Numerical control gear processing machine tool health early warning method based on low-frequency sampling operation data

Abstract

The application discloses a numerical control gear processing machine tool health early warning method based on low-frequency sampling operation data, comprising: acquiring low-frequency sampling internal operation data of a numerical control gear processing machine tool, dividing candidate stable processing data segments according to program segments, tools and instruction / feedback changes, and evaluating the availability of the segments through five-dimensional scores of sample sufficiency, data integrity, state stability, instruction feedback consistency and boundary remoteness, and screening the segments meeting the availability threshold to enter subsequent analysis. The available data segments are classified into the same working condition data segment set according to the program segments, tools, rotating speed intervals and feeding intervals, and the same working condition health benchmark is established based on historical normal available segments; the health deviation of the current segment relative to the health benchmark is calculated, and the same working condition sequence deviation is accumulated in combination with the deviation reference value, the segment availability score and the deviation direction consistency coefficient, and the health early warning result is output according to the accumulated risk value.

Description

Technical Field

[0001] This invention belongs to the field of CNC gear machining machine tool condition monitoring and equipment health management technology, specifically a health early warning method for CNC gear machining machine tools based on low-frequency sampled operating data. This method can construct low-frequency sampled data segments based on the machine tool control status, gear machining condition information, spindle / workpiece / feed axis command feedback consistency, and operating response data under low sampling frequency conditions. Through data segment availability evaluation, comparison with health benchmarks under the same operating conditions, and cumulative deviations under the same operating conditions, it monitors and provides early warnings for health deviations during processes such as hobbing, shaping, grinding, shaving, honing, or gear composite machining. Background Technology

[0002] CNC gear machining tools are key equipment in the gear manufacturing process, widely used in hobbing, shaping, grinding, shaving, honing, and other gear machining applications. Gear parts typically have high requirements for tooth profile accuracy, tooth direction accuracy, tooth pitch accuracy, and tooth surface quality. Changes in the condition of factors such as machine tools, cutting tools, fixtures, workpiece blanks, cooling and lubrication, and machining processes can all affect the final machining quality. Compared with general machining processes, gear machining is characterized by strong repeatability of process cycles, a generating or synchronous motion relationship between the cutting tool and the workpiece, phased or periodic changes in cutting load, and frequent repetition of the same working conditions in batch processing. Therefore, it is suitable to use historical operating data under the same working conditions for health monitoring of the machining status.

[0003] Among existing CNC machine tool condition monitoring methods, some rely on external high-frequency sensor signals such as vibration, acoustic emission, high-frequency current, and cutting force, and extract fault features through spectrum analysis, time-frequency analysis, or machine learning models. While these methods have good characterization capabilities for high-frequency dynamic anomalies, long-term deployment in industrial settings typically requires additional sensors, acquisition equipment, and signal lines, resulting in high implementation costs. Furthermore, they may be limited by equipment modification conditions, interface permissions, site environment, maintenance costs, and the conditions for large-scale deployment.

[0004] In many CNC gear machining environments, companies can more easily and stably obtain internal machine tool operating data from CNC systems, PLCs, servo drives, spindle drives, workpiece axis drives, or industrial gateways. This data includes program number, program segment number, tool number, process stage, spindle commanded speed, actual spindle speed, workpiece axis commanded speed, actual workpiece axis speed, commanded feed rate, actual feed rate, electronic gearbox synchronization status, spindle load, spindle power, spindle current, feed axis load, feed axis current, position deviation, alarm status, and operating status words. This type of data is typically read periodically at a low sampling frequency, with sampling periods potentially reaching or approaching seconds. Because low-frequency sampling data struggles to reflect high-frequency information such as vibration and shock, meshing frequency, bearing characteristic frequency, or abrupt changes in acoustic emission, traditional fault diagnosis methods relying on high-frequency signals are difficult to directly apply to such low-frequency data scenarios.

[0005] Existing monitoring methods based on fixed thresholds typically directly determine whether spindle load, current, power, feed axis load, or position deviation exceeds preset alarm limits. While simple to implement, this method struggles to adapt to varying operating conditions during CNC gear machining. Significant differences in machine tool response can exist depending on program segments, cutting tools, gear models, machining stages, spindle speed, workpiece axis speed, feed rate, cutting allowance, material batches, and cooling / chip removal status. Using a uniform threshold across different operating conditions can easily misinterpret normal variations as abnormalities. Furthermore, under the same operating conditions, in the early stages of tool wear, changes in grinding wheel condition, deterioration of cutting load, abnormal cooling / chip removal, or slow shifts in feed follow-up, relevant variables may not yet exceed fixed alarm limits, but there may already be an increase in average values, increased fluctuations, a persistently high proportion, or a trend shift, leading to a risk of missed detections with the fixed threshold method.

[0006] Some existing methods incorporate tool, step, program segment, or process status information to segment, label, or perform correlation analysis on operational data. For example, some solutions acquire CNC system program segments, tool, speed, and feed information, set condition anchor points according to program segment switching, tool changing, and speed / feed changes, fragment the machine tool operation process, and perform real-time diagnosis and control at the edge nodes. These methods improve the systematic nature of diagnosis in multi-machine-tool, multi-process scenarios; however, they directly perform threshold judgment or model matching after segmentation, without considering the impact of quality differences between different data segments on subsequent health assessments, and lack a mechanism for judging the cumulative time-series deviations of multiple data segments under the same operating condition.

[0007] Some solutions construct multi-condition health benchmarks based on historical operating data, using hidden Markov models or Gaussian mixture models to model the normal operating state under different conditions. The health status is assessed by calculating the likelihood probability or deviation of the current state relative to the health benchmark. These methods achieve adaptive health assessment to some extent, but they typically use a single sampling point or a single data sample as the evaluation unit. Under low-frequency sampling conditions, the number of effective sampling points within a single processing segment is limited. The health deviation of a single data segment is easily affected by the sampling phase, periodic fluctuations in gear processing load, or local entry / exit disturbances, resulting in insufficient stability for single deviation judgments.

[0008] Another approach is to predict faults by calculating the deviation between monitoring indicators and health benchmarks and then accumulating this deviation over time. While this method has some predictive capability under varying load conditions, its load condition segmentation is rather coarse (e.g., dividing load bands by load index), making it unsuitable for gear machining scenarios requiring precise differentiation of program segments, tools, speed ranges, and feed ranges. Furthermore, this type of method typically includes all data segments in the deviation accumulation calculation, failing to consider the interference of low-quality segments on the accumulation results, and lacking an exemption mechanism for "normal fluctuation ranges." This can easily lead to the continuous accumulation of small deviations within historical normal fluctuation ranges, resulting in false alarms.

[0009] Furthermore, existing methods generally lack a systematic evaluation mechanism for the usability of data segments. Under low-frequency sampling conditions, the number of sampling points within a single machining segment is limited and may be affected by factors such as program segment switching, tool changing, spindle start / stop, workpiece axis synchronization adjustment, feed acceleration / deceleration, rapid traverse, data loss, and asynchrony between status words and response data. If the suitability of low-frequency data segments for health assessment is not determined, and all segments are directly used for health deviation calculation, health baseline update, or early warning triggering, it is easy to cause false alarms, missed alarms, or contamination of the health baseline by low-quality segments. Even if some solutions perform data cleaning, missing value imputation, or outlier removal during the data preprocessing stage, they do not treat "whether the data segment has health assessment value" as an independent evaluation dimension, nor do they couple the segment usability score with the cumulative deviation under the same working condition.

[0010] In summary, the existing technology for health warning of CNC gear processing machine tools under low-frequency sampling conditions has the following main shortcomings: (1) It lacks a systematic evaluation mechanism for the availability of low-frequency data segments, and cannot identify which segments are truly suitable for health evaluation; (2) The establishment and updating of the health benchmark under the same working condition lacks data quality control, and low-quality segments may contaminate the benchmark; (3) The stability of single segment health deviation judgment is insufficient, and it is easily affected by the sampling phase and periodic load fluctuations under low-frequency sampling conditions; (4) It lacks a deviation accumulation mechanism for the data segment sequence under the same working condition, and it is difficult to distinguish between continuous health deviation and single occasional disturbance; (5) The warning results lack gear processing working condition context (program segment, tool, process stage, speed / feed range, etc.), which is not conducive to maintenance personnel quickly locating the problem background.

[0011] Therefore, it is necessary to propose a health early warning method applicable to low-frequency sampling internal operating data of CNC gear processing machine tools. This method can screen low-frequency sampling data segments with health evaluation value without relying on external high-frequency sensors, and compare health deviations and make cumulative judgments of deviations under the same or similar gear processing conditions in chronological order. This will improve the stability, interpretability, and field deployment adaptability of health early warning under low-frequency data conditions. Summary of the Invention

[0012] In view of this, the purpose of this invention is to provide a health early warning method for CNC gear processing machine tools based on low-frequency sampling operation data, which aims to solve the problems of limited single-segment information, low-quality segments contaminating the benchmark, and occasional disturbances easily causing false alarms under low-frequency sampling conditions, and to achieve stable and reliable health early warning without relying on high-frequency sensors.

[0013] To achieve the above objectives, the present invention provides the following technical solution: A health early warning method for CNC gear machining tools based on low-frequency sampled operating data includes the following steps: S1: Acquire low-frequency sampling internal operating data of the CNC gear processing machine tool during the processing process, the low-frequency sampling internal operating data including control status data and health response data; S2: Based on the control status data, according to the changes in the command values ​​and actual feedback values ​​of the machine tool operating status, program segment number, tool number, spindle speed, workpiece axis speed and feed rate, candidate stable machining data segments are divided from the continuous low-frequency sampling data; S3: Calculate the fragment availability score for the candidate stable processing data fragment. The fragment availability score is used to evaluate whether the data fragment can be used as a basis for health evaluation, and is used for at least the following processes: determining whether the data fragment is allowed to enter the subsequent health deviation calculation, determining whether the data fragment is allowed to participate in the health benchmark update under the same working condition, and as a weighting factor in the cumulative deviation of the subsequent data fragment sequence under the same working condition. S4: Based on the fragment availability score, candidate stable processing data fragments that meet the availability threshold are determined as available data fragments, and candidate stable processing data fragments that do not meet the availability threshold are determined as low availability data fragments. S5: For the available data segments, construct a set of data segments under the same working conditions according to their corresponding gear processing conditions and machine tool operating status, and establish a health benchmark for each working condition based on historical normal available data segments. S6: Compare the health response characteristics of the currently available data segment with the corresponding health benchmark under the same working conditions, and calculate the health deviation of the currently available data segment; S7: Accumulate the deviation of multiple available data segments in the same working condition data segment sequence according to the time order, and calculate the cumulative risk value; S8: Output the health warning result of the CNC gear processing machine tool based on the accumulated risk value.

[0014] Furthermore, in step S1, the control status data includes one or more of the following: program number, program segment number, tool number, NC running status, PLC status word, spindle start / stop status, spindle commanded speed, spindle actual speed, workpiece axis commanded speed, workpiece axis actual speed, commanded feed rate, actual feed rate, spindle ratio, feed ratio, electronic gearbox synchronization status, tool life count, alarm status, pause status, tool change status, rapid traverse status, machining stage identifier, gear machining process type, coolant start / stop status, and lubrication status. The health response data includes one or more of the following: spindle load rate, spindle power, spindle current, workpiece axis load rate, feed axis load rate, feed axis current, position deviation, synchronization deviation, feed axis speed, machining cycle time, and segment duration.

[0015] Furthermore, in step S2, the method for dividing candidate stable processing data segments from continuous low-frequency sampling data is as follows: Based on the control status data, exclude sampling points for alarms, emergency stops, pauses, manual interventions, tool changes, rapid traverses, spindle start / stop, workpiece axis synchronization establishment processes, idling, or non-machining states; The boundaries of data segments are determined based on the changes in program segment number, tool number, machining stage, spindle speed command, workpiece axis speed command, and feed rate command, and the moment of change is used as the dividing position between adjacent data segments. For a data segment formed by consecutive sampling points between adjacent boundary positions, if the deviation between the actual spindle speed and the spindle command speed, the deviation between the actual workpiece spindle speed and the workpiece spindle command speed, and the deviation between the actual feed rate and the commanded feed rate are all within the preset allowable range, and the above deviations remain relatively stable within a number of consecutive sampling points, the data segment is determined as a candidate stable machining data segment.

[0016] Furthermore, in step S3, fragment usability scoring... Sufficiency of sample scoring Data integrity score State stability score Instruction feedback consistency score and boundary distance score Weighted average yields:

[0017] in: , , , and Let represent the weighting coefficients, where each weighting coefficient is non-negative and satisfies ... .

[0018] Furthermore, the sample sufficiency score Determined based on the relationship between the number of valid sampling points within a segment and the preset minimum number of valid sampling points:

[0019] in: This represents the number of valid sampling points within the segment; This is the preset minimum number of valid sampling points; The data integrity score Determined based on the proportion of missing or invalid sampling points within a segment to the total number of sampling points:

[0020] in: This represents the number of missing or invalid sampling points within the segment. This represents the total number of sampling points within the segment; The state stability score The determination is based on whether there are changes in the program segment number, tool number, NC running status, PLC status word, spindle start / stop status, electronic gearbox synchronization status, alarm status, or pause status within the segment:

[0021] in: The number of sampling points within a segment where a state change or exclusionary state marker occurs; The instruction feedback consistency score Determined based on the consistency between the actual feedback from the spindle, workpiece axis, and feed system and the corresponding command values:

[0022] in: The feedback of the instruction constitutes a comprehensive deviation; This is the upper limit of the allowable deviation; The boundary distance score Determined based on the time distance between candidate stable machining data segments and program segment switching, tool change, spindle start / stop, workpiece axis synchronization adjustment, rapid traverse, pause, alarm, or acceleration / deceleration state boundaries:

[0023] in: The time interval between the candidate stable processed data segment and the nearest state boundary; This is the preset minimum boundary interval.

[0024] Furthermore, in step S4, the availability threshold is determined based on the availability score distribution of candidate stable processing data segments in historically confirmed normal processing records: A batch of historically confirmed normal processing records were subjected to the same data segmentation and availability score calculation process as the online monitoring process to obtain a comprehensive availability score for multiple candidate stable processing data segments. ; The lowest quantile of the overall availability score is taken as the initial availability threshold. :

[0025] in: This represents the comprehensive availability score of candidate stable processing data segments within historically confirmed normal processing records. quantiles; This indicates the proportion of low-availability segments that are allowed to be removed from historical, normal, candidate, stable processing data segments; Based on the initial availability threshold, the final availability threshold is determined by adjusting for the dispersion of health response characteristics under the same operating conditions and the number of retained historical candidate stable processing data segments. .

[0026] Furthermore, by combining the dispersion of health response characteristics under the same operating conditions and the number of retained historical candidate stable processing data segments, a correction is made to determine the final availability threshold. The method is as follows: by As the initial threshold, select several data points with scores no lower than the overall availability score distribution of historically confirmed normal processing records. Candidate threshold ; For each candidate threshold , retain satisfaction Historical candidate stable processing data segments, and based on the retained segments, establish corresponding candidate health benchmarks and health response characteristic dispersion under the same working conditions; Among multiple candidate thresholds that meet the requirements of the dispersion of health response characteristics under the same working condition not exceeding the preset stability requirement and the number of retained historical candidate stable processing data segments not less than the preset minimum sample number, the candidate threshold with the smallest value is selected as the final availability threshold. :

[0027] in: Indicates at the candidate threshold The dispersion of the health response characteristics under the same working conditions established by retaining the lower segment; This indicates the upper limit of the allowed dispersion. Indicates at the candidate threshold The number of historical candidate stable processed data segments to retain; This indicates the minimum number of samples required to establish a health benchmark under the same working conditions.

[0028] Furthermore, in step S5, the method for constructing a set of data segments under the same working condition is as follows: When two or more available data segments meet the basic working conditions, they are grouped into the same set of data segments with the same working conditions. The basic working conditions include: the program number or machining task number is the same, the program segment number is the same or belongs to the same preset program segment range, the tool number is the same, the spindle command speed, the workpiece axis command speed and the command feed speed are the same or fall into the same preset range, and the corresponding sampling ranges are all in a stable machining state. When auxiliary working condition information can be obtained, the gear machining process type, gear machining process stage, gear model, workpiece type, material batch, blank allowance range, cooling state, spindle ratio, feed ratio, or cumulative tool machining time are used as auxiliary working condition labels to further subdivide or label the data segment set of the same working condition.

[0029] Furthermore, in step S5, the method for establishing a health benchmark under the same working conditions is as follows: For each set of data segments under the same working condition, health response features are extracted based on historical normal and usable data segments. The health response features include one or more of the following: average spindle load, average spindle load fluctuation, average spindle power, average spindle current, average workpiece axis load, average feed axis load, average feed axis current, average position deviation, average position deviation fluctuation, average synchronization deviation, average synchronization deviation fluctuation, segment duration, and machining cycle time. Calculate the baseline central value and baseline fluctuation scale of each health response characteristic in the historical normal available data segment as the health baseline under the same working condition:

[0030]

[0031] in: The reference center value; Used as the benchmark fluctuation scale; For the first The first historical data segment that is normally available Individual health response characteristics; This represents the number of historically available data segments. Data segments that are allowed to participate in health baseline updates must simultaneously meet the following conditions: the segment availability score is not lower than the preset availability threshold, the current comprehensive health deviation does not exceed the preset baseline update threshold, the effective deviation increment is zero or lower than the preset update allowable value, the cumulative risk value is at the normal level, and the data segment has not triggered a health warning and has not been manually confirmed as abnormal.

[0032] Furthermore, in step S6, the method for calculating the health deviation of the currently available data segment is as follows: Let the currently available data segment be the first... Each health response characteristic is Its corresponding center value in the same working condition health benchmark is The fluctuation scale is Then the normalized deviation of this feature is:

[0033] in: Indicates the first Normalized deviation of each health response feature; To prevent positive numbers with a denominator of zero; Alternatively, for features that increase with increased cutting load or deteriorated machining conditions, a one-sided deviation calculation method can be used:

[0034] The overall health deviation is obtained by weighting the data according to the importance of each health response feature:

[0035] in: This represents the overall health deviation of the currently available data segment; This indicates the number of health response features involved in the calculation; Indicates the first The weight coefficients of each health response feature are non-negative and satisfy the following conditions: ; Record one or more health response features that contribute the most as the primary deviation variables.

[0036] Furthermore, in step S7, the method for calculating the cumulative risk value is as follows: Multiple available data segments from the same set of data segments under the same working condition are arranged in chronological order according to the start time or end time of the segments to form a sequence of data segments under the same working condition. Set deviation from reference value The deviation reference value represents the healthy deviation boundary corresponding to the normal fluctuation range under the same working conditions; Current health deviation Not exceeding the deviation from the reference value When, no new risk increment is generated; when Exceed When the fluctuation exceeds the normal range, the portion exceeding the normal fluctuation range is considered as the effective deviation increment:

[0037] in: Indicates the first The effective deviation increment of each available data segment; Constructing a cumulative risk value based on the effective deviation increment:

[0038] in: Indicates the first data segment in the same working condition sequence. The cumulative risk value after processing each available data segment; This represents the cumulative risk value after processing the previous available data segment; This represents the historical risk decay coefficient, with a value ranging from 0 to 1. This indicates the availability score of the current data segment; This represents the coefficient of consistency in deviation direction; This represents the effective deviation increment of the current data segment.

[0039] Furthermore, in step S8, the method for outputting the health warning result of the CNC gear processing machine tool is as follows: Multiple risk thresholds are set based on the cumulative risk value distribution of data segment sequences under the same working conditions in historically confirmed normal processing records; When the cumulative risk value is below the first risk threshold, a normal status is output; when the cumulative risk value reaches the first risk threshold but is below the second risk threshold, a slight deviation warning is output; when the cumulative risk value reaches the second risk threshold but is below the third risk threshold, a continuous deviation warning is output; when the cumulative risk value reaches the third risk threshold, a high-risk deviation warning is output. The health warning results include the health deviation level, cumulative risk value, corresponding set of data segments under the same working conditions, working condition label, main deviation variables, and maintenance and inspection prompts. The working condition label includes one or more of the following: program segment number, tool number, gear machining stage, spindle speed range, workpiece axis speed range, feed rate range, gear model, material batch, and blank allowance range. The maintenance check prompts determine the inspection direction based on the main deviation variables: if the main deviation variables are concentrated in the spindle load, spindle power, or spindle current, the prompts indicate to check the tool wear condition, grinding wheel condition, cutting parameters, blank allowance, cooling and chip removal condition, or machining condition; if the main deviation variables are concentrated in the workpiece axis load, feed axis load, feed axis current, position deviation, or synchronization deviation, the prompts indicate to check the workpiece axis or feed axis running condition, electronic gearbox synchronization condition, trajectory following condition, fixture clamping condition, cutting load, or process parameters.

[0040] Furthermore, it also includes summarizing and analyzing the early warning results of multiple sets of data segments under the same working conditions: If multiple sets of data segments under the same working condition with the same tool number but corresponding to different program segments or different cutting parameters all show health deviations, and the main deviation variables are concentrated in spindle load, spindle power or spindle current, then the maintenance and inspection priority corresponding to that tool should be increased. If multiple sets of data segments with the same program segment, the same gear machining stage, or the same gear model but corresponding to different cutting tools all show health deviations, then the priority of checking the process parameters, blank allowance, material condition, or cooling and chip removal status of that machining stage should be increased.

[0041] The beneficial effects of this invention are as follows: This invention relates to a health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data. Through a technical framework of segment availability evaluation, establishment of benchmarks for the same working condition, and accumulation of sequence deviations, it achieves stable early warning of the health status of CNC gear processing machine tools using only low-frequency sampled internal operating data, and has achieved the following technical effects.

[0042] First, this invention employs a five-dimensional segment availability scoring mechanism (sample sufficiency, data integrity, state stability, instruction feedback consistency, and boundary deviation) to effectively identify and exclude low-quality data segments caused by acceleration / deceleration, tool changes, missing data, unstable instruction feedback, etc., thus preventing them from entering the health baseline update and early warning judgment process. This mechanism also uses the availability score as a weighting factor for deviation accumulation, allowing high-quality segments to play a greater role in risk assessment, thereby significantly reducing the risk of false alarms and baseline contamination caused by inconsistent data quality under low-frequency sampling conditions.

[0043] Secondly, by constructing a set of data segments under the same working conditions, this invention accurately compares the current data segment with historical normal and usable segments under the same program segment, tool, speed range, and feed range. This eliminates the interference of normal response differences between different machining conditions on health judgment, so that health warnings truly focus on state deviations under the same working conditions, rather than normal fluctuations caused by changes in working conditions.

[0044] Furthermore, the present invention deviates from the reference value. The settings ensure that health deviations within the historical normal fluctuation range do not generate risk increments; this is achieved through the historical attenuation coefficient λ in the cumulative deviation of the same operating condition sequence and the segment availability score. Consistency coefficient of deviation direction The triple weighting allows persistent health deviations to gradually accumulate and strengthen, while single, occasional disturbances are unlikely to trigger high-level warnings. This mechanism effectively mitigates the shortcomings of limited single-segment information and susceptibility to occasional disturbances under low-frequency sampling conditions.

[0045] Finally, the early warning results output by this invention include health deviation level, cumulative risk value, working condition label (program segment number, tool number, speed range, feed range, etc.) and main deviation variables. Maintenance personnel can quickly locate the machining background and specific manifestations of the health deviation based on this, providing a clear direction for targeted inspections (tool wear, cutting parameters, cooling and chip removal, feed / synchronization status, etc.), significantly improving the practicality and interpretability of the early warning results.

[0046] In summary, the health early warning method for CNC gear processing machine tools based on low-frequency sampling operation data of this invention achieves early identification and stable early warning of health risks such as tool wear, abnormal cutting load, and deviation of feed follow-up status during gear processing without increasing hardware costs. It has good adaptability to industrial field deployment and promotion value. Attached Figure Description

[0047] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1This is a flowchart of the health early warning method for CNC gear processing machine tools based on low-frequency sampling operation data according to the present invention. Detailed Implementation

[0048] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0049] This invention aims to propose a health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data. This invention addresses the problems of limited sample points in a single data segment under low-frequency sampling conditions, significant impact of gear machining condition switching, insufficient stability in single-segment anomaly judgment, and susceptibility of health benchmarks to contamination by low-quality segments. It acquires internal operating data from low-frequency sampling of CNC gear machining machines and, based on machine tool operating status, program segments, cutting tools, spindle speed, workpiece axis speed, feed rate, and changes in command and actual feedback values, divides candidate stable machining data segments from continuous low-frequency sampling data. Availability evaluation is performed on these candidate stable machining data segments, and only those segments meeting availability requirements are used for subsequent health analysis. Available data segments are categorized into sets of data segments under the same working condition according to gear machining conditions. Based on historical normal machining records obtained through segmentation, availability evaluation, and same-condition classification, a same-condition health benchmark is established. The health deviation of the current available data segment relative to the same-condition health benchmark is calculated, and the effective deviation is accumulated in chronological order for the same-condition data segment sequence. Finally, based on the accumulated risk value, the health deviation level, working condition label, main deviation variables, and maintenance check prompts are output.

[0050] In this invention, the fragment availability score is used not only to determine whether a data fragment is allowed to enter the health deviation calculation, but also to restrict low-quality fragments from participating in the same-condition health benchmark update, and as a weighting factor in the same-condition deviation accumulation to adjust the contribution of the current fragment to the cumulative risk value. By coupling the data fragment availability evaluation with the same-condition health deviation accumulation, false alarms and benchmark contamination caused by state switching, insufficient sample points, unstable command feedback, or missing data under low-frequency sampling conditions can be reduced.

[0051] Specifically, such as Figure 1 As shown in the figure, the health early warning method for CNC gear processing machine tools based on low-frequency sampling operation data in this embodiment includes the following steps.

[0052] S1: Acquire low-frequency sampling internal operating data of the CNC gear processing machine tool during the processing process. The low-frequency sampling internal operating data includes control status data and health response data.

[0053] In this embodiment, low-frequency sampling internal operating data refers to machine tool control status data and health response data that are periodically read from the CNC system, PLC, servo driver, spindle driver, workpiece axis driver, electronic gearbox, industrial gateway, or data acquisition module at a sampling period of seconds or close to seconds.

[0054] The control status data describes the current operating status of the machine tool, the gear machining stage, and the process conditions. It includes one or more of the following: program number, program segment number, tool number, NC running status, PLC status word, spindle start / stop status, spindle commanded speed, actual spindle speed, workpiece axis commanded speed, workpiece axis actual speed, commanded feed rate, actual feed rate, spindle ratio, feed ratio, electronic gearbox synchronization status, tool life count, alarm status, pause status, tool change status, rapid traverse status, machining stage identifier, gear machining process type, coolant start / stop status, and lubrication status.

[0055] Health response data is used to characterize the operational response of a CNC gear machining tool under the current machining state, including one or more of the following: spindle load rate, spindle power, spindle current, workpiece axis load rate, feed axis load rate, feed axis current, position deviation, synchronization deviation, feed axis speed, machining cycle time, and segment duration. Based on health response data, segment characteristics such as mean, fluctuation value, rate of change, proportion of persistently high values, machining cycle time offset, or command feedback deviation statistics can also be calculated in subsequent data segments.

[0056] Where available, workpiece type, gear model, module, number of teeth, helix angle, tooth width, machining task number, production batch, material grade, material batch, heat treatment status, blank allowance, blank batch, cooling status, fixture status, tool specifications, cumulative cutting time of the tool, and number of processed parts can be used as auxiliary working condition labels to improve the accuracy of subsequent classification of the same working condition.

[0057] After data acquisition, the original low-frequency sampled internal operating data can be sorted by time, aligned by fields, and marked with missing values ​​and outliers. For data from different sources, such as CNC system data, PLC status data, spindle drive data, and servo drive data, alignment can be performed based on the sampling timestamp or sampling sequence number, so that data records at the same sampling time contain corresponding control status data and health response data.

[0058] S2: Based on the control status data, according to the changes in the command values ​​and actual feedback values ​​of the machine tool operating status, program segment number, tool number, spindle speed, workpiece axis speed and feed rate, candidate stable machining data segments are divided from the continuous low-frequency sampling data.

[0059] In this embodiment, the method for dividing candidate stable processing data segments from continuous low-frequency sampling data is as follows.

[0060] Based on the control status data, sampling points in states such as alarm, emergency stop, pause, manual intervention, tool change, rapid traverse, spindle start / stop, workpiece axis synchronization establishment, idling, or non-machining are excluded. Specifically, sampling intervals that are clearly unsuitable for stable machining health evaluation are excluded based on the control status data. If a sampling point corresponds to a machine tool in an alarm, emergency stop, pause, manual intervention, tool change, rapid traverse, spindle start / stop, workpiece axis synchronization establishment, idling, or non-machining state, then that sampling point will not participate in the construction of candidate stable machining data segments.

[0061] Data segment boundaries are determined based on changes in program segment number, tool number, machining stage, spindle speed command, workpiece axis speed command, and feed rate command. The moments of change are used as the boundaries between adjacent data segments. Specifically, after excluding the aforementioned sampling points, data segment boundaries are determined based on changes in program segment number, tool number, spindle speed command, workpiece axis speed command, and feed rate command. When the program segment number changes, the tool number changes, the machining stage changes, the spindle speed command changes significantly, the workpiece axis speed command changes significantly, or the feed rate command changes significantly, the corresponding moment of change is used as the boundary between adjacent data segments.

[0062] For data segments formed by consecutive sampling points between adjacent boundary positions, it is further determined whether they meet the stable machining conditions. If the data segment satisfies the following: the deviation between the actual spindle speed and the commanded spindle speed, the deviation between the actual workpiece axis speed and the commanded workpiece axis speed, and the deviation between the actual feed rate and the commanded feed rate are all within the preset allowable range, and the above deviations remain relatively stable within a series of consecutive sampling points, then the data segment is determined as a candidate stable machining data segment. For CNC gear machining machines with electronic gearboxes or multi-axis synchronous control, it is also possible to further determine whether the synchronization state or linkage deviation between the tool spindle, workpiece axis, and feed axis meets the preset requirements.

[0063] For example, when the commanded speed of the tool spindle or workpiece axis changes during gear hobbing, if the actual speed is still gradually approaching the commanded speed, or the synchronization state of the electronic gearbox is not yet stable, the corresponding sampling point should not be used as stable machining data. Only when the actual feedback of the spindle, workpiece axis and feed axis enters the allowable range and remains relatively stable within a number of consecutive sampling points can subsequent consecutive sampling points be included in the candidate stable machining data segment.

[0064] Each candidate stable machining data segment should include at least the segment start time, segment end time, number of sampling points, program number or program segment number, tool number, spindle speed range, workpiece axis speed range, feed rate range, machine tool operating status indicator, and corresponding health response data sequence. When obtainable through machining tasks, process documents, program segment mapping, or status identification, auxiliary operating condition information such as gear machining process type, gear machining process stage, workpiece model, material batch, and blank allowance range may also be included. To avoid statistical instability caused by excessively short segments, a minimum number of sampling points or a minimum duration requirement can be set.

[0065] S3: Calculate the availability score of the candidate stable processing data segment. The availability score is used to evaluate whether the data segment can be used as a basis for health assessment, and is used for at least the following processes: First, determine whether the data segment is allowed to enter the subsequent health deviation calculation; Second, determine whether the data segment is allowed to participate in the health benchmark update under the same working condition; Third, as a weighting factor in the cumulative deviation of the subsequent data segment sequence under the same working condition, it is used to adjust the contribution of the health deviation of the data segment to the cumulative risk value.

[0066] The availability evaluation is based at least on the number of samples of sampling points within the candidate stable machining data segment, data integrity, machine tool control state stability, consistency between spindle / workpiece axis / feed axis command values ​​and actual feedback values, and the time distance between the data segment and state boundaries such as program segment switching, tool change, spindle start / stop, workpiece axis synchronization adjustment, rapid movement, alarm, pause or obvious acceleration / deceleration.

[0067] In this embodiment, fragment availability scoring Sufficiency of sample scoring Data integrity score State stability score Instruction feedback consistency score and boundary distance score Weighted average yields:

[0068] in: , , , and Let represent the weighting coefficients, where each weighting coefficient is non-negative and satisfies ... Each weighting coefficient can be set based on the monitored object, the completeness of data fields, and on-site experience. When prior experience is lacking, each weighting coefficient can be equally weighted, for example, all can be set to 0.2. If a certain type of data source is missing a lot, or if a certain type of state boundary or linkage deviation has a more significant impact on the operational response during gear processing, the weight of the corresponding sub-score can be increased.

[0069] Each sub-score can be calculated using a normalized method, with the value range limited to between 0 and 1; the closer the value is to 1, the more suitable the corresponding evaluation item is for health status evaluation. The following formula is only one optional implementation method. In practical applications, the calculation method or threshold parameters of each scoring item can be adjusted according to the machine tool type, CNC system data fields, and on-site process requirements.

[0070] (1) Sample adequacy score .

[0071] Sample adequacy score The number of valid sampling points within a candidate stable processing data segment is used to characterize whether it meets the health assessment requirements. This number can be determined based on the relationship between the number of valid sampling points within the segment and the preset minimum number of valid sampling points.

[0072] in: This represents the number of valid sampling points within the segment; This is the preset minimum number of valid sampling points. It can be determined based on the sampling period and the minimum duration of the stable processing segment. For example, when the sampling period is 5 seconds and it is desired that the segment covers a stable processing process of no less than 30 seconds, Option 6 is acceptable.

[0073] (2) Data integrity score .

[0074] Data integrity score The criteria for characterizing missing or invalid records within candidate stable processing data segments can be determined based on the proportion of missing sampling points, invalid fields, or anomaly markers to the total number of sampling points within the segment.

[0075] in: This represents the number of missing or invalid sampling points within the segment. This represents the total number of sampling points within the segment.

[0076] (3) State stability score .

[0077] State stability score To characterize whether the machine tool control state remains consistent within a candidate stable machining data segment, it can be determined based on whether there are changes in the program segment number, tool number, NC running status, PLC status word, spindle start / stop status, electronic gearbox synchronization status, alarm status, pause status, etc. within the segment:

[0078] in: This represents the number of sampling points within a segment where a state change or exclusionary state marker occurs.

[0079] (4) Instruction feedback consistency score .

[0080] Instruction feedback consistency score This score is used to characterize whether the actual feedback from the spindle, workpiece axis, and feed system is consistent with the corresponding command values. For CNC gear machining tools, the coordination between the spindle, workpiece axis, and feed axis affects the gear surface machining quality and machining stability. Therefore, this score is used not only for data quality evaluation but also to identify whether a segment is in a stable machining state. In this embodiment, the command feedback consistency score... Determined based on the consistency between the actual feedback from the spindle, workpiece axis, and feed system and the corresponding command values:

[0081] in: To provide comprehensive feedback on the instruction deviation, This represents the upper limit of the allowable deviation. It can be obtained by weighting the relative deviation between the actual spindle speed and the spindle commanded speed, the relative deviation between the actual workpiece spindle speed and the workpiece spindle commanded speed, the relative deviation between the actual feed rate and the commanded feed rate, and the available electronic gearbox synchronization deviation. Based on the allowable speed following error range of the CNC system, servo driver, or spindle driver, the spindle and workpiece axis speed stability accuracy requirements, the electronic gearbox synchronization accuracy requirements, and the comprehensive deviation of command feedback in historically confirmed normal and stable machining segments, the following parameters can be considered. The high quantile boundary is determined.

[0082] Specifically, in this embodiment, the command feedback deviations of the spindle, workpiece axis, and feed system are first uniformly converted into dimensionless relative deviations. Let the commanded spindle speed and the actual speed be respectively... and The commanded rotational speed and the actual rotational speed of the workpiece axis are respectively and The commanded feed rate and the actual feed rate are respectively and The relative deviation of the spindle speed Relative deviation of workpiece shaft speed Relative deviation of feed rate Calculate according to the following formulas:

[0083]

[0084]

[0085] in, To prevent a positive number with a zero denominator due to a zero command value, when the electronic gearbox synchronization deviation can be obtained, let the current electronic gearbox synchronization deviation be... The upper limit of the allowable synchronization deviation given by the equipment or CNC system is Then the synchronization deviation of the electronic gearbox is made dimensionless. :

[0086] After obtaining the above dimensionless deviation, the instruction feedback comprehensive deviation is obtained. Calculated using a weighted method:

[0087]

[0088] in, , , and is a non-negative weighting coefficient. When prior weights are lacking, equal weights are applied to the deviation terms that can actually be obtained; when the synchronization deviation of the electronic gearbox cannot be obtained, let Then, the remaining weights are renormalized so that their sum is 1. Through the above processing, each deviation term is a dimensionless quantity and can be directly weighted and combined.

[0089] Upper limit of allowable deviation The preferred method is to determine the data from historically confirmed normal and stable processing segments. Specifically, the command feedback comprehensive deviation is calculated for historically confirmed normal and stable processing segments using the same method described above, forming a historical comprehensive deviation sample set. And take its 95th percentile as the upper limit of the allowable deviation:

[0090] If the CNC system, spindle driver, servo driver, or electronic gearbox provides the upper limit of the individual allowable deviation for spindle speed, workpiece axis speed, feed rate, or electronic gearbox synchronization deviation, then the upper limit of each individual allowable deviation can be first converted to the value corresponding to the maximum allowable deviation of the spindle speed, workpiece axis speed, feed rate, or electronic gearbox synchronization deviation. , , and The same dimensionless expression form, and adopts the integrated deviation with instruction feedback. By weighting the data with the same weighting coefficients, we obtain the upper limit of the overall allowable deviation for the equipment. And take the 95th percentile of the historical normal composite deviation. and The smaller one as This is to avoid availability determinations exceeding the limits allowed by the equipment or process.

[0091] (5) Boundary distance score .

[0092] Boundary distance score This is used to characterize the time distance between candidate stable machining data segments and state boundaries such as program segment switching, tool change, spindle start / stop, workpiece axis synchronization adjustment, rapid traverse, pause, alarm, or significant acceleration / deceleration. If the segment is far from the above state boundaries, then... Higher; if the fragment is adjacent to the above state boundary, then Reduced. In this embodiment, the boundary distance score is... Determined based on the time distance between candidate stable machining data segments and program segment switching, tool change, spindle start / stop, workpiece axis synchronization adjustment, rapid traverse, pause, alarm, or acceleration / deceleration state boundaries:

[0093] in: The time interval between the candidate stable processed data segment and the nearest state boundary; This is the preset minimum boundary interval. It can be determined based on the stable recovery time of the gear processing machine tool after state switching. Specifically, it can be set by combining the machine tool acceleration / deceleration transition time, the electronic gearbox synchronization establishment time, the response stabilization time after program segment switching, and the time required for each feedback quantity to recover to a stable range after spindle start / stop, workpiece axis synchronization adjustment, or feed speed change.

[0094] Specifically, the stable recovery time is calculated for each state boundary event in the historical confirmed normal processing record. Let the... The occurrence times of state boundaries such as sub-program segment switching, spindle start / stop, workpiece axis synchronization adjustment, or feed rate changes are: Searching backwards from that moment, when the relative deviations of the spindle speed, workpiece spindle speed, and feed rate fall within their respective allowable ranges, and the synchronization state returns to normal when the electronic gearbox synchronization state can be acquired, and when at least two consecutive sampling points meet the above conditions, the first sampling moment that meets the conditions is recorded as... . No. The steady-state recovery time corresponding to the sub-state boundary for:

[0095] The allowable ranges for the relative deviations of each component of the spindle, workpiece axis, and feed system can be taken as the 95th percentile boundary of historically confirmed normal and stable machining segments. If the CNC system or driver specifies a stricter upper limit for allowable deviations, the stricter limit should be adopted. The 95th percentile of the stable recovery time for multiple stable boundary conditions of the same type is taken as the preset minimum boundary interval corresponding to that type of boundary condition. :

[0096] For different types of state boundaries, the corresponding calculations can be performed separately. Furthermore, during online segment evaluation, the corresponding value is selected based on the type of the nearest state boundary; if a uniform minimum boundary interval is used, then the corresponding values ​​for each type of state boundary can be taken. The maximum value in. Therefore, It can be calculated from the actual stable recovery process in historically confirmed normal processing records.

[0097] In a preferred embodiment of this example, in addition to based on the comprehensive usability score In addition to determining the usability of candidate stable processing data segments, lower limits can be set for one or more of the following: data integrity score, state stability score, instruction feedback consistency score, and boundary distance score. The overall usability score of a candidate stable processing data segment... Even if a data segment reaches the availability threshold, but its key sub-score is below the corresponding lower limit, it can still be marked as a low-availability data segment. This process avoids data segments that are unstable due to weighted compensation, have unstable instruction feedback, or are close to the state transition boundary being mistakenly identified as available data segments.

[0098] S4: Based on the fragment availability score, candidate stable processing data fragments that meet the availability threshold are determined as available data fragments, and candidate stable processing data fragments that do not meet the availability threshold are determined as low availability data fragments.

[0099] Specifically, based on the availability score of the candidate stable processing data segments obtained in step S3, the candidate stable processing data segments are screened. Let the overall availability score of the candidate stable processing data segments be... The availability threshold is .when When, the candidate stable processing data segment is determined as a usable data segment; when When this happens, the candidate stable processing data segment is identified as a low-availability data segment.

[0100] In this embodiment, the availability threshold The preferred method is to determine the availability score distribution of candidate stable machining data segments from historically confirmed normal machining records. The historically confirmed normal machining records are consistent with the historical normal machining records used to establish the same working condition health benchmark. They refer to continuous low-frequency sampling data collected within a time range when the machine tool did not experience alarms, was not manually confirmed as abnormal, the gear machining quality or production records did not show any abnormalities, and the tool was in the new tool stable machining stage or normal use stage.

[0101] Specifically, a batch of historically confirmed normal processing records undergoes the same data segmentation and availability score calculation process as the online monitoring process to obtain a comprehensive availability score for multiple candidate stable processing data segments. The above comprehensive availability scores are sorted by numerical value, and the lowest quantile is used as the initial availability threshold. ,Right now:

[0102] in: This represents the comprehensive availability score of candidate stable processing data segments within historically confirmed normal processing records. quantiles; This is used to indicate the proportion of low-availability segments that can be removed from historically normal candidate stable processing data segments. For example, when you want to remove the lowest 10%, 15%, or 20% of segments with the lowest availability scores from historically normal candidate stable processing data segments, you can take [the appropriate percentage]. , or In this way, the availability threshold It can be determined by the availability score distribution of historical normal processing data, rather than relying entirely on human experience.

[0103] In an optional implementation of this embodiment, the initial availability threshold can also be determined by combining the stability of the health baseline under the same operating conditions. Perform an upward adjustment. Specifically, with As the initial threshold, select several data points with scores no lower than the overall availability score distribution of historically confirmed normal processing records. Candidate threshold For example, usability scores corresponding to 15%, 20%, 25%, 30%, 40%, 50%, or 60% quantiles can be selected as candidate thresholds. For each candidate threshold... , retain satisfaction Historical candidate stable processing data segments are used to establish corresponding candidate health benchmarks and health response characteristic dispersion based on the retained segments.

[0104] If at a certain candidate threshold If, under the same operating conditions, the dispersion of the health response characteristics meets the preset stability requirement, and the number of retained historical candidate stable processing data segments is not less than the preset minimum sample number, then the candidate threshold can be used as the corrected availability threshold. Preferably, among multiple candidate thresholds that meet the above conditions, the candidate threshold with the smallest value is selected as the final availability threshold. ,Right now:

[0105] in: Indicates at the candidate threshold The dispersion of the health response characteristics under the same working conditions established by retaining the lower segment; This indicates the upper limit of the allowed dispersion. Indicates at the candidate threshold The number of historical candidate stable processed data segments to retain; This indicates the minimum number of samples required to establish a health benchmark under the same working conditions.

[0106] To ensure that the minimum sample size and the dispersion of health response characteristics under the same working conditions have a clear and uniform calculation scale, an initial availability threshold is used. Once determined, for each set of data segments under the same working condition, select those that meet the following criteria. Historically confirmed normal candidate stable processing data segments, and extracted data from each segment that participated in the establishment of health benchmarks. A health response characteristic. Let the first... The first historical segment Each health response characteristic is Then, firstly, based on the initial historical segment of the same working condition, the first... The distribution of each health response feature is calculated using a fixed reference scale. :

[0107] in, and These represent the 95th and 5th percentiles of the same health response characteristic in the aforementioned historical segments, respectively. This represents the reference scale of change of the health response feature under historical normal conditions, and is applied to different candidate thresholds. It remains unchanged during the comparison process. When When the result is 0, add a positive number to the denominator. To avoid division by zero.

[0108] Minimum number of samples The preferred method is to determine the accuracy based on the estimation of the health baseline center value. Let the first historical segment mentioned above be the... The sample standard deviation of each health response feature is The critical value of the standard normal distribution corresponding to the selected confidence level is... The allowable health benchmark center estimation error accounts for a percentage of the reference scale. The proportion is Then the first Minimum number of samples corresponding to each health response feature It can be calculated using the following formula:

[0109] in, Determined by the selected confidence level; in a preferred embodiment, a 95% confidence level is used, corresponding to . This represents the proportion of the allowable health baseline center estimation error relative to the historical normal reference scale, and can be selected from 0.1 to 0.2 depending on the required baseline estimation accuracy. In one implementation, it is taken as... Health response characteristics for all participants in establishing health benchmarks were calculated separately. The maximum value among them is taken as the minimum sample size for the data segment set under the same working condition. ,at the same time Not less than 2, to ensure that the benchmark fluctuation scale can be calculated:

[0110] If the number of historically confirmed normal segments currently available in a certain set of data segments under the same operating condition is less than the calculated number... If the health baseline for the same working condition is not established or updated for the time being, historical normal data can continue to be accumulated; alternatively, basic same working condition classification can be appropriately adopted according to the aforementioned hierarchical same working condition classification method to increase the number of historical samples for the same working condition.

[0111] For any candidate threshold Suppose that satisfies The number of historically confirmed normal candidate stable processing data segments is For the retained fragment, the first... Calculate the cross-fractional standard deviation for each health response feature. and using a fixed reference scale Normalize it to obtain the first Normalized dispersion of each health response feature :

[0112] Candidate thresholds are obtained by weighting the normalized dispersion of each health response feature. Comprehensive dispersion under :

[0113] in, For the first The dispersion weights of each health response feature are assigned, with each weight being a non-negative number and the sum of the weights being 1; equal weights can be used when prior weights are lacking. Since... In all candidate thresholds The bottom remains unchanged. It can be used to compare the relative stability of the health benchmark under the same working conditions after screening different candidate thresholds.

[0114] Upper limit of allowable overall dispersion The preferred method is determined by the normal dispersion distribution in historically confirmed normal processing records. Specifically, it will satisfy... The historical confirmed normal segments under the same working conditions are arranged in chronological order, and their number is assumed to be . ,by Using the sliding window length as the sliding step size and one segment as the sliding step, consecutive historical segment windows are formed sequentially. For the first... A window, according to The same calculation method is used to obtain the historical normal comprehensive dispersion. ;when At that time, it can form Each window is used as a reference, and the 95th percentile of the historical normal overall dispersion of each window is taken as the reference value. :

[0115] Therefore, for each candidate threshold Simultaneously calculate and Only when and Only when the candidate threshold meets both the health baseline stability and minimum sample size requirements is the candidate threshold considered to satisfy both the requirements. The candidate threshold with the smallest value among those meeting the conditions is then selected as the final usability threshold. This allows for the retention of as many historical normal data fragments as possible while ensuring stable health baselines and sufficient sample size.

[0116] Using the above method, the availability threshold It can reflect the availability score distribution of candidate stable processing data segments in historical normal processing data, and can avoid the threshold being too low, causing low-quality segments to enter the healthy baseline, or the threshold being too high, resulting in an insufficient number of available segments.

[0117] Available data segments are used for subsequent comparisons with health benchmarks under the same operating conditions, calculation of health deviations, and accumulation of deviations in data segment sequences under the same operating conditions; their availability scores are also used to adjust the weight of the data segment in the cumulative risk calculation. Low availability data segments are not used to trigger formal health warnings, nor are they used to update health benchmarks under the same operating conditions; when it is necessary to trace the operation process, low availability data segments and their causes of low availability can be stored.

[0118] Specifically, low availability may be caused by one or more of the following: insufficient number of valid sampling points, high data loss rate, changes in machine tool control status, large deviations between actual and commanded values ​​of spindle speed, workpiece axis speed, or feed rate, unstable electronic gearbox synchronization, or the segment being too close to the state boundary such as program segment switching or tool switching. By addressing these issues, it is possible to avoid directly using data segments that lack health assessment value for health deviation calculations or health baseline updates.

[0119] It should be noted that identifying a candidate stable processing data segment as a usable data segment only indicates that the data segment meets the basic data quality requirements for inclusion in the health deviation calculation; it does not necessarily mean that the data segment can participate in the health baseline update under the same working conditions. Whether participation in the health baseline update is allowed requires further judgment based on the data segment's health deviation, cumulative risk value, warning level, manual confirmation results, or processing quality records.

[0120] S5: For the available data segments, construct a set of data segments under the same working conditions according to their corresponding gear processing conditions and machine tool operating status, and establish a health benchmark for each working condition based on historical normal available data segments.

[0121] In this embodiment, the available data segments obtained in step S4 are used to construct a set of data segments under the same working conditions based on their corresponding gear processing conditions and machine tool operating states. A data segment refers to a piece of data obtained by dividing continuous low-frequency sampling data according to time continuity, machine tool state changes, and working condition parameter changes; a set of data segments under the same working conditions refers to a data set formed by merging multiple available data segments generated at different times but with the same gear processing working conditions or within the same preset working condition range.

[0122] In one embodiment of this example, when two or more available data segments meet the basic operating conditions, they can be grouped into the same set of data segments under the same operating conditions. Specifically, the basic operating conditions include: the program number or machining task number is the same, the program segment number is the same or belongs to the same preset program segment range, the tool number is the same, the spindle command speed, workpiece axis command speed, and command feed rate are the same or fall into the same preset range, and the corresponding sampling ranges are all in a stable machining state.

[0123] When it is possible to obtain auxiliary working condition information such as gear machining process type, gear machining process stage, gear model, workpiece type, material batch, blank allowance range, cooling state, spindle ratio, feed ratio, or cumulative tool machining time, the above auxiliary working condition information can also be used as auxiliary working condition labels to further subdivide or label the set of data segments with the same working condition.

[0124] In another implementation of this embodiment, a hierarchical classification method based on the same working condition can be adopted. When the auxiliary working condition labels are complete and the number of historical samples is sufficient, a refined classification based on the same working condition is used, for example, the working condition is determined by program segment number, tool number, gear model, material batch, blank allowance range, spindle speed range, workpiece axis speed range, and feed rate range. When the auxiliary working condition labels are missing or the number of samples is insufficient, a basic classification based on the same working condition is used, for example, the working condition is determined by program segment number, tool number, machining status, spindle speed range, and feed rate range, and the warning result is marked to indicate that the lack of auxiliary working condition information or insufficient samples may affect the confidence level of the judgment.

[0125] For continuous numerical parameters such as spindle speed command, workpiece axis speed command, and commanded feed rate, interval division can be used to determine whether they belong to the same working condition. For example, if the spindle speed command of two data segments both fall within the same preset speed range, the workpiece axis speed command both fall within the same preset speed range, and the commanded feed rate both fall within the same preset feed range, then they can be considered to meet the requirements for classification as the same working condition in terms of speed and feed conditions. Alternatively, a relative difference threshold can be used for judgment; that is, when the relative difference between the spindle speed command, workpiece axis speed command, and commanded feed rate of two data segments does not exceed a preset threshold, they are considered to meet the requirements for classification as the same working condition.

[0126] During the health baseline establishment phase, the aforementioned historically confirmed normal processing records are preferably used as the initial baseline data source. These historically confirmed normal processing records have already undergone data segmentation, availability score calculation, and availability threshold calibration as described above, and can be further used to establish health baselines under the same operating conditions.

[0127] In this embodiment, for each set of data segments under the same working condition, a health benchmark for that working condition is established based on the historically normal and usable data segments therein. Specifically, health response features are extracted for each historically normal and usable data segment. These health response features may include one or more of the following features: average spindle load, average spindle load fluctuation, average spindle power, average spindle current, average workpiece axis load, average feed axis load, average feed axis current, average position deviation, position deviation fluctuation, average synchronization deviation, synchronization deviation fluctuation, segment duration, machining cycle time, and the proportion of sustained high load.

[0128] For a given set of data segments under the same operating conditions, let the number of historically normal and usable data segments be . , No. The first historical data segment that is normally available Each health response characteristic is Then, under this working condition, the first... The baseline center value of each health response feature It can be represented as:

[0129] Under this working condition, the first The baseline fluctuation scale of a health response characteristic It can be represented as:

[0130] in: The reference center value; Used as the benchmark fluctuation scale; For the first The first historical data segment that is normally available Individual health response characteristics; This represents the number of historically available data segments.

[0131] In addition to the mean and standard deviation, quantile boundaries can be calculated based on historical, normally usable data segments, such as the upper quartile, 90th percentile, or 95th percentile. For health response characteristics such as spindle load, spindle power, spindle current, workpiece axis load, and feed axis load, which may increase with the deterioration of machining conditions, the high quantile boundary can be used as the health reference boundary under that condition. For characteristics such as position deviation, synchronization deviation, and segment duration, upper boundaries, lower boundaries, or bilateral boundaries can be set according to the actual process meaning.

[0132] During the health baseline update phase, not all available data segments participate in the baseline update. Data segments allowed to participate in the health baseline update must simultaneously meet the following conditions: the segment availability score is not lower than the preset availability threshold; the current overall health deviation does not exceed the preset baseline update threshold; the effective deviation increment is zero or lower than the preset update allowable value; the cumulative risk value is at the normal level; and the data segment has not triggered a health warning, has not been manually confirmed as abnormal, and the processing quality or production records do not show any abnormalities.

[0133] Data segments that have triggered minor deviation alerts, continuous deviation warnings, or high-risk deviation warnings, data segments that have generated significant and effective deviation increments, and data segments with manually confirmed anomalies or abnormal machining quality records are not included in health baseline updates. This prevents data on tool wear, abnormal cutting loads, abnormal blank allowances, or other deteriorating machining conditions from being absorbed into the health baseline, which would cause the health baseline to gradually rise.

[0134] Through the above-mentioned data segment set and health benchmark construction process under the same working conditions, the current data segment can be judged for health deviation under the same or similar gear processing working conditions, avoiding direct mixing and comparison of data under different program segments, different tools, different gear models, different spindle speeds, different workpiece shaft speeds, different feed rates, or different blank allowances.

[0135] S6: Compare the health response characteristics of the currently available data segment with the corresponding health benchmark under the same operating conditions, and calculate the health deviation of the currently available data segment.

[0136] In this embodiment, the method for calculating the health deviation of the currently available data segment is as follows.

[0137] Let the currently available data segment be the first... Each health response characteristic is Its corresponding center value in the same working condition health benchmark is The fluctuation scale is Then the normalized deviation of this feature can be expressed as:

[0138] in: Indicates the first Normalized deviation of each health response feature; To prevent positive numbers with a denominator of zero.

[0139] For features that increase with increased cutting load or deteriorated machining conditions, a one-sided deviation calculation method is used. Specifically, for features such as spindle load, spindle power, spindle current, workpiece axis load, and feed axis load that may increase with increased cutting load, tool wear, abnormal cooling and chip removal, or deteriorated machining conditions, a one-sided deviation calculation method can be used.

[0140] That is, when the current characteristic value is lower than or close to the historical normal center value, it is not considered to constitute a positive health deviation; when the current characteristic value is significantly higher than the health benchmark under the same working conditions, the deviation increases. For characteristics such as positional deviation, synchronization deviation, or processing cycle time, bilateral deviation or unilateral deviation can be adopted according to their process meaning.

[0141] After obtaining the normalized deviation of each health response feature, the features can be weighted according to their importance to obtain the comprehensive health deviation of the currently available data segment:

[0142] in: This represents the overall health deviation of the currently available data segment; This indicates the number of health response features involved in the calculation; Indicates the first The weight coefficients of each health response feature are non-negative and satisfy the following conditions: .

[0143] When prior weights are lacking, the weights of each health response feature can be made equal. If the monitoring target focuses on tool wear risk or cutting load deviation, the weights of features such as spindle load, spindle power, and spindle current can be appropriately increased. If the monitoring target focuses on feed-related operational response deviation, the weights of features such as feed axis load, feed axis current, position deviation, or synchronization deviation can be increased.

[0144] In this embodiment, for the currently available data segment, the first... Weighting coefficients of each health response feature Its normalized deviation product As a characteristic of this health response, the overall health deviation The contribution value of each health response feature is determined, and the feature with the largest contribution value is identified as the primary deviation variable for the current data segment. When two or more health response features have the same contribution value and are both at their maximum value, they can be recorded together as primary deviation variables. For example, when the spindle load average and spindle power average correspond to... When the values ​​are large, the spindle load and spindle power can be used as the main deviation variables for this segment; when the workpiece axis load, feed axis load, position deviation, or synchronization deviation corresponds to... When the deviation is large, the corresponding variable can be used as the primary deviation variable. This primary deviation variable is used for subsequent consistency judgment of deviation direction and interpretation of warning results.

[0145] S7: Accumulate the deviations of multiple available data segments in the same working condition data segment sequence in chronological order, and calculate the cumulative risk value.

[0146] Because the number of sampling points contained in a single data segment in low-frequency sampling data is limited, the health deviation of a single data segment may be affected by the sampling phase, periodic fluctuations in gear machining load, local entry and exit disturbances, or occasional process disturbances. Therefore, this embodiment does not use the health deviation of a single data segment as the sole basis for early warning, but instead performs a cumulative deviation judgment on multiple available data segments in the same working condition data segment set in chronological order.

[0147] In this embodiment, the method for calculating the cumulative risk value is as follows.

[0148] Multiple available data segments from the same set of data segments under the same working condition are arranged in chronological order according to their start time or end time to form a sequence of data segments under the same working condition.

[0149] Set deviation from reference value The deviation reference value represents the healthy deviation boundary corresponding to the normal fluctuation range under the same operating conditions. Specifically, for the first data segment in the same operating condition data sequence... There are available data segments, and their segment availability score is 1. The overall health deviation is To prevent small deviations within the normal range from accumulating continuously, this embodiment sets a deviation reference value. The deviation from the reference value Used to indicate the healthy deviation boundary corresponding to the normal fluctuation range under the same operating conditions. Deviation reference value The health deviation can be determined based on the distribution of historical normal available data segments, such as the 90th or 95th percentile of historical normal health deviation, or it can be set according to field experience or process requirements.

[0150] Current health deviation Not exceeding the deviation from the reference value At that time, it is considered that the deviation of this segment is still within the normal fluctuation range and does not generate new risk increment; when the current health deviation is... Exceeding the reference value When the fluctuation exceeds the normal range, the portion exceeding this range is considered the effective deviation increment. The effective deviation increment can be expressed as:

[0151] in: Indicates the first The effective deviation increment of each available data segment.

[0152] Constructing a cumulative risk value based on the effective deviation increment:

[0153] in: Indicates the first data segment in the same working condition sequence. The cumulative risk value after processing each available data segment; This represents the cumulative risk value after processing the previous available data segment; This represents the historical risk decay coefficient, with a value ranging from 0 to 1. This indicates the availability score of the current data segment; This represents the coefficient of consistency in deviation direction; This represents the effective deviation increment of the current data segment.

[0154] Specifically, the historical risk decay coefficient This is used to gradually reduce the risk contribution that occurred earlier as the sequence of operating condition data segments progresses. Suppose we want a certain historical risk contribution to gradually decrease over time. After subsequent data segments under the same operating conditions are available, the contribution rate will decrease to the original contribution ratio. ,in It is a positive integer. ,but Determine by the following formula:

[0155] When using the half-decay rule, one can take... And based on the desired risk memory length calculate . The smaller, The smaller the value, the more sensitive the cumulative risk value is to changes in recent data segments; The larger, The closer the cumulative risk value is to 1, the longer the historical deviation information can be retained over a longer period of time.

[0156] Deviation direction consistency coefficient Used to characterize the current number Whether the main deviation variables and deviation directions of each available data segment are consistent with the previous available data segment under the same working condition. For the first available data segment in the sequence of data segments under the same working condition, take... For subsequent available data segments, if the main deviation variable of the current segment is the same as that of the previous segment, and the direction of deviation is consistent, then take... Otherwise take For health response features calculated using unilateral positive deviation, the deviation direction is considered consistent as long as the main deviation variables are the same. For health response features calculated using bilateral deviation, if the health response features of the current segment and the previous segment are relative to the center value of the health baseline under the same working condition... If the difference values ​​have the same sign, that is, if both deviations are higher than the reference center or both are lower than the reference center, then the deviation directions are considered to be the same.

[0157] By introducing deviation from the reference value Small deviations within the normal range will not continuously increase the cumulative risk value; by introducing a historical risk decay coefficient λ, the early risk contribution can gradually weaken as the sequence progresses; by introducing a fragment availability score... Data segments with higher availability contribute more to the cumulative risk value, while data segments with lower availability, even with some deviation, do not have a significant impact on the cumulative risk value; this is addressed by introducing a deviation direction consistency coefficient. This can further suppress the impact of single, occasional deviations on health warning results.

[0158] S8: Output the health warning result of the CNC gear processing machine tool based on the accumulated risk value.

[0159] In this embodiment, a health warning result for the CNC gear machining machine is output based on the cumulative risk value obtained in step S7. The health warning result indicates that, under a specific data segment set corresponding to the gear machining conditions, the machine tool's operating response has continuously deviated from the historical health benchmark for the same operating conditions. This result is not directly equivalent to a specific fault root cause diagnosis result.

[0160] In one embodiment of this example, multiple risk thresholds are set based on the cumulative risk value distribution of data segment sequences under the same working conditions in historically confirmed normal processing records. That is, in this embodiment, the cumulative risk value is divided into different health warning levels. For example, when the cumulative risk value is below the first risk threshold, a normal state is output; when the cumulative risk value reaches the first risk threshold but is below the second risk threshold, a slight deviation warning is output; when the cumulative risk value reaches the second risk threshold but is below the third risk threshold, a continuous deviation warning is output; and when the cumulative risk value reaches the third risk threshold, a high-risk deviation warning is output.

[0161] In a preferred embodiment of this example, the first risk threshold, the second risk threshold, and the third risk threshold can be determined based on the cumulative risk value distribution of the data segment sequence under the same working condition in historically confirmed normal processing records. For example, a data segment sequence under the same working condition can be constructed based on historically confirmed normal processing records, and the cumulative risk value distribution under historical normal conditions can be calculated in the same way as online monitoring; then, different high quantile boundaries of the cumulative risk value distribution can be taken as risk thresholds of different levels.

[0162] For example:

[0163] in: These represent the first risk threshold, the second risk threshold, and the third risk threshold, respectively. , and These represent the 90th, 95th, and 99th percentiles of the historical normal cumulative risk value distribution, respectively. This represents the cumulative risk value calculated from a sequence of data segments under the same operating conditions within historically confirmed normal processing records. The quantile values ​​mentioned above can be adjusted based on false alarm tolerance, maintenance response requirements, and on-site management strategies.

[0164] In this embodiment, the health warning result may include the following: health deviation level, cumulative risk value, corresponding set of data segments under the same working condition, working condition label, main deviation variables, and maintenance and inspection prompts. The working condition label includes one or more of the following: program segment number, tool number, spindle speed range, workpiece axis speed range, feed rate range, machining status, gear model, gear machining process stage, material batch, blank allowance range, and cooling status, used to indicate the machining background in which the health deviation occurs. The main deviation variables are used to explain that the health deviation is mainly reflected in operational response data such as spindle load, spindle power, spindle current, workpiece axis load, feed axis load, feed axis current, position deviation, or synchronization deviation.

[0165] When generating maintenance check prompts, the direction of inspection can be determined based on the main deviation variables. If the main deviation variables are concentrated in spindle load, spindle power, or spindle current, the prompt can indicate the need to check tool wear condition, grinding wheel condition, cutting parameters, blank allowance, cooling and chip removal condition, or machining condition. If the main deviation variables are concentrated in workpiece axis load, feed axis load, feed axis current, position deviation, or synchronization deviation, the prompt can indicate the need to check workpiece axis or feed axis running condition, electronic gearbox synchronization condition, trajectory following condition, fixture clamping condition, cutting load, or process parameters. The above maintenance check prompts are used to assist maintenance personnel in determining the direction of inspection and should not be used as the sole basis for determining the root cause of the fault.

[0166] For example, when the working condition label corresponding to a certain set of data segments with the same working condition is tool T01, program segment N120, spindle speed range of 1000 to 1200 r / min, workpiece axis speed range of 80 to 100 r / min, and feed rate range of 80 to 100 mm / min, and the cumulative risk value of this set reaches the high-risk deviation threshold, and the main deviation variables are spindle load and spindle power, the system can output: "High-risk health deviation detected when tool T01 executes program segment N120 and is under the corresponding spindle speed, workpiece axis speed, and feed conditions; the main deviation variables are spindle load and spindle power; it is recommended to check the tool wear condition, cutting parameters, blank allowance, cooling and chip removal condition, or machining condition."

[0167] In one optional embodiment of this example, the method further includes summarizing and analyzing the warning results of multiple sets of data segments under the same working condition. The method is as follows: if multiple sets of data segments under the same working condition with the same tool number but corresponding to different program segments or different cutting parameters all show health deviations, and the main deviation variables are concentrated in the spindle load, spindle power or spindle current, then the maintenance inspection priority corresponding to that tool can be increased; if multiple sets of data segments with the same program segment, the same gear machining stage or the same gear model but corresponding to different tools all show health deviations, then the inspection priority of the process parameters, blank allowance, material condition or cooling and chip removal status of that machining stage can be increased.

[0168] The following section provides a specific example to further illustrate the detailed implementation of the health early warning method for CNC gear processing machine tools based on low-frequency sampling operation data.

[0169] Specifically, this embodiment describes the specific implementation of the invention using a CNC gear hobbing machine to process similar gears in batches. This embodiment is only used to illustrate the technical solution of the present invention and does not limit the scope of protection of the present invention. This embodiment can also be applied to CNC gear processing machine tools that perform gear shaping, grinding, shaving, honing, or gear composite machining, which have repetitive machining conditions, can acquire low-frequency internal operating data, and can classify working conditions.

[0170] In this embodiment, the machine tool under monitoring provides control status data through the CNC system and PLC, and operational response data through the spindle driver, workpiece axis driver, and servo driver. The data sampling period is 5 seconds, meaning that machine tool operation data is read once every 5 seconds. The collected fields include program segment number, tool number, NC operating status, PLC status word, machining stage identifier, spindle command speed, actual spindle speed, workpiece axis command speed, actual workpiece axis speed, commanded feed rate, actual feed rate, electronic gearbox synchronization status, spindle load rate, spindle power, spindle current, workpiece axis load rate, feed axis load rate, feed axis current, and position deviation. If the MES or process documents can provide gear model, material batch, blank allowance, production batch, cumulative cutting time of the tool, and cooling status, these are also recorded as auxiliary operating condition labels.

[0171] After data collection, data from different sources are aligned according to sampling time, and missing, invalid, or obviously abnormal fields are marked, ensuring that each sampling record simultaneously contains corresponding control status data and health response data. Subsequently, candidate stable machining data segments are segmented from continuous low-frequency sampling data. When the machine tool is in an alarm, pause, tool change, rapid traverse, spindle start / stop, workpiece axis synchronization establishment, unstable electronic gearbox synchronization, or non-machining state, the corresponding sampling point does not participate in the construction of candidate stable machining data segments; when the program segment number changes, the tool number changes, or there are significant changes in the spindle command speed, workpiece axis command speed, or commanded feed rate, the corresponding change time is used as the boundary of the data segment.

[0172] For a data segment formed between adjacent boundaries, if the deviation between the actual spindle speed and the spindle commanded speed, the deviation between the actual workpiece spindle speed and the workpiece spindle commanded speed, and the deviation between the actual feed rate and the commanded feed rate are all within the allowable range in continuous sampling points, and the electronic gearbox synchronization is stable, then the data segment is determined as a candidate stable machining data segment.

[0173] For each candidate stable processing data segment, a segment availability score is calculated. In this embodiment, the segment availability score is obtained by weighting the sample sufficiency score, data integrity score, state stability score, instruction feedback consistency score, and boundary distance score. If no prior weights are available, the weight of each sub-score is set to 0.2. The preset minimum number of effective sampling points is determined based on the sampling period and the expected stable processing duration: when the sampling period is 5 seconds and it is desired that each segment covers a stable processing period of no less than 30 seconds, the minimum number of effective sampling points is 6.

[0174] Fragment availability threshold The availability score is determined based on the distribution of comprehensive availability scores for candidate stable processing data segments in historically confirmed normal processing records. In this embodiment, the same data segmentation and availability score calculation as the online monitoring process are first performed on the historically confirmed normal processing records to obtain the comprehensive availability scores for multiple candidate stable processing data segments. The historically confirmed normal processing records refer to gear processing records where no machine tool alarms occurred, no abnormalities were manually confirmed, gear processing quality or production records showed no abnormalities, and the corresponding cutting tool was in the stable processing stage or normal use stage. Then, the low quantile of the above comprehensive availability score is taken as the initial availability threshold, and corrected by combining the dispersion of the health response characteristics under the same working condition and the number of historical segments retained. If the availability threshold determined according to the above process is close to 0.7, then... The value is set to 0.7. This value is only an example value in this embodiment. Under other machine tools, other sampling periods, or other gear machining conditions, it can be re-determined based on the corresponding historical confirmation of normal machining records. If the availability score of the candidate stable machining data segment is not lower than 0.7, it is determined as a usable data segment; if it is lower than 0.7, it is determined as a low availability data segment.

[0175] For example, a candidate stable machining data segment belongs to program segment N120, tool T01, and the stable gear hobbing stage. The sampling period is 5 seconds, and the segment contains 8 valid sampling points with no missing points. The program segment number, tool number, and running status are consistent. The average relative deviation between the actual spindle speed and the commanded speed is less than 2%, the average relative deviation between the actual workpiece axis speed and the commanded speed is less than 2%, and the average relative deviation between the actual feed rate and the commanded feed rate is less than 3%. Furthermore, the segment is more than 20 seconds away from the most recent program segment switch or tool change boundary. This segment has an availability score higher than 0.7 and can be included in the health deviation calculation and participate in the health baseline update when normal conditions are met. Another candidate segment, although with a high spindle load, is only 5 seconds away from the program segment switch boundary, and the actual spindle speed is still in the transition process of following the commanded speed. Therefore, this segment is marked as a low availability segment, does not participate in the formal health warning trigger, and does not participate in the health baseline update under the same operating conditions.

[0176] After obtaining usable data segments, a set of data segments for the same working condition is constructed based on the program segment number, tool number, machining stage, spindle speed range, workpiece axis speed range, and feed rate range. When auxiliary information such as gear model, material batch, blank allowance, production batch, cooling status, and cumulative cutting time of the tool can be obtained, refined same-working-condition classification is prioritized; when auxiliary information is missing or the number of samples under a certain refined working condition is insufficient, basic same-working-condition classification is used, and the explanation of missing auxiliary information or insufficient samples is retained in the subsequent warning results.

[0177] During the health baseline establishment phase, this implementation method uses the above-described method to determine the fragment availability threshold. Historical confirmed normal processing records serve as the initial benchmark data source. These historical confirmed normal processing records have undergone data segmentation, availability score calculation, and availability threshold calibration as described above, and can be further used to establish health benchmarks under the same operating conditions.

[0178] In this embodiment, six health response features are extracted for each historically usable data segment: mean spindle load, spindle load fluctuation, mean spindle power, mean workpiece axis load, mean feed axis load, and position deviation fluctuation. If the electronic gearbox synchronization deviation can be obtained, the mean or fluctuation of the synchronization deviation can also be used as health response features. For each set of data segments under the same operating condition, the mean, standard deviation, and high quantile boundary of the above health response features in the historically usable data segments are calculated as the health benchmark under that operating condition.

[0179] During online monitoring, when a new available data segment is generated, it is matched to the corresponding set of data segments under the same working condition based on its program segment number, tool number, machining stage, spindle speed range, workpiece axis speed range, and feed rate range. Then, the health response features of this data segment are extracted and compared with the health benchmark under the same working condition to calculate the normalized deviation of each health response feature. For the average spindle load, average spindle power, average workpiece axis load, and average feed axis load, a one-sided deviation method is used, meaning a positive deviation is only counted when the current feature value is higher than the historical normal center value under the same working condition. For spindle load fluctuations, position deviation fluctuations, and synchronization deviation fluctuations, the deviation is calculated based on the degree of increase relative to the historical normal fluctuation scale.

[0180] After obtaining the deviation of each health response feature, the comprehensive health deviation of the data segment is calculated according to the preset weights, and then calculated using the weight coefficients of each health response feature. Deviation from normalization product As the contribution value of this feature to the overall health deviation, the health response feature with the largest contribution value is determined as the main deviation variable of the current data segment. For each set of data segments under the same working condition, the available data segments are arranged in chronological order of their end times to form a sequence of data segments under the same working condition. A deviation reference value is set. In this embodiment, the 95th percentile of the comprehensive health deviation of the historical normal and usable data segment under the same working conditions is taken; in other embodiments, the 90th percentile or other percentile boundaries may also be taken according to maintenance requirements.

[0181] When the overall health deviation of the current data segment does not exceed When the deviation from the expected health standard is considered to be within the normal fluctuation range and does not generate new risk increments, then when it exceeds the expected level, the deviation is considered to be within the normal fluctuation range and does not generate new risk increments. At that time, only the excess portion is considered as valid deviation increment. The cumulative risk value is calculated according to... Calculation. In this embodiment, a semi-decay rule is adopted, and it is desired that a certain historical risk contribution decays to approximately 50% of its original contribution after three subsequent usable data segments under the same operating conditions. This can be approximated as 0.8. For the first usable data segment in a sequence of data segments under the same operating conditions, Set to 1; for subsequent segments, when the principal deviation variable of the current segment is the same as that of the previous segment, and the deviation direction of the two-sided deviation feature relative to the center value of the health baseline is consistent, Take 1 if the main deviation variable is the same, otherwise take 0.5. For unilateral positive deviation features, the same main deviation variable can be considered as having the same direction.

[0182] For example, in a sequence of data segments under the same operating conditions, if the average spindle load and average spindle power of multiple consecutive available data segments are higher than the health benchmark under the same operating conditions, and the main deviation variables remain consistent, then the portion of each segment exceeding the deviation reference value will gradually accumulate under the influence of availability score and directional consistency coefficient. This causes the accumulated risk value to progress from a normal state to a slight deviation warning, and then to a continuous deviation warning. If only a segment experiences a short-term load increase, but the availability of that segment is low, or subsequent segments do not show the same directional deviation, then its contribution to the accumulated risk value is weakened, and it is less likely to directly trigger a high-level warning.

[0183] Health warning results are output based on cumulative risk values. In this embodiment, based on the aforementioned historical confirmed normal processing records, the cumulative risk value distribution of each data segment sequence under the same working condition is calculated, and three risk thresholds are set accordingly. Specifically, the same data segment division, availability score calculation, availability threshold screening, same working condition classification, health deviation calculation, and cumulative risk value calculation are first performed on the historical confirmed normal processing records, as in the online monitoring process, to obtain the cumulative risk value distribution of each data segment sequence under the same working condition under historical normal conditions; then, the 90th percentile of the historical normal cumulative risk value distribution is used as the first risk threshold, the 95th percentile as the second risk threshold, and the 99th percentile as the third risk threshold.

[0184] When the cumulative risk value obtained from online monitoring is lower than the first risk threshold, a normal status is output; when the cumulative risk value reaches the first risk threshold but is lower than the second risk threshold, a slight deviation warning is output; when the cumulative risk value reaches the second risk threshold but is lower than the third risk threshold, a continuous deviation warning is output; when the cumulative risk value reaches the third risk threshold, a high-risk deviation warning is output. The above quantile values ​​can be adjusted according to the false alarm tolerance, maintenance response requirements, and on-site management strategies.

[0185] The health warning results include the health deviation level, cumulative risk value, corresponding set of data segments under the same working condition, working condition label, main deviation variables, and maintenance and inspection prompts. The working condition label includes one or more of the following: program segment number, tool number, gear machining stage, spindle speed range, workpiece axis speed range, feed rate range, gear model, material batch, and blank allowance range, used to indicate the machining background in which the health deviation occurs; the main deviation variables are used to explain in which type of operational response data the health deviation is mainly reflected.

[0186] For example, when the working condition label corresponding to a certain set of data segments with the same working condition is tool T01, program segment N120, stable hobbing stage, spindle speed range of 1000 to 1200 r / min, workpiece axis speed range of 80 to 100 r / min, and feed rate range of 80 to 100 mm / min, and the cumulative risk value of this set reaches the high-risk deviation threshold, and the main deviation variable is the average spindle load or the average spindle power, the system outputs: "High-risk health deviation detected when tool T01 executes program segment N120 and is under the corresponding speed and feed conditions; the main deviation variable is the average spindle load or the average spindle power; it is recommended to check the tool wear condition, cutting parameters, blank allowance, cooling and chip removal condition, or machining condition."

[0187] In another scenario, when the main deviation variables of a set of data segments under the same working condition are the average workpiece axis load, the average feed axis load, the position deviation fluctuation value, or the synchronization deviation fluctuation value, the system outputs that there is a deviation in the feed or synchronization-related operating response under this working condition, and prompts to check the workpiece axis running status, electronic gearbox synchronization status, feed axis trajectory following status, fixture clamping status, cutting load, or process parameters. The above maintenance check prompts are used to assist maintenance personnel in determining the direction of inspection and are not the sole basis for determining the root cause of the fault.

[0188] Compared to directly using a fixed threshold for judgment, this implementation reduces the impact of normal response differences caused by different program segments, tool types, gear models, speeds, feed rates, material batches, or blank allowance conditions on the early warning results by using a health benchmark under the same working conditions. Compared to directly using a single segment deviation judgment, this implementation suppresses single occasional fluctuations by accumulating effective deviations of the same working condition segment sequence. Compared to directly using all segments for health benchmark updates, this implementation reduces the contamination of the health benchmark by segments such as state switching, insufficient sampling points, unstable command feedback, or missing data by segment availability evaluation.

[0189] This implementation method enables the elimination of interference from low-availability data fragments on health assessments using only low-frequency sampling of internal operating data, and identifies persistent health deviations under the same operating conditions, thereby providing a basis for health warnings and maintenance checks of CNC gear processing machine tools.

[0190] Compared with the prior art, the present invention has at least the following beneficial effects.

[0191] First, this invention is applicable to health early warning under low-frequency sampling of internal operating data in CNC gear processing machine tools. The method of this invention does not rely on external high-frequency vibration sensors, acoustic emission sensors, or high-frequency acquisition devices. Instead, it utilizes low-frequency internal operating data that can be periodically read from the CNC system, PLC, servo driver, spindle driver, workpiece axis driver, electronic gearbox, or industrial gateway for health status analysis. Therefore, this invention is more suitable for deployment in existing CNC gear processing machine tools and industrial field data acquisition conditions, reducing reliance on additional sensors and high-frequency acquisition hardware.

[0192] Secondly, this invention avoids directly using unsuitable data segments for health assessment by evaluating the availability of candidate stable machining data segments. For data segments with insufficient sampling points, high data loss rates, unstable states, large deviations in spindle / workpiece / feed axis command feedback, unstable electronic gearbox synchronization, or those approaching state boundaries such as program segment switching, tool changing, or spindle start / stop, this invention can prevent them from entering the health baseline update process or reduce their impact on health deviation calculation and cumulative risk calculation. Therefore, it can reduce false alarms and health baseline contamination caused by low-availability data segments under low-frequency sampling conditions.

[0193] Third, by constructing a set of data segments under the same working conditions and establishing corresponding health benchmarks, this invention can reduce the interference of varying working conditions in CNC gear machining on health warning results. Since machine tool operation responses differ under different program segments, tool types, gear models, machining stages, spindle speeds, workpiece axis speeds, feed rates, material batches, and blank allowances, this invention compares the current data segment with historically normal and usable data segments under the same or similar working conditions, rather than directly using a globally fixed threshold for judgment. Therefore, it can reduce false alarms caused by changes in normal working conditions and identify health deviation phenomena that have not yet exceeded fixed alarm limits but have shown continuous deviations under the same working conditions.

[0194] Fourth, this invention improves the stability of health warnings under low-frequency sampling conditions by accumulating deviations from data segment sequences under the same operating conditions. The health deviation of a single low-frequency data segment may be affected by occasional disturbances, sampling phase, periodic fluctuations in gear machining load, or local process fluctuations. This invention arranges available data segments in the same set of data segments under the same operating conditions in chronological order and combines deviation reference values, segment availability scores, historical risk attenuation coefficients, and deviation direction consistency coefficients for cumulative judgment. This allows persistent deviations to gradually enhance the warning response, while small fluctuations within the normal range and single occasional deviations are less likely to directly trigger high-level warnings.

[0195] Fifth, this invention can provide health warning results with gear machining condition context. The warning results not only include the health deviation level and cumulative risk value, but also can be associated with the corresponding set of data segments under the same working condition, program segment number, tool number, gear machining stage, spindle speed range, workpiece axis speed range, feed rate range, gear model, material batch, blank allowance range, and main deviation variables. Maintenance personnel can use this information to determine in which machining context the health deviation mainly occurs and in which type of operational response variable it is mainly reflected, thereby enabling targeted checks on tool status, grinding wheel status, cutting parameters, blank allowance, cooling and chip removal status, workpiece axis or feed axis operating status, electronic gearbox synchronization status, or process parameters.

[0196] Sixth, the early warning results of this invention are not directly equivalent to specific fault root cause diagnosis, but are used to indicate persistent health deviations observable in low-frequency sampling data. This method is more in line with the characterization capabilities of internal operating data in low-frequency sampling, and is particularly suitable for early warning of risks such as tool wear risk, changes in grinding wheel condition, abnormal deviations in cutting load, abnormal cooling and chip removal, deviations in feed-related operating responses, or deterioration of gear machining conditions, and can provide auxiliary basis for equipment maintenance and machining process inspection.

[0197] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A health early warning method for CNC gear machining tools based on low-frequency sampled operating data, characterized in that: The steps include the following: S1: Acquire low-frequency sampling internal operating data of the CNC gear processing machine tool during the processing process, the low-frequency sampling internal operating data including control status data and health response data; S2: Based on the control status data, according to the changes in the command values ​​and actual feedback values ​​of the machine tool operating status, program segment number, tool number, spindle speed, workpiece axis speed and feed rate, candidate stable machining data segments are divided from the continuous low-frequency sampling data; S3: Calculate the fragment availability score for the candidate stable processing data fragment. The fragment availability score is used to evaluate whether the data fragment can be used as a basis for health evaluation, and is used for at least the following processes: determining whether the data fragment is allowed to enter the subsequent health deviation calculation, determining whether the data fragment is allowed to participate in the health benchmark update under the same working condition, and as a weighting factor in the cumulative deviation of the subsequent data fragment sequence under the same working condition. S4: Based on the fragment availability score, candidate stable processing data fragments that meet the availability threshold are determined as available data fragments, and candidate stable processing data fragments that do not meet the availability threshold are determined as low availability data fragments. S5: For the available data segments, construct a set of data segments under the same working conditions according to their corresponding gear processing conditions and machine tool operating status, and establish a health benchmark for each working condition based on historical normal available data segments. S6: Compare the health response characteristics of the currently available data segment with the corresponding health benchmark under the same working conditions, and calculate the health deviation of the currently available data segment; S7: Accumulate the deviation of multiple available data segments in the same working condition data segment sequence according to the time order, and calculate the cumulative risk value; S8: Output the health warning result of the CNC gear processing machine tool based on the accumulated risk value.

2. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 1, characterized in that: In step S1, the control status data includes one or more of the following: program number, program segment number, tool number, NC running status, PLC status word, spindle start / stop status, spindle commanded speed, spindle actual speed, workpiece axis commanded speed, workpiece axis actual speed, commanded feed rate, actual feed rate, spindle ratio, feed ratio, electronic gearbox synchronization status, tool life count, alarm status, pause status, tool change status, rapid traverse status, machining stage identifier, gear machining process type, coolant start / stop status, and lubrication status. The health response data includes one or more of the following: spindle load rate, spindle power, spindle current, workpiece axis load rate, feed axis load rate, feed axis current, position deviation, synchronization deviation, feed axis speed, machining cycle time, and segment duration.

3. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 1, characterized in that: In step S2, the method for dividing candidate stable processing data segments from continuous low-frequency sampling data is as follows: Based on the control status data, exclude sampling points for alarms, emergency stops, pauses, manual interventions, tool changes, rapid traverses, spindle start / stop, workpiece axis synchronization establishment processes, idling, or non-machining states; The boundaries of data segments are determined based on the changes in program segment number, tool number, machining stage, spindle speed command, workpiece axis speed command, and feed rate command, and the moment of change is used as the dividing position between adjacent data segments. For a data segment formed by consecutive sampling points between adjacent boundary positions, if the deviation between the actual spindle speed and the spindle command speed, the deviation between the actual workpiece spindle speed and the workpiece spindle command speed, and the deviation between the actual feed rate and the commanded feed rate are all within the preset allowable range, and the above deviations remain relatively stable within a number of consecutive sampling points, the data segment is determined as a candidate stable machining data segment.

4. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 1, characterized in that: In step S3, fragment usability score. Sufficiency of sample scoring Data integrity score State stability score Instruction feedback consistency score and boundary distance score Weighted average yields: in: , , , and Let represent the weighting coefficients, where each weighting coefficient is non-negative and satisfies ... .

5. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 4, characterized in that: The sample sufficiency score Determined based on the relationship between the number of valid sampling points within a segment and the preset minimum number of valid sampling points: in: This represents the number of valid sampling points within the segment; This is the preset minimum number of valid sampling points; The data integrity score Determined based on the proportion of missing or invalid sampling points within a segment to the total number of sampling points: in: This represents the number of missing or invalid sampling points within the segment. This represents the total number of sampling points within the segment; The state stability score The determination is based on whether there are changes in the program segment number, tool number, NC running status, PLC status word, spindle start / stop status, electronic gearbox synchronization status, alarm status, or pause status within the segment: in: The number of sampling points within a segment where a state change or exclusionary state marker occurs; The instruction feedback consistency score Determined based on the consistency between the actual feedback from the spindle, workpiece axis, and feed system and the corresponding command values: in: The feedback of the instruction constitutes a comprehensive deviation; This is the upper limit of the allowable deviation; The boundary distance score Determined based on the time distance between candidate stable machining data segments and program segment switching, tool change, spindle start / stop, workpiece axis synchronization adjustment, rapid traverse, pause, alarm, or acceleration / deceleration state boundaries: in: The time interval between the candidate stable processed data segment and the nearest state boundary; This is the preset minimum boundary interval.

6. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 1, characterized in that: In step S4, the availability threshold is determined based on the availability score distribution of candidate stable processing data segments in historically confirmed normal processing records: A batch of historically confirmed normal processing records were subjected to the same data segmentation and availability score calculation process as the online monitoring process to obtain a comprehensive availability score for multiple candidate stable processing data segments. ; The lowest quantile of the overall availability score is taken as the initial availability threshold. : in: This represents the comprehensive availability score of candidate stable processing data segments within historically confirmed normal processing records. quantiles; This indicates the proportion of low-availability segments that are allowed to be removed from historical, normal, candidate, stable processing data segments; Based on the initial availability threshold, the final availability threshold is determined by adjusting for the dispersion of health response characteristics under the same operating conditions and the number of retained historical candidate stable processing data segments. .

7. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 6, characterized in that: The final availability threshold is determined by combining the dispersion of health response characteristics under the same operating conditions with the number of retained historical candidate stable processing data segments for correction. The method is as follows: by As the initial threshold, select several data points with scores no lower than the overall availability score distribution of historically confirmed normal processing records. Candidate threshold ; For each candidate threshold , retain satisfaction Historical candidate stable processing data segments, and based on the retained segments, establish corresponding candidate health benchmarks and health response characteristic dispersion under the same working conditions; Among multiple candidate thresholds that meet the requirements of the dispersion of health response characteristics under the same working condition not exceeding the preset stability requirement and the number of retained historical candidate stable processing data segments not less than the preset minimum sample number, the candidate threshold with the smallest value is selected as the final availability threshold. : in: Indicates at the candidate threshold The dispersion of the health response characteristics under the same working conditions established by retaining the lower segment; This indicates the upper limit of the allowed dispersion. Indicates at the candidate threshold The number of historical candidate stable processed data segments to retain; This indicates the minimum number of samples required to establish a health benchmark under the same working conditions.

8. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 1, characterized in that: In step S5, the method for constructing a set of data segments under the same working condition is as follows: When two or more available data segments meet the basic working conditions, they are grouped into the same set of data segments with the same working conditions. The basic working conditions include: the program number or machining task number is the same, the program segment number is the same or belongs to the same preset program segment range, the tool number is the same, the spindle command speed, the workpiece axis command speed and the command feed speed are the same or fall into the same preset range, and the corresponding sampling ranges are all in a stable machining state. When auxiliary working condition information can be obtained, the gear machining process type, gear machining process stage, gear model, workpiece type, material batch, blank allowance range, cooling state, spindle ratio, feed ratio, or cumulative tool machining time are used as auxiliary working condition labels to further subdivide or label the data segment set of the same working condition.

9. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 1, characterized in that: In step S5, the method for establishing a health benchmark under the same working conditions is as follows: For each set of data segments under the same working condition, health response features are extracted based on historical normal and usable data segments. The health response features include one or more of the following: average spindle load, average spindle load fluctuation, average spindle power, average spindle current, average workpiece axis load, average feed axis load, average feed axis current, average position deviation, average position deviation fluctuation, average synchronization deviation, average synchronization deviation fluctuation, segment duration, and machining cycle time. Calculate the baseline central value and baseline fluctuation scale of each health response characteristic in the historical normal available data segment as the health baseline under the same working condition: in: The reference center value; Used as the benchmark fluctuation scale; For the first The first historical data segment that is normally available Individual health response characteristics; This represents the number of historically available data segments. Data segments that are allowed to participate in health baseline updates must simultaneously meet the following conditions: the segment availability score is not lower than the preset availability threshold, the current comprehensive health deviation does not exceed the preset baseline update threshold, the effective deviation increment is zero or lower than the preset update allowable value, the cumulative risk value is at the normal level, and the data segment has not triggered a health warning and has not been manually confirmed as abnormal.

10. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 1, characterized in that: In step S6, the method for calculating the health deviation of the currently available data segment is as follows: Let the currently available data segment be the first... Each health response characteristic is Its corresponding center value in the same working condition health benchmark is The fluctuation scale is Then the normalized deviation of this feature is: in: Indicates the first Normalized deviation of each health response feature; To prevent positive numbers with a denominator of zero; Alternatively, for features that increase with increased cutting load or deteriorated machining conditions, a one-sided deviation calculation method can be used: The overall health deviation is obtained by weighting the data according to the importance of each health response feature: in: This represents the overall health deviation of the currently available data segment; This indicates the number of health response features involved in the calculation; Indicates the first The weight coefficients of each health response feature are non-negative and satisfy the following conditions: ; Record one or more health response features that contribute the most as the primary deviation variables.

11. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 1, characterized in that: In step S7, the method for calculating the cumulative risk value is as follows: Multiple available data segments from the same set of data segments under the same working condition are arranged in chronological order according to the start time or end time of the segments to form a sequence of data segments under the same working condition. Set deviation from reference value The deviation reference value represents the healthy deviation boundary corresponding to the normal fluctuation range under the same working conditions; Current health deviation Not exceeding the deviation from the reference value At the same time, no new risk increment is generated; when Exceed When the fluctuation exceeds the normal range, the portion exceeding the normal fluctuation range is considered as the effective deviation increment: in: Indicates the first The effective deviation increment of each available data segment; Constructing a cumulative risk value based on the effective deviation increment: in: Indicates the first data segment in the same working condition sequence. The cumulative risk value after processing each available data segment; This represents the cumulative risk value after processing the previous available data segment; This represents the historical risk decay coefficient, with a value ranging from 0 to 1. This indicates the availability score of the current data segment; This represents the coefficient of consistency in deviation direction; This represents the effective deviation increment of the current data segment.

12. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 1, characterized in that: In step S8, the method for outputting the health warning result of the CNC gear processing machine tool is as follows: Multiple risk thresholds are set based on the cumulative risk value distribution of data segment sequences under the same working conditions in historically confirmed normal processing records; When the cumulative risk value is lower than the first risk threshold, the normal status is output. When the cumulative risk value reaches the first risk threshold but is lower than the second risk threshold, a slight deviation warning is output. When the cumulative risk value reaches the second risk threshold but is below the third risk threshold, a continuous deviation warning is output. When the accumulated risk value reaches the third risk threshold, a high-risk deviation warning is issued. The health warning results include the health deviation level, cumulative risk value, corresponding set of data segments under the same working conditions, working condition label, main deviation variables, and maintenance and inspection prompts. The working condition label includes one or more of the following: program segment number, tool number, gear machining stage, spindle speed range, workpiece axis speed range, feed rate range, gear model, material batch, and blank allowance range. The maintenance check prompts determine the inspection direction based on the main deviation variables: if the main deviation variables are concentrated in the spindle load, spindle power, or spindle current, the prompts indicate to check the tool wear condition, grinding wheel condition, cutting parameters, blank allowance, cooling and chip removal condition, or machining condition; if the main deviation variables are concentrated in the workpiece axis load, feed axis load, feed axis current, position deviation, or synchronization deviation, the prompts indicate to check the workpiece axis or feed axis running condition, electronic gearbox synchronization condition, trajectory following condition, fixture clamping condition, cutting load, or process parameters.

13. The health early warning method for CNC gear processing machine tools based on low-frequency sampled operating data according to claim 12, characterized in that: It also includes summarizing and analyzing the early warning results of multiple sets of data segments under the same operating conditions: If multiple sets of data segments under the same working condition with the same tool number but corresponding to different program segments or different cutting parameters all show health deviations, and the main deviation variables are concentrated in spindle load, spindle power or spindle current, then the maintenance and inspection priority corresponding to that tool should be increased. If multiple sets of data segments with the same program segment, the same gear machining stage, or the same gear model but corresponding to different cutting tools all show health deviations, then the priority of checking the process parameters, blank allowance, material condition, or cooling and chip removal status of that machining stage should be increased.

Images