Intelligent CNC machine tool power prediction method and system

Abstract

The present invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for predicting power of intelligent CNC machine tools. The method comprises the following steps: collecting historical operation records of intelligent CNC machine tools to obtain an integrated historical operation record data set; identifying power consumption patterns of intelligent CNC machine tools based on the integrated historical operation record data set to obtain a power consumption pattern data set; dividing the integrated historical operation record data set according to the power consumption pattern data set to obtain several groups of historical operation record data sets; mining work task features of each group of historical operation record data sets to obtain each work task feature data; constructing an adaptive model for intelligent CNC machine tools based on the power consumption pattern data set, several groups of historical operation record data sets and each work task feature data to obtain a robot power prediction model. The present invention effectively improves the accuracy and adaptability of power prediction.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for predicting power of an intelligent numerically controlled machine tool. Background Art

[0002] In the field of industrial automation, the power consumption of robots is a key factor affecting production costs and energy efficiency. However, existing power prediction technologies for intelligent CNC machine tools often have the problem of insufficient accuracy, which limits their application in actual production. On the one hand, the power consumption of intelligent CNC machine tools is affected by many factors, such as task type, load weight, and moving path. The complexity and variability of these factors make it a challenge to accurately predict power consumption. On the other hand, existing prediction models usually lack in-depth analysis of the robot's historical operating data and cannot fully capture the subtle differences in power consumption patterns.

[0003] In addition, the operating environment and working conditions of intelligent CNC machine tools often change, which requires the power prediction model to be able to adapt to different working scenarios and conditions. However, traditional prediction methods often use static models, which are difficult to cope with dynamically changing environments. This limitation leads to a large deviation between the prediction results and the actual power consumption, affecting the accuracy of production planning and the efficiency of energy management. Summary of the invention

[0004] Based on this, it is necessary for the present invention to provide a method and system for predicting power of intelligent CNC machine tools to solve at least one of the above technical problems.

[0005] To achieve the above object, a method for predicting power of an intelligent CNC machine tool comprises the following steps:

[0006] Step S1: collecting historical operation records of the intelligent CNC machine tool to obtain an integrated historical operation record data set; identifying the power consumption pattern of the intelligent CNC machine tool based on the integrated historical operation record data set to obtain a power consumption pattern data set;

[0007] Step S2: dividing the integrated historical operation record data set according to the power consumption mode data set to obtain several groups of historical operation record data sets; mining the work task characteristics of each group of historical operation record data sets to obtain each work task characteristic data;

[0008] Step S3: constructing an adaptive model for the intelligent CNC machine tool based on the power consumption mode data set, several groups of historical operation record data sets and characteristic data of each work task to obtain a robot power prediction model;

[0009] Step S4: constructing a working virtual environment for the intelligent CNC machine tool according to the integrated historical operation record data set to obtain a robot working virtual environment; performing intensive training on the robot power prediction model based on the robot working virtual environment to obtain an optimized robot power prediction model;

[0010] Step S5: Based on the preset federated learning framework, the optimized robot power prediction model is deployed to several edge devices of the target factory to obtain a distributed robot power prediction model set; the real-time task data stream of the robot is obtained; based on the distributed robot power prediction model set, the power of the intelligent CNC machine tool is predicted according to the actual task data of the robot to obtain a robot prediction power data set.

[0011] The present invention can identify the power consumption mode of the robot by collecting and analyzing the historical operation records of the intelligent CNC machine tool, thereby laying the foundation for building a more accurate power prediction model. This can better understand and predict the energy consumption behavior of the robot under different working conditions. By dividing and mining the integrated data set, the model can identify the specific features of different work tasks, so that the prediction model can not only adapt to a single task, but also adapt to a variable combination of work tasks, thereby enhancing the generalization ability and adaptability of the model. By constructing a working virtual environment and conducting intensive training, the power prediction model can be learned and optimized in a simulated dynamic environment, which improves the model's ability to respond to changes in actual working conditions and reduces the impact of environmental changes on prediction accuracy. By using the federated learning framework to deploy the optimized model on the edge device, the distributed operation of the model is realized. This not only reduces the dependence on the central server, but also can realize fast and real-time localized power prediction, and improves the response speed and reliability of the system. By collecting the robot's task data stream in real time and combining the distributed model for power prediction, continuous and real-time power consumption prediction can be provided. This is of great significance for energy management and optimization in the production process, and helps to achieve more refined production scheduling and energy use. In summary, the present invention effectively improves the accuracy and adaptability of power prediction.

[0012] Preferably, the present invention further provides an intelligent CNC machine tool power prediction system for executing the intelligent CNC machine tool power prediction method as described above, the intelligent CNC machine tool power prediction system comprising:

[0013] The power consumption pattern recognition module is used to collect historical operation records of the intelligent CNC machine tool to obtain an integrated historical operation record data set; based on the integrated historical operation record data set, the power consumption pattern of the intelligent CNC machine tool is recognized to obtain a power consumption pattern data set;

[0014] The work task feature mining module is used to divide the integrated historical operation record data set according to the power consumption mode data set to obtain several groups of historical operation record data sets; perform work task feature mining on each group of historical operation record data sets to obtain each work task feature data;

[0015] A prediction model building module is used to build an adaptive model for the intelligent CNC machine tool based on a power consumption mode data set, several sets of historical operation record data sets and characteristic data of each work task, so as to obtain a robot power prediction model;

[0016] The model reinforcement training module is used to construct a working virtual environment for the intelligent CNC machine tool according to the integrated historical operation record data set to obtain a robot working virtual environment; based on the robot working virtual environment, the robot power prediction model is reinforced and trained to obtain an optimized robot power prediction model;

[0017] The power prediction module is used to deploy the optimized robot power prediction model to several edge devices in the target factory based on the preset federated learning framework to obtain a distributed robot power prediction model set; obtain the robot's real-time task data stream; and perform power prediction on the intelligent CNC machine tool based on the distributed robot power prediction model set according to the robot's actual task data to obtain a robot prediction power data set.

[0018] In the present invention, through the in-depth analysis of historical operation data by the power consumption pattern recognition module, the power consumption pattern can be accurately identified, providing reliable basic data for subsequent predictions. The work task feature mining module can extract key features from historical data, which are crucial for understanding the energy consumption behavior of the robot under different tasks. The prediction model construction module uses the extracted feature data to build an adaptive model, so that the prediction model can automatically adjust according to different tasks and conditions, thereby improving the flexibility and accuracy of the prediction. The model reinforcement training module trains the model by building a virtual environment, enhances the model's adaptability to actual work scenarios, and optimizes the prediction results. The power prediction module adopts a federated learning framework to realize the distributed deployment of the model on edge devices, reduce the pressure of centralized computing, and improve the scalability and response speed of the system. By acquiring and processing the robot's task data stream in real time, it provides instant power prediction and provides real-time data support for production scheduling and energy management. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0020] Figure 1 A schematic flow chart of the steps of a method for predicting power of an intelligent CNC machine tool according to an embodiment is shown.

[0021] Figure 2 A detailed flow chart of step S4 of an embodiment is shown.

[0022] Figure 3 A detailed flow chart of step S45 of an embodiment is shown. DETAILED DESCRIPTION

[0023] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0024] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0025] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0026] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for predicting power of an intelligent numerically controlled machine tool, comprising the following steps:

[0027] Step S1: collecting historical operation records of the intelligent CNC machine tool to obtain an integrated historical operation record data set; identifying the power consumption pattern of the intelligent CNC machine tool based on the integrated historical operation record data set to obtain a power consumption pattern data set;

[0028] Step S2: dividing the integrated historical operation record data set according to the power consumption mode data set to obtain several groups of historical operation record data sets; mining the work task characteristics of each group of historical operation record data sets to obtain each work task characteristic data;

[0029] Step S3: constructing an adaptive model for the intelligent CNC machine tool based on the power consumption mode data set, several groups of historical operation record data sets and characteristic data of each work task to obtain a robot power prediction model;

[0030] Step S4: constructing a working virtual environment for the intelligent CNC machine tool according to the integrated historical operation record data set to obtain a robot working virtual environment; performing intensive training on the robot power prediction model based on the robot working virtual environment to obtain an optimized robot power prediction model;

[0031] Step S5: Based on the preset federated learning framework, the optimized robot power prediction model is deployed to several edge devices of the target factory to obtain a distributed robot power prediction model set; the real-time task data stream of the robot is obtained; based on the distributed robot power prediction model set, the power of the intelligent CNC machine tool is predicted according to the actual task data of the robot to obtain a robot prediction power data set.

[0032] The present invention can identify the power consumption mode of the robot by collecting and analyzing the historical operation records of the intelligent CNC machine tool, thereby laying the foundation for building a more accurate power prediction model. This can better understand and predict the energy consumption behavior of the robot under different working conditions. By dividing and mining the integrated data set, the model can identify the specific features of different work tasks, so that the prediction model can not only adapt to a single task, but also adapt to a variable combination of work tasks, thereby enhancing the generalization ability and adaptability of the model. By constructing a working virtual environment and conducting intensive training, the power prediction model can be learned and optimized in a simulated dynamic environment, which improves the model's ability to respond to changes in actual working conditions and reduces the impact of environmental changes on prediction accuracy. By using the federated learning framework to deploy the optimized model on the edge device, the distributed operation of the model is realized. This not only reduces the dependence on the central server, but also can realize fast and real-time localized power prediction, and improves the response speed and reliability of the system. By collecting the robot's task data stream in real time and combining the distributed model for power prediction, continuous and real-time power consumption prediction can be provided. This is of great significance for energy management and optimization in the production process, and helps to achieve more refined production scheduling and energy use. In summary, the present invention effectively improves the accuracy and adaptability of power prediction.

[0033] In this embodiment, a suitable industrial Internet of Things (IIoT) platform, such as ThingWorx, Siemens Mindsphere or IBM Watson IoT, is selected and configured. By connecting the robot control system and the sensor, the data collection frequency (e.g., once per second) is set to ensure that the collected data includes but is not limited to the robot operation status, location information, and historical operation records of the error log. A centralized data storage system is established, such as a MySQL or MongoDB database, and an ETL (Extract, Transform, Load) tool (such as Talend or Apache Nifi) is used to clean, transform and integrate data to form an integrated historical operation record data set. A data mining tool (such as Scikit-learn in Python or Caret in R) is used to identify the power consumption pattern of the integrated historical operation record data set, extract key features (such as spectral features, time domain features), and obtain a power consumption pattern data set. The integrated historical operation record data set is classified according to the power consumption pattern data set, and a clustering algorithm (such as K-means or DBSCAN) is used to divide the data set to obtain several groups of historical operation record data sets. The natural language processing (NLP) technology and machine learning methods are used to mine the characteristics of each group of historical operation record data sets, identify the characteristics of each task (such as load, path), and obtain the characteristic data of each work task. Based on the power consumption mode data set, several groups of historical operation record data sets and the characteristic data of each work task, an adaptive model is built using a machine learning framework (such as TensorFlow or PyTorch) to obtain a preliminary robot power prediction model. By integrating the historical operation record data sets, a robot working virtual environment is built using simulation software (such as Gazebo or V-REP). Based on the virtual environment, the preliminary robot power prediction model is reinforced and trained using a reinforcement learning algorithm (such as Deep Q-Learning or Policy Gradient) to optimize the model performance and obtain an optimized robot power prediction model. Then, based on a preset federated learning framework (such as TensorFlow Federated or Federated AI Technology Enabler), the optimized robot power prediction model is distributed to multiple edge node devices in the target factory through a preset model distribution mechanism, a localized model instance is built, and a distributed robot power prediction model set is obtained. The real-time task data stream of the robot is obtained by integrating the data acquisition module of the intelligent CNC machine tool control system. Finally, based on the distributed robot power prediction model set and the actual task data of the robot, the model is used to predict power, and the computing power and real-time data processing technology of the edge computing platform are combined to obtain the robot prediction power data set.

[0034] Preferably, step S1 comprises the following steps:

[0035] Step S11: collecting historical operation records of the intelligent CNC machine tool to obtain an integrated historical operation record data set;

[0036] Specifically, select and configure a suitable Industrial Internet of Things (IIoT) platform, such as Siemens Mindsphere. By connecting the robot control system and sensors, set the data collection frequency to once per second, and ensure that the collected data includes but is not limited to the robot operation status, location information, and error log historical operation records. In addition, establish a centralized data storage system, such as a MySQL database, to centrally store and manage all collected historical operation records to form an integrated historical operation record data set.

[0037] Step S12: extracting the device ID from the integrated historical operation record data set to obtain a robot device ID data set;

[0038] Specifically, we can use the Pandas data processing library in Python to read the integrated historical operation record data set, and extract the column containing the device ID by filtering and cleaning the data. Then, we can deduplicate the device ID column to ensure that each device ID is unique, thus forming a robot device ID data set.

[0039] Step S13: performing usage time statistics on the integrated historical operation record data set to obtain a robot usage time data table;

[0040] Specifically, select appropriate data processing and statistical methods, such as using the Python programming language and using the Pandas library for data processing. Then, read the integrated historical operation record data set and convert the timestamp column to the datetime type. Next, group the data by device ID and count the total usage time of each device. The specific operation includes calculating the earliest and latest usage timestamps of each device, and taking the difference between the two to obtain the total usage time of each robot, and finally obtaining the robot usage time data table.

[0041] Step S14: collecting the service life of retired intelligent CNC machine tools to obtain a robot service life data set;

[0042] Specifically, work with the factory or robot management department to obtain a list of all retired intelligent CNC machine tools, including each robot model, equipment ID, and retirement time. Next, query the detailed records of these robots from the beginning of use to retirement, which can be obtained from the factory's equipment management system or maintenance log. This data is organized and stored in a centralized data storage system, such as a MySQL database, to form a robot service life data set.

[0043] Step S15: Divide the intelligent CNC machine tool into use stages according to the robot service life data set and the robot use time data table to obtain use stage data;

[0044] Specifically, please refer to the sub-steps of step S15 for the specific processing process of this embodiment.

[0045] Step S16: Based on the robot device ID data set and the robot usage time data table, the integrated historical operation record data set is labeled and divided according to the usage stage data to obtain several stage historical operation record data sets;

[0046] Specifically, the robot device ID data set, the robot usage time data table, and the usage stage data are extracted from the database. Then, the integrated historical operation record data set is read, and each record is matched and labeled with the corresponding usage stage according to the device ID. For example, for each historical operation record, its device ID is first determined, and then the record is labeled as early, middle, or late according to the usage time and usage stage data of the device. Then, the labeled historical operation record data set is divided according to the usage stage to form several stages of historical operation record data sets.

[0047] Step S17: Based on the historical operation record data sets of several stages, the power consumption pattern of the intelligent CNC machine tool is identified to obtain a power consumption pattern data set.

[0048] Specifically, please refer to the sub-steps of step S17 for the specific processing process of this embodiment.

[0049] The present invention provides a unified data foundation for subsequent data analysis and model building by integrating the standardized processing of historical operation record data sets. Secondly, the extraction and management of device IDs enable clear identification of the robot's usage and performance, which helps to carry out effective equipment maintenance and optimization. In addition, the statistics of usage time and the collection of service life provide important information for evaluating the robot's workload, life cycle, and fault prevention. By dividing the use stage and building a staged data set, the understanding of power consumption patterns in different stages is further deepened, thereby improving the accuracy and adaptability of the prediction model.

[0050] Preferably, step S15 comprises the following steps:

[0051] Step S151: collecting historical fault records of retired intelligent CNC machine tools to obtain a fault record data set;

[0052] Specifically, we can work with the factory or robot management department to obtain the fault records of all retired intelligent CNC machine tools, including but not limited to the time, fault type, fault description and repair measures of each fault. The data source can be the factory's equipment management system, maintenance log or repair report. Then, these fault record data are organized into a structured format, such as a spreadsheet or database table, and finally a fault record data set is obtained.

[0053] Step S152: performing major fault detection on the fault record data set to obtain major fault detection result data;

[0054] Specifically, you can define what a major failure is, such as a failure that causes the robot to shut down for more than 24 hours, a failure that requires the replacement of key components, or a failure that occurs repeatedly. Next, use data analysis tools (such as the Pandas library in Python or the R language) to analyze the fault record dataset and filter out records that meet the major failure criteria. For example, you can write a script to automatically identify the type of fault and the length of time it takes to repair it, and mark all records that meet the major failure criteria. Then, organize these major failure records into a new dataset to form major failure detection result data, and provide a detailed description of each record, including the cause of the failure, the scope of impact, and the repair measures.

[0055] Step S153: if the major fault detection result data shows that there is no major fault record, the robot service life data set is used as the final effective service life data set; if the major fault detection result data shows that there is a major fault record, the major fault robot ID is extracted from the fault record data set to obtain a fault robot ID data set, wherein at least one or more fault robot ID data exists in the fault robot ID data set;

[0056] Specifically, check the major fault detection result dataset and count whether there are major fault records. If the test results show that there are no major fault records, directly use the previously collected robot service life dataset as the final effective service life dataset. If the test results show that there are major fault records, further processing is required. Use data processing tools (such as Python or Excel) to extract the robot ID in the major fault record to form a faulty robot ID dataset. Ensure that there is at least one or more faulty robot IDs in the faulty robot ID dataset.

[0057] Step S154: extracting and removing corresponding life data from the robot service life data set according to the faulty robot ID data set, to obtain a major fault service life data set and an initial effective service life data set, wherein the major fault service life data set contains at least one or more major fault service life data;

[0058] Specifically, extract the faulty robot ID dataset and the robot service life dataset from the database. Then, use data processing tools (such as the Pandas library or Excel in Python) to filter out the corresponding life data in the robot service life dataset based on the faulty robot ID dataset. These data will be marked as major fault service life data to form a major fault service life dataset. Then, these major fault service life data are removed from the robot service life dataset, and the remaining part is used as the initial valid service life dataset. Ensure that there is at least one or more major fault service life data in the major fault service life dataset.

[0059] Step S155: Calculate the average life of the initial effective service life data set to obtain average service life data;

[0060] Specifically, you can use data processing tools (such as the Pandas library in Python or Excel) to add up all robot service life values ​​and divide them by the total number of robots to get the average service life data.

[0061] Step S156: performing deviation calculation on the major failure service life data set and the average service life data to obtain a service life deviation data set, wherein the service life deviation data set contains at least one or more service life deviation data;

[0062] Specifically, data processing tools (such as the Pandas library in Python or Excel) can be used to compare the service life of each robot with major failures with the average service life data, calculate the difference, and form a life deviation data set. Ensure that there is at least one or more life deviation data in the life deviation data set.

[0063] Step S157: discarding the major fault service life data corresponding to the life deviation data set that is greater than or equal to the preset life deviation threshold, and storing the major fault service life data corresponding to the life deviation data set that is less than the preset life deviation threshold in the life deviation data set into the initial effective service life data set, to obtain the final effective service life data set;

[0064] Specifically, a life deviation threshold value, such as 10%, can be set. Use data processing tools (such as the Pandas library in Python or Excel) to screen the life deviation data and find out the life deviation records that are greater than or equal to the threshold value. The major failure life data corresponding to these records are discarded. Then, the major failure life data corresponding to the records with life deviations less than the threshold value are added to the initial effective life data set, and the updated initial effective life data set is saved as the final effective life data set.

[0065] Step S158: dividing the final effective service life data set into use stages according to a preset stage division ratio to obtain use stage division data;

[0066] Specifically, the use phase division ratio can be defined, for example, 20% for the initial stage, 60% for the middle stage, and 20% for the late stage. Use data processing tools (such as Excel or the Pandas library in Python) to divide the service life data of each robot according to a preset ratio. The specific steps include multiplying the total service life of each robot by the ratio of different stages to determine the time range of each stage. Then, organize the division results into a structured data table to form the use phase division data.

[0067] Step S159: Divide the intelligent CNC machine tool into usage stages according to the usage stage division data and the robot usage time data table to obtain usage stage data.

[0068] Specifically, data processing tools (such as the Pandas library in Python or Excel) can be used to match the usage time data of each robot with its corresponding usage stage division data. The specific steps include determining which stage each robot is in, early, middle or late, based on its usage time. For example, if the usage time of a robot is within the first 20% of its total life, it is marked as early; within the range of 20%-80%, it is marked as middle; within the last 20%, it is marked as late. Then, the matched usage stage data is organized into a structured data table to obtain the usage stage data.

[0069] The present invention eliminates interference factors by eliminating abnormal data and calculating the average life span, making the assessment of the robot's service life more accurate. The quantitative analysis of life deviation further identifies the key factors that affect the life span, and the reliability of life prediction is improved by setting the life deviation threshold for data screening and optimization. The preset stage division ratio realizes the effective segmentation of life data, providing refined management for the maintenance and performance monitoring of the robot at different use stages. Combined with the dynamic identification of the use stage of the use time data table, the maintenance strategy can be flexibly adjusted to achieve customization of the maintenance strategy, reduce maintenance costs, and extend the working life of the robot. Accurate use stage and life data support the construction of the prediction model and enhance the ability to predict future power consumption and performance changes.

[0070] Preferably, step S17 comprises the following steps:

[0071] Step S171: extracting task information from each operation record in the stage historical operation record data set to obtain a historical task data set;

[0072] Specifically, data processing tools (such as the Pandas library in Python or Excel) can be used to parse each operation record in the stage historical operation record data set, extract task-related information, such as task ID, task type, start time, end time, and task description, and organize the extracted task information into a structured data table to form a historical task data set.

[0073] Step S172: extracting task features from the historical task data set to obtain a task load data set and a task moving path data set;

[0074] Specifically, data processing tools (such as the Pandas library or Excel in Python) can be used to extract features from each task record in the historical task dataset. For the task load data set, the load information in the task description is extracted to record the load of the robot in each task; for the task movement path data set, the location information in the task is parsed to record the movement path of the robot during the task execution. Geographic Information System (GIS) tools or path analysis algorithms can be used to process and record the movement path. Then, the extracted load and movement path data are organized into two independent data tables, forming the task load data set and the task movement path data set respectively.

[0075] Step S173: Segmenting the task load data set according to a preset load segmentation threshold to obtain load segmentation data;

[0076] Specifically, the thresholds for weight segmentation can be defined, such as light weight (0-10 kg), medium weight (10-50 kg), and heavy weight (over 50 kg). Extract the task weight data set from the database, which includes the weight information of each task. Then, use data processing tools (such as the Pandas library in Python or Excel) to segment each weight data and divide the weight data into different levels according to the preset weight threshold. For example, if the weight of a task is 15 kg, it is classified as medium weight. Then, organize the segmentation results into a new data table to form weight segmentation data.

[0077] Step S174: performing path clustering on the task moving path data set to obtain task path classification data;

[0078] Specifically, a suitable clustering algorithm (such as K-means, DBSCAN or hierarchical clustering) can be selected to perform cluster analysis on each mobile path data in the task mobile path data set, and similar paths can be classified into the same category. The specific steps include calculating the feature vector of each path (such as path length, number of turning points, average speed), and clustering these feature vectors using the selected clustering algorithm. Finally, the clustering results are organized into a structured data table to form task path classification data, and the classification results of each path are recorded.

[0079] Step S175: dividing the stage historical operation record data set into load segments according to the load segment data to obtain the load segment historical operation record data set;

[0080] Specifically, data processing tools (such as the Pandas library in Python or Excel) can be used to match the load segmentation data with the stage historical operation record dataset. According to the task ID or other unique identifier of each operation record, the corresponding load segmentation information is added to the operation record. Then, according to the load segmentation information, the operation record is divided into different load segmentation groups, such as light load group, medium load group and heavy load group. The segmentation results are organized into a structured data table to form a load segmentation historical operation record dataset.

[0081] Step S176: dividing the load segment historical operation record data set into the same type of paths according to the task path classification data to obtain a load path division record data set;

[0082] Specifically, data processing tools (such as the Pandas library in Python or Excel) can be used to match the path classification data with the load segmentation historical operation record dataset. According to the task ID or other unique identifier of each operation record, the corresponding path classification information is added to the operation record. Then, according to the path classification information, the load segmentation historical operation record dataset is divided into different path groups, such as path group A, path group B, and path group C. The division results are organized into a structured data table to form a load path division record dataset.

[0083] Step S177: embed the power consumption mode into the load path division record data set to obtain the power consumption mode embedded data, and perform power consumption mode identification on the power consumption mode embedded data to obtain the power consumption mode data. When several stages of historical operation record data sets are traversed, the power consumption mode data set is obtained.

[0084] Specifically, each record is extracted from the load path partition record dataset, such as operating status, load information, and path classification details. Subsequently, real-time or historical data related to power consumption, such as current sensor and voltage data, is obtained, and the data is preprocessed and aligned using data processing tools (e.g., the Pandas library in Python). Next, feature extraction techniques are used to extract key features, such as spectral features and time domain features, from the preprocessed current and voltage data to reflect the power consumption characteristics of the robot under different tasks and paths. Finally, the extracted power consumption features are embedded into each load path partition record through principal component analysis (PCA) or spectral analysis methods to form a power consumption pattern embedded dataset.

[0085] The present invention provides rich raw data for feature analysis and pattern recognition through comprehensive task information extraction, which enhances the understanding of the robot's energy consumption behavior. Secondly, the precise extraction of task features, especially the key features of load and movement path, lays the foundation for building a segmented power consumption model. The meticulous segmentation and classification processing further enhances the understanding of the diversity of energy consumption patterns, and provides structured data support for model construction through targeted data set division. The refined embedding technology of power consumption patterns combines abstract energy consumption behaviors with specific task characteristics, significantly improving the accuracy of pattern recognition. In addition, the efficient pattern recognition process provides a reliable data set for building a prediction model, while enhancing the generalization ability of the model, improving the prediction accuracy, and helping to achieve accurate energy consumption prediction and management.

[0086] Preferably, step S2 comprises the following steps:

[0087] Step S21: assigning a corresponding power consumption mode label to each operation record in the integrated historical operation record data set according to the power consumption mode data set to obtain a power consumption mode label set;

[0088] Specifically, for each operation record in the integrated historical operation record data set, a predefined algorithm or rule (for example, based on feature matching or a machine learning model) can be used to match and determine its corresponding power consumption mode label. The matched power consumption mode label is assigned to each historical operation record data to form a power consumption mode label set.

[0089] Step S22: dividing the integrated historical operation record data set according to the power consumption mode label set to obtain a plurality of groups of historical operation record data sets, wherein a group of historical operation record data sets corresponds to a power consumption mode;

[0090] Specifically, the records in the integrated historical operation record data set can be grouped according to the power consumption mode label set. Each set of historical operation record data sets contains all operation records associated with a specific power consumption mode. For example, if there are three power consumption modes (high power consumption, medium power consumption, and low power consumption), three sets of historical operation record data sets will be generated, each containing all operation records associated with these modes.

[0091] Step S23: performing time-series segmentation on each group of historical operation record data sets according to a preset time window to obtain a plurality of historical operation segment data sets;

[0092] Specifically, the appropriate time window size can be set according to the analysis requirements and data characteristics. The time window can be a time period of fixed length, such as every hour, every day, or every working cycle. For each set of historical operation record data sets related to the power consumption mode, it is segmented according to the set time window. The data in each time window is regarded as a time series data segment. Several historical operation segment data sets are generated, each of which contains historical operation records within a specific time window. These segment data sets reflect the operation status and power consumption characteristics of the robot in different time periods.

[0093] Step S24: extracting motion semantic features from each historical running segment data set to obtain each motion semantic feature data set;

[0094] Specifically, appropriate tools and algorithms, such as machine learning libraries in Python (e.g., scikit-learn) or deep learning frameworks (e.g., TensorFlow, PyTorch), can be used to extract motion semantic features, such as speed change patterns, acceleration distribution, and motion trajectory shapes. For each record in each historical run segment dataset, the selected feature extraction algorithm is applied to convert the raw data into motion semantic feature data, thereby obtaining each motion semantic feature dataset. These features can capture the motion behaviors and characteristics of the robot under different work tasks.

[0095] Step S25: performing task clustering on the motion semantic feature data set to obtain a work task cluster;

[0096] Specifically, a motion semantic feature dataset can be used as input data. Each sample data represents the motion semantic feature of a historical running segment. According to the characteristics of the data and the clustering requirements, an appropriate clustering algorithm is selected, such as K-means clustering, hierarchical clustering, and DBSCAN. The selection of clustering algorithm depends on the distribution of the data and the clustering target. The selected clustering algorithm is used to cluster the motion semantic feature dataset. The algorithm will group the data points into multiple clusters based on the similarity of the features. Each cluster represents a work task with similar motion features, and finally a work task cluster set is obtained.

[0097] Step S26: Characterizing the task features of the work task clusters based on the motion semantic feature data set to obtain feature data of each work task.

[0098] Specifically, appropriate feature characterization methods can be selected according to the requirements of task features and analysis objectives, such as statistical features (such as mean, variance), frequency domain features, time series features, or more complex feature characterization methods (such as feature extraction of deep learning models). Characterize the tasks in each work task cluster. By performing statistical analysis on all motion semantic feature data in the cluster, representative features of each work task cluster can be obtained, and the obtained work task feature data can be integrated and stored to finally obtain the feature data of each work task.

[0099] By assigning power consumption mode labels to operation records, the present invention not only enhances the interpretability of the data set, but also enables the data to be directly associated with specific energy consumption characteristics. Secondly, the refined organization and time series segmentation of the data set provide structured data resources for time series analysis, which helps to capture the dynamic changes in power consumption. In addition, in-depth motion semantic feature mining and effective task clustering not only deepen the understanding of the robot's motion state, but also help to identify the energy consumption characteristics of different tasks, providing a basis for task-specific energy consumption optimization. By clearly characterizing task characteristics, the purposefulness and efficiency of feature selection are enhanced, thereby improving the accuracy of the prediction model.

[0100] Preferably, step S3 comprises the following steps:

[0101] Step S31: performing heterogeneous data fusion on the power consumption mode data set, several groups of historical operation record data sets and each work task feature data to obtain a fused robot operation feature data set;

[0102] Specifically, data integration tools or programming languages ​​(such as the Pandas library in Python) can be used to integrate and align power consumption pattern data sets from different sources, several groups of historical operation record data sets, and various work task feature data. Ensure that each piece of historical operation record data can be associated with its corresponding power consumption pattern label and work task feature data. According to the analysis requirements and predetermined goals, select appropriate power consumption pattern features, historical operation record features, and work task features for merging. Fuse the integrated and aligned power consumption pattern data, historical operation record data, and work task feature data to finally obtain a fused robot operation feature data set. This can be achieved through a database join operation or a data frame merge operation.

[0103] Step S32: extracting power-related features from the fusion robot operation feature data set to obtain a power-related feature data set;

[0104] Specifically, appropriate feature extraction methods and tools can be selected according to the goal of power-related feature extraction. These features include but are not limited to current, voltage, power factor, and energy consumption. Power-related features are extracted for each fusion robot operation feature data. This involves statistical analysis, frequency domain analysis, and time domain analysis. A variety of techniques and methods are used to extract key features representing power characteristics from the raw data. The extracted power-related feature data are integrated and stored to obtain a power-related feature data set.

[0105] Step S33: performing modal decomposition on the power-related feature data set to obtain a modal power feature subset group;

[0106] Specifically, appropriate modal decomposition methods can be selected according to the properties of the power-related feature data set and the objectives of the analysis. Such as wavelet transform, singular value decomposition (SVD), and empirical mode decomposition (EMD). The selected modal decomposition method is applied to the power-related feature data set. These methods can decompose complex power feature data into components of different frequencies or modes, revealing the hidden structures and patterns in the data. Based on the results of modal decomposition, a modal power feature subset group is generated. Each subset represents the power change characteristics under different frequencies or modes, reflecting the power consumption pattern of the robot under different operating conditions.

[0107] Step S34: performing modal coding on the modal power feature subset group to obtain a modal power feature vector set;

[0108] Specifically, a suitable modal encoding method can be selected according to the properties of the modal power feature subset and the needs of analysis. Such as neural network-based encoding methods and autoencoders. The selected encoding method is applied to each modal power feature subset group. These methods can encode high-dimensional modal power feature representations into low-dimensional feature vectors, retaining important information of the data while reducing the dimension of the data. The modal power feature vector set of each modal power feature subset group is obtained through the encoding process. Each feature vector is usually represented as a numerical vector, reflecting the abstract representation of the power feature under the corresponding mode.

[0109] Step S35: performing attention fusion on the modal power feature vector set to obtain the robot fusion power feature encoding;

[0110] Specifically, a suitable attention mechanism can be selected to fuse the modal power feature vector set. For example, the self-attention mechanism (Self-Attention) and the multi-head attention mechanism are used to capture the important relationships and dependencies between different modal power features. The selected attention mechanism is applied to fuse the modal power feature vector set. In the self-attention mechanism, the weight between each feature vector and other feature vectors can be calculated to obtain the weighted sum representation of each feature vector. The robot fused power feature encoding is obtained, that is, the feature vector set after fusion through the attention mechanism. These encodings usually have stronger representation capabilities and can more accurately reflect the overall characteristics of the robot under different power consumption modes.

[0111] Step S36: Adaptively predict the robot fusion power feature coding to obtain a robot power prediction model.

[0112] Specifically, a suitable prediction model can be selected according to the complexity and requirements of the prediction task. Such as recurrent neural network (RNN), long short-term memory network (LSTM), transformer, and their variants and combinations. The selected prediction model is applied to the robot fusion power feature encoding data for training. During the training process, the model optimizes parameters by learning the complex relationship between input features and prediction targets (such as future power consumption). The trained model is used to predict the new robot power feature encoding. The prediction process can be a single-step prediction or a multi-step prediction, depending on the specific application scenario and requirements. The prediction results are evaluated and optimized, usually including calculating the prediction error (such as mean square error, mean absolute error), and adjusting the model parameters or optimizing the model structure according to the evaluation results, and finally obtaining the robot power prediction model.

[0113] The present invention effectively integrates multi-source data through heterogeneous data fusion technology, constructs a comprehensive robot operation feature data set, and greatly enriches the in-depth analysis of the robot's energy consumption characteristics. The targeted feature extraction process focuses on energy consumption-related data, significantly improving the prediction accuracy of the model. The application of modal decomposition technology deepens the understanding of the potential structure of the data, while modal encoding further refines the feature representation and enhances the data understanding ability of the model. The introduction of the attention mechanism enables the model to focus on important features, which not only improves the prediction accuracy but also enhances the interpretability of the model. The adaptive prediction method gives the model the ability to dynamically respond to real-time data and enhances its adaptability to changing operating conditions. In addition, the prediction model exhibits strong generalization capabilities and can provide accurate power consumption predictions across tasks and environments.

[0114] Preferably, step S4 comprises the following steps:

[0115] Step S41: extracting the mechanism model from the integrated historical operation record data set to obtain a robot working mechanism model set;

[0116] Specifically, we can select appropriate mechanism modeling methods according to the specific working characteristics and requirements of the robot, such as physical modeling, data-driven modeling, and hybrid modeling. The selected modeling method is applied to the historical operation record data set. For example, for physical modeling, a mathematical model can be established based on the dynamic and kinematic characteristics of the robot; for data-driven modeling, machine learning techniques can be used to learn patterns and laws from historical data. According to the specific information in the data set, the mechanism model is parameterized, and finally the robot working mechanism model set is obtained.

[0117] Step S42: extracting the environment scene of the intelligent CNC machine tool to obtain a static environment scene data set and a dynamic environment scene data set;

[0118] Specifically, static environment scenes include fixed environmental factors of robot operation, such as the geographical location, building structure, and layout of the workplace. This information can be obtained through map data, building plans, or on-site measurements. Dynamic environment scenes involve environmental factors that change over time, such as temperature, humidity, and light changes. These data can be collected in real time by sensors (such as environmental sensors, weather stations) or historical records, and finally static environment scene datasets and dynamic environment scene datasets are obtained.

[0119] Step S43: constructing a virtual environment for the robot working mechanism model set using the static environment scene data set and the dynamic environment scene data set to obtain a robot working virtual environment;

[0120] Specifically, a static environment scene dataset and a dynamic environment scene dataset can be used to construct a virtual environment model for the robot to work in. This includes converting the geographical and structural information of the environment into a computer model, such as a scene model in a three-dimensional modeling software. The data of dynamic environmental factors (such as temperature and humidity) and static environmental factors (such as building structure) are integrated into the virtual environment model, and finally a virtual environment for the robot to work in is obtained.

[0121] Step S44: extracting the process flow from the integrated historical operation record data set to obtain a robot process flow data set;

[0122] Specifically, the operation sequence or workflow executed by each robot can be extracted from the integrated historical operation record data set. Each sequence includes a series of operation steps or events, which records the activity history of the robot. The extracted process flow data is formatted and sorted to obtain the robot process flow data set.

[0123] Step S45: Based on the robot working virtual environment and the robot process flow data set, the robot power prediction model is subjected to intensive training to obtain an optimized robot power prediction model.

[0124] Specifically, please refer to the sub-steps of step S45 for the specific processing process of this embodiment.

[0125] The present invention significantly improves the scientificity and practicality of power prediction of intelligent CNC machine tools by comprehensively applying advanced technical means such as mechanism model extraction, environmental scene simulation, virtual environment construction, process flow mapping and intensive training. On the technical level, it first deepens the understanding of the working mechanism of the robot and provides a solid theoretical support for prediction. Secondly, it comprehensively simulates different environmental scenes to enhance the adaptability and generalization ability of the model. The constructed virtual environment provides the model with simulation training conditions close to actual application, and the accurate mapping of the process flow ensures the practicality and operational guidance value of the prediction results. Through intensive training, the model continues to learn and evolve in the virtual environment, improving the prediction accuracy and stability. In addition, the present invention demonstrates the exploration and application of the cutting-edge technology, provides strong technical support for the application of intelligent CNC machine tools in intelligent manufacturing, and ensures the robustness of the system in the face of uncertainty and changes in the production environment.

[0126] Preferably, step S45 includes the following steps:

[0127] Step S451: constructing a virtual environment simulator based on the robot working virtual environment and the robot process flow data set to obtain a working virtual environment simulator;

[0128] Specifically, a virtual environment simulator can be designed and developed based on the robot's working virtual environment to simulate the robot's working environment under different process flows. The simulator can include a simulation engine, a physics engine, and an environmental interaction module to achieve accurate simulation of robot operations and environmental impacts. The robot process data set is integrated with the virtual environment simulator to ensure that the simulator can generate realistic working scenes according to the specific process flow. Finally, a working virtual environment simulator is obtained.

[0129] Step S452: Initializing the neural strategy network of the robot power prediction model to obtain an initialized strategy network model;

[0130] Specifically, a neural network model suitable for the robot power prediction task can be selected, such as a deep neural network (DNN), a convolutional neural network (CNN), or a recurrent neural network (RNN). Initialize the neural network structure and initial parameters according to the task requirements and data characteristics. Use the robot process flow data set as training data. These data include but are not limited to the operation steps, timestamps, and key information on power consumption in the process flow. Use the initialized neural network model to train the robot process flow data set. During the training process, the model learns the power consumption patterns and prediction capabilities under different process flows. Finally, the initialization strategy network model is obtained.

[0131] Step S453: designing a power efficiency reward function for the initialized strategy network model according to the robot process flow data set to obtain a strategy network reward function model;

[0132] Specifically, a suitable reward function can be designed based on the power consumption pattern in the robot process data set and the prediction ability of the policy network to evaluate the power efficiency of the robot under a specific process. The reward function can be based on the error between the actual value and the predicted value of power consumption, the fluctuation of power consumption, or other indicators related to power efficiency. Use mathematical modeling tools or programming languages ​​(such as Python) to implement the designed reward function model. Ensure that the reward function can accurately reflect the power performance of the robot under different process flows and can be effectively used in subsequent simulation and training processes.

[0133] Step S454: performing a working simulation of the intelligent CNC machine tool based on the working virtual environment simulator, and performing policy gradient sampling on the current robot state in the virtual simulation environment simulator according to the policy network reward function model to obtain a policy gradient sampling data set;

[0134] Specifically, the policy network model initialized in step S452 can be loaded into the working virtual environment simulator. Ensure that the policy network can effectively interact with the simulation environment simulator and can output corresponding operation policies according to the environment state. According to the policy network reward function model, the state of the robot in the current virtual simulation environment is sampled by policy gradient. This includes executing robot actions according to the output of the current policy network and evaluating the effects of these actions according to the reward function. During the simulation process, a policy gradient sampling data set is collected, including robot state, operation actions, environmental feedback, and corresponding reward values, and finally a policy gradient sampling data set is obtained.

[0135] Step S455: storing the policy gradient sampling data set into a preset experience replay pool, and performing data batch sampling from the preset experience replay pool to obtain a policy optimization training data set, and optimizing the policy network parameters of the policy network reward function model using the policy optimization training data set to obtain an optimized policy network model;

[0136] Specifically, a preset experience replay pool can be created, which usually uses an efficient data structure (such as a circular buffer) to store the policy gradient sampling data set. This pool allows a large amount of historical sampled data to be efficiently stored and managed during training. The policy gradient sampling data set is stored in the preset experience replay pool. Data batches are randomly or according to a certain strategy extracted from the preset experience replay pool. These data batches are used to train the policy network to avoid overfitting and improve training efficiency. The policy network parameters are optimized using data batches. This includes loading the policy network model using a deep learning framework (such as TensorFlow or PyTorch), defining an optimizer (such as Adam optimizer), and calculating gradients and updating model parameters based on the sampled data batches to maximize the reward function performance of the policy network. The performance indicators of the policy network, such as the average value of the reward function, are monitored during the training process. Through repeated iterative optimization, a policy network model that performs well in the virtual environment simulator is obtained, and finally an optimized policy network model is obtained.

[0137] Step S456: Migrate and fine-tune the model parameters of the robot power prediction model according to the optimization strategy network model to obtain an optimized robot power prediction model.

[0138] Specifically, the optimization strategy network model can be loaded into the working virtual environment simulator. The robot power prediction model is loaded in the simulation environment. According to the optimization strategy network model, some parameters of the power prediction model are adjusted or migrated. This step is completed by passing the output of the strategy network as the input of the power prediction model or performing joint training. Fine-tune and verify the model in the working virtual environment simulator. This includes using historical operation record data sets for simulation testing to evaluate the power prediction accuracy and effect of the optimization model under different process flows. Based on the simulation results and the accuracy of the power prediction, the model is evaluated for performance and necessary adjustments and optimizations are performed to finally obtain the optimized robot power prediction model.

[0139] The present invention provides a training platform close to actual application for the policy network by constructing a highly simulated working virtual environment simulator, thereby ensuring the authenticity of the simulation environment. The initialization of the neural policy network sets the starting point for the learning and optimization of the model, while the customized power efficiency reward function design clarifies the optimization goal of the model and guides the model to evolve in the direction of improving power efficiency. By performing work simulation and policy gradient sampling in a virtual environment, the model learns the optimal behavior strategy under different states. The policy gradient sampling data storage and batch sampling of the experience replay pool enable the model to learn and optimize the strategy from historical data, thereby improving the stability and efficiency of learning. The continuous optimization of the policy network and the migration and fine-tuning of the model parameters further improve the performance of the power prediction model, so that the model can provide high-precision prediction results under various working conditions. The present invention enhances the robustness of the system in the face of uncertainty and dynamic changes in the production process. In addition, accurate power prediction provides reliable data support for energy management and production scheduling, which helps enterprises to achieve the optimization of energy use and the intelligence of job scheduling, significantly improving production efficiency and reducing energy consumption costs.

[0140] Preferably, step S5 comprises the following steps:

[0141] Step S51: Based on the preset federated learning framework, the optimized robot power prediction model is distributed to multiple edge node devices of the target factory through a preset model distribution mechanism, and a localized model instance is constructed on each edge node device to obtain a distributed robot power prediction model set;

[0142] Specifically, a suitable federated learning framework can be selected and configured, such as Google's FederatedLearning framework or OpenMined's PySyft. The optimized robot power prediction model is uploaded to the federated learning server. Then, the optimized robot power prediction model is distributed to multiple edge node devices in the target factory through a preset model distribution mechanism (such as HTTP or MQTT protocol). After each edge node device receives the model, it builds a model instance locally and initializes relevant parameters and environment. During the model distribution process, a secure transmission protocol (such as SSL / TLS) is used to ensure the security of data transmission. After completing the model distribution and localization construction, a distributed robot power prediction model set is obtained, covering all edge node devices in the target factory.

[0143] Step S52: collecting real-time task data of the intelligent CNC machine tool to obtain a real-time task data stream of the robot;

[0144] Specifically, a task scheduling module can be integrated into the intelligent CNC machine tool control system, which is responsible for issuing and managing the robot's work tasks. Work tasks are generated and assigned in real time through a task scheduling system (such as an MES system or an ERP system). These task information include but are not limited to task ID, task type, task priority, start time, end time, and task parameters. The task scheduling system is connected to an edge computing platform (such as NVIDIA Jetson or Raspberry Pi) to ensure that task information can be transmitted to the edge node device in real time. Whenever a new task is issued, the edge node device immediately receives and records the task data. In order to ensure the timeliness and accuracy of the data, a message queue (such as Kafka or RabbitMQ) can be used to realize real-time transmission and processing of task data. The data processing module on the edge node device parses and stores the received task data to form a continuous real-time task data stream for the robot.

[0145] Step S53: Slice and split the robot real-time task data stream according to the preset edge node task allocation strategy to obtain a sliced ​​robot real-time task data set;

[0146] Specifically, the edge node task allocation strategy can be determined, and the strategy can be configured based on the task type, task priority, and time window parameters. For example, data sharding can be performed according to the time window of the task, and the task data in each time window can be used as a data slice. Use a data sharding tool (such as Apache Kafka Streams or Apache Flink) to split the real-time task data stream. After the task data stream enters the sharding tool, it is sharded according to the preset time window (such as every minute or every hour) to generate multiple sharded task data sets. Each sharded task data set contains all task information within the time window, including but not limited to task ID, task type, start time, end time, and task parameters. After the sharding is completed, the sharded task data set is stored in the local storage or distributed storage system (such as HDFS or Cassandra) of the edge computing platform, and finally a sharded robot real-time task data set is obtained.

[0147] Step S54: Distribute the sharded robot real-time task data set to each robot power prediction model in the distributed robot power prediction model set according to the preset task distribution rules, and use the robot power prediction model to perform power prediction calculations on the intelligent CNC machine tool by combining the corresponding sharded robot real-time task data to obtain the robot predicted power data set.

[0148] Specifically, a preset task distribution rule can be determined, which can be configured based on the computing power and load balancing strategy of the edge node. For example, a polling algorithm, a minimum load priority algorithm, or a task priority distribution algorithm can be used to distribute tasks. Using the task scheduling module of a task distribution tool (such as Apache NiFi or Kubernetes), the sharded task data set is distributed to each edge node device in the distributed robot power prediction model set one by one according to the preset task distribution rule. After each edge node device receives the sharded task data set, it inputs the task data into the localized robot power prediction model. The robot power prediction model performs power prediction calculations by analyzing task parameters (such as task type, load, path) and combining historical power consumption patterns. The prediction model is run on the edge node using a machine learning framework (such as TensorFlow or PyTorch) to calculate the predicted power data for each task. After the prediction calculation is completed, the predicted power data of all sharded tasks are summarized and stored in the local database of the edge node or sent to the central server to form a complete robot prediction power data set.

[0149] The distributed model deployment achieved by the present invention through the federated learning framework effectively sinks the computing power to the edge nodes, optimizes resource allocation, and at the same time, the localized model instances built on the edge nodes reduce data transmission delays and improve the real-time performance and efficiency of the model. The collection and processing of real-time task data enhances the accuracy and practicality of the prediction, while the intelligent distribution mechanism ensures the efficient management of task data. Parallelized prediction calculations greatly improve the processing speed, and the continuous optimization and personalized adaptation of distributed models enhance the generalization ability and stability of the model. In addition, refined energy management benefits from the accuracy of real-time power prediction, and the fast feedback of real-time prediction results improves the flexibility of production scheduling. The robustness of the system benefits from the fault tolerance of the distributed architecture, and even if some nodes fail, it does not affect the overall performance.

[0150] Preferably, the present invention further provides an intelligent CNC machine tool power prediction system for executing the intelligent CNC machine tool power prediction method as described above, the intelligent CNC machine tool power prediction system comprising:

[0151] The power consumption pattern recognition module is used to collect historical operation records of the intelligent CNC machine tool to obtain an integrated historical operation record data set; based on the integrated historical operation record data set, the power consumption pattern of the intelligent CNC machine tool is recognized to obtain a power consumption pattern data set;

[0152] The work task feature mining module is used to divide the integrated historical operation record data set according to the power consumption mode data set to obtain several groups of historical operation record data sets; perform work task feature mining on each group of historical operation record data sets to obtain each work task feature data;

[0153] A prediction model building module is used to build an adaptive model for the intelligent CNC machine tool based on a power consumption mode data set, several sets of historical operation record data sets and characteristic data of each work task, so as to obtain a robot power prediction model;

[0154] The model reinforcement training module is used to construct a working virtual environment for the intelligent CNC machine tool according to the integrated historical operation record data set to obtain a robot working virtual environment; based on the robot working virtual environment, the robot power prediction model is reinforced and trained to obtain an optimized robot power prediction model;

[0155] The power prediction module is used to deploy the optimized robot power prediction model to several edge devices in the target factory based on the preset federated learning framework to obtain a distributed robot power prediction model set; obtain the robot's real-time task data stream; and perform power prediction on the intelligent CNC machine tool based on the distributed robot power prediction model set according to the robot's actual task data to obtain a robot prediction power data set.

[0156] In the present invention, through the in-depth analysis of historical operation data by the power consumption pattern recognition module, the power consumption pattern can be accurately identified, providing reliable basic data for subsequent predictions. The work task feature mining module can extract key features from historical data, which are crucial for understanding the energy consumption behavior of the robot under different tasks. The prediction model construction module uses the extracted feature data to build an adaptive model, so that the prediction model can automatically adjust according to different tasks and conditions, thereby improving the flexibility and accuracy of the prediction. The model reinforcement training module trains the model by building a virtual environment, enhances the model's adaptability to actual work scenarios, and optimizes the prediction results. The power prediction module adopts a federated learning framework to realize the distributed deployment of the model on edge devices, reduce the pressure of centralized computing, and improve the scalability and response speed of the system. By acquiring and processing the robot's task data stream in real time, it provides instant power prediction and provides real-time data support for production scheduling and energy management.

[0157] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0158] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for predicting power of intelligent CNC machine tools, characterized in that: The following steps are involved: Step S1: collecting historical operation records of the intelligent CNC machine tool to obtain an integrated historical operation record data set; Based on the integrated historical operation record data set, the power consumption pattern of the intelligent CNC machine tool is identified to obtain a power consumption pattern data set; step S1 includes the following steps: Step S11: collecting historical operation records of the intelligent CNC machine tool to obtain an integrated historical operation record data set; Step S12: extracting the device ID from the integrated historical operation record data set to obtain a robot device ID data set; Step S13: performing usage time statistics on the integrated historical operation record data set to obtain a robot usage time data table; Step S14: collecting the service life of retired intelligent CNC machine tools to obtain a robot service life data set; Step S15: Divide the intelligent CNC machine tool into use stages according to the robot service life data set and the robot use time data table to obtain use stage data; Step S15 includes the following steps: Step S151: collecting historical fault records of retired intelligent CNC machine tools to obtain a fault record data set; Step S152: performing major fault detection on the fault record data set to obtain major fault detection result data; Step S153: if the major fault detection result data shows that there is no major fault record, the robot service life data set is used as the final effective service life data set; if the major fault detection result data shows that there is a major fault record, the major fault robot ID is extracted from the fault record data set to obtain a fault robot ID data set, wherein at least one or more fault robot ID data exists in the fault robot ID data set; Step S154: extracting and removing corresponding life data from the robot service life data set according to the faulty robot ID data set, to obtain a major fault service life data set and an initial effective service life data set, wherein the major fault service life data set contains at least one or more major fault service life data; Step S155: Calculate the average life of the initial effective service life data set to obtain average service life data; Step S156: performing deviation calculation on the major failure service life data set and the average service life data to obtain a service life deviation data set, wherein the service life deviation data set contains at least one or more service life deviation data; Step S157: discarding the major fault service life data corresponding to the life deviation data set that is greater than or equal to the preset life deviation threshold, and storing the major fault service life data corresponding to the life deviation data set that is less than the preset life deviation threshold in the life deviation data set into the initial effective service life data set, to obtain the final effective service life data set; Step S158: dividing the final effective service life data set into use stages according to a preset stage division ratio to obtain use stage division data; Step S159: Divide the intelligent CNC machine tool into use stages according to the use stage division data and the robot use time data table to obtain use stage data; Step S16: Based on the robot device ID data set and the robot usage time data table, the integrated historical operation record data set is labeled and divided according to the usage stage data to obtain several stage historical operation record data sets; Step S17: Based on the historical operation record data sets of several stages, the power consumption pattern of the intelligent CNC machine tool is identified to obtain a power consumption pattern data set; Step S17 includes the following steps: Step S171: extracting task information from each operation record in the stage historical operation record data set to obtain a historical task data set; Step S172: extracting task features from the historical task data set to obtain a task load data set and a task moving path data set; Step S173: Segmenting the task load data set according to a preset load segmentation threshold to obtain load segmentation data; Step S174: performing path clustering on the task moving path data set to obtain task path classification data; Step S175: dividing the stage historical operation record data set into load segments according to the load segment data to obtain the load segment historical operation record data set; Step S176: dividing the load segment historical operation record data set into the same type of paths according to the task path classification data to obtain a load path division record data set; Step S177: embedding the power consumption mode into the load path division record data set to obtain the power consumption mode embedded data, and performing power consumption mode recognition on the power consumption mode embedded data to obtain the power consumption mode data. When traversing through several stages of historical operation record data sets, the power consumption mode data set is obtained. Step S2: dividing the integrated historical operation record data set according to the power consumption mode data set to obtain several groups of historical operation record data sets; mining the work task characteristics of each group of historical operation record data sets to obtain each work task characteristic data; Step S3: constructing an adaptive model for the intelligent CNC machine tool based on the power consumption mode data set, several groups of historical operation record data sets and characteristic data of each work task to obtain a robot power prediction model; Step S4: constructing a working virtual environment for the intelligent CNC machine tool according to the integrated historical operation record data set to obtain a robot working virtual environment; performing intensive training on the robot power prediction model based on the robot working virtual environment to obtain an optimized robot power prediction model; Step S4 includes the following steps: Step S41: extracting the mechanism model from the integrated historical operation record data set to obtain a robot working mechanism model set; Step S42: extracting the environment scene of the intelligent CNC machine tool to obtain a static environment scene data set and a dynamic environment scene data set; Step S43: constructing a virtual environment for the robot working mechanism model set using the static environment scene data set and the dynamic environment scene data set to obtain a robot working virtual environment; Step S44: extracting the process flow from the integrated historical operation record data set to obtain a robot process flow data set; Step S45: Based on the robot working virtual environment and the robot process flow data set, the robot power prediction model is reinforced trained to obtain an optimized robot power prediction model; Step S45 includes the following steps: Step S451: constructing a virtual environment simulator based on the robot working virtual environment and the robot process flow data set to obtain a working virtual environment simulator; Step S452: Initializing the neural strategy network of the robot power prediction model to obtain an initialized strategy network model; Step S453: designing a power efficiency reward function for the initialized strategy network model according to the robot process flow data set to obtain a strategy network reward function model; Step S454: performing a working simulation of the intelligent CNC machine tool based on the working virtual environment simulator, and performing policy gradient sampling on the current robot state in the virtual simulation environment simulator according to the policy network reward function model to obtain a policy gradient sampling data set; Step S455: storing the policy gradient sampling data set into a preset experience replay pool, and performing data batch sampling from the preset experience replay pool to obtain a policy optimization training data set, and optimizing the policy network parameters of the policy network reward function model using the policy optimization training data set to obtain an optimized policy network model; Step S456: performing model parameter migration and fine-tuning on the robot power prediction model according to the optimization strategy network model to obtain an optimized robot power prediction model; Step S5: Based on the preset federated learning framework, the optimized robot power prediction model is deployed to several edge devices of the target factory to obtain a distributed robot power prediction model set; the real-time task data stream of the robot is obtained; based on the distributed robot power prediction model set, the power of the intelligent CNC machine tool is predicted according to the actual task data of the robot to obtain a robot prediction power data set.

2. The power prediction method for intelligent CNC machine tools according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: assigning a corresponding power consumption mode label to each operation record in the integrated historical operation record data set according to the power consumption mode data set to obtain a power consumption mode label set; Step S22: dividing the integrated historical operation record data set according to the power consumption mode label set to obtain a plurality of groups of historical operation record data sets, wherein a group of historical operation record data sets corresponds to a power consumption mode; Step S23: performing time-series segmentation on each group of historical operation record data sets according to a preset time window to obtain a plurality of historical operation segment data sets; Step S24: extracting motion semantic features from each historical running segment data set to obtain each motion semantic feature data set; Step S25: performing task clustering on the motion semantic feature data set to obtain a work task cluster; Step S26: Characterizing the task features of the work task clusters based on the motion semantic feature data set to obtain feature data of each work task.

3. The power prediction method for intelligent CNC machine tools according to claim 1 is characterized in that: Step S5 includes the following steps: Step S51: Based on the preset federated learning framework, the optimized robot power prediction model is distributed to multiple edge node devices of the target factory through a preset model distribution mechanism, and a localized model instance is constructed on each edge node device to obtain a distributed robot power prediction model set; Step S52: collecting real-time task data of the intelligent CNC machine tool to obtain a real-time task data stream of the robot; Step S53: Slice and split the robot real-time task data stream according to the preset edge node task allocation strategy to obtain a sliced ​​robot real-time task data set; Step S54: Distribute the sharded robot real-time task data set to each robot power prediction model in the distributed robot power prediction model set according to the preset task distribution rules, and use the robot power prediction model to perform power prediction calculations on the intelligent CNC machine tool by combining the corresponding sharded robot real-time task data to obtain the robot predicted power data set.

4. An intelligent CNC machine tool power prediction system, characterized in that: Used to execute the intelligent CNC machine tool power prediction method as claimed in claim 1, the intelligent CNC machine tool power prediction system comprises: The power consumption pattern recognition module is used to collect historical operation records of the intelligent CNC machine tool to obtain an integrated historical operation record data set; based on the integrated historical operation record data set, the power consumption pattern of the intelligent CNC machine tool is recognized to obtain a power consumption pattern data set; The work task feature mining module is used to divide the integrated historical operation record data set according to the power consumption mode data set to obtain several groups of historical operation record data sets; perform work task feature mining on each group of historical operation record data sets to obtain each work task feature data; A prediction model building module is used to build an adaptive model for the intelligent CNC machine tool based on a power consumption mode data set, several sets of historical operation record data sets and characteristic data of each work task, so as to obtain a robot power prediction model; The model reinforcement training module is used to construct a working virtual environment for the intelligent CNC machine tool according to the integrated historical operation record data set to obtain a robot working virtual environment; based on the robot working virtual environment, the robot power prediction model is reinforced and trained to obtain an optimized robot power prediction model; The power prediction module is used to deploy the optimized robot power prediction model to several edge devices in the target factory based on the preset federated learning framework to obtain a distributed robot power prediction model set; obtain the robot's real-time task data stream; and perform power prediction on the intelligent CNC machine tool based on the distributed robot power prediction model set according to the robot's actual task data to obtain a robot prediction power data set.

Images