An optical lens grinding process monitoring method and system based on digital twinning

Abstract

This invention belongs to the field of data processing technology, specifically relating to a method and system for monitoring the optical lens grinding process based on digital twins. The method includes: calculating the material removal rate based on the difference in lens volume at each moment and the sampling period; obtaining the dynamic window length based on the material removal rate, spindle torque, and rated maximum torque, and calculating the variance and vibration amplitude of the spindle torque sequence and triaxial acceleration sequence accordingly; obtaining the standard removal rate, standard discrete variance, and standard amplitude to calculate the judgment scale factor; obtaining the deviation value at the current moment based on the difference between the material removal rate and the theoretical removal rate and calculating the absolute deviation, and calculating the dynamic neighborhood radius in conjunction with the judgment scale factor; constructing a three-dimensional vector for each moment using the standard removal rate, standard discrete variance, and standard amplitude; and using the dynamic neighborhood radius to improve the density clustering algorithm for anomaly detection. This invention improves the anomaly detection rate in the lens grinding process.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. More specifically, this invention relates to a method and system for monitoring the optical lens grinding process based on digital twins. Background Technology

[0002] In the manufacturing of precision optical components, grinding process monitoring is a key technology for ensuring lens surface quality and processing efficiency. Its core task is to track the material removal status, grinding head wear, and machine tool vibration characteristics in real time during the grinding process. This aims to determine the processing progress and identify physical defects such as chipping and scratches, thereby ensuring proper processing alignment and reducing scrap rates.

[0003] Existing grinding process monitoring methods typically employ theoretical cutting parameter estimation or signal interception based on conventional observation windows. These methods primarily rely on static deviation comparison strategies of the original sensor signals and determine whether the current processing status is abnormal by using preset empirical cutting rates or safety envelopes.

[0004] However, in actual machine tool monitoring scenarios, the physical conditions of grinding are not constant. Instead, they drift with the cumulative wear of the tool, and generate energy fluctuations and random noise at different processing stages. When the data fluctuates significantly due to impact during the rough grinding stage, these unstable physical noises can mask the true cutting characteristics. Traditional methods cannot dynamically identify and adjust the noise reduction scale of the data, resulting in the monitoring results failing to accurately reflect the true material peeling progress.

[0005] Secondly, in the absence of a boundary adaptive mechanism, baseline drift caused by normal tool wear may be misjudged due to similar local values, leading to a loose monitoring and early warning logic. Traditional static comparison strategies ignore the natural contraction of physical vibration energy as material removal slows down, which can easily amplify minute machine tool servo noise as an anomaly in the extremely stable stage at the end of processing, or miss high-frequency weak defects, ultimately reducing the reliability and statistical effectiveness of the full-cycle monitoring and early warning system for grinding. Summary of the Invention

[0006] To address the technical problems in existing monitoring solutions, such as inaccurate physical progress due to tool wear, feature extraction failure caused by static data windows, and the tendency for fixed judgment criteria to generate false alarms under baseline drift, this invention provides solutions in the following aspects.

[0007] In a first aspect, the present invention provides a method for monitoring the optical lens grinding process based on digital twins, comprising: acquiring a triaxial acceleration sequence and a spindle torque sequence; obtaining the difference in lens volume between the current moment and the previous moment, and combining the sampling period to obtain the material removal rate and normalize it to a standard removal rate; obtaining the dynamic window length at each moment based on the material removal rate, spindle torque, and rated maximum torque; using the normalized variance of the spindle torque sequence within the dynamic window at each moment as the standard discrete variance at each moment; calculating the vibration amplitude based on the triaxial acceleration sequence within the dynamic window and normalizing it to obtain the standard amplitude at each moment; and calculating the standard discrete variance, standard amplitude, and standard deviation at each moment. The removal rate is used to obtain the judgment scale factor at each time step; the theoretical removal rate is obtained based on the difference between the rated lens volume and the target lens volume combined with the planned processing time; the deviation value at each time step is obtained based on the difference between the material removal rate and the theoretical removal rate at each time step, and the absolute deviation at each time step is the absolute value of the difference between the deviation value at each time step and the previous time step; the dynamic neighborhood radius at each time step is obtained based on the judgment scale factor, absolute deviation, and reference radius at each time step; the standard removal rate, standard variance, and standard amplitude at each time step are constructed into a three-dimensional vector at each time step; the density clustering algorithm is improved using the dynamic neighborhood radius to cluster the three-dimensional vectors at each time step, thereby realizing anomaly detection in the grinding process.

[0008] This invention performs adaptive noise reduction on the spindle torque sequence and triaxial acceleration sequence by integrating the material removal rate and dynamic window length. It generates a judgment scale factor by combining the standard discrete variance, standard amplitude and standard removal rate, and uses the theoretically predicted absolute deviation to correct the dynamic neighborhood radius in the feature space. It measures the cutting peel thickness and abnormal energy distribution of optical lens grinding under variable transient conditions, as well as the tolerance span of the clustering algorithm search boundary for load pulsation and cumulative tool wear. It provides a basis for defining the real-time physical progress of the rough and fine grinding stages and distinguishing between natural wear and instantaneous chipping. It suppresses the false alarm caused by the continuous drift of the processing reference and the switching of working conditions caused by the static intercept window and fixed deviation threshold. It also reduces the feature clustering error caused by the energy contraction of physical vibration at the end of fine grinding.

[0009] Preferably, the material removal rate satisfies the following relationship: ; In the formula, For a moment Material removal rate; For a moment The volume of the lens; For a moment The volume of the lens; The sampling period.

[0010] This invention calculates the volume difference between the lens volume at the current moment and the lens volume at the previous moment, and integrates the volume difference with the sampling period to calculate the material removal rate. It measures the grinding state of the optical lens and the instantaneous cutting and peeling thickness, providing a benchmark for defining the real-time physical progress of optical lens grinding. This reduces the physical progress alignment deviation caused by the attenuation of contact stress due to grinding head wear, and prevents false alarms caused by relying on preset cutting rates to determine the processing progress during the transition between rough and fine grinding stages.

[0011] Preferably, the length of the dynamic window satisfies the following relationship: ; In the formula, For a moment The dynamic window length; To adjust the base; Let be the material removal rate at time t; The minimum window threshold; For a moment Spindle torque; This is the rated maximum torque; This is the rounding function. () is the maximum value function. It is an exponential function with the natural constant as the base.

[0012] This invention extracts the spindle torque and material removal rate at the current moment, and calculates the dynamic window length by combining the preset rated maximum torque and minimum window threshold. It measures the span of the data interval under different grinding transients, providing a calculation basis for feature extraction and adaptive noise reduction of spindle torque sequence and triaxial acceleration sequence. It suppresses the limitation that the preset static data interception window cannot peel off abnormal features under variable grinding transients, and reduces the false negative rate of high-frequency weak fine grinding defects caused by the mean-masking effect.

[0013] Preferably, the decision criterion factor satisfies the following relationship: ; In the formula, The decision scale factor for time t; It is a minimal scale constant; To determine the adjustment constant; Let be the standard deviation of time t; The standard amplitude at time t; Let be the standard removal rate at time t. It is a function with maximum value. It is an exponential function with the natural constant as the base.

[0014] This invention obtains the standard removal rate, standard discrete variance, and standard amplitude by performing a minimum-maximum normalization algorithm on the material removal rate, variance, and vibration amplitude. It then generates a decision scale factor by combining a minimum scale constant and a decision adjustment constant to measure the tolerance boundary of the basic search space. This provides an indicator for defining the abnormal energy distribution in the extremely stable phase and reduces the false alarms of grinding machines caused by minute random vibrations.

[0015] Preferably, the dynamic neighborhood radius satisfies the following relationship: ; In the formula, For a moment The dynamic neighborhood radius; The reference radius; The decision scale factor for time t; This is the penalty coefficient; For a moment The absolute deviation, It is an exponential function with the natural constant as the base.

[0016] This invention calculates the absolute deviation by extracting the measured material removal rate and the theoretical removal rate, and integrates the judgment scale factor and the preset benchmark radius to perform exponential suppression adjustment on the absolute deviation to construct a dynamic neighborhood radius. This measures the search span and working condition evolution characteristics in the three-dimensional feature space, providing a distance constraint basis for the density clustering algorithm to count the number of sample points in the neighborhood space. It suppresses the benchmark interference caused by the cumulative wear of the processing tool leading to normal gradual offset, and reduces the underreporting of real chipping defects caused by the inability of a fixed deviation threshold to distinguish between natural wear and instantaneous chipping.

[0017] Preferably, the acquisition of the triaxial acceleration sequence and the spindle torque sequence includes: acquiring the triaxial acceleration sequence using a sensor array and reading the spindle torque sequence using a CNC interface.

[0018] Preferably, obtaining the difference between the lens volume at the current moment and the previous moment includes: using a laser profilometer to scan and obtain the lens volume, and importing the lens volume into a digital twin architecture; extracting the lens volume at the current moment and the lens volume at the previous moment from the digital twin architecture; and calculating the difference between the lens volume at the previous moment and the lens volume at the current moment.

[0019] Preferably, the method of using the dynamic neighborhood radius improved density clustering algorithm to cluster the three-dimensional vectors at each time step to achieve anomaly detection in the grinding process includes: inputting the three-dimensional vector as the test point into a feature space containing historical processing sample points; searching the neighborhood space with the test point as the center and the dynamic neighborhood radius as the radius within the feature space; counting the number of sample points contained in the neighborhood space; obtaining a preset minimum threshold; and performing anomaly detection.

[0020] Preferably, the anomaly detection includes: determining the test point as an isolated anomaly when the number of sample points is less than a minimum threshold; and outputting an early warning signal when the test point is determined to be an isolated anomaly, thereby realizing real-time identification of abnormal states.

[0021] Secondly, the present invention provides a digital twin-based optical lens grinding process monitoring system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned digital twin-based optical lens grinding process monitoring method is implemented.

[0022] By adopting the above technical solution, a computer program is generated from the above-mentioned method for monitoring the optical lens grinding process based on digital twins and stored in a memory so that it can be loaded and executed by a processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0023] The beneficial effects of this invention are as follows: This invention extracts the volume difference between adjacent period optical lenses and calculates the material removal rate based on the sampling period, and combines this with adjusting the dynamic window length based on the rated maximum torque to measure the transient cutting characteristics and resistance fluctuations of optical lens grinding at different stages; it generates a judgment scale factor using standard discrete variance, standard amplitude, and standard removal rate, and integrates theoretical prediction absolute deviation to correct the dynamic neighborhood radius, thereby measuring the boundary tolerance span and search scale of the density clustering algorithm in the feature space; based on the dynamic neighborhood radius, it performs abnormal state identification in the feature space, providing an objective basis for defining natural wear and instantaneous chipping of optical lenses, suppressing false alarms caused by fixed judgment criteria in extremely stable stages, and reducing the false alarm rate and missed detection rate of minor defects caused by continuous drift of the processing reference. Attached Figure Description

[0024] Figure 1 This is a flowchart of a method for monitoring the optical lens grinding process based on digital twins; Figure 2 This is a schematic diagram of feature space anomaly detection based on improved clustering. Detailed Implementation

[0025] 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, not all, of the embodiments of the present invention. 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.

[0026] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] This invention discloses a method for monitoring the grinding process of optical lenses based on digital twins, referring to... Figure 1 This includes steps S1-S4: S1. Collect the triaxial acceleration sequence and spindle torque sequence of the grinding process, obtain the difference in lens volume between the current moment and the previous moment, and calculate the material removal rate at the current moment in combination with the sampling period.

[0028] It should be noted that optical lens grinding relies on a preset cutting rate to determine the processing progress. However, the contact stress attenuation caused by grinding head wear can lead to the grinding machine failing to align with the actual peeling thickness, resulting in inaccurate progress in optical lens grinding and misjudgment during the transition between rough and fine grinding stages. Therefore, this invention calculates the material removal rate by measuring the volume difference of optical lenses in adjacent cycles, thus measuring the grinding status of the optical lens and providing a benchmark for defining the real-time progress of optical lens grinding.

[0029] Specifically, the sampling frequency is set; the sampling period is calculated based on the sampling frequency; a three-axis acceleration sequence is acquired using a sensor array, and a spindle torque sequence is read using a CNC interface; the lens volume is obtained by scanning with a laser profilometer and imported into a digital twin architecture; the lens volume at the current moment and the lens volume at the previous moment are extracted from the digital twin architecture; the difference between the lens volume at the previous moment and the lens volume at the current moment is calculated; and the material removal rate is calculated based on the difference between the lens volume at the previous moment and the lens volume at the current moment and the sampling period.

[0030] It should be added that a triaxial acceleration sequence refers to the sequence of accelerations on the x, y, and z axes of an object.

[0031] Specifically, the material removal rate satisfies the expression: ; In the formula, For a moment Material removal rate; For a moment The volume of the lens; For a moment The volume of the lens; The sampling period.

[0032] in, The larger the value, the greater the calculated stage volume difference, resulting in a higher material removal rate. The higher the value, the more material is removed during the instantaneous cutting, and the more likely it is to determine the timing. In the transient state of coarse grinding; The smaller the value, the smaller the calculated stage volume difference, resulting in a higher material removal rate. The smaller the value, the less material is removed during the instantaneous cutting, and the more likely it is that the machining process has entered the instantaneous state of fine grinding.

[0033] For example, the sampling frequency determines the time-domain resolution of the sensor in extracting high-frequency vibrations and transient mechanical features of the machine tool. An empirical range is [1000, 5000]. In this embodiment, the sampling frequency is set to 2000Hz. In other embodiments, the implementer can set the sampling frequency according to the precision requirements of the grinding process and the system's computing power. For example, when it is necessary to identify extremely weak high-frequency scratches during the fine grinding stage, the sampling frequency can be appropriately increased to obtain more refined transient waveforms and high-frequency features; when it is only necessary to monitor the macroscopic cutting state during the rough grinding stage, the sampling frequency can be appropriately reduced to reduce the system's data processing load and ensure the real-time performance of the monitoring.

[0034] S2. Based on the real-time physical characteristics of material removal rate and spindle torque, dynamically adjust the data capture window length at the current moment, and calculate the variance and vibration amplitude of the torque sequence accordingly.

[0035] It should be noted that optical lens grinding involves peeling impact during the rough grinding stage, while exhibiting extremely faint scratch characteristics at the final fine grinding stage. This characteristic leads to a data extraction environment with an unbalanced signal-to-noise ratio during optical lens grinding, making it impossible for the preset static data extraction window to peel away abnormal features under the variable grinding transients, thus inducing false alarms in rough grinding and missed detections of defects in fine grinding. Therefore, this invention combines the rated maximum torque with the material removal rate to adjust the dynamic window length, performing adaptive noise reduction during the optical lens grinding stage, providing a data foundation for identifying the physical characteristics of optical lens grinding.

[0036] Specifically, the spindle torque at the current moment is obtained based on the spindle torque sequence; the rated maximum torque is obtained; the preset adjustment base is obtained; the preset minimum window threshold is set; and the dynamic window length is calculated based on the material removal rate, instantaneous torque value, rated maximum torque and minimum window threshold.

[0037] Specifically, the dynamic window length satisfies the expression: ; In the formula, For a moment The dynamic window length; To adjust the base; Let be the material removal rate at time t; The minimum window threshold; For a moment Spindle torque; This is the rated maximum torque; This is the rounding function. () is the maximum value function. It is an exponential function with the natural constant as the base.

[0038] in, The larger the value, the more material is removed during the instantaneous cutting, and the more likely it is to be determined that the current moment is in the rough grinding transient state. (Dynamic window length) The larger the value, the larger the range of the intercepted data, thus effectively smoothing out the background noise of large-scale contact and preventing false alarms caused by random shocks; The smaller the value, the more likely it is that the machining process has entered the transient state of fine grinding; dynamic window length. The smaller the value, the smaller the range of data intercepted, thus maintaining sensitivity to high-frequency minor defects.

[0039] The larger the value, The larger the value, the faster the time. The greater the cutting resistance, the lower the material removal rate. It has an amplification effect, increasing the dynamic window length. The larger the value, the more the data range is stretched to compensate, thus smoothing out random shocks and preventing false alarms. The smaller, The closer the value is to 1, the more stable the physical cutting conditions become at the end of the machining process, thus increasing the dynamic window length. The closer it converges to the minimum window threshold, the more sensitive it is to detecting minute, high-frequency defects.

[0040] For example, the adjustment base determines the sensitivity of the dynamic window length to changes in rate, with an empirical range of [100, 500]. In this embodiment, the adjustment base is set to 300. In other embodiments, the implementer can set the adjustment base according to the initial rotational speed of the grinding machine. For example, when the initial rotational speed is high, the adjustment base can be appropriately increased to obtain a longer smoothing range; when the initial rotational speed is low, the adjustment base can be appropriately decreased to accurately capture minute high-frequency features.

[0041] For example, the minimum window threshold determines the minimum number of sampling points required for feature extraction, with an empirical range of [10, 50]. In this embodiment, the minimum window threshold is set to 20. In other embodiments, implementers can set the minimum window threshold according to the hardware sampling frequency of the sensor. For example, when the hardware sampling frequency is high, the minimum window threshold can be appropriately increased to ensure statistical significance; when the hardware sampling frequency is low, the minimum window threshold can be appropriately decreased to reduce computational latency.

[0042] Furthermore, the variance of the principal shaft torque sequence within the dynamic window is calculated. The root mean square of the triaxial acceleration sequence within the dynamic window is calculated to obtain the vibration amplitude.

[0043] S3. Based on the standard removal rate, standard discrete variance, and standard amplitude, calculate the judgment scale factor that characterizes the tolerance boundary of the current operating condition.

[0044] It should be noted that optical lens grinding exhibits surface homogeneity at the final grinding stage, and the physical vibration energy contracts as material removal slows down. This causes the fixed judgment criteria to fail during the extremely stable phase when monitoring optical lens grinding, making it prone to false alarms triggered by minute random vibrations. Therefore, this invention combines a minimum scale constant with the standard removal rate to generate a judgment scale factor, maintaining the basic search space for optical lens grinding and providing a criterion for defining abnormal energy distributions in optical lens grinding.

[0045] Specifically, the standard removal rate is calculated using minimum-maximum normalization for the material removal rate; the standard discrete variance is calculated using minimum-maximum normalization for the variance; the standard amplitude is calculated using the minimum-maximum normalization algorithm for the vibration amplitude; and the judgment scale factor for the current moment is calculated based on the standard discrete variance, standard amplitude, preset judgment adjustment constant, and preset minimum scale constant at the current moment.

[0046] Specifically, the determination scale factor satisfies the expression: ; In the formula, The decision scale factor for time t; It is a minimal scale constant; To determine the adjustment constant; Let be the standard deviation of time t; The standard amplitude at time t; Let be the standard removal rate at time t. It is a function with maximum value. It is an exponential function with the natural constant as the base.

[0047] Among them, standard amplitude When the value approaches zero, the decision scale factor... The closer the value is to the minimum scale constant This maintains the basic search space, thereby preventing cluster collapse from causing false alarms.

[0048] The larger the value, A larger value indicates more severe cutting load pulsation, which increases the severity of the judgment criterion factor. The larger the value, the greater the tolerance range of the judgment boundary, thus effectively accommodating normal fluctuations in the early stage of processing and preventing false alarms. The smaller the value, The smaller the value, the more stable the load has become, making the decision scale factor more effective. The smaller the value, the smaller the tolerance range of the judgment boundary, thus maintaining extreme sensitivity to small fluctuations.

[0049] The smaller the value, The closer the value is to 2, the greater the standard amplitude of the base. The value is within the closed interval [0,1], so that The smaller the value, the easier it is to determine the scale factor. The smaller the value, the more the processing has entered the final stage of fine grinding, resulting in a smaller judgment boundary span, thus maintaining extreme sensitivity to minor anomalies. The higher the value, the more likely the processing is in a rough grinding stage. The closer the value is to 1, the more... The closer the value is to the base value itself, the better the decision scale factor becomes. The numerical adaptive amplification results in a larger judgment boundary span, thereby preventing false alarms caused by severe vibrations.

[0050] For example, determining the adjustment constant The basic scaling ratio of the judgment scale factor is determined, with an empirical range of [0.5, 2]. In this embodiment, the judgment adjustment constant is set to 1.2. In other embodiments, the implementer can set the judgment adjustment constant according to the material hardness of the lens to be processed. For example, when the material hardness is high, the judgment adjustment constant can be appropriately increased to accommodate severe cutting vibrations; when the material hardness is low, the judgment adjustment constant can be appropriately decreased to strictly control micro-scratches on the surface.

[0051] For example, the minimum scale constant The physical baseline of the anti-collapse search space is determined, with an empirical value range of [0.01, 0.1]. In this embodiment, the minimum scale constant is... Set to 0.05. In other embodiments, implementers can set the minimum scale constant according to the background servo noise of the CNC machine tool. For example, when the background servo noise is large, the minimum scale constant can be appropriately increased to prevent false alarms caused by background noise; when the background servo noise is extremely small, the minimum scale constant can be appropriately decreased to retain ultra-weak defect signals.

[0052] S4. Calculate the absolute deviation between physical perception and theoretical prediction, use the decision scale factor and absolute deviation to correct the dynamic neighborhood radius, and perform anomaly recognition in the feature space.

[0053] It should be noted that optical lens grinding typically uses a fixed deviation threshold to determine anomalies. However, the processing tools experience cumulative wear, causing continuous drift in the processing reference. This physical characteristic leads to a gradual shift in measured data over time, making it impossible for the fixed deviation threshold to distinguish between natural wear and instantaneous chipping, thus causing false alarms in the later stages of grinding. Therefore, this invention utilizes a judgment scale factor to dominate the reference radius and combines it with exponential suppression adjustment of the absolute deviation to generate a dynamic neighborhood radius, providing a basis for density clustering algorithms to identify abnormal states.

[0054] Specifically, the theoretical removal rate is obtained by dividing the difference between the rated lens volume and the target lens volume by the planned processing time; the deviation value at the current moment is the difference between the material removal rate and the theoretical removal rate; the absolute deviation is obtained by taking the absolute value of the difference between the deviation value at the current moment and the deviation value at the previous moment; and the dynamic neighborhood radius is calculated based on the judgment scale factor, the absolute deviation, the preset penalty coefficient, and the preset reference radius.

[0055] It should be added that the rated lens volume and the target lens volume are not fixed constants, but rather basic physical quantities pre-calculated based on the design drawings of the optical element to be processed. Specifically, the rated lens volume is the initial volume of the blank before grinding, obtained by integrating parameters such as the initial outer diameter and center thickness; the target lens volume is the theoretical final volume that meets the optical design specifications, calculated by parameters such as the target curvature surface shape and the light-transmitting aperture. During the initialization phase, the above volume parameters will be pre-fixed into the configuration library as the absolute benchmark for calculating the global material removal rate.

[0056] Specifically, the dynamic neighborhood radius satisfies the expression: ; In the formula, For a moment The dynamic neighborhood radius; The reference radius; The decision scale factor for time t; This is the penalty coefficient; For a moment The absolute deviation, ) is an exponential function with the natural constant as the base.

[0057] in, The larger the value, the more drastic the fluctuations in the grinding process conditions, and the greater the dynamic neighborhood radius. The larger the value, the larger the search span, which can better prevent false alarms in unsteady operating conditions. The smaller the value, the more stable the grinding process, and the smaller the dynamic neighborhood radius. The smaller the value, the smaller the search span, thus maintaining greater sensitivity to detecting minute anomalies.

[0058] The smaller the value, the more likely the measured cutting state is to maintain parallel evolution. The closer the value is to 1, the greater the dynamic neighborhood radius. The more likely it is to maintain the basic span, the more likely it is to successfully filter out slow, gradual errors, thereby preventing false alarms. The larger the value, the more likely a sudden change has occurred in the measured cutting condition. The closer the value is to 0, the greater the dynamic neighborhood radius. The smaller the value, the more likely it is that the chipping will occur during the cutting process, thus preventing the actual chipping defects from being missed.

[0059] For example, the baseline radius determines the initial neighborhood span within the three-dimensional feature space, with an empirical value range of [0.1, 0.5]. In this embodiment, the baseline radius is set to 0.2. In other embodiments, implementers can set the baseline radius according to the scaling scale of the feature normalization algorithm. For example, when the normalization scaling scale is concentrated, the baseline radius can be appropriately reduced to improve the accuracy of abnormal clustering; when the normalization scaling scale is divergent, the baseline radius can be appropriately increased to prevent normal data points from being misjudged.

[0060] For example, the penalty coefficient determines the degree of contraction of the neighborhood search span with respect to the absolute deviation, and its empirical value range is [1, 5]. In this embodiment, the penalty coefficient is set to 2.5. In other embodiments, the implementer can set the penalty coefficient according to the wear resistance of the grinding wheel abrasive. For example, when the wear resistance is good, the penalty coefficient can be appropriately increased to accelerate the interception of sudden chipping; when the wear resistance is poor, the penalty coefficient can be appropriately decreased to accommodate more severe natural wear deviations.

[0061] Furthermore, the standard removal rate, standard discrete variance, and standard amplitude at the current moment are constructed into a three-dimensional vector; the three-dimensional vector is input as the test point into a feature space containing historical processed sample points; within the feature space, a neighborhood space is searched with the test point as the center and the dynamic neighborhood radius as the radius; the number of sample points contained in the neighborhood space is counted; a preset minimum threshold is obtained; when the number of sample points is less than the minimum threshold, the test point is determined to be an isolated anomaly; when the test point is determined to be an isolated anomaly, an early warning signal is output to achieve real-time identification of abnormal states.

[0062] For example, the minimum threshold determines the minimum clustering threshold for core data points in the density clustering algorithm, with an empirical range of [3, 15]. In this embodiment, the minimum threshold is set to 5. In other embodiments, implementers can set the minimum threshold according to the data redundancy of the sensor network. For example, when the data redundancy is high, the minimum threshold can be appropriately increased to enhance anti-interference and noise resistance; when the data redundancy is low, the minimum threshold can be appropriately decreased to prevent the missed detection of isolated weak defective signals.

[0063] For example, Figure 2 This diagram illustrates anomaly detection based on improved clustering in the feature space. In the three-dimensional feature space comprised of the standard removal rate, standard discrete variance, and standard amplitude, sample points in the current real-time normal processing state closely overlap with historical normal sample clusters, forming a smooth, gradual transition zone. This demonstrates the system's ability to accommodate natural tool wear and normal transitions between rough and fine grinding conditions. Simultaneously, isolated anomaly points are separated from the normal main cluster. This is because a sudden increase in absolute deviation due to feature mutation triggers a dynamic shrinkage of the neighborhood radius, isolating the anomaly points due to insufficient neighboring samples. This spatial distribution reflects how the invention, by dynamically adjusting the search boundary of density clustering, can distinguish between the natural drift of the processing baseline and the actual condition. It effectively overcomes the false alarms and missed alarms that are prone to occur when traditional fixed judgment thresholds change operating conditions, thus improving the robustness and accuracy of the monitoring and early warning system.

[0064] This invention also discloses a digital twin-based optical lens grinding process monitoring system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a digital twin-based optical lens grinding process monitoring method according to the present invention.

[0065] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for monitoring the optical lens grinding process based on digital twins, characterized in that, include: Collect triaxial acceleration sequences and spindle torque sequences; Obtain the difference between the lens volume at the current moment and the previous moment, combine it with the sampling period to obtain the material removal rate and normalize it to the standard removal rate; obtain the dynamic window length at each moment based on the material removal rate, spindle torque, and rated maximum torque; The normalized variance of the spindle torque sequence within the dynamic window at each moment is used as the standard discrete variance at each moment. The vibration amplitude is calculated and normalized based on the triaxial acceleration sequence within the dynamic window to obtain the standard amplitude at each moment. The judgment scale factor at each moment is obtained based on the standard discrete variance, standard amplitude, and standard removal rate. The theoretical removal rate is obtained based on the difference between the rated lens volume and the target lens volume combined with the planned processing time. The deviation value at each moment is obtained based on the difference between the material removal rate and the theoretical removal rate at each moment, and the absolute deviation at each moment is the absolute deviation at each moment, calculated as the absolute deviation between the deviation values ​​at each moment and the previous moment. The dynamic neighborhood radius at each moment is obtained based on the judgment scale factor, absolute deviation, and reference radius. The standard removal rate, standard discrete variance, and standard amplitude at each moment are used to construct a three-dimensional vector at each moment. The density clustering algorithm improved by the dynamic neighborhood radius is used to cluster the three-dimensional vectors at each moment to achieve anomaly detection in the grinding process.

2. The method for monitoring the optical lens grinding process based on digital twins according to claim 1, characterized in that, The material removal rate satisfies the following relationship: ； In the formula, For a moment Material removal rate; For a moment The volume of the lens; For a moment The volume of the lens; The sampling period.

3. The method for monitoring the optical lens grinding process based on digital twins according to claim 1, characterized in that, The length of the dynamic window satisfies the following relationship: ; In the formula, For a moment The dynamic window length; To adjust the base; Let be the material removal rate at time t; The minimum window threshold; For a moment Spindle torque; This is the rated maximum torque; This is the rounding function. () is the maximum value function. It is an exponential function with the natural constant as the base.

4. The method for monitoring the optical lens grinding process based on digital twins according to claim 1, characterized in that, The decision criterion factor satisfies the following relationship: ； In the formula, The decision scale factor for time t; It is a minimal scale constant; To determine the adjustment constant; Let be the standard deviation of time t; The standard amplitude at time t; Let be the standard removal rate at time t. It is a function with maximum value. It is an exponential function with the natural constant as the base.

5. The method for monitoring the optical lens grinding process based on digital twins according to claim 1, characterized in that, The dynamic neighborhood radius satisfies the following relationship: ； In the formula, For a moment The dynamic neighborhood radius; The reference radius; The decision scale factor for time t; This is the penalty coefficient; For a moment The absolute deviation It is an exponential function with the natural constant as the base.

6. The method for monitoring the optical lens grinding process based on digital twins according to claim 1, characterized in that, The acquisition of the triaxial acceleration sequence and the spindle torque sequence includes: acquiring the triaxial acceleration sequence using a sensor array and reading the spindle torque sequence using a CNC interface.

7. The method for monitoring the optical lens grinding process based on digital twins according to claim 1, characterized in that, Before obtaining the difference between the lens volume at the current moment and the previous moment, the method further includes: using a laser profilometer to scan and obtain the lens volume.

8. The method for monitoring the optical lens grinding process based on digital twins according to claim 1, characterized in that, The method of using a dynamic neighborhood radius-improved density clustering algorithm to cluster three-dimensional vectors at each time step to achieve anomaly detection in the grinding process includes: inputting the three-dimensional vector as the test point into a feature space containing historical processing sample points; searching the neighborhood space with the test point as the center and the dynamic neighborhood radius as the radius within the feature space; counting the number of sample points contained in the neighborhood space; obtaining a preset minimum threshold; and performing anomaly detection.

9. A method for monitoring the optical lens grinding process based on digital twins according to claim 8, characterized in that, The anomaly detection includes: determining the test point as an isolated anomaly when the number of sample points is less than a minimum threshold; and outputting an early warning signal when the test point is determined to be an isolated anomaly, thereby achieving real-time identification of abnormal states.

10. A monitoring system for the optical lens grinding process based on digital twins, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a digital twin-based optical lens grinding process monitoring method according to any one of claims 1-9.

Images