A method and system for dynamic offloading of edge cloud collaborative computing tasks for numerical control machining process flow

Abstract

This invention discloses a method and system for dynamic task allocation in edge-cloud collaborative computing for the entire CNC machining process. The system includes a process analysis module for real-time identification of process boundaries and technological features; a machining task perception module for automatically generating a list of various computing tasks required for the current process based on process features (including tool wear monitoring, spindle vibration monitoring, thermal error compensation, machining parameter optimization, workpiece quality prediction, contour error compensation, etc.); a process-driven dynamic decision engine for independently deciding whether to allocate each computing task to the edge or the cloud; and a CNC instruction collaboration module for converting the decision results into compensation instructions and writing them into the CNC system. This invention combines process flow features with various computing task types throughout the CNC machining process, achieving refined dynamic allocation at the process and task levels. It also ensures task continuity through the migration of intermediate computing results during process switching, making it suitable for various intelligent application scenarios throughout the entire CNC machining process.

Description

[0001] This invention belongs to the intersection of CNC machining technology and edge computing, specifically involving a dynamic task distribution method and system for edge-cloud collaborative computing for CNC machine tool machining process flow. It is applicable to CNC machining scenarios with high requirements for real-time performance and accuracy, such as tool wear monitoring, spindle status monitoring, machining parameter optimization, workpiece quality prediction, and contour error compensation. Background Technology

[0002] CNC machine tools, as the "mother machines of industry," are evolving towards intelligence and networking. Industry forecasts predict that by 2026, over 45% of CNC equipment will be interconnected through the Industrial Internet of Things (IIoT), and edge computing modules directly embedded in CNC controllers can reduce data processing latency to below 10 milliseconds. Against this backdrop, edge-cloud collaborative computing architecture has become a key technological path to resolve the contradiction between real-time performance and computing power in CNC machining.

[0003] When intelligent machine tools operate in continuous machining environments, the coupling effect of multi-component degradation and machining task updates leads to massive multi-source sensor data streams and a large number of computational tasks with complex dependencies. Strict real-time constraints and complex data dependency structures pose challenges to traditional single-computing frameworks. Currently, research on edge-cloud collaborative computing for CNC machine tools mainly focuses on the following aspects: Patent CN109933004B discloses a latency-optimized load task migration algorithm in an edge computing environment, which schedules and sorts tasks based on load balancing and task quality. However, this patent takes latency optimization and load balancing as the single optimization objective and does not combine the characteristics of CNC machining process flow to make differentiated flow decisions.

[0004] Patent CN121325765A discloses a CNC machine tool data acquisition device and data processing method with distributed edge computing capabilities. It forms a distributed edge computing cluster in a local area network through a resource discovery and communication module, realizing dynamic allocation and load balancing of computing tasks. However, this solution focuses on load balancing within the distributed cluster and does not involve a dynamic traffic splitting mechanism between edge nodes and the cloud.

[0005] The paper "STEP-NC enabled edge–cloud collaborative manufacturing system for compliant CNC machining" proposes an edge-cloud collaborative manufacturing system based on STEP-NC, supporting the exchange of machining data marked by machining steps, and proposing an environment-based dynamic task delivery and data subscription method to improve data traceability. However, this system uses machining step tags for data distribution, mainly for data traceability purposes, and focuses limitedly on the dynamic distribution decision algorithm for computing tasks between the edge and the cloud.

[0006] The paper "An Adaptive Hybrid Edge-Cloud Collaborative Offloading Method for Large-Scale Computational Tasks of Intelligent Machine Tool" proposes a hybrid offloading mechanism combining unilateral and multilateral cloud collaboration. This mechanism dynamically switches collaboration modes based on compute node status, task characteristics, dependency complexity, and resource availability. Its effectiveness was verified on a digital twin five-axis machining center, resulting in an average task processing time improvement of 27.36% and a latency reduction of 37.03%. However, this study primarily uses task characteristics as the basis for offloading decisions and does not incorporate the technological characteristics of the CNC machining process flow into the decision-making dimension.

[0007] In summary, existing edge-cloud collaborative computing task offloading methods have the following technical shortcomings: (i) Disconnect between task offloading decision and process characteristics: Existing task offloading methods mainly focus on minimizing latency and load balancing as optimization objectives, lacking awareness and utilization of the characteristics of CNC machining processes. Different processes have significant differences in the real-time requirements, accuracy requirements, and data security levels of computing tasks, but existing methods do not incorporate these process characteristics into the task offloading decision model; (ii) Limited methods for identifying process boundaries: Existing technologies for identifying processes mostly rely on step labels in the STEP-NC standard or statistical predictions based on historical data. The former requires the CNC system to support the STEP-NC interface, limiting its applicability; the latter is an offline prediction method and cannot perceive process boundaries in real time. No existing technology has proposed a process boundary identification method based on real-time parsing of G / M codes.

[0008] (iii) Lack of task continuity during process switching: When a workpiece switches from one process to the next, the intermediate calculation results accumulated at the edge (such as the cumulative value of tool wear and the workpiece quality trend) cannot be transferred with the process, resulting in a waste of computing resources and a decrease in prediction accuracy.

[0009] (iv) The decision results are disconnected from the closed loop of the CNC system: Most existing edge-cloud collaborative solutions are limited to the data acquisition layer or monitoring layer. The calculation results are difficult to be fed back to the CNC system in real time, and the machining parameters based on the diversion decision cannot be compensated in real time.

[0010] Therefore, there is an urgent need for a method and system for dynamic task allocation of edge-cloud collaborative computing that can sense the characteristics of CNC machining processes, achieve dynamic task allocation at process boundaries, and ensure task continuity during process switching. Summary of the Invention

[0011] The purpose of this invention is to provide a dynamic task splitting method and system for edge-cloud collaborative computing oriented to CNC machining process flow. By analyzing the CNC machining program in real time to identify process boundaries and technological features, a multi-dimensional constraint decision model is established driven by process features to realize the dynamic splitting of computing tasks between the edge and the cloud, ensuring task continuity during process switching, and feeding the decision results back to the CNC system in real time to achieve closed-loop compensation, thereby maximizing computing accuracy and resource utilization while meeting the real-time constraints of machining.

[0012] To achieve the above objectives, the present invention adopts the following technical solution: On one hand, the present invention provides a dynamic task offloading system for edge-cloud collaborative computing oriented to CNC machining process flow, characterized in that it includes: S10, Process analysis module, configured to analyze CNC machining programs in real time, identify process boundaries and extract the process features of each process, wherein the process features include at least process type, machining allowance, surface quality requirements, process priority and cutting parameters; S20. Machining task perception module, deployed at the edge, configured to identify the type of calculation task triggered by the current process. The calculation task type includes one or more of the following: tool wear monitoring task, spindle vibration monitoring task, thermal error compensation task, machining parameter optimization task, workpiece quality prediction task, and contour error compensation task. S30, a multi-dimensional state awareness module, is deployed at the edge and configured to collect real-time data on the edge's computing load, memory usage, inference queue length, as well as current network bandwidth, latency, and data transmission quality parameters. S40, a process-driven dynamic decision engine, configured to construct a multi-objective constraint decision model based on process characteristics, current computing task type, edge real-time status and network status, and dynamically determine the destination of various computing tasks generated by the current process. The destination includes edge execution, cloud execution or hierarchical execution mode. S50, the CNC instruction coordination module, is configured to convert the flow splitting decision results (including tool wear compensation value, thermal error compensation value, contour error compensation value, optimized machining parameters, etc.) into compensation instructions that can be recognized by the CNC system, and write them into the bias register or macro variable of the CNC system via fieldbus or OPC UA protocol.

[0013] Furthermore, the machining task perception module automatically infers the types of computational tasks that may be generated by analyzing the process characteristics: for example, the finishing process automatically triggers the contour error compensation task and the high-frequency vibration monitoring task; the roughing process automatically triggers the tool wear monitoring task and the machining parameter optimization task; and the long-term continuous machining process automatically triggers the thermal error compensation task.

[0014] Furthermore, the process-driven dynamic decision engine employs an extended five-dimensional constraint decision function: Sdecision = f (Pprocess, Rtask, Ledge, Nnetwork, Dsecurity) in: • Pprocess is the process feature vector; • Rtask is the real-time requirement factor for the current computation task type (e.g., tool wear monitoring requires a response time of ≤10ms, while process parameter optimization allows a delay of ≥500ms). • Ledge is the real-time load parameter at the edge; • Nnetwork is a network state parameter; • Dsecurity is a data security level parameter.

[0015] The decision engine calculates the task allocation score based on the decision function. When the score is higher than a preset first threshold, the task is executed at the edge. When the score is lower than a preset second threshold, the task is allocated to the cloud for execution. When the score is between the first and second thresholds, a layered execution mode of edge screening and cloud-based fine judgment is adopted.

[0016] Furthermore, the system also includes a tool condition monitoring module, but the present invention is not limited thereto. It also includes a spindle vibration monitoring module, a thermal error monitoring module, a workpiece quality monitoring module, etc. Each module can be dynamically enabled or disabled according to process characteristics.

[0017] Furthermore, the system also includes a process switching task migration module, which is configured to migrate intermediate calculation results such as the tool wear accumulation value, spindle vibration characteristic baseline, thermal error trend, and workpiece quality prediction deviation accumulated at the edge of the current process to the edge node corresponding to the subsequent process when the workpiece is detected to switch from the current process to the next process. The differential migration strategy is used to transmit only the changes.

[0018] On the other hand, the present invention also provides a method for dynamic task distribution of edge-cloud collaborative computing for the entire CNC machining process, characterized by including the following steps: Step S1: CNC machining program parsing and process boundary identification. The G-code / M-code instruction stream of the CNC system is acquired in real time. Using tool change commands or significant changes in feed rate as signals, process boundaries are identified and the technological features of the current process are extracted.

[0019] Step S2: Task perception and generation throughout the machining process. Based on the process characteristics, an automatic list of calculation tasks required for the current process is generated. The calculation tasks include at least one of tool wear monitoring, spindle vibration monitoring, thermal error compensation, machining parameter optimization, workpiece quality prediction, and contour error compensation.

[0020] Step S3: Multi-dimensional status acquisition. Acquire real-time load status, network connectivity status, and data security level of the current process at the edge.

[0021] Step S4: Process-driven dynamic decision-making. Based on the process characteristics extracted in Step S1, the types of computational tasks generated in Step S2 and their real-time requirements, and the multi-dimensional status collected in Step S3, a decision model is constructed. A routing score is calculated for each computational task to determine the routing destination of each task.

[0022] Step S5: Task offloading and execution. Based on the decision results of step S4, different computing tasks are allocated to edge devices for execution, cloud devices for execution, or a tiered execution mode.

[0023] Step S6: Closed-loop compensation feedback. The calculated tool wear compensation value, thermal error compensation value, optimized feed rate, and other results from the edge or cloud are converted into compensation commands that the CNC system can recognize, and written into the bias register or macro variable of the CNC system via the fieldbus.

[0024] Step S7: Process Switching Task Migration. When a process switch is detected, the various intermediate calculation results accumulated in the current process are migrated to the edge nodes corresponding to the subsequent processes.

[0025] Step S8: Edge-Cloud Collaborative Evolution. The local machining data collected at the edge is anonymized and uploaded to the cloud. The cloud uses global data to train and update model parameters such as tool wear prediction, thermal error compensation, and workpiece quality prediction. After generating a lightweight model through knowledge distillation, it is distributed to the edge.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Process-Driven Refined Task Offloading Decisions: For the first time, the process characteristics of CNC machining processes (process type, machining allowance, surface quality requirements, process priority) are used as the core input for computational task offloading decisions, enabling offloading strategies to adapt to the differentiated needs of different processes. For example, the high-frequency vibration monitoring task in the finishing process is forced to be executed at the edge to ensure real-time performance, while the large-scale process parameter optimization task in the roughing process is offloaded to the cloud to utilize high computing power.

[0027] 2. Process boundary identification through real-time parsing of G-code / M-code: By deploying a lightweight instruction parser at the edge, process boundaries are identified in real time using tool change instructions and significant changes in feed rate as signals. This eliminates the need to rely on the STEP-NC standard interface, resulting in wider applicability and a lower application threshold.

[0028] 3. Ensuring task continuity during process switching: For the first time, a mechanism for transferring intermediate calculation results during process switching is proposed, which transfers knowledge such as the accumulated tool wear value and workpiece quality trend accumulated at the edge to subsequent processes, avoiding cold starts and repeated calculations, and improving the efficiency of edge-cloud collaboration.

[0029] 4. Closed-loop integration of decision-making and CNC system: The result of the split decision is converted into compensation instructions in real time and written into the CNC bias register to realize the millisecond-level closed loop of "edge cloud decision-making - CNC execution", which makes up for the defect of the existing edge cloud collaboration solution and the CNC system being disconnected.

[0030] 5. Tool status adaptive flow division based on process characteristics: The flow division strategy is dynamically adjusted according to the tool wear stage. During the period of rapid tool wear, the monitoring frequency is automatically increased and edge execution is forced, which effectively ensures machining safety. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention.

[0032] Figure 2 This is a flowchart of the method of the present invention.

[0033] Figure 3 This is a schematic diagram illustrating the working principle of the process-driven dynamic decision engine of this invention.

[0034] Figure 4 This is a schematic diagram of the workflow of the process switching task migration module of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the embodiments of this invention will be further described in detail below with reference to the accompanying drawings. Figure 1 As shown, the present invention provides a dynamic task distribution system for edge-cloud collaborative computing of the entire CNC machining process, including an edge terminal deployed on the CNC machine tool and a cloud terminal deployed on a remote server. The edge terminal and the cloud terminal are connected through the Industrial Internet.

[0036] The entire process is as follows Figure 2As shown, it includes 8 steps. Taking the machining of aerospace structural parts by a five-axis machining center as an example, the CNC program includes three operations: Operation 1 is roughing (tool T01, feed rate F2000, spindle speed S5000), Operation 2 is semi-finishing (tool T02, feed rate F1200, spindle speed S8000), and Operation 3 is finishing (tool T03, feed rate F600, spindle speed S12000).

[0037] The system uses G-code parsing to automatically generate a list of corresponding calculation tasks at the start of each process: • Process 1 (Rough Machining): The main tasks are tool wear monitoring (high real-time requirement, 10ms response) and machining parameter optimization (low real-time requirement, 500ms latency allowed). Based on process characteristics (rough machining, medium priority), tool status (new tool), edge load (30%), and good network conditions, the decision engine decides: the tool wear monitoring task will be executed at the edge (score 0.91), while the machining parameter optimization task will be offloaded to the cloud (score 0.28, low real-time requirement and high computing power requirement).

[0038] • Process 2 (Semi-finishing): Automatically generates three tasks: tool wear monitoring, spindle vibration monitoring, and thermal error compensation. Considering that semi-finishing has certain requirements for surface quality, the decision engine assigns spindle vibration monitoring (real-time requirement 20ms) and thermal error compensation (real-time requirement 50ms) to the edge for execution, while tool wear monitoring (the tool has used 30% of its life but has not yet reached the critical wear stage) is still assigned to the edge for execution.

[0039] • Process 3 (Finishing): Automatically generate three tasks: contour error compensation (real-time requirement 5ms), high-frequency vibration monitoring (real-time requirement 10ms), and workpiece quality prediction (real-time requirement 100ms). The decision engine forces contour error compensation and high-frequency vibration monitoring to be executed at the edge (scores 0.95 and 0.93), while workpiece quality prediction, due to its allowance for slightly larger delays and requirement for a high-precision model, is assigned to be executed in the cloud (score 0.42).

[0040] Throughout the machining process, the system dynamically senses process transitions. When process 1 ends and process 2 begins, the accumulated tool wear value (0.12mm) from process 1 and the optimized machining parameters (optimized feed rate curve) are migrated to the edge nodes of process 2 for reference in subsequent processes. The workflow of the process transition task migration module is as follows: Figure 4 As shown.

[0041] This embodiment focuses on demonstrating how the tool wear monitoring task can be collaboratively offloaded with other tasks within the workflow framework, but the scope of protection of this invention is not limited to tool wear monitoring. Figure 3The diagram shows the working principle of the dynamic decision engine driven by the process step. During the semi-finishing process of step 2, when the edge-end real-time inference indicates that tool T02 has entered a period of rapid wear (remaining life <15%), the decision engine automatically triggers the following adjustments: increasing the sampling frequency of the tool wear monitoring task from 1Hz to 10Hz, switching the task from the original "edge-end routine execution" to "edge-end forced high-priority execution," and temporarily suspending the machining parameter optimization task for this step to ensure the real-time nature of the safety warning. After the tool change (M06 T03), the system automatically restores the default flow distribution strategy.

[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A dynamic task distribution system for edge-cloud collaborative computing across the entire CNC machining process, characterized in that, include: S10, Process Analysis Module, configured to analyze CNC machining programs in real time, identify process boundaries, and extract the process features of each process; S20. Machining task perception module, configured to automatically generate a list of calculation tasks required for the current process based on the process characteristics of the process. The list of calculation tasks includes at least one of tool wear monitoring task, spindle vibration monitoring task, thermal error compensation task, machining parameter optimization task, workpiece quality prediction task, and contour error compensation task. S30, a multi-dimensional state awareness module, is deployed at the edge and configured to collect real-time data on the edge's computing load, memory usage, inference queue length, as well as current network bandwidth, latency, and data transmission quality parameters. S40, a process-driven dynamic decision engine, configured to construct a multi-objective constraint decision model based on process characteristics, the real-time requirements of the current computing task, the real-time status of the edge, and the network status, and dynamically determine the routing destination for each computing task, wherein the routing destination includes edge execution, cloud execution, or layered execution mode. S50, the CNC instruction coordination module, is configured to convert the diversion decision result into a compensation instruction that the CNC system can recognize and write it into the bias register or macro variable of the CNC system.

2. The system according to claim 1, characterized in that, The machining task perception module automatically infers the following by analyzing the process type: when the process type is finishing, it automatically triggers the contour error compensation task and the high-frequency vibration monitoring task; when the process type is roughing, it automatically triggers the tool wear monitoring task and the machining parameter optimization task; when the continuous machining time exceeds the preset threshold, it automatically triggers the thermal error compensation task.

3. The system according to claim 1, characterized in that, The process-driven dynamic decision engine employs a decision function that incorporates factors related to task real-time requirements. Sdecision = f (Pprocess, Rtask, Ledge, Nnetwork, Dsecurity) Rtask represents the maximum response time required by the current computation task.

Images