A method and system for in-machine detection based milling and grinding process self-optimization

Abstract

The application discloses a kind of milling and grinding process self-optimization method and system based on in-machine detection, and relates to precision machining technical field.The kind of milling and grinding process self-optimization method and system based on in-machine detection, including S1, the acquisition and normalization processing of in-machine multi-source signal and surface topography are completed, feature deviation component is generated and is stored to construct milling and grinding process database;S2, defect cause coupling analysis is carried out through multi-source feature deviation data;S3, geometric driving strength is evaluated by comprehensive cause correlation measure, geometric deviation characterization and load feature;S4, reconstruct the benefits of the collaborative evaluation of defect representation change and adjustment intensity before and after structure reconstruction, and complete rollback correction, configuration solidification and sample precipitation closed-loop verification.The problem that tool state in-machine detection lags behind and defect quantitative characteristics cannot drive tool geometric parameters online closed-loop optimization, thereby resulting in limited efficiency, quality consistency and tool life is solved.

Description

Technical Field

[0001] This invention relates to the field of precision machining technology, specifically to a self-optimization method and system for milling and grinding processes based on in-machine inspection. Background Technology

[0002] With the increasing demands for workpiece shape and position accuracy and surface integrity in fields such as aerospace, precision molds, and medical devices, precision machining methods integrating milling and grinding are widely used in the manufacturing of hard and brittle materials and difficult-to-machine alloy parts. Current process assurance methods mainly rely on offline inspection and post-processing traceability. On the production site, surface and geometric accuracy are often confirmed after machining using roughness testers, contour measurements, microscopic observation, and coordinate measuring machines. At the same time, tool replacement or regrinding is performed based on machining time and the number of parts machined.

[0003] For example, the invention patent with publication number CN110888394B discloses a tool axis optimization method for wear control of ball end mills in CNC machining of curved surfaces. Based on the input tool path and geometric information, the tool-workpiece meshing area at each tool position point is constructed, and the interference-free tool axis space at each tool position point is calculated. Then, the tool cutting edge is divided into multiple cutting intervals along the tool axis. Based on the calculated tool-workpiece meshing area, the cutting length of each cutting interval at each tool position point under the initial tool axis is calculated to obtain the cutting length of each cutting interval of the tool edge for machining the entire part. The tool wear amount of each cutting interval is calculated based on the wear rate. Finally, a fixed tool axis strategy is adopted to determine the tool axis direction in which tool wear is evenly distributed within each cutting interval of the tool edge from the interference-free tool axis space at each tool position point. By adjusting the tool axis to achieve a uniform distribution of tool wear, the problem of premature tool failure due to concentrated tool wear in localized areas can be effectively avoided.

[0004] For example, the invention patent with publication number CN108549320B discloses a method for controlling milling parameters and tool wear of titanium alloys based on roughness. It establishes an initial process parameter domain for precision milling of titanium alloy components and conducts orthogonal experiments based on this initial process parameter domain. The surface roughness of the test component in the milling feed and cutting width directions is measured, and an absolute sensitivity model of surface roughness to precision milling is established. The range of surface roughness variation corresponding to the stable and unstable domains of each parameter is determined until the optimization target of surface roughness falls within the range corresponding to the stable domain. Tool wear experiments are conducted based on the stable domain to obtain the selection range of tool wear and surface roughness. By setting different precision milling process parameters and conducting experiments, the stable domain of the process parameters is obtained, and tool wear experiments are performed to obtain the relationship curve between the flank wear and surface roughness, thereby achieving control of the surface roughness of the precision milling of titanium alloy components.

[0005] The industry is gradually exploring the use of defect classification, feature deviation assessment, and load feature analysis to assist in the timely adjustment of process parameters and the tracing of defect causes. However, it still faces the following challenges: On the one hand, the quantitative coupling relationship between defect causes and multi-source detection features is still unclear. Defect identification and cause classification rely heavily on expert rules or experience bases, lacking adaptive closed-loop correlation models. On the other hand, the dynamic matching ability between tool geometry parameter adjustment schemes and working conditions, material properties, and historical machining data is limited. Existing optimization measures are mostly empirical corrections, lacking systematic data-driven and feedback verification mechanisms. In addition, the construction of process databases and the archiving of data throughout the machining process are still incomplete, making it difficult to support multiple rounds of sample accumulation and knowledge reinjection. As a result, process optimization and defect prevention have long been limited to post-analysis and manual adjustment, affecting production efficiency and quality consistency.

[0006] Therefore, in order to address the above problems, there is an urgent need for a self-optimization method and system for milling and grinding processes based on in-machine inspection. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides a self-optimization method and system for milling and grinding processes based on in-machine inspection. This solves the problems of lagging in-machine tool condition detection and the inability of quantitative defect characteristics to drive online closed-loop optimization of tool geometric parameters, which leads to limited efficiency, quality consistency, and tool life.

[0009] Technical solution

[0010] To achieve the above objectives, the present invention provides the following technical solution: a milling process self-optimization method and system based on in-machine inspection, comprising: S1, completing the acquisition and normalization processing of multi-source signals and surface morphology in the in-machine environment, generating characteristic deviation components, deviation scales, load characteristic scales, and defect indices before geometric adjustment, storing them, and then constructing a milling process database; S2, performing defect cause coupling analysis through multi-source characteristic deviation data, and realizing defect classification and geometric adjustment guidance based on the defect cause analysis results; S3, comprehensively evaluating the geometric driving strength by integrating cause correlation measurement, geometric deviation characterization, and load characteristic evaluation, and realizing the execution of load reduction conservative machining control and tool geometry reverse reconstruction; S4, evaluating the reconstruction benefits by synergistically assessing the changes in defect characterization and adjustment strength before and after structural reconstruction, and completing closed-loop verification of backtracking correction, configuration solidification, and sample sedimentation.

[0011] Further, the specific steps for acquiring and normalizing on-machine multi-source signals and surface morphology are as follows: Acquiring on-machine signal data: The spindle vibration acceleration time sequence is obtained through the vibration sensor in the multi-source detection module; the cutting force triaxial component time sequence is obtained through the triaxial force sensor in the tool holder area; the acoustic emission count rate and envelope energy time sequence are obtained through the acoustic emission sensor; and the tool near-cutting zone temperature time sequence is obtained through the temperature sensor. Each measurement value is bound to the sampling time, and vibration signal sequences, cutting force signal sequences, acoustic emission signal sequences, and temperature signal sequences are established according to the machining time slice. Acquiring surface morphology data: During machining breaks, the vision unit and confocal unit scan the toolpath coverage area to acquire arithmetic mean roughness, surface contour morphology field, and spatial spectrum data. The texture main frequency and several auxiliary frequencies are identified from the spatial spectrum. The process involves several steps: First, extracting the vertical dimension, width, and root thickness of the burrs using image processing methods. Then, writing the arithmetic mean roughness, texture frequency index, and burr size index, along with the processing time slice identifier, into the surface morphology record. Next, collecting workpiece material property data: reading the elastic modulus, yield strength, hardness level, thermal conductivity, and coefficient of linear expansion from the material grade record, and writing the material property parameters, processing batch identifier, and time slice identifier into the material property record. Finally, collecting tool geometry and operating condition data: reading the current set values ​​of the rake angle, clearance angle, inclination angle, and cutting edge radius, along with the corresponding geometric reference values, from the tool geometry parameter record table. Obtaining the feed rate, spindle speed rate, axial depth of cut, and radial width of cut from the CNC device operation record, and writing the tool geometry parameters, geometric reference values, and processing condition parameters, along with the time slice identifier, into the operating condition record table.

[0012] Further, the specific steps for generating and storing characteristic deviation components, deviation scales, load characteristic scales, and pre-geometric adjustment defect indices to construct a milling process database are as follows: Denoising, de-drifting, and outlier removal are performed on the acquired vibration signals, cutting force signals, acoustic emission signals, and temperature signals. At a unified time granularity, the vibration amplitude characteristics, vibration frequency characteristics, cutting force fluctuation characteristics, acoustic emission energy characteristics, and temperature peak characteristics of each time slice are calculated and compared with the nominal values ​​in the reference sample to generate item-by-item characteristic deviation components. The total number of characteristic deviation components is recorded as the number of characteristic components. Based on the differences between the current rake angle, clearance angle, cutting edge inclination angle, and cutting edge radius and the geometric reference values, the deviation scale is obtained through normalization and amplitude synthesis. Simultaneously, the geometric deviation scale is read from the geometric parameter configuration table and written into the geometric deviation scale field to provide the deviation scale. The reference level in geometrically driven evaluation is as follows: Load characteristic scales are synthesized based on the normalized results of vibration amplitude characteristics and cutting force fluctuation characteristics at a unified time granularity. Simultaneously, load scales are read from the working condition parameter configuration table and written into the load scale field to provide the reference level of the load characteristic scale in geometrically driven evaluation. The defect index output process is executed using arithmetic mean roughness, texture frequency index, burr size index, acoustic emission energy characteristics, and temperature peak characteristics as inputs: Each item is scaled to between zero and one according to the upper and lower reference limits to obtain five dimensionless deviation sub-indices. These five dimensionless deviation sub-indices are then multiplied in a fixed order, and the fifth root of the product is taken. The resulting single value is recorded as the defect index before geometric adjustment. After storing the standardized and normalized tool geometry parameters, workpiece material property parameters, and machining process parameters, a milling process database is constructed.

[0013] Furthermore, the specific steps for defect cause coupling analysis using multi-source feature deviation data are as follows: Obtain the feature deviation components and the number of feature components; within each time slice, select the maximum value from the absolute values ​​of all feature deviation components as the multidimensional feature deviation maxima; square each feature deviation component, multiply it by the corresponding feature component adjustment coefficient, add a negative sign before the product, and then input it into the natural exponential function for calculation, adding one to the output result to obtain the intermediate value corresponding to each feature; multiply all intermediate values ​​sequentially to form a product, and then perform a geometric mean operation on the product according to the number of feature components to obtain the geometric mean term; negative sign the multidimensional feature deviation maxima before inputting it into the natural exponential function to obtain the deviation attenuation term; finally, multiply the deviation attenuation term by the geometric mean term to obtain the defect identification confidence value.

[0014] Furthermore, the specific steps for defect classification and geometric adjustment guidance based on the defect cause analysis results are as follows: By comparing the defect identification confidence value and confidence threshold in real time, when the defect identification confidence value is less than the confidence threshold, the feed rate and spindle speed are reduced, the axial depth of cut and radial width of cut remain unchanged, the acoustic emission and vibration sampling time is extended and the surface morphology scanning frequency is increased, without triggering tool geometry parameter reconstruction, only the current multi-source features and defect identification confidence value are archived to the defect sample set; when the defect identification confidence value is greater than or equal to the confidence threshold, the cause-driven geometry optimization process is executed: the cause type is mapped to an executable tool geometry adjustment action, when the cause points to increased back face friction, the back angle is increased; when the cause points to the expansion of the plastic deformation zone, the rake angle is increased and the cutting edge radius is decreased; when the cause points to poor chip removal and material lateral flow, the cutting edge inclination angle is adjusted to change the chip flow direction and cut... Burr reduction is achieved by adjusting the back angle and cutting edge tilt angle in tandem to weaken the excitation source when the flutter excitation is enhanced. After geometric reconstruction, a new round of surface scanning is triggered. The changes in arithmetic mean roughness, texture frequency, burr size, acoustic emission energy, and temperature peak are compared and the defect feature vector is updated. When the comparison results show that the roughness decreases, the texture frequency tends to stabilize, the burr size shrinks, the acoustic emission energy weakens, and the temperature peak drops, the current geometric parameter configuration is maintained and the subsequent processing continues. When the comparison results show that any index is still greater than the corresponding quality control threshold, the comprehensive defect index is sorted in descending order based on the change in comprehensive defect index caused by each geometric adjustment action. The one with the largest decrease in comprehensive defect index is selected and retained. The remaining geometric adjustment actions are rolled back and the reconstruction process is executed again. In m consecutive rounds of verification, the comprehensive defect index is always kept below the quality control upper limit.

[0015] Furthermore, the specific steps for evaluating the geometric driving strength by integrating causal correlation measurement, geometric deviation characterization, and load characteristic evaluation are as follows: obtain the defect identification confidence value, deviation scale, geometric deviation scale, load characteristic scale, and load scale; divide the deviation scale by the geometric deviation scale to obtain the geometric ratio term; perform hyperbolic tangent function operation on the geometric ratio term to obtain the geometric saturation term; divide the load characteristic scale by the load scale to obtain the load ratio term; perform hyperbolic tangent function operation on the load ratio term to obtain the load saturation term; multiply the defect identification confidence value by the geometric saturation term to obtain the coupled geometric term; multiply the coupled geometric term by the load saturation term to obtain the geometric driving response value.

[0016] Furthermore, the specific steps for implementing load-reducing conservative machining control and tool geometry reverse reconstruction are as follows: By comparing the geometric drive response value and the response threshold in real time, when the geometric drive response value is less than the response threshold, the feed rate and spindle speed are reduced, while the rake angle, clearance angle, inclination angle, and cutting edge radius are constrained to maintain their current settings. Only the defect feature vector, workpiece material property parameters, and machining process parameter vector, along with the geometric drive response value, are archived into the training dataset. When the geometric drive response value is greater than or equal to the response threshold, the defect feature vector, workpiece material property parameters, machining process parameter vector, defect performance data from historical machining samples, working condition data, and tool geometry parameter data are input to construct a multi-dimensional geometric control feature sequence covering the same machining time window. Through sliding window feature extraction, multi-source feature fusion, and dynamic feature filtering, the analysis is first based on the cutting mechanics mechanism and friction behavior. The basic geometric adjustment solutions for the rake angle, clearance angle, inclination angle, and cutting edge radius are calculated based on the material forming law. Then, an incremental correction is performed on the basic geometric adjustment solutions using a data-driven solution framework with multivariate regression and gradient boosting as the core, forming a tool geometry inverse solution model. The tool geometry inverse solution model adjusts the feature weights and parameter configurations according to the disturbance amplitude of the defect scenario data, calculates and outputs a geometric adjustment vector composed of the rake angle increment, clearance angle increment, inclination angle adjustment, and cutting edge radius adjustment, and recalculates the geometric drive response value. If it is less than the response threshold, the current geometric parameter configuration is maintained; if the geometric drive response value is still greater than or equal to the response threshold, the feature sequence is updated and the tool geometry inverse solution model is called again to output a new geometric adjustment vector, until the geometric drive response value falls back to less than the response threshold, the geometric parameter configuration is maintained, and the subsequent geometric parameter dynamic reconstruction module is entered.

[0017] Furthermore, the specific steps for collaboratively evaluating the reconstruction benefits by assessing the changes in defect characterization and adjustment intensity before and after structural reconstruction are as follows: Obtain the geometrically driven response value and the defect index before geometric adjustment; execute the defect index output process on the arithmetic mean roughness, texture frequency index, burr size index, acoustic emission energy characteristics, and temperature peak characteristics obtained in the new round of data collection, and record the obtained values ​​as the defect index after geometric adjustment; calculate the defect index before geometric adjustment minus the defect index after geometric adjustment to obtain the defect improvement difference; add the defect index before geometric adjustment to a minimum positive number to obtain a stable scale term; divide the defect improvement difference by the stable scale term to obtain an improvement ratio term; perform hyperbolic tangent function operation on the improvement ratio term to obtain an improvement saturation term; multiply the geometrically driven response value by the improvement saturation term to obtain a reconstruction evaluation value that comprehensively reflects the matching degree between the investment in this round of geometric reconstruction and the defect improvement effect.

[0018] Furthermore, the specific steps for completing the closed-loop verification of backtracking correction, configuration solidification, and sample sedimentation are as follows: By comparing the reconstruction evaluation value and the evaluation threshold in real time, when the reconstruction evaluation value is less than the evaluation threshold, it is determined that the tooling reconstruction benefit corresponding to the geometric adjustment vector in the previous round is insufficient. The geometric reconstruction actuator restores the rake angle, clearance angle, blade tilt angle, and blade radius to the geometric adjustment vector combination recorded in the previous round, reduces the feed rate and spindle speed, and tightens the axial depth of cut and radial width of cut. At the same time, the on-machine multi-source detection module adds an additional acoustic emission, vibration, and surface morphology acquisition. The edge computing unit outputs the failed markers and carries the vibration peaks, acoustic emission energy, and surface roughness before and after reconstruction. The comparison results of the degree, texture frequency, and burr size before and after are used to archive the current geometric adjustment amount, multi-source detection results, and reconstruction evaluation value as negative samples into the milling process database. These samples, along with the failed markers, the comparison results before and after, and the updated defect feature vector, are sent back to the tool geometry reverse solution module as input for the next round of solution. When the reconstruction evaluation value is greater than or equal to the evaluation threshold, the current combination of rake angle, clearance angle, inclination angle, and cutting edge radius remains unchanged. The geometric adjustment vector of this round and the comparison results before and after reconstruction are used to form a data pair and archived into the milling process database. This data is used to update the correlation between subsequent defect identification and geometric reverse solution, thereby completing the verification and knowledge feedback loop.

[0019] Furthermore, a second aspect of the present invention provides a milling process self-optimization system based on in-machine detection, applying an in-machine detection-based milling process self-optimization method, comprising: an in-machine multi-source detection module, used to complete the acquisition and normalization processing of in-machine multi-source signals and surface morphology, generating characteristic deviation components, deviation scales, load characteristic scales, and defect indices before geometric adjustment, and storing them to construct a milling process database; a defect diagnosis and cause correlation module, used to perform defect cause coupling analysis through multi-source characteristic deviation data, and realize defect classification and geometric adjustment guidance based on the defect cause analysis results; a tool geometry reverse optimization module, used to integrate cause correlation measurement, geometric deviation characterization, and load characteristic evaluation of geometric drive strength, and realize the execution of load reduction conservative machining control and tool geometry reverse reconstruction; and a geometric parameter dynamic reconstruction module, used to complete closed-loop verification of backtracking correction, configuration solidification, and sample sedimentation by collaboratively evaluating the changes in defect characterization and adjustment strength before and after structural reconstruction.

[0020] Beneficial effects

[0021] The present invention has the following beneficial effects:

[0022] (1) This invention constructs a multidimensional feature deviation maximum value for the feature deviation component and calculates the defect identification confidence value by combining the feature component adjustment coefficient and geometric mean. This achieves the effect of quantifying the degree of defect cause matching and unifying the classification judgment basis, effectively solving the problem of insufficient classification consistency caused by the reliance on a single signal indicator and experience interpretation in the prior art for defect diagnosis.

[0023] (2) This invention constructs a geometric driving response value by integrating defect identification confidence value, deviation scale, geometric deviation scale, load characteristic scale and load scale, and suppresses the amplification effect of extreme fluctuations on the evaluation results with a hyperbolic tangent saturation link, thereby achieving a more stable effect of controllable mapping of processing state driving intensity and selection of processing control branches, effectively solving the problem of unclear basis for control strategy switching caused by the lack of unified intensity characterization in the prior art.

[0024] (3) In this invention, when the geometric drive response value meets the triggering condition, defect feature vector, material property parameters, machining process parameter vector and historical sample information are introduced. First, a basic solution for geometric adjustment is formed, then incremental correction is performed and the geometric adjustment vector is output. In this way, the quantitative features of defects are mapped to the adjustment amount of the front angle, back angle, blade tilt angle and blade arc radius. This effectively solves the problem that in the prior art, defect features are difficult to guide tool geometric adjustment in reverse and rely on trial and error for a long time.

[0025] (4) In this invention, by forming a sample sedimentation of the failure mark, the comparison results before and after reconstruction, the geometric adjustment amount and the reconstruction evaluation value and feeding it back into the tool geometry reverse solution process, positive and negative samples are formed and the correlation is continuously updated, thereby realizing the effect of knowledge accumulation driven by closed-loop verification and the improvement of the quality of subsequent solution input. This effectively solves the problem that the lack of closed-loop backfeeding in the prior art makes it difficult to structure and reuse historical experience.

[0026] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0027] Figure 1 This is a flowchart of a milling process self-optimization method based on in-machine inspection according to the present invention;

[0028] Figure 2 This is a structural diagram of a milling process self-optimization system based on in-machine inspection according to the present invention;

[0029] Figure 3 This is a three-dimensional surface diagram showing the geometric drive response and process adjustment effect under different milling conditions of the present invention;

[0030] Figure 4 This is a flowchart illustrating the reconstructed evaluation and parameter adjustment decision-making process of this invention. Detailed Implementation

[0031] 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.

[0032] Please see Figures 1-4 This invention provides a technical solution: a milling process self-optimization method and system based on in-machine detection, comprising: S1, completing the acquisition and normalization processing of multi-source signals and surface morphology in the machine, generating characteristic deviation components, deviation scales, load characteristic scales, and defect indices before geometric adjustment, and storing them to construct a milling process database; S2, performing defect cause coupling analysis through multi-source characteristic deviation data, and realizing defect classification and geometric adjustment guidance based on the defect cause analysis results; S3, comprehensively evaluating the geometric driving strength by integrating cause correlation measurement, geometric deviation characterization, and load characteristic evaluation, and realizing the execution of load reduction conservative machining control and tool geometry reverse reconstruction; S4, evaluating the reconstruction benefits by synergistically assessing the changes in defect characterization and adjustment strength before and after structural reconstruction, and completing closed-loop verification of backtracking correction, configuration solidification, and sample sedimentation.

[0033] Specifically, the steps for acquiring and normalizing on-machine multi-source signals and surface morphology data are as follows: Acquiring on-machine signal data: The spindle vibration acceleration time sequence is obtained through the vibration sensor in the multi-source detection module; the cutting force triaxial component time sequence is obtained through the triaxial force sensor in the tool holder area; the acoustic emission count rate and envelope energy time sequence are obtained through the acoustic emission sensor; and the tool near-cutting zone temperature time sequence is obtained through the temperature sensor. Each measured value is bound to the sampling time, and then divided into continuous, non-overlapping small time segments according to the processing time-slice division rules. Within each time segment, the corresponding measured values ​​are accumulated and arranged, thereby establishing vibration signal sequences, cutting force signal sequences, acoustic emission signal sequences, and temperature signal sequences according to the processing time segments. This aligns the multi-source signals at a unified time granularity, providing a unified analysis window for subsequent extraction of characteristic deviation components, deviation scales, and load characteristic scales, as well as the calculation of defect indices before geometric adjustment. Acquiring surface morphology data: During processing breaks, the vision unit and confocal unit scan the toolpath coverage area to acquire the arithmetic mean roughness value, surface contour morphology field, and spatial spectrum data. The dominant texture frequency and several auxiliary frequency components are identified in the inter-spectral density. Image processing methods are used to extract the vertical dimension, width, and root thickness of the burrs. The arithmetic mean roughness value, dominant texture frequency index, and burr size index, along with the processing time slice identifier, are written into the surface morphology record, thus correlating surface quality information with on-machine signal data at the time slice level. Workpiece material property data is collected: elastic modulus, yield strength, hardness level, thermal conductivity, and coefficient of linear expansion are read from the material grade record. The material property parameters, along with the processing batch identifier and time slice identifier, are written into the material... Material properties are recorded to correlate material effects within the same time slice; tool geometry and working condition data are collected: the current set values ​​of rake angle, clearance angle, inclination angle, and cutting edge radius, as well as the corresponding geometric reference values, are read from the tool geometry parameter record table; the feed rate, spindle speed rate, axial depth of cut, and radial width of cut are obtained from the CNC device operation record; the tool geometry parameters, geometric reference values, and machining working condition parameters, along with the time slice identifier, are written into the working condition record table, so that on-machine signals, surface morphology, material properties, and tool geometry working conditions form a unified data closed loop in the machining time slice dimension.

[0034] In this implementation scheme, machine vibration, cutting force, acoustic emission, temperature signals, surface morphology, workpiece material properties, tool geometry parameters, and machining conditions are synchronously acquired and normalized at a uniform machining time slice granularity. This constructs a basic dataset containing characteristic deviation components, the number of characteristic components, deviation scale, load characteristic scale, and defect indices before geometric adjustment, providing consistent and complete input data support for subsequent processing.

[0035] Specifically, the steps for generating and storing characteristic deviation components, deviation scales, load characteristic scales, and pre-geometric adjustment defect indices to construct a milling process database are as follows: Denoising, de-drifting, and outlier removal are performed on the acquired vibration signals, cutting force signals, acoustic emission signals, and temperature signals. At a uniform time granularity, the vibration amplitude characteristics, vibration frequency characteristics, cutting force fluctuation characteristics, acoustic emission energy characteristics, and temperature peak characteristics for each time slice are calculated. These are then compared with the nominal values ​​in the reference samples to generate item-by-item characteristic deviation components. The reference samples are those that have passed quality acceptance in the milling process database and… The nominal values ​​are obtained by filtering historical records that are consistent with or within the same process range as the current machining record, including machining batch, material grade, tool model, rake angle, clearance angle, inclination angle, cutting edge radius, feed rate, spindle speed rate, axial depth of cut, and radial width of cut. The nominal values ​​are taken as the statistical center values ​​of the filtered records at the corresponding time slice and corresponding feature items, and written into the reference sample nominal value table. The total number of feature deviation components is recorded as the feature component quantity, and the feature component quantity and feature deviation components are written together into the multi-source feature record. The values ​​are then calculated based on the differences between the current rake angle, clearance angle, inclination angle, and cutting edge radius and the geometric reference values. The deviation scale is obtained through normalization and amplitude synthesis. Simultaneously, the geometric deviation scale is read from the geometric parameter configuration table and written into the geometric deviation scale field. The geometric parameter configuration table includes at least field names, field meanings, applicable material grades, applicable tool models, corresponding machining condition ranges, geometric reference values, geometric deviation scales, and version identifiers. The geometric deviation scale is statistically obtained from the distribution of differences between the rake angle, clearance angle, inclination angle, and cutting edge radius relative to the geometric reference values ​​in the reference sample and is updated synchronously with the version identifier. This scale is used to provide a reference level for the deviation scale in geometrically driven evaluation. Based on the vibration at a unified time granularity... The normalized results of the vibration amplitude characteristics and cutting force fluctuation characteristics are combined to form the load characteristic scale. At the same time, the load scale is read from the working condition parameter configuration table and written into the load scale field. The working condition parameter configuration table includes at least the field name, field meaning, applicable material grade, applicable tool model, corresponding feed rate range, spindle speed range, axial depth of cut range, radial width of cut range, load scale and version identifier. The load scale is obtained by statistically analyzing the fluctuation distribution of vibration amplitude characteristics and cutting force fluctuation characteristics in the same working condition range in the reference sample, and is used to give the reference level of the load characteristic scale in geometric drive evaluation.A defect index output process is executed using arithmetic mean roughness, texture frequency index, burr size index, acoustic emission energy characteristic, and temperature peak characteristic as inputs. This process employs consistent direction and scaling rules for each item: each item is scaled to between zero and one using upper and lower reference limits to obtain a dimensionless deviation sub-index. Specifically, the arithmetic mean roughness, burr size index, acoustic emission energy characteristic, and temperature peak characteristic are constructed as dimensionless deviation sub-indexes according to the direction where larger values ​​indicate more significant defects. The texture frequency index is constructed as a dimensionless deviation sub-index based on its deviation from the nominal texture frequency of the reference sample, such that a larger value indicates a more significant deviation from the reference sample state. The upper and lower reference limits are taken from the milling process database. Accepted records for the same material grade, tool type, and operating condition range are assigned boundary values ​​and written into a reference boundary table. The reference boundary table includes at least the material grade, tool type, operating condition range identifier, index name, upper reference limit, lower reference limit, and version identifier. After obtaining five dimensionless deviation sub-indices, they are multiplied in a fixed order, and the fifth root of the product is taken. The resulting single value is recorded as the defect index before geometric adjustment, ensuring that the defect index before geometric adjustment increases synchronously when any deviation of a sub-indicator increases, forming a consistent defect characterization scale. The standardized and normalized tool geometry parameters, workpiece material property parameters, and machining process parameters, along with the reference sample nominal value table, reference boundary table, and version identifier, are stored to construct a milling process database.

[0036] In this implementation scheme, noise reduction, drift reduction, and outlier removal are performed on vibration signals, cutting force signals, acoustic emission signals, and temperature signals. Vibration amplitude characteristics, vibration frequency characteristics, cutting force fluctuation characteristics, acoustic emission energy characteristics, and temperature peak characteristics are extracted according to the processing time slice. Characteristic deviation components are generated by comparing with the nominal values ​​of reference samples, and the number of characteristic components is counted. Simultaneously, the deviation scale is obtained based on the differences between the rake angle, clearance angle, cutting edge inclination angle, and cutting edge radius relative to the geometric reference value, and a geometric deviation scale is associated with it. Load characteristics are obtained based on the vibration amplitude characteristics and cutting force fluctuation characteristics. The scale is correlated with the load scale, and then the arithmetic mean roughness, texture frequency index, burr size index, acoustic emission energy characteristics and temperature peak characteristics are mapped into dimensionless deviation sub-indices in a unified direction and aggregated to obtain the defect index before geometric adjustment. Together with the reference upper limit, reference lower limit, reference sample nominal value table, reference boundary table and version identifier, they are written into the milling process database. In this way, the scattered on-machine inspection information is transformed into a comparable, traceable and reusable quantitative characterization, providing a stable data foundation for defect diagnosis, geometric adjustment guidance, tool geometry reverse reconstruction and verification recharge.

[0037] Specifically, the steps for defect cause coupling analysis using multi-source feature deviation data are as follows: During defect identification, firstly, the feature deviation components and their quantities are obtained. Within each time slice, the maximum value among the absolute values ​​of all feature deviation components is selected as the multidimensional feature deviation maxima. For each feature deviation component, it is first squared and then multiplied by the corresponding feature component adjustment coefficient. This adjustment coefficient is obtained through regression analysis and cross-validation based on historical defect sample sets, typically ranging from 0.1 to 5.0, used to characterize the sensitivity and contribution of different detection features in defect identification, avoiding subjective human intervention and ensuring data-driven objectivity. A negative sign is added before the product, and the result is fed into the natural exponential function for calculation. One is added to the output to obtain the intermediate value for each feature. Subsequently, all intermediate values ​​are multiplied sequentially to form a product, and then a geometric mean is calculated on the product according to the number of feature components to obtain the geometric mean term. Simultaneously, the multidimensional feature deviation maxima are negatively squared and fed into the natural exponential function for calculation to obtain the deviation attenuation term. Finally, the deviation attenuation term is multiplied by the geometric mean term to output the defect identification confidence value, providing a quantitative basis for subsequent defect type judgment and process decision-making.

[0038] The specific calculation method for the defect identification confidence value is as follows:

[0039] ;

[0040] In the formula, D represents the defect identification confidence value, used to characterize the matching strength between the current multi-source detection features and the target causal pattern; max This represents the maximum value of the multidimensional characteristic deviation, used to reflect the degree of deviation of the current operating condition from the reference operating condition; This represents the i-th feature deviation component, used to describe the deviation ratio of a single detection index relative to the reference value; represents the adjustment coefficient of the i-th feature component, used to adjust the influence intensity of the corresponding feature in cause identification; n represents the number of feature components participating in the product term and geometric mean operation, which is determined by the number of features currently selected for coupling calculation; e represents the natural exponent base, which is a commonly used mathematical constant.

[0041] In this implementation scheme, by realizing the normalization and fusion of multi-dimensional feature information and the allocation of sensitivity weights, the final output is a defect identification confidence value that can comprehensively reflect the degree of matching between the current working condition and the target defect pattern, providing a scientific and quantifiable decision-making basis for subsequent defect classification and geometric parameter adjustment.

[0042] Specifically, the steps for determining defect types and providing geometric adjustment guidance based on defect cause analysis results are as follows: By comparing the defect identification confidence value and confidence threshold in real time, when the defect identification confidence value is less than the confidence threshold, the feed rate and spindle speed are reduced in the CNC device according to the current working conditions, while keeping the axial depth of cut and radial width of cut unchanged. The multi-source observation density under the same defect scenario is increased by extending the acoustic emission and vibration sampling time and increasing the surface morphology scanning frequency. Tool geometry parameter reconstruction is not triggered; only the vibration, cutting force, acoustic emission, temperature, and surface parameters of the current event are recorded. Multi-source features composed of surface morphology, along with defect identification confidence values, are written into the defect sample set to provide a data foundation for subsequent confidence threshold adjustment and causal mode update. When the defect identification confidence value is not less than the confidence threshold, the causal-driven geometry optimization process is initiated, mapping the causal type output from the defect identification stage to specific tool geometry adjustment actions. When the causal cause points to increased back face friction, the back angle is increased; when the causal cause points to the expansion of the plastic deformation zone, the rake angle is increased and the cutting edge radius is decreased; when the causal cause points to poor chip removal and material lateral flow, the cutting edge inclination angle is adjusted to change the chip flow direction and reduce burrs. When the cause of a burr is indicated by increased flutter excitation, the vibration excitation intensity is reduced by coordinating the back angle and the blade tilt angle. After geometric reconstruction, a new round of surface scanning is triggered by the on-machine multi-source detection module. The changes in arithmetic mean roughness, texture frequency, burr size, acoustic emission energy, and temperature peak are quantitatively compared and the defect feature vector is updated. At the same time, the comprehensive defect index is calculated based on the updated multi-source features. When the comparison results show that the arithmetic mean roughness decreases, the texture frequency tends to stabilize, the burr size shrinks, the acoustic emission energy decreases, and the temperature peak falls back, and the comprehensive defect index does not exceed the quality control upper limit, the current geometric parameter configuration is maintained and the established processing flow continues. When the comparison results show that at least one quality characterization value still exceeds the corresponding quality control threshold, a sorting sequence is constructed based on the changes in the comprehensive defect index brought about by different geometric adjustment actions. The geometric adjustment action with the highest ranking is selected and retained, and the remaining geometric adjustment actions are rolled back. The geometric reconstruction and detection process is re-executed based on the retained actions. The comprehensive defect index is continuously monitored during the continuous m rounds of verification to keep the comprehensive defect index below the quality control upper limit.

[0043] In this implementation scheme, by classifying and judging the confidence value and confidence threshold of defect identification, when the confidence level is less than the confidence threshold, feed reduction and multi-source detection encryption are performed to collect defect samples. The corresponding tool geometry adjustment action is triggered according to the cause type. The adjustment results are screened and cyclically corrected by combining the differences in arithmetic mean roughness, texture frequency, burr size, acoustic emission energy and temperature peak before and after. This makes multi-source detection, defect classification, geometric reconstruction and quality judgment form a closed-loop control link, thereby realizing a graded intervention strategy for different defect causes and a stable quality control effect during the processing.

[0044] Specifically, the steps for evaluating the geometric driving strength by integrating causal correlation measurement, geometric deviation characterization, and load characteristic assessment are as follows: Obtain the defect identification confidence value, deviation scale, geometric deviation scale, load characteristic scale, and load scale. The defect identification confidence value comes from the coupled calculation results of the multi-source characteristic deviation components and the number of characteristic components in the current machining time slice by the defect diagnosis and causal correlation module. The deviation scale is obtained by synthesizing the amplitude of the differences between the rake angle, clearance angle, cutting edge inclination angle, and cutting edge radius relative to the geometric reference value. The geometric deviation scale is read from the calibration record in the geometric parameter configuration table and bound to the tool model, material grade, and working condition file version. The load characteristic scale is synthesized from the vibration amplitude characteristics and cutting force fluctuation characteristics. The load scale is read from the load grading calibration record in the working condition parameter configuration table and bound to the machining batch and time slice identifier. The deviation scale... Dividing the geometric deviation scale by the geometric ratio term yields the geometric ratio term, and performing a hyperbolic tangent function operation on the geometric ratio term yields the geometric saturation term, ensuring that the contribution of geometric deviation to the response remains within a controlled range and avoiding extreme ratio amplification. Dividing the load characteristic scale by the load scale yields the load ratio term, and performing a hyperbolic tangent function operation on the load ratio term yields the load saturation term, ensuring that the contribution of load fluctuation to the response remains within a controlled range and avoiding numerical mutations caused by transient shocks. Multiplying the defect identification confidence value by the geometric saturation term yields the coupled geometric term, achieving a joint characterization of the cause's credibility and the influence of geometric deviation. Multiplying the coupled geometric term by the load saturation term yields the geometrically driven response value, thus forming a quantitative result of geometric adjustment requirements for the same machining time window, and providing consistent input for triggering judgments of response threshold comparison, load reduction control, and tool geometry inverse solution.

[0045] The specific calculation method for the geometry-driven response value is as follows:

[0046] ;

[0047] In the formula, This represents the geometry drive response value, used to describe the intensity of the current tool geometry adjustment requirement; G represents the defect identification confidence value, which is a quantitative result of the reliability of the causal mode; G represents the deviation scale, which is used to reflect the sensitivity of the cutting edge inclination angle and the radius of the cutting edge blunt circle to the target defect. The geometric deviation scale is used to define the position where G enters the saturation region during the calculation; It represents the load characteristic scale, reflecting the combined magnitude of vibration amplitude and cutting force fluctuation during the machining process; This indicates the load scale, used to define the position where L enters the saturation region in the calculation.

[0048] In this embodiment, the processing time for group 1 is 15 minutes, the material is 45# steel, the defect identification confidence value is 0.32, the deviation scale is 0.18, the load characteristic scale is 0.73, the geometric deviation scale is 0.52, the load scale is 0.82, and the geometric drive response value is 0.086; the processing time for group 2 is 42 minutes, the material is 304 stainless steel, the defect identification confidence value is 0.58, the deviation scale is 0.41, the load characteristic scale is 1.12, the geometric deviation scale is 0.48, the load scale is 0.76, and the geometric drive response value is 0.517; the processing time for group 3 is 78 minutes, the material is titanium alloy TC4, the defect identification confidence value is 0.83, the deviation scale is 0.35, and the load... The first group had a feature scale of 2.08, a geometric deviation scale of 0.55, a load scale of 0.79, and a geometric drive response value of 0.941. The fourth group had a machining time of 125 minutes, used aluminum alloy 7075, had a defect identification confidence value of 0.67, a deviation scale of 1.25, a load feature scale of 1.38, a geometric deviation scale of 0.50, a load scale of 0.80, and a geometric drive response value of 0.924. The fifth group had a machining time of 189 minutes, used mold steel SKD11, had a defect identification confidence value of 1.28, a deviation scale of 0.82, a load feature scale of 2.65, a geometric deviation scale of 0.53, a load scale of 0.84, and a geometric drive response value of 1.416.

[0049] Table 1. Milling process self-optimization adjustment data based on multi-parameter evaluation

[0050]

[0051] like Figure 3 As shown, this is a three-dimensional surface diagram of the geometric drive response and process adjustment effect under different milling conditions provided in the embodiments of this application. Combined with the data in Table 1... Figure 3It is evident that the process response varies significantly under different materials and processing conditions, with characteristic parameters closely matching actual working conditions. For example, when titanium alloy TC4 experiences sudden chatter, the load characteristic scale reaches 2.08, and the geometric drive response value rises to 0.941, triggering the system to perform a coordinated adjustment of reducing the rake angle and speed. During the break-in phase of new tools, the parameters of 45 steel are all low compared to other materials, with a geometric drive response value of only 0.086, resulting in stable processing and demonstrating good stability in the initial working conditions. When mold steel SKD11 encounters material hard points, the defect cause coupling value reaches as high as 1.28, and the geometric drive response value exceeds 1.416, prompting the system to immediately initiate a combined control of increasing the rake angle and emergency speed reduction, reflecting a strong response capability to extreme working conditions. This indicates that the properties of the processed material and the real-time working conditions jointly determine the optimization path, and the geometric drive response mechanism plays a significant role in the fusion and judgment of multi-source information. For example, when machining 7075 aluminum alloy for 125 minutes, the high geometric deviation triggered an adjustment of the cutting edge angle to improve chip removal. Conversely, 304 stainless steel, despite a machining time of only 42 minutes, entered a monitoring state earlier due to increased load characteristics. This pattern provides a reliable basis for adaptive optimization of milling processes, allowing for real-time adjustment of geometric parameters and process strategies for different materials and tool conditions. It also provides effective data support for machining quality prediction, anomaly warning, and intelligent decision-making, contributing to a comprehensive improvement in machining consistency, tool life, and overall process stability.

[0052] In this implementation scheme, the defect identification confidence value, deviation scale, and load characteristic scale are saturated and coupled under the constraints of geometric deviation scale and load scale to output geometric drive response value. This compresses the degree of tool geometric deviation, the degree of machining load fluctuation, and the defect identification conclusion into the same comparable dimension and controlled dynamic range, supporting the subsequent trigger selection for load reduction conservative machining control and tool geometric reverse reconstruction, and reducing the interference of transient noise and extreme deviation on decision stability.

[0053] Specifically, the steps for implementing load reduction conservative machining control and tool geometry reverse reconstruction are as follows: By comparing the geometric drive response value and the response threshold in real time, when the geometric drive response value is less than the response threshold, the system will synchronously reduce the feed rate and spindle speed rate, while locking the rake angle, clearance angle, inclination angle and cutting edge radius without adjustment, and archive the defect feature vector, workpiece material property parameters and machining process parameter vector together with the current geometric drive response value into the training dataset for subsequent model training and working condition archiving. When the geometric drive response value is greater than or equal to the response threshold, a multi-dimensional geometric control feature sequence covering the same machining time window is constructed by combining the currently collected defect feature vector, workpiece material property parameters, machining process parameter vector, defect performance data of historical machining samples, working condition data, and tool geometric parameter data. Through sliding window feature extraction, multi-source feature fusion, and dynamic feature screening, the basic solutions for geometric adjustment of the rake angle, clearance angle, inclination angle, and cutting edge radius are first calculated using basic theories such as cutting mechanics mechanism constraints, friction behavior analysis, and material forming law. Then, a data-driven framework with multivariate regression and gradient enhancement class integrated structure as the core is used to incrementally correct the basic solution, forming a tool geometry inverse solution model that combines theoretical guidance and data self-adaptation capabilities. The model adjusts feature weights and parameter configurations in real time based on the disturbance magnitude of defect scenario data. It calculates and outputs a geometric adjustment vector composed of the rake angle increment, clearance angle increment, cutting edge inclination angle adjustment, and cutting edge radius adjustment. It then recalculates the geometric drive response value. If the response value falls below the response threshold, the current geometric parameter configuration remains unchanged. If the geometric drive response value is still greater than or equal to the response threshold, the feature sequence is updated and the tool geometry inverse solution model is called again to output a new geometric adjustment vector. This process is repeated until the geometric drive response value drops below the response threshold. Finally, the optimized geometric parameter combination is locked, and the model proceeds to the subsequent geometric parameter dynamic reconstruction module.

[0054] In this implementation scheme, by dynamically comparing the geometric drive response value and the response threshold, intelligent diversion and precise response to the tool geometry parameter adjustment needs under different working conditions are achieved: when the adjustment needs are not obvious, a robust strategy of reducing load and archiving data is prioritized to avoid unnecessary structural changes and ensure machining stability; when the adjustment needs are significant, the optimal adjustment combination of the rake angle, clearance angle, inclination angle and cutting edge radius is progressively calculated and corrected by combining the dual constraints of the mechanism model and the data-driven method until the response value during the machining process drops to a reasonable range, effectively improving tool adaptability and machining quality consistency, while providing a high-quality parameter basis and feature data support for subsequent dynamic reconstruction and continuous optimization.

[0055] Specifically, the steps for collaboratively evaluating the reconstruction benefits by assessing the changes in defect characterization and adjustment intensity before and after structural reconstruction are as follows: Obtain the geometrically driven response value and the defect index before geometric adjustment; perform dimensionless and fusion calculations on the arithmetic mean roughness, texture frequency index, burr size index, acoustic emission energy characteristics, and temperature peak characteristics obtained in the new round of data collection, following the aforementioned defect index output process; register the output results as the defect index after geometric adjustment; based on the above, subtract the defect index after geometric adjustment from the defect index before geometric adjustment to obtain the defect improvement difference; then, superimpose the defect index before geometric adjustment with a minimum positive number to form... The stable scale term is constructed by dividing the difference in defect improvement by the stable scale term. An improvement ratio term is then constructed by applying a hyperbolic tangent mapping to the improvement ratio term to obtain an improvement saturation term. The geometrically driven response value is multiplied by the improvement saturation term to obtain a reconstruction evaluation value that comprehensively reflects the matching degree between the investment in this round of geometric reconstruction and the effect of defect improvement. The minimum positive value is much smaller than the amplitude of commonly used defect indicators. It is selected by statistical analysis of the distribution range of defect indicators before geometric adjustment in historical process data and kept consistent throughout the entire reconstruction evaluation calculation process to avoid numerical fluctuations and calculation instability risks caused by the stable scale term approaching zero.

[0056] In this implementation plan, a reconstruction benefit assessment quantity is constructed by using the geometrically driven response value and the defect index before and after geometric adjustment. The calculation process of the minimum positive number stable defect improvement ratio is used to provide quantitative criteria for subsequent geometric parameter retention, rollback and processing load adjustment based on the reconstruction benefit assessment results, thereby achieving a fine evaluation of the matching degree between geometric reconstruction investment and defect improvement effect.

[0057] Specifically, the steps for completing the closed-loop verification of rollback correction, configuration solidification, and sample sedimentation are as follows: Reconstruct the evaluation value and evaluation threshold through real-time comparison, such as... Figure 4This is a flowchart of the reconstruction evaluation and parameter adjustment decision-making process in this embodiment. When the reconstruction evaluation value is less than the evaluation threshold, it is determined that the tool reconstruction benefit corresponding to the geometric adjustment vector in the previous round is insufficient. The geometric reconstruction execution mechanism restores the rake angle, clearance angle, inclination angle, and cutting edge radius to the geometric adjustment vector combination recorded in the previous round, and writes the restored rake angle, clearance angle, inclination angle, and cutting edge radius into the tool geometric parameter record table. The CNC device simultaneously adjusts the feed rate and spindle speed rate down to the load reduction control rate associated with the geometric adjustment vector combination in the working condition record table, and adjusts the axial depth of cut and radial width of cut to the target values ​​corresponding to the load reduction control in the working condition record table to suppress vibration peak fluctuations and acoustic emission energy transitions. At the same time, the on-machine multi-source detection module adds an acoustic emission, vibration, and surface morphology acquisition. The edge computing unit compares the acquisition results before reconstruction with those after reconstruction based on the machining time slice identifier. The results are paired and compared, and the output of the new and old comparison results are marked with failure and include the vibration peak, acoustic emission energy, surface roughness, texture frequency and burr size before and after reconstruction. The geometric adjustment amount, multi-source detection results and reconstruction evaluation value, together with the evaluation threshold, failure mark, and new and old comparison results are written into the milling process database to form negative samples. The updated defect feature vector and negative sample index are sent back to the tool geometry reverse solution module as input for the next round of solution. When the reconstruction evaluation value is greater than or equal to the evaluation threshold, the current combination of rake angle, clearance angle, cutting edge inclination angle and cutting edge radius remains unchanged. The geometric adjustment vector of this round and the comparison results before and after reconstruction are written into the milling process database as a data pair. At the same time, the reconstruction evaluation value and evaluation threshold are written into the database to form a verification sample, which is used to update the correlation between subsequent defect identification and geometric reverse solution, thereby completing the verification and knowledge feedback loop.

[0058] In this implementation plan, the input and output of the previous round of geometric adjustment vectors are judged in a closed loop by comparing the reconstructed evaluation value with the evaluation threshold in real time. When the judgment fails, the system promptly reverts to the geometric adjustment vector combination recorded in the previous round and simultaneously implements load reduction control and additional on-machine data acquisition. It outputs the failure mark and the comparison results of the vibration peak, acoustic emission energy, surface roughness, texture frequency and burr size before and after reconstruction. The geometric adjustment amount, multi-source detection results and reconstructed evaluation value are precipitated as negative samples and sent back to the tool geometry reverse solution module to encourage subsequent solutions to avoid low-yield adjustment paths. When the judgment passes, the current combination of rake angle, clearance angle, cutting edge inclination angle and cutting edge radius is solidified and the geometric adjustment vector and comparison results of this round are archived to form a reusable positive sample to update the correlation between defect identification and geometric reverse solution. This achieves a verification closed loop of "reverting and correcting - configuration solidification - sample precipitation - knowledge reinjection", reduces the number of trial and error and improves the consistency and traceability of geometric adjustment decisions.

[0059] Specifically, this embodiment provides a milling process self-optimization system based on in-machine detection, applied to a milling process self-optimization based on in-machine detection, including: an in-machine multi-source detection module, used to complete the acquisition and normalization processing of in-machine multi-source signals and surface morphology, generate characteristic deviation components, deviation scales, load characteristic scales and defect indices before geometric adjustment, and store them to construct a milling process database; a defect diagnosis and cause correlation module, used to perform defect cause coupling analysis through multi-source characteristic deviation data, and realize defect classification and geometric adjustment guidance based on the defect cause analysis results; a tool geometry reverse optimization module, used to integrate cause correlation measurement, geometric deviation characterization and load characteristic evaluation of geometric driving strength, realize the execution of load reduction conservative machining control and tool geometry reverse reconstruction; and a geometric parameter dynamic reconstruction module, used to complete the closed-loop verification of backtracking correction, configuration solidification and sample sedimentation by coordinating the evaluation of defect characterization changes and adjustment strength before and after structural reconstruction.

[0060] In this implementation plan, by modularly integrating multi-source sensing, intelligent diagnosis, geometric reverse control and dynamic reconstruction, the process of multi-dimensional feature acquisition, defect cause identification, parameter adaptive adjustment and closed-loop verification in the milling process is fully connected, providing intelligent support for the dynamic optimization of process quality and knowledge accumulation under complex working conditions.

[0061] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0062] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A self-optimization method for milling and grinding processes based on in-machine inspection, characterized in that, Includes the following steps: S1 completes the acquisition and normalization of multi-source signals and surface morphology in the machine, generates characteristic deviation components, deviation scales, load characteristic scales and defect indices before geometric adjustment, and stores them to build a milling process database. S2 uses multi-source characteristic deviation data to perform defect cause coupling analysis, and uses the defect cause analysis results to determine the defect type and provide geometric adjustment guidance. S3 integrates causal correlation measurement, geometric deviation characterization, and load characteristic evaluation of geometric drive strength to achieve load reduction conservative machining control and tool geometry reverse reconstruction; S4, through the collaborative evaluation of the changes in defect characterization and adjustment intensity before and after structural reconstruction, the reconstruction benefits are realized, and the closed-loop verification of rollback correction, configuration solidification and sample sedimentation is completed. The generation of characteristic deviation components, deviation scales, load characteristic scales, and pre-geometric adjustment defect indices includes: performing denoising, de-drifting, and outlier removal processing on the acquired vibration signals, cutting force signals, acoustic emission signals, and temperature signals; calculating the vibration amplitude characteristics, vibration frequency characteristics, cutting force fluctuation characteristics, acoustic emission energy characteristics, and temperature peak characteristics for each time slice at a uniform time granularity; and generating item-by-item characteristic deviation components by comparing them with the nominal values ​​in the reference sample; and based on the differences between the current rake angle, clearance angle, cutting edge inclination angle, and cutting edge radius and the geometric reference values, after regression... The deviation scale is obtained by normalization and amplitude synthesis; the load characteristic scale is synthesized based on the normalization results of vibration amplitude characteristics and cutting force fluctuation characteristics under a unified time granularity; the defect index output process is executed with arithmetic mean roughness, texture main frequency index, burr size index, acoustic emission energy characteristics and temperature peak characteristics as inputs: each item is scaled to between zero and one according to the upper and lower reference limits to obtain five dimensionless deviation sub-indices, and then the five dimensionless deviation sub-indices are multiplied in a fixed order and the fifth root of the product is taken as the single value obtained as the defect index before geometric adjustment. The specific steps for defect cause coupling analysis using multi-source feature deviation data are as follows: Obtain the feature deviation components and the number of feature components; within each time slice, select the maximum value from the absolute values ​​of all feature deviation components as the multidimensional feature deviation maxima; square each feature deviation component, multiply it by the corresponding feature component adjustment coefficient, add a negative sign before the product, and then input it into the natural exponential function for calculation, adding one to the output result to obtain the intermediate value corresponding to each feature; multiply all intermediate values ​​sequentially to form a product, and then perform a geometric mean operation on the product according to the number of feature components to obtain the geometric mean term; negative sign the multidimensional feature deviation maxima before inputting it into the natural exponential function to obtain the deviation attenuation term; finally, multiply the deviation attenuation term by the geometric mean term to obtain the defect identification confidence value. The method of determining the type of defect and providing geometric adjustment guidance based on the defect cause analysis results includes: comparing the defect identification confidence value and the confidence threshold in real time; when the defect identification confidence value is less than the confidence threshold, reducing the feed rate and spindle speed, keeping the axial depth of cut and radial width of cut unchanged, extending the acoustic emission and vibration sampling time and increasing the surface morphology scanning frequency, not triggering tool geometry parameter reconstruction, and only archiving the current multi-source features and defect identification confidence value to the defect sample set; The specific steps for the comprehensive causal correlation measurement, geometric deviation characterization, and load characteristic evaluation of geometric driving strength are as follows: Obtain the defect identification confidence value, deviation scale, geometric deviation scale, load characteristic scale, and load scale; divide the deviation scale by the geometric deviation scale to obtain the geometric ratio term; perform a hyperbolic tangent function operation on the geometric ratio term to obtain the geometric saturation term; divide the load characteristic scale by the load scale to obtain the load ratio term; perform a hyperbolic tangent function operation on the load ratio term to obtain the load saturation term; multiply the defect identification confidence value by the geometric saturation term to obtain the coupled geometric term; multiply the coupled geometric term by the load saturation term to obtain the geometric driving response value. The implementation of load reduction conservative machining control and tool geometry reverse reconstruction includes: when the geometric drive response value is greater than or equal to the response threshold, inputting defect feature vectors, workpiece material property parameters, machining process parameter vectors, defect performance data from historical machining samples, working condition data, and tool geometry parameter data, constructing a multi-dimensional geometric control feature sequence covering the same machining time window; through sliding window feature extraction, multi-source feature fusion, and dynamic feature filtering, first calculating the basic geometric adjustment solutions for the rake angle, clearance angle, inclination angle, and cutting edge radius based on cutting mechanics mechanisms, friction behavior analysis, and material forming laws; and then using a data-driven solution with a multivariate regression and gradient enhancement integrated structure as the core. The solution framework performs incremental correction on the basic solution of geometric adjustment, forming a tool geometry inverse solution model. The tool geometry inverse solution model adjusts the feature weights and parameter configurations according to the disturbance amplitude of the defect scenario data, calculates and outputs a geometric adjustment vector composed of the rake angle increment, clearance angle increment, cutting edge inclination angle adjustment, and cutting edge radius adjustment, and recalculates the geometric drive response value. If it is less than the response threshold, the current geometric parameter configuration is maintained; if the geometric drive response value is still greater than or equal to the response threshold, the feature sequence is updated and the tool geometry inverse solution model is called again to output a new geometric adjustment vector, until the geometric drive response value falls back to less than the response threshold, the geometric parameter configuration is maintained, and the subsequent geometric parameter dynamic reconstruction module is entered. The specific steps for the collaborative evaluation of reconstruction benefits through the changes in defect characterization and adjustment intensity before and after structural reconstruction are as follows: Obtain the geometrically driven response value and the defect index before geometric adjustment; execute the defect index output process on the arithmetic mean roughness, texture frequency index, burr size index, acoustic emission energy characteristics, and temperature peak characteristics obtained in the new round of data collection, and record the obtained values ​​as the defect index after geometric adjustment; calculate the defect index before geometric adjustment minus the defect index after geometric adjustment to obtain the defect improvement difference; add the defect index before geometric adjustment to a minimum positive number to obtain a stable scale term; divide the defect improvement difference by the stable scale term to obtain an improvement ratio term; perform hyperbolic tangent function operation on the improvement ratio term to obtain an improvement saturation term; multiply the geometrically driven response value by the improvement saturation term to obtain a reconstruction evaluation value that comprehensively reflects the matching degree between the investment in this round of geometric reconstruction and the defect improvement effect; The specific steps for completing the closed-loop verification of backtracking correction, configuration solidification, and sample sedimentation are as follows: By comparing the reconstruction evaluation value and the evaluation threshold in real time, when the reconstruction evaluation value is less than the evaluation threshold, it is determined that the tooling reconstruction benefit corresponding to the geometric adjustment vector in the previous round is insufficient. The geometric reconstruction actuator restores the rake angle, clearance angle, blade tilt angle, and blade radius to the geometric adjustment vector combination recorded in the previous round, reduces the feed rate and spindle speed, and tightens the axial depth of cut and radial width of cut. At the same time, the on-machine multi-source detection module adds an acoustic emission, vibration, and surface morphology acquisition. The edge computing unit outputs the failed marker and carries the vibration peaks, acoustic emission energy, and surface roughness before and after reconstruction. The comparison results of the texture frequency and burr size are used to archive the current geometric adjustment amount, multi-source detection results, and reconstruction evaluation value as negative samples into the milling process database. These samples, along with the failed markers, the comparison results, and the updated defect feature vector, are sent back to the tool geometry reverse solution module as input for the next round of solution. When the reconstruction evaluation value is greater than or equal to the evaluation threshold, the current combination of rake angle, clearance angle, inclination angle, and cutting edge radius remains unchanged. The geometric adjustment vector of this round and the comparison results before and after reconstruction are used to form a data pair and archived into the milling process database. This data is used to update the correlation between subsequent defect identification and geometric reverse solution, thereby completing the verification and knowledge feedback loop.

2. The milling process self-optimization method based on in-machine inspection according to claim 1, characterized in that: The specific steps for completing the acquisition and normalization processing of in-machine multi-source signals and surface morphology are as follows: On-machine signal data acquisition: Spindle vibration acceleration timing is acquired via a vibration sensor within the multi-source detection module; the cutting force triaxial component timing is acquired via a triaxial force sensor in the tool holder area; acoustic emission count rate and envelope energy timing are acquired via an acoustic emission sensor; and the tool near-cutting zone temperature timing is acquired via a temperature sensor. Each measurement value is bound to the sampling time, and vibration signal sequences, cutting force signal sequences, acoustic emission signal sequences, and temperature signal sequences are established according to machining time slices. Surface topography data acquisition: During machining breaks, a vision unit and a confocal unit scan the toolpath coverage area to acquire arithmetic mean roughness, surface contour topography field, and spatial spectrum data. The dominant texture frequency and several auxiliary frequency components are identified from the spatial spectrum, and vertical burr components are extracted using image processing methods. The dimensions, burr width, and burr root thickness are recorded in the surface morphology record along with the arithmetic mean roughness, texture frequency index, and burr size index, along with the machining time slice identifier. Workpiece material property data is collected: elastic modulus, yield strength, hardness level, thermal conductivity, and coefficient of linear expansion are read from the material grade record, and the material property parameters, machining batch identifier, and time slice identifier are written into the material property record. Tool geometry and working condition data are collected: the current set values ​​of rake angle, clearance angle, inclination angle, and cutting edge radius, along with the corresponding geometric reference values, are read from the tool geometry parameter record table; the feed rate, spindle speed rate, axial depth of cut, and radial width of cut are obtained from the CNC device operation record; and the tool geometry parameters, geometric reference values, and machining working condition parameters, along with the time slice identifier, are written into the working condition record table.

3. The milling process self-optimization method based on in-machine inspection according to claim 1, characterized in that: The step of generating and storing characteristic deviation components, deviation scales, load characteristic scales, and defect indices before geometric adjustment to construct a milling process database further includes: The total number of characteristic deviation components is recorded as the number of characteristic components; the geometric deviation scale is read from the geometric parameter configuration table and written into the geometric deviation scale field to provide a reference level for the deviation scale in the geometrically driven evaluation; the load scale is read from the working condition parameter configuration table and written into the load scale field to provide a reference level for the load characteristic scale in the geometrically driven evaluation; after storing the standardized and normalized tool geometry parameters, workpiece material property parameters and machining process parameters, a milling process database is constructed.

4. The milling process self-optimization method based on in-machine inspection according to claim 1, characterized in that: The method of classifying defects and guiding geometric adjustments based on defect cause analysis results also includes: when the defect identification confidence value is greater than or equal to the confidence threshold, executing a cause-driven geometric optimization process: mapping the cause type to executable tool geometric adjustment actions; increasing the clearance angle when the cause points to increased back face friction; increasing the rake angle and decreasing the cutting edge radius when the cause points to expanded plastic deformation zone; adjusting the cutting edge inclination angle to change the chip flow direction and reduce burrs when the cause points to increased chatter excitation; and weakening the excitation source by coordinating the clearance angle and cutting edge inclination angle when the cause points to enhanced chatter excitation. After completing geometric reconstruction, a new round of surface scanning is triggered to adjust the arithmetic mean roughness and texture frequency. The changes in burr size, acoustic emission energy, and temperature peak are compared and the defect feature vector is updated. When the comparison results show that the roughness decreases, the texture frequency tends to stabilize, the burr size shrinks, the acoustic emission energy weakens, and the temperature peak drops, the current geometric parameter configuration is maintained and the subsequent processing flow continues. When the comparison results show that any index is still greater than the corresponding quality control threshold, the comprehensive defect index is sorted in descending order based on the change in comprehensive defect index caused by each geometric adjustment action. The one with the largest decrease in comprehensive defect index is selected and retained. The remaining geometric adjustment actions are rolled back and the reconstruction process is executed again. In m consecutive rounds of verification, the comprehensive defect index is always kept below the quality control upper limit.

5. The milling process self-optimization method based on in-machine inspection according to claim 1, characterized in that: The implementation of load reduction conservative machining control and tool geometry reverse reconstruction also includes: By comparing the geometric drive response value and the response threshold in real time, when the geometric drive response value is less than the response threshold, the feed rate and spindle speed are reduced. At the same time, the rake angle, clearance angle, inclination angle and cutting edge radius are constrained to maintain the current settings. Only the defect feature vector, workpiece material property parameters and machining process parameter vector are archived together with the geometric drive response value into the training dataset.

6. A milling process self-optimization system based on in-machine inspection, employing the milling process self-optimization method based on in-machine inspection as described in any one of claims 1-5, characterized in that... ,include: The in-machine multi-source detection module is used to complete the acquisition and normalization processing of in-machine multi-source signals and surface morphology, generate characteristic deviation components, deviation scales, load characteristic scales and defect indices before geometric adjustment, and store them to build a milling process database. The defect diagnosis and cause association module is used to perform defect cause coupling analysis through multi-source feature deviation data, and to realize defect classification and geometric adjustment guidance based on the defect cause analysis results; The tool geometry reverse optimization module is used to integrate causal correlation measurement, geometric deviation characterization and load characteristic evaluation of geometric drive strength, and realize the execution of load reduction conservative machining control and tool geometry reverse reconstruction. The dynamic reconstruction module for geometric parameters is used to collaboratively evaluate the changes in defect characterization and adjustment intensity before and after structural reconstruction, and to complete the closed-loop verification of rollback correction, configuration solidification and sample sedimentation.

Images