A tool identification method and identification device

Abstract

The present invention discloses a tool identification method and identification device, the method comprising: obtaining a digital twin model of the tools in the tool magazine; based on the digital twin model, calculating the similarity matrix between the tools, and generating a tool association network topology structure according to a preset similarity threshold; in the process of the tools entering and exiting the tool magazine, collecting tool status data, and fusing it with the tool history usage data to form a tool comprehensive evaluation data set; based on the tool comprehensive evaluation data set and the tool association network topology structure, calculating the tool health index through a preset performance evaluation algorithm; according to the tool health index and the current processing task requirements, generating a tool usage plan and a preventive maintenance plan through an intelligent scheduling algorithm. The technical solution of the present invention can accurately evaluate tool performance, dynamically monitor tool health status, and optimize tool use and maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of tool management of tool magazines, and in particular to a tool identification method and an identification device. Background Art

[0002] Tool magazines are an important part of modern CNC machine tools, used to store and manage the various tools required in the machine tool processing process. Tool magazines are usually equipped with robotic arms or rotating structures, and use automated mechanisms to quickly replace tools to meet the needs of complex parts processing for diversified tools. Each tool has a unique storage location in the tool magazine and is called in sequence according to the processing steps. The efficient operation of the tool magazine is directly related to the stability of the production process and the processing quality. However, due to the complexity of the processing environment, the performance of the tool may change dynamically over time, which poses new technical challenges to traditional tool magazine management.

[0003] At present, the tool management method of the tool magazine mainly relies on regular offline inspection and manual replacement strategies. Operators usually stop the machine to check the wear of the tool or replace the tool at preset time intervals to ensure processing accuracy. Although this fixed-cycle management method is simple, it lacks the ability to evaluate the dynamic changes of tool performance in real time, and it is difficult to effectively capture minor performance degradation problems caused by cutting force, coolant corrosion, material reaction, etc., resulting in the risk of sudden tool failure during processing. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problem that the existing tool magazine tool management method lacks the ability to evaluate the dynamic changes of tool performance in real time.

[0005] A first aspect of the present invention provides a tool identification method, the tool identification method comprising:

[0006] Acquire a digital twin model of a tool in a tool magazine, wherein the digital twin model includes geometric feature parameters, material property information, and processing task data of the tool;

[0007] Based on the geometric feature parameters, material property information and processing task data in the digital twin model, a similarity matrix between tools is calculated, and a tool association network topology structure is generated according to a preset similarity threshold;

[0008] When tools are in and out of the tool magazine, tool status data is collected and integrated with tool history usage data to form a comprehensive tool evaluation data set.

[0009] Based on the tool comprehensive evaluation data set and the tool associated network topology structure, calculating the tool health index by a preset performance evaluation algorithm;

[0010] According to the tool health index and the current processing task requirements, a tool usage plan and a preventive maintenance plan are generated through an intelligent scheduling algorithm.

[0011] Optionally, the step of obtaining a digital twin model of a tool in a tool magazine, wherein the digital twin model includes geometric feature parameters, material property information, and processing task data of the tool, includes:

[0012] Identify the tools in the tool magazine and obtain the unique identification code of the tools;

[0013] querying a tool information database according to the unique identification code;

[0014] If corresponding complete information is found in the tool information database, the corresponding complete information includes geometric feature parameters, material property information and processing task data of the tool, and a digital twin model is constructed based on the corresponding complete information;

[0015] If the corresponding complete information is not found, perform the following steps:

[0016] Perform three-dimensional scanning on the tool to obtain the geometric profile data of the tool;

[0017] Based on the geometric profile data, combined with a preset tool parameter template, the geometric feature parameters of the tool are generated;

[0018] Obtain the material composition data of the tool and determine the material property information of the tool based on the material performance database;

[0019] Access the production management system, obtain the historical processing task records associated with the tool, and extract the processing task data;

[0020] The geometric feature parameters, material property information and processing task data are integrated to construct a digital twin model.

[0021] Optionally, the building of the digital twin model includes:

[0022] Performing parameterization processing on the geometric feature parameters to obtain a high-precision three-dimensional geometric model of the tool;

[0023] Mapping the material property information onto the high-precision three-dimensional geometric model through a multi-scale material model to obtain a tool virtual entity with microstructure information;

[0024] Performing deep learning analysis on the processing task data to obtain the evolution law of tool performance;

[0025] Associating the tool performance evolution law with the tool virtual entity to obtain a dynamic performance prediction model;

[0026] Constructing a two-way data interaction interface between the virtual entity of the tool and the actual tool to achieve real-time status synchronization;

[0027] The high-precision three-dimensional geometric model, the tool virtual entity, the dynamic performance prediction model and the two-way data interaction interface are integrated to obtain a digital twin model of the tool.

[0028] Optionally, the step of constructing a bidirectional data interaction interface between the virtual tool entity and the actual tool to achieve real-time state synchronization includes:

[0029] Design tool data exchange protocol to define the data structure and communication format between tool virtual entity and actual tool;

[0030] Based on the tool data exchange protocol, the actual tool working parameters are collected and processed to generate a standardized tool status data packet;

[0031] Using the tool status data packet to update corresponding parameters in the virtual entity, and generating a control instruction data packet according to an optimization algorithm of the virtual entity;

[0032] The control instruction data packet is transmitted to the actual tool through the tool data exchange protocol to complete two-way data interaction and status synchronization.

[0033] Optionally, the method of calculating the similarity matrix between tools based on the geometric feature parameters, material property information and processing task data in the digital twin model, and generating a tool association network topology structure according to a preset similarity threshold, includes:

[0034] Performing feature vectorization processing on geometric feature parameters, material property information and processing task data in the digital twin model to obtain a tool feature vector set;

[0035] Using a multi-dimensional similarity calculation algorithm, the vectors in the tool feature vector set are compared pairwise to obtain a similarity matrix between tools;

[0036] According to the preset similarity threshold, binarizing the inter-tool similarity matrix to obtain a tool association adjacency matrix;

[0037] Based on the tool association adjacency matrix, a tool association network topology structure is constructed by using a graph theory algorithm.

[0038] Optionally, during the process of the tool entering and exiting the tool magazine, tool status data is collected and merged with the tool history usage data to form a tool comprehensive evaluation data set, including:

[0039] When the tool enters and exits the tool magazine, the tool status data is obtained, wherein the tool status data includes a tool geometry profile change diagram, a surface microstructure diagram, and a tool-handle interface stress distribution diagram;

[0040] Extracting tool history usage data of tools entering and leaving the tool magazine from the production management system, wherein the tool history usage data includes a processing process parameter sequence and a processing quality index record;

[0041] Performing cross-modal feature extraction on the tool state data to obtain a tool state feature vector;

[0042] Using a time series prediction algorithm, generating tool future usage trend prediction data based on the tool historical usage data;

[0043] A data fusion algorithm based on an attention mechanism is used to adaptively assign weights to and integrate the tool state feature vector and the future usage trend prediction data to obtain a comprehensive tool evaluation data set.

[0044] Optionally, the calculating of the tool health index by a preset performance evaluation algorithm based on the tool comprehensive evaluation data set and the tool associated network topology structure includes:

[0045] Performing wavelet decomposition processing on the tool comprehensive evaluation data set to obtain tool wear trend characteristics, machining accuracy fluctuation characteristics and thermal stress distribution characteristics;

[0046] Calculate the network characteristics of each tool according to the tool-associated network topology to obtain the network centrality and clustering coefficient of the tool;

[0047] The tool wear trend characteristics, machining accuracy fluctuation characteristics, thermal stress distribution characteristics, network centrality and clustering coefficient are input into a pre-constructed five-layer feedforward neural network to obtain an initial tool health score;

[0048] Back-propagation updates the weights of the last two layers of the five-layer feedforward neural network to obtain an optimized neural network model for the current task;

[0049] The tool wear trend characteristics, machining accuracy fluctuation characteristics, thermal stress distribution characteristics, network centrality and clustering coefficient are input into the optimized neural network model, and processed by the Softmax function to obtain a corrected health score of the tool;

[0050] The tool health index is obtained by performing Mamdani fuzzy reasoning operation on the modified health score, the difficulty coefficient of the current machining task, the material properties and the precision requirements.

[0051] Optionally, the tool usage plan and preventive maintenance plan are generated by an intelligent scheduling algorithm according to the tool health index and the current processing task requirements, including:

[0052] Performing multi-objective optimization processing on the tool health index and the current machining task requirements to obtain a tool-task matching score;

[0053] The tool-task matching score, the number of available tools in the tool magazine, and the process sorting constraints are input into an intelligent scheduling algorithm based on deep reinforcement learning to obtain a tool usage plan;

[0054] Comparing and analyzing the expected usage time of each tool in the tool usage plan with its health index, and obtaining the critical usage time point of the tool;

[0055] The tool critical use time point and production plan are input into an intelligent maintenance algorithm based on constraint programming to obtain a tool preventive maintenance plan.

[0056] A second aspect of the present invention provides a discrimination device, the discrimination device comprising:

[0057] A digital twin model acquisition module is used to acquire a digital twin model of a tool in a tool magazine, wherein the digital twin model includes geometric feature parameters, material property information, and processing task data of the tool;

[0058] A similarity calculation module, used to calculate the similarity matrix between tools based on the geometric feature parameters, material property information and processing task data in the digital twin model, and generate a tool association network topology structure according to a preset similarity threshold;

[0059] The data fusion module is used to collect tool status data during the process of tools entering and leaving the tool magazine, and fuse it with the tool history usage data to form a comprehensive tool evaluation data set;

[0060] A health evaluation module, used to calculate the tool health index through a preset performance evaluation algorithm based on the tool comprehensive evaluation data set and the tool associated network topology structure;

[0061] The intelligent scheduling module is used to generate a tool usage plan and a preventive maintenance plan through an intelligent scheduling algorithm according to the tool health index and the current processing task requirements.

[0062] The technical solution provided in the embodiment of the present application has the following advantages:

[0063] First, at the data level, this solution uses a digital twin model to achieve all-round modeling of the static properties and dynamic behaviors of the tool. The digital twin model is centered on the geometric characteristic parameters, material property information, and processing task data of the tool. This process not only includes high-precision three-dimensional scanning modeling of the tool's geometric profile, but also combines the material's microscopic property data and the historical task data accumulated by the tool in actual processing. By organically integrating these multi-dimensional information, the digital twin model can dynamically reflect the actual performance of the tool under different processing tasks. This modeling method enables the management system to no longer rely solely on fixed-cycle offline inspections, but to be able to grasp the performance status of the tool in real time, thereby breaking the traditional method's reliance on time intervals.

[0064] Secondly, at the analysis level, this solution further utilizes the geometric features, material properties and processing task data in the digital twin model, calculates the similarity matrix between tools, and generates the tool association network topology based on the graph theory algorithm. The similarity calculation here uses multi-dimensional feature vectorization to ensure that the state of each tool can be quantified in a high-precision mathematical form. The construction of the tool association network is based on the threshold set in the similarity matrix, and a topological map reflecting the association between tools is generated through the adjacency matrix. In this process, the role of the tool in the overall system is quantified through network features such as centrality and clustering coefficient, which can quickly identify key tools and high-risk tool groups. This networked evaluation method not only effectively improves the pertinence of performance evaluation, but also dynamically reflects the matching degree between tools and task requirements, thereby achieving accurate performance analysis.

[0065] Thirdly, in terms of real-time data collection and fusion, this solution generates a comprehensive tool evaluation data set by monitoring the geometric profile changes, surface microstructure and interface stress distribution of the tool, combined with the historical usage data of the tool. Cross-modal feature extraction technology is used to unify these data from different sources into feature representations in the same vector space, and a time series prediction algorithm is used to analyze the future performance trends of the tool. This dynamic data fusion method not only makes up for the problem of evaluation bias caused by traditional tool management relying only on a single data source, but also provides a scientific basis for performance prediction. For example, by analyzing the wear trend of the tool, the time point of its performance degradation can be predicted in advance, providing reliable support for subsequent scheduling and maintenance plans.

[0066] Finally, in terms of decision optimization, the solution is based on the analysis results of the comprehensive tool evaluation data set and the associated network, and the tool usage plan and preventive maintenance plan are formulated through the intelligent scheduling algorithm. In the specific implementation process, the calculation of the health index is a key link. Through the fusion analysis of characteristics such as wear trend, processing accuracy fluctuation, thermal stress distribution, and combined with the network centrality and clustering characteristics of the tool, the deep neural network model is used to calculate the accurate health score. These scores are optimized through Mamdani fuzzy reasoning, and finally generate an accurate health index, providing a clear basis for tool use and maintenance decisions. The intelligent scheduling algorithm combines the tool health index with the processing task requirements through multi-objective optimization, dynamically adjusts the use order of the tool, and optimizes the maintenance plan, avoiding production interruptions and equipment losses caused by sudden tool failure. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0068] Figure 1 A schematic diagram of an embodiment of a tool identification method in an embodiment of the present invention;

[0069] Figure 2 FIG. 1 is a schematic diagram of an embodiment of a discrimination device in an embodiment of the present invention.

[0070] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back...), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0073] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, "and / or" in the full text includes three solutions. Taking A and / or B as an example, it includes technical solution A, technical solution B, and technical solution that satisfies both A and B. In addition, the technical solutions between the various embodiments can be combined with each other, which must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0074] An embodiment of the present application provides a tool identification method. Figure 1 A flow chart of a tool identification method provided in an embodiment of the present application. In this embodiment, the method includes:

[0075] See also Figure 1 , obtaining a digital twin model of a tool in a tool magazine, wherein the digital twin model includes geometric feature parameters, material property information, and processing task data of the tool;

[0076] In one embodiment of the present invention, the digital twin model of the tool in the tool magazine is obtained, and the digital twin model includes the geometric feature parameters, material property information and processing task data of the tool, including: identifying the tool in the tool magazine to obtain the unique identification code of the tool; querying the tool information database according to the unique identification code; if the corresponding complete information is found in the tool information database, the corresponding complete information includes the geometric feature parameters, material property information and processing task data of the tool, and constructing the digital twin model based on the corresponding complete information; if the corresponding complete information is not found, the following steps are performed: three-dimensional scanning of the tool to obtain the geometric contour data of the tool; based on the geometric contour data, combined with a preset tool parameter template, generating the geometric feature parameters of the tool; obtaining the material composition data of the tool, and determining the material property information of the tool according to the material performance database; accessing the production management system, obtaining the historical processing task records associated with the tool, and extracting the processing task data; integrating the geometric feature parameters, material property information and processing task data to construct a digital twin model.

[0077] Specifically, first, in order to identify the tools in the tool magazine, it is necessary to obtain the unique identification code of the tool. The unique identification code is a code that can uniquely mark each tool. It is usually assigned when the tool is produced, or generated by a management system, such as based on RFID tags or QR code technology. The core of identity recognition is to quickly capture the unique identification code of the tool through a scanning device (such as an RFID scanner or a QR code recognition device). The unique identification code carries the basic information of the tool and is the starting point of the entire identification and query process. For example, in the tool magazine of a complex multi-axis CNC machine tool, scanning the unique identification code of each tool can quickly determine its corresponding record in the tool information database.

[0078] Next, the tool information database is queried using the unique identification code obtained. If the corresponding complete information is found in the database, and this information includes the geometric feature parameters, material property information, and processing task data of the tool, the digital twin model of the tool can be directly constructed based on these data. The geometric feature parameters here are the precise three-dimensional geometric data that describe the shape of the tool, such as the blade angle of the tool, the curvature of the cutting surface, etc.; the material property information is the data about the microstructure and physical properties of the tool material, such as hardness, wear resistance, heat resistance, etc.; the processing task data is the usage record accumulated by the tool in historical tasks, including the type of processing material, processing parameters (such as rotation speed, feed speed) and task completion quality. The integrity of these data is crucial because they determine whether the digital twin model can accurately reflect the actual characteristics of the tool. For example, in the scenario of processing titanium alloy parts, if the database has stored the usage records of the tool in similar tasks, a digital twin model that highly matches the actual tool behavior can be quickly generated for subsequent performance evaluation and optimization.

[0079] However, when the complete information of the tool is not found in the database, an alternative strategy is needed to supplement the data and complete the construction of the digital twin model. At this time, the first step is to perform a three-dimensional scan of the tool to obtain the geometric contour data of the tool. Three-dimensional scanning is a common reverse modeling technology that can capture the contour of the tool through laser scanning or structured light scanning equipment to generate a high-precision three-dimensional geometric model. The significance of this step is that even if the database is missing information, the geometric features of the tool can be restored through physical modeling. For example, when a new tool put into the tool magazine has not yet been registered in the system, its shape data can be quickly generated through three-dimensional scanning, thereby solving the problem of missing data.

[0080] Next, based on the geometric profile data obtained by 3D scanning, combined with the preset tool parameter template, the geometric feature parameters of the tool are generated. The preset tool parameter template here is a standardized parameter set established based on common tool types (such as milling cutters, drills, etc.), which sets a mathematical model of typical features for each type of tool. Based on the scan data, by comparing it with the parameter template, characteristic values ​​such as tool length, blade width, helix angle, etc. can be extracted to further enrich the description of geometric features.

[0081] Then, by detecting the chemical composition data of the tool material, its material property information is determined. The material composition can be obtained by a micro-area X-ray spectrometer or a portable spectrometer. By matching the detected composition data with the material property database, the material property information such as hardness and corrosion resistance of the tool can be further confirmed. For example, if the test results show that the tool is made of a cemented carbide with a high tungsten content, it can be inferred that it has good heat resistance but is more susceptible to brittle fracture under high impact loads. These data will be integrated into the model to better evaluate the applicability of the tool under different task conditions.

[0082] Finally, it is necessary to extract the historical processing task records associated with the tool from the production management system to supplement the processing task data. By querying the task records associated with the tool's unique identification code, the tool's operating parameters (such as spindle speed, cutting force, etc.) during the processing and the completion of the task can be obtained. These data not only reflect the historical use of the tool, but also provide support for analyzing the performance trend of the tool. For example, in the processing of aerospace parts, if the record shows that a tool exhibits obvious wear under long-term high temperature, these historical data can be used as key inputs in the model.

[0083] After integrating geometric feature parameters, material property information and processing task data, a complete digital twin model of the tool is formed. This model has the ability to dynamically update and perform multi-dimensional analysis, and can serve as the basis for tool performance evaluation and scheduling optimization. In practical applications, such as when processing complex medical device components, the digital twin model can provide real-time feedback on the health status of the tool, helping the management system predict the life of the tool and adjust the usage plan.

[0084] Through these steps, the method not only realizes the accurate identification and comprehensive modeling of tool information, but also solves the problem of modeling interruption caused by missing database information. Based on these implementation methods, the entire system can dynamically adapt to the diversity of tool performance and the complexity of the processing environment, improving the scientific nature of tool management and the continuity of production.

[0085] In one embodiment of the present invention, the construction of the digital twin model includes: parameterizing the geometric feature parameters to obtain a high-precision three-dimensional geometric model of the tool; mapping the material property information to the high-precision three-dimensional geometric model through a multi-scale material model to obtain a tool virtual entity with microstructure information; performing deep learning analysis on the processing task data to obtain the law of tool performance evolution; associating the tool performance evolution law with the tool virtual entity to obtain a dynamic performance prediction model; constructing a two-way data interaction interface between the tool virtual entity and the actual tool to achieve real-time state synchronization; integrating the high-precision three-dimensional geometric model, the tool virtual entity, the dynamic performance prediction model and the two-way data interaction interface to obtain a digital twin model of the tool.

[0086] Specifically, first of all, parameterization of geometric feature parameters is the starting point for establishing a digital twin model of the tool. This step generates a high-precision three-dimensional geometric model by expressing the geometric shape of the tool in a parametric way. Specifically, based on the point cloud data or scan data of the tool, a parametric curve fitting method is used to extract the blade profile, curvature distribution and other geometric features of the tool. A parametric model is established through these features, for example, spline curves are used to approximate the blade profile and complex geometric shapes are expressed as a set of mathematical equations. The resulting high-precision three-dimensional geometric model can accurately describe the key structure of the tool. For example, when processing complex blades, the model can accurately reproduce the tool edge angle and the curvature distribution of the cutting surface, thereby simulating the real cutting behavior of the tool in the simulation. The advantage of this method is that through parametric processing, the model is not only more accurate, but also has high versatility and is suitable for different types of tools.

[0087] After generating a high-precision three-dimensional geometric model, the material property information is mapped to the geometric model through a multi-scale material model, thereby generating a virtual entity of the tool with microstructure information. The multi-scale material model associates the grain structure and texture direction of the tool material with its overall performance parameters (such as hardness and fatigue resistance) by combining micro and macro data. The mapping process specifically includes: extracting microstructure data that matches the tool material from the material performance database, such as grain size and distribution characteristics; then embedding these data into specific areas of the geometric model through the finite element method (FEM), so that the virtual entity not only has a geometric shape, but also can simulate the material response of the tool under actual processing conditions. For example, under high-temperature cutting conditions, the virtual entity can exhibit behaviors such as thermal expansion and plastic deformation, thereby more realistically reflecting the use characteristics of the tool. The advantage of this mapping step is that by embedding material properties into the geometric model, the virtual entity can comprehensively reflect the geometry and material properties of the tool, providing important support for subsequent performance analysis.

[0088] Next, deep learning analysis is performed on the processing task data to extract the performance evolution law of the tool. Processing task data usually includes the process parameters (such as spindle speed, feed speed), task conditions (such as cutting force, thermal stress) and result data (such as surface roughness, wear degree) of the tool in different processing tasks. By constructing a time series prediction model, such as a deep learning framework based on LSTM (Long Short-Term Memory Network), the time dependence and implicit patterns of these data are analyzed. For example, the model can learn the nonlinear effect of cutting speed on tool wear, thereby predicting the performance attenuation trend of the tool in continuous processing. Taking high-strength alloy processing as an example, deep learning can reveal the key nodes of tool thermal stress accumulation under long-term processing, and then predict the service life of the tool. The core significance of this analysis is that by extracting the performance evolution law, the model can provide accurate performance predictions and provide data support for the optimal scheduling of tools.

[0089] Subsequently, the tool performance evolution law is associated with the tool virtual entity to generate a dynamic performance prediction model. The dynamic performance prediction model can dynamically simulate the performance changes of the tool under different processing conditions by integrating geometric features, material properties and performance laws. Specifically, the performance evolution law is introduced into the virtual entity, and the current state of the tool is reflected in real time by adjusting the material property parameters or stress distribution data in the entity. For example, the model can predict the wear rate of a tool when processing high-hardness steel in a virtual environment, and simulate the deformation of the blade by adjusting the dimensional parameters of the geometric model. The advantage of this dynamic performance prediction model is that it can provide detailed status of tool performance in real time, providing an accurate basis for subsequent scheduling optimization and maintenance strategies.

[0090] In order to realize real-time data interaction between the virtual entity of the tool and the actual tool, a two-way data interaction interface is constructed. The interface realizes two-way data transmission by designing a standardized communication protocol (such as MQTT or REST API). On the one hand, the operating parameters of the tool (such as temperature and stress distribution) are collected in real time and sent to the virtual entity; on the other hand, the virtual entity generates optimization instructions through simulation analysis and transmits them back to the actual tool. For example, when the interface detects that the temperature of the tool exceeds the preset threshold, the virtual entity can generate instructions to adjust the cutting speed or coolant flow, thereby reducing thermal stress and extending the tool life. The advantage of this interaction is that it ensures that the virtual model can be consistent with the actual tool at any time, enhancing the real-time and applicability of the model.

[0091] Finally, the high-precision three-dimensional geometric model, tool virtual entity, dynamic performance prediction model and two-way data interaction interface are integrated to build a complete digital twin model. Through this integration, the digital twin model can uniformly manage the geometric features, material properties and performance status of the tool, and dynamically update through real-time interaction. For example, when multi-axis CNC machine tools are processing complex parts, the digital twin model can provide real-time feedback on the health status of the tool, predict the tool replacement time in advance, and thus optimize the processing process. The advantage of this integration is that it comprehensively improves the visualization, controllability and predictability of tool performance, making it more efficient and reliable in complex processing tasks.

[0092] In summary, through rigorous step design, from geometric modeling to material mapping, deep learning analysis, dynamic prediction and real-time interaction, the construction of the tool digital twin model is fully realized. This model solves the problem of difficult real-time evaluation of performance changes in traditional tool management, and provides solid technical support for the efficient execution of processing tasks.

[0093] In one embodiment of the present invention, the bidirectional data interaction interface between the tool virtual entity and the actual tool is constructed to achieve real-time status synchronization, including: designing a tool data exchange protocol to define the data structure and communication format between the tool virtual entity and the actual tool; based on the tool data exchange protocol, collecting and processing the working parameters of the actual tool to generate a standardized tool status data packet; using the tool status data packet to update the corresponding parameters in the virtual entity, and generating a control instruction data packet according to the optimization algorithm of the virtual entity; transmitting the control instruction data packet to the actual tool through the tool data exchange protocol to complete bidirectional data interaction and status synchronization.

[0094] Specifically, first of all, the design of the tool data exchange protocol is the basis of the entire two-way data interaction, which defines the data structure and communication format between the virtual entity and the actual tool. The main contents of the protocol include parameter identifiers, data types, transmission frequency and message formats. For example, the protocol can specify that the identifiers of tool operating parameters (such as temperature and stress distribution) are specifically encoded, using JSON or XML as the data format, and the transmission frequency is once per second. The significance of this protocol design is to ensure the standardization and compatibility of data exchange, especially in a complex environment where multiple devices are interconnected, it can effectively avoid data conflicts or losses. For example, during the processing process, if the real-time stress value of the tool and the working parameters of the cooling system need to be transmitted synchronously, the data exchange protocol can ensure that key data is transmitted on time by specifying priorities and timestamps, providing a basis for system optimization.

[0095] After the protocol design is completed, the working parameters of the actual tool are collected and processed based on the tool data exchange protocol to generate a standardized tool status data packet. The working parameters include the temperature, vibration frequency, cutting force and wear of the actual tool during operation. These data can be obtained in real time through sensors (such as temperature sensors and vibration sensors). The collected raw data are usually not in a uniform format or contain noise, and need to be preprocessed, including denoising, normalization and formatting. For example, temperature data is filtered to remove outliers and stored in standard units of degrees Celsius. After processing, these data are encapsulated into a standardized tool status data packet according to the structure defined by the data exchange protocol, which contains parameter identifiers, data values ​​and timestamps. The advantage of standardization is that no matter how complex the data source is, after unified encapsulation, the virtual entity can directly parse these data to provide a basis for status updates.

[0096] Using the tool status data packet mentioned above, the corresponding parameters in the virtual entity can be dynamically updated so as to always keep consistent with the actual tool status. For example, if the status data packet shows that the temperature of the actual tool rises to 90°C, the temperature parameters of the virtual entity will be updated synchronously, and the thermal expansion coefficient of the tool material properties will be adjusted according to the mapped material model. This dynamic update not only enables the virtual entity to truly reflect the current status of the tool, but also provides real-time input for subsequent optimization algorithms. In addition, the optimization algorithm of the virtual entity generates a control instruction data packet based on the updated status parameters. For example, when the cutting force exceeds the set threshold, the virtual entity generates an instruction to adjust the feed speed or cutting depth through analysis to ensure that the tool operates within a safe range. The control instruction data packet is also encapsulated by the protocol and contains an instruction identifier, target value, and execution priority.

[0097] Next, the generated control instruction data packet is transmitted to the actual tool through the tool data exchange protocol, thereby completing two-way data interaction and state synchronization. During the transmission process, the protocol ensures that the data packet can reach the target tool quickly and accurately, and trigger the corresponding execution module. For example, when the instruction data packet indicates to reduce the feed speed, the actual tool control system will immediately execute the adjustment instruction, and at the same time, the adjusted parameters (such as the new feed speed) will be packaged into a state data packet again, and fed back to the virtual entity to form a closed-loop control. The biggest advantage of this interaction is that it achieves a high degree of coordination between the virtual entity and the actual tool, and can dynamically optimize the tool usage strategy according to the real-time status. For example, when processing high-strength materials, if the tool performance degrades due to continuous high temperature, the virtual entity can immediately generate a control instruction to reduce the cutting speed, which is transmitted to the actual tool to prevent failure, while monitoring the new operating status and further optimizing subsequent instructions.

[0098] Through the above implementation process, the virtual entity of the tool and the actual tool can achieve real-time two-way data interaction. Specifically, the data exchange protocol ensures the standardization and efficiency of information transmission, the status data packet provides real-time operating parameters, and the control instruction data packet realizes the dynamic optimization of the tool status. The advantage of the entire process is that it can not only accurately reflect the current performance of the tool, but also actively intervene in the operation of the actual tool through the analysis and prediction capabilities of the virtual entity, thereby improving the intelligent level of tool management, extending the tool life, and optimizing the processing quality. This interactive method is suitable for the dynamic adjustment needs in complex processing environments and provides reliable technical support for processing.

[0099] Please continue reading Figure 1 , based on the geometric feature parameters, material property information and processing task data in the digital twin model, calculating the similarity matrix between tools, and generating a tool association network topology structure according to a preset similarity threshold;

[0100] In one embodiment of the present invention, the inter-tool similarity matrix is ​​calculated based on the geometric feature parameters, material property information and processing task data in the digital twin model, and the tool association network topology structure is generated according to a preset similarity threshold, including: performing feature vector quantization processing on the geometric feature parameters, material property information and processing task data in the digital twin model to obtain a tool feature vector set; using a multi-dimensional similarity calculation algorithm to perform pairwise comparison on the vectors in the tool feature vector set to obtain an inter-tool similarity matrix; according to the preset similarity threshold, binarizing the inter-tool similarity matrix to obtain a tool association adjacency matrix; based on the tool association adjacency matrix, constructing a tool association network topology structure through a graph theory algorithm.

[0101] Specifically, first, when the geometric feature parameters, material property information and processing task data in the digital twin model are feature vectorized, the core goal is to convert multiple attributes of each tool (geometry, material, historical task data, etc.) into a unified mathematical representation. This process extracts the different features of each tool (such as the blade shape, material hardness, processing conditions, etc.) into vector form to ensure that various types of data can be compared and calculated in the same dimension. Specifically, the geometric features can be converted into numerical features through information such as the blade angle, edge shape, and surface finish of the tool, and combined with the material properties of the tool (such as hardness, corrosion resistance, etc.) and processing task data (such as cutting speed, feed rate, and processing material) to form the feature vector of the tool. For example, a milling cutter has a vector containing features such as blade angle, cutting width, and material hardness, which can fully express the working characteristics of the tool. The advantage of feature vectorization is that it converts the multi-dimensional information of the tool into numerical values, which can be used for subsequent calculations and analysis efficiently.

[0102] After obtaining the feature vector set of the tool, it is necessary to compare each tool in the tool feature vector set with each other through a multi-dimensional similarity calculation algorithm to calculate the similarity matrix between the tools. The purpose of similarity calculation is to quantify the similarity between each tool and other tools in multiple dimensions (such as geometry, material properties, and processing tasks). For example, by calculating the Euclidean distance of geometric features between tools, the cosine similarity of material properties, and the Pearson correlation of processing task data, a matrix of similarity between tools can be obtained. Each element in this matrix represents the similarity between two tools in a specific dimension, and the higher the value, the more similar their features are. For example, if two tools have similar geometry and material properties and have been used for similar processing tasks, the similarity between them will be very high. Through this similarity calculation, we can better understand the similarity between tools and their impact on the entire processing process. The advantage of multi-dimensional similarity calculation is that it can comprehensively consider data from multiple aspects to ensure that the similarity between tools can be fully and accurately reflected.

[0103] Based on the similarity matrix, we further perform binarization to obtain the tool association adjacency matrix. The purpose of binarization is to simplify the similarity matrix, remove those tool pairs with low similarity, and retain only those tool pairs with high similarity. For example, when the similarity between two tools exceeds the set threshold, we believe that they have a certain correlation, so they can be represented as "1" in the tool association adjacency matrix, otherwise they are represented as "0". In this way, the adjacency matrix forms a simplified tool relationship graph, where nodes represent tools and edges represent the correlation between tools. The advantage of binarization is that by setting an appropriate similarity threshold, we can eliminate unnecessary complex relationships and retain tool pairs with practical significance, thereby simplifying calculations and improving processing efficiency in subsequent analysis.

[0104] Based on the tool association adjacency matrix, the tool association network topology is constructed using graph theory algorithms. The graph theory algorithm mainly uses the data in the adjacency matrix to construct a relationship network between tools. In this network, each tool is a node, and the edges between nodes represent the association between tools. This process can further analyze the relationship between tools through common graph theory algorithms such as the shortest path algorithm, centrality calculation, and community discovery. By analyzing the graph structure, key tools (such as nodes with high centrality) or tool groups with strong correlation (such as tools in the same community) can be identified, thereby providing a basis for tool management. For example, some tools show similar performance in multiple processing tasks, and the "group" they form in the tool association network can be used as an optimization object for centralized management and scheduling. The advantage of the graph theory algorithm is that it can intuitively display the association structure between tools, helping the management system to optimize the configuration of tools and assign tasks.

[0105] In general, the implementation of the above steps can accurately analyze the similarity of tools and construct an association network from multiple dimensions. Through feature vectorization, similarity calculation and graph analysis, we can fully understand the relationship between tools and improve processing efficiency and quality by optimizing tool usage strategies. Especially in complex processing environments, such as the processing of aerospace parts, the performance of tools often has an important impact on processing quality. Therefore, accurate association analysis between tools based on such technical means is crucial to improving the stability and efficiency of the entire production process.

[0106] Please continue reading Figure 1 ,During the process of tools entering and leaving the tool magazine, the tool status data is collected and integrated with the tool history usage data to form a comprehensive tool evaluation data set;

[0107] In one embodiment of the present invention, during the process of the tool entering and exiting the tool magazine, tool status data is collected and fused with the tool history usage data to form a tool comprehensive evaluation data set, including: during the process of the tool entering and exiting the tool magazine, tool status data is obtained, and the tool status data includes a tool geometry profile change map, a surface microstructure map and a tool-handle interface stress distribution map; tool history usage data of the tools entering and exiting the tool magazine is extracted from the production management system, and the tool history usage data includes a processing parameter sequence and a processing quality index record; cross-modal feature extraction is performed on the tool status data to obtain a tool status feature vector; a time series prediction algorithm is used to generate future tool usage trend prediction data based on the tool history usage data; a data fusion algorithm with an attention mechanism is used to adaptively weight and integrate the tool status feature vector and the future usage trend prediction data to obtain a tool comprehensive evaluation data set.

[0108] Specifically, first, during the process of the tool entering and exiting the tool magazine, the tool status data needs to be obtained. These data include tool geometry profile change diagram, surface microstructure diagram, and tool-handle interface stress distribution diagram. The geometry profile change diagram records the changes in the blade shape, tool length, etc. during use by scanning or measuring the actual shape of the tool. For example, through laser scanning technology, a high-precision image of the tool surface geometry profile can be obtained, reflecting the morphological changes caused by wear of the tool during long-term use. The surface microstructure diagram reflects the microscopic morphology of the tool surface material, such as microcracks and corrosion spots on the blade part, which are key factors affecting tool performance. The surface microstructure diagram generated by microscopy technology (such as electron scanning microscope, SEM) can help determine whether the tool has microstructural damage and warn of possible failure modes in advance. The tool-handle interface stress distribution diagram measures the stress distribution on the contact surface between the tool and the handle through sensors, reflecting the stress of the tool in the working state. Through stress measurement equipment, such as strain gauges, the force changes of the tool in use can be captured in real time. Especially in a high-load processing environment, excessive stress may cause tool failure, so this data is crucial for tool health management.

[0109] After obtaining these tool status data, the next task is to extract the tool usage history data of entering and leaving the tool magazine from the production management system. The tool usage history data includes the processing parameter sequence and the processing quality index record. The processing parameter sequence includes various process parameters used by the tool in different processing tasks, such as cutting speed, feed rate, cutting depth, etc. These data are usually recorded by CNC machine tools and can be directly extracted from the production management system. The processing quality index record includes the quality performance of the tool during the processing, such as surface roughness, cutting force, processing accuracy, etc. These indicators reflect the working status of the tool in actual work and can provide valuable reference for the service life and health status of the tool.

[0110] Next, the acquired tool status data needs to be subjected to cross-modal feature extraction to obtain the tool status feature vector. Cross-modal feature extraction refers to converting data from different sources and types (such as geometric change maps, microstructure maps, and stress distribution maps) into a unified representation that can be compared and analyzed - the tool status feature vector. Since these data sources have different modes, they need to be converted into digital vectors through feature extraction algorithms to facilitate subsequent analysis and processing. For example, the geometric profile change map can extract the key features of the tool surface morphology through edge detection technology, and the surface microstructure map can extract the degree of surface damage through image processing algorithms. This information will eventually be converted into a numerical vector to express the health status of the tool during use. Through cross-modal feature extraction, the tool status data can be analyzed in a unified feature space, thereby realizing the fusion of multi-dimensional information.

[0111] Next, a time series prediction algorithm is used to generate tool usage trend prediction data based on the tool historical usage data. Time series prediction algorithms such as LSTM (Long Short-Term Memory Network) can analyze the time dependency in historical data and predict tool performance changes in future use. For example, under certain process conditions, after a period of use, a tool may show a trend of increased wear. Through the historical processing process parameter sequence and processing quality index records, the LSTM model can identify the key factors affecting tool performance and predict its future usage status and remaining life. The advantage of this process is that through time series prediction, the system can identify potential risks of the tool in advance and avoid sudden failure of the tool in subsequent processing tasks.

[0112] After obtaining the future usage trend prediction data of the tool, the data fusion algorithm of the attention mechanism is used to adaptively weight and integrate the tool state feature vector and the future usage trend prediction data. The attention mechanism is an algorithm that can dynamically adjust the weight according to the importance of the input data. Here, the attention mechanism ensures that key factors (such as geometric wear and material fatigue of the tool) can get more attention in the tool state assessment by adaptively allocating weights to different data sources. For example, suppose a tool shows greater wear in a specific processing task, but due to the strong wear resistance of its material, the future usage trend prediction data shows that it still has a certain service life. The attention mechanism will automatically adjust the weight of the feature vector so that the future usage trend data occupies a higher proportion in the evaluation, thereby achieving a more accurate comprehensive evaluation.

[0113] Finally, a comprehensive evaluation data set of the tool is obtained through the data fusion algorithm. This data set combines the tool's state feature vector and future usage trend prediction data, and can fully reflect the tool's health status and future performance. The comprehensive evaluation data set can not only help the management system evaluate the tool's current state in real time, but also provide optimization decisions based on the predicted trend data. For example, for tools that are expected to fail, the system can plan the tool replacement time or adjust the processing technology in advance to avoid processing interruptions or equipment damage.

[0114] In summary, this process provides an accurate tool health assessment and usage trend prediction system by extracting valuable features from multidimensional data and integrating them through fusion algorithms. Through cross-modal feature extraction, time series prediction, and data fusion of attention mechanisms, the system can dynamically optimize tool usage strategies, improve the stability and efficiency of the machining process, avoid potential machining risks, and improve the level of intelligent production. The implementation of this technology will significantly optimize tool management and maintenance strategies, and improve production efficiency and equipment reliability in the manufacturing industry.

[0115] Please continue reading Figure 1 , based on the tool comprehensive evaluation data set and the tool associated network topology structure, calculating the tool health index through a preset performance evaluation algorithm;

[0116] In one embodiment of the present invention, the tool health index is calculated based on the tool comprehensive evaluation data set and the tool associated network topology structure through a preset performance evaluation algorithm, including: performing wavelet decomposition processing on the tool comprehensive evaluation data set to obtain tool wear trend characteristics, machining accuracy fluctuation characteristics and thermal stress distribution characteristics; performing network feature calculation on each tool according to the tool associated network topology structure to obtain the network centrality and clustering coefficient of the tool; inputting the tool wear trend characteristics, machining accuracy fluctuation characteristics, thermal stress distribution characteristics, network centrality and clustering coefficient into a pre-constructed five-layer feedforward neural network to obtain an initial tool health score; back-propagating and updating the weights of the last two layers of the five-layer feedforward neural network to obtain an optimized neural network model for the current task; inputting the tool wear trend characteristics, machining accuracy fluctuation characteristics, thermal stress distribution characteristics, network centrality and clustering coefficient into the optimized neural network model, and processing through a Softmax function to obtain a revised health score of the tool; performing Mamdani fuzzy reasoning operation on the revised health score, the difficulty coefficient of the current machining task, material properties and precision requirements to obtain a tool health index.

[0117] Specifically, first, the tool comprehensive evaluation data set contains multi-dimensional information such as tool wear, machining accuracy, thermal stress, etc., and wavelet decomposition processing is an effective signal processing method that can extract features of different frequencies from complex signals. For the tool status data, wavelet decomposition can decompose the time series signal into multiple sub-signals of different frequency bands, thereby capturing the change trend of the tool at different scales. Specifically, the tool wear trend characteristics, machining accuracy fluctuation characteristics, and thermal stress distribution characteristics are decomposed based on the signals collected during the tool operation (such as vibration, temperature, cutting force, etc.). The wear trend characteristics reflect the change in the wear rate of the tool over time during use; the machining accuracy fluctuation characteristics reflect the accuracy stability of the tool during the machining process; and the thermal stress distribution characteristics reflect the thermal stress changes on the tool surface through temperature sensor data. After these signals are decomposed by wavelet, different frequency components can be effectively separated, so that the performance of the tool under different use conditions can be more clearly identified. For example, if the tool experiences high-frequency thermal shock under certain cutting conditions, then wavelet decomposition can extract these sudden high-frequency characteristics to help determine the potential damage or failure mode of the tool. The advantage of wavelet decomposition is that it can analyze in two dimensions, time and frequency, and thus can very effectively extract useful information from the complex signals of the tool.

[0118] Next, based on the tool association network topology, the network characteristics of the tool are calculated to obtain the network centrality and clustering coefficient of the tool. The tool association network topology is constructed by the tool similarity matrix generated in the previous steps. Each tool can be regarded as a node in the graph, and the association between tools is represented by the edges in the graph. Network centrality is an important concept in graph theory, which indicates the importance of a node in the entire network, and is usually measured by calculating the shortest path from the node to other nodes. For a tool, centrality can reflect the criticality of the tool in the entire tool library. For example, if a tool is frequently used in multiple processing tasks and has stable performance, its centrality is high, indicating that it is more important in the tool library. The clustering coefficient is another graph theory feature that is used to measure the degree of connection between the neighboring nodes of a node. For tools, the clustering coefficient can help identify which tools have similar working modes or performance characteristics in use, and then more reasonably group and schedule them in management. By calculating the centrality and clustering coefficient of the tool, important tools and tool groups with similar working characteristics can be identified, providing data support for subsequent tool management and optimization.

[0119] After obtaining the wear trend characteristics, machining accuracy fluctuation characteristics, thermal stress distribution characteristics, network centrality and clustering coefficient of the tool, these characteristics are input into a pre-built five-layer feedforward neural network for processing to obtain the initial tool health score. Feedforward Neural Network (FNN) is a basic neural network architecture in deep learning, consisting of an input layer, multiple hidden layers and an output layer. In this scheme, the input layer receives various characteristics of the tool, and after layer-by-layer processing by the hidden layer, the output layer will give the initial health score of the tool. The calculation of the initial health score is based on the multi-dimensional characteristics of the tool. The network learns the relationship between the historical data and health status of the tool and outputs a score value to reflect the current health status of the tool. For example, through network training, the model learns how to evaluate the health status of the tool based on its wear, thermal stress and other characteristics. This process can quickly and accurately calculate the initial health score of the tool.

[0120] Then, the weights of the last two layers of the five-layer feedforward neural network are back-propagated and updated to obtain the optimized neural network model for the current task. Back-propagation is a key algorithm in neural network training, which adjusts the weights of the network by minimizing the prediction error. In each iteration, the neural network calculates the error based on the gap between the input data and the true health status, and propagates the error backwards through the back-propagation algorithm, and updates the network weights through the gradient descent algorithm. The optimized neural network model can more accurately calculate the health score based on the current characteristics of the tool and adapt to the complexity of different machining tasks. The advantage of this optimization process is that it can improve the prediction accuracy of the model in continuous updates, especially when faced with new tool data, it can quickly adjust and give accurate health assessments.

[0121] Next, the tool wear trend characteristics, machining accuracy fluctuation characteristics, thermal stress distribution characteristics, network centrality and clustering coefficient are input into the optimized neural network model and processed by the Softmax function to obtain the corrected health score of the tool. The Softmax function is often used in classification tasks. It converts the continuous value of the output into a probability distribution to ensure the interpretability of the result. Here, the Softmax function is used to normalize the optimized health score and convert the score values ​​of all tools into a probability distribution that conforms to certain rules, ensuring that the health score of each tool is within a reasonable range and can be compared with other tools. The advantage of this process is that through the normalization of the Softmax function, the corrected health score can more intuitively reflect the relative advantages and disadvantages of the tool health status, which is convenient for subsequent decision-making and tool scheduling.

[0122] Finally, Mamdani fuzzy reasoning is performed on the corrected health score, the difficulty coefficient of the current machining task, the material properties and the precision requirements to obtain the tool health index. Mamdani fuzzy reasoning is a classic fuzzy logic algorithm that is usually used to process uncertainty and fuzzy information. In this scheme, Mamdani fuzzy reasoning combines the corrected health score of the tool with information such as task difficulty and material properties, and derives the tool health index through fuzzy rules. For example, if the tool health score is low and the difficulty of the current machining task is high, the fuzzy reasoning system will increase the necessity of tool replacement or maintenance. By performing fuzzy reasoning on different input conditions, the system can flexibly generate a tool health index to reflect whether the tool is suitable to continue to perform the current task or whether it needs maintenance or replacement.

[0123] In summary, through the application of a series of algorithms such as wavelet decomposition, graph analysis, neural network optimization and fuzzy reasoning, this solution achieves a comprehensive assessment of tool health status. This process not only improves the accuracy of tool health assessment, but also enhances the adaptability and intelligence level of the system, providing strong data support for tool management and optimization decisions. The combination of these technologies not only effectively improves the intelligence level of tool management, but also provides a scientific basis for tool maintenance and optimization in complex manufacturing tasks.

[0124] Please continue reading Figure 1 , according to the tool health index and the current processing task requirements, a tool usage plan and a preventive maintenance plan are generated through an intelligent scheduling algorithm.

[0125] In one embodiment of the present invention, a tool usage plan and a preventive maintenance plan are generated through an intelligent scheduling algorithm based on the tool health index and the current processing task requirements, including: performing multi-objective optimization processing on the tool health index and the current processing task requirements to obtain a tool-task matching score; inputting the tool-task matching score, the number of available tools in the tool magazine, and the process sorting constraints into an intelligent scheduling algorithm based on deep reinforcement learning to obtain a tool usage plan; comparing and analyzing the expected usage time of each tool in the tool usage plan with its health index to obtain a tool critical usage time point; inputting the tool critical usage time point and the production plan into an intelligent maintenance algorithm based on constraint programming to obtain a tool preventive maintenance plan.

[0126] Specifically, first, the tool health index and the current machining task requirements are multi-objective optimized to obtain the tool-task matching score. The tool health index reflects the current working status and expected life of the tool, while the current machining task requirements include factors such as the difficulty of the task, the precision requirements, and the required tool characteristics. The core task of the multi-objective optimization process is to combine these two factors to ensure that the tool that best suits the current task is selected within the tool health range. This process usually involves an optimization algorithm that balances the relationship between health status and task requirements by setting weights. For example, for a tool with greater wear, if its health index is low, the optimization system will reduce its matching score according to the difficulty and precision requirements of the task, while for a tool with a higher health index, the system will increase its matching score. Finally, the tool-task matching score obtained provides a quantitative basis for the degree of adaptation of each tool to the current task. The advantage of this process is that it can fully consider the health status of the tool and the characteristics of the machining task, thereby providing more accurate decision support for tool selection.

[0127] Next, the tool-task matching score, the number of available tools in the tool magazine, and the process sorting constraints are input into the intelligent scheduling algorithm based on deep reinforcement learning to obtain the tool usage plan. Deep reinforcement learning is a powerful algorithm that can optimize decisions in complex environments. Here, the reinforcement learning algorithm continuously optimizes the tool usage plan through multiple attempts and feedback. The intelligent scheduling system uses the tool-task matching score as the main input, combined with the number of available tools in the tool magazine and the process sorting constraints to optimize scheduling. The process sorting constraint usually means that in multiple processes, each tool can only be used in a specific step, so the algorithm needs to consider the order of each process to avoid unreasonable tool scheduling. The deep reinforcement learning algorithm selects the optimal tool and task pairing based on the current state (tool health index, task requirements, tool availability, etc.) at each decision, and generates a tool usage plan. Through repeated training and updating, the intelligent scheduling system can gradually improve the efficiency of tool use and ensure the smooth progress of processing tasks. The advantage of this process is that deep reinforcement learning can automatically handle complex scheduling tasks, ensure the efficient use of tools, and find the optimal solution under multi-dimensional constraints.

[0128] After obtaining the tool usage plan, the next task is to compare and analyze the expected usage time of each tool in the tool usage plan with its health index to obtain the critical usage time point of the tool. The expected usage time refers to the expected working time of the tool under the current task and conditions, while the health index reflects the degree of wear and remaining service life of the tool. When the expected usage time of the tool is close to the critical value reflected by its health index, the system will identify the critical usage time point of the tool. This critical point represents the moment when the tool may fail or degrade in performance. By comparing and analyzing the expected usage time and health index of the tool, the system can set a reasonable replacement or maintenance time for each tool. For example, if the health index of the tool is low and the expected usage time is long, the system will identify the potential risks of the tool in advance and mark it as a tool that needs to be replaced or maintained. The advantage of this analysis is that the tool maintenance or replacement can be planned in advance based on the health status and usage time of the tool, avoiding processing interruptions caused by tool failure.

[0129] Subsequently, the critical use time point of the tool and the production plan are input into the intelligent maintenance algorithm based on constraint programming to obtain the preventive maintenance plan for the tool. The constraint programming algorithm is mainly used to deal with optimization problems with multiple constraints. In the tool maintenance plan, the constraints include the critical use time point of the tool, the task arrangement in the production plan, and the availability of the tool. Through constraint programming, the system can reasonably arrange the maintenance or replacement time of the tool while ensuring that the production task is completed on time. Specifically, the intelligent maintenance algorithm will select a suitable time window for preventive maintenance of the tool in combination with the current production plan. For example, if the health index of a tool is low and close to the critical use time point, the system will choose to perform preventive maintenance before it is actually used, thereby avoiding failure in the middle of the task. In this way, the preventive maintenance plan not only improves the service life of the tool, but also ensures the smooth progress of the production task. The advantage of this process is that the constraint programming algorithm can find the optimal maintenance plan under multiple conditions, avoid the impact of sudden failures on production, and improve the overall stability of production.

[0130] In summary, through the application of a series of intelligent algorithms such as multi-objective optimization, deep reinforcement learning, expected usage time analysis and constraint programming, this solution achieves accurate scheduling and optimized maintenance of tools. Through these methods, the system can evaluate the health of the tool in real time during use, and optimize the tool usage plan and maintenance timing according to the processing task requirements and production plan. Ultimately, the tool health index and usage plan can not only improve production efficiency, but also reduce the risk of production interruption caused by tool failure, and ensure the efficient and stable operation of the manufacturing process. The implementation of this technical solution provides strong support for tool management and maintenance in complex manufacturing environments.

[0131] The above describes the tool identification method in the embodiment of the present invention. The following describes the identification device in the embodiment of the present invention. Figure 2 , an embodiment of the discrimination device in the embodiment of the present invention includes:

[0132] A digital twin model acquisition module 101 is used to acquire a digital twin model of a tool in a tool magazine, wherein the digital twin model includes geometric feature parameters, material property information, and processing task data of the tool;

[0133] A similarity calculation module 102 is used to calculate the similarity matrix between tools based on the geometric feature parameters, material property information and processing task data in the digital twin model, and generate a tool association network topology structure according to a preset similarity threshold;

[0134] The data fusion module 103 is used to collect tool status data during the process of tools entering and exiting the tool magazine, and fuse it with the tool history usage data to form a tool comprehensive evaluation data set;

[0135] A health evaluation module 104, configured to calculate a tool health index through a preset performance evaluation algorithm based on the tool comprehensive evaluation data set and the tool associated network topology structure;

[0136] The intelligent scheduling module 105 is used to generate a tool usage plan and a preventive maintenance plan through an intelligent scheduling algorithm according to the tool health index and the current processing task requirements.

[0137] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A tool identification method, characterized in that: include: Acquire a digital twin model of a tool in a tool magazine, wherein the digital twin model includes geometric feature parameters, material property information, and processing task data of the tool; Based on the geometric feature parameters, material property information and processing task data in the digital twin model, a similarity matrix between tools is calculated, and a tool association network topology structure is generated according to a preset similarity threshold; When tools are in and out of the tool magazine, tool status data is collected and integrated with tool history usage data to form a comprehensive tool evaluation data set. Based on the tool comprehensive evaluation data set and the tool associated network topology structure, the tool health index is calculated by a preset performance evaluation algorithm, and the tool health index is calculated by a preset performance evaluation algorithm based on the tool comprehensive evaluation data set and the tool associated network topology structure, including: performing wavelet decomposition processing on the tool comprehensive evaluation data set to obtain tool wear trend characteristics, machining accuracy fluctuation characteristics and thermal stress distribution characteristics; performing network feature calculation on each tool according to the tool associated network topology structure to obtain the network centrality and clustering coefficient of the tool; and combining the tool wear trend characteristics, machining accuracy fluctuation characteristics and thermal stress distribution characteristics into a network centrality and clustering coefficient. , network centrality and clustering coefficient are input into a pre-built five-layer feedforward neural network to obtain an initial tool health score; the weights of the last two layers of the five-layer feedforward neural network are back-propagated and updated to obtain an optimized neural network model for the current task; the tool wear trend characteristics, machining accuracy fluctuation characteristics, thermal stress distribution characteristics, network centrality and clustering coefficient are input into the optimized neural network model, and processed by the Softmax function to obtain a corrected tool health score; Mamdani fuzzy reasoning operation is performed on the corrected health score, the difficulty coefficient of the current machining task, material properties and precision requirements to obtain a tool health index; According to the tool health index and the current processing task requirements, a tool usage plan and a preventive maintenance plan are generated through an intelligent scheduling algorithm.

2. The tool identification method according to claim 1, characterized in that: The step of obtaining a digital twin model of a tool in a tool magazine, wherein the digital twin model includes geometric feature parameters, material property information, and processing task data of the tool, includes: Identify the tools in the tool magazine and obtain the unique identification code of the tools; querying a tool information database according to the unique identification code; If corresponding complete information is found in the tool information database, the corresponding complete information includes geometric feature parameters, material property information and processing task data of the tool, and a digital twin model is constructed based on the corresponding complete information; If the corresponding complete information is not found, perform the following steps: Perform three-dimensional scanning on the tool to obtain the geometric profile data of the tool; Based on the geometric profile data, combined with a preset tool parameter template, the geometric feature parameters of the tool are generated; Obtain the material composition data of the tool and determine the material property information of the tool based on the material performance database; Access the production management system, obtain the historical processing task records associated with the tool, and extract the processing task data; The geometric feature parameters, material property information and processing task data are integrated to construct a digital twin model.

3. The tool identification method according to claim 2, characterized in that: The construction of the digital twin model includes: Performing parameterization processing on the geometric feature parameters to obtain a high-precision three-dimensional geometric model of the tool; Mapping the material property information onto the high-precision three-dimensional geometric model through a multi-scale material model to obtain a tool virtual entity with microstructure information; Performing deep learning analysis on the processing task data to obtain the evolution law of tool performance; Associating the tool performance evolution law with the tool virtual entity to obtain a dynamic performance prediction model; Constructing a two-way data interaction interface between the virtual entity of the tool and the actual tool to achieve real-time status synchronization; The high-precision three-dimensional geometric model, the tool virtual entity, the dynamic performance prediction model and the two-way data interaction interface are integrated to obtain a digital twin model of the tool.

4. The tool identification method according to claim 3, characterized in that: The step of constructing a two-way data interaction interface between the virtual tool entity and the actual tool to achieve real-time state synchronization includes: Design tool data exchange protocol to define the data structure and communication format between tool virtual entity and actual tool; Based on the tool data exchange protocol, the actual tool working parameters are collected and processed to generate a standardized tool status data packet; Using the tool status data packet to update corresponding parameters in the virtual entity, and generating a control instruction data packet according to an optimization algorithm of the virtual entity; The control instruction data packet is transmitted to the actual tool through the tool data exchange protocol to complete two-way data interaction and status synchronization.

5. The tool identification method according to claim 1, characterized in that: The method of calculating the similarity matrix between tools based on the geometric feature parameters, material property information and processing task data in the digital twin model, and generating a tool association network topology structure according to a preset similarity threshold, includes: Performing feature vectorization processing on geometric feature parameters, material property information and processing task data in the digital twin model to obtain a tool feature vector set; Using a multi-dimensional similarity calculation algorithm, the vectors in the tool feature vector set are compared pairwise to obtain a similarity matrix between tools; According to the preset similarity threshold, binarizing the inter-tool similarity matrix to obtain a tool association adjacency matrix; Based on the tool association adjacency matrix, a tool association network topology structure is constructed by using a graph theory algorithm.

6. The tool identification method according to claim 1, characterized in that: During the process of tools entering and exiting the tool magazine, tool status data is collected and merged with tool history usage data to form a comprehensive tool evaluation data set, including: When the tool enters and exits the tool magazine, the tool status data is obtained, wherein the tool status data includes a tool geometry profile change diagram, a surface microstructure diagram, and a tool-handle interface stress distribution diagram; Extracting tool history usage data of tools entering and leaving the tool magazine from the production management system, wherein the tool history usage data includes a processing process parameter sequence and a processing quality index record; Performing cross-modal feature extraction on the tool state data to obtain a tool state feature vector; Using a time series prediction algorithm, generating tool future usage trend prediction data based on the tool historical usage data; A data fusion algorithm based on an attention mechanism is used to adaptively assign weights to and integrate the tool state feature vector and the future usage trend prediction data to obtain a comprehensive tool evaluation data set.

7. The tool identification method according to claim 1, characterized in that: According to the tool health index and the current processing task requirements, a tool usage plan and a preventive maintenance plan are generated through an intelligent scheduling algorithm, including: Performing multi-objective optimization processing on the tool health index and the current machining task requirements to obtain a tool-task matching score; The tool-task matching score, the number of available tools in the tool magazine, and the process sorting constraints are input into an intelligent scheduling algorithm based on deep reinforcement learning to obtain a tool usage plan; Comparing and analyzing the expected usage time of each tool in the tool usage plan with its health index, and obtaining the critical usage time point of the tool; The tool critical use time point and production plan are input into an intelligent maintenance algorithm based on constraint programming to obtain a tool preventive maintenance plan.

8. A discrimination device, characterized in that: The identification device adopts the tool identification method according to any one of claims 1 to 7, and the identification device comprises: A digital twin model acquisition module is used to acquire a digital twin model of a tool in a tool magazine, wherein the digital twin model includes geometric feature parameters, material property information, and processing task data of the tool; A similarity calculation module, used to calculate the similarity matrix between tools based on the geometric feature parameters, material property information and processing task data in the digital twin model, and generate a tool association network topology structure according to a preset similarity threshold; The data fusion module is used to collect tool status data during the process of tools entering and leaving the tool magazine, and fuse it with the tool history usage data to form a comprehensive tool evaluation data set; A health evaluation module, used to calculate the tool health index through a preset performance evaluation algorithm based on the tool comprehensive evaluation data set and the tool associated network topology structure; The intelligent scheduling module is used to generate a tool usage plan and a preventive maintenance plan through an intelligent scheduling algorithm according to the tool health index and the current processing task requirements.

Images