model to output the action selected from the action space per unit time of the spiral spike to be monitored as a process parameter adjustment strategy;

[0143] Further, the use of the automated control system to dynamically adjust the process parameters includes adjusting the corresponding process parameters according to the adjustment values ​​of the process parameters in the process adjustment strategy, for example, in the action space, the temperature process parameter corresponding to is 1 o C, then the extrusion temperature of the spiral spike to be monitored is increased by 1 degree Celsius;

[0144] Furthermore, the image processing feedback information refers to the difference between various quality parameters after adjustment by the process adjustment strategy and before adjustment;

[0145] The iterative training of the deep reinforcement learning model based on image processing feedback information is to use the difference of various quality parameters as the training samples of the deep reinforcement learning model, so that the deep reinforcement learning model can learn the influence of the process adjustment strategy in the sample on the various quality parameters of the product. After executing the process adjustment strategy, various quality parameters reach the set goals.

[0146] Example 2

[0147] like Figure 2 As shown, the spiral spike process parameter optimization system based on image deep learning includes a process data collection module, a product quality assessment module, a strategy generation module and a control feedback module; wherein each module is electrically connected;

[0148] A process data collection module collects spiral spike target information, real-time image data and a set of process parameters, and sends the spiral spike target information, real-time image data and a set of process parameters to a product quality assessment module;

[0149] The product quality assessment module uses image processing technology to analyze various real-time quality parameters of the spiral spikes in the real-time image data; based on the real-time quality parameters and the spiral spike target information, a quality assessment function is constructed, and based on the real-time quality parameters and the spiral spike target information, the quality of the extruded product is analyzed using the quality assessment function, and the quality of the extruded product is sent to the strategy generation module;

[0150] The strategy generation module sends the extrusion product quality to the automation control system, generates a process adjustment strategy based on the extrusion product quality and the process parameter set using a deep reinforcement learning model, and sends the process adjustment strategy to the control feedback module;

[0151] The control feedback module uses an automated control system to dynamically adjust process parameters based on the process adjustment strategy. The automated control system receives image processing feedback information in real time and iteratively trains the deep reinforcement learning model based on the image processing feedback information.

[0152] Example 3

[0153] The present application also provides an electronic device. The electronic device may include one or more processors and one or more memories. The memories store computer-readable codes, and when the computer-readable codes are executed by the one or more processors, the method for optimizing the process parameters of a spiral spike based on image deep learning as described above may be executed.

[0154] The electronic device may include a bus, one or more CPUs, ROM, RAM, a communication port connected to a network, an input / output component, a hard disk, etc. A storage device in the electronic device, such as a ROM or a hard disk, may store the spiral spike process parameter optimization method based on image deep learning provided in the present application.

[0155] Furthermore, the electronic device may also include a user interface. Of course, this architecture is only exemplary, and when implementing different devices, one or more components in the electronic device may be omitted according to actual needs.

[0156] Example 4

[0157] The present application also discloses a computer-readable storage medium. Computer-readable instructions are stored on the computer-readable storage medium. When the computer-readable instructions are executed by the processor, the spiral spike process parameter optimization method based on image deep learning according to the embodiment of the present application can be executed. The computer-readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache). Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0158] In addition, according to the implementation of the present application, the process described above with reference to the embodiment of the present application can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, the non-transitory machine-readable storage medium stores machine-readable instructions, the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided by the present application, and when the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0159] The methods, apparatuses, and devices of the present application may be implemented in many ways. For example, the methods, apparatuses, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers recording media storing programs for executing the method according to the present application.

[0160] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0161] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0162] The above preset parameters or preset thresholds are all set by technicians in this field according to actual conditions or obtained through large-scale data simulation.

[0163] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for optimizing process parameters of spiral spikes based on image deep learning, characterized in that: The following steps are involved: Step 1: Collect spiral spike target information, real-time image data and process parameter set; Step 2: Use image processing technology to analyze various real-time quality parameters of the spiral spikes in the real-time image data; construct a quality evaluation function based on the real-time quality parameters and the spiral spike target information; Step 3: Based on the real-time quality parameters and spiral spike target information, use the quality evaluation function to analyze the quality of the extruded product; Step 4: Generate a process adjustment strategy using a deep reinforcement learning model based on the extrusion product quality and process parameter set; Step 5: Based on the process adjustment strategy, the process parameters are dynamically adjusted using the automated control system. The automated control system receives the image processing feedback information in real time and iteratively trains the deep reinforcement learning model based on the image processing feedback information. The method for collecting the target information of the spiral spike is as follows: Collecting the real-time expected values ​​of various quality parameters of the spiral spikes expected to be produced in each unit time in the production plan, and collecting the material properties and surface treatment information of the raw materials to be extruded in the production plan; the real-time expected values ​​of various quality parameters, the material properties and surface treatment information of the raw materials constitute the target information of the spiral spikes; The method of constructing the quality evaluation function based on the real-time quality parameters and the spiral spike target information is as follows: For any unit of time during the extrusion process: Calculate the difference between the real-time value of each quality parameter in the real-time quality parameter and the real-time expected value of the corresponding quality parameter in the spiral spike target information at the corresponding time as the quality parameter difference of the quality parameter; Based on the differences of various quality parameters, a quality assessment function is constructed; The method of using the deep reinforcement learning model to generate a process adjustment strategy is: Set the deep reinforcement learning model to the Actor-Critic network model; Defining a learning environment for a deep reinforcement learning model; the learning environment includes a state space, an action space, and a reward function; The state space is defined as a set of current process parameters and product quality indicators; The state space includes real-time readings of temperature, pressure, speed, and product quality expressed using a comprehensive quality score calculated by a quality assessment function; Defining the action space as a range of adjustable process parameters; the action space includes increasing or decreasing values ​​of temperature, pressure and speed; The reward function is used to encourage the deep reinforcement learning model to optimize product quality, and the reward function is constructed based on the implemented actions and the comprehensive quality score; Initialize the Actor-Critic network model; The Actor-Critic network model is preliminarily trained using the action space, selected actions, and rewards recorded in the past extrusion production process to collect state, action, and reward triplets under different production conditions; Initialize the weights of the Actor network and the Critic network; In each unit of time, the Actor network selects an action based on the current state, and after executing the action, obtains a new state and reward; the Critic network evaluates the value of the new state and updates the value function; Use gradient ascent to update the Actor network to maximize long-term rewards; Use the temporal difference error to update the Critic network to improve the accuracy of state assessment; The Actor network in the trained deep reinforcement learning model is used to output the action selected from the action space per unit time of the spiral spike to be monitored as a process parameter adjustment strategy.

2. The method for optimizing process parameters of spiral spikes based on image deep learning according to claim 1 is characterized in that: The real-time image data is collected in the following manner: During the multi-layer composite extrusion production process of the bimetallic composite spiral spike, a camera installed at the outlet of the extruder to monitor the extruded product in real time is used to collect images of the extruded spiral spikes per unit time as real-time image data.

3. The method for optimizing process parameters of spiral spikes based on image deep learning according to claim 2 is characterized in that: The process parameter set is collected in the following manner: During the multi-layer composite extrusion production process of the bimetal composite spiral spikes, various extrusion parameters of the extruder are collected in real time through the control background of the extruder.

4. The method for optimizing process parameters of spiral spikes based on image deep learning according to claim 3 is characterized in that: The method of using image processing technology to analyze various real-time quality parameters of the spiral spike in real-time image data is as follows: Performing image preprocessing on the real-time image of the spiral spike in the real-time image data; Perform edge detection on the real-time image of the spiral spike to obtain the sideline edge in the real-time image, and then find the edge of the spiral spike by detecting the brightness change position in the real-time image; the brightness change position is determined by setting a brightness change threshold; Extract the edge lines of the detected spiral spikes and identify the contours of the spiral spikes through contour detection technology; Perform frequency domain analysis on the extracted spiral spike profile to identify the spiral periodicity characteristics, and calculate the real-time pitch of the spiral spike, that is, the period of spiral repetition, by analyzing the peak position in the spectrum; The detected contour of the spiral spike is fitted into a circle or an ellipse by using the shape fitting algorithm of the least square method, and the maximum diameter is determined from the fitting result as the diameter measurement value of the spiral spike, and the diameter measurement value is the real-time diameter; The minimum circumscribed rectangle of the spiral spike is used to obtain the length of the spiral spike, and the image pixel length is converted into the actual physical length, which is the real-time length, and the image calibration data is used for calibration; The real-time pitch, real-time diameter and real-time length of the spiral spike constitute real-time quality parameters.

5. The method for optimizing process parameters of spiral spikes based on image deep learning according to claim 4 is characterized in that: The method of analyzing the quality of extruded products based on real-time quality parameters and spiral spike target information and using a quality evaluation function is as follows: Collecting real-time extrusion images of the spiral spike to be monitored during the extrusion process; Then, the real-time values ​​of various quality parameters corresponding to the real-time extrusion image are analyzed by image processing technology to form the real-time quality parameters of the spiral spike to be monitored; Then, according to the real-time quality parameters of the spiral spike to be monitored and the target information of the spiral spike, the difference of each quality parameter at each unit time is calculated respectively; The difference of each quality parameter is substituted into the quality evaluation function, and the quality of the extruded product is calculated by the quality evaluation function.

6. A spiral spike process parameter optimization system based on image deep learning, which is used to implement the spiral spike process parameter optimization method based on image deep learning as described in any one of claims 1 to 5, characterized in that: It includes a process data collection module, a product quality assessment module, a strategy generation module and a control feedback module; wherein each module is electrically connected; A process data collection module collects spiral spike target information, real-time image data and a set of process parameters, and sends the spiral spike target information, real-time image data and a set of process parameters to a product quality assessment module; The product quality assessment module uses image processing technology to analyze various real-time quality parameters of the spiral spikes in the real-time image data; based on the real-time quality parameters and the spiral spike target information, a quality assessment function is constructed, and based on the real-time quality parameters and the spiral spike target information, the quality of the extruded product is analyzed using the quality assessment function, and the quality of the extruded product is sent to the strategy generation module; The strategy generation module sends the extrusion product quality to the automation control system, generates a process adjustment strategy based on the extrusion product quality and the process parameter set using a deep reinforcement learning model, and sends the process adjustment strategy to the control feedback module; The control feedback module uses an automated control system to dynamically adjust process parameters based on the process adjustment strategy. The automated control system receives image processing feedback information in real time and iteratively trains the deep reinforcement learning model based on the image processing feedback information.

Images