Manufacturing process parameters and detection optimization methods and systems based on dual-source axis position tables and AI agents

Abstract

This invention provides a method and system for optimizing manufacturing process parameters and inspection based on a dual-source axis position table and an AI agent, belonging to the field of intelligent manufacturing technology. This invention generates and aligns virtual and real axis position tables, extracts axis position difference features, and fuses them with part geometric deviation detection data to construct a causal reasoning model of error source—process parameters—equipment status. Under a multi-objective optimization framework, it outputs process parameter correction instructions and inspection scanning strategies, and combines incremental learning of the AI ​​agent to achieve iterative updates, thereby improving the accuracy of dynamic error diagnosis, the efficiency of process and inspection collaborative optimization, and the system convergence stability.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for optimizing production process parameters and detection based on a dual-source axis position table and an AI intelligent agent. Background Technology

[0002] The CNC machining and inspection process generally adopts a serial workflow of "offline process setting - online machining - post-inspection": the machining side mainly relies on the tool path and machine tool parameters generated by simulation, while the inspection side mostly uses fixed or regularized scanning paths to acquire point cloud / three-coordinate data, and then judges dimensional and geometric errors based on this. Process optimization is mostly driven by human experience, and anomaly diagnosis mainly traces back to possible equipment or parameter problems through the statistical correlation of the final part deviation; the data source is usually single-source (either simulation or actual measurement), and the two lack consistency analysis after spatiotemporal alignment.

[0003] As the application of digital twins and AI in manufacturing expands, the industry is evolving from outcome monitoring to process awareness, causal diagnosis, and closed-loop optimization. Key trends include: the fusion of virtual / real multi-source data; interpretable error attribution under physical constraints; and the collaborative optimization and online updating of processing parameters and detection strategies. The goal is to shorten the optimization cycle from multiple rounds of trial cutting to a few rounds of convergent closed-loop control.

[0004] Existing solutions still suffer from three major shortcomings: First, the simulation and measured axis motion data are not strictly aligned in time and modeled for differences, making it difficult to decouple dynamic errors such as servo hysteresis, thermal drift, and backlash. Second, AI models tend to focus on prediction rather than "causal reasoning under physical constraints," making it difficult to output operable error contribution rates and explanations of physical mechanisms. Third, machining and inspection are often optimized separately, and the scanning path and sampling density are not adaptively reconstructed using process-side error information, resulting in inspection redundancy and slow optimization convergence. These deficiencies restrict the stable achievement of high-precision parts and the improvement of efficiency. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a manufacturing process parameter and detection optimization method and system based on a dual-source axis position table and an AI agent. By introducing time-series alignment difference modeling of virtual and real dual-source axis position tables, combining causal reasoning and physical constraint embedding mechanisms of AI agents, and realizing the collaborative updating of processing parameters correction and detection scanning strategies under a multi-objective optimization framework, the accuracy of dynamic error diagnosis, the efficiency of process and detection linkage optimization, and the overall convergence stability of the system are significantly improved.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for optimizing manufacturing process parameters and inspection based on a dual-source axis positioning table and an AI agent, comprising:

[0008] Based on the digital model and process plan of the target part, a machining trajectory to be executed is generated, and the machining trajectory is run in a simulation environment to obtain a virtual axis position table containing timestamps and multi-axis positions.

[0009] The machining trajectory is executed on the machine tool, and a real axis position table containing timestamps and multi-axis positions is collected in real time. After machining is completed, detection data characterizing the geometric deviation distribution of the part is obtained.

[0010] The virtual axis table is time-aligned with the real axis table, a multi-axis position difference sequence is calculated to form axis difference features, and the axis difference features are fused with the detection data to form joint spatiotemporal features for diagnosis.

[0011] Under the premise of complying with the kinematics and dynamics boundaries and process safety constraints of the machine tool, a causal reasoning model of error source-process parameter-equipment state is constructed based on the joint spatiotemporal characteristics and the preset AI agent. The model outputs the contribution weight and mechanism explanation for multiple candidate error sources. With "geometric tolerance satisfaction, processing cost, and inspection time" as multiple objectives, the model solves for the process parameter correction instruction set for the processing process and the adaptive scanning strategy for the inspection process.

[0012] The process parameter correction instruction set and the adaptive scanning strategy are respectively sent to the processing and inspection stages for execution, and a new round of inspection data of the real axis position table and geometric deviation distribution is collected. The causal reasoning model is continuously updated through the incremental learning mechanism of the AI ​​agent until the preset accuracy is met or the convergence criterion is reached.

[0013] Preferably, a machining trajectory to be executed is generated based on the digital model and process plan of the target part, and the machining trajectory is run in a simulation environment to obtain a virtual axis position table containing timestamps and multi-axis positions, including:

[0014] Extract geometric features from the three-dimensional digital model of the target part; the geometric features include at least the surface contour, boundary conditions and machining allowance of the part;

[0015] Obtain process constraints from the process plan; the process constraints include at least the machining sequence, cutting posture restrictions, and feed rate range.

[0016] In a CNC simulation environment, the geometric features and process constraints are used as inputs to generate a machining trajectory to be executed and run dynamically.

[0017] During dynamic operation, multi-axis position data is output according to a preset sampling frequency, and a timestamp is added to each position data to form the virtual axis position table.

[0018] Preferably, the machining trajectory is executed on a machine tool, a real axis position table including timestamps and multi-axis positions is acquired in real time, and detection data characterizing the geometric deviation distribution of the part is obtained after machining, including:

[0019] The machining trajectory is loaded onto the CNC machine tool and the machining process is started, driving the multi-axis servo system to move according to the machining trajectory;

[0020] During the machining process, the position command value and position feedback value of each axis are acquired in real time through the feedback channel of the machine tool control system, and a sampling timestamp is marked for each set of position command value and position feedback value of each axis to form a real-time multi-axis position sequence.

[0021] The multi-axis position sequence is stored as a real axis position table in the order of sampling time, which is used for subsequent comparison with the virtual axis position table;

[0022] After machining, the parts are geometrically inspected to collect surface point data based on a coordinate measuring machine or laser scanner.

[0023] The collected point data is fitted and aligned to obtain the detection data that characterizes the geometric deviation distribution between the part and the target digital model.

[0024] Preferably, the virtual axis position table is temporally aligned with the real axis position table, a multi-axis position difference sequence is calculated to form axis position difference features, and the axis position difference features are fused with the detection data to form joint spatiotemporal features for diagnosis, including:

[0025] The virtual axis position table and the real axis position table are unified in terms of time reference. Different sampling clocks are aligned using monotonic time mapping. Under the condition of satisfying the machine tool motion continuity constraint, the optimal time registration function is obtained to minimize the mean square error between the two in the entire time domain. The time registration formula is as follows:

[0026] ;

[0027] in, For a moment The true multi-axis position vector; A virtual multi-axis position vector; It is a monotonic time registration function; This is the optimal time registration function; for The derivative; The time smoothing penalty coefficient is obtained from the machine tool servo bandwidth calibration; The allowable deviation for time scaling is determined by the machine tool synchronization accuracy specification; and These represent the total duration of the real and virtual sequences, respectively.

[0028] A multi-axis position difference sequence is calculated on a unified time scale to form an axis position difference matrix, which is used to characterize the time-varying distribution of dynamic errors. The definition formula of the axis position difference matrix is:

[0029] ;

[0030] in, For a moment The multi-axis position difference vector reveals the dynamic errors of servo hysteresis, thermal drift, and backlash;

[0031] The axis difference matrix and the detected geometric deviation field are spatially correlated and time-traced in the coordinate system of the part surface, and the spatiotemporal projection result from process error to shape deviation is completed under the tool center path mapping of the tool trajectory. The mapping and fusion formula is as follows:

[0032] ;

[0033] in, A spatial point on the surface of the workpiece; The geometric deviation obtained from the detection; This represents the path of the tool center on the workpiece surface. Geometric sensitivity to shaft disturbances and tool / surface errors; Differences in process-side location;

[0034] An adaptive fusion mechanism is introduced into the spatiotemporal projection results to balance the contributions of the process side and the detection side. The fusion weights and kernel function are defined as follows:

[0035]

[0036] in, The empirical variance of geometric deviation; The empirical variance of axis position differences; For adaptive fusion coefficients; The kernel function is scaled by the feed step size; The core width is determined by the feed rate. With sampling interval Automatically determined.

[0037] Preferably, the process parameter correction instruction set for the machining process includes at least adjustments to one or more of the following: feed, acceleration / deceleration profile, tool compensation, and cooling strategy; the adaptive scanning strategy for the inspection process includes at least the configuration of one or more of the following: scanning path, sampling density, and measurement priority.

[0038] Preferably, under the premise of complying with the kinematics and dynamics boundaries and process safety constraints of the machine tool, a causal reasoning model of error source-process parameter-equipment state is constructed based on the joint spatiotemporal characteristics and the preset AI agent. This model outputs contribution weights and mechanism explanations for multiple candidate error sources. Furthermore, using "geometric tolerance satisfaction, machining cost, and inspection time" as multiple objectives, a process parameter correction instruction set for the machining process and an adaptive scanning strategy for the inspection process are obtained, including:

[0039] The AI ​​agent accesses the joint spatiotemporal features, integrates the dynamic error features extracted from the sequence of axis difference features with the geometric deviation of the 3D scan / point cloud on the detection side and the comparison with CAD, and forms a comprehensive input set for attribution reasoning;

[0040] The AI ​​agent, embedded in the boundaries of machine tool dynamics and kinematics, uses Bayesian networks to perform causal reasoning, establishes the relationship between error sources, process parameters, and equipment status, and outputs the contribution weight, confidence assessment, and physical mechanism description of each candidate error source.

[0041] The AI ​​agent generates an interpretable attribution report; the attribution report includes at least the dominant error source, contribution weight and corresponding physical mechanism, which is used to drive collaborative optimization decisions on the processing and detection sides.

[0042] The AI ​​agent constructs and solves a collaborative optimization problem with the goal of improving geometric tolerance satisfaction and reducing machining costs and inspection time, under the premise of satisfying machine tool kinematics, dynamics and process safety constraints. This enables the machining side and the inspection side to converge simultaneously and produce two types of control variables in the same optimization round.

[0043] The AI ​​agent generates a set of process parameter correction instructions; the set of process parameter correction instructions includes at least one or more adjustments to feed rate, acceleration / deceleration profile, tool compensation and cooling strategy, and modifies relevant CNC instruction segments as necessary to suppress error propagation chains;

[0044] The AI ​​agent generates an adaptive scanning strategy based on the error spatial distribution and the optimization trade-off.

[0045] Preferably, the adaptive scanning strategy includes at least one or more configurations of scanning path, sampling density and measurement priority, and increases sampling density and local coverage in areas of high confidence in curvature or deviation.

[0046] Preferably, the process parameter correction instruction set and the adaptive scanning strategy are respectively sent to the processing and inspection stages for execution, collecting a new round of inspection data of the real axis position table and geometric deviation distribution, so as to continuously update the causal inference model through the incremental learning mechanism of the AI ​​agent until the preset accuracy is met or the convergence criterion is reached, including:

[0047] The AI ​​agent is used to apply the process parameter correction instruction set obtained from the previous round of optimization to the machine tool machining process, and the adaptive scanning strategy is used for the geometric inspection process; the entire process from the start of execution of the process parameter correction instruction set and the adaptive scanning strategy to the generation of the new optimized output is defined as iterative loop and start-up execution.

[0048] During the iterative rounds, timestamped multi-axis position feedback is collected to generate a new true axis position table, and surface deviations based on 3D scans or point clouds are obtained to form a geometric deviation distribution; the true axis position table, geometric deviation distribution, and execution logs are compiled into an iterative data package for learning and evaluation;

[0049] Based on the AI ​​agent, the target evaluation vector is calculated according to the iterative data package. The components are geometric tolerance satisfaction, processing cost and detection time, which are used to quantify the performance of the iterative rounds and serve as the basis for subsequent optimization and stopping judgment.

[0050] Based on the AI ​​agent, the causal reasoning model is incrementally learned and updated using the iterative data package as a new sample, while maintaining the embedding of machine tool kinematics and dynamics boundaries and process safety constraints to ensure the physical consistency and interpretability of the attribution.

[0051] Based on the updated causal inference results and the target evaluation vector, the re-optimization of the processing side process parameters and the detection side scanning strategy is solved simultaneously, generating process parameter correction instructions and adaptive scanning strategies for the next iteration round, thus achieving the coordinated convergence of the processing-detection dual closed loop;

[0052] When the target evaluation vector meets the preset accuracy threshold or the improvement in consecutive iterations is less than the preset threshold, the AI ​​agent determines that the convergence criterion has been met and terminates the iteration; if the convergence criterion has not been met, the AI ​​agent continues to execute a new iteration until the accuracy is met or the stopping condition is triggered.

[0053] A manufacturing process parameter and inspection optimization system based on a dual-source axis positioning table and an AI agent includes:

[0054] The machining trajectory generation unit is used to generate a machining trajectory to be executed based on the digital model and process plan of the target part, and run the machining trajectory in a simulation environment to obtain a virtual axis position table containing timestamps and multi-axis positions.

[0055] The real data acquisition unit is used to execute the machining trajectory on the machine tool, acquire a real axis position table containing timestamps and multi-axis positions in real time, and obtain detection data characterizing the geometric deviation distribution of the part after machining is completed;

[0056] The temporal alignment and feature fusion unit is used to temporally align the virtual axis table with the real axis table, calculate the multi-axis position difference sequence to form axis difference features, and fuse the axis difference features with the detection data into joint spatiotemporal features for diagnosis.

[0057] The AI ​​agent causal reasoning and optimization unit is used to construct a causal reasoning model of error source-process parameter-equipment state based on the joint spatiotemporal features and the preset AI agent, under the premise of complying with the kinematic and dynamic boundaries and process safety constraints of the machine tool. It outputs the contribution weight and mechanism explanation for multiple candidate error sources, and solves the process parameter correction instruction set for the machining process and the adaptive scanning strategy for the inspection process with "geometric tolerance satisfaction, machining cost and inspection time" as multiple objectives.

[0058] The closed-loop execution and incremental learning unit is used to send the process parameter correction instruction set and the adaptive scanning strategy to the processing and inspection stages for execution, and to collect a new round of detection data of the real axis position table and geometric deviation distribution. The causal reasoning model is continuously updated through the incremental learning mechanism of the AI ​​agent until the preset accuracy is met or the convergence criterion is reached.

[0059] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0060] This invention introduces a dual-source comparison mechanism between a virtual axis position table and a real axis position table, enabling real-time quantification of dynamic errors such as servo hysteresis, thermal drift, and backlash. This breaks through the bottleneck of being unable to accurately locate errors by relying on a single data source, and significantly improves the precision and real-time performance of error capture.

[0061] This invention constructs a causal reasoning model under physical constraints using an AI agent, which can output the dominant error sources and their contribution weights, and provide corresponding physical mechanism explanations, thus avoiding the uninterpretability of existing "black box predictions" and improving the usability and reliability of diagnosis.

[0062] This invention proposes an adaptive scanning strategy based on AI agents, which can dynamically adjust the scanning path and sampling density according to the geometric deviation distribution and curvature characteristics. This reduces unnecessary redundant data acquisition and improves the detection resolution of key areas, achieving a dual optimization of detection efficiency and effectiveness.

[0063] This invention integrates process parameter optimization and detection strategy optimization into the same optimization round for simultaneous solution. Under the premise of satisfying machine tool dynamics and process safety constraints, it achieves rapid convergence of the machining-detection dual closed loop, thereby shortening the iteration cycle and significantly reducing trial cutting costs and time.

[0064] This invention uses an incremental learning mechanism of AI agents to absorb new data and update the causal reasoning model in each iteration, enabling the optimization process to dynamically evolve with changes in equipment status and operating conditions, thus maintaining the long-term effectiveness and robustness of the optimization strategy.

[0065] This invention considers geometric tolerance satisfaction, processing cost, and inspection time simultaneously in multi-objective optimization. By quantifying the objective evaluation vector and setting convergence criteria, the optimization process can meet quality requirements while also taking into account economy and efficiency, thus having stronger engineering applicability in actual production environments. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0068] Figure 2 This is a schematic diagram of the overall process framework provided for embodiments of the present invention;

[0069] Figure 3 This is a schematic diagram of the hierarchical structure of the AI ​​agent provided in an embodiment of the present invention. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0072] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1As shown, this invention provides a manufacturing process parameter and detection optimization method based on a dual-source axis position table and an AI agent, including:

[0073] Step 100: Generate the machining trajectory to be executed based on the digital model and process plan of the target part, and run the machining trajectory in the simulation environment to obtain a virtual axis position table containing timestamps and multi-axis positions;

[0074] Step 200: Execute the machining trajectory on the machine tool, collect the real axis position table containing timestamps and multi-axis positions in real time, and obtain the detection data characterizing the geometric deviation distribution of the part after machining is completed;

[0075] Step 300: Perform temporal alignment between the virtual axis table and the real axis table, calculate the multi-axis position difference sequence to form axis difference features, and fuse the axis difference features with the detection data into joint spatiotemporal features for diagnosis;

[0076] Step 400: Under the premise of complying with the kinematic and dynamic boundaries and process safety constraints of the machine tool, construct a causal reasoning model of error source-process parameter-equipment state based on the joint spatiotemporal characteristics and the preset AI agent, output the contribution weight and mechanism explanation for multiple candidate error sources, and solve the process parameter correction instruction set for the machining process and the adaptive scanning strategy for the inspection process with "geometric tolerance satisfaction, machining cost and inspection time" as multiple objectives.

[0077] Step 500: The process parameter correction instruction set and adaptive scanning strategy are issued to the processing and inspection stages respectively for execution. A new round of inspection data of real shaft position table and geometric deviation distribution is collected. The causal reasoning model is continuously updated through the incremental learning mechanism of AI agent until the preset accuracy is met or the convergence criterion is reached.

[0078] In one specific embodiment, step 100 first extracts the geometric features required for machining from the three-dimensional digital model of the target part. The geometric features include at least the surface contour, boundary conditions, and machining allowance of the part, which are parameter information that can directly affect the tool path generation. Then, process constraints are obtained from the process plan. The process constraints include at least the machining sequence, cutting posture restrictions, and feed rate range to ensure that the trajectory generation process is executed within the scope of process specifications and equipment capabilities.

[0079] After obtaining the geometric features and process constraints, these are loaded as inputs into the CNC simulation environment. The simulation environment generates the machining trajectory to be executed, and the built-in CNC interpolation algorithm dynamically runs the trajectory. During dynamic operation, multi-axis position data is continuously output according to a preset sampling frequency, and each position data point is timestamped, thus obtaining a virtual axis position table containing timestamps and multi-axis positions. This virtual axis position table reflects the theoretical motion trajectory of each axis of the machine tool under ideal working conditions, providing a benchmark for subsequent comparison and difference analysis with actual production data.

[0080] The "virtual axis position table" mentioned in this invention refers to a time-series data table generated by CNC machining simulation software running the machining trajectory to be executed under ideal working conditions. This table records the theoretical position coordinates and corresponding timestamps of each motion axis of the machine tool at each sampling moment, and can be expanded to include derived parameters such as speed or acceleration. The function of the virtual axis position table is to provide an ideal benchmark for subsequent dual-source data comparison, thereby quantifying the dynamic deviations caused by servo hysteresis, thermal deformation or external disturbances during the actual machining process.

[0081] Specifically, in step 200 of this embodiment, when the CNC machine tool loads and executes the machining trajectory, the multi-axis servo motor is driven to perform linked motion according to the trajectory. During the motion, the machine tool control system collects the control command values ​​and actual position feedback values ​​of each axis in real time through the feedback channel, and simultaneously collects signals from temperature sensors and vibration sensors to characterize the dynamic effects of thermal drift and mechanical vibration during machining. After marking each set of data with a unified sampling timestamp, the sampling results are stored in chronological order as a multi-axis position sequence, thereby forming a real axis position table for subsequent comparative analysis with the virtual axis position table.

[0082] After machining, the part is transferred to a coordinate measuring machine (CMM) or laser scanner to acquire 3D point cloud data covering key surfaces and feature points. By fitting the point cloud and aligning it with the coordinate system, the measurement data is differentially analyzed with the 3D digital model of the target part to obtain a geometric deviation distribution reflecting the difference between the actual machined surface and the theoretical geometric shape. This deviation distribution is stored in a color map or numerical matrix format for subsequent fusion with axis features to diagnose the sources of machining errors.

[0083] Specifically, in this embodiment, the real axis position table refers to the axis motion trajectory data recorded in real time during the actual machining process of the machine tool. It includes timestamps, multi-axis coordinate positions, and speed information, and can be expanded to include operating condition parameters such as temperature and vibration. Its function is to provide a real physical data foundation for subsequent comparison with virtual axis position tables, error tracing, and causal reasoning models.

[0084] In a preferred embodiment, step 300 first unifies the virtual axis position table and the real axis position table to the same time reference. Specifically, using the clock acquired by the actual machine tool as a reference, the virtual sequence and the real sequence are aligned through a monotonic time mapping method. During this process, continuity constraints are introduced to ensure that the time scaling ratio remains within ±2%, and a smoothing penalty is used to suppress non-physical abrupt changes in the time axis, thereby obtaining a stable time correspondence and ensuring that the virtual and real motion trajectories maintain minimal error across the entire time domain.

[0085] After unifying the time scale, this embodiment differs the real position data and virtual position data at each moment to form a multi-axis position difference sequence. This sequence is then stacked over time to form a difference matrix, which is used to characterize the time-varying distribution of dynamic errors such as servo hysteresis, thermal drift, and backlash. The rows of the difference matrix represent the sampling time, the columns represent each motion axis or axis combination, and the values ​​in the matrix are the dynamic error amplitudes at that moment.

[0086] After obtaining the difference matrix, this embodiment maps it to the geometric deviation field obtained from the surface detection of the part. The method is as follows: in the surface coordinate system of the part, the center path of the machine tool is mapped to the geometric surface of the workpiece, and the axis difference is mapped to the surface deviation using the geometric sensitivity relationship. In order to reflect the influence range between different points, the scale factor calculated by the feed rate and sampling interval is used as the influence radius, so that the surface points closer to the tool path are given a higher weight for process error transmission, thereby obtaining the spatiotemporal projection result from process error to topographic deviation.

[0087] Building upon spatiotemporal projection, this embodiment introduces an adaptive fusion mechanism based on statistical variance to balance the contributions of process-side and detection-side data. Specifically, the variance ratio of the geometric deviation field and the difference matrix is ​​calculated and used as the fusion weight coefficient. When the detection-side data fluctuates significantly, the corresponding weight decreases; when the process-side error is large, the corresponding weight increases. Simultaneously, a kernel function with the feed step size as the scale is introduced during spatial projection, and the kernel width is automatically determined by the feed rate and sampling interval. In this way, the process-side and detection-side information can be dynamically balanced in the fused joint spatiotemporal features, avoiding diagnostic distortion due to anomalies in unilateral data.

[0088] Specifically, the virtual axis position table is a time-series data table generated by the CNC simulation environment running the machining trajectory under ideal working conditions. The real axis position table is a time-series data table generated by the machine tool control system collecting position feedback and working condition sensor data during the machining process. The axis position difference matrix is ​​the time-series stacking result of the difference between the real and virtual positions after unifying the time scale. The geometric deviation field is the distribution of the deviation on the part surface after being obtained by equipment such as coordinate measuring machine or laser scanner and aligned with the CAD model. The joint spatiotemporal features are a set of diagnostic features that integrate the axis position difference matrix and the geometric deviation field into a unified spatiotemporal domain through path mapping, sensitivity analysis and adaptive fusion.

[0089] In the above steps, monotonic time mapping originates from time series registration technology, used to strictly and monotonically map the virtual time axis to the real time axis; smoothing penalty originates from servo bandwidth constraints, used to suppress non-physical abrupt changes that occur during time registration; geometric sensitivity originates from machine tool kinematics, used to quantify the impact of positional disturbances of each axis on the tool tip and workpiece surface errors; kernel function originates from spatial weighted modeling, used to control the range of influence of process errors on different spatial points. Their applications in this embodiment are respectively ensuring time alignment stability, maintaining physical consistency, establishing the transmission relationship between process errors and geometric deviations, and achieving adaptive control of spatial influences.

[0090] Furthermore, the time scaling ratio is preferably no more than ±2%, the sampling frequency is preferably set to 1000 Hz, the servo jitter high-frequency threshold is preferably 50 Hz, the detection point cloud density in areas with curvature exceeding 0.1 mm is preferably increased to 200 points per square centimeter, and the point spacing in high-risk areas is preferably no more than 0.1 mm.

[0091] In a preferred embodiment, step 400 of this embodiment first accesses joint spatiotemporal features through an AI agent. These joint spatiotemporal features are composed of the difference information between the virtual axis position table and the real axis position table after alignment, as well as the geometric deviation distribution. Simultaneously, temperature and vibration sensor data from the working condition side are introduced to reflect the effects of thermal drift and mechanical vibration. To ensure the consistency of the input data, a unified time reference and a consistent coordinate system are used. The path representation of the machining trajectory is aligned with the 3D scan or point cloud to the coordinate domain of the part surface and organized into batch data according to the process sequence for subsequent causal inference and optimization.

[0092] After data preparation, the AI ​​agent performs causal inference within the bounds of machine tool kinematics and dynamics. These boundaries include rigidity limits, upper limits for velocity and acceleration, and the range of achievable postures, to suppress explanations inconsistent with the equipment's physical capabilities. The inference framework employs an interpretable probabilistic causal method to establish causal relationships between error sources, process parameters, and equipment status, and outputs the contribution weight and confidence assessment for each candidate error source. Based on these results, an attribution report is automatically generated, including at least a textual description of the dominant error source, its contribution weight, and the corresponding physical mechanism, to guide subsequent parameter and strategy optimization.

[0093] After obtaining the attribution results, the AI ​​agent simultaneously constructs a multi-objective collaborative optimization problem for both machining and inspection channels. The objectives include improving geometric tolerance satisfaction, reducing machining costs, and shortening inspection time. The optimization is solved within the machine tool kinematics, dynamics, and process safety boundaries. The machining channel outputs a set of process parameter correction instructions, including adjustments to feed rate, acceleration / deceleration profile, tool compensation, and cooling strategies. If necessary, parameters are updated for relevant sections of the CNC program to weaken the error propagation chain. The inspection channel outputs an adaptive scanning strategy, including configurations for scanning path, sampling density, and measurement priority. This dynamically improves the coverage and resolution of critical areas by combining the part's curvature and deviation confidence region.

[0094] The process parameter correction instruction set and adaptive scanning strategy are respectively sent to the machine tool and measuring equipment for execution, collecting a new round of real axis position tables and geometric deviation distributions to form an iterative data package containing trajectory execution records, sensor readings, and detection results. The AI ​​agent calculates the target evaluation vector based on the iterative data package, which includes at least three indicators: geometric tolerance satisfaction, processing cost, and detection time. Based on this, incremental learning and updates are performed, allowing the causal model and optimization strategy to continuously self-correct as equipment status and operating conditions change. When a preset accuracy threshold is met or the improvement margin is below the threshold for several consecutive rounds, convergence is determined, and the current strategy is solidified; otherwise, the loop continues.

[0095] Specifically, in this embodiment, the AI ​​agent refers to an integrated intelligent decision-making entity that performs multi-source data fusion, causal reasoning under physical constraints, and multi-objective collaborative optimization; the joint spatiotemporal features refer to a set of diagnostic features formed by fusing axis position difference information and geometric deviation distribution within a unified time reference and a unified surface coordinate domain, which simultaneously contains spatial distribution and temporal evolution attributes; the attribution report refers to a structured result based on causal reasoning output used to guide optimization decisions, which at least includes the dominant error source, contribution weight, and physical mechanism description; the process parameter correction instruction set refers to a set of parameter adjustments that can be directly applied to machine tools or programs, including feed rate, acceleration / deceleration profile, tool compensation, and cooling strategies; and the adaptive scanning strategy refers to a detection scheme that is dynamically adjusted based on geometric and working condition information, which at least includes path, sampling density, and measurement priority, and automatically improves coverage and resolution in high-confidence areas of curvature or deviation.

[0096] Optionally, the time scaling ratio in this embodiment is preferably controlled within ±2% to ensure alignment stability; the sampling frequency is preferably 1000 times per second to capture high-frequency error components; the servo jitter judgment threshold is preferably in the frequency band above 50 Hz; when the local curvature exceeds 0.1 per millimeter, the point cloud density of the key area is preferably increased to 200 points per square centimeter; the single correction amplitude of the machining side feed rate is preferably no more than ±10%, and the smoothing window of the acceleration and deceleration profile is preferably between 10 milliseconds and 30 milliseconds; the convergence criterion is preferably set as the target evaluation vector being less than 2% in three consecutive rounds of improvement or the geometric tolerance meeting the process standard.

[0097] like Figure 2 As shown, the overall process of this invention includes steps such as part modeling, process planning, simulation operation, actual machining, inspection feedback, and intelligent optimization. The process plan generated by the part modeling and machining process planning software is imported into the simulation software to output a virtual simulation machine tool trajectory point table. Simultaneously, the machining machine tool records the actual production machine tool trajectory point table during the actual production process and completes the machining of the target workpiece. The target workpiece obtains surface geometric data through 3D scanning detection, and after fitting and alignment processing, geometric deviation data and scanning trajectories are formed. These virtual axis position tables, real axis position tables, and geometric deviation distribution data are input to the AI ​​agent.

[0098] The AI ​​agent compares and analyzes the input data, outputting error attribution results and optimization instructions. Specifically, this includes: analyzing the differences between virtual and real machine tool trajectory point tables, performing spatiotemporal fusion with 3D geometric deviation data to obtain joint features suitable for diagnostics; based on this, the agent generates machining code optimization schemes, process parameter correction schemes, process flow optimizations, and scanning trajectory optimizations, thereby ensuring geometric tolerances while balancing machining and inspection efficiency, achieving bidirectional optimization of process and inspection driven by virtual-real comparison.

[0099] like Figure 3 As shown, the AI ​​agent structure of this embodiment adopts a layered design, encompassing an input layer, a feature extraction layer, an attribution inference layer, an optimization decision layer, and an output layer. The input layer receives geometric deviation data, shaft position differences, and process parameters. The feature extraction layer extracts spatial and temporal features from the input and embeds physical constraints such as machine tool kinematics and dynamics. The attribution inference layer uses a probabilistic graphical model-based network to calculate the contribution weights and confidence levels of different error sources, outputting explanatory results containing error attribution relationships.

[0100] After attribution reasoning is completed, the optimization decision layer includes a processing optimizer and a scanning optimizer, which output processing correction instructions and adaptive scanning schemes, respectively. The optimization result layer then integrates both to generate a co-converged optimization result. The output layer transforms these results into executable processing correction instructions, scanning trajectory schemes, and error attribution reports. To ensure the continuity and adaptability of optimization, the system also includes a reinforcement learning feedback mechanism, which continuously updates the causal reasoning model and optimization strategy using newly acquired real data, forming a closed-loop optimization process of processing and detection.

[0101] In a preferred embodiment, step 500 of this embodiment utilizes an AI agent to directly apply the process parameter correction instruction set obtained from the previous round of optimization to the machine tool machining process, and sends the adaptive scanning strategy to a coordinate measuring machine or laser scanning system to perform geometric inspection, thereby initiating a new iteration round. Within the iteration round, the machine tool control system collects timestamped multi-axis position feedback and forms a new true axis position table. The inspection equipment acquires three-dimensional point cloud or surface deviation data covering the key curved surfaces of the target workpiece, and aggregates this data with the execution log into an iteration data package. The AI ​​agent calculates a target evaluation vector based on the iteration data package. This vector includes at least three components: geometric tolerance satisfaction, machining cost, and inspection time, used to measure the overall performance of the iteration round and as the basis for subsequent optimization and convergence determination.

[0102] After calculating the target evaluation vector, the AI ​​agent uses the iterative data package as new samples to perform incremental learning updates on the causal inference model. During the update process, it maintains the embedding of machine tool kinematics, dynamic boundaries, and process safety constraints to ensure the physical consistency and interpretability of the attribution process. Based on the updated inference results and the target evaluation vector, the AI ​​agent simultaneously optimizes the machining-side process parameters and the inspection-side scanning strategy, generating a process parameter correction instruction set and an adaptive scanning strategy for the next iteration, achieving coordinated convergence of the machining-inspection dual closed loop. When the target evaluation vector meets a preset accuracy threshold, or its improvement in multiple consecutive iterations is less than 2%, the AI ​​agent determines that convergence has been achieved and terminates the iteration. If the convergence condition is not met, a new iteration continues until the accuracy requirement is met or the stopping condition is triggered.

[0103] An iterative round refers to a complete closed-loop process from the issuance and execution of process parameter correction instructions and adaptive scanning strategies until new optimization results are produced. The iterative data package is a comprehensive data set composed of the real axis position table, geometric deviation distribution, and execution log collected within the iterative round. The target evaluation vector is a multi-objective quantitative indicator calculated by the AI ​​agent based on the iterative data package, used to measure the comprehensive performance of accuracy, cost, and efficiency. The convergence criterion is a standard used to determine whether the iterative process can be stopped, which usually includes two types of conditions: a preset accuracy threshold and a lower limit of improvement.

[0104] In a preferred embodiment, the system's input layer receives multi-source heterogeneous data, including geometric deviation data (the deviation matrix between the 3D measurement model and the digital model of the part), axis position table differences (time-by-time position errors after time alignment of the virtual and real axis position tables), and process parameters (cutting speed, feed rate, coolant flow rate, etc.). The input layer supports real-time data stream input of no less than 1000 times / second, with a timestamp alignment accuracy of no more than 1 millisecond, to ensure that virtual and real data remain comparable in high-speed machining scenarios. Subsequently, the system enters the feature extraction layer: curvature analysis is performed on the geometric deviation data, marking areas with curvature higher than 0.1 mm as key feature regions; frequency domain analysis is performed on the axis position table differences, identifying components higher than 50 Hz as servo jitter symptoms; and a unified spatiotemporal feature vector is formed by combining the process parameters for subsequent causal inference.

[0105] After feature preparation, the attribution inference layer embeds the machine tool kinematics and dynamics boundaries as physical constraints into the inference process, including rigidity limits, velocity and acceleration boundaries, and reachable attitude ranges, to eliminate explanations inconsistent with equipment capabilities. Based on this, a causal error chain centered on a Bayesian network is constructed, connecting "error features—process parameters—equipment status," calculating the contribution rate of multiple candidate error sources and providing confidence assessments. The system automatically generates an error attribution report, including at least the dominant error source, contribution rate ranking, and physical mechanism description, to guide subsequent synchronous optimization on the machining and inspection sides. After obtaining the attribution results, the optimization decision layer initiates two optimization channels in parallel: the machining optimizer uses the attribution report and process parameters as input, employing a policy gradient-based reinforcement learning method to generate machining correction instructions; the scanning optimizer uses spatial features and geometric deviation distribution as input, employing a curvature-driven adaptive method to generate a scanning trajectory scheme and increase point cloud density in key feature regions. The output layer transforms the optimization results into executable objects: one type is a set of G-code parameter adjustments that can be directly revised (e.g., acceleration curves), and the other type is a scan trajectory file (preferably in STL format) that can be downloaded to the detection device. For regions with curvature higher than 0.1 mm, the point cloud density is increased to 200 points per square centimeter to ensure detection resolution in critical areas. The entire process is coupled with a reinforcement learning feedback mechanism: a reward function is constructed based on the data returned from production and detection, driving the two channels to converge continuously in subsequent rounds. "Multi-source heterogeneous data access" refers to the system simultaneously accessing data from different sources and with different structures, such as geometric deviation matrices, virtual and real axis position difference sequences, and process parameters, and completing the consistency of timestamps and coordinate domains; "Feature extraction layer" refers to the processing layer that reduces and integrates spatial, temporal, and process dimensions within a unified spatiotemporal domain; "Attribution reasoning layer" refers to the processing layer that performs interpretable reasoning on error sources under physical constraints and outputs contribution rates and confidence assessments; "Optimization decision layer" refers to the processing layer that performs multi-objective collaborative solutions based on attribution results for both processing and inspection channels and generates executable solutions; "Reinforcement learning feedback mechanism" refers to the self-learning process that uses a new round of production and inspection data to construct reward signals and iteratively update strategies to achieve continuous convergence; "Output layer" refers to the interface layer that organizes optimization results into instructions and trajectory files that can be directly applied to machine tools and inspection equipment.

[0106] Corresponding to the above method, this embodiment also provides a manufacturing process parameter and inspection optimization system based on a dual-source axis position table and an AI agent, including:

[0107] The machining trajectory generation unit is used to generate a machining trajectory to be executed based on the digital model and process plan of the target part, and run the machining trajectory in a simulation environment to obtain a virtual axis position table containing timestamps and multi-axis positions.

[0108] The real data acquisition unit is used to execute the machining trajectory on the machine tool, acquire a real axis position table containing timestamps and multi-axis positions in real time, and obtain detection data characterizing the geometric deviation distribution of the part after machining is completed;

[0109] The temporal alignment and feature fusion unit is used to temporally align the virtual axis table with the real axis table, calculate the multi-axis position difference sequence to form axis difference features, and fuse the axis difference features with the detection data into joint spatiotemporal features for diagnosis.

[0110] The AI ​​agent causal reasoning and optimization unit is used to construct a causal reasoning model of error source-process parameter-equipment state based on the joint spatiotemporal features and the preset AI agent, under the premise of complying with the kinematic and dynamic boundaries and process safety constraints of the machine tool. It outputs the contribution weight and mechanism explanation for multiple candidate error sources, and solves the process parameter correction instruction set for the machining process and the adaptive scanning strategy for the inspection process with "geometric tolerance satisfaction, machining cost and inspection time" as multiple objectives.

[0111] The closed-loop execution and incremental learning unit is used to send the process parameter correction instruction set and the adaptive scanning strategy to the processing and inspection stages for execution, and to collect a new round of detection data of the real axis position table and geometric deviation distribution. The causal reasoning model is continuously updated through the incremental learning mechanism of the AI ​​agent until the preset accuracy is met or the convergence criterion is reached.

[0112] The beneficial effects of this invention are as follows:

[0113] (1) This invention establishes a stable and consistent time reference and coordinate domain based on the dual-source alignment of the virtual axis position table and the real axis position table, so that dynamic errors such as servo lag, thermal drift, and backlash in the machining process are transformed from "not directly observable" to "time-quantifiable". In terms of capability indicators, the input layer supports data stream access of no less than 1000 times / second and time alignment accuracy of no more than 1 millisecond, which significantly suppresses the impact of sampling jitter and clock drift on diagnostic accuracy. Compared with the existing technology that focuses on offline simulation or single-source monitoring, this invention uses dual-source difference modeling to align simulation and actual measurement under all working conditions, which significantly improves the observability and reproducibility of errors, and provides a solid data foundation for subsequent reasoning and optimization.

[0114] (2) The AI ​​agent of the present invention performs causal reasoning under the constraints of machine tool kinematics, dynamics and process safety, and outputs the dominant error source, contribution weight and physical mechanism description, avoiding the black box prediction that "only gives the result without explaining the reason". Unlike the existing technology that mainly relies on experience compensation or pure data fitting, the present invention incorporates hard constraints of physical boundaries into the reasoning process, effectively avoiding out-of-limit solutions and unimplementable solutions; unlike the existing technology that focuses on the rule adjustment of the detection side, the present invention can open up the mechanism chain of "error generation - error propagation - geometric shaping", and produce directly executable and traceable optimization basis, reducing repeated manual trial cutting and reliance on experience.

[0115] (3) This invention generates two types of control variables simultaneously within the same iteration round: a set of process parameter correction instructions on the machining side (feed, acceleration / deceleration profile, tool compensation, cooling strategy, etc.) and an adaptive scanning strategy on the detection side (path, density, priority). This "dual-channel collaborative optimization" bypasses the serial bottleneck of "machining first and then separately adjusting detection" or "detecting first and then separately adjusting machining," enabling optimization to proceed synchronously with convergence as the goal. In terms of engineering strategy, this invention automatically increases the point cloud density in areas with large curvature or high deviation confidence, and reduces redundant sampling in low-risk areas, thereby simultaneously shortening the detection time and improving the resolution of critical areas. Compared with the existing technology that only adapts the scanning path or density, this invention links the detection strategy with the process parameters, significantly improving the convergence speed and overall process efficiency.

[0116] (4) This invention uses an incremental learning closed loop as the main line. After each round of execution, it absorbs new real axis position tables and geometric deviation distributions, automatically updates the causal model and optimization strategy, and can adaptively evolve with the drift of equipment status and environmental conditions. The output format is directly geared towards the field: one end generates a set of machining program parameters that can be written back, and the other end generates a scan trajectory file that can be distributed, reducing system integration costs. Unlike the existing technologies that focus on "offline compensation" or "detection-side adaptation" respectively, this invention realizes a closed-loop self-evolution of data-mechanism-decision-execution, which not only improves the stability of geometric tolerance compliance, but also achieves comprehensive benefits in terms of cost and cycle time, and has clear engineering feasibility and scalability.

[0117] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0118] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A production manufacturing process parameter and detection optimization method based on a dual-source axis position table and an AI agent, characterized in that, Comprise: Based on the target part of the digital model and process planning to generate the machining trajectory to be executed, and run in the simulation environment to get the virtual axis position table containing time stamp and multi-axis position; On the machine tool to execute the machining trajectory, real-time collection of real axis position table containing time stamp and multi-axis position, and obtain the detection data characterizing the part geometric deviation distribution after processing is completed; The virtual axis position table and the real axis position table are time series aligned, the multi-axis position difference sequence is calculated to form the axis position difference feature, and the axis position difference feature and the detection data are fused into the joint space-time feature for diagnosis; Under the premise of conforming to the kinematics and dynamics boundary of machine tool and process safety constraints, based on the joint space-time feature and the preset AI agent, the causal reasoning model of error source-process parameter-equipment state is constructed, the contribution weight and mechanism explanation for multiple candidate error sources are output, and the "geometric tolerance satisfaction degree, processing cost, detection time" is taken as the multi-objective, and the process parameter correction instruction set and the adaptive scanning strategy for the detection process are solved; The process parameter correction instruction set and the adaptive scanning strategy are respectively issued to the processing and detection link to execute, and a new round of real axis position table and geometric deviation distribution detection data are collected, so as to continuously update the causal reasoning model through the incremental learning mechanism of AI agent until the preset precision is met or the convergence criterion is reached; The virtual axis position table and the real axis position table are time series aligned, the multi-axis position difference sequence is calculated to form the axis position difference feature, and the axis position difference feature and the detection data are fused into the joint space-time feature for diagnosis, comprising: The time base of the virtual axis position table and the real axis position table is unified, and the optimal time registration function is obtained under the condition of satisfying the machine tool motion continuity constraint, so that the least mean square error of the two in the whole time domain is obtained, and the time registration formula is: ； wherein, is the real multi-axis position vector at time ; is the virtual multi-axis position vector; is a monotonic time registration function; is an optimal time registration function; is the derivative of ; is a time-smoothing penalty coefficient derived from the machine tool servo bandwidth specification; is a time-stretching allowed deviation determined from the machine tool synchronization accuracy specification; and are the total time duration of the real and virtual sequences, respectively; The multi-axis position difference sequence is calculated on the unified time scale and the axis position difference matrix is formed to represent the time-varying distribution of dynamic error, and the definition formula of the axis position difference matrix is: ； wherein, is the multi-axis position difference vector at time t, revealing dynamic errors of servo hysteresis, thermal drift, reverse gap; The axis position difference matrix and the geometric deviation field detected are corresponded in space and retraced in time in the part surface coordinate system, and the time and space projection results from process error to topographic deviation are completed under the tool center path mapping of the tool path, and the mapping and fusion formula is: ； wherein, is a spatial point on the surface of the workpiece; is a detected geometric deviation; is a path of the tool center on the surface of the workpiece; is a geometric sensitivity from shafting disturbances to tool / surface errors; is a process side positional discrepancy.

2. The dual-source axial table and AI agent-based production manufacturing process parameter and detection optimization method according to claim 1, wherein, Based on the target part of the digital model and process planning to generate the machining trajectory to be executed, and run in the simulation environment to get the virtual axis position table containing time stamp and multi-axis position, comprising: Extracting geometric features from the three-dimensional digital model of the target part; The geometric features at least include the surface profile, boundary conditions and machining allowance of the part; Obtain process constraints from process planning; The process constraints at least include machining sequence, cutting posture restriction and feed speed range; In the numerical control simulation environment, the geometric features and the process constraints are taken as input to generate the machining trajectory to be executed and dynamically run; In the dynamic running process, multi-axis position data is output according to a preset sampling frequency, and a time stamp is added to each piece of position data to form the virtual axis position table.

3. The dual-source axial table and AI agent-based production manufacturing process parameter and detection optimization method of claim 1, wherein, The machining trajectory is executed on the machine tool, a real axis position table containing time stamps and multi-axis positions is collected in real time, and detection data representing the geometric deviation distribution of the part is obtained after machining is completed, including: The machining trajectory is loaded on the numerical control machine tool and the machining process is started, and the multi-axis servo system is driven to move according to the machining trajectory; In the machining process, the position command value and the position feedback value of each axis are obtained in real time through the feedback channel of the machine tool control system, and each set of position command value and position feedback value of each axis is marked with a sampling time stamp to form a real-time multi-axis position sequence; The multi-axis position sequence is stored in the real axis position table in the order of sampling time for subsequent comparison with the virtual axis position table; After machining is completed, the part is geometrically detected to collect part surface point position data based on a coordinate measuring machine or a laser scanner; The collected point position data is fitted and aligned to obtain the detection data representing the geometric deviation distribution between the part and the target model.

4. The dual-source axial table and AI agent-based production manufacturing process parameter and detection optimization method of claim 1, wherein, The virtual axis position table and the real axis position table are time-aligned, a multi-axis position difference sequence is calculated to form an axis position difference feature, and the axis position difference feature and the detection data are fused into a joint space-time feature for diagnosis, further including: An adaptive fusion mechanism is introduced in the space-time projection result to balance the contributions of the process side and the detection side, and the fusion weight and the kernel function are defined as: wherein, is the empirical variance of the geometric deviation; is the empirical variance of the axis position difference; is the adaptive fusion coefficient; is the kernel function scaled by the feed step; is the kernel width, scaled by the feed speed and the sampling interval is determined automatically.

5. The dual-source axial table and AI agent-based production manufacturing process parameter and detection optimization method of claim 1, wherein, The process parameter correction instruction set for the machining process at least includes adjustment of one or more of feed, acceleration / deceleration profile, tool compensation, and cooling strategy.

6. The dual-source axial table and AI agent-based production manufacturing process parameter and detection optimization method of claim 1, wherein, Under the premise of meeting the kinematic and dynamic boundaries and process safety constraints of the machine tool, a causal reasoning model of error source-process parameter-equipment state is constructed based on the joint space-time feature and a preset AI agent, the contribution weight and mechanism explanation for multiple candidate error sources are output, and the geometric tolerance satisfaction degree, machining cost, and detection time are used as multiple objectives to solve the process parameter correction instruction set for the machining process and the adaptive scanning strategy for the detection process, including: The AI agent accesses the joint space-time feature, fuses the dynamic error feature extracted from the sequence of axis position difference features and the three-dimensional scanning / point cloud geometric deviation on the detection side and the comparison relationship with CAD to form a comprehensive input set for attribution reasoning; The AI agent uses a Bayesian network to perform causal reasoning under the condition of embedding the kinematic and dynamic boundaries of the machine tool, establishes the relationship between the error source, the process parameter, and the equipment state, and outputs the contribution weight and confidence evaluation of each candidate error source and the physical mechanism explanation; The AI agent generates an interpretable attribution report; the attribution report at least includes the dominant error source, the contribution weight, and the corresponding physical mechanism, which is used to drive the collaborative optimization decision of the machining and detection sides. The AI agent constructs and solves a collaborative optimization problem under the premise of meeting the kinematics and dynamics of the machine tool and process safety constraints, aiming to improve the geometric tolerance satisfaction and reduce the machining cost and detection time, so that the machining side and the detection side converge simultaneously and output two types of control quantities in the same optimization round; The AI agent generates a process parameter correction instruction set, which at least includes adjustments to one or more of the feed rate, acceleration and deceleration profile, tool compensation, and cooling strategy, and modifies the relevant numerical control instruction paragraphs as necessary to suppress the error propagation chain; The AI agent generates an adaptive scanning strategy based on the error space distribution and optimization trade-off results.

7. The dual-source axial table and AI agent-based production manufacturing process parameter and detection optimization method of claim 1, wherein, The adaptive scanning strategy at least includes configuration of one or more of the scanning path, sampling density, and measurement priority, and increases the sampling density and local coverage in high curvature or deviation confidence areas.

8. The dual-source axial table and AI agent-based production manufacturing process parameter and detection optimization method of claim 1, wherein, The process parameter correction instruction set and the adaptive scanning strategy are respectively issued to the machining and detection links for execution, and the detection data of the real axis position table and the geometric deviation distribution of the new round are collected to continuously update the causal inference model through the incremental learning mechanism of the AI agent until the preset accuracy is met or the convergence criterion is reached, including: The AI agent applies the process parameter correction instruction set obtained from the previous optimization to the machine tool machining process, and uses the adaptive scanning strategy for geometric detection process; the whole process from the start of the execution of the process parameter correction instruction set and the adaptive scanning strategy to the generation of new optimization output is defined as an iteration round and starts execution; In the iteration round, time-stamped multi-axis position feedback is collected to generate a new real axis position table, and surface deviation based on three-dimensional scanning or point cloud is obtained to form a geometric deviation distribution; the real axis position table, geometric deviation distribution, and execution log are collected as an iteration data package for learning and evaluation; Based on the AI agent, the target evaluation vector is calculated based on the iteration data package, with components in turn being geometric tolerance satisfaction, machining cost, and detection time, for quantifying the performance of the iteration round and as a basis for subsequent optimization and stopping criterion; Based on the AI agent, the causal inference model is updated by the AI agent based on the iteration data package as a new sample, and the embedding of the kinematics and dynamics boundaries of the machine tool and the process safety constraints is maintained to ensure the physical consistency and explainability of the attribution; According to the updated causal inference results and the target evaluation vector, the re-optimization of the machining side process parameters and the detection side scanning strategy is simultaneously solved to generate the process parameter correction instruction and the adaptive scanning strategy for the next iteration round, realizing the collaborative convergence of the machining-detection double closed loop; When the target evaluation vector meets the preset accuracy threshold or the improvement amplitude in the continuous iteration rounds is lower than the preset threshold, the AI agent determines that the convergence criterion is reached and terminates the iteration; if the convergence criterion is not met, the AI agent continues to perform a new iteration round until the accuracy is met or the stopping condition is triggered.

9. A production manufacturing process parameter and detection optimization system based on dual-source axis table and AI agent, characterized in that, The manufacturing process parameter and detection optimization method based on dual-source axis positioning table and AI agent as described in any one of claims 1 to 8, wherein the manufacturing process parameter and detection optimization system based on dual-source axis positioning table and AI agent comprises: The machining trajectory generation unit is used to generate a machining trajectory to be executed based on the digital model and process plan of the target part, and run the machining trajectory in a simulation environment to obtain a virtual axis position table containing timestamps and multi-axis positions. The real data acquisition unit is used to execute the machining trajectory on the machine tool, acquire a real axis position table containing timestamps and multi-axis positions in real time, and obtain detection data characterizing the geometric deviation distribution of the part after machining is completed; The temporal alignment and feature fusion unit is used to temporally align the virtual axis table with the real axis table, calculate the multi-axis position difference sequence to form axis difference features, and fuse the axis difference features with the detection data into joint spatiotemporal features for diagnosis. The AI ​​agent causal reasoning and optimization unit is used to construct a causal reasoning model of error source-process parameter-equipment state based on the joint spatiotemporal features and the preset AI agent, under the premise of complying with the kinematic and dynamic boundaries and process safety constraints of the machine tool. It outputs the contribution weight and mechanism explanation for multiple candidate error sources, and solves the process parameter correction instruction set for the machining process and the adaptive scanning strategy for the inspection process with "geometric tolerance satisfaction, machining cost and inspection time" as multiple objectives. The closed-loop execution and incremental learning unit is used to send the process parameter correction instruction set and the adaptive scanning strategy to the processing and inspection stages for execution, and to collect a new round of detection data of the real axis position table and geometric deviation distribution. The causal reasoning model is continuously updated through the incremental learning mechanism of the AI ​​agent until the preset accuracy is met or the convergence criterion is reached.

Images