A hobbing machine machining precision prediction data processing method

Abstract

The present application relates to the technical field of numerical control machine tool processing prediction, and discloses a hobbing machine tool processing precision prediction data processing method.The method comprises starting full-cycle data acquisition after receiving a prediction request.First, a dynamic precision threshold sequence matching the current processing sequence is generated.By analyzing the deviation of real-time load fluctuation from the threshold, a potential attenuation trajectory of processing precision is drawn.This trajectory is associated and matched with a historical database, a candidate process influence point set causing attenuation is located and contribution degree tracing is performed, and key interference elements are identified.According to the key elements, the existing prediction model is structurally reconstructed to obtain an adaptive model.Real-time processing data is input into the model, iterative calculation is performed, and the precision evolution profile in the future time window is output.According to the compliance degree of the profile and the quality target, a corresponding parameter adjustment strategy is assigned to the machine tool control system.The present application realizes dynamic, traceable prediction and self-adaptive control of processing precision.

Description

Technical Field

[0001] This invention relates to the field of CNC machine tool machining prediction technology, specifically a data processing method for predicting machining accuracy on a gear hobbing machine. Background Technology

[0002] In the field of gear hobbing, predicting machine tool machining accuracy is crucial for ensuring workpiece quality and enabling preventative maintenance. Current technologies generally employ static accuracy thresholds or models based on fixed rules for accuracy monitoring and early warning. These methods typically pre-determine a uniform accuracy tolerance range for a specific machine tool model or standard process, or use the model permanently after training. However, gear hobbing is a dynamic process; parameters such as spindle speed and tool feed change in real time depending on the machining sequence. When dealing with different workpieces and different processes, fixed thresholds or models struggle to accurately reflect the true accuracy boundaries under specific machining conditions, leading to insensitive early warnings or frequent false alarms.

[0003] When machining accuracy shows a deviation trend, conventional methods often rely on the operator's experience for troubleshooting or simply attribute the cause by monitoring whether a single parameter exceeds its limit. These methods lack a systematic analysis of the accuracy decay trajectory under the coupled effects of multiple parameters during machining, and are unable to quickly correlate real-time abnormal patterns with massive amounts of historical process data. Therefore, it is difficult to accurately and automatically locate the root cause of the problem from the complex set of process parameters, resulting in delayed and untargeted adjustment decisions that affect machining efficiency and quality stability. Summary of the Invention

[0004] The purpose of this invention is to provide a method for processing data to predict the machining accuracy of gear hobbing machines, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for processing data to predict the machining accuracy of a gear hobbing machine, the method comprising: Receive a hobbing accuracy prediction request from the control system, and initiate full-cycle data acquisition for the specified workpiece machining sequence based on the hobbing accuracy prediction request. Based on the historical data stream of the spindle speed and tool feed of the gear hobbing machine, a dynamic accuracy threshold sequence is generated that matches the current machining sequence; By analyzing the deviation relationship between real-time load fluctuations during the processing and the dynamic accuracy threshold sequence, a potential attenuation trajectory of processing accuracy is plotted. The potential trajectory of machining accuracy decay is correlated and matched with multiple sets of historical process data in the process parameter database to locate the set of candidate process influence points that lead to the occurrence of accuracy decay trajectory. Perform an independent contribution source analysis on each process influence point in the candidate process influence point set to identify the key interference elements; Based on key interference factors, the currently deployed accuracy prediction model is structurally reconstructed to generate an adaptive accuracy prediction model; The real-time acquired processing status data stream is input into the adaptive accuracy prediction model, which iteratively calculates and outputs the accuracy evolution profile within a specified future time window. Based on the degree of conformity between the accuracy evolution profile and the machining quality target, corresponding parameter adjustment strategies are assigned to the real-time control system of the gear hobbing machine.

[0006] Preferably, receiving the hobbing accuracy prediction request from the control system and initiating full-cycle data acquisition for the specified workpiece machining sequence based on the hobbing accuracy prediction request specifically involves: Analyze the workpiece identifier and target machining accuracy level contained in the gear hobbing accuracy prediction request; Retrieve the corresponding processing procedure plan based on the workpiece identifier, and obtain the definition of all processing stages covered by the specified workpiece processing sequence; At the start of each machining stage, multiple types of sensors deployed on the gear hobbing machine spindle, tool magazine, and workpiece fixture are activated simultaneously. By configuring a unified timestamp, the system continuously collects raw readings of vibration spectrum, temperature gradient, and deformation displacement generated by various types of sensors during the processing stage, forming a full-cycle data set corresponding to the processing stage.

[0007] Preferably, the generation of a dynamic accuracy threshold sequence matching the current machining sequence based on the historical data stream of the gear hobbing machine spindle speed and tool feed specifically involves: Extract completed processing task records from the historical data warehouse that have similar process characteristics to the current processing sequence; For each completed machining task record, the time-series variation curves of the gear hobbing machine spindle speed and the tool feed are separated. The time-series variation curves of spindle speed and tool feed are coupled and segmented to divide multiple feature machining intervals; Calculate the average deviation between the actual processing accuracy and the theoretical processing accuracy within each feature processing interval, and use the average deviation as the local accuracy threshold of the feature processing interval; According to the temporal order of the current processing sequence, the selected local precision thresholds are arranged and combined to construct a dynamic precision threshold sequence.

[0008] Preferably, the step of analyzing the deviation relationship between real-time load fluctuations during the processing and the dynamic accuracy threshold sequence to plot the potential attenuation trajectory of processing accuracy specifically involves: Real-time power data from the spindle motor is received and converted into a real-time load fluctuation index that reflects changes in cutting force. In the dynamic precision threshold sequence, determine the current local precision threshold corresponding to the current processing moment; Calculate the deviation rate of the real-time load fluctuation index relative to the reference load in the processing stage, and map the deviation rate to a correction coefficient for the current local accuracy threshold. The current local precision threshold is adjusted using a correction factor to obtain the dynamic permissible precision boundary. A function that continuously records the dynamic allowable accuracy boundary as a function of time is used to depict the potential decay trajectory of machining accuracy.

[0009] Preferably, the step of associating and matching the potential attenuation trajectory of machining accuracy with multiple sets of historical process data in the process parameter database to locate the set of candidate process influence points that lead to the occurrence of the accuracy attenuation trajectory specifically involves: Multiple sets of historical process data are retrieved from the process parameter database. Each set of historical process data contains historical processing parameter configurations and their corresponding historical processing accuracy trajectories. Calculate the potential decay trajectory of machining accuracy and the morphological similarity of each historical machining accuracy trajectory at each time point; All historical processing accuracy trajectories with morphological similarity exceeding a preset threshold are selected, and their corresponding historical processing parameter configurations are extracted into associated process parameter groups. Compare the differences between the current processing parameter configuration and each associated process parameter group, and summarize all the differences into a set of candidate process influence points.

[0010] Preferably, the step of performing an independent contribution tracing analysis on each process influence point in the candidate process influence point set to identify key interference elements specifically involves: For a single process influence point in the candidate process influence point set, modify the parameter represented by the process influence point individually in the simulation environment, while keeping the remaining parameters consistent with the current processing parameter configuration; Run a machining process simulation to obtain the simulated machining accuracy trajectory after parameter modification; Calculate the degree of overlap between the simulated machining accuracy trajectory and the potential decay trajectory of machining accuracy, and define the degree of overlap as the intervention intensity value of the process influence point; Traverse all process influence points in the candidate process influence point set and calculate the intervention intensity value for each process influence point; A predetermined number of process influence points ranked first by intervention intensity value are marked as key interference elements.

[0011] Preferably, the step of structurally reconstructing the currently deployed accuracy prediction model based on key interference elements to generate an adaptive accuracy prediction model specifically involves: Read the original input layer structure of the currently deployed accuracy prediction model, which defines multiple original input variables; Check whether the process parameters corresponding to the key interference elements are all included in the original input variables; The process parameters corresponding to key interference elements that were not included in the original input variables are added as new input variables to the input layer; Establish initial weight connections between the newly added input variables and the hidden layer nodes of the accuracy prediction model, complete the structural reconstruction of the accuracy prediction model, and obtain an adaptive accuracy prediction model.

[0012] Preferably, the step of inputting the real-time collected processing status data stream into the adaptive accuracy prediction model, iteratively calculating and outputting the accuracy evolution profile within a specified future time window, specifically involves: At fixed time intervals, the latest processing status data is extracted from the full-cycle dataset to form a data slice. The data slices and the current key interference element parameter values ​​are input into the adaptive accuracy prediction model. The adaptive accuracy prediction model outputs prediction accuracy values ​​at a series of discrete time points from the current time to the end of a specified future time window; By connecting the predicted accuracy values ​​at a series of discrete time points in chronological order, an accuracy evolution profile is formed within a specified future time window.

[0013] Preferably, the method further includes optimizing the adaptive accuracy prediction model after outputting the accuracy evolution profile, the steps of which include: After the end of the specified future time window is reached, the actual machining accuracy measurement values ​​within the time window are collected; Calculate the error sequence between the actual machining accuracy measurement value and the predicted accuracy value at the corresponding time point in the accuracy evolution profile; Based on the magnitude of the error sequence, the weight values ​​corresponding to the newly added input variables in the adaptive accuracy prediction model are adjusted in reverse. The model with adjusted weights is used as the adaptive accuracy prediction model for the next round of prediction calculations.

[0014] Preferably, the parameter adjustment strategy assigned to the real-time control system of the gear hobbing machine based on the degree of conformity between the accuracy evolution profile and the machining quality target is as follows: Compare all predicted accuracy values ​​in the accuracy evolution profile with the minimum accuracy standard required by the machining quality target; If all prediction accuracy values ​​are higher than the minimum accuracy standard, then assign a strategy to the real-time control system to maintain the current parameters. If the prediction accuracy values ​​at multiple consecutive time points are lower than the minimum accuracy standard, then a targeted parameter correction strategy is assigned to the real-time control system based on the key interference factors corresponding to these time points. The parameter correction strategy includes the process parameters that need to be adjusted and suggestions on their adjustment direction and magnitude.

[0015] Compared with the prior art, the beneficial effects of the present invention are: A dynamic accuracy threshold sequence is generated based on historical data streams of gear hobbing machine spindle speed and tool feed, replacing the traditional static fixed thresholds. This transforms the accuracy assessment benchmark from a preset constant into a dynamic variable that responds in real-time to changes in the specific machining sequence. By continuously analyzing the actual operating data of the spindle and feed, it constructs an accuracy tolerance range that matches the current process load characteristics. This solves the problem of insufficient adaptability of static thresholds for different workpieces and processes, elevating the early warning mechanism from universal applicability to individual, precise adaptation, thereby reducing the probability of false alarms and missed alarms, and making the monitoring of accuracy status more realistic and sensitive.

[0016] By analyzing the deviation between real-time load fluctuations and dynamic threshold sequences, a potential accuracy degradation trajectory is plotted. This trajectory is then correlated and matched with a historical process parameter database to pinpoint a set of candidate process influence points leading to the degradation. This method represents a shift from passively receiving abnormal signals to actively tracing the causal chain of the process. It no longer views isolated exceedances of individual parameters but treats abnormal patterns in the processing as a holistic trajectory. Through rapid comparison with a historical case library, it automatically identifies several combinations of process parameters most likely triggering the current trajectory. This reduces the time required from problem discovery to root cause location, transforming experience-based fuzzy troubleshooting into precise screening based on data patterns, providing a direct and reliable basis for real-time optimization and adjustment of process parameters. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the working principle of the gear hobbing machine tool machining accuracy prediction data processing method described in this invention. Figure 2 A flowchart for receiving prediction requests and collecting data throughout the entire lifecycle; Figure 3 A flowchart illustrating the potential degradation trajectory of machining accuracy; Figure 4 A map for identifying key interference elements; Figure 5 This is a contour prediction map for accuracy evolution. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example Please see Figure 1 This invention provides a method for processing data to predict the machining accuracy of a gear hobbing machine, the method comprising: S1: Receives a hobbing accuracy prediction request from the control system, and initiates full-cycle data acquisition for the specified workpiece machining sequence based on the hobbing accuracy prediction request.

[0020] S2: Based on the historical data stream of the gear hobbing machine spindle speed and tool feed, generate a dynamic accuracy threshold sequence that matches the current machining sequence.

[0021] S3: By analyzing the relationship between real-time load fluctuations and deviations from the dynamic accuracy threshold sequence during the processing, the potential decay trajectory of processing accuracy is plotted.

[0022] S4: Match the potential decay trajectory of machining accuracy with multiple sets of historical process data in the process parameter database to locate the set of candidate process influence points that cause the accuracy decay trajectory.

[0023] S5: Perform an independent contribution source analysis on each process influence point in the candidate process influence point set to identify the key interference elements.

[0024] S6: Based on key interference factors, structurally reconstruct the currently deployed accuracy prediction model to generate an adaptive accuracy prediction model.

[0025] S7: Input the real-time acquired processing status data stream into the adaptive accuracy prediction model, iteratively calculate and output the accuracy evolution profile within a specified future time window.

[0026] S8: Based on the degree of conformity between the accuracy evolution profile and the machining quality target, assign the corresponding parameter adjustment strategy to the real-time control system of the gear hobbing machine.

[0027] See Figure 2Step S1 specifically includes: parsing the workpiece identifier and target machining accuracy level contained in the hobbing machining accuracy prediction request. Retrieving the corresponding machining process plan based on the workpiece identifier to obtain the definitions of all machining stages covered by the specified workpiece machining sequence. At the start of each machining stage, simultaneously activating multiple types of sensors deployed on the hobbing machine spindle, tool magazine, and workpiece fixture. Configuring a unified timestamp, continuously collecting the raw readings of vibration spectrum, temperature gradient, and deformation displacement generated by the multiple types of sensors during the machining stage, forming a full-cycle data set corresponding to the machining stage.

[0028] Furthermore, step S2 specifically includes: extracting completed machining task records from the historical data warehouse that have similar process characteristics to the current machining sequence. For each completed machining task record, separating the time-series variation curves of the hobbing machine spindle speed and tool feed is performed. The time-series variation curves of the spindle speed and tool feed are coupled and segmented to divide multiple characteristic machining intervals. The average deviation between the actual machining accuracy and the theoretical machining accuracy within each characteristic machining interval is calculated, and this average deviation is used as the local accuracy threshold of the characteristic machining interval. According to the time sequence of the current machining sequence, the selected multiple local accuracy thresholds are arranged and combined to construct a dynamic accuracy threshold sequence.

[0029] In practice, the gear hobbing accuracy prediction request is issued by the control system. This request includes a workpiece identifier and a target machining accuracy level. After parsing the workpiece identifier "Gear-001" and the target machining accuracy level IT6, the corresponding machining process plan is retrieved based on the workpiece identifier "Gear-001." This retrieves the definitions of all machining stages covered by the specified workpiece machining sequence, which includes roughing, semi-finishing, and finishing stages. At the start of each machining stage, multiple types of sensors deployed on the gear hobbing machine spindle, tool magazine, and workpiece fixture are simultaneously activated. These sensors include vibration sensors, temperature sensors, and displacement sensors. A unified timestamp is configured, and the raw readings of vibration spectra, temperature gradients, and deformation displacements generated by these sensors during the machining stage are continuously collected, forming a full-cycle data set corresponding to each machining stage. For example, the full-cycle data set for the roughing stage includes a sequence of vibration spectrum amplitude values, a temperature gradient change curve, and a sequence of measured deformation displacement values.

[0030] In practice, completed machining task records with similar process characteristics to the current machining sequence are extracted from the historical data warehouse. Process characteristic similarity is matched based on workpiece material, gear module, and number of teeth. For example, if the current machining sequence is a steel gear with a module of 3 and 30 teeth, machining records of steel gears with a module of 3 and a number of teeth between 25 and 35 are extracted from the historical data warehouse. For each completed machining task record, the time-series variation curves of the hobbing machine spindle speed and tool feed are separated. The spindle speed variation curve changes over time in revolutions per minute, while the tool feed variation curve changes over time in millimeters per revolution. The time-series variation curves of the spindle speed and tool feed are coupled and segmented to divide multiple characteristic machining intervals. The coupling segmentation is determined based on the co-current abrupt change point of the spindle speed change rate and the tool feed rate change rate. For example, the boundary of the characteristic machining interval is defined when the spindle speed change rate exceeds 5 revolutions per second and the tool feed rate change rate exceeds 0.1 millimeters per second.

[0031] In some embodiments, the average deviation between the actual machining accuracy and the theoretical machining accuracy within each feature machining interval is calculated, and the average deviation is used as the local accuracy threshold for the feature machining interval. The average deviation is calculated using the following formula:

[0032] in: This represents the average deviation. This indicates the number of sampling points within the feature processing interval. This represents the actual machining accuracy value of the k-th sampling point. This represents the theoretical machining accuracy value at the k-th sampling point. The actual machining accuracy value comes from the tooth pitch deviation value collected by the online measurement system in the historical machining task records, while the theoretical machining accuracy value is calculated based on the machine tool kinematic model and tool geometric parameters. Following the temporal order of the current machining sequence, multiple selected local accuracy thresholds are arranged and combined to construct a dynamic accuracy threshold sequence. For example, if the current machining sequence includes three feature machining intervals, the dynamic accuracy threshold sequence consists of three local accuracy thresholds arranged in chronological order. ".

[0033] In some embodiments, the raw readings of vibration spectrum, temperature gradient, and deformation displacement are preprocessed before aggregation, including the removal of anomalous noise and unit standardization. Optionally, the completed machining task records in the historical data warehouse also include tool wear status indicators, which are weighted and adjusted when calculating the average deviation. Optionally, the segmentation results of the feature machining intervals are manually verified through a visualization interface to ensure the accuracy of the coupled segmentation. It is understood that the full-cycle dataset provides a multi-dimensional sensor data foundation for subsequent analysis, and the dynamic accuracy threshold sequence reflects the accuracy fluctuation pattern in historical machining. It is understood that the coupled segmentation of the time-series variation curves of spindle speed and tool feed can capture key transient stages in the machining process.

[0034] See Figure 3 Step S3 specifically includes: receiving instantaneous power data from the spindle motor in real time and converting the instantaneous power data into a real-time load fluctuation index reflecting changes in cutting force. Determining the current local accuracy threshold corresponding to the current machining moment from the dynamic accuracy threshold sequence. Calculating the offset rate of the real-time load fluctuation index relative to the reference load of the machining stage, and mapping this offset rate to a correction coefficient for the current local accuracy threshold. Adjusting the current local accuracy threshold using the correction coefficient to obtain the dynamic permissible accuracy boundary. Continuously recording the function of the dynamic permissible accuracy boundary changing over time, and depicting this function as a potential decay trajectory of machining accuracy.

[0035] Furthermore, step S4 specifically includes: retrieving multiple sets of historical process data from the process parameter database, each set of historical process data containing historical processing parameter configurations and their corresponding historical processing accuracy trajectories. Calculating the morphological similarity between the potential decay trajectory of processing accuracy and each historical processing accuracy trajectory at various time points. Filtering out all historical processing accuracy trajectories with morphological similarity exceeding a preset threshold, and extracting their corresponding historical processing parameter configurations as associated process parameter groups. Comparing the differences between the current processing parameter configuration and each associated process parameter group, and summarizing all differences into a candidate process influence point set.

[0036] In practical implementation, instantaneous power data from the spindle motor is received in real time, collected once per second. For example, the instantaneous power data sequence collected at a certain machining moment might be 5.1kW, 5.3kW, and 5.8kW. This instantaneous power data is converted into a real-time load fluctuation index reflecting changes in cutting force. The real-time load fluctuation index is obtained by calculating the standard deviation of the instantaneous power data within the sliding time window. In the dynamic accuracy threshold sequence, the current local accuracy threshold corresponding to the current machining moment is determined. The dynamic accuracy threshold sequence is predefined based on the machining time axis. The current machining moment falls within the second characteristic machining interval, therefore, the current local accuracy threshold is the local accuracy threshold Q corresponding to this interval. The offset rate of the real-time load fluctuation index relative to the machining stage reference load is calculated. The machining stage reference load is the average instantaneous power during a stable period after the start of the current machining stage. The offset rate is calculated using the formula:

[0037] in: Indicates the offset rate. This represents the real-time load fluctuation indicator at the current moment. This represents the reference load during the processing stage. The offset rate is mapped to a correction coefficient for the current local accuracy threshold; the mapping relationship is a preset piecewise linear function, and the correction coefficient is denoted as... Adjusting the current local precision threshold using correction factors yields the dynamic permissible precision boundary. The calculation formula is A function that continuously records the dynamic allowable accuracy boundary changes over time is plotted as a potential decay trajectory of machining accuracy. This trajectory is represented as a curve that fluctuates over time, where the ordinate of each point represents the allowable accuracy boundary value at that moment.

[0038] In practice, multiple sets of historical process data are retrieved from the process parameter database. This database stores records of all machining tasks from the past year. Each set of historical process data includes historical machining parameter configurations and their corresponding historical machining accuracy trajectories. The historical machining parameter configurations include cutting speed, feed rate, depth of cut, and coolant flow rate. The historical machining accuracy trajectory is a sequence showing the change in actual measured accuracy over time in historical tasks. The morphological similarity between the potential decay trajectory of machining accuracy and each historical machining accuracy trajectory at each time point is calculated. Morphological similarity is calculated using a dynamic time warping algorithm, which measures the minimum cumulative distance between two sequences after considering time scaling and converts the reciprocal of the distance into a similarity value between 0 and 1. All historical machining accuracy trajectories with morphological similarity exceeding a preset threshold (set to 0.85) are selected and marked as highly similar trajectories. Their corresponding historical machining parameter configurations are extracted as associated process parameter groups. For example, if three highly similar historical machining accuracy trajectories are selected, three associated process parameter groups are extracted accordingly. Compare the differences between the current machining parameter configuration and each associated process parameter group. In the current machining parameter configuration, the cutting speed is 120 m / min, while in an associated process parameter group, the cutting speed is 115 m / min. Therefore, the cutting speed is a difference. All the differences are summarized into a set of candidate process influence points. The set of candidate process influence points may include parameter names such as {cutting speed, feed rate, coolant flow rate}.

[0039] In some embodiments, the mapping strategy from offset rate to correction coefficient is calibrated based on experimental data, with different offset rate ranges corresponding to different linear mapping slopes. In some embodiments, the calculation of morphological similarity considers not only the trajectory shape but also the similarity of the first-order difference sequence of the trajectory to capture commonalities in changing trends. Optionally, the reference load during the machining stage is not a fixed value but a rolling average that is dynamically updated based on the tool wear state. Optionally, the preset threshold used when filtering historical machining accuracy trajectories can be adaptively adjusted according to the target machining accuracy level; the higher the accuracy level, the stricter the threshold setting. It can be understood that the combination of real-time load fluctuation indicators and dynamic accuracy threshold sequences enables dynamic adaptive adjustment of machining allowable boundaries. It can be understood that by matching historical trajectories with morphological similarity, the potential range of process parameters that may lead to the current accuracy decay mode can be quickly located.

[0040] Step S5 specifically includes: For a single process influence point in the candidate process influence point set, modify the parameter represented by that process influence point individually in the simulation environment, keeping the remaining parameters consistent with the current processing parameter configuration. Run the processing simulation to obtain the simulated processing accuracy trajectory after parameter modification. Calculate the overlap between the simulated processing accuracy trajectory and the potential attenuation trajectory of processing accuracy, and define this overlap as the intervention intensity value of the process influence point. Traverse all process influence points in the candidate process influence point set and calculate the intervention intensity value for each process influence point. Mark a predetermined number of process influence points with the highest intervention intensity values ​​as key interference elements.

[0041] Step S6 specifically includes: reading the original input layer structure of the currently deployed accuracy prediction model, which defines multiple original input variables; checking whether the process parameters corresponding to the key interference elements are all included in the original input variables; adding the process parameters corresponding to the key interference elements that are not included in the original input variables as new input variables to the input layer; establishing the initial weight connection relationship between the new input variables and the hidden layer nodes of the accuracy prediction model, completing the structural reconstruction of the accuracy prediction model, and obtaining an adaptive accuracy prediction model.

[0042] In practical implementation, for a single process influence point in the candidate process influence point set, such as the "coolant flow rate" parameter, the value represented by the "coolant flow rate" parameter is modified individually in the simulation environment. The coolant flow rate is changed from 15 liters per minute in the current machining parameter configuration to 10 liters per minute, while keeping the other parameters consistent with the current machining parameter configuration. These other parameters include cutting speed, feed rate, and depth of cut. The machining process simulation is run, with the simulation environment built based on the machine tool dynamics model and cutting force model. The simulated machining accuracy trajectory after parameter modification is obtained. This trajectory consists of simulated tooth profile error values ​​at a series of simulation time points. The overlap between the simulated machining accuracy trajectory and the potential decay trajectory of machining accuracy is calculated. The overlap is obtained by calculating the correlation coefficient between the two trajectory sequences at corresponding time periods. The correlation coefficient is defined as the intervention intensity value of the process influence point "coolant flow rate," denoted as [insert value here]. The formula for its calculation is:

[0043] in: Indicates the intervention intensity value. This indicates the number of data points in the trajectory sequence. The simulated machining accuracy trajectory is represented in the first... Precision value of each point. This represents the average value of all points on the simulated machining accuracy trajectory. The potential decay trajectory representing machining accuracy is in the first... Precision boundary values ​​for each point. This represents the average value of all points on the potential decay trajectory of machining accuracy.

[0044] Traverse all process influence points in the candidate process influence point set, which includes "cutting speed", "feed rate", and "coolant flow rate". Repeat the above simulation and calculation process for each process influence point to calculate the intervention intensity value of each process influence point. For example, obtain the intervention intensity value of the cutting speed. The intervention intensity value of the feed rate is 0.92. The interference intensity value of coolant flow is 0.87. The value is 0.95. A preset number of process influence points with the highest intervention intensity values ​​are marked as key interference elements. The preset number is set to 2. Therefore, the top two process influence points with the highest intervention intensity values ​​are "coolant flow rate" (…). ) and "cutting speed" ) was marked as a key interference element.

[0045] In the specific implementation, the original input layer structure of the currently deployed accuracy prediction model is read. The original input layer structure defines multiple original input variables, including spindle vibration amplitude, spindle temperature, and workpiece fixture displacement. It is checked whether the process parameters corresponding to the key interference elements are all included in the original input variables. The key interference elements "coolant flow rate" and "cutting speed" are process parameters. The check reveals that the original input variables only include spindle vibration amplitude, spindle temperature, and workpiece fixture displacement, but not "coolant flow rate" and "cutting speed". The process parameters corresponding to the key interference elements not included in the original input variables are added as new input variables to the input layer, namely "coolant flow rate" and "cutting speed". An initial weight connection relationship is established between the new input variables and the hidden layer nodes of the accuracy prediction model. The initial weight is set to a random decimal close to zero, for example, uniformly distributed in the range of -0.01 to 0.01. This completes the structural reconstruction of the accuracy prediction model, resulting in an adaptive accuracy prediction model.

[0046] In some embodiments, when calculating the intervention intensity value, the overlap can also be measured by the reciprocal of the dynamic time-warped distance between the two trajectory sequences, and then normalized. In some embodiments, the preset number is determined based on the distribution of intervention intensity values ​​of key interference elements, and the elbow rule is used to automatically determine the sorting cutoff point. Optionally, when modifying the process influence point parameter in the simulation environment, the modification magnitude is set based on the historical operating range of the parameter, for example, selecting a modification value within ±20% of the parameter's standard value. Optionally, the initial weight connection relationship between the newly added input variable and the hidden layer node can also be initialized by assigning values ​​to the linear regression coefficients of the parameter and the accuracy value in historical data. It can be understood that through independent contribution source analysis, the degree of simulation impact of each process parameter change on the current accuracy decay mode can be quantified. It can be understood that structurally reconstructing the model input layer based on key interference elements enables the accuracy prediction model to directly perceive and respond to major process disturbance factors.

[0047] See Figure 4 This chart presents the results of an analysis of the impact of various process parameters on precision degradation during gear hobbing. The different process influence points are arranged horizontally according to their impact strength values, with key interference elements highlighted in bright red and general process influence points represented in cyan. Each bar is labeled with its specific impact strength value, reflecting the correlation between the change in that process parameter and the trajectory of machining precision degradation. This visualization method allows engineers to quickly identify the key process parameters that have the most significant impact on machining precision stability, providing a clear direction for subsequent model reconstruction and process optimization. The overall chart design is professional and well-organized, using color differentiation and numerical labeling to intuitively present complex analysis results.

[0048] Step S7 specifically includes: extracting the latest processing status data from the full-cycle dataset at fixed time intervals to form a data slice. This data slice, along with the current key interference element parameter values, is then input into the adaptive accuracy prediction model. The adaptive accuracy prediction model outputs predicted accuracy values ​​at a series of discrete time points from the current time to the end of a specified future time window. These predicted accuracy values ​​are then connected chronologically to form the accuracy evolution profile within the specified future time window.

[0049] Step S7 further includes optimizing the adaptive accuracy prediction model after outputting the accuracy evolution profile. Specifically, after reaching the end of a specified future time window, the actual machining accuracy measurements within that time window are collected. The error sequence between the actual machining accuracy measurements and the predicted accuracy values ​​at the corresponding time points in the accuracy evolution profile is calculated. Based on the magnitude of the error sequence, the weight values ​​corresponding to the newly added input variables in the adaptive accuracy prediction model are adjusted in reverse. The model with adjusted weights is used as the adaptive accuracy prediction model for the next round of prediction calculations.

[0050] In practice, the latest processing status data for a specific period is extracted from the full-cycle dataset at fixed time intervals, with each interval set to 5 seconds. The full-cycle dataset is continuously updated. At time T, readings of the vibration spectrum, temperature gradient, and deformation displacement from time T-20 seconds to time T are extracted from the full-cycle dataset, forming a data slice. This data slice is a multi-dimensional time series matrix. The data slice, along with the key interference element parameter values ​​at the current time, are input into the adaptive accuracy prediction model. The key interference element parameter values ​​include coolant flow rate and cutting speed, for example, a coolant flow rate of 10 liters per minute and a cutting speed of 120 meters per minute. The adaptive accuracy prediction model outputs prediction accuracy values ​​at a series of discrete time points from the current time T to the end of a future specified time window. The future specified time window is 15 seconds long, with discrete time points at T+5 seconds, T+10 seconds, and T+15 seconds, respectively. The output prediction accuracy values ​​are 8.2 micrometers, 8.9 micrometers, and 9.7 micrometers. By connecting a series of prediction accuracy values ​​in chronological order, an accuracy evolution profile is formed within a specified future time window. The accuracy evolution profile is represented as a curve pointing from the current moment to the future, with the accuracy value (error) gradually increasing.

[0051] In practice, after the end of the specified future time window, the actual machining accuracy measurements within that time window are collected. These measurements are obtained using an online measurement probe at three times: T+5 seconds, T+10 seconds, and T+15 seconds. An error sequence is calculated between the actual machining accuracy measurements and the predicted accuracy values ​​at corresponding time points in the accuracy evolution profile. This error sequence is the set of differences between the actual and predicted values ​​at each corresponding time point. Based on the magnitude of the error sequence, the weight values ​​corresponding to the newly added input variables in the adaptive accuracy prediction model are adjusted in reverse. This adjustment is based on the error backpropagation principle. Before calculating the adjustment, the error and input values ​​are normalized to eliminate the influence of physical dimensions. The weight adjustment is calculated using the following formula:

[0052] in: This indicates the addition of a new input variable node. To hidden layer nodes The change in the connection weights that needs to be adjusted is a dimensionless coefficient adjustment value. This represents the preset learning rate, which is a positive decimal less than 1. M represents the number of data points in the error sequence; in this example, M is 3. Indicates the first error in the error sequence. The error value at each time point has the dimension of length. The reference quantity representing the error is a constant with the same length dimension. Indicates the first At each time point, add a new input variable node. The input value, the dimension of which depends on the parameter type, This indicates the corresponding newly added input variable node. The input reference is a constant with the same dimensions. The model with adjusted weights is used as the adaptive accuracy prediction model for the next round of prediction calculations, which will begin at time T+5 seconds using the model with updated weight parameters.

[0053] In some embodiments, the reference value of the error Select the lowest accuracy standard value required for the processing quality target. In some embodiments, add an input variable node. Input reference quantity Select the setpoint or rated value for this process parameter in the processing sequence. Optionally, the length of the specified future time window is proportional to the remaining time of the current processing stage, but a maximum upper limit is set. Optionally, the learning rate... After setting the initial values, the model can be dynamically fine-tuned based on the improvement rate of prediction accuracy after the most recent weight adjustments. This can be understood as normalizing the error and input values, enabling dimensionless weighted calculations of quantities with different physical meanings, thus ensuring the consistency of the formula's dimensions. Furthermore, adjusting the model weights in reverse based on the normalized error sequence achieves adaptive learning and progressive optimization of the prediction model during operation (see Table 1).

[0054] Table 1: Comparison of Predicted and Actual Measurements within Future Time Window

[0055] See Figure 5This diagram illustrates a comparative analysis of the predicted and actual measured values ​​of machining accuracy evolution of a gear hobbing machine within a future time window. The solid blue line represents the predicted accuracy evolution trend, connecting the predicted accuracy values ​​at three future time points; the purple dots represent the actual measured accuracy values ​​at the corresponding time points; and the orange dashed line marks the minimum accuracy standard line for machining quality requirements. The gray connecting lines between the predicted trend line and the actual measurement points visually indicate the magnitude and direction of the prediction error, while the semi-transparent blue area represents the confidence interval of the prediction results. This dynamic prediction visualization helps operators identify potentially dangerous periods where accuracy may exceed limits, providing decision support for real-time parameter adjustments and demonstrating the practical application value of predictive maintenance in intelligent manufacturing.

[0056] Step S8 specifically includes: comparing all predicted accuracy values ​​in the accuracy evolution profile with the minimum accuracy standard required by the machining quality target. If all predicted accuracy values ​​are higher than the minimum accuracy standard, a strategy to maintain the current parameters is assigned to the real-time control system. If there are multiple consecutive time points where the predicted accuracy values ​​are lower than the minimum accuracy standard, a targeted parameter correction strategy is assigned to the real-time control system based on the key interference elements corresponding to these time points. The parameter correction strategy includes the process parameters that need to be adjusted and suggestions on their adjustment direction and magnitude.

[0057] In practice, the degree of conformity between the accuracy evolution profile and the machining quality target is determined through a comparison operation. The accuracy evolution profile includes predicted accuracy values ​​at multiple future time points from the current moment. The minimum accuracy standard required by the machining quality target is a clearly defined numerical threshold, such as a tooth pitch error not exceeding 8 micrometers. All predicted accuracy values ​​in the accuracy evolution profile are compared with the minimum accuracy standard required by the machining quality target. The predicted accuracy values ​​at T+5 seconds, T+10 seconds, and T+15 seconds are 7.8 micrometers, 8.2 micrometers, and 7.9 micrometers, respectively, while the minimum accuracy standard is 8 micrometers. If all predicted accuracy values ​​are higher than the minimum accuracy standard, a strategy to maintain the current parameters is assigned to the real-time control system. Maintaining the current parameters means not changing any currently executing process parameter settings. If the prediction accuracy values ​​at multiple consecutive time points are lower than the minimum accuracy standard, then a targeted parameter correction strategy is assigned to the real-time control system based on the key interference elements corresponding to these time points. For example, if the prediction accuracy values ​​at three consecutive time points T+60 seconds, T+120 seconds, and T+180 seconds are 8.3 micrometers, 8.5 micrometers, and 8.7 micrometers, respectively, all of which exceed the minimum accuracy standard of 8 micrometers, the parameter correction strategy includes the process parameters that need to be adjusted and suggestions on their adjustment direction and magnitude.

[0058] In practical implementation, the generation logic of targeted parameter correction strategies is based on mapping rules, which define the correspondence between key interference elements, prediction deviation patterns, and parameter adjustment suggestions. The cumulative deviation between the prediction accuracy value corresponding to consecutive time points below the standard and the minimum accuracy standard is calculated. The calculation formula is:

[0059] in: This represents the cumulative deviation. This represents the set of time points where the predicted accuracy value is continuously lower than the minimum accuracy standard. Indicates at a point in time The prediction accuracy value, This indicates the minimum accuracy standard. Based on the key interference element and cumulative deviation, a preset strategy mapping table is consulted. The strategy mapping table stores the suggested parameter adjustment direction and magnitude for different key interference elements under different deviation ranges. For example, if the key interference element is "cutting speed" and the cumulative deviation... Within the 1.0 to 2.0 microsecond range, the parameter correction strategy recommends reducing the cutting speed by 5 meters per minute. The parameter correction strategy assigned to the real-time control system is issued in the form of structured instructions, which explicitly include the process parameter identifier, adjustment direction, and specific magnitude value.

[0060] In some embodiments, a small negative tolerance band is set when determining whether the predicted accuracy value is "below" the minimum accuracy standard. For example, the "below" judgment is only triggered when the predicted accuracy value is 0.1 micrometers or more above the minimum accuracy standard, in order to avoid frequent strategy switching near the critical point. In some embodiments, the adjustment range of the parameter correction strategy is not a fixed value, but is positively correlated with the slope of the predicted accuracy decay curve; the larger the slope, the larger the recommended adjustment range. Optionally, when multiple key interference factors exist, the generated parameter correction strategy may include coordinated adjustment suggestions for multiple process parameters, with priority indicators attached to the adjustment suggestions. Optionally, when assigning a parameter correction strategy to the real-time control system, a log record containing the decision basis is generated, including the predicted accuracy point that triggered the judgment, the key interference factors used, the calculated cumulative deviation, and the details of the finally assigned strategy. It can be understood that comparing the accuracy evolution profile with a clear quality standard provides a clear basis for control decisions. It can be understood that generating specific adjustment suggestions based on key interference factors and quantified deviations makes the parameter adjustment of the control system more targeted and operable.

[0061] It should be noted that, although embodiments of the present invention have been shown and described herein, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for processing data to predict machining accuracy on a gear hobbing machine, characterized in that, Perform the following steps in sequence: Receive a hobbing accuracy prediction request from the control system, and initiate full-cycle data acquisition for the specified workpiece machining sequence based on the hobbing accuracy prediction request. Based on the historical data stream of the spindle speed and tool feed of the gear hobbing machine, a dynamic accuracy threshold sequence is generated that matches the current machining sequence; By analyzing the deviation relationship between real-time load fluctuations during the processing and the dynamic accuracy threshold sequence, a potential attenuation trajectory of processing accuracy is plotted. The potential trajectory of machining accuracy decay is correlated and matched with multiple sets of historical process data in the process parameter database to locate the set of candidate process influence points that lead to the occurrence of accuracy decay trajectory. Perform an independent contribution source analysis on each process influence point in the candidate process influence point set to identify the key interference elements; Based on key interference factors, the currently deployed accuracy prediction model is structurally reconstructed to generate an adaptive accuracy prediction model; The real-time acquired processing status data stream is input into the adaptive accuracy prediction model, which iteratively calculates and outputs the accuracy evolution profile within a specified future time window. Based on the degree of conformity between the accuracy evolution profile and the machining quality target, corresponding parameter adjustment strategies are assigned to the real-time control system of the gear hobbing machine.

2. The method for predicting machining accuracy on a gear hobbing machine tool according to claim 1, characterized in that, The step of receiving a hobbing accuracy prediction request from the control system and initiating full-cycle data acquisition for the specified workpiece machining sequence based on the hobbing accuracy prediction request specifically involves: Analyze the workpiece identifier and target machining accuracy level contained in the gear hobbing accuracy prediction request; Retrieve the corresponding processing procedure plan based on the workpiece identifier, and obtain the definition of all processing stages covered by the specified workpiece processing sequence; At the start of each machining stage, multiple types of sensors deployed on the gear hobbing machine spindle, tool magazine, and workpiece fixture are activated simultaneously. By configuring a unified timestamp, the system continuously collects raw readings of vibration spectrum, temperature gradient, and deformation displacement generated by various types of sensors during the processing stage, forming a full-cycle data set corresponding to the processing stage.

3. The method for predicting machining accuracy on a gear hobbing machine tool according to claim 2, characterized in that, The historical data stream of the gear hobbing machine spindle speed and tool feed is used to generate a dynamic accuracy threshold sequence that matches the current machining sequence, specifically: Extract completed processing task records from the historical data warehouse that have similar process characteristics to the current processing sequence; For each completed machining task record, the time-series variation curves of the gear hobbing machine spindle speed and the tool feed are separated. The time-series variation curves of spindle speed and tool feed are coupled and segmented to divide multiple feature machining intervals; Calculate the average deviation between the actual processing accuracy and the theoretical processing accuracy within each feature processing interval, and use the average deviation as the local accuracy threshold of the feature processing interval; According to the temporal order of the current processing sequence, the selected local precision thresholds are arranged and combined to construct a dynamic precision threshold sequence.

4. The method for predicting machining accuracy on a gear hobbing machine tool according to claim 3, characterized in that, The process involves analyzing the deviation between real-time load fluctuations during processing and the dynamic accuracy threshold sequence to plot a potential attenuation trajectory of processing accuracy. Specifically: Real-time power data from the spindle motor is received and converted into a real-time load fluctuation index that reflects changes in cutting force. In the dynamic precision threshold sequence, determine the current local precision threshold corresponding to the current processing moment; Calculate the deviation rate of the real-time load fluctuation index relative to the reference load in the processing stage, and map the deviation rate to a correction coefficient for the current local accuracy threshold. The current local precision threshold is adjusted using a correction factor to obtain the dynamic permissible precision boundary. A function that continuously records the dynamic allowable accuracy boundary as a function of time is used to depict the potential decay trajectory of machining accuracy.

5. The method for predicting machining accuracy on a gear hobbing machine tool according to claim 4, characterized in that, The process involves associating and matching the potential attenuation trajectory of machining accuracy with multiple sets of historical process data in the process parameter database to locate the set of candidate process influence points that lead to the occurrence of the accuracy attenuation trajectory. Specifically: Multiple sets of historical process data are retrieved from the process parameter database. Each set of historical process data contains historical processing parameter configurations and their corresponding historical processing accuracy trajectories. Calculate the potential decay trajectory of machining accuracy and the morphological similarity of each historical machining accuracy trajectory at each time point; All historical processing accuracy trajectories with morphological similarity exceeding a preset threshold are selected, and their corresponding historical processing parameter configurations are extracted into associated process parameter groups. Compare the differences between the current processing parameter configuration and each associated process parameter group, and summarize all the differences into a set of candidate process influence points.

6. The method for predicting machining accuracy on a gear hobbing machine tool according to claim 5, characterized in that, The method involves performing an independent contribution tracing analysis on each process influence point in the candidate process influence point set to identify key interference elements, specifically: For a single process influence point in the candidate process influence point set, modify the parameter represented by the process influence point individually in the simulation environment, while keeping the remaining parameters consistent with the current processing parameter configuration; Run a machining process simulation to obtain the simulated machining accuracy trajectory after parameter modification; Calculate the degree of overlap between the simulated machining accuracy trajectory and the potential decay trajectory of machining accuracy, and define the degree of overlap as the intervention intensity value of the process influence point; Traverse all process influence points in the candidate process influence point set and calculate the intervention intensity value for each process influence point; A predetermined number of process influence points ranked first by intervention intensity value are marked as key interference elements.

7. The method for predicting machining accuracy on a gear hobbing machine tool according to claim 6, characterized in that, The process involves structurally reconstructing the currently deployed accuracy prediction model based on key interference elements to generate an adaptive accuracy prediction model, specifically as follows: Read the original input layer structure of the currently deployed accuracy prediction model, which defines multiple original input variables; Check whether the process parameters corresponding to the key interference elements are all included in the original input variables; The process parameters corresponding to key interference elements that were not included in the original input variables are added as new input variables to the input layer; Establish initial weight connections between the newly added input variables and the hidden layer nodes of the accuracy prediction model, complete the structural reconstruction of the accuracy prediction model, and obtain an adaptive accuracy prediction model.

8. The method for predicting machining accuracy on a gear hobbing machine tool according to claim 7, characterized in that, The process involves inputting the real-time collected processing status data stream into an adaptive accuracy prediction model, iteratively calculating and outputting the accuracy evolution profile within a specified future time window. At fixed time intervals, the latest processing status data is extracted from the full-cycle dataset to form a data slice. The data slices and the current key interference element parameter values ​​are input into the adaptive accuracy prediction model. The adaptive accuracy prediction model outputs prediction accuracy values ​​at a series of discrete time points from the current time to the end of a specified future time window; By connecting the predicted accuracy values ​​at a series of discrete time points in chronological order, an accuracy evolution profile is formed within a specified future time window.

9. The method for predicting machining accuracy on a gear hobbing machine tool according to claim 8, characterized in that, The method further includes optimizing the adaptive accuracy prediction model after outputting the accuracy evolution profile, the steps of which include: After the end of the specified future time window is reached, the actual machining accuracy measurement values ​​within the time window are collected; Calculate the error sequence between the actual machining accuracy measurement value and the predicted accuracy value at the corresponding time point in the accuracy evolution profile; Based on the magnitude of the error sequence, the weight values ​​corresponding to the newly added input variables in the adaptive accuracy prediction model are adjusted in reverse. The model with adjusted weights is used as the adaptive accuracy prediction model for the next round of prediction calculations.

10. The method for predicting machining accuracy on a gear hobbing machine tool according to claim 1, characterized in that, Based on the degree of conformity between the accuracy evolution profile and the machining quality target, the real-time control system of the gear hobbing machine is assigned a corresponding parameter adjustment strategy, specifically: Compare all predicted accuracy values ​​in the accuracy evolution profile with the minimum accuracy standard required by the machining quality target; If all prediction accuracy values ​​are higher than the minimum accuracy standard, then assign a strategy to the real-time control system to maintain the current parameters. If the prediction accuracy values ​​at multiple consecutive time points are lower than the minimum accuracy standard, then a targeted parameter correction strategy is assigned to the real-time control system based on the key interference factors corresponding to these time points. The parameter correction strategy includes the process parameters that need to be adjusted and suggestions on their adjustment direction and magnitude.

Images