Industrial equipment control system and method based on AI

Abstract

The invention discloses an AI-based industrial equipment control system and method, and belongs to the technical field of industrial automation control, the system collects multi-dimensional operation data of industrial equipment in real time through a data collection module, the multi-dimensional operation data comprises appearance parameters and working condition parameters, and after abnormal value elimination and standardization processing are carried out through a data processing module, the data are sent to a server; inputting a pre-trained DRL model to generate a plurality of groups of control parameter adjustment suggestions, performing Monte Carlo simulation verification on the adjustment suggestions through a digital twin simulation environment, screening out an optimal scheme, generating a control instruction, issuing the control instruction to industrial equipment for execution through an OPC UA protocol, and dynamically optimizing model parameters through an adaptive learning module; according to the method, the problems of poor static strategy adaptability, verification mechanism deficiency and weak self-learning ability in traditional industrial control are solved, the dynamic response ability, safety and self-adaptability of the control system are improved, and collaborative improvement of equipment energy efficiency optimization and production quality stability is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial automation control, and specifically discloses an AI-based industrial equipment control system and method. Background Art

[0002] With the advancement of Industry 4.0, the current industrial equipment control technology supports a network architecture for dynamic intelligent reconfiguration, which can quickly adapt to the requirements of different production scenarios, allows flexible adjustment of control strategies through programming, and adapts to diverse production processes; the industrial control computer collects and analyzes data in real time, provides decision-making support for production optimization, and reduces human misjudgment; the industrial equipment control technology has become the core driving force for promoting industrial automation upgrading by integrating hardware reliability, software intelligence, and user-friendliness.

[0003] However, the traditional industrial equipment control technology exposes the following limitations:

[0004] (1) Defects in static control strategies: Most existing systems adopt pre-programmed PID control or rule-based control logic, which are difficult to adapt to the dynamic changes of complex working conditions; especially in multi-physical field coupling scenarios, such as the coordinated regulation of multiple parameters such as holding time, injection speed, cooling speed, and holding pressure, the adjustment of control parameters lags behind, resulting in a decrease in equipment energy efficiency and even potential safety hazards;

[0005] (2) Lack of verification mechanism: The traditional method directly issues control instructions to physical devices, lacking a virtual verification link; although some control systems introduce a simulation module, they only rely on a single physical model for prediction and do not integrate a dynamic correction mechanism driven by real-time data, resulting in a large deviation between the simulation results and the actual working conditions;

[0006] (3) Weak adaptability: Existing AI control solutions rely on a large amount of labeled data, and the model update cycle is long; in the face of long-term variables such as equipment aging and raw material property changes, the control strategy is difficult to evolve autonomously and still requires frequent manual intervention;

[0007] In view of the above problems, there is an urgent need for an AI-based industrial equipment control system and method. Summary of the Invention

[0008] To overcome the above-mentioned defects of the prior art, the present invention provides an AI-based industrial equipment control system and method. The multi-dimensional operation data of industrial equipment, including appearance parameters and working conditions parameters, is collected in real time by a data acquisition module. After outlier removal and standardization processing by a data processing module, it is input into a pre-trained DRL model to generate multiple groups of control parameter adjustment suggestions. The adjustment suggestions are verified by Monte Carlo simulation in a digital twin simulation environment. After screening out the optimal solution, control instructions are generated and sent to the industrial equipment for execution via the OPC UA protocol. At the same time, the model parameters are dynamically optimized by an adaptive learning module, effectively solving the problems mentioned in the background technology.

[0009] To achieve the above object, the present invention provides the following technical solutions: an AI-based industrial equipment control system and method, characterized in that it includes a data acquisition module, a data processing module, an AI control core module, an adaptive learning module, and industrial equipment, wherein the data acquisition module is installed on the industrial equipment; specifically includes the following steps:

[0010] S1. The data acquisition module collects the multi-dimensional operation data of the industrial equipment in real time, and the data processing module performs outlier removal and standardization processing on the multi-dimensional operation data to obtain preprocessed data;

[0011] S2. Input the preprocessed data into the pre-trained DRL model to output multiple groups of control parameter adjustment suggestions;

[0012] S3. Verify the feasibility of the control parameter adjustment suggestions in the digital twin simulation environment, and select the optimal control parameter adjustment suggestions according to the verification results;

[0013] S4. Generate control instructions according to the verified optimal control parameter adjustment suggestions;

[0014] S5. Send the control instructions to the industrial equipment via the OPC UA protocol to adjust the production process and feedback the execution data to the adaptive learning module.

[0015] Combining all the above technical solutions, the positive effects of the present invention are as follows:

[0016] (1) Improved dynamic adaptability: By combining the DRL model and the digital twin simulation environment, the system can respond to the dynamic changes of complex working conditions in real time, replacing the traditional static control strategy and reducing the delay of control parameter adjustment;

[0017] (2) Enhanced safety and reliability: Through Monte Carlo simulation verification and multi-physical field coupling simulation in the digital twin environment, the feasibility is verified before the control instructions are sent, effectively avoiding equipment failures caused by incorrect parameter adjustment;

[0018] (3) Optimized self-learning ability: The adaptive learning module supports online incremental learning and fault diagnosis. The model update cycle is shortened to the hourly level, and it can adapt to long-term variables such as equipment aging and raw material property changes, reducing the frequency of manual intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described with reference to the accompanying drawings. However, the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the following drawings.

[0020] Figure 1 It is a flowchart of the method implementation steps of the present invention.

[0021] Figure 2 It is a flowchart for outputting control parameter adjustment suggestions in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0023] See Figure 1 As shown, the present invention proposes an AI-based industrial equipment control system and method, which is characterized in that it includes a data acquisition module, a data processing module, an AI control core module, an adaptive learning module, and industrial equipment, wherein the data acquisition module is installed on the industrial equipment;

[0024] In a more specific application of the present invention, the data acquisition module is used to collect appearance data parameters and working condition parameters. The above appearance parameters specifically include: the size, warpage, and surface defect data of the product; the above working condition parameters specifically include the qualification rate, pressure inside the mold, and injection speed; the data acquisition module can specifically be a combination of a machine vision unit and a sensor array. Among them, the machine vision unit uses a combination of a high-resolution industrial camera and a laser scanner to achieve millimeter-level detection of the product appearance size, warpage, and surface defects; the sensor array is composed of pressure sensors and speed sensors deployed at key parts of the production equipment; the pressure sensor is specifically an embedded piezoelectric sensor, and the speed sensor is installed on the injection molding machine screw to measure the instantaneous speed during the injection molding stage.

[0025] It should be further noted that the qualification rate can be counted by integrating data management software in the unit, recording the number of qualified / unqualified products in real time, and calculating the qualification rate.

[0026] The data processing module is used to perform noise reduction, outlier removal, and standardization on the data collected by the data acquisition module. The outlier removal method is to use the dynamic threshold method, calculate the mean μ and standard deviation σ based on the historical data distribution within the sliding time window, and remove the real-time data points that exceed the range of mean ± 3σ; the standardization method is to perform Min-Max normalization on the multi-source heterogeneous data, unify the dimension to the interval [0, 1], and form a standardized data matrix for the model to use.

[0027] The AI control core module includes a deep reinforcement learning model and a digital twin simulation environment, which are used to generate control parameter adjustment suggestions after inputting the preprocessed data and verify the feasibility of the parameter adjustment suggestions;

[0028] The adaptive learning module includes an online incremental learning unit: the online incremental learning unit dynamically updates the DRL model parameters according to the actual feedback data after the industrial equipment executes the control instructions;

[0029] It should be further noted that the specific implementation steps for the online incremental learning unit to dynamically update the DRL model parameters are as follows:

[0030] (1) Data feedback and collection: After the industrial equipment executes the control instructions, the multi-dimensional operation data and working condition parameters after execution are collected in real time through the data acquisition module; specifically, it includes appearance parameters such as size, warpage, surface defects, qualification rate, mold pressure, and injection speed;

[0031] (2) Feedback data preprocessing: The data processing module performs outlier removal using the dynamic threshold method and Min-Max normalization on the feedback data to generate a standardized data matrix;

[0032] (3) Update parameters: Using the online policy gradient algorithm, taking the feedback data as new training samples, combined with the historical experience replay pool, perform mini-batch gradient descent optimization on the policy network of the DRL model. The update formula is: where θ is the policy network parameter, α is the learning rate, R(s, a) is the reward function, which is calculated by weighted averaging of the energy efficiency ratio, qualification rate, and safety score. θ is the gradient operator of the parameter, representing the gradient direction of the expected return. is the expected return, that is, the expected value of the reward obtained by taking action a in state s under the policy.

[0033] It should be further noted that in practice, the expected return cannot be directly calculated and needs to be approximated by sampling:

[0034] In the formula, N represents the number of sampled trajectories, T represents the time step of a single trajectory. represents the policy network in state selecting action The probability, represents the discounted reward accumulated starting from time step t, where γ is the discount factor in the formula, and r k is the immediate reward obtained at time step k;

[0035] (4) Model verification and deployment: The updated DRL model needs to be verified by Monte Carlo simulation in the digital twin simulation environment to ensure the feasibility of its control recommendations. After passing the verification, the new model parameters are deployed to the AI control core module in real time.

[0036] Example calculation of the accumulated discounted reward:

[0037] Parameter setting: The rewards obtained from time step t to t + 2 are r t = 1, r t+1 = 2, r t+2 = 3, and the discount factor γ = 0.9;

[0038] Calculation process: G t = 1 + 0.9×2 + 0.9 2 ×3 = 1 + 1.8 + 2.43 = 5.23;

[0039] Example calculation of parameter update:

[0040] Parameter setting: Energy efficiency ratio = 82%, Qualification rate = 95%, Safety score = 90,

[0041] Reward function: R(s, a) = 0.4×Energy efficiency ratio + 0.5×Qualification rate + 0.1×Safety score,

[0042] Policy network parameter θ old [0.5, -0.3], Sampling trajectory N = 2, Time step T = 3,

[0043] Calculation process:

[0044] Trajectory 1:

[0045] At t = 0,

[0046] At t = 1,

[0047] At t = 2,

[0048] Total gradient of Trajectory 1: 2.857 + 2.0 + 1.0 = 5.857;

[0049] Trajectory 2:

[0050] At t = 0,

[0051] When t = 1,

[0052] When t = 2,

[0053] Total gradient of trajectory 1: 3.75 + 3.0 + 0.5 = 7.25;

[0054] Average gradient: (5.857 + 7.25) / 2 ≈ 6.553;

[0055]

[0056] Conclusion: Adjust the policy network parameters from [0.5, -0.3] to [0.5655, -0.2345];

[0057] The industrial equipment can be production equipment for intelligent manufacturing equipment, specifically an injection molding machine;

[0058] The specific implementation of the present invention is as Figure 1 shown, and specifically includes the following steps:

[0059] S1. The data acquisition module real-time collects multi-dimensional operation data of industrial equipment, and the data processing module performs outlier removal and standardization processing on the multi-dimensional operation data to obtain preprocessed data;

[0060] The data acquisition module cooperates with a high-resolution industrial camera and a laser scanner to collect appearance parameters of the product, including size, warpage, qualification rate, and surface defect data, and obtains the pressure and injection speed in the mold through a pressure sensor and a speed sensor; the data processing module performs noise reduction, outlier removal, and standardization processing on the data collected by the data acquisition module. The outlier removal method is to use the dynamic threshold method, calculate the mean μ and standard deviation σ based on the historical data distribution within a sliding time window, remove the real-time data points exceeding the range of mean ± 3σ, and fill the missing data based on the linear interpolation algorithm of the time series; the standardization processing method is to perform Min-Max normalization on multi-source heterogeneous data, unify the dimension to the [0, 1] interval, and form a standardized data matrix for the model to use.

[0061] It should be further noted that the method of using the dynamic threshold method for noise reduction and outlier removal is as follows: taking the pressure data as an example, calculate the mean and standard deviation of the pressure data within 10 minutes. If a certain pressure data deviates from the mean by more than 3 times the standard deviation, it is determined as an outlier and removed. Assuming the pressure data is 105 MPa, 120 MPa, 104 MPa, 103 MPa, 101 MPa, the calculated mean is 103 MPa, the standard deviation is 1.43 MPa, and 120 MPa deviates from the mean by more than 3 times the standard deviation, so it is removed.

[0062] The linear interpolation algorithm based on time series to fill in missing data means calculating linear interpolation according to the parameter values at adjacent moments before and after to fill in the missing data. For example, if the pressure data at a certain moment is missing, it is filled with the average value of the pressure values at the moments before and after. Taking the pressure data at a certain moment as an example, assuming that t1, t2, and t3 are three consecutive moments, the measured pressure at t1 is 103 MPa, the data at t2 is missing, and the measured pressure data at t3 is 105 MPa, then the pressure data at t2 is filled as (103 + 105) / 2 = 104 MPa.

[0063] The method for performing Min - Max normalization on the data is as follows. Taking the pressure data as an example: Set P max = 110 MPa, P min = 100 MPa, the pressure data P at a certain moment is 105 MPa, after normalization, P norm = (105 - 100) / (110 - 100) = 0.5 MPa.

[0064] S2. Input the pre - processed data into the pre - trained DRL model, and output multiple groups of control parameter adjustment suggestions;

[0065] It should be explained that the control parameter adjustment suggestions described in the above step S2 include the adjustment of injection speed, holding pressure, and cooling speed. The specific adjustment formulas are as follows:

[0066] Injection speed adjustment formula: In the formula, V new is the adjusted injection speed; V base is the reference injection speed; ΔL is the difference between the measured size and the target size; N defect is the number of surface defects per unit area; k1, k2 are material - related correction coefficients, which are set according to the characteristics of the injection material; L target is the theoretical size of the plastic part under ideal molding conditions.

[0067] Example of injection speed adjustment:

[0068] Scenario: When a certain injection molding machine produces plastic parts, the measured size is 2 mm smaller than the target size, that is, ΔL = - 2 mm, and the surface defect density is 5 pieces / m 2 ;

[0069] Parameter setting: Reference injection speed V base = 80 mm / s, target size L target = 10 mm, material correction coefficients k1 = 0.1, k2 = 0.05

[0070] Calculation process:

[0071] Conclusion: After adjustment, the injection speed is increased to 100.2 mm / s to compensate for dimensional deficiencies and reduce the defect rate.

[0072] Packing pressure adjustment formula: In the formula, P hold is the packing pressure; P max is the maximum packing pressure, which is determined by the material viscosity; t hold is the packing time; t gate is the gate freeze time; ΔD is the difference between the measured shrinkage rate and the theoretical value; λ and μ are process coupling coefficients, which are determined through DOE experiments.

[0073] Example of packing pressure adjustment:

[0074] Scenario: In a certain injection molding process, the measured shrinkage rate is 0.5% higher than the theoretical value, i.e., ΔD = 0.5%, and the packing pressure needs to be optimized.

[0075] Parameter setting: The maximum packing pressure P max = 100 MPa, the packing time t hold = 10 s, the gate freeze time t gate = 5 s, the process coupling coefficient λ = 0.2, μ = 0.1;

[0076] Calculation process: P hold = 100·(1 - e -0.2·10 / 5 ) + 0.1·0.5 ≈ 33.02 MPa

[0077] Conclusion: The packing pressure is adjusted to 33.02 MPa to suppress the shrinkage rate deviation.

[0078] Cooling speed adjustment formula: In the formula, V cool is the cooling speed; W meas is the measured warpage; W crit is the critical warpage threshold, ΔT mold is the temperature difference between the mold cavity and the core, η is the heat conduction correction coefficient; W crit , ΔT mold , and the values of η are set according to expert opinions;

[0079] It should be further noted that the calculation method of the above warpage is: W = h / (2L) × 100%, where h is the maximum warpage height in the formula, L is the measurement reference length, and the measurement reference length is the diagonal length of the plastic part;

[0080] Example of warpage calculation:

[0081] Parameter setting: The maximum warpage height of a certain plastic part is 2 mm, and the diagonal length of the plastic part is 200 mm;

[0082] Calculation process: The warpage of the plastic part W = 2 / 2×200×100% = 0.5%.

[0083] Example of cooling rate adjustment:

[0084] Scenario description: In a certain injection molding production line, the measured warpage of the plastic part is 0.6%, that is, W meas = 0.6%, and the temperature difference between the mold cavity and the core is 2°C, that is, ΔT mold = 2°C. It is necessary to optimize the cooling rate to further improve the quality stability.

[0085] Parameter setting: Benchmark cooling rate: V ref = 120 L / min, Critical warpage threshold: W crit = 1%, Thermal conductivity correction coefficient: η = 0.03;

[0086] Calculation process:

[0087] Conclusion: Adjust the cooling rate to 50.88 L / min.

[0088] S3. Verify the feasibility of the control parameter adjustment suggestions in the digital twin simulation environment, and select the optimal control parameter adjustment suggestions according to the verification results;

[0089] As Figure 2 shown, in the above step S3, the method of the feasibility verification is to input the control parameter adjustment suggestions into the digital twin simulation environment and simulate running at least one complete production cycle; if the energy efficiency ratio of the equipment in the simulation results is significantly improved and the control parameters do not exceed the safety threshold, it is determined that the verification is passed and the control parameter adjustment suggestions are output; if the simulation results do not pass the verification, the previous stable control parameter adjustment suggestions are output and an alarm log is generated, and at the same time, the manual intervention channel is started.

[0090] The method for selecting the above optimal control parameter adjustment suggestions is as follows:

[0091] In the digital twin simulation environment, the system conducts multi-dimensional performance evaluations on the simulation results of each control parameter adjustment suggestion, including: energy efficiency ratio improvement rate, product qualification rate change, equipment safety index, process stability; through the weighted scoring model comprehensively calculate the scores of each plan, and select the suggestion with the highest total score; in the above scoring model: E is the energy efficiency ratio improvement rate, Q is the change in the qualification rate, S is the equipment safety score, P is the process stability score, k is the weight coefficient, A is the scoring score, where k is determined by historical experience and expert suggestions;

[0092] Example of calculating the scores of each plan:

[0093] Parameter setting: Weight k of the energy efficiency ratio improvement rate E= 0.4, the weight k of the change in the pass rate Q = 0.4, the weight k of equipment safety S = 0.1, the weight k of process stability P = 0.1;

[0094] Scenario 1: The energy efficiency ratio improvement rate E = 15%, the change in the pass rate Q = +2%, the equipment safety score S = 1.0, and the process stability score P = 1.0;

[0095] Scenario 2: The energy efficiency ratio improvement rate E = 12%, the change in the pass rate Q = +3%, the equipment safety score S = 1.0, and the process stability score P = 0.625;

[0096] Scenario 3: The energy efficiency ratio improvement rate E = 10%, the change in the pass rate Q = +1%, the equipment safety score S = 0.5, and the process stability score P = 0.417;

[0097] Scenario 4: The energy efficiency ratio improvement rate E = 8%, the change in the pass rate Q = +0%, the equipment safety score S = 1.0, and the process stability score P = 0.333;

[0098] Calculation process:

[0099] The score A1 of Scenario 1 = (15 / 15 × 0.4) + (2 / 3 × 0.4) + (1 × 0.1) + (1 × 0.1) ≈ 0.866;

[0100] The score A2 of Scenario 2 = (12 / 15 × 0.4) + (3 / 3 × 0.4) + (1 × 0.1) + (0.625 × 0.1) = 0.8825;

[0101] The score A3 of Scenario 3 = (10 / 15 × 0.4) + (1 / 3 × 0.4) + (0.5 × 0.1) + (0.417 × 0.1) = 0.4917;

[0102] The score A4 of Scenario 4 = (8 / 15 × 0.4) + (0 / 3 × 0.4) + (1 × 0.1) + (0.333 × 0.1) ≈ 0.346;

[0103] Conclusion: Scenario 2 has the highest score, and Scenario 2 is selected as the optimal control parameter adjustment recommendation;

[0104] S4. Generate a control instruction according to the control parameter adjustment recommendation after verification;

[0105] In the above step S4, the method for generating the control instruction is as follows: one-to-one correspondence between the control parameters and the control variables of the device. For example, map the accommodation speed to the screw propulsion rate parameter of the injection molding machine, map the holding pressure to the pressure control node of the hydraulic system, and map the cooling speed to the opening control node of the cooling water valve; and encapsulate the instruction according to the data format of the OPC UA protocol.

[0106] S5. Send the control instruction to the industrial device through the OPC UA protocol to adjust the production process and feedback the execution data to the adaptive learning module.

[0107] The control instruction is transmitted to the injection molding machine through the secure channel of the OPC UA protocol. Specifically, in implementation, the system first parses the control instruction into data points recognizable by the device, then writes the data points recognizable by the device into the injection molding machine to execute the instruction. At the same time, the system monitors the instruction execution status. If the device feeds back a successful execution signal, it is recorded in the log; if it times out or returns an error code, the adaptive learning module is triggered to roll back to the previous stable parameter set, and an alarm is pushed to the human-machine interface.

[0108] The above content is only an example and illustration of the structure of the present invention. Those skilled in the art of this technology make various modifications or supplements or use similar methods to replace the specific embodiments described. As long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should fall within the protection scope of the present invention.

Claims

1. A method for controlling industrial equipment based on AI, characterized in that; Specifically, it includes the following steps: S1. The data acquisition module collects multi-dimensional operation data of industrial equipment in real time, and the data processing module eliminates outliers and standardizes the multi-dimensional operation data to obtain preprocessed data; S2. Input the preprocessed data into the pre-trained DRL model to output multiple groups of control parameter adjustment suggestions; S3. Verify the feasibility of the control parameter adjustment suggestions in the digital twin simulation environment, and select the optimal control parameter adjustment suggestions according to the verification results; S4. Generate control instructions according to the optimal control parameter adjustment suggestions that have passed the verification; S5. Send the control instructions to the industrial equipment through the OPC UA protocol to adjust the production process and feedback the execution data to the adaptive learning module.

2. The industrial equipment control method based on AI according to claim 1, wherein: The preprocessed data specifically includes: appearance data parameters and working condition parameters; the appearance parameters specifically include: the size, warpage degree, and surface defect data of the product; the working condition parameters specifically include the qualification rate, the pressure in the mold, and the injection speed.

3. The industrial equipment control method based on AI according to claim 1, characterized in that: The outlier elimination uses the dynamic threshold method to eliminate abnormal data points caused by sensor noise and communication packet loss; the method of standardization is to perform Min-Max normalization on multi-source heterogeneous data, unify the dimension to the interval [0, 1], and form a standardized data matrix for the model to use.

4. The industrial equipment control method based on AI according to claim 1, wherein: The control parameter adjustment suggestions specifically include adjusting the injection speed, cooling speed, and holding pressure.

5. The industrial equipment control method based on AI according to claim 4, characterized in that: The formula for adjusting the injection molding speed is as follows: In the formula, V new is the adjusted injection molding speed; V base is the reference injection molding speed; ΔL is the difference between the measured size and the target size; N defect is the number of surface defects per unit area; k1 and k2 are material-related correction factors, which are set according to the characteristics of the injection molding material; L target is the theoretical size of the plastic part under ideal molding conditions.

6. The industrial equipment control method based on AI according to claim 4, characterized in that: The formula for adjusting the holding pressure is as follows: In the formula, P hold is the holding pressure; P max is the maximum holding pressure, which is determined by the material viscosity; t hold is the holding time; t gate is the gate freezing time; ΔD is the difference between the measured shrinkage rate and the theoretical value; λ and μ are process coupling coefficients determined through DOE experiments.

7. The industrial equipment control method based on AI according to claim 4, wherein: The formula for adjusting the cooling rate is as follows: In the formula, V cool is the cooling rate; W meas is the measured warpage; W crit is the critical warpage threshold, ΔT mold is the temperature difference between the mold cavity and the core, and η is the heat conduction correction coefficient; W crit 、ΔT mold The values of η are set according to expert opinions.

8. The industrial equipment control method based on AI according to claim 1, wherein: The digital twin simulation environment is constructed by combining physical modeling and data-driven methods, and supports the Monte Carlo simulation verification of control strategies.

9. The industrial equipment control method based on AI according to claim 1, wherein: The method of feasibility verification is: input the control parameter adjustment suggestions into the digital twin simulation environment and simulate running at least one complete production cycle; if the energy efficiency ratio of the equipment in the simulation results is significantly improved and the control parameters do not exceed the safety threshold, it is determined to pass the verification and the control parameter set is output; if the simulation results do not pass the verification, the previous stable control parameter set is output and an alarm log is generated, and at the same time, the manual intervention channel is started.

10. An AI-based industrial equipment control system as claimed in claim 1, characterized in that: It includes a data acquisition module, a data processing module, an AI control core module, an adaptive learning module, and industrial equipment, where the data acquisition module is installed on the industrial equipment.

Images