Full-process quality monitoring method and device in the production process of optical lenses

Abstract

The present invention relates to the technical field of quality control, and discloses a full-process quality monitoring method and device in the production process of optical lenses. The method first obtains the first and second deformation parameter sets, processes them to obtain a flow front feature vector and a flow vector matrix, and establishes a flow field classification rule based on this; then retrieves the historical production data set to construct a flow-process mapping model to obtain a process control strategy, and at the same time obtains the relevant parameters of the melt to obtain the three-dimensional flow field distribution and perform multi-physical field coupling simulation; finally, different process optimization strategies are triggered according to the simulation results and classification conditions; the method realizes the precise monitoring of the entire production process of optical lenses, effectively improving the product quality and production efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of quality control, and more specifically, to a full-process quality monitoring method and device in the production process of optical lenses. Background Art

[0002] In the production process of optical lenses, quality control is a crucial link. With the continuous development of optical technology, the performance requirements for optical lenses are getting higher and higher, such as high refractive index, aspherical shape, etc., which pose higher challenges to quality monitoring in the production process.

[0003] Chinese Patent with the authorization announcement number CN117974719B discloses a processing tracking and detection method for optical lenses, which can track the processing process by integrating machine vision and deep neural network, and can detect processing anomalies in a timely manner. However, this method mainly focuses on the tracking and anomaly detection of the processing process, and pays less attention to the reconstruction of the flow field of the molten material in the mold cavity during the injection molding stage. In the injection molding of optical lenses, the flow state of the material directly affects the product quality, such as uneven refractive index distribution. This method cannot solve the problem of real-time reconstruction of the flow field of the molten material in the mold cavity during the injection molding process, and cannot provide an effective basis for optimizing the injection molding process.

[0004] Chinese Patent with the authorization announcement number CN119047695B provides a lens production quality management system based on optical characteristics, which can obtain various data in the lens production process and perform steady-state evaluation and index output. However, this system has deficiencies in real-time monitoring and controlling the flow state of the molten material in the mold cavity. For the injection molding process of high refractive index aspherical lenses, traditional pressure sensors can only obtain single-point data and cannot reconstruct the three-dimensional flow field, resulting in a problem that the uneven refractive index distribution rate is as high as 3.5%. This system cannot effectively solve this problem.

[0005] The existing technology cannot obtain comprehensive flow field information in the quality monitoring of the optical lens production process, it is difficult to accurately judge the flow state of the material, lacks fine classification of different flow anomaly modes and personalized parameter optimization control, and cannot achieve precise control of the molding process. Summary of the Invention

[0006] To overcome the above-mentioned defects of the prior art, the present invention provides a full-process quality monitoring method and device in the production process of optical lenses, aiming to solve the problem of real-time reconstruction of the flow field of molten materials in the mold cavity during the injection molding stage, and improve the production quality and efficiency of optical lenses. By embedding a distributed fiber Bragg grating array and ultrasonic guided wave transmitters on the mold surface, combined with a multi-physical field coupling model, it is possible to obtain and analyze the physical field information in the mold cavity in real time, accurately reconstruct the three-dimensional flow field, effectively solve the problem of uneven refractive index distribution, and improve the product quality; accelerate the detection speed of abnormal working conditions, adjust process parameters in a timely manner; reduce material waste and production costs; shorten the molding cycle and improve production efficiency.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A full-process quality monitoring method in the production process of optical lenses, including:

[0009] Obtain the first deformation parameter set and the second excitation parameter set, perform wavelet packet decomposition on the first deformation parameter set, and extract the high-frequency components of the n1 layer as the flow front feature vector F f ; Generate a flow vector matrix M according to the second excitation parameter set v ; Establish a flow field classification rule according to the flow front feature vector F f and the flow vector matrix M v , and obtain the classification result of the flow anomaly mode; the classification result of the flow anomaly mode includes the first-level flow mode, the second-level flow mode, and the third-level flow mode;

[0010] Retrieve the n2 batch historical production data set, and based on the historical production data set, use a generative adversarial network to construct a flow-process mapping model; according to the flow front feature vector F f , the flow vector matrix M v and the trained flow-process mapping model, obtain a process control strategy;

[0011] Obtain the melt viscosity η0 in the mold cavity and the heat transfer coefficient h0 between the melt and the mold; based on η0 and h0, obtain the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity; according to the three-dimensional flow field distribution v0(x, y, z, t), obtain the accurate three-dimensional transient flow field distribution v(x, y, z, t); according to the classification result of the flow anomaly mode, perform multi-physical field coupling simulation on the accurate three-dimensional transient flow field distribution v(x, y, z, t) to obtain a multi-physical field coupling simulation result;

[0012] According to the multi-physical field coupling simulation result and the process control strategy, trigger different process optimization strategies for different classification results of the flow anomaly mode.

[0013] Further, the obtaining of the first deformation parameter set and the second excitation parameter set includes:

[0014] Within 0.5 s after the mold is closed, surface deformation time series data is collected through a distributed fiber Bragg grating array to generate a first deformation parameter set, and the first deformation parameter set includes a pressure fluctuation value sequence P d (t) and a temperature gradient sequence T d (t);

[0015] Synchronously start an ultrasonic guided wave transmitter, apply a directional excitation wave along the melt flow direction, detect the reflected wave phase difference through the Doppler effect, and generate a second excitation parameter set, and the second excitation parameter set includes an x-axis flow velocity distribution V e (x) and a y-axis shear stress distribution τ e (y).

[0016] Further, the wavelet packet decomposition of the first deformation parameter set and the extraction of the high-frequency components of the n1 layer as the flow front feature vector F f include:

[0017] Using the Daubechies8 wavelet basis function, perform 8-layer wavelet packet decomposition on the pressure fluctuation value sequence P d (t); in the high-frequency subband coefficients of the n1 layer, extract the wavelet energy spectrum peak value and the corresponding scale index; form a 16-dimensional flow front feature vector F from the extracted wavelet energy spectrum peak value and the corresponding scale index f ; the n1 is a positive integer, and the value range is [5, 8].

[0018] Further, the generation of the flow vector matrix M v includes: using the x-axis flow velocity distribution V e (x) and the y-axis shear stress distribution τ e (y) as the row vector and column vector of the flow vector matrix M v respectively.

[0019] Further, the establishment of the flow field classification rule and the obtaining of the classification result of the flow anomaly mode include:

[0020] When the flow front feature vector F f ≥0.7 and the included angle θ of the main component of the flow vector matrix M v <15°, it is determined as a first-level flow mode, and the included angle θ of the main component is the direction of the first principal component of M v ;

[0021] When 0.4 ≤ F f <0.7 and 15° ≤ θ < 30°, it is determined as a second-level flow mode;

[0022] Determine the remaining cases as the three - level flow pattern.

[0023] Further, the historical production data set includes historical flow field characteristic quantities and corresponding process parameter optimization quantities; the historical flow field characteristic quantities include the historical flow front characteristic vector F f , hist and the historical flow vector matrix M v,hist ; the process parameter optimization quantities include the historical pressure optimization quantity ΔP hist and the historical temperature optimization quantity ΔT hist ;

[0024] The construction of the flow - process mapping model using the generative adversarial network includes:

[0025] Take the historical flow front characteristic vector F f , hist and the historical flow vector matrix M v,hist as the input X, and take the corresponding historical pressure optimization quantity ΔP hist and the historical temperature optimization quantity ΔT hist as the output Y to form a training data pair (X, Y); build a flow - process mapping model and train the flow - process mapping model according to the training data pair (X, Y).

[0026] Further, the obtaining of the process control strategy includes:

[0027] If it is the two - level flow pattern or the three - level flow pattern, then input the flow front characteristic vector F f and the flow vector matrix M v into the trained flow - process mapping model to obtain the pressure adjustment quantity ΔP pred and the temperature adjustment quantity ΔT pred ;

[0028] If it is the one - level flow pattern, then ΔP pred = 0, ΔT pred = 0;

[0029] Combine ΔP pred and ΔT pred to form a process control strategy (ΔP pred , ΔT pred ).

[0030] Further, the heat transfer coefficient h0 is obtained by inverse calculation from the temperature gradient sequence T d (t) and the pre - constructed three - dimensional transient heat transfer equation;

[0031] The obtaining of the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity includes: based on η0 and h0, using the pre-constructed generalized Newtonian fluid constitutive equation and the three-dimensional transient flow control equation, and numerically solving by the finite volume method to obtain the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity.

[0032] Further, the obtaining of the accurate three-dimensional transient flow field distribution v(x, y, z, t) according to the three-dimensional flow field distribution v0(x, y, z, t) includes:

[0033] Matching the x-axis velocity distribution V e (x) with v0(x, y, z, t), correcting η0 and h0, and obtaining the optimized melt viscosity η opt and the optimized heat transfer coefficient h opt ;

[0034] Substituting the optimized melt viscosity η opt and the optimized heat transfer coefficient h opt into the generalized Newtonian fluid constitutive equation and the three-dimensional transient flow control equation, and numerically solving again to obtain the accurate three-dimensional transient flow field distribution v(x, y, z, t) that matches the actual flow state.

[0035] Further, the matching of the x-axis velocity distribution V e (x) with v0(x, y, z, t) and correcting η0 and h0 includes:

[0036] For the x-axis velocity distribution V e (x) and v0(x, y, z, t), extracting N matching points on the y0 - z0 cross-section of V e (x), and establishing the least square objective function of the velocity deviation;

[0037] Based on the gradient descent method, iteratively optimizing and solving the least square objective function to obtain the optimized melt viscosity η opt and the optimized heat transfer coefficient h opt .

[0038] Further, the triggering of different process optimization strategies for the classification results of different flow anomaly patterns includes:

[0039] Inputting the multi-physical field coupling simulation results into the pre-trained refractive index prediction network to obtain the real-time refractive index distribution n(x, y, z, t);

[0040] Determining the target refractive index distribution n tgt (x, y, z), and calculating the difference between n(x, y, z, t) and the target refractive index distribution n tgtThe deviation Δn(x, y, z, t) of (x, y, z) is double-integrated over space and time for Δn(x, y, z, t) to obtain the cumulative deviation amount Δn total ;

[0041] Set the deviation threshold θn. Based on Δn total and θn, different process optimization strategies are triggered according to the classification results of different flow anomaly patterns.

[0042] Furthermore, the step of triggering different process optimization strategies according to Δn total and θn according to the classification results of different flow anomaly patterns includes:

[0043] For the first-level flow pattern, if Δn total < θn, the current process control strategy (ΔP pred , ΔT pred ) remains unchanged;

[0044] For the second-level flow pattern or the first-level flow pattern, if Δn total ≥ θn, process parameter optimization based on global sensitivity analysis is triggered;

[0045] For the third-level flow pattern, an adaptive parameter regulation strategy is triggered.

[0046] A full-process quality monitoring device in the production process of optical lenses, which is used to implement the full-process quality monitoring method in the production process of the above-mentioned optical lenses. The device includes:

[0047] A flow anomaly detection module: used to obtain a first deformation parameter set and a second excitation parameter set, perform wavelet packet decomposition on the first deformation parameter set, and extract the high-frequency components of the n1 layer as the flow front feature vector F f ; Generate a flow vector matrix M according to the second excitation parameter set v ; Establish a flow field classification rule according to the flow front feature vector F f and the flow vector matrix M v , and obtain the classification result of the flow anomaly pattern; the classification result of the flow anomaly pattern includes a first-level flow pattern, a second-level flow pattern, and a third-level flow pattern;

[0048] A process control strategy generation module: used to retrieve the historical production data set of n2 batches, and based on the historical production data set, construct a flow-process mapping model using a generative adversarial network; according to the flow front feature vector F f , the flow vector matrix M v and the trained flow-process mapping model, obtain the process control strategy;

[0049] Simulation module: It is used to obtain the melt viscosity η0 in the mold cavity and the heat transfer coefficient h0 between the melt and the mold; based on η0 and h0, obtain the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity; according to the three-dimensional flow field distribution v0(x, y, z, t), obtain the accurate three-dimensional transient flow field distribution v(x, y, z, t); according to the classification results of the flow anomaly patterns, perform multi-physics field coupling simulation on the accurate three-dimensional transient flow field distribution v(x, y, z, t) to obtain the multi-physics field coupling simulation results;

[0050] Process optimization module: According to the multi-physics field coupling simulation results and the process control strategy, trigger different process optimization strategies for the classification results of different flow anomaly patterns.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] The present invention realizes multi-dimensional precise control of the injection molding process. In the data acquisition link, the distributed fiber Bragg grating array and the ultrasonic guided wave transmitter work together to obtain rich and complementary data, providing a reliable basis for subsequent analysis. Through wavelet packet decomposition, constructing a flow vector matrix and establishing a flow field classification rule, different flow anomaly patterns can be finely distinguished, and potential problems can be discovered in advance. The flow-process mapping model constructed based on historical data can give accurate process control strategies according to the real-time flow state, improving the pertinence and effectiveness of process adjustment. The accurate measurement of melt parameters and flow field simulation, combined with multi-physics field coupling simulation, deeply reveals the physical change law of materials in the mold cavity. Finally, the process optimization strategy triggered according to the simulation results and classification effectively reduces the uneven refractive index distribution rate, reduces quality fluctuations and defect rates, shortens the molding cycle, and reduces material waste. Overall, this method breaks through the traditional limitations, improves the real-time performance of quality prediction, accelerates the optimization speed, promotes the injection molding process towards intelligence and digitization, significantly improves the quality and efficiency of optical lens production, and enhances the competitiveness of enterprises in the industry. Description of the Drawings

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0054] Figure 1 It is the principle flow chart of the full-process quality monitoring method in the production process of optical lenses in the present invention;

[0055] Figure 2Flowchart of the method for obtaining the first deformation parameter set and the second excitation parameter set in the full-process quality monitoring method during the production process of the optical lens of the present invention;

[0056] Figure 3 Flowchart of the method for performing wavelet packet decomposition on the first deformation parameter set and extracting the high-frequency components of the n1 layer as the flow front feature vector F in the full-process quality monitoring method during the production process of the optical lens of the present invention f ;

[0057] Figure 4 Flowchart of the method for matching the x-axis flow velocity distribution V e (x) with v0(x, y, z, t) and correcting η0 and h0 in the full-process quality monitoring method during the production process of the optical lens of the present invention;

[0058] Figure 5 Flowchart of the method for triggering different process optimization strategies in the full-process quality monitoring method during the production process of the optical lens of the present invention;

[0059] Figure 6 Flowchart of the method for constructing a quadratic response surface model in the full-process quality monitoring method during the production process of the optical lens of the present invention;

[0060] Figure 7 Functional module diagram of the full-process quality monitoring device during the production process of the optical lens in the present invention. Specific embodiments

[0061] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0062] Embodiment 1

[0063] Please refer to Figure 1 as shown. This embodiment provides a full-process quality monitoring method during the production process of an optical lens, including:

[0064] Step S1000: Obtain the first deformation parameter set and the second excitation parameter set, and based on the first deformation parameter set and the second excitation parameter set, establish a flow field classification rule to obtain the classification result of the flow anomaly mode;

[0065] Further, step S1000 includes:

[0066] Step S1100: Obtain the first deformation parameter set and the second excitation parameter set;

[0067] Furthermore, as Figure 2 shown, step S1100 includes:

[0068] Step S1110, within 0.5 s after the mold is closed, collect the surface deformation time-series data through a distributed fiber Bragg grating array to generate a first deformation parameter set, where the first deformation parameter set includes a pressure fluctuation value sequence P d (t) and a temperature gradient sequence T d (t);

[0069] Step S1120, synchronously start an ultrasonic guided wave transmitter, apply a directional excitation wave along the melt flow direction, detect the reflected wave phase difference through the Doppler effect to generate a second excitation parameter set, where the second excitation parameter set includes an x-axis flow velocity distribution V e (x) and a y-axis shear stress distribution τ e (y).

[0070] Specifically, the main purpose of step S1100 is to obtain the key data for constructing the initial flow field classification rules, that is, the first deformation parameter set and the second excitation parameter set, laying a data foundation for the subsequent precise analysis of the material flow state during the injection molding process. This step collects physical quantity data related to material flow from different angles through specific detection techniques and means, and these data complement each other, being able to comprehensively reflect the dynamic characteristics of material flow during the injection molding process.

[0071] The distributed fiber Bragg grating array is a monitoring device based on fiber optic sensing technology, with a layout density of 300 per m². This high-density layout method can achieve high-resolution perception of the mold surface deformation field. The working principle of the fiber Bragg grating is based on its sensitivity to external physical quantities (such as strain, temperature). When the external environment changes, the reflection spectrum of the grating will change accordingly. By demodulating and analyzing the grating reflection spectrum, the strain and temperature data of each point on the mold surface can be accurately obtained.

[0072] In actual operation, by filtering the strain data continuously collected in time series, the noise interference in the data can be removed, and the changing trend of the pressure fluctuation can be restored. For example, during the injection molding process, when the molten material is injected into the mold cavity, it will generate pressure on the mold surface, and this pressure change will cause changes in the mold surface strain. By collecting the strain data through the fiber Bragg grating and filtering it, the pressure fluctuation value sequence P d (t) reflecting the pressure change can be obtained. Similarly, by calculating the spatial gradient of the collected temperature data, the temperature gradient distribution T d (t) can be obtained. The temperature gradient refers to the rate of change of temperature with position in space, which can reflect the direction and intensity of heat transfer within the mold. By obtaining the temperature gradient sequence Td (t), the temperature difference at different positions inside the mold during the injection molding process can be understood. The beneficial effects of this step are significant. It can timely capture the dynamic deformation characteristics of the mold surface caused by material flow at the initial stage of molding and obtain the sequence of pressure fluctuation values P d (t) and the spatio-temporal distribution data of the two key physical quantities of the temperature gradient sequence T d (t), which has important significance in many aspects. From the perspective of process optimization, the pressure fluctuation and temperature gradient data can reflect the flow uniformity and filling state of the material in the cavity. If the pressure fluctuation is too large, it may mean that there is a turbulent flow phenomenon in the material flow, which is likely to cause defects in the product. By monitoring these data, the injection molding process parameters, such as injection speed, temperature, etc., can be adjusted in time to ensure the product quality. From the perspective of production process monitoring, these data provide a basis for real-time monitoring of the injection molding process, facilitating operators to timely discover potential problems, prevent production failures, and improve production efficiency.

[0073] Ultrasonic guided wave is an elastic wave that propagates in solid or fluid media, with characteristics such as long propagation distance and sensitivity to medium characteristics. In this step, a directional ultrasonic guided wave transmitter is installed at the mold gate, and a Doppler effect detector array is arranged on the side wall perpendicular to the gate direction. When the transmitter emits ultrasonic guided waves, the guided waves propagate in the melt, and due to the flow of the melt, a Doppler frequency shift effect will occur.

[0074] The Doppler effect refers to the phenomenon that when there is relative motion between the wave source and the observer, the frequency of the wave received by the observer will change. In this scenario, when the ultrasonic guided wave propagates in the flowing melt, the flow of the melt causes the frequency of the reflected wave to change relative to the transmitted wave. By detecting the frequency change of the reflected wave signal and combining the spatial distribution characteristics of the guided wave field, the velocity vector field of the material flow can be inversely calculated, and thus the x-axis velocity distribution V e (x) can be obtained. At the same time, the material will be subjected to the shear force of the cavity wall during the flow process, and this shear force will cause a slight change in the phase of the reflected wave. By demodulating and analyzing the phase difference distribution of the reflected wave and combining the geometric dimension parameters of the cavity, the shear stress distribution τ e (y) can be calculated.

[0075] For example, in the actual injection molding production of a certain plastic part, assume that the frequency change and phase difference of the reflected wave are measured by the ultrasonic guided wave detection system. After calculation, at a certain moment, the x-axis velocity distribution V e (x) shows that the flow velocity of the melt in a certain area of the cavity is relatively fast, while the y-axis shear stress distribution τ e (y) indicates that the shear force received in this area is relatively large. This may mean that the melt flow in this area is not very stable and defects are likely to occur. The second set of excitation parameters V e (x), τe (y) Characterizes the dynamic characteristics of material flow from a rheological perspective and complements the first set of deformation parameters. The first set of deformation parameters mainly reflects the deformation and temperature changes on the mold surface caused by material flow, while the second set of excitation parameters delves into the melt interior to obtain the dynamic distributions of key parameters such as flow velocity and shear stress. The fusion of this multi-source data together constitutes the multi-source heterogeneous data foundation for comprehensively perceiving the injection molding process. The active ultrasonic guided wave detection method is adopted to overcome the defect of insufficient sensitivity of traditional passive sensors. Traditional passive sensors can usually only detect some macroscopic physical quantity changes and it is difficult to accurately measure the microscopic flow parameters inside the melt. However, the ultrasonic guided wave detection method can penetrate into the complex structure interior to detect the microscopic motion parameters of the fluid. By optimizing the frequency selection and emission direction of the ultrasonic guided wave, the signal-to-noise ratio of the phase difference signal can be maximized, the measurement sensitivity can be improved, making the obtained data more accurate and reliable, providing strong support for subsequent accurate analysis of the material flow state and optimization of the injection molding process, and playing an important role in improving product quality stability, reducing the scrap rate, and enhancing production efficiency.

[0076] Step S1200, based on the first set of deformation parameters and the second set of excitation parameters, establish a flow field classification rule to obtain the classification result of the flow anomaly pattern;

[0077] Further, step S1200 includes:

[0078] Step S1210, perform wavelet packet decomposition on the first set of deformation parameters, and extract the high-frequency components of the n1-th layer as the flow front feature vector F f ; n1 is a positive integer, and the value range is [5, 8];

[0079] Further, as Figure 3 shown, step S1210 includes:

[0080] Step S1211, use the Daubechies8 wavelet basis function to perform 8-layer wavelet packet decomposition on the pressure fluctuation value sequence P d (t);

[0081] Step S1212, in the high-frequency subband coefficients of the n1-th layer, extract the wavelet energy spectrum peak value and the corresponding scale index;

[0082] Step S1213, compose the extracted wavelet energy spectrum peak value and the corresponding scale index into a 16-dimensional flow front feature vector F f .

[0083] Specifically, during the injection molding production process of optical lenses, perform wavelet packet decomposition on the first set of deformation parameters to extract the flow front feature vector F fIt is a key link for monitoring the flow state. The first set of deformation parameters includes the pressure fluctuation value sequence P d (t) and the temperature gradient sequence T d (t), where the pressure fluctuation value sequence P d (t) reflects the dynamic change of the pressure on the mold surface caused by the material flow during injection molding, and the temperature gradient sequence T d (t) reflects the temperature difference distribution on the mold surface due to material flow and heat transfer.

[0084] Wavelet packet decomposition is a time-frequency analysis method. In this step, the pressure fluctuation value sequence P d (t) is decomposed by 8 layers using the Daubechies8 wavelet basis function. The Daubechies8 wavelet basis function has good compact support and regularity, and can well characterize the signal features in the time domain and frequency domain. When performing wavelet packet decomposition, the signal will be continuously decomposed into sub-bands of different frequencies. The low-frequency sub-bands contain the main trend information of the signal, and the high-frequency sub-bands contain the details and local change information of the signal. From the perspective of signal feature extraction, this value range can accurately capture the key features of the flow front. During the injection molding process, it can sensitively perceive the changes related to flow instability in the high-frequency components of the pressure fluctuation value sequence P d (t). The 16-dimensional flow front feature vector F f is formed by the extracted wavelet energy spectrum peak value and scale index, which accurately characterizes the state of the flow front. In terms of computing resources and processing efficiency, the value of [5,8] achieves a balance, avoiding insufficient feature mining due to too small n1 affecting anomaly recognition, or excessive consumption of computing resources and prolongation of processing time due to too large n1, which reduces real-time performance. In terms of the stability and reliability of the flow field classification rules, this value range ensures the stability of the flow front feature vector F f , reduces classification errors, and enables the classification model to operate stably even when the process parameters fluctuate in different batches of production, providing a reliable basis for process adjustment and quality control, and ensuring the consistency of product quality.

[0085] After completing the 8-layer wavelet packet decomposition, the wavelet energy spectrum peak value and the corresponding scale index are extracted from the high-frequency sub-band coefficients of the n1-th layer (the value range of n1 is [5,8]). The wavelet energy spectrum peak value represents the energy concentration degree of the signal in this frequency sub-band, and the scale index reflects the change characteristics of the signal at different time scales. For example, when the material flow becomes unstable, such as in the case of turbulence, the energy in the high-frequency band of the pressure fluctuation value sequence P d (t) will increase significantly, the wavelet energy spectrum peak value of the corresponding frequency sub-band will increase significantly, and the scale index will also change accordingly. The extracted wavelet energy spectrum peak values and the corresponding scale indices are used to form a 16-dimensional flow front feature vector F f .

[0086] The beneficial effects of this process are remarkable. From the monitoring perspective, the flow front feature vector F f can sensitively reflect the flow instability during the filling process of the molten material in the cavity. Traditional injection molding process monitoring relies on empirical models or simple statistical rules and is difficult to accurately capture complex three-dimensional flow changes. However, this method uses wavelet packet decomposition to analyze the pressure fluctuation signal from both the time domain and the frequency domain, can extract high-dimensional features, and enhance the ability to identify abnormal flow patterns. For example, in actual production, if a certain dimension value of the flow front feature vector F f is abnormally high, combined with process experience and historical data, it can quickly judge that flow stratification or curling may occur, so as to give an early warning and avoid producing defective optical lens products. From the perspective of optimizing production, accurate monitoring of flow instability provides an accurate basis for subsequent process parameter adjustment, helps to optimize the injection molding process, improve product quality and production efficiency, and reduce the scrap rate. In terms of cost control, timely discovery and solution of flow anomalies reduce the waste of raw materials and rework costs caused by product defects, and improve the economic benefits of the enterprise.

[0087] Step S1220, generate a flow vector matrix M according to the second set of excitation parameters v ;

[0088] Taking the x-axis direction flow velocity distribution V e (x) and the y-axis direction shear stress distribution τ e (y) as the row vector and column vector of the flow vector matrix M v respectively.

[0089] Specifically, the second set of excitation parameters includes the x-axis direction flow velocity distribution V e (x) and the y-axis direction shear stress distribution τ e (y). The x-axis direction flow velocity distribution V e (x) reflects the change in the flow velocity of the molten material in the mold cavity along the x direction, and the y-axis direction shear stress distribution τ e (y) reflects the stress distribution state generated by the shear force of the cavity wall on the material during flow in the y direction.

[0090] Taking the x-axis direction flow velocity distribution V e (x) and the y-axis direction shear stress distribution τ e (y) as the row vector and column vector of the flow vector matrix M v respectively. Compared with simply superimposing or splicing the two, this method can explore the spatial correlation characteristics of the flow velocity field and the shear stress field. During the injection molding process, the flow velocity of the material and the shear stress it receives affect each other. For example, near the cavity wall, the material flow velocity will decrease due to wall friction, and at the same time, a large shear stress will be generated. Constructing them into the flow vector matrix Mv , which can more comprehensively describe the flow state of materials in a two-dimensional plane.

[0091] The data are processed using a convolutional encoder, which can automatically extract the spatial features and internal connections in the data. The convolution operation slides the convolution kernel on the data to extract features from the local area, which can effectively capture the spatial variation of the flow velocity field and the shear stress field. For example, at a certain moment, when the flow velocity of the material in a certain area of ​​the cavity suddenly accelerates, the convolutional encoder can identify the relationship between the flow velocity change in the area and the surrounding shear stress distribution, and explore the physical mechanism contained in the distribution of the second excitation parameter set. The flow vector matrix M generated in this way v It is a more abstract and robust representation of the material flow state, which can better reflect the actual flow conditions of the material in the cavity.

[0092] Step S1230: according to the flow front feature vector F f and the flow vector matrix M v , establishing a flow field classification rule, and obtaining a classification result of a flow anomaly pattern, wherein the classification result of the flow anomaly pattern includes a primary flow pattern, a secondary flow pattern, and a tertiary flow pattern;

[0093] Further, step S1230 includes:

[0094] Step S1231, when the flow front feature vector F f ≥0.7 and the flow vector matrix M v When the principal component angle θ is less than 15°, it is determined to be a primary flow mode, and the principal component angle θ is M v The first principal component direction of

[0095] Step S1232, when 0.4≤F f <0.7 and 15°≤θ<30°, it is determined to be a secondary flow mode;

[0096] Step S1233, determine the remaining situations as the third-level flow mode.

[0097] Specifically, step S1230 aims to calculate the flow front feature vector F f and the flow vector matrix M v , establish a set of scientific and reasonable flow field classification rules, so as to obtain the classification results of flow abnormality patterns, providing a key basis for subsequent process adjustment and quality control. This step realizes the effective evaluation and classification of material flow stability during the injection molding process by comprehensively considering multiple key parameters reflecting the material flow state.

[0098] In step S1231, when the flow front feature vector F f≥0.7 and the flow vector matrix M v When the included angle θ of the principal components is < 15°, it is determined as the first-level flow pattern. Here, the flow front characteristic vector F f is a key index extracted by performing wavelet packet decomposition on the sequence of pressure fluctuation values P d (t) in the first deformation parameter set. It can sensitively reflect the flow instability of the molten material during the filling process in the mold cavity. F f The larger the value, the more significant the characteristics of flow instability. And the flow vector matrix M v The included angle θ of the principal components refers to v the direction of the first principal component of M. This included angle reflects the deviation degree between the material flow direction and the main axis of the mold cavity. When θ < 15°, it means that the deviation between the material flow direction and the main axis of the mold cavity is small, the material flow is relatively smooth and close to the ideal laminar flow state. In actual injection molding production, assume that an optical lens is produced. During the injection molding process, it is monitored that the flow front characteristic vector F f reaches 0.8, and at the same time, the included angle θ of the principal components of the flow vector matrix M v is 10°. According to the rules, it can be determined as the first-level flow pattern. This indicates that the flow state of the material in the mold cavity is good at this time, and basically no adjustment of the process parameters is required, and the probability of product defects is relatively low. This accurate determination method can help producers timely understand the production status, avoid unnecessary process adjustments, save production time and costs, and improve production efficiency. At the same time, accurately identifying a stable production state helps to accumulate high-quality production data, provides strong support for the subsequent consistency control of product quality, and ensures the stability of product quality.

[0099] Step S1232 stipulates that when 0.4 ≤ F f < 0.7 and 15° ≤ θ < 30°, it is determined as the second-level flow pattern. In this interval, the flow front characteristic vector F f indicates that the material flow has begun to show a certain degree of instability, but has not reached a serious level; and the range of the included angle θ of the principal components indicates that the deviation of the material flow direction relative to the main axis of the mold cavity has increased, and the flow begins to deviate from the ideal state. For example, when producing another optical lens with a special shape, if it is detected that F fIf is 0.5 and θ is 20°, it belongs to the secondary flow pattern. In this case, although the material flow has not got out of control yet, there are already potential risks. It is necessary to fine-tune the process parameters, such as appropriately adjusting the injection pressure or temperature, to ensure that the material can fill the cavity more evenly and avoid quality problems such as uneven local thickness of the product and surface defects. Through timely fine-tuning, the generation of defects can be effectively prevented, the qualified rate of products can be improved, the production cost can be reduced, and the competitiveness of products in the market can be enhanced. At the same time, the accurate judgment of the secondary flow pattern also helps the enterprise to optimize the production process, continuously improve the production technology level, and accumulate experience in dealing with different flow states.

[0100] Step S1233 determines the remaining cases as the tertiary flow pattern. This means that in these cases, the material flow state reflected by the flow front eigenvector F f and the included angle θ with the main component is poor, and the degree of flow instability is relatively high. For example, when producing an optical lens with a complex structure, if F f is less than 0.4 and θ is greater than 30°, it will be determined as the tertiary flow pattern. At this time, the flow of the material in the cavity may exhibit serious unstable phenomena such as turbulence and stratification, which have a great impact on the product quality. In response to this situation, it is necessary to actively intervene in the process conditions and take more radical adjustment measures, such as greatly adjusting the injection speed and optimizing the mold temperature distribution, to suppress the violent fluctuations of the flow and ensure the product quality. The accurate definition and timely treatment of such a severely unstable flow state can effectively reduce the generation of waste products and reduce the economic losses of the enterprise. At the same time, through the research and treatment of the tertiary flow pattern, it helps to promote the innovation and development of the injection molding process, develop more advanced process control methods, and improve the production level of the entire industry.

[0101] Overall, step S1230 and its sub-steps achieve a detailed classification of the material flow state during the injection molding process through clear classification rules. This classification method has various beneficial effects. From the perspective of quality control, it can promptly detect abnormal situations in material flow, take corresponding measures for different degrees of abnormalities, effectively prevent and reduce product defects, and improve product quality. From the perspective of production management, it helps to reasonably arrange production resources. For stable primary flow patterns, unnecessary monitoring and adjustment can be reduced, and resources can be concentrated on dealing with unstable flow states; for secondary and tertiary flow patterns, process parameters can be accurately adjusted according to specific situations, avoiding waste of resources caused by blind adjustment and improving production efficiency. From the perspective of process optimization, the large amount of data and processing experience accumulated under different flow patterns provide a basis for optimizing the injection molding process, helping enterprises continuously improve production processes, enhance their own technical strength, and gain an advantageous position in the market competition. In addition, this classification model also has good scalability and universality. It is not only applicable to the current production of optical lenses, but after appropriate adjustment, it can also be applied to the production process monitoring and quality control of other injection molded products, providing strong technical support for the development of the entire injection molding industry.

[0102] Step S2000, retrieve the historical production data set of n2 batches. Based on the historical production data set, use a generative adversarial network to construct a flow-process mapping model; according to the flow front feature vector F f , the flow vector matrix M v and the trained flow-process mapping model, obtain the process control strategy;

[0103] Furthermore, step S2000 includes:

[0104] Step S2100, retrieve the historical production data set of n2 batches. The historical production data set includes historical flow field feature quantities and corresponding process parameter optimization quantities; the historical flow field feature quantities include the historical flow front feature vector F f , hist and the historical flow vector matrix M v,hist ; the process parameter optimization quantities include the historical pressure optimization quantity ΔP hist and the historical temperature optimization quantity ΔT hist ;

[0105] Specifically, the purpose of step S2100 is to obtain historical production data for constructing a flow-process mapping model, providing a basis for subsequent accurate prediction and adjustment of process parameters. The historical production data set of n2 batches covers rich production process information. Among them, the historical flow field feature quantities include the historical flow front feature vector F f,hist and the historical flow vector matrix M v,hist . The historical flow front feature vector F f,histRecords the unstable characteristics of the material flow front in past production. For example, during different injection molding stages, whether abnormal conditions such as turbulence and stratification occur at the material flow front. This information is obtained through the analysis of the sequence of pressure fluctuation values at that time. The historical flow vector matrix M v,hist Characterizes the velocity field distribution of historical samples during the flow filling stage, specifically manifested as the velocity distribution in the x-axis direction and the shear stress distribution in the y-axis direction, etc., reflecting the flow velocity of the material at different positions in the cavity and the shear force received.

[0106] The process parameter optimization amounts include the historical pressure optimization amount ΔP hist and the historical temperature optimization amount ΔT hist , which are quantitative indicators for optimizing and adjusting the two key process parameters of pressure and temperature for different flow anomalies. For example, when it is monitored that the material flow shows anomalies, which may lead to product defects, the injection molding pressure and temperature will be adjusted, and the specific values of these adjustments are recorded as the historical pressure optimization amount ΔP hist and the historical temperature optimization amount ΔT hist .

[0107] Obtaining a large number (such as n2 taking 100) of historical production data sets has important significance in many aspects. From the perspective of model training, rich data can more comprehensively cover various possible production conditions, making the subsequent constructed flow-process mapping model more generalizable and capable of adapting to the process parameter prediction requirements under different production conditions. Through the analysis of these data, the potential relationship between the flow field characteristics and process parameter optimization can be mined. For example, it is found that when the historical flow front feature vector F f,hist exceeds a certain threshold, and the x-axis direction velocity distribution in the historical flow vector matrix M v,hist shows anomalies, it is often necessary to make adjustments to the pressure and temperature in a specific direction and amplitude. This provides accurate samples for subsequent model training and helps improve the accuracy of model prediction. From the perspective of production optimization, these historical data provide a strong reference for process adjustment in the current production process. When similar flow anomalies to the historical ones occur, the historical optimization experience can be quickly borrowed, reducing the trial-and-error cost, improving production efficiency, reducing the scrap rate, and enhancing the stability of product quality.

[0108] Step S2200, based on the historical production data set, construct a flow-process mapping model using a generative adversarial network;

[0109] Furthermore, step S2200 includes:

[0110] Step S2210, the historical flow front feature vector F f , hist and the historical flow vector matrix M v,histTake the corresponding historical pressure optimization amount ΔP as the input X hist and the historical temperature optimization amount ΔT hist as the output Y to form a training data pair (X, Y);

[0111] Step S2220: Build a flow-process mapping model and train the flow-process mapping model according to the training data pair (X, Y);

[0112] The flow-process mapping model includes a generator G and a discriminator D. Construct a generator loss function L G and a discriminator loss function L D , and by minimizing the generator loss function L G and the discriminator loss function L D , alternately train the generator G and the discriminator D until the Nash equilibrium is reached.

[0113] The generator loss function L G includes:

[0114]

[0115] where:

[0116] : Represents the mathematical expectation, which is used to calculate the average value of a random variable. Here, it is used to calculate the expectation of variables under different distributions.

[0117] : Represents the distribution of real data (i.e., historical flow field characteristic quantities, such as and ). It is obtained from the historical production dataset and represents the true distribution of flow field characteristics in the actual production process.

[0118] : Samples drawn from the real data distribution , that is, the specific values of historical flow field characteristic quantities.

[0119] : The discriminator, whose input is the process parameter combination generated by the generator or the real process parameter combination , and the output is a probability value indicating that the input is a real sample.

[0120] : The generator, whose input is the historical flow field characteristic quantity , and the output is the generated process parameter combination , with the aim of mimicking the real process parameters as much as possible.

[0121] : The actual optimized process parameters (such as and ), which are obtained from the historical production dataset, represent the optimized values of the process parameters actually carried out for specific flow field characteristics.

[0122] : The weight coefficient is used to balance the importance of the two terms in the generator loss function. It can be adjusted within a certain range (such as ) through methods such as cross-validation to achieve the best training effect. A larger value will pay more attention to the closeness between the generated process parameters and the actual process parameters, while a smaller value will pay more attention to the ability to deceive the discriminator.

[0123] The first term : Measures the probability that the discriminator judges the process parameter combination generated by the generator as a real sample. The larger this probability is, the better, that is, it makes it difficult for the discriminator to distinguish the generated parameters from the real parameters. This term prompts the generator to generate combinations closer to the actual process parameter distribution.

[0124] The second term : Represents the mean square error between the generated process parameter combination and the actual process parameters. This term ensures that the parameters generated by the generator can not only deceive the discriminator but also be numerically close to the actual optimized process parameters, thus ensuring the practicality of the generated parameters.

[0125] As the process parameters generated by the generator get closer and closer to the actual process parameters , the value of the second term will gradually decrease. At the same time, if the generated makes it more difficult for the discriminator to distinguish it from the real sample, the value of the first term will also decrease. Generally speaking, when the process parameters generated by the generator meet the requirements better, the value is smaller.

[0126] The discriminator loss function L D includes:

[0127]

[0128] The first term : Measures the probability that the discriminator correctly identifies the real process parameter sample as real. The discriminator hopes that this probability is as large as possible, that is, the judgment of the real sample is more accurate.

[0129] The second term : Measures the probability that the discriminator correctly identifies the sample generated by the generator as fake. The discriminator also hopes that this probability is as large as possible, that is, the ability to identify the generated sample is stronger.

[0130] The entire loss function prompts the discriminator to continuously improve its ability to distinguish between real samples and generated samples, thereby driving the generator to generate process parameters closer to the real ones.

[0131] When the discriminator's recognition probability for real process parameter samples is higher, the value of the first term is smaller; when the probability that the discriminator recognizes the generated samples as fake is higher, the value of the second term is smaller. Therefore, when the discriminator's discrimination ability is stronger, the value of is smaller. During the training process, the discriminator continuously optimizes itself, making

[0132] gradually decrease, and at the same time also prompting the generator to improve to cope with the improvement of the discriminator, ultimately reaching the Nash equilibrium.

[0132] The essence of sub-step S2210 is to establish the association between the flow field characteristics and the corresponding process parameter optimization strategies. For example, in a set of training data pairs, if the historical flow front feature vector F f,hist shows obvious instability in the material flow front, and the historical flow vector matrix M v,hist indicates abnormal flow velocity and shear stress distribution, the corresponding historical pressure optimization amount ΔP hist and the historical temperature optimization amount ΔT hist record the adjustment values of the injection pressure and temperature made at that time to solve the flow problem. Through a large number of such data pairs, the process parameter optimization required under different flow states can be fully presented, providing sufficient information for the model to learn.

[0133] The goal of the generator G is to generate a process parameter combination G(u) that is as close as possible to the real process parameter optimization amount v based on the input historical flow field characteristic quantity u. The first term G in its loss function L measures the probability that the generated process parameter combination is judged as a real sample by the discriminator. The generator hopes that this probability is as large as possible, that is, making it difficult for the discriminator to distinguish between the generated parameters and the real parameters, and prompting the generator to generate a process parameter combination that more conforms to the real distribution; the second term represents the mean square error between the generated process parameter combination and the real process parameters, ensuring that the generated parameters can not only deceive the discriminator but also be numerically close to the real optimization amount, guaranteeing the practicality of the generated parameters. For example, if the pressure and temperature adjustment parameters generated by the generator differ greatly from the real optimization amount, the mean square error value of the second term will be very large, prompting the generator to adjust the generated parameters.

[0134] The goal of the discriminator D is to maximize the ability to distinguish between real samples and generated samples. During the training process, the discriminator continuously optimizes itself to improve the ability to distinguish between real samples and generated samples, thereby driving the improvement of the generator. For example, when the discriminator can accurately identify unreasonable process parameters generated by the generator, the generator will adjust the generation strategy according to the feedback of the discriminator to generate parameters closer to the real situation. As the training progresses, the generator and the discriminator continuously play games until they finally reach a Nash equilibrium. At this time, the generator G can directly generate optimized process parameters from the given flow characteristics.

[0135] The flow-process mapping model obtained by training through the adversarial generation network can accurately predict appropriate process parameter adjustment schemes based on the real-time monitored flow field characteristics, effectively avoiding product defects caused by improper process parameters and improving product quality. In terms of improving production efficiency, it reduces the time and resource consumption required for traditional trial-and-error process adjustment, quickly provides accurate process parameter suggestions for production, speeds up the production process, and improves production efficiency. From the perspective of cost control, it reduces the scrap rate, reduces raw material waste and rework costs, and improves the economic benefits of the enterprise. At the same time, this model also provides a powerful tool for the optimization research of the optical lens production process, helps to deeply explore the complex relationship between the flow field and process parameters, and promotes the technological progress of the industry.

[0136] Step S2300, according to the flow front feature vector F f 、the flow vector matrix M v and the trained flow-process mapping model, obtain the pressure adjustment amount ΔP pred and the temperature adjustment amount ΔT pred , and form a process control strategy (ΔP pred , ΔT pred ) with ΔP pred and ΔT pred .

[0137] If it is a first-level flow mode, then ΔP pred = 0, ΔT pred = 0;

[0138] If it is a second-level flow mode or a third-level flow mode, then input the flow front feature vector F f and the flow vector matrix M v into the trained flow-process mapping model to obtain the pressure adjustment amount ΔP pred and the temperature adjustment amount ΔT pred .

[0139] Specifically, the flow front feature vector F f reflects the flow instability of the molten material during filling in the mold cavity. For example, when F fThe numerical value is abnormally high, which may indicate problems such as turbulence and delamination in the material flow; the flow vector matrix M v characterizes the flow state of the material in terms of flow velocity and shear stress, such as the x-axis flow velocity distribution V e (x) and the y-axis shear stress distribution τ e (y). Their changes can reflect the flow characteristics of the material at different positions in the cavity.

[0140] When it is judged as the first-level flow pattern, it means that the material flow is in the most stable quasi-laminar flow state. At this time, ΔP pred = 0, ΔT pred = 0, that is, there is no need to adjust the current injection pressure and temperature parameters. This is because in this stable state, the existing process parameters can ensure that the material fills the cavity evenly and stably. Continuing to maintain the current parameters can ensure the stability of the product quality and avoid introducing new unstable factors due to unnecessary parameter adjustments. For example, in a certain optical lens production workshop, when the flow front characteristic vector F f and the flow vector matrix M v meet the determination conditions of the first-level flow pattern, by maintaining the current injection pressure and temperature parameters, the quality of the produced lens products is stable and the yield is high.

[0141] If it is the second-level flow pattern or the third-level flow pattern, it indicates that the material flow has deviated from the ideal state to varying degrees or has a high degree of instability. At this time, input the flow front characteristic vector F f and the flow vector matrix M v into the trained flow-process mapping model. The model will output the pressure adjustment amount ΔP pred and the temperature adjustment amount ΔT pred according to the relationship between the flow field characteristics learned previously and the process parameter optimization. These adjustment amounts are obtained based on the learning and analysis of a large amount of historical production data and can specifically solve the current flow anomaly problem. For example, in the second-level flow pattern, if the flow front characteristic vector F f shows a slight instability trend in the material flow and there is a local anomaly in the x-axis flow velocity distribution in the flow vector matrix M v , the model may output a small pressure adjustment amount ΔP pred and an appropriate temperature adjustment amount ΔT pred . By fine-tuning the injection pressure and temperature, the material flow can be restored to stability and the possibility of product defects can be reduced. In the third-level flow pattern, the model will output more significant pressure and temperature adjustment amounts according to more complex flow anomalies to suppress the violent fluctuations of the flow and ensure the product quality.

[0142] The beneficial effects of this step are reflected in multiple aspects. From the perspective of product quality assurance, by adjusting process parameters in a timely and accurate manner, product defects caused by abnormal flow, such as bubbles, uneven thickness, and internal stress concentration, are effectively avoided, improving the optical performance and overall quality of the product. In terms of production efficiency improvement, based on the rapid response of the model, an adjustment plan can be quickly given after detecting abnormal flow, reducing production interruptions and trial-and-error time, and accelerating the production process. From the perspective of cost control, the scrap rate is reduced, raw material waste and rework costs are decreased, and the production efficiency of the enterprise is improved. At the same time, this data-driven process control strategy provides strong support for the automation and intelligent control of the optical lens production process, contributing to the improvement of the production technology level of the entire industry.

[0143] Step S3000, obtain the melt viscosity η0 in the mold cavity and the heat transfer coefficient h0 between the melt and the mold; based on η0 and h0, obtain the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity; match the x-axis velocity distribution V e (x) with v0(x, y, z, t) to correct η0 and h0 to obtain the optimized melt viscosity η opt and the optimized heat transfer coefficient h opt ; according to η opt and h opt , obtain the accurate three-dimensional transient flow field distribution v(x, y, z, t); according to the classification result of the abnormal flow pattern, perform multi-physics field coupling simulation on the accurate three-dimensional transient flow field distribution v(x, y, z, t) to obtain the multi-physics field coupling simulation result;

[0144] Furthermore, step S3000 includes:

[0145] Step S3100, obtain the melt viscosity η0 in the mold cavity and the heat transfer coefficient h0 between the melt and the mold; the heat transfer coefficient h0 is obtained by inverse calculation from the temperature gradient sequence T d (t) and the pre-constructed three-dimensional transient heat transfer equation;

[0146] Specifically, the melt viscosity η0 reflects the ability of the melt to resist flow deformation internally, and it directly affects the flow velocity and uniformity of the material in the mold cavity. For example, when the melt viscosity is high, the material flow is relatively difficult, which may lead to insufficient filling or uneven flow; while when the melt viscosity is low, although the flow is relatively easy, problems such as flash may be caused. In this embodiment, a high-pressure capillary rheometer is used to measure the melt viscosity η0 in real time. Compared with the traditional method of looking up empirical values in a table, this in-situ measurement method can more accurately reflect the rheological characteristics of the melt under actual injection molding process conditions. Because during the injection molding process, the viscosity of the melt is dynamically affected by various factors such as temperature, pressure, and shear rate. It is difficult for the traditional table-lookup method to take into account these real-time changing factors, while in-situ measurement can track the change of the melt viscosity in real time.

[0147] The heat transfer coefficient h0 is used to characterize the efficiency of heat transfer between the melt and the mold, and it is obtained by inverse calculation from the temperature gradient sequence T d (t) and the pre-constructed three-dimensional transient heat transfer equation. The temperature gradient sequence T d (t) is obtained by collecting with a distributed fiber Bragg grating array within 0.5 s after the mold is closed in the early stage. The three-dimensional transient heat transfer equation is established based on the basic principles of heat transfer, comprehensively considering the heat transfer process in time and space. Through this calculation process, the heat transfer coefficient h0 between the melt and the mold at different times and positions can be obtained. Accurately obtaining the heat transfer coefficient h0 is of great significance for analyzing the heat distribution and transfer law during the injection molding process. During the injection molding process, the heat exchange between the melt and the mold will affect the temperature distribution of the melt, and further affect its viscosity and flow performance. If the heat transfer coefficient is inaccurate, the temperature change of the melt in the cavity cannot be accurately simulated, resulting in deviations in the analysis of the melt flow state. Therefore, accurately obtaining the melt viscosity η0 and the heat transfer coefficient h0 lays a solid data foundation for accurately simulating the flow and heat transfer process of the melt in the mold cavity later, helps to improve the understanding and control ability of the injection molding process, and thus effectively guarantees the production quality of optical lenses.

[0148] Step S3200, based on η0 and h0, using the pre-constructed generalized Newtonian fluid constitutive equation and three-dimensional transient flow control equation, numerically solve by the finite volume method to obtain the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity;

[0149] Specifically, polymer melts exhibit non-Newtonian rheological properties during the injection molding process, that is, their viscosity changes with changes in factors such as shear rate. The generalized Newtonian fluid constitutive equation is a mathematical model used to describe this property, which can accurately characterize the rheological behavior of polymer melts under different shear conditions. For example, during the injection molding process, the melt is subjected to a higher shear rate near the gate. At this time, the generalized Newtonian fluid constitutive equation can reflect the changes in the melt viscosity, which in turn affects the flow velocity and pressure distribution of the melt.

[0150] The three-dimensional transient flow control equations include mass, momentum and energy conservation equations, which describe the flow behavior of polymer melt in the cavity from the physical essence. The mass conservation equation ensures that the mass of the melt will not be created or disappeared in the injection molding process, ensuring the physical rationality of the calculation; the momentum conservation equation describes the momentum change of the melt during the flow process, which is closely related to the speed and force of the melt; the energy conservation equation takes into account the energy conversion and transfer of the melt during the flow process, such as the energy change caused by frictional heat generation.

[0151] The finite volume method is a commonly used numerical calculation method, and its core idea is to discretize the cavity space and time. Specifically, the cavity is divided into a series of tiny control volumes, and the partial differential form of the control equation is integrated in each control volume to convert it into a set of algebraic equations. Then, an iterative algorithm is used to solve these algebraic equations with high precision. In each iteration, the value of the variable is continuously updated according to the result of the previous iteration until a certain convergence condition is met. In this way, the distribution of physical quantities such as velocity, pressure, and temperature of the melt in the cavity at different times and positions can be obtained, that is, the three-dimensional flow field distribution v0 (x, y, z, t).

[0152] This step obtains the three-dimensional flow field distribution v0 (x, y, z, t) through precise numerical simulation, which provides detailed information for in-depth understanding of the flow law of the melt during the injection molding process. For example, the flow path of the melt in the cavity and whether there are flow dead corners can be analyzed, thereby providing a basis for optimizing mold design. Secondly, accurate flow field distribution helps to predict possible defects in the product, such as weld marks, bubbles, etc. By discovering potential problems in advance, process parameters can be adjusted in time to reduce scrap rates and improve product quality. In addition, this method based on physical equations and numerical calculations is more scientific and reliable than traditional empirical methods, and can more accurately guide actual production, improve production efficiency, reduce production costs, and promote technological progress in the injection molding production process of optical lenses.

[0153] Step S3300: Distribute the velocity V in the x-axis direction. e(x) is matched with v0(x, y, z, t), η0 and h0 are corrected, and the optimized melt viscosity η is obtained. opt and the optimized heat transfer coefficient h opt ;

[0154] Furthermore, as Figure 4 shown, step S3300 includes:

[0155] Step S3310, for the x-axis velocity distribution V e (x) and v0(x, y, z, t), N matching points are extracted on the y0-z0 cross-section of V e (x), and a least squares objective function for the velocity deviation is established;

[0156] The least squares objective function includes:

[0157]

[0158] where:

[0159] is the number of matching points;

[0160] is the th matching point's coordinates;

[0161] and are the corresponding and coordinates;

[0162] is the current time;

[0163] is the th measured flow velocity value at the matching point;

[0164] is the th simulated flow velocity value at the matching point;

[0165] is the melt viscosity to be corrected;

[0166] is the heat transfer coefficient to be corrected.

[0167] Step S3320, based on the gradient descent method, iteratively optimizes and solves the least squares objective function to obtain the optimized melt viscosity η opt and the optimized heat transfer coefficient h opt .

[0168] Specifically, first, N matching points are extracted on the cross-section y0-z0 of V e (x). These matching points are the key positions for comparing the measured flow velocity and the simulated flow velocity. For example, in a specific optical lens injection mold, according to the structural characteristics of the mold and the main path of the melt flow, the cross-section y0-z0 is selected in the area where the melt flow is more critical, and N points are uniformly or non-uniformly selected as matching points according to actual needs on this cross-section. By comparing the flow velocity values in the x-axis direction of the actually measured V e (x) (i.e., the measured flow velocity values ) with the flow velocity values of the corresponding points calculated from the three-dimensional flow field distribution v0(x, y, z, t) (i.e., the simulated flow velocity values ), the difference between the flow field simulated by the currently used melt viscosity η0 and heat transfer coefficient h0 and the actual situation can be evaluated.

[0169] Based on these matching points, a least-squares objective function of the velocity deviation is established. The significance of the least-squares objective function is to find the optimal melt viscosity η and heat transfer coefficient h by minimizing the sum of the squares of the deviation between the measured flow velocity and the simulated flow velocity. Since the melt viscosity η and heat transfer coefficient h will affect the calculation result of the three-dimensional flow field distribution v0(x, y, z, t), and thus affect the simulated flow velocity value, adjusting these two parameters to make the objective function J(η, h) reach the minimum value can make the simulated flow velocity closer to the measured flow velocity, so that the simulated flow field is more in line with the actual situation.

[0170] Subsequently, the least-squares objective function is iteratively optimized and solved based on the gradient descent method. The gradient descent method is a commonly used optimization algorithm. It calculates the gradient of the objective function with respect to the variables (i.e., the melt viscosity η and heat transfer coefficient h), and then gradually adjusts the values of the variables along the opposite direction of the gradient, so that the value of the objective function gradually decreases. In each iteration process, the gradient of the objective function is calculated according to the current variable values, and then the variables are updated according to a certain step size. For example, in the first iteration, the gradient of the objective function is calculated according to the initial melt viscosity η0 and heat transfer coefficient h0, and then the values of these two parameters are adjusted; in the second iteration, the gradient is calculated again based on the parameter values updated in the first iteration and the parameters are further adjusted, and so on, until the value of the objective function converges to a smaller value. At this time, the obtained melt viscosity η opt and heat transfer coefficient h opt are the optimized parameters.

[0171] Through this matching and optimization process, this step can significantly improve the simulation accuracy of the melt flow state in the mold cavity. Accurate simulation results help to understand the actual flow behavior of the melt during the injection molding process more deeply, providing a reliable basis for subsequent accurate prediction of product quality and optimization of process parameters. In terms of product quality assurance, more accurate simulation can detect potential quality problems in advance, such as internal stress concentration and appearance defects in the product caused by uneven melt flow. By adjusting the process parameters in a timely manner, these problems can be effectively avoided, and the yield rate of the product can be improved. The optimized melt viscosity and heat transfer coefficient provide more accurate references for optimizing the injection molding process, helping engineers to formulate more reasonable process plans, improve production efficiency, reduce production costs, and enhance the competitiveness of products in the market.

[0172] Step S3400, substitute the optimized melt viscosity η opt and the optimized heat transfer coefficient h opt into the generalized Newtonian fluid constitutive equation and the three-dimensional transient flow control equation, and perform numerical solution again to obtain the accurate three-dimensional transient flow field distribution v(x, y, z, t) that matches the actual flow state;

[0173] Specifically, the generalized Newtonian fluid constitutive equation and the three-dimensional transient flow control equation are the core mathematical models describing the flow and heat transfer behavior of polymer melts during the injection molding process. In previous calculations, due to certain errors in the melt viscosity η0 and the heat transfer coefficient h0, there was a deviation between the simulated three-dimensional flow field distribution v0(x, y, z, t) and the actual situation. The optimized melt viscosity η opt and the heat transfer coefficient h opt are obtained by correcting the original parameters based on the actually measured flow velocity data (i.e., the x-axis flow velocity distribution V e (x)), which are more in line with the physical characteristics in the actual injection molding process.

[0174] When substituting the optimized parameters into the equations for re-solving, the same finite volume discretization method and numerical algorithm as in step S3200 are used to perform high-resolution solution of the cavity flow field in terms of time and space. The finite volume discretization method divides the cavity space into multiple tiny control volumes, and solves the physical quantities within each control volume, thereby realizing the discretization process of the entire cavity flow field. The numerical algorithm is used to iteratively solve the discretized algebraic equations. By continuously updating the values of the variables, the calculation results gradually approach the real flow field state. During this process, due to the use of more accurate parameters, the boundary conditions in the constitutive equation and the control equation can better adapt to the actual process state. For example, at the junction of the melt and the mold wall surface, the optimized heat transfer coefficient h optIt can more accurately describe the heat transfer situation, making the setting of boundary conditions more in line with the actual heat transfer process, thereby improving the accuracy of numerical simulation.

[0175] The melt flow field distribution v(x, y, z, t) obtained through this step highly coincides with the actual injection molding process in terms of spatio-temporal scale. This precise flow field distribution provides a solid data foundation for subsequent defect prediction and quality control. In terms of defect prediction, based on the precise flow field distribution, it is possible to more accurately judge whether the flow of the melt in the cavity is uniform and whether defects such as weld lines and bubbles will form in certain areas. For example, by analyzing the convergence of the melt in the flow field, the possible locations of weld lines can be predicted; by observing the flow velocity and pressure distribution, it can be determined whether bubbles will be generated due to the inability to discharge gas. In terms of quality control, the precise flow field distribution helps to optimize process parameters, such as adjusting the injection pressure, temperature, etc., to ensure uniform flow of the melt in the cavity, reduce the generation of defects, and improve product quality. In addition, this precise flow field distribution can also provide a reference for mold design, helping engineers to improve the mold structure, further enhance the stability of the injection molding process and product quality, and enhance the competitiveness of enterprises in the field of optical lens production.

[0176] Step S3500: According to the classification results of the flow anomaly mode, perform multi-physics field coupling simulation on the precise three-dimensional transient flow field distribution v(x, y, z, t) to obtain the multi-physics field coupling simulation results.

[0177] Furthermore, step S3500 includes:

[0178] Step S3510: Under the first-level flow mode, through thermo-fluid-solid coupling analysis, obtain the temperature field T(x, y, z, t), pressure field P(x, y, z, t), stress field σ(x, y, z, t), and product deformation displacement field δ(x, y, z, t) to form the first-level flow simulation results;

[0179] Step S3520: Under the second-level flow mode, through thermo-fluid-stress coupling analysis, obtain the temperature field T(x, y, z, t), pressure field P(x, y, z, t), secondary flow velocity field v2(x, y, z, t), and stress field σ(x, y, z, t) to form the second-level flow simulation results;

[0180] Step S3530: Under the third-level flow mode, through thermo-fluid-stress-electromagnetic multi-field coupling analysis, obtain the temperature field T(x, y, z, t), pressure field P(x, y, z, t), tertiary flow velocity field v3(x, y, z, t), stress field σ(x, y, z, t), and electromagnetic induction intensity field E(x, y, z, t) to form the third-level flow simulation results;

[0181] Step S3540: Combine the primary flow simulation results, secondary flow simulation results, and tertiary flow simulation results to form the multi-physics field coupling simulation results.

[0182] Specifically, step S3500 aims to perform multi-physics field coupling simulation on the precise three-dimensional transient flow field distribution v(x, y, z, t) based on the classification results of flow anomaly patterns, so as to obtain the multi-physics field coupling simulation results, providing a key basis for subsequent process optimization. By considering the interactions between multiple physical fields, this step deeply reveals the behavioral characteristics of polymer melts under different flow anomaly patterns, which is of great significance for improving the production quality of optical lenses.

[0183] In step S3510, for the primary flow mode, a thermal-fluid-solid coupling analysis method is adopted to obtain the temperature field T(x, y, z, t), pressure field P(x, y, z, t), stress field σ(x, y, z, t), and product deformation displacement field δ(x, y, z, t), which constitute the primary flow simulation results. Thermal-fluid-solid coupling analysis is an analytical method that comprehensively considers the interactions of heat conduction, fluid flow, and solid mechanics. During the injection molding process of optical lenses, when in the primary flow mode, the flow of the melt in the mold cavity is relatively stable. For example, when producing a common circular optical lens, the melt fills the cavity at a relatively uniform speed. At this time, through thermal-fluid-solid coupling analysis, it is possible to accurately simulate the heat transfer, pressure distribution during the filling process, and the impact on the mold wall and the final product shape. By obtaining the temperature field T(x, y, z, t), the temperature changes of the melt at different positions and times in the cavity can be understood, which is crucial for controlling the curing process of the product. Excessive or too low temperature may lead to product defects, such as internal stress concentration and poor surface quality. The analysis of the pressure field P(x, y, z, t) helps to master the pressure distribution during the filling process of the melt, ensuring that the melt can evenly fill all corners of the cavity and avoiding underfilling or overfilling. The stress field σ(x, y, z, t) reflects the magnitude and distribution of the stress on the product during the molding process, while the product deformation displacement field δ(x, y, z, t) is directly related to the final shape accuracy of the product. Through the comprehensive analysis of these physical fields, potential quality problems of the product in the primary flow mode, such as minor deformations or internal stress concentration areas, can be predicted in advance, providing strong support for subsequent process optimization, helping to improve the product qualification rate and reduce production costs.

[0184] In step S3520, for the secondary flow pattern, thermo-fluid-stress coupling analysis is performed to obtain the temperature field T(x, y, z, t), pressure field P(x, y, z, t), secondary flow velocity field v2(x, y, z, t), and stress field σ(x, y, z, t), which constitute the secondary flow simulation results. Compared with the primary flow pattern, the flow of the melt in the secondary flow pattern begins to show a certain degree of instability. Thermo-fluid-stress coupling analysis focuses on the mutual influence between heat, fluid, and stress. Taking the production of an optical lens with a certain curvature as an example, in the secondary flow pattern, the flow velocity and direction of the melt may exhibit some fluctuations. Through thermo-fluid-stress coupling analysis, the impact of these fluctuations on product quality can be studied more deeply. The secondary flow velocity field v2(x, y, z, t) reflects the secondary flow characteristics of the melt on the basis of the primary flow. This secondary flow may be caused by factors such as the interaction between the melt and the mold wall, temperature differences, etc. Analyzing the secondary flow velocity field helps to detect local flow anomalies of the melt in the cavity, such as possible vortices or regions with uneven flow velocities. Combining the analysis results of the temperature field, pressure field, and stress field can provide a more comprehensive understanding of the physical state changes of the product during the molding process. For example, when it is found that a certain area has a high temperature and stress concentration, and at the same time the secondary flow velocity is large, it can be judged that there may be potential quality risks in this area, such as defects like bubbles and sink marks. By discovering these problems in advance, process parameters can be adjusted accordingly, such as optimizing the temperature distribution and adjusting the injection pressure, so as to effectively avoid or reduce the generation of product defects and improve the stability and consistency of product quality.

[0185] In step S3530, under the three - level flow mode, a thermal - fluid - stress - electromagnetic multi - field coupling analysis is carried out to obtain the temperature field T(x, y, z, t), pressure field P(x, y, z, t), three - level flow velocity field v3(x, y, z, t), stress field σ(x, y, z, t) and electromagnetic induction intensity field E(x, y, z, t), which constitute the three - level flow simulation results. The flow instability degree of the melt under the three - level flow mode is relatively high, involving more complex physical phenomena. The thermal - fluid - stress - electromagnetic multi - field coupling analysis comprehensively considers the interactions of various physical processes such as heat conduction, fluid flow, stress distribution, and electromagnetic induction. For example, when producing lenses with special optical performance requirements, electromagnetic effects may be introduced during the injection molding process to improve product quality. Through this multi - field coupling analysis, the influence of the electromagnetic induction intensity field E(x, y, z, t) on the melt flow and product forming can be studied. The three - level flow velocity field v3(x, y, z, t) further describes the velocity change of the melt under complex flow conditions, which contains more flow details and instability factors. Through the comprehensive analysis of these physical fields, the forming process of the product under a highly unstable flow state can be understood more deeply. For example, when it is found that the electromagnetic induction intensity field is strong in a certain area, and at the same time, the stress field and temperature field in this area also show abnormalities, and the three - level flow velocity field presents complex changes, it can be judged that serious quality problems may occur in this area, such as uneven internal structure and inconsistent optical performance of the product. Based on these analysis results, more targeted process improvement measures can be taken, such as adjusting electromagnetic parameters and optimizing the mold structure, to ensure that the product can still meet the quality requirements under complex flow conditions, improve the product yield rate, and expand the applicable range of the production process of optical lenses.

[0186] Step S3540 constitutes the multi - physical - field coupling simulation results from the first - level flow simulation results, the second - level flow simulation results, and the third - level flow simulation results. The multi - physical - field coupling simulation results integrate multiple physical - field information under different flow anomaly modes. Starting from multiple physical mechanisms such as heat conduction, pressure, flow control, solidification shrinkage, residual stress, and electromagnetic induction, they comprehensively reveal the defect generation rules of polymer melts under different flow anomaly modes. This is the theoretical basis for realizing accurate quality prediction and optimization control. By comparing the simulation results under different levels of flow patterns, it can be clearly seen that as the flow instability increases, the change trends of each physical field and the complexity of their interactions. For example, from the first - level to the third - level flow pattern, the non - uniformity of the temperature field may gradually increase, the fluctuations of the pressure field become more intense, and the distribution of the stress field becomes more complex. These changes are closely related to the generation of product defects. Based on the multi - physical - field coupling simulation results, production enterprises can formulate corresponding preventive measures and process optimization plans in advance. For possible defects such as bubbles, sink marks, and deformations, the process parameters can be adjusted before production to avoid a large number of defective products during production and reduce production costs. At the same time, these results also provide data support and theoretical basis for the improvement and innovation of the optical lens production process, helping to promote the development of the industry technology and improve the quality control level and production efficiency of the entire optical lens production industry.

[0187] Step S4000, according to the multi - physical - field coupling simulation results and the process control strategy, triggers different process optimization strategies for the classification results of different flow anomaly modes.

[0188] Further, step S4000 includes:

[0189] Step S4100, according to the multi - physical - field coupling simulation results, obtains the real - time refractive index distribution n(x, y, z, t); according to the real - time refractive index distribution n(x, y, z, t), for the classification results of different flow anomaly modes, triggers different process optimization strategies;

[0190] Further, as Figure 5 shown, step S4100 includes:

[0191] Step S4110, inputs the multi - physical - field coupling simulation results into a pre - trained refractive index prediction network to obtain the real - time refractive index distribution n(x, y, z, t);

[0192] Step S4120, determines the target refractive index distribution n tgt (x, y, z), calculates the deviation Δn(x, y, z, t) between n(x, y, z, t) and the target refractive index distribution n tgt (x, y, z), and performs double integration of Δn(x, y, z, t) in space and time to obtain the cumulative deviation amount Δntotal ;

[0193] Step S4130, set the deviation threshold θn, and according to Δn total and θn, for the classification results of different flow anomaly patterns, trigger different process optimization strategies.

[0194] For the first-level flow pattern, if Δn total < θn, then maintain the current process control strategy (ΔP pred , ΔT pred ) unchanged;

[0195] For the second-level flow pattern or the first-level flow pattern, if Δn total ≥ θn, then enter step S4200 to trigger the optimization of process parameters based on global sensitivity analysis;

[0196] For the third-level flow pattern, enter step S4300 to trigger a strong robust adaptive parameter regulation strategy.

[0197] Specifically, step S4100 aims to obtain the real-time refractive index distribution n(x, y, z, t) according to the multi-physics field coupling simulation results, and trigger different process optimization strategies based on this distribution and the classification results of flow anomaly patterns, so as to achieve precise quality control of the optical lens production process. This step links the physical quantity information obtained from the multi-physics field coupling simulation with the optical performance indicators, providing a clear direction and basis for process optimization.

[0198] In step S4110, the multi-physics field coupling simulation results are input into a pre-trained refractive index prediction network to obtain the real-time refractive index distribution n(x, y, z, t). The refractive index prediction network here adopts a three-dimensional CNN (Convolutional Neural Network) architecture. The three-dimensional CNN architecture is a deep learning model specifically used to process three-dimensional data. It can automatically extract deep features from the input three-dimensional data through a combination of multiple convolutional layers, pooling layers, and fully connected layers. In this embodiment, the network can learn and extract the key features affecting the refractive index from the spatio-temporal distributions of the flow field, temperature field, and stress field. The principle is that the convolutional kernels in the convolutional layer slide on the three-dimensional data, and local features are extracted through convolutional operations. Different convolutional kernels can capture feature information of different scales and directions; the pooling layer is used to downsample the data, reducing the data volume while retaining the main features, improving the computational efficiency and generalization ability of the model; the fully connected layer integrates the features extracted previously and outputs the final prediction result.

[0199] When training this network, a large number of process-performance datasets are used. These datasets contain the flow field, temperature field, and stress field data corresponding to optical lenses produced under different process conditions (such as different injection pressures, temperatures, melt viscosities, etc.), as well as the refractive index data obtained from actual measurements corresponding to them. Using the flow field, temperature field, and stress field data in these datasets as inputs and the corresponding refractive index data as the true labels, the network is trained. During the training process, the network continuously adjusts its own parameters (such as the weights of convolution kernels, the connection weights of fully connected layers, etc.) to make the predicted refractive index distribution as close as possible to the true labels. For example, during the training process, the network will learn the influence laws of factors such as melt flow velocity and direction on the refractive index based on the input flow field information; from the temperature field data, it will master the relationship between temperature changes and the refractive index; by learning the stress field data, it can understand how the stress distribution changes the optical properties of materials and thus affects the refractive index. In this way, the network establishes a mapping relationship between the flow state and the optical performance, thereby realizing real-time quality prediction. When the new multi-physical field coupling simulation results are input into the trained network, the network can quickly and accurately predict the real-time refractive index distribution n(x, y, z, t). This prediction method based on deep learning can capture the non-linear relationship between complex physical fields and the refractive index more accurately compared with the traditional prediction methods based on empirical formulas or simple models, improving the accuracy and reliability of the prediction and providing more accurate data support for subsequent process optimization.

[0200] In step S4120, the target refractive index distribution n tgt (x, y, z) is determined by the product design model, which reflects the spatial distribution characteristics of the ideal optical performance. For example, for a specific optical lens, its design requirements specify specific refractive indices at different positions to achieve specific optical functions such as focusing and imaging. These specific refractive index distributions are the target refractive index distribution n tgt (x, y, z). The difference Δn(x, y, z, t) between the real-time predicted n(x, y, z, t) and n tgt (x, y, z) quantitatively characterizes the instantaneous deviation degree of the optical performance. If the value of Δn(x, y, z, t) is large, it indicates that there is a large difference between the optical performance of the product in the current production process and the ideal state, which may affect the final quality of the product. The double integral of Δn(x, y, z, t) in space and time gives the cumulative deviation amount Δn total , because the quality of an optical lens is an index that comprehensively considers spatial and time factors. Spatially, the refractive index deviations at different positions will all affect the overall optical performance; temporally, the changes in the refractive index during the entire injection molding process will also accumulate and affect the product quality. The Δn obtained through the double integral totalIt can more comprehensively evaluate the overall quality level of the product. For example, at a certain moment, the refractive index deviation in a certain area of the product may be small, but if the deviation in this area persists throughout the production process, the Δn obtained through double integration total will be large, indicating that there are relatively serious quality problems in this area of the product and process adjustments are needed. This method of evaluating product quality by calculating the cumulative deviation amount can more accurately reflect the actual situation of product quality and provide a more reasonable basis for subsequent process optimization.

[0201] Step S4130 sets a deviation threshold θn. Based on Δn total and θn, different process optimization strategies are triggered for the classification results of different flow anomaly patterns. The deviation threshold θn is a reference value set according to the quality requirements of the product and production experience. For the first-level flow pattern, if Δn total < θn, the current process parameters remain unchanged. This is because in the first-level flow pattern, the material flow is relatively stable. When the cumulative deviation amount is small, it indicates that the current process parameters can meet the product quality requirements and no adjustment is needed. This can avoid unnecessary process changes, ensure the stability and consistency of the production process, and reduce production costs. For example, when producing a batch of optical lenses, the calculated Δn total is less than the set θn. At this time, continuing to maintain the current process parameters such as injection pressure and temperature can ensure the stable quality of the subsequent produced lenses and reduce the quality fluctuations that may be caused by parameter adjustments.

[0202] For the second-level flow pattern or in the first-level flow pattern, if Δn total ≥ θn (for example, 0.05, where 0.05 is an example value and can be determined according to specific products and processes), step S4200 is entered to trigger the optimization of process parameters based on global sensitivity analysis. This is because when the cumulative deviation amount reaches a certain level, it indicates that the current process parameters may need to be adjusted to improve product quality. Global sensitivity analysis can determine which process parameters have a greater impact on product quality (measured by refractive index deviation here), and thus targetedly adjust these parameters. For example, through global sensitivity analysis, it is found that temperature has a greater impact on refractive index deviation. Then, the process parameters related to temperature, such as mold temperature and melt temperature, can be focused on adjusting to reduce refractive index deviation and improve the optical performance of the product. This data- and analysis-based process optimization method can find the optimization direction more efficiently compared to the traditional trial-and-error method, reduce the blindness of process adjustment, and improve production efficiency and product quality.

[0203] For the three-level flow mode, step S4300 is entered to trigger a strong and robust adaptive parameter control strategy. The flow instability in the three-level flow mode is relatively high, and the product quality problem is relatively serious, requiring more powerful control means. The strong and robust adaptive parameter control strategy can quickly and flexibly adjust the process parameters according to real-time production data, suppress the violent fluctuations of the flow, and restore the production process to stability as soon as possible to ensure product quality. For example, in the three-level flow mode, the flow state of the melt is monitored in real time by sensors. When it is found that the flow fluctuates violently, the strategy can quickly adjust the injection pressure, speed and other parameters to restore the melt flow to relative stability, thereby avoiding serious defects in the product and improving the product qualification rate. This method of adopting different process optimization strategies for different flow modes and quality deviations can be accurately controlled according to the actual situation of the production process, effectively improve the production quality and production efficiency of optical lenses, reduce the scrap rate, and enhance the competitiveness of enterprises in the market.

[0204] Step S4200, process parameter optimization based on global sensitivity analysis;

[0205] Further, step S4200 includes:

[0206] Step S4210, constructing a quadratic response surface model based on the deviation Δn(x, y, z, t) and multi-physics field coupling simulation results;

[0207] like Figure 6 As shown, further, step S4210 includes:

[0208] Step S4211, using the Sobol method to perform global sensitivity analysis, obtain the temperature sensitivity coefficient ST of the deviation Δn(x, y, z, t) to the temperature field T(x, y, z, t), and the pressure sensitivity coefficient SP of the deviation Δn(x, y, z, t) to the pressure field P(x, y, z, t);

[0209] Step S4212, using ST and SP as independent variables, Δn total As the dependent variable, a quadratic response surface model Δn was fitted by least squares regression. total =f(ST,SP).

[0210] Step S4220: d (t) and T d (t), and (ΔP pred , ΔT pred ) is input into the quadratic response surface model to obtain the quality deviation prediction function Δn of the process parameter adjustment ΔP and ΔT total (ΔP, ΔT), where ΔP is the pressure optimization adjustment amount, and ΔT is the temperature optimization adjustment amount;

[0211] Calculate P d (t) and T d The corresponding current pressure sensitivity coefficient SP1 and current temperature sensitivity coefficient ST1 of (t);

[0212] Superimpose the process control strategy (ΔP pred , ΔT pred ) with the current pressure sensitivity coefficient SP1 and current temperature sensitivity coefficient ST1 to obtain the adjusted temperature sensitivity index ST adj and the adjusted pressure sensitivity index SP adj ;

[0213] Substitute ST adj and SP adj into the quadratic response surface model to obtain the mass deviation prediction value Δn total,adj .

[0214] Step S4230, obtain the quality target, calculate the mass deviation Δm between the mass deviation prediction value Δn total,adj and the quality target, evaluate the optimization effect of the process control strategy (ΔP pred , ΔT pred ), if the optimization effect does not meet the requirements, further fine-tune the process control strategy (ΔP pred , ΔT pred ).

[0215] Specifically, step S4200 focuses on the optimization of process parameters based on global sensitivity analysis. Its purpose is to achieve precise adjustment of process parameters by deeply analyzing the influence of process parameters on product quality, thereby improving product quality. Sub-step S4210 aims to construct a quadratic response surface model based on the deviation Δn(x, y, z, t) and the multi-physical field coupling simulation results. The Sobol method is a global sensitivity analysis method used to evaluate the influence degree of model input variables on the output results. By calculating the Sobol index of variables, the contribution of each input variable to the output is quantified. In this embodiment, the temperature sensitivity coefficient ST of the deviation Δn(x, y, z, t) to the temperature field T(x, y, z, t) and the pressure sensitivity coefficient SP of the deviation Δn(x, y, z, t) to the pressure field P(x, y, z, t) are obtained using the Sobol method. These two coefficients are key indicators for quantifying the influence of temperature and pressure factors on the refractive index deviation in the entire cavity, reflecting the sensitivity of optical performance to temperature and pressure changes during the molding process. For example, if the temperature sensitivity coefficient ST is large, it indicates that a small change in the temperature field T(x, y, z, t) will have a large impact on the refractive index deviation Δn(x, y, z, t), meaning that attention needs to be paid to the adjustment of temperature parameters during process optimization. Taking ST and SP as independent variables, Δn totalis the dependent variable, and a quadratic response surface model Δn is fitted through least - squares regression total = f(ST, SP). Least - squares regression is a commonly used curve - fitting method. By minimizing the sum of the squares of the errors between the observed values and the model - predicted values, it finds the functional relationship that best represents the data trend. This quadratic response surface model establishes a non - linear mapping relationship between the optical quality deviation and the process parameter sensitivity (i.e., product quality), and can intuitively show how the temperature and pressure sensitivity coefficients affect the overall quality level of the product. For example, through this model, it can be clearly seen how the product quality deviation Δn total will change when the temperature sensitivity coefficient ST increases within a certain range, providing a theoretical basis for subsequent process parameter adjustment.

[0216] Sub - step S4220 inputs the previously obtained pressure fluctuation value sequence P d (t), temperature gradient sequence T d (t), and the process control strategy (ΔP pred , ΔT pred ) obtained according to the flow - process mapping model into the quadratic response surface model, so as to obtain the quality deviation prediction function Δn total (ΔP, ΔT) of the process parameter adjustment amounts ΔP and ΔT. This step comprehensively considers the current process state (reflected by P d (t) and T d (t)) and the preliminary process adjustment strategy (ΔP pred , ΔT pred ), further refining the prediction of the product quality deviation and providing a more accurate basis for evaluating the effectiveness of the process control strategy.

[0217] In sub - step S4230, the quality target of the product is obtained, and the quality deviation Δm between the quality deviation prediction value Δn total,adj and the quality target is calculated to evaluate the optimization effect of the process control strategy (ΔP pred , ΔT pred ). If the optimization effect does not meet the requirements, the process control strategy (ΔP pred , ΔT pred ) is further fine - tuned. For example, when the quality deviation Δm exceeds the allowable range, it indicates that the current process control strategy fails to effectively improve the product quality. According to the quadratic response surface model and the sensitivity analysis results, the pressure adjustment amount ΔP pred and the temperature adjustment amount ΔT pred are fine - tuned, such as appropriately increasing or decreasing the pressure adjustment amount and fine - tuning the amplitude of the temperature adjustment, until the quality deviation Δm meets the quality target requirements.

[0218] The beneficial effects of this step are reflected in multiple aspects. From the perspective of product quality improvement, by accurately analyzing the impact of temperature and pressure on product quality and optimizing process parameters accordingly, it is possible to effectively reduce product quality deviations and improve the consistency and stability of the optical performance of optical lenses. In terms of production efficiency, it avoids the time waste and resource consumption caused by blindly adjusting process parameters. By making adjustments based on scientific sensitivity analysis and model prediction, it is possible to quickly find a better combination of process parameters and shorten the production cycle. From the perspective of cost control, it reduces the scrap rate caused by unqualified product quality and lowers production costs. In addition, this data-driven process optimization method provides strong support for the continuous improvement of the optical lens production process and helps enterprises gain an advantageous position in the market competition.

[0219] Step S4300, a strong-robust adaptive parameter regulation strategy.

[0220] Specifically, step S4300 implements a strong-robust adaptive parameter regulation strategy for the three-stage flow pattern, which is a key link to ensure the production quality of optical lenses under complex flow conditions. In the three-stage flow pattern, the flow of materials in the cavity is highly unstable, and traditional process regulation methods are difficult to effectively solve the problem. Therefore, a more intelligent and adaptive regulation strategy needs to be adopted.

[0221] In actual operation, based on the electromagnetic induction intensity field E(x, y, z, t) in the three-stage flow pattern obtained in step S3530, the power spatial distribution of the induction heating system is optimized. The electromagnetic induction intensity field E(x, y, z, t) reflects the magnetic field intensity at different positions in the cavity under the action of electromagnetic induction. By analyzing this distribution, it is possible to determine which areas require more or less energy input, and then make targeted adjustments to the power spatial distribution of the induction heating system. For example, if the electromagnetic induction intensity is weak in a certain area, resulting in unstable material flow, the power of the induction heating system in this area can be appropriately increased to raise the material temperature, reduce the melt viscosity, improve the flow performance, and promote uniform material flow.

[0222] Performing high-frequency dynamic voltage regulation on the key parts with abnormal three-stage flow is also an important measure in this step. The key parts usually refer to the areas where turbulence, large velocity gradients, or abnormal pressures are likely to occur during the flow process, and these parts have a significant impact on product quality. High-frequency dynamic voltage regulation means quickly adjusting the voltage within a short time to adapt to the complex flow requirements of materials at the key parts. For example, when it is monitored that the material flow rate is too fast at a certain key part, which may lead to defects, the voltage at this part is reduced through high-frequency dynamic voltage regulation to reduce the induction heating power, lower the material temperature, increase the melt viscosity, and thus slow down the flow rate and stabilize the flow state.

[0223] Establishing a thermal-fluid-stress-electromagnetic multi-field coupling control loop is the core of achieving strong and robust adaptive parameter regulation. The multi-field coupling control loop correlates multiple physical processes such as heat conduction, material flow, stress distribution, and electromagnetic induction to form a closed-loop control system. In this system, the changes in each physical field are monitored in real time, and the process parameters are dynamically adjusted according to the monitoring data. For example, when it is detected that the change in the thermal field affects the flow state of the material, which in turn leads to abnormal stress distribution, the control loop will automatically adjust parameters such as the power and voltage of the induction heating system according to the results of multi-physical field coupling simulation and the preset control strategy, change the thermal field distribution, and then affect the viscosity and flow velocity of the material, and finally adjust the stress distribution to restore the entire system to a stable state.

[0224] The beneficial effects of this step are significant. In terms of product quality assurance, through the precise regulation of complex flow states, the violent fluctuations of the flow are effectively suppressed, and defects caused by unstable flow, such as bubbles, weld lines, and internal stress concentration, are reduced, improving the quality and yield rate of optical lenses. In terms of process stability, the multi-field coupling control loop realizes the real-time monitoring and dynamic adjustment of the process, enhancing the anti-interference ability of the process. Even when the production conditions fluctuate, the stability of product quality can be guaranteed. From the perspective of production efficiency improvement, it avoids production interruptions and rework caused by frequent quality problems, improving production efficiency. At the same time, this advanced regulation strategy helps to promote the development of optical lens production technology towards intelligence and refinement, enhances the core competitiveness of enterprises, and provides new ideas and methods for the technological progress of the industry.

[0225] Example 2

[0226] Based on Example 1, this example provides a full-process quality monitoring device in the production process of optical lenses, as Figure 7 shown, including:

[0227] Flow anomaly detection module: used to obtain the first deformation parameter set and the second excitation parameter set, perform wavelet packet decomposition on the first deformation parameter set, and extract the high-frequency components of the n1 layer as the flow front feature vector F f ; generate a flow vector matrix M according to the second excitation parameter set v ; establish a flow field classification rule according to the flow front feature vector F f and the flow vector matrix M v , and obtain the classification result of the flow anomaly mode; the classification result of the flow anomaly mode includes the first-level flow mode, the second-level flow mode, and the third-level flow mode;

[0228] Process control strategy generation module: used to retrieve the historical production data set of n2 batches, and based on the historical production data set, construct a flow-process mapping model using a generative adversarial network; according to the flow front feature vector F f , the flow vector matrix M v and the trained flow-process mapping model to obtain a process control strategy;

[0229] Simulation module: used to obtain the melt viscosity η0 in the mold cavity and the heat transfer coefficient h0 between the melt and the mold; based on η0 and h0, obtain the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity; according to the three-dimensional flow field distribution v0(x, y, z, t), obtain the accurate three-dimensional transient flow field distribution v(x, y, z, t); according to the classification results of the flow anomaly patterns, perform multi-physical field coupling simulation on the accurate three-dimensional transient flow field distribution v(x, y, z, t) to obtain the multi-physical field coupling simulation results;

[0230] Process optimization module: according to the multi-physical field coupling simulation results and the process control strategy, trigger different process optimization strategies for different classification results of the flow anomaly patterns.

[0231] In the flow anomaly detection module, the obtaining of the first deformation parameter set and the second excitation parameter set includes:

[0232] Step S1110, within 0.5 s after the mold is closed, collect the surface deformation time series data through a distributed fiber Bragg grating array to generate the first deformation parameter set, and the first deformation parameter set includes the pressure fluctuation value sequence P d (t) and the temperature gradient sequence T d (t);

[0233] Step S1120, synchronously start the ultrasonic guided wave transmitter, apply a directional excitation wave along the melt flow direction, and detect the reflected wave phase difference through the Doppler effect to generate the second excitation parameter set, and the second excitation parameter set includes the x-axis flow velocity distribution V e (x) and the y-axis shear stress distribution τ e (y).

[0234] In the flow anomaly detection module, the wavelet packet decomposition of the first deformation parameter set and the extraction of the high-frequency components of the n1 layer as the flow front feature vector F f includes:

[0235] Step S1211, use the Daubechies8 wavelet basis function to perform 8-layer wavelet packet decomposition on the pressure fluctuation value sequence P d (t);

[0236] Step S1212, in the high-frequency sub-band coefficients of the n1 layer, extract the wavelet energy spectrum peak value and the corresponding scale index;

[0237] Step S1213: Construct a 16-dimensional flow front feature vector F from the extracted peak values of the wavelet energy spectrum and the corresponding scale indices. f 。

[0238] In the process control strategy generation module, the construction of the flow-process mapping model using the adversarial generative network based on the historical production data set includes:

[0239] Step S2210: Use the historical flow front feature vector F f , hist and the historical flow vector matrix M v,hist as the input X, and use the corresponding historical pressure optimization amount ΔP hist and the historical temperature optimization amount ΔT hist as the output Y to form a training data pair (X, Y);

[0240] Step S2220: Build a flow-process mapping model and train the flow-process mapping model according to the training data pair (X, Y).

[0241] The flow-process mapping model includes a generator G and a discriminator D. Construct a generator loss function L G and a discriminator loss function L D , and alternately train the generator G and the discriminator D by minimizing the generator loss function L G and the discriminator loss function L D until the Nash equilibrium is reached.

[0242] In the simulation module, the obtaining of the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity includes:

[0243] Obtain the melt viscosity η0 in the mold cavity and the heat transfer coefficient h0 between the melt and the mold; the heat transfer coefficient h0 is obtained by inverse calculation from the temperature gradient sequence T d (t) and the pre-constructed three-dimensional transient heat transfer equation;

[0244] Based on η0 and h0, use the pre-constructed generalized Newtonian fluid constitutive equation and three-dimensional transient flow control equation, and numerically solve by the finite volume method to obtain the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity.

[0245] In the process optimization module, the triggering of different process optimization strategies includes:

[0246] Step S4110: Input the multi-physical field coupling simulation results into the pre-trained refractive index prediction network to obtain the real-time refractive index distribution n(x, y, z, t);

[0247] Step S4120: Determine the target refractive index distribution n tgt (x, y, z), calculate the deviation Δn(x, y, z, t) between n(x, y, z, t) and the target refractive index distribution n tgt (x, y, z), and perform a double integral of Δn(x, y, z, t) over space and time to obtain the cumulative deviation amount Δn total ;

[0248] Step S4130: Set the deviation threshold θn. Based on Δn total and θn, for the classification results of different flow anomaly patterns, trigger different process optimization strategies.

[0249] For the first-level flow pattern, if Δn total < θn, then maintain the current process control strategy (ΔP pred , ΔT pred ) unchanged;

[0250] For the second-level flow pattern or the first-level flow pattern, if Δn total ≥ θn, then enter Step S4200 to trigger the optimization of process parameters based on global sensitivity analysis;

[0251] For the third-level flow pattern, then enter Step S4300 to trigger a strong robust adaptive parameter regulation strategy.

[0252] The methods and apparatuses of the present application may be implemented in many ways. For example, the methods and apparatuses of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is only for illustration, and the steps of the method of the present application are not limited to the above specific described order unless otherwise specifically stated.

[0253] In addition, parts of the above technical solutions provided in the embodiments of the present application that are consistent with the corresponding technical solutions in the prior art in terms of implementation principles are not described in detail to avoid excessive elaboration.

[0254] As described above in the specific embodiments, the purpose, technical solutions, and beneficial effects of the present invention are further described in detail. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A full-process quality monitoring method in the production process of an optical lens, characterized in that The method includes: Obtain the first deformation parameter set and the second excitation parameter set, perform wavelet packet decomposition on the first deformation parameter set, and extract the high-frequency components of the n1 layer as the flow front feature vector F f ; Generate a flow vector matrix M according to the second excitation parameter set v ; According to the flow front feature vector F f and the flow vector matrix M v , establish a flow field classification rule to obtain the classification result of the flow anomaly pattern; the classification result of the flow anomaly pattern includes the first-level flow pattern, the second-level flow pattern, and the third-level flow pattern; The obtaining of the first deformation parameter set and the second excitation parameter set includes: Within 0.5 s after the mold is closed, collect the surface deformation time-series data through a distributed fiber Bragg grating array to generate a first deformation parameter set, where the first deformation parameter set includes a pressure fluctuation value sequence P d (t) and a temperature gradient sequence T d (t); Synchronously start the ultrasonic guided wave transmitter, apply a directional excitation wave along the melt flow direction, detect the reflected wave phase difference through the Doppler effect, and generate a second excitation parameter set, where the second excitation parameter set includes an x-axis flow velocity distribution V e (x) and a y-axis shear stress distribution τ e (y); Retrieve the historical production data set of the n2 batch. Based on the historical production data set, use a generative adversarial network to construct a flow-process mapping model; according to the flow front feature vector F f , the flow vector matrix M v and the trained flow-process mapping model, obtain the process control strategy; Obtaining the melt viscosity η0 in the mold cavity and the heat transfer coefficient h0 between the melt and the mold; based on η0 and h0, obtaining the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity; according to the three-dimensional flow field distribution v0(x, y, z, t), obtaining the accurate three-dimensional transient flow field distribution v(x, y, z, t); according to the classification results of the flow anomaly patterns, performing multi-physical field coupling simulation on the accurate three-dimensional transient flow field distribution v(x, y, z, t) to obtain the multi-physical field coupling simulation results; According to the multi-physical field coupling simulation results and the process control strategy, different process optimization strategies are triggered for different classification results of the flow anomaly patterns.

2. The full-process quality monitoring method in the production process of an optical lens according to claim 1, wherein, Wavelet packet decomposition is performed on the first set of deformation parameters, and the high-frequency components of the n1 layer are extracted as the flow front feature vector F f including: Using the Daubechies8 wavelet basis function, the pressure fluctuation value sequence P d (t) is decomposed by 8-layer wavelet packet decomposition; in the high-frequency subband coefficients of the n1-th layer, the peak value of the wavelet energy spectrum and the corresponding scale index are extracted; the extracted peak value of the wavelet energy spectrum and the corresponding scale index are used to form a 16-dimensional flow front feature vector F f ; where n1 is a positive integer, and its value range is [5, 8].

3. The full-process quality monitoring method in the production process of the optical lens according to claim 2, characterized in that, The generation of the flow vector matrix M v includes: the x-axis velocity distribution V e (x) and the y-axis shear stress distribution τ e (y) are respectively used as the row vector and column vector of the flow vector matrix M v of.

4. The full-process quality monitoring method in the production process of an optical lens according to claim 1, wherein The establishing of the flow field classification rules to obtain the classification results of the flow anomaly patterns includes: When the flow front feature vector F f ≥ 0.7 and the flow vector matrix M v The principal component included angle θ < 15°, it is determined as the first-level flow pattern, and the principal component included angle θ is the direction of the first principal component of M v ; When 0.4 ≤ F f < 0.7 and 15° ≤ θ < 30°, it is determined as the secondary flow pattern; Determining the remaining cases as the third-level flow pattern.

5. The full-process quality monitoring method in the production process of an optical lens according to claim 1, characterized in that The historical production data set includes historical flow field characteristic quantities and corresponding process parameter optimization quantities; the historical flow field characteristic quantities include a historical flow front feature vector F f,hist and a historical flow vector matrix M v,hist ; the process parameter optimization quantities include a historical pressure optimization quantity ΔP hist and a historical temperature optimization quantity ΔT hist ; The constructing of the flow-process mapping model using the generative adversarial network includes: Take the historical flow front feature vector F f,hist and the historical flow vector matrix M v,hist as the input X, and take the corresponding historical pressure optimization amount ΔP hist and the historical temperature optimization amount ΔT hist as the output Y to form a training data pair (X, Y); build a flow-process mapping model and train the flow-process mapping model according to the training data pair (X, Y).

6. The full-process quality monitoring method in the production process of an optical lens according to claim 1, characterized in that The obtaining of the process control strategy includes: If it is a secondary flow mode or a tertiary flow mode, the flow front feature vector F f and the flow vector matrix M v are input into the trained flow-process mapping model to obtain the pressure adjustment amount ΔP pred and the temperature adjustment amount ΔT pred ; If it is a first-level flow mode, then ΔP pred = 0, ΔT pred = 0; Combine ΔP pred and ΔT pred to form a process control strategy (ΔP pred , ΔT pred ).

7. The full-process quality monitoring method during the production process of an optical lens according to claim 2, characterized in that, The heat transfer coefficient h0 is obtained by inverse calculation from the temperature gradient sequence T d (t) and a pre-established three-dimensional transient heat transfer equation; The obtaining of the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity includes: based on η0 and h0, using the pre-constructed generalized Newtonian fluid constitutive equation and the three-dimensional transient flow control equation, and numerically solving by the finite volume method to obtain the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity.

8. The full-process quality monitoring method in the production process of an optical lens according to claim 7, characterized in that, The obtaining of the accurate three-dimensional transient flow field distribution v(x, y, z, t) according to the three-dimensional flow field distribution v0(x, y, z, t) includes: Match the x-axis velocity distribution V e (x) with v0(x, y, z, t), correct η0 and h0, and obtain the optimized melt viscosity η opt and the optimized heat transfer coefficient h opt ; Substitute the optimized melt viscosity η opt and the optimized heat transfer coefficient h opt into the generalized Newtonian fluid constitutive equation and the three-dimensional transient flow control equation, and perform numerical solution again to obtain the accurate three-dimensional transient flow field distribution v(x, y, z, t) that matches the actual flow state.

9. The full-process quality monitoring method in the production process of an optical lens according to claim 8, characterized in that, The matching of the x-axis velocity distribution V e (x) with v0(x, y, z, t) and the correction of η0 and h0 include: For the x-axis flow velocity distribution V e (x) and v0(x, y, z, t), N matching points are extracted on the y0-z0 cross-section of V e (x), and a least-squares objective function for the velocity deviation is established; Iteratively optimize and solve the least-squares objective function based on the gradient descent method to obtain the optimized melt viscosity η opt and the optimized heat transfer coefficient h opt .

10. The full-process quality monitoring method in the production process of an optical lens according to claim 6, characterized in that, The triggering of different process optimization strategies for different classification results of the flow anomaly patterns includes: Inputting the multi-physical field coupling simulation results into the pre-trained refractive index prediction network to obtain the real-time refractive index distribution n(x, y, z, t); Determine the target refractive index distribution n tgt (x, y, z), and calculate the deviation Δn(x, y, z, t) between n(x, y, z, t) and the target refractive index distribution n tgt (x, y, z). Perform a double integral of Δn(x, y, z, t) over space and time to obtain the cumulative deviation amount Δn total ; Set the deviation threshold θn according to Δn total Based on total and θn, for the classification results of different flow anomaly patterns, trigger different process optimization strategies.

11. The full-process quality monitoring method during the production process of an optical lens according to claim 10, wherein Said according to Δn total and θn, for the grading results of different flow anomaly patterns, trigger different process optimization strategies including: For the primary flow mode, if Δn total < θn, then maintain the current process control strategy (ΔP pred , ΔT pred ) unchanged; For the secondary flow pattern or the primary flow pattern, if Δn total ≥ θn, then the optimization of process parameters based on global sensitivity analysis is triggered; For the third-level flow pattern, an adaptive parameter regulation strategy is triggered.

12. An all-process quality monitoring device in the production process of an optical lens, which is used to implement the all-process quality monitoring method in the production process of the optical lens described in any one of claims 1-11, characterized in that, The device includes: Flow anomaly detection module: used to obtain the first deformation parameter set and the second excitation parameter set, perform wavelet packet decomposition on the first deformation parameter set, and extract the high-frequency components of the n1 layer as the flow front feature vector F f ; generate a flow vector matrix M according to the second excitation parameter set v ; according to the flow front feature vector F f and the flow vector matrix M v , establish a flow field classification rule to obtain the classification result of the flow anomaly pattern; the classification result of the flow anomaly pattern includes the first-level flow pattern, the second-level flow pattern, and the third-level flow pattern; Process control strategy generation module: used to retrieve the historical production data set of batch n2, and based on the historical production data set, construct a flow-process mapping model using a generative adversarial network; according to the flow front feature vector F f , the flow vector matrix M v and the trained flow-process mapping model to obtain a process control strategy; A simulation module: used to obtain the melt viscosity η0 in the mold cavity and the heat transfer coefficient h0 between the melt and the mold; based on η0 and h0, obtain the three-dimensional flow field distribution v0(x, y, z, t) of the melt in the mold cavity; according to the three-dimensional flow field distribution v0(x, y, z, t), obtain the accurate three-dimensional transient flow field distribution v(x, y, z, t); according to the classification results of the flow anomaly patterns, perform multi-physical field coupling simulation on the accurate three-dimensional transient flow field distribution v(x, y, z, t) to obtain the multi-physical field coupling simulation results; A process optimization module: according to the multi-physical field coupling simulation results and the process control strategy, trigger different process optimization strategies for different classification results of the flow anomaly patterns.

Images