Hardware tool industry order AI production scheduling system and method based on MOM platform

Abstract

The invention discloses a hardware tool industry order AI production scheduling system and method based on an MOM platform, and belongs to the technical field of enterprise resource scheduling. The system comprises a data collection preprocessing and constraint module, an AI production scheduling module, a dynamic adjustment module and a visual interaction module. The data collection preprocessing and constraint module is responsible for collecting order data and production data, preprocessing the order data and the production data and constructing constraint conditions; the AI production scheduling module is responsible for generating an optimized production scheduling scheme by using an algorithm; the dynamic adjustment module is responsible for abnormal condition detection, classification processing and production scheduling scheme adjustment; the visual interaction module is responsible for realizing data visualization and a manual intervention production scheduling scheme by utilizing a front-end and back-end development technology; compared with a traditional order production scheduling method, the method has the technical advantage of customized AI production scheduling, solves the problem of low production efficiency caused by low speed of the traditional production scheduling method, and is more efficient in order production scheduling.

Description

Technical Field

[0001] This invention relates to the field of enterprise resource scheduling technology, and in particular to an AI-based production scheduling system and method for hardware tool orders based on the MOM platform. Background Technology

[0002] The hardware tools industry is an important part of the global light industry, covering multiple fields such as tool hardware, building hardware, and home hardware. Its products are widely used in people's daily lives and industrial production. The MOM platform is a technology platform for integrated manufacturing operations management. It can integrate multiple aspects of an enterprise, such as production planning, resource scheduling, and material management, to realize the automation and intelligence of the production process. In the hardware tools industry, the application of the MOM platform can significantly improve the efficiency and accuracy of order processing, optimize the allocation of production resources, and reduce production costs.

[0003] In the hardware tool manufacturing industry, the core and critical link is order scheduling and processing, which directly affects production efficiency. However, the hardware tool industry still faces difficulties and challenges. For example, traditional order scheduling methods often rely too much on manual scheduling, which is a rather cumbersome process. Technical staff need to repeatedly check various data and determine whether they meet production constraints, which consumes a lot of time and energy. For large orders, the scheduling cycle is relatively long, which will frequently lead to hardware tool production efficiency falling short of expectations. Therefore, manual scheduling can no longer meet the market demand for rapid response. Summary of the Invention

[0004] This invention discloses an AI-based production scheduling system and method for the hardware tool industry based on the MOM platform. The system collects order and production data in real time by calling the MOM platform's API interface. After preprocessing and constraint construction, it predicts production parameters using a random forest regression algorithm and constructs and outputs a globally optimized production scheduling scheme using a deep reinforcement learning algorithm combined with a neural network architecture. Based on this scheme, the system continuously calls the MOM platform's API interface to obtain real-time production data. Using constraints and industry characteristics as a basis, it constructs a constraint database and compares real-time production data with the globally optimized production scheduling scheme to calculate the actual deviation. The system dynamically adjusts the globally optimized production scheduling scheme based on the actual deviation and utilizes front-end and back-end development technologies to achieve data visualization and interactive functions. This solves the problem of low production efficiency caused by slow scheduling speeds in traditional scheduling methods, achieving a high-efficiency production scheduling effect. Specifically, it includes: The data collection, preprocessing and constraint module, AI scheduling module, dynamic adjustment module and visualization interaction module are connected in sequence. The data collection, preprocessing, and constraint module collects order and production data, cleans, extracts features, integrates, and preprocesses the collected data, constructs constraints, and outputs the preprocessed data and constraints. The AI ​​scheduling module, based on the preprocessed data and constraints, uses a random forest regression algorithm to predict order production parameters and a deep reinforcement learning algorithm to generate an optimized scheduling plan, outputting the optimized plan. The dynamic adjustment module, based on the optimized scheduling plan, identifies production site anomalies, categorizes these anomalies, and dynamically adjusts the optimized scheduling plan accordingly, outputting production site anomaly logs and the adjusted, optimized scheduling plan. The visualization and interaction module, based on the production site anomaly logs and the adjusted, optimized scheduling plan, uses front-end and back-end development technologies to monitor scheduling in real time and record production site anomaly logs; it completes the visualization data display, interactive functions, and plan execution.

[0005] Furthermore, the data collection, preprocessing, and constraint module specifically includes a data collection submodule, a preprocessing submodule, a constraint construction submodule, and an integration and output submodule; The data collection submodule uses API interface technology through the MOM platform to collect order data and production data, and outputs the order data and production data. The preprocessing submodule cleans the order data and production data and stores them in the database. Then, it uses a direct extraction method to extract the features of the order data and production data and outputs the features of the order data and production data. The constraint construction submodule sets production process constraints and time cost constraints, and sets specific constraints according to the characteristics of the hardware industry. It integrates the above constraints and outputs the integrated constraint conditions. The integrated output submodule generates three data tables based on order feature data, production feature data, and integration constraints. These data tables are then integrated into a structured data model, which is then output as the structured data model.

[0006] Furthermore, the AI ​​production scheduling module specifically includes a production parameter submodule and a reinforcement learning submodule; the production parameter submodule uses a random forest regression algorithm to predict production parameters based on a structured data model and outputs the prediction results; The reinforcement learning submodule uses deep reinforcement learning algorithms to generate an optimized production scheduling plan that meets the integration constraints within the structured data model based on the prediction results, and outputs the optimized production scheduling plan.

[0007] Furthermore, the production parameter submodule specifically includes a dataset partitioning unit, an RF model training unit, and a prediction output unit; The dataset is divided into training, validation, and test sets according to the structured data model. The RF model training unit, based on the training set, uses the Random Forest Regressor from the scikit-learn library integrated in Python to build an instantiated model. Based on the validation set, it uses Grid Search combined with cross-validation to derive the optimal hyperparameter combination, and then trains the final RF model on the instantiated model using this hyperparameter combination. Based on the test set, it calculates three different evaluation indices—mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²)—and then uses a weighted summation algorithm to calculate the final RF model evaluation score, used to assess model reliability. The weighted summation formula is as follows: Final score =ω mse *MSE score +ω mae *MAE score +ω r2 *R 2 score Final score ω is the final model evaluation score. mse ω is the weighting coefficient for the mean square error; mae ω is the weighting coefficient for the mean absolute error; r2 The weighting coefficients for the coefficients of determination; MSE score This is the mean squared error (MAE) value. score R is the mean absolute error value. 2 score The value of the determining factor; The prediction output unit, based on the final RF model, inputs the structured data model into the RF model to obtain the prediction results. The prediction results are then combined with the structured data model to process missing values ​​and normalize the data, resulting in integrated data.

[0008] Furthermore, the reinforcement learning submodule specifically includes an environment setup unit, an initialization unit, an experience collection and training unit, and a production scheduling scheme generation unit; The environment building unit, based on the integrated data and openAI's Gym interface, creates a simulated production environment class to ensure that there is a production scheduling environment that conforms to the Gym interface, and outputs an environment instance class; The initialization unit uses the PPO algorithm and neural network architecture (MLP) based on the Stable-Baselines3 library to create an agent for environmental interaction and learning strategies, and outputs a PPOAgent instance. The experience collection training unit sets the total number of rounds based on the PPOAgent instance, puts it back into the environment instance class to perform actions, observes the results, and collects the experience data required for learning, and outputs the experience sequence after the total number of rounds. Based on the experience sequence, it updates the internal parameters of PPOAgent to improve its strategy and outputs the final PPOAgent. The production scheduling unit obtains the order's production cycle plan through the MOM platform API interface based on the final PPOAgent, puts the production cycle plan into the final PPOAgent to generate an optimized production scheduling plan, and outputs the optimized production scheduling plan.

[0009] Furthermore, the dynamic adjustment module specifically includes a production rule base submodule, an abnormal situation detection submodule, and a dynamic classification adjustment submodule; The production rule library submodule constructs a standard production rule library based on the integrated constraints within the structured data model and other characteristics of the hardware tool manufacturing industry. The anomaly detection submodule acquires real-time production site data through the MOM platform API interface based on the standard production rule library. After standardizing and converting the data, it packages and integrates it to obtain real-time production data. The real-time production data is then compared with the optimized production scheduling plan to calculate the actual deviation value. This deviation value is then compared with the standard production rule library to determine the data and output the anomaly result. The dynamic classification and adjustment submodule categorizes abnormal situations based on the abnormal results and records them in logs. Using dynamic adjustment methods, it makes targeted modifications and adjustments to the optimized production scheduling plan and outputs production site data, the adjusted optimized production scheduling plan, and abnormal situation log data.

[0010] Furthermore, the visualization and interaction module specifically includes a visualization sub-module and an interaction sub-module; The visualization submodule records abnormal situation log data based on production site data and the adjusted optimized production scheduling plan, and uses the layout components of the front-end JavaScript framework and visualization library to design the data display dashboard (UI) and administrator interface. The interactive submodule, based on the data display dashboard (UI) and administrator interface, uses backend development combined with database application technology to design interactive functions, enabling manual input and correction of data, and optimizing the production scheduling plan based on the corrected data. Finally, it produces a highly feasible, highly adaptable and efficient final optimized production scheduling plan and delivers it for execution.

[0011] Furthermore, the specific implementation steps of the system as a whole are as follows: Step 1) Collect order data and production data. After data cleaning, feature extraction and integration preprocessing of the collected data, construct constraints and output the preprocessed data and constraints. Step 2) Receive the preprocessed data and constraints output in Step 1), use the random forest regression algorithm to predict order production parameters, and use the deep reinforcement learning algorithm to generate an optimized production scheduling plan, and output the optimized production scheduling plan. Step 3) Receive the output of Step 2), optimize the production scheduling plan, obtain abnormal situations on the production site, classify the abnormal situations on the production site, dynamically adjust and optimize the production scheduling plan according to the categories, and output the abnormal situation log on the production site and the adjusted and optimized production scheduling plan. Step 4) Receive the production site anomaly logs and adjusted and optimized production scheduling plan output in Step 3), use front-end and back-end development technologies to monitor production scheduling in real time and record the production site anomaly logs; complete the visualization data display, interactive functions and plan execution.

[0012] Technical effects: 1) This invention utilizes AI scheduling technology based on the MOM platform. Compared with traditional scheduling methods, it uses the random forest algorithm to predict order parameters and combines deep reinforcement learning algorithm and neural network architecture to construct a learning environment instance and a learned agent instance. The agent instance can automatically learn and update its internal parameters in the learning environment instance, continuously learn scheduling optimization and update strategies, and finally use the agent instance to generate an optimized scheduling plan, solving the problem of low production efficiency caused by slow scheduling speed in traditional systems.

[0013] 2) This invention utilizes front-end and back-end development technologies to design a simple visual data dashboard based on front-end visual components. This dashboard can display and dynamically update data in the hardware production process in real time, and also designs a system administrator console interface. Compared with traditional systems, it has a higher level of visualization and solves the problem of low data visualization in traditional systems.

[0014] 3) This invention utilizes a front-end and back-end interactive design, combining the data update and visualization functions required by the front-end development with back-end development technology to implement logic code execution and schedule and allocate the operation rules of the entire system. It enables manual modification of the production schedule plan in the system administrator console interface, providing the system with a more secure guarantee. Compared with traditional systems, it has more user-friendly operability and solves the problem of insufficient adaptability of traditional systems. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the AI-based production scheduling method for hardware tool orders on the MOM platform. Figure 2 Architecture diagram of the data collection, preprocessing, and constraint modules; Figure 3 Architecture diagram of the AI ​​scheduling module; Figure 4 Architecture diagram for the production parameter submodule; Figure 5 Architecture diagram for the reinforcement learning submodule; Figure 6 To dynamically adjust the module architecture diagram; Figure 7 This is an architecture diagram of the visual interaction module. Detailed Implementation

[0017] To make the technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention discloses an AI-based production scheduling system and method for hardware tool industry orders based on the MOM platform. The system is characterized by comprising: a data collection, preprocessing and constraint module, an AI production scheduling module, a dynamic adjustment module and a visualization interaction module, which are connected in sequence.

[0019] according to Figure 1The data collection, preprocessing, and constraint module collects order and production data, performs data cleaning, feature extraction, and integration preprocessing, and constructs constraints, outputting the preprocessed data and constraints. The AI ​​scheduling module, based on the preprocessed data and constraints, uses a random forest regression algorithm to predict order production parameters and a deep reinforcement learning algorithm to generate an optimized scheduling plan, outputting the optimized plan. The dynamic adjustment module, based on the optimized scheduling plan, identifies production site anomalies, categorizes these anomalies, and dynamically adjusts the optimized scheduling plan according to the categories, outputting a production site anomaly log and the adjusted optimized scheduling plan. The visualization and interaction module, based on the production site anomaly log and the adjusted optimized scheduling plan, uses front-end and back-end development technologies to monitor scheduling in real time and record production site anomaly logs, completing the visualization data display, interactive functions, and plan execution.

[0020] according to Figure 2 In the specific implementation, the data collection, preprocessing and constraint module specifically includes a data collection submodule, a preprocessing submodule, a constraint construction submodule and an integration output submodule; The data collection submodule uses API interface technology through the MOM platform to collect order data and production data, and outputs the order data and production data. The preprocessing submodule cleans the order data and production data and stores them in the database. Then, it uses a direct extraction method to extract the features of the order data and production data and outputs the features of the order data and production data. The constraint construction submodule sets production process constraints and time cost constraints, and sets specific constraints according to the characteristics of the hardware industry. It integrates the above constraints and outputs the integrated constraint conditions. The integrated output submodule generates three data tables based on order feature data, production feature data, and integration constraints. It then integrates these data tables into a structured data model and outputs the structured data model. The method for extracting data from the above modules is as follows: Data is extracted by calling the API interface of the MOM platform and the customer management system (such as ERP or CRM). Order data includes tool type, delivery date, order quantity, and order specifications; production data includes equipment status, worker shifts, worker job responsibilities, raw material inventory, and raw material supply chain information. The data is stored in JSON format. Data cleaning is performed using the Pandas library, and missing values ​​are handled using SQL query filtering. Missing fields and values ​​in the data are queried, filtered, and filled (e.g., missing supply chain information, outdated equipment status, etc.). A standard score is used to calculate the total deviation value, and corrections are made for large deviations (e.g., abnormally high order quantity, far exceeding the normal range). The standard score calculation formula is as follows:

[0021] Its Z-value represents the total deviation of anomalies; x represents a single data point to be tested; μ represents the overall mean of the data; σ represents the overall standard deviation. Orders that have passed the standard score test are ensured to match equipment specifications and calibration time (if the order creation time is later than the delivery time, time alignment with the MOM platform is performed). The preprocessed data is packaged and its features are extracted using direct extraction methods, such as order features (batch size, etc.) and production features (including equipment processing speed and worker technical skills, etc.). After obtaining feature values, the MinMaxScaler and StandardScaler tools from the Python-based Scikit-learn library are used for normalization and standardization. Constraints are then set according to the characteristics of the hardware industry (e.g., rapid tool wear and frequent tool changes are necessary during high-speed cutting or machining of hard materials). Finally, the normalized and standardized feature data (including order and production features) and the set constraint data are integrated using SQL storage technology to form a structured data model (containing order, production feature data, and constraint data) to provide data support for subsequent modules.

[0022] according to Figure 3 In the specific implementation, the AI ​​production scheduling module specifically includes a production parameter submodule and a reinforcement learning submodule; The production parameter submodule uses a random forest regression algorithm based on a structured data model to predict production parameters and outputs the prediction results; its purpose is to provide initial decision-making basis for the reinforcement learning submodule. The reinforcement learning submodule uses deep reinforcement learning algorithms based on the prediction results to generate an optimized production scheduling plan that meets the integration constraints within the structured data model, and outputs the optimized production scheduling plan. Its purpose is to generate a globally optimized production scheduling plan that satisfies the multi-objective optimization of production efficiency, cost, and on-time delivery rate.

[0023] according to Figure 4 In the specific implementation, the production parameter submodule specifically includes a dataset partitioning unit, an RF model training unit, and a prediction output unit; The dataset is divided into training, validation, and test sets according to the structured data model. The RF model training unit, based on the training set, uses the Random Forest Regressor from the scikit-learn library integrated in Python to build an instantiated model. Based on the validation set, it uses Grid Search combined with cross-validation to derive the optimal hyperparameter combination, and then trains the final RF model on the instantiated model using this hyperparameter combination. Based on the test set, it calculates three different evaluation indices—mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²)—and then uses a weighted summation algorithm to calculate the final RF model evaluation score, used to assess model reliability. The weighted summation formula is as follows: Final score =ω mse *MSE score +ω mae *MAE score +ω r2 *R 2 score Final score ω is the final model evaluation score. mse ω is the weighting coefficient for the mean square error; mae ω is the weighting coefficient for the mean absolute error; r2 The weighting coefficients for the coefficients of determination; MSE score This is the mean squared error (MAE) value. score R is the mean absolute error value. 2 score The value of the determining factor; The prediction output unit, based on the final RF model, inputs the structured data model into the RF model to obtain the prediction results. It then integrates the prediction results with the structured data model by processing missing values ​​and normalizing them to obtain the integrated data. As described above, the training set is used to train the Random Forest (RF) model, accounting for 70% of the model's data. The validation set is used to adjust the model's hyperparameters and evaluate its performance to prevent overfitting, accounting for 15% of the model's data. The test set is used to verify the model's reliability, accounting for 15% of the RF model's data. It utilizes Python programming to call the Random Forest regressor, establishes an instantiated model, and after hyperparameter tuning, verifies the model's reliability metrics. The evaluation metrics are divided into three parts. Firstly, the Mean Absolute Error (MAE) is a commonly used evaluation index that measures the average deviation between predicted and actual values. It reflects the overall accuracy of the model's predictions by calculating the average absolute value of the prediction errors across all samples. The mathematical formula is as follows:

[0024] Its MAE score is the mean absolute error; n is the number of samples; This represents the actual value of the i-th sample. Let be the predicted value for the i-th sample; Let be the absolute error of the i-th sample (the result is the absolute value); Secondly, the mean squared error (MSE) primarily measures the average of the sum of squares of the differences between predicted and actual values. A smaller MSE indicates a better predictive performance. The mathematical formula is:

[0025] Its MSE score Here, n is the mean squared error; n is the number of samples. This represents the actual value of the i-th sample. Let be the predicted value for the i-th sample; It is the squared error of the i-th sample (which eliminates positive and negative biases by squaring, while amplifying larger errors). Third, the coefficient of determination R² is a core indicator that measures the explanatory power of a regression model for the variation of the dependent variable (the explained variable). Its theoretical basis is based on three core concepts: total sum of squares (SST), residual sum of squares (SSE), and regression sum of squares. The formula for the coefficient of determination R² is derived from the evolution of these relationships:

[0026] R² is the coefficient of determination; 1 is the measure (in practice, the closer the coefficient of determination is to 1, the stronger the model's ability to interpret the data and the better the fit); SSE is the sum of squared residuals; SST is the sum of squares. Based on the three evaluation index parameters mentioned above, in practical applications, relying on only one evaluation index parameter is not sufficient to determine the reliability of the model. Therefore, a weighted summation method is used to integrate the three evaluation index parameters (the weighted summation formula is as shown above), and finally a combined index value is calculated, which can be used to evaluate the reliability of the model. according to Figure 5 In the specific implementation, the reinforcement learning sub-module specifically includes an environment class building unit, an initialization unit, an experience collection and training unit, and a production scheduling scheme generation unit; The environment building unit creates a simulated production environment class based on the integrated data and the openAI Gym interface, enabling PPOAgent to learn in the simulated production environment class, ensuring a production scheduling environment that conforms to the Gym interface, and outputting an environment instance class. The initialization unit, based on the environment instance class, uses the PPO algorithm and neural network architecture (MLP) within the Stable-Baselines3 library to create an agent for environment interaction and learning strategies, and outputs a PPOAgent instance. The experience collection training unit sets the total number of rounds based on the PPOAgent instance, puts it back into the environment instance class to perform actions, observes the results, and collects the experience data required for learning, and outputs the experience sequence after the total number of rounds. Based on the experience sequence, it updates the internal parameters of PPOAgent to improve its strategy and outputs the final PPOAgent. The production scheduling unit obtains the order's production cycle plan through the MOM platform API interface based on the final PPOAgent, puts the production cycle plan into the final PPOAgent to generate an optimized production scheduling plan, and outputs the optimized production scheduling plan. Based on the above, when building an environment using OpenAI's Gym interface, this interface should inherit from the gym.env class. After initialization, it should return an initial state vector and call the step method to implement step(action). The action is an integer index of a specific order. Its internal logic is to update the WIP queue and resource usage based on the action, calculate the new state and reward index, and return a new set of states and reward indices after the process is completed. At the same time, it defines the action and state space, and finally integrates them to output a complete environment instance class. Based on the above, the process of collecting experience is as follows: the Agent controls all instances in the vector environment, interacts with the environment, and collects a batch of experience data, including state, action, reward, next state, and whether it has ended. Then, it calls the vec_env.reset method in the Stable-Baselines3 library to reset the environment. After the Agent selects an action, it obtains feedback, adds the collected feedback experience to the experience replay buffer, updates its policy based on the feedback experience, and closes the environment instance class after training for the set total number of rounds. It outputs the updated PPOAgent and saves it for subsequent loading and use.

[0027] according to Figure 6 In the specific implementation, the dynamic adjustment module specifically includes a production rule base submodule, an abnormal situation detection submodule, and a dynamic classification adjustment submodule; The production rule base submodule constructs a standard production rule base based on the integrated constraints within the structured data model and other characteristics of the hardware tool manufacturing industry, including heat treatment requirements (such as quenching temperature requirements) and the constraint characteristics of machining tools. Its purpose is to provide rule comparison data for the anomaly detection submodule. The anomaly detection submodule acquires real-time production site data through the MOM platform API interface based on the standard production rule library. After standardizing and converting the data, it packages and integrates it to obtain real-time production data. The real-time production data is then compared with the optimized production scheduling plan to calculate the actual deviation value. This deviation value is then compared with the standard production rule library to determine the data and output the anomaly result. The dynamic classification and adjustment submodule categorizes abnormal situations based on the abnormal results (such as processing abnormalities, personnel allocation abnormalities, etc.) and records logs (including a brief description of the abnormal situation, the time of occurrence of the abnormality, and the severity of the event). The classification results are then used to make targeted modifications and adjustments to the optimized production scheduling plan based on the dynamic adjustment method, and output production site data, the adjusted optimized production scheduling plan, and abnormal situation log data.

[0028] according to Figure 7 In the specific implementation, the visualization interaction module specifically includes a visualization sub-module and an interaction sub-module; The visualization submodule, based on production site data, the adjusted optimized production scheduling plan, and abnormal situation log data, uses the layout components of the front-end JavaScript framework and visualization library to design the data display dashboard (UI) and administrator interface. The interactive submodule, based on the data display dashboard (UI) and administrator interface, uses Python as the basic language combined with MySQL database application technology to design interactive functions, enabling manual intervention, input of correction data, and optimization of the production scheduling plan based on the correction data. Finally, it produces a highly feasible, highly adaptable, and highly efficient final optimized production scheduling plan and delivers it for execution.

[0029] In specific implementation, the overall implementation steps of the system are as follows: Step 1) Collect order data and production data. After data cleaning, feature extraction and integration preprocessing of the collected data, construct constraints and output the preprocessed data and constraints. Step 2) Receive the preprocessed data and constraints output in Step 1), use the random forest regression algorithm to predict order production parameters, and use the deep reinforcement learning algorithm to generate an optimized production scheduling plan, and output the optimized production scheduling plan. Step 3) Receive the output of Step 2), optimize the production scheduling plan, obtain abnormal situations on the production site, classify the abnormal situations on the production site, dynamically adjust and optimize the production scheduling plan according to the categories, and output the abnormal situation log on the production site and the adjusted and optimized production scheduling plan. Step 4) Receive the production site anomaly logs and adjusted and optimized production scheduling plan output in Step 3), use front-end and back-end development technologies to monitor production scheduling in real time and record the production site anomaly logs; complete the visualization data display, interactive functions and plan execution.

[0030] Working Principle: The system is designed using Python programming. It obtains real-time order and production data related to hardware production by calling the APIs of MOM platforms and customer management systems (such as ERP or CRM). This data undergoes cleaning, feature extraction, and preprocessing, while simultaneously setting constraints. These constraints should include the basic characteristics of the hardware manufacturing industry and the production rules to be followed. During system design, a database is created using basic SQL storage techniques, and the data is stored in this database. Further, this data is extracted, and a standard score algorithm is used to check for missing fields and values ​​and handle any issues. Order specifications are confirmed to be compatible with equipment, and calibration times are ensured to be consistent. Next, feature vectors are extracted using a direct extraction method, and then normalized and standardized. SQL techniques are then used to re-store the feature vectors and constraints in the database according to a structured data model. Finally, the structured data model is transferred to the AI ​​scheduling module. An instantiated model is built using the random forest regressor from the scikit-learn library integrated into Python. After training, this instantiated model can be used to predict order parameters. The predicted parameters and the structured data model are then integrated... The integrated data is then used to create an environment instance based on the Gym interface. An agent is created using the PPO algorithm and neural network architecture. After training and strategy improvement, a PPOAgent is obtained, which can be used to generate production scheduling plans. The generated production scheduling plan is then transmitted to the dynamic adjustment module. Based on the production scheduling plan, the module again calls the MOM platform's API to obtain real-time production site data. Combining the integrated constraints within the structured data model and other characteristics of the hardware tool manufacturing industry, a standard production rule database is created using SQL technology. The deviation between the production scheduling plan and the real-time production site data is calculated and stored in the standard production rule database to determine whether it conforms to the production rules. The production scheduling plan is adjusted in real time, and anomaly logs are recorded promptly. Furthermore, a dashboard displaying the data and a user-interventionable client are designed using front-end and back-end programming. User input data can be transmitted to the system for modification of the production scheduling plan. An anomaly log database is also created using SQL technology to store data from each troubleshooting session, along with the original and modified data, for subsequent regression analysis.

[0031] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of variations or substitutions within the technical scope disclosed in the present invention, but such variations or substitutions still fall within the spirit and connotation of the present invention and have not departed from the spirit and connotation of the present invention. Therefore, any invention application made under such circumstances should be covered within the scope of protection of the present invention.

Claims

1. A hardware tool industry order AI scheduling system based on the MOM platform, characterized in that, It includes a data collection, preprocessing and constraint module, an AI scheduling module, a dynamic adjustment module and a visualization interaction module, which are connected in sequence. The data collection, preprocessing, and constraint module collects order data and production data, performs data cleaning, feature extraction, and integration preprocessing on the collected data, constructs constraints, and outputs the preprocessed data and constraints. The AI ​​production scheduling module uses a random forest regression algorithm to predict order production parameters based on preprocessed data and constraints, and uses a deep reinforcement learning algorithm to generate an optimized production scheduling plan, outputting the optimized production scheduling plan. The dynamic adjustment module obtains abnormal situations on the production site based on the optimized production scheduling plan, classifies the abnormal situations, dynamically adjusts and optimizes the production scheduling plan according to the category, and outputs a log of abnormal situations on the production site and the adjusted and optimized production scheduling plan. The visual interaction module uses front-end and back-end development technologies to monitor production scheduling in real time and record production site anomaly logs based on production site anomaly logs and adjusted and optimized production scheduling plans. Complete the visualization of data, interactive functions, and implementation of solutions.

2. The hardware tool industry order AI scheduling system based on the MOM platform according to claim 1, characterized in that, The data collection, preprocessing, and constraint module specifically includes a data collection submodule, a preprocessing submodule, a constraint construction submodule, and an integration and output submodule. The data collection submodule collects order and production data through the MOM platform using API interface technology, and outputs the order and production data. The preprocessing submodule cleans the order data and production data and stores them in the database. Then, it uses a direct extraction method to extract the features of the order data and production data and outputs the features of the order data and production data. The constraint construction submodule sets production process constraints and time cost constraints, and sets specific constraints according to the characteristics of the hardware industry. It integrates the above constraints and outputs the integrated constraint conditions. The integrated output submodule generates three data tables based on order feature data, production feature data, and integration constraints. These data tables are then integrated into a structured data model, which is then output as the structured data model.

3. The hardware tool industry order AI scheduling system based on the MOM platform according to claim 2, characterized in that, The AI ​​production scheduling module specifically includes a production parameter submodule and a reinforcement learning submodule; The production parameter submodule uses a random forest regression algorithm based on a structured data model to predict production parameters and outputs the prediction results. The reinforcement learning submodule uses deep reinforcement learning algorithms based on the prediction results to generate an optimized production scheduling plan that meets the integration constraints within the structured data model, and outputs the optimized production scheduling plan.

4. The hardware tool industry order AI scheduling system based on the MOM platform according to claim 3, characterized in that, The production parameter submodule specifically includes a dataset partitioning unit, an RF model training unit, and a prediction output unit. The dataset is divided into training, validation, and test sets according to the structured data model. The RF model training unit, based on the training set, uses the Random Forest Regressor from the scikit-learn library integrated in Python to build an instantiated model. Based on the validation set, it uses Grid Search combined with cross-validation to derive the optimal hyperparameter combination, and then trains the final RF model on the instantiated model using this hyperparameter combination. Based on the test set, it calculates three different evaluation indices—mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²)—and then uses a weighted summation algorithm to calculate the final RF model evaluation score, used to assess model reliability. The weighted summation formula is as follows: Final score =ω mse *MSE score +oh mae *MAE score +oh r2 *R 2 score Final score ω is the final model evaluation score. mse ω is the weighting coefficient for the mean square error; mae ω is the weighting coefficient for the mean absolute error; r2 The weighting coefficients for the coefficients of determination; MSE score This is the mean squared error (MAE) value. score R is the mean absolute error value. 2 score The value of the determining factor; The prediction output unit, based on the final RF model, inputs the structured data model into the RF model to obtain the prediction results. The prediction results are then combined with the structured data model to process missing values ​​and normalize the data, resulting in integrated data.

5. The hardware tool industry order AI scheduling system based on the MOM platform according to claim 4, characterized in that, The reinforcement learning submodule specifically includes an environment setup unit, an initialization unit, an experience collection and training unit, and a production scheduling scheme generation unit. The environment building unit, based on the integrated data and openAI's Gym interface, creates a simulated production environment class to ensure that there is a production scheduling environment that conforms to the Gym interface, and outputs an environment instance class; The initialization unit uses the PPO algorithm and neural network architecture (MLP) based on the Stable-Baselines3 library to create an agent for environmental interaction and learning strategies, and outputs a PPOAgent instance. The experience collection training unit sets the total number of rounds based on the PPOAgent instance, puts it back into the environment instance class to perform actions, observes the results, and collects the experience data required for learning, and outputs the experience sequence after the total number of rounds. Based on the experience sequence, it updates the internal parameters of PPOAgent to improve its strategy and outputs the final PPOAgent. The production scheduling unit obtains the order's production cycle plan through the MOM platform API interface based on the final PPOAgent, puts the production cycle plan into the final PPOAgent to generate an optimized production scheduling plan, and outputs the optimized production scheduling plan.

6. The hardware tool industry order AI scheduling system based on the MOM platform according to claim 5, characterized in that, The dynamic adjustment module specifically includes a production rule base submodule, an abnormal situation detection submodule, and a dynamic classification adjustment submodule; The production rule library submodule constructs a standard production rule library based on the integrated constraints within the structured data model and other characteristics of the hardware tool manufacturing industry. The abnormal situation detection submodule obtains real-time production site data through the MOM platform API interface based on the standard production rule library, and packages and integrates the data after standardizing the format conversion to obtain real-time production data. The actual deviation value is calculated by combining real-time production data with optimized production scheduling plan. The deviation value is then compared with the standard production rule library to determine the data and output the abnormal results. The dynamic classification and adjustment submodule categorizes abnormal situations based on the abnormal results and records them in logs. Using dynamic adjustment methods, it makes targeted modifications and adjustments to the optimized production scheduling plan and outputs production site data, the adjusted optimized production scheduling plan, and abnormal situation log data.

7. The hardware tool industry order AI scheduling system based on the MOM platform according to claim 6, characterized in that, The visualization and interaction module specifically includes a visualization sub-module and an interaction sub-module; The visualization submodule records abnormal situation log data based on production site data and the adjusted optimized production scheduling plan, and uses the layout components of the front-end JavaScript framework and visualization library to design the data display dashboard (UI) and administrator interface. The interactive submodule, based on the data display dashboard (UI) and administrator interface, uses backend development combined with database application technology to design interactive functions, enabling manual input and correction of data, refining and optimizing the production scheduling plan based on the corrected data, and finally delivering the optimized production scheduling plan for execution.

8. A hardware tool industry order AI scheduling method based on a MOM platform, characterized in that, The method includes the following steps: Step 1) Collect order data and production data. After data cleaning, feature extraction and integration preprocessing of the collected data, construct constraints and output the preprocessed data and constraints. Step 2) Based on the preprocessed data and constraints output in Step 1), use the random forest regression algorithm to predict order production parameters, and use the deep reinforcement learning algorithm to generate an optimized production scheduling plan, and output the optimized production scheduling plan. Step 3) Based on the optimized production scheduling plan output in Step 2), obtain abnormal situations on the production site, classify the abnormal situations on the production site, dynamically adjust and optimize the production scheduling plan according to the categories, and output the abnormal situation log and the adjusted and optimized production scheduling plan. Step 4) Based on the production site anomaly logs and the adjusted and optimized production scheduling plan output in Step 3), use front-end and back-end development technologies to monitor production scheduling in real time and record production site anomaly logs; complete the visualization data display, interactive functions and plan execution.

Images