Spectacle case stamping forming collaborative control method and system based on multi-process fusion

Abstract

The application relates to the technical field of industrial automation control, and discloses a glasses box stamping forming collaborative control method and system based on multi-process fusion, which comprises the following steps: acquiring a material falling process device torque signal and a mold vibration signal; constructing a material falling signal time sequence according to the signal and performing time sequence prediction to obtain a size deviation; adjusting a forming process mold closing pressure according to the size deviation; acquiring a signal of the forming process under the target mold closing pressure and constructing a forming signal time sequence to perform state identification; when the state is abnormal, determining a trimming beat adjustment amount according to the abnormal state and a preset mechanical mapping relationship, acquiring a trimming process signal and constructing a trimming signal time sequence to perform quality evaluation; and if the quality evaluation value is less than a preset evaluation threshold value, generating a shutdown control instruction to control each process to stop running. The method can realize collaborative control of glasses box stamping forming multi-processes, and improve product precision and the qualified rate.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control technology, and in particular to a collaborative control method and system for the stamping and forming of eyeglass cases based on multi-process integration. Background Technology

[0002] Currently, the stamping and forming production of eyeglass cases typically involves multiple continuous processes, including blanking, forming, and trimming. The quality of each process is passed down sequentially and ultimately determines the finished product's pass rate. As a precision metal product, eyeglass cases require dimensional tolerances of ±0.05mm for hinge mating surfaces and ±0.08mm for flange contours. While the overall dimensional tolerance can be relaxed to ±0.2mm, the high precision requirements of functional components mean that even a slight deviation in any process can lead to product failure. In high-speed continuous stamping (SPM≥60) scenarios, ensuring stable collaboration between processes in the industrial control system under high-cycle production has become crucial for manufacturing enterprises to improve product yield.

[0003] In a current technology, traditional stamping production lines treat blanking, forming, and trimming as independent workstations. Each process operates independently according to pre-set fixed parameters, with material transfer between processes relying solely on the predetermined rhythm of the conveyor belt. However, during actual stamping, equipment status signals such as the current and torque of the servo motor and the vibration frequency of the die under stress can subtly shift due to factors such as batch differences in materials, die wear, and lubrication variations. In the current technology, these changes are not incorporated into the information transmission between processes, and there is no real-time signal interaction between processes. As a result, any minor deviations in upstream processes cannot be detected and adjusted in time by downstream processes. Downstream processes continue processing according to the original parameters and rhythm, thus amplifying the deviations at each stage.

[0004] In summary, existing technologies suffer from the problem that the dynamic parameters and cycle time adjustments of upstream and downstream processes cannot be coordinated with real-time equipment status signals, causing deviations to accumulate and amplify step by step along the production line. Summary of the Invention

[0005] This invention provides a collaborative control method and system for eyeglass case stamping based on multi-process integration, in order to solve the problem in the prior art that the dynamic parameters and cycle time adjustments of upstream and downstream processes cannot be coordinated with the real-time equipment status signals, causing deviations to accumulate and amplify step by step along the production line.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a collaborative control method for eyeglass case stamping based on multi-process fusion, comprising: Acquire equipment torque signals and mold vibration signals during the blanking process; Based on the equipment torque signal and the mold vibration signal, a blanking signal time series is constructed, and time series prediction is performed on the blanking signal time series to obtain the dimensional deviation of the blanking process; Adjust the mold closing pressure of the molding process according to the dimensional deviation to obtain the target mold closing pressure; The equipment torque signal and mold vibration signal of the molding process under the target mold closing pressure are obtained, and a molding signal time series is constructed. The state of the molding signal time series is identified to obtain the state identification result. When the state identification result is an abnormal state, the trimming cycle adjustment amount is determined according to the abnormal state and combined with the preset mechanical mapping relationship to adjust the running cycle of the trimming process. The equipment torque signal and mold vibration signal of the trimming process under the adjusted running cycle are obtained, and the trimming signal time series is constructed. The trimming signal time series is quality evaluated to obtain the trimming quality evaluation value. If the edge trimming quality assessment value is less than the preset assessment threshold, a shutdown control command is generated based on the edge trimming quality assessment value to control the blanking process, the forming process and the edge trimming process to stop running.

[0007] Secondly, the present invention provides a collaborative control system for stamping and forming of eyeglass cases based on multi-process integration, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0008] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention acquires the equipment torque signal and mold vibration signal of the blanking process, constructs the blanking signal time series, and uses a deep neural network to learn the time-dependent features and predict the size deviation, thereby realizing the real-time online quantitative perception of blanking size deviation. This solves the problem that blanking deviation in traditional stamping production lines can only be detected by offline sampling inspection, which leads to feedback lag. Finally, it realizes the online dynamic monitoring of the processing status of the blanking process, providing reliable data basis for the active compensation of downstream processes.

[0010] (2) By establishing a deviation resistance mapping relationship between dimensional deviation and mold closing pressure, the present invention transforms blanking deviation into target mold closing pressure in the forming process in a feedforward manner, thereby realizing active compensation of blanking deviation in the forming process; and further, by linking the forming process state recognition with the preset mechanical mapping relationship, the cutting cycle adjustment amount is dynamically calculated to adapt to the abnormal forming state, which solves the problem in the prior art that the downstream process is unaware of the upstream deviation, resulting in the deviation being amplified step by step, and finally realizes the adaptive adjustment of the forming process to the upstream material fluctuation, significantly improving the stamping production line's adaptability to material batch differences and mold state changes.

[0011] (3) This invention achieves quantitative evaluation of edge cutting quality by extracting features of edge cutting signals in multiple stages, reducing principal components and calculating the degree of quality deviation. When the evaluation value is lower than the preset threshold, a stop control command is generated to stop the three processes of material feeding, forming and edge cutting simultaneously. This solves the problem that quality abnormalities in traditional production lines can only be discovered by sampling of finished products, resulting in batches of scrap. Finally, it realizes real-time perception of quality abnormalities in the stamping process and rapid interlocking of multiple processes, effectively avoiding the generation of batches of scrap and improving the finished product accuracy and pass rate of eyeglass case stamping. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the collaborative control method for eyeglass case stamping based on multi-process fusion provided in the first embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Reference Figure 1 The first embodiment of the present invention provides a collaborative control method for the stamping and forming of eyeglass cases based on multi-process fusion, including the following steps: S1, acquire the equipment torque signal and mold vibration signal of the blanking process; S2, construct a blanking signal time series based on the equipment torque signal and the mold vibration signal, perform time series prediction on the blanking signal time series, and obtain the dimensional deviation of the blanking process; S3, adjust the mold closing pressure of the molding process according to the dimensional deviation to obtain the target mold closing pressure; S4, acquire the equipment torque signal and mold vibration signal of the molding process under the target mold closing pressure, construct the molding signal time series, perform state identification on the molding signal time series, and obtain the state identification result; S5, when the state identification result is an abnormal state, the trimming cycle adjustment amount is determined according to the abnormal state and combined with the preset mechanical mapping relationship to adjust the running cycle of the trimming process, the equipment torque signal and mold vibration signal of the trimming process under the adjusted running cycle are obtained, and the trimming signal time series is constructed. The trimming signal time series is quality evaluated to obtain the trimming quality evaluation value. S6, if the edge trimming quality assessment value is less than the preset assessment threshold, then a stop control command is generated based on the edge trimming quality assessment value to control the blanking process, the forming process and the edge trimming process to stop running.

[0015] In step S1, the equipment torque signal and mold vibration signal of the blanking process are acquired.

[0016] In one implementation, a torque sensor mounted on the main spindle of the stamping press collects the equipment torque signal during the blanking process, and a piezoelectric accelerometer / vibration sensor mounted on the die base collects the die vibration signal. The sampling frequency of the torque sensor is set to 1000 Hz, and the sampling frequency of the vibration sensor is set to 5000 Hz. The collected signals are time-series data, denoted as the original torque time series and the original vibration time series, respectively.

[0017] It should be noted that the sampling frequency is set based on the Nyquist sampling theorem, which states that the sampling frequency must be at least twice the highest frequency component of the signal to avoid spectral aliasing and signal distortion during sampling. In the stamping blanking process, the torque signal mainly reflects the load changes of the servo motor, and its frequency components are concentrated within the periodic variation range of the stamping stroke. Through actual measurement and analysis of the torque signals of multiple batches of stamping processes using a spectrum analyzer during the debugging phase, it was found that the signal energy in the 0 to 200 Hz frequency band accounts for more than 95% of the total energy, while the energy in the frequency band above 200 Hz accounts for less than 5%. Based on this, setting the sampling frequency to 1000 Hz can completely cover the harmonic components of the torque signal, while reserving sufficient bandwidth margin for possible higher harmonics, ensuring the accuracy of torque signal acquisition.

[0018] It should be further noted that the vibration signal originates from the impact response during the die punching process, and its frequency components are much higher than those of the torque signal. When the sheet metal fractures under the punching force, the impact excitation on the die base generates a broadband vibration response with effective frequency components reaching 2000 Hz, including the natural frequency of the die structure and high-frequency stress wave components during material fracture. According to the Nyquist sampling theorem, a sampling frequency of at least 4000 Hz is required to accurately acquire vibration signals below 2000 Hz. Setting the sampling frequency of the vibration sensor to 5000 Hz not only meets the basic requirements of the Nyquist theorem but also provides an additional 25% bandwidth margin, ensuring that all frequency components below 2000 Hz in the vibration signal can be completely acquired without spectral aliasing.

[0019] It is worth noting that the equipment torque signal reflects the load change of the servo motor during the stamping process, while the die vibration signal reflects the force and vibration response characteristics of the die at the moment of blanking. Together, they constitute the original signal source characterizing the stamping state of the blanking process.

[0020] In step S2, a blanking signal time series is constructed based on the equipment torque signal and the mold vibration signal, and time series prediction is performed on the blanking signal time series to obtain the dimensional deviation of the blanking process.

[0021] The process of constructing a material feeding signal time series based on the equipment torque signal and the mold vibration signal includes: The equipment torque signal and the mold vibration signal are time-aligned to obtain a synchronous torque sequence and a synchronous vibration sequence. The synchronous torque sequence and the synchronous vibration sequence are combined into a multivariate synchronous time sequence, which is used as the material dropping signal time sequence.

[0022] In one implementation, the system uses the time axis of the torque signal as a reference to resample the vibration signal. Since the sampling frequency of the torque signal is 1000 Hz and the sampling frequency of the vibration signal is 5000 Hz, the number of sampling points per unit time differs, and direct splicing would lead to time point misalignment. The specific resampling operation involves using each sampling moment of the torque signal as the target time point, finding the two closest sampling points before and after that moment in the vibration signal's time series, and calculating the corresponding vibration signal value at that moment through linear interpolation. This downsamples the vibration signal from 5000 Hz to 1000 Hz, ensuring a one-to-one numerical correspondence between the two sequences at each time point, resulting in a synchronized torque sequence and a synchronized vibration sequence. Subsequently, the system arranges the synchronized torque sequence and synchronized vibration sequence row by row in chronological order. Each row contains a sampling moment, the torque value at that moment, and the vibration value at that moment, forming a three-column multivariate synchronized time series, i.e., the material dropping signal time series.

[0023] It should be noted that the purpose of downsampling is to ensure that the torque signal and vibration signal have a consistent sampling rate in the time dimension, so as to meet the basic requirement of the subsequent time series prediction model for equally spaced sampling of input data. The downsampling process is only for the construction of the material dropping signal time series, used for time series prediction in step S2. In subsequent steps S4 and S5, the signals of the forming process and the trimming process are independently re-acquired. The vibration signal still maintains the original sampling rate of 5000 Hz for frequency domain feature extraction, and its effective frequency range can still reach 2500 Hz (Nyquist frequency), which is sufficient to cover the 2000 Hz effective frequency component of the vibration signal, and there is no problem of high-frequency information loss.

[0024] It should be further explained that the purpose of the time-series alignment process is to eliminate the sampling point misalignment problem caused by different sensors having different sampling frequencies. Although the torque sensor and vibration sensor are both triggered by the same clock source, their hardware sampling circuits are independent, and the difference in sampling frequencies makes it impossible for their sampling times to be naturally aligned. If alignment processing is not performed and the two sets of raw data are directly spliced ​​together in sequence, a chaotic situation will occur where the same timestamp corresponds to multiple sampling points, resulting in a distortion of the time series received by the subsequent time-series prediction model in the time dimension, affecting the prediction accuracy.

[0025] It should be further explained that the rationale for linear interpolation is based on the assumption that the vibration signal changes approximately linearly between adjacent sampling points. The vibration signal is sampled at 5000 Hz, and the time interval between adjacent sampling points is 0.2 milliseconds. Within such a short time interval, the amplitude change of the vibration signal is extremely small, and the error introduced by linear interpolation is negligible in engineering practice. After resampling, the sampling frequency of the synchronous vibration sequence is reduced to 1000 Hz, consistent with the torque sequence, and the values ​​of both at each time step correspond to the actual physical state at the same moment.

[0026] It is worth noting that the synchronous torque sequence and synchronous vibration sequence are combined into a multivariate synchronous time series, rather than simply added together or processed separately, because torque and vibration reflect different aspects of the physical information during the stamping process, and there is a complementary relationship between them. The peak torque is positively correlated with the thickness and hardness of the material; the thicker or harder the material, the greater the peak torque required by the servo motor during the stamping process. The spectral distribution of the vibration signal is closely related to the wear state of the die cutting edge. When the die cutting edge wears, the high-frequency impact component and the energy distribution of specific frequency bands at the moment of stamping will undergo regular changes. Both together map the trend of the profile deviation of the blanked part. The torque signal reflects the overall deformation resistance of the material, while the vibration signal reflects the local contact state of the cutting edge. By combining them into a multivariate sequence, the subsequent time series prediction model can simultaneously utilize the information from both channels to capture complex features that cannot be expressed by a single channel, thereby improving the accuracy of dimensional deviation prediction.

[0027] It should be further explained that the sampling window length is determined during the equipment debugging phase by detecting the upper and lower stop positions of the press slide using photoelectric sensors, and measuring the complete duration of a single stamping stroke. Based on this duration, the sampling window length is rounded up and an additional 20% margin is added to ensure complete coverage of the entire stamping cycle from the start of blanking to the complete retraction of the die. Once determined during equipment debugging, the sampling window length is permanently configured as a parameter in the system configuration file. When the production cycle time changes, for example, from 60 SPM to 80 SPM, the stamping cycle time shortens accordingly, requiring recalibration of the sampling window length to ensure that the collected signals always completely cover a full stamping stroke, without missing signal characteristics from critical blanking stages.

[0028] For example, within a 1-second sampling window, the torque signal contains 1000 sampling points, and the vibration signal contains 5000 sampling points. After resampling, the vibration signal is downsampled to 1000 sampling points, corresponding one-to-one with the torque signal in time. The combined material drop signal time series is a 1000-row, 3-column matrix, where the first column is the timestamp, the second column is the torque value, and the third column is the vibration value. The i-th row of the matrix corresponds to the torque and vibration synchronization data at the i-th sampling moment.

[0029] Specifically, time-series prediction is performed on the blanking signal time series to obtain the dimensional deviation of the blanking process, including: The time series of the material dropping signal is subjected to time-series feature extraction to obtain a time-dependent feature vector; The time-dependent feature vector is regressed to obtain the size deviation.

[0030] In one implementation, the system employs a pre-defined Long Short-Term Memory (LSTM) network as a temporal feature extractor. The torque and vibration values ​​of the material dropping signal time series are input into the LSM network in chronological order. The network's internal loop processes the input data step-by-step and outputs a hidden state vector at the last time step. This hidden state vector, which integrates the temporal dependency information of the entire sequence, is the temporal dependency feature vector. Then, the system inputs this temporal dependency feature vector into a fully connected regression layer. This regression layer maps the feature vector to a scalar output, which is the predicted dimensional deviation of the material dropping process, in millimeters.

[0031] It should be noted that the temporal dependency feature vector refers to the feature representation extracted from the time series that can characterize the dynamic law of the sequence's evolution over time. During the stamping process, the blanking size deviation does not suddenly form at a certain moment, but rather accumulates gradually during continuous stamping due to factors such as changes in material thickness and tool wear. Therefore, the size deviation is closely related to the signal change trend over a period of time. The Long Short-Term Memory (LSTM) network, through its internal forget gate, input gate, and output gate mechanisms, can selectively remember and forget information in the sequence, capturing dependencies over long time spans. The output hidden state vector can effectively characterize the temporal dependency features of the entire sequence.

[0032] It should be further explained that the construction method of the Long Short-Term Memory (LSTM) network involves collecting a large number of time-series samples of blanking signals under different working conditions on an actual production line or through stamping simulation experiments, and simultaneously measuring the corresponding actual size deviation value for each sample. The size deviation is measured by setting up an online visual inspection device or laser rangefinder after the blanking process to measure the size of the blanked material, obtaining a size deviation label corresponding to the signal sequence. The mean squared error is used as the loss function, and the network parameters are updated through a backpropagation algorithm until the prediction error of the model converges on the validation set.

[0033] It should be noted that the total number of training samples is no less than 10,000, covering at least 5 different material batches and 3 mold wear stages. During the training process, the samples are divided into training set and validation set in an 8:2 ratio. After each training cycle, the mean squared error loss value on the validation set is calculated. Training is stopped when the validation set loss value no longer decreases for 10 consecutive training cycles, and this model is used as the final prediction model.

[0034] It's worth noting that the regression calculation for size deviation prediction uses a fully connected regression layer instead of a classifier because size deviation is a continuous quantity, requiring a regression method to output a precise value. The fully connected regression layer uses a linear activation function, making the range of output values ​​unrestricted and covering possible positive and negative deviations. During training, the regression layer and the Long Short-Term Memory network are trained end-to-end, enabling the temporal feature extraction part to automatically learn the most discriminative temporal features for the size deviation prediction task.

[0035] For example, suppose the time series of blanking signals collected from the blanking process contains 500 sampling points, where the torque column and vibration column each constitute 500-dimensional inputs. The hidden layer dimension of the Long Short-Term Memory network is 64, and the hidden state vector of the last time step is a 64-dimensional temporal dependency feature vector. This 64-dimensional vector is input into a fully connected regression layer, the weight matrix of which is 64 rows and 1 column, and the output is a scalar. If the scalar is positive (0.05), it means that the predicted size of the blanking billet in this batch is 0.05 mm larger than the target size; if it is negative, it means that the size is smaller than the target size.

[0036] In step S3, the mold closing pressure of the molding process is adjusted according to the dimensional deviation to obtain the target mold closing pressure, including: Based on the dimensional deviation and the preset deviation resistance mapping relationship, the deformation resistance compensation gradient is determined; The pressure compensation amount is calculated based on the dimensional deviation and the deformation resistance compensation gradient. The pressure compensation amount is superimposed on the pre-acquired reference mold closing pressure to obtain the target mold closing pressure.

[0037] In one implementation, after obtaining the dimensional deviation of the blanking process, the system first calls a preset deviation resistance mapping table. This table uses the dimensional deviation value as the lookup key and the deformation resistance compensation gradient as the output value. The deformation resistance compensation gradient describes the increment of the additional mold closing pressure required for the molding process corresponding to a unit dimensional deviation, in kilonewtons per millimeter. The system looks up the corresponding gradient value in the table based on the currently predicted dimensional deviation, then multiplies the absolute value of the dimensional deviation by this gradient value to obtain the pressure compensation amount, in kilonewtons. Finally, the system reads the preset reference mold closing pressure from the current operating parameters of the molding process, adds the pressure compensation amount to the reference mold closing pressure, and obtains the target mold closing pressure.

[0038] It should be noted that the preset deviation resistance mapping relationship is a mathematical model describing the physical relationship between blanking size deviation and forming mold closing pressure compensation requirements. Its physical basis lies in the fact that when the blanking blank size deviates due to material thickness fluctuations or tool wear, the mold closing pressure in the forming process needs to be adjusted accordingly. If the blank size is too large, the material volume during forming will be too large, requiring a larger mold closing force to ensure the material fully fills the mold cavity; if the blank size is too small, the material volume will be insufficient, and excessive mold closing force will lead to excessive material flow or even flash. Therefore, a pressure compensation amount positively correlated with the size deviation needs to be superimposed on the baseline mold closing pressure. The deviation resistance mapping relationship is determined through stamping forming mechanics simulation and experimental calibration, and is fixed in the control system in tabular form.

[0039] It should be further explained that the deviation resistance mapping relationship was determined through a two-stage process of simulation pre-calibration and experimental fine calibration. In the simulation stage, a stamping model was established using Dynaform software, with sheet metal dimensional deviations covering -0.15mm to +0.15mm (step size 0.03mm, 11 levels in total). The corresponding data of deviation and required die-closing pressure were obtained by solving for each deviation. In the experimental stage, blank samples (30 pieces each, 330 pieces in total) were prepared for the above 11 deviation levels. Trial stamping was performed within the range of 90% to 110% of the reference die-closing pressure (step size 1%). The dimensions of the formed workpiece were measured, and the least squares linear regression method was used to fit the blank deviation. With mold closing pressure The relationship, the fitting equation slope in This is the compensation gradient. When the deviation between simulation and experimental results is within ±5%, the experimental value is taken as the standard; otherwise, the sample size is increased and the results are recalibrated. Finally, the compensation gradient values ​​corresponding to each deviation level are solidified into a two-dimensional table.

[0040] It should be further noted that the pressure compensation amount is calculated by multiplying the absolute value of the dimensional deviation by the gradient, rather than directly using the positive or negative sign of the deviation. This is because the adjustment direction of the mold closing pressure needs to be determined separately based on the sign of the deviation. In practical applications, if the dimensional deviation is positive (the blank is too large), the pressure compensation amount is a positive sum; if the dimensional deviation is negative (the blank is too small), the pressure compensation amount is a negative sum, indicating a reduction in the mold closing pressure. An alternative solution is to directly provide the signed pressure compensation amount based on the deviation sign in the deviation resistance mapping table, simplifying the online calculation steps.

[0041] It is worth noting that the reference mold closing pressure is the standard mold closing pressure value set under normal operating conditions for the molding process. It is determined by both mold design parameters and material properties and is usually recorded in the molding process parameter table. The target mold closing pressure obtained by adding the pressure compensation amount to the reference mold closing pressure is the mold closing pressure setting value that the molding process should use in the next stamping cycle. The system sends this target mold closing pressure to the servo pressure controller of the molding process, which takes effect in the next stamping cycle.

[0042] For example, suppose the predicted dimensional deviation of a batch of blanks is a positive value of 0.08 mm, and the deformation resistance compensation gradient corresponding to this deviation value in the deviation resistance mapping table is 12.5 kN / mm. Then the pressure compensation is 0.08 mm multiplied by 12.5 kN / mm, which equals 1 kN. The reference mold closing pressure of the current forming process is 80 kN, and after superposition, the target mold closing pressure is 81 kN. The system sends 81 kN to the forming process controller, and the mold closing pressure is adjusted to 81 kN in the next stamping cycle.

[0043] In step S4, the equipment torque signal and mold vibration signal of the molding process under the target mold closing pressure are acquired, and a molding signal time series is constructed. The state of the molding signal time series is then identified to obtain the state identification result.

[0044] Specifically, acquiring the equipment torque signal and mold vibration signal of the molding process under the target mold closing pressure, and constructing a molding signal time series, includes: Real-time equipment torque and real-time mold vibration of the molding process are collected under the target mold closing pressure; The real-time equipment torque and the real-time mold vibration are time-aligned to obtain a synchronous equipment torque sequence and a synchronous mold vibration sequence. The torque sequence of the synchronization device and the vibration sequence of the synchronization mold are spliced ​​together to obtain the molding signal time sequence.

[0045] In one implementation, after receiving the target mold-closing pressure, the controller of the forming process applies this pressure to close the mold in the next stamping cycle. During the execution of the stamping cycle, the system synchronously acquires real-time equipment torque signals and real-time mold vibration signals through a torque sensor on the forming process spindle and a vibration sensor on the mold base. The sampling frequency of the signal acquisition is consistent with that of the blanking process, with the torque signal at 1000 Hz and the vibration signal at 5000 Hz. After acquisition, the system performs time alignment processing in the same way as in step S2, resampling the vibration signal to 1000 Hz and corresponding it one-to-one with the torque signal in time to obtain a synchronous equipment torque sequence and a synchronous mold vibration sequence. Subsequently, the system splices these two synchronous sequences in chronological order to form a forming signal time sequence containing both torque and vibration channels.

[0046] It should be noted that the real-time equipment torque in the molding process reflects the load changes of the servo motor during mold closing, pressurization, and pressure holding, while the real-time mold vibration reflects the vibration response of the mold during mold closing impact and material flow. These two signals together constitute the basic data for identifying the state of the molding process. Signals acquired under the target mold closing pressure are more reflective of the actual molding effect after dimensional deviation compensation compared to signals acquired under the reference pressure. If the mold closing pressure is improperly adjusted under abnormal conditions, the torque and vibration signals during the molding process will exhibit characteristic patterns different from those under normal conditions. These characteristic patterns are key evidence for subsequent state identification.

[0047] It should be further noted that the time sequence alignment and splicing method is the same as the method for constructing the material dropping signal time series in step S2, employing linear interpolation resampling and column splicing to ensure that the forming signal time series has a consistent data format with the material dropping signal time series in both the time and channel dimensions. This consistent format allows subsequent processing flows to reuse the same feature extraction and pattern recognition algorithms, reducing system development complexity.

[0048] The process of performing state identification on the time series of the formed signal to obtain state identification results includes: Extract the statistical features in the time domain and the spectral features in the frequency domain of the shaped signal time series to obtain the time domain feature vector and the frequency domain feature vector; The time-domain feature vector and the frequency-domain feature vector are fused to obtain a fused feature vector; The similarity between the fused feature vector and each template feature vector in the preset state type is calculated, and the state type corresponding to the template feature vector with the highest similarity is selected as the state recognition result of the molding process.

[0049] In one implementation, the system first extracts time-domain features from the torque and vibration channels of the molding signal time series. The time-domain statistical features include six features: mean, peak value, root mean square, kurtosis, skewness, and standard deviation. These are extracted from both channels to obtain 12-dimensional time-domain feature vectors. Next, the system performs a Fast Fourier Transform (FFT) on both channels to convert the time-domain signals to the frequency domain, extracting frequency-domain features such as the energy concentration frequency, dominant frequency amplitude, and frequency band energy proportion. These are extracted from both channels to obtain 8-dimensional frequency-domain feature vectors. Then, the system concatenates the 12-dimensional time-domain feature vector and the 8-dimensional frequency-domain feature vector to obtain a 20-dimensional fused feature vector. Finally, the system calculates the Euclidean distance between this fused feature vector and the feature vectors of each template in a preset state type library, selecting the state type corresponding to the template with the smallest distance as the state recognition result for the molding process.

[0050] It should be noted that the aforementioned time-domain statistical features characterize the signal's distribution along the time axis. The mean reflects the signal's DC offset level, the peak value reflects the signal's maximum instantaneous impact intensity, the root mean square reflects the signal's effective energy, the kurtosis reflects the sharpness of the impact component in the signal, the skewness reflects the asymmetry of the signal distribution, and the standard deviation reflects the signal's fluctuation amplitude. These features describe the time-domain behavior of torque and vibration during the molding process from different perspectives, effectively distinguishing between normal and abnormal states.

[0051] It should be further explained that the frequency domain spectral characteristics characterize the energy distribution of the signal at different frequency components. Under normal conditions, the torque and vibration signals in the forming process exhibit the fundamental frequency and its harmonic components corresponding to the stamping cycle, with a relatively stable energy distribution. When abnormal conditions such as abnormal mold wear, material hardness fluctuations, or poor lubrication occur, new high-frequency components or abnormally increased amplitudes at specific frequencies will appear in the vibration signal spectrum. By extracting spectral characteristics through Fast Fourier Transform, abnormal changes that are difficult to detect in the time domain can be amplified and revealed in the frequency domain.

[0052] It is worth noting that the fused feature vector combines time-domain and frequency-domain features, which can simultaneously utilize the advantages of both types of features and overcome the limitation of a single feature domain being insensitive to certain abnormal patterns. The similarity calculation uses Euclidean distance as a metric, which is simple and efficient to calculate and suitable for real-time execution on industrial controllers. The template feature vectors in the preset state type library are obtained through offline training. Specifically, for four state types—normal state, slight mold wear, insufficient lubrication, and material hardness—signal samples of 100 stamping cycles (400 samples in total) are collected for each. The manufacturing methods for each abnormal state are as follows: slight mold wear is naturally formed after 2000 no-load stamping cycles on the surface of a new mold; insufficient lubrication is achieved by reducing the amount of lubricating oil supplied to the sliding surface of the mold during the forming process (50% of the normal amount); and material hardness is achieved by selecting sheet metal from the same batch with a hardness 10 Brinell hardness higher than the normal value. For each state type, 100 sets of fused feature vectors are collected. First, the mean and standard deviation of each dimension are calculated. Outliers exceeding the mean ± 3 times the standard deviation are removed. Then, the remaining samples are averaged dimension by dimension to obtain the template feature vector for each state type. During online recognition, the system calculates the Euclidean distance between the current fused feature vector and each template. The one with the smallest distance is the current state.

[0053] In step S5, when the state identification result is an abnormal state, the trimming cycle adjustment amount is determined according to the abnormal state and combined with the preset mechanical mapping relationship to adjust the running cycle of the trimming process. The equipment torque signal and mold vibration signal of the trimming process under the adjusted running cycle are obtained, and a trimming signal time series is constructed. The trimming signal time series is quality evaluated to obtain the trimming quality evaluation value.

[0054] The step of adjusting the cutting rhythm is determined based on the abnormal state and a preset mechanical mapping relationship, including: Based on the abnormal state, obtain the corresponding material hardness change and cooling time delay; Based on the change in material hardness and the cooling time delay, the predicted value of cutting resistance is calculated according to a preset mechanical mapping relationship. The tool wear compensation value is matched with the tool wear compensation value through a preset tool wear compensation curve based on the predicted cutting resistance value; The feed rate of the current trimming process is corrected using the tool wear compensation value to obtain the target feed rate; The target feed rate is converted into the trimming cycle adjustment amount.

[0055] It should be noted that when the status recognition result is normal, the system keeps the current running cycle of the trimming process unchanged and directly enters the step of acquiring the equipment torque signal and mold vibration signal of the trimming process under the current running cycle, and constructing the trimming signal time series for quality assessment.

[0056] In one implementation, after obtaining the abnormal state identification result of the forming process, the system first queries the abnormal state mapping table to obtain the material hardness change and cooling time delay corresponding to the abnormal state. For example, if the abnormal state is "material too hard", a positive value for the material hardness change indicates an increase in hardness, and a positive value for the cooling time delay indicates that the forced cooling interval between two adjacent cutting edges needs to be increased; if the abnormal state is "insufficient lubrication", a smaller value for the material hardness change and a negative value for the cooling time delay indicate a decrease in cooling efficiency. Then, the system substitutes the material hardness change and cooling time delay into a preset mechanical mapping formula to calculate the predicted cutting resistance value. Next, the system looks up the corresponding tool wear compensation value on the tool wear compensation curve based on the predicted cutting resistance value. This compensation value represents the feed rate correction amount required by the tool when bearing the predicted cutting resistance. Finally, the system adds the tool wear compensation value to the feed rate of the current cutting edge process to obtain the target feed rate, and then converts the target feed rate into the cutting edge cycle adjustment amount according to the conversion relationship between feed rate and cycle time.

[0057] It is worth noting that the abnormal state mapping table is constructed by conducting stamping tests on the same batch of standard sheet metal during the equipment debugging phase for four state types: normal state, slight mold wear, insufficient lubrication, and material hardness. For the slight mold wear state, a mold that has completed 200,000 stamping cycles is used for the test. For the insufficient lubrication state, the lubricating oil supply to the sliding surface of the mold in the forming process is reduced to 50% of the normal amount for the test. For the material hardness state, sheet metal with a hardness 15 Brinell hardness higher than the standard value is selected for the test. Each state is tested 50 times, and the change in material hardness and the required cooling time delay are measured for each test. The average value is taken as the stored value of that state in the mapping table. The mapping table is fixed in the control system in the form of a database. When the system is running online, it can directly look up the corresponding mapping parameters in the table based on the state identification results of the forming process.

[0058] It should be noted that the preset mechanical mapping relationship is the core mathematical model connecting the abnormal forming state and the trimming cycle adjustment. Its physical basis lies in the fact that the change in material hardness directly originates from the "material too hard" state identified in the forming process. This state indicates that the hardness of the current batch of sheet material is higher than the normal value. Since the same batch of sheet material is supplied to the blanking, forming, and trimming processes simultaneously, this hardness deviation still exists when the material is transferred to the trimming process, and therefore can be used as a feedforward adjustment basis for the trimming process. The mechanical mapping formula is as follows: ,in This is the predicted cutting resistance value (in Newtons). The reference cutting resistance (in Newtons) is measured through cutting tests when the material hardness is at the standard value (150HB) under normal working conditions. It is 380N. This refers to the change in material hardness, expressed in Brinell hardness, and typically ranges from -20 to +20 HB. This is the adjustment amount for the cooling interval deviating from the reference value. A positive value indicates that the cooling interval is shortened (material temperature increases), and a negative value indicates that the cooling interval is extended (material temperature decreases). The value range is generally from -3 to +5 seconds. This is the hardness influence coefficient, with units of 1 / HB. It is calibrated through cutting tests, and its typical value range is 0.003 to 0.0081 / HB. The coefficient representing the influence of cooling time is 1 / s. It is calibrated through cutting tests, and its typical value range is -0.02 to 0.05 1 / s. and The calibration was performed using a standard cutting tool and standard sheet metal on a testing machine. The cutting resistance was measured under different combinations of hardness grades (140HB, 150HB, 160HB) and cooling times (0s, 3s, 5s). The data were then substituted into the above formula, and the least squares method was used for fitting and solving. and The estimated value.

[0059] It should be further noted that the calibration process for the tool wear compensation curve is as follows: ISO 3685 standard is used as the basis for tool wear judgment, with a flank wear VB of 0.3 mm as the critical wear standard. Using a universal tool wear testing machine, different cutting resistance levels (200 N to 800 N, in 50 N increments) are applied. Continuous cutting is performed at each resistance level until the flank wear VB reaches 0.3 mm, and the cumulative cutting length or cutting time during this process is recorded. Based on the tool life data at each resistance level, a power function fitting is used to establish the cutting resistance... With safety feed rate correction value Relationship curve ,in and The coefficients represent the fit. There are 13 sets of experimental data points, and the goodness of fit R0 is... 2 The value should be no less than 0.95. After calibration, the curve is discretized into 13 interpolation points and fixed in the control system. In actual use, the compensation value corresponding to any resistance is obtained through linear interpolation.

[0060] It is worth noting that the conversion relationship between feed rate and cycle time is the adjustment amount of the cutting cycle. ,in The cutting stroke length is in millimeters and is determined by the product drawing. This represents the current feed rate (in millimeters per second). Target feed rate (in millimeters per second); This is the beat adjustment amount, in seconds. A positive value indicates extending the beat, and a negative value indicates shortening the beat. The system will calculate the amount... The control unit for the trimming process sends the adjustment to the operating cycle time in the next processing cycle. The cycle time of the trimming process refers to the time interval between two adjacent trimming operations, which is determined by the feed rate, stroke length, and auxiliary time. The faster the feed rate, the shorter the single trimming cycle and the shorter the cycle time; that is, a negative cycle time adjustment indicates a faster cycle time. Conversely, the slower the feed rate, the longer the cycle time; a positive cycle time adjustment indicates a slower cycle time. The system converts the difference between the target feed rate and the current feed rate into a cycle time adjustment value and sends it to the control unit for the trimming process to adjust the operating cycle time in the next processing cycle.

[0061] The quality assessment of the time series of the edge-cutting signal to obtain the edge-cutting quality assessment value includes: The cutting signal time series is divided into cutting stages, and features are extracted from the cutting signal time series corresponding to each cutting stage to obtain an initial feature vector; The initial feature vector is reduced in dimension and compressed to obtain the core feature vector; The deviation between the core feature vector and the preset quality standard is calculated and used as the edge cutting quality evaluation value.

[0062] In one implementation, during the edge trimming process at an adjusted cycle time, the system collects torque and vibration signals for that process and constructs an edge trimming signal time series using the same methods as steps S2 and S4. Then, based on the processing characteristics of the edge trimming process, the system divides the edge trimming signal time series into three sub-sequences: the entry stage, the steady-state cutting stage, and the exit stage. For each sub-sequence, the system extracts the mean and peak values ​​of the torque signal and the root mean square and kurtosis of the vibration signal within that stage, resulting in a total of 12-dimensional initial feature vectors (3 stages multiplied by 4 features). Next, the system uses principal component analysis (PCA) to reduce the dimensionality of the 12-dimensional initial feature vector. The specific number of principal components to retain is determined by the cumulative variance contribution rate. After sorting the feature values ​​from largest to smallest, the variance contribution rates of each principal component are accumulated sequentially. When the accumulated value first reaches 95%, the number of principal components included in the current accumulation is recorded, and the principal components with the previous number of principal components are retained as the dimensionality reduction result. In practical applications, the number of principal components is usually 3 to 5, with the specific value fluctuating slightly depending on the distribution of the sample data, to obtain the core feature vector. Finally, the system calculates the Euclidean distance between the core feature vector and the preset quality standard vector, and uses this distance as the edge trimming quality evaluation value; the smaller the distance, the better the edge trimming quality.

[0063] It should be noted that the trimming process is divided into cutting stages because the signal characteristics of different cutting stages differ significantly. In the entry stage, the tool just contacts the workpiece, resulting in a large impact load and a noticeable impact component in the signal; in the steady-state cutting stage, the tool is cutting stably within the workpiece, and the signal is relatively stable; in the exit stage, the tool is about to leave the workpiece, the load decreases rapidly, and the signal decays quickly. Processing these three stages separately allows for the extraction of features from each stage, avoiding aliasing that masks stage-specific information.

[0064] It should be further explained that the principal component analysis algorithm is a linear dimensionality reduction method. Its core idea is to project the original high-dimensional data into a low-dimensional space through a linear transformation, while preserving as much variance information as possible from the original data. In edge trimming quality assessment, the 12-dimensional initial feature vector may contain redundancy, and some features may be highly correlated. Through principal component analysis, a few uncorrelated principal components can be extracted. These principal components concentrate the main variance information in the original data, while removing noise dimensions, thus improving the stability and anti-interference ability of the quality assessment.

[0065] It is worth noting that the preset quality standard vector is obtained by averaging the core feature vectors extracted from 100 known qualified edge-cutting signal samples during the offline calibration phase. The criteria for determining qualified products are as follows: each edge-cutting workpiece is inspected using a coordinate measuring machine. Workpieces with hinge mating surface dimensional deviations within ±0.05mm, flange contour deviations within ±0.08mm, and edge burr heights not exceeding 0.1mm are marked as qualified; otherwise, they are discarded. Considering the potential correlation between the components of the core feature vector, the deviation is calculated using Mahalanobis distance instead of Euclidean distance. Mahalanobis distance decouples the correlation between features by introducing the inverse of the covariance matrix, thus more accurately measuring the true deviation between the current feature vector and the standard feature pattern. The smaller the calculated Mahalanobis distance, the closer the current edge-cutting state is to the standard qualified state, and the higher the quality assessment value. In practice, the reciprocal of the Mahalanobis distance is normalized to a value between 0 and 1 as the quality assessment value. The larger the value, the better the quality, which facilitates subsequent threshold comparison. Specifically, normalization requires pre-calculating the maximum and minimum values ​​of the reciprocal of the Mahalanobis distance for qualified product samples. During online calculation, the difference between the current reciprocal of the Mahalanobis distance and the minimum value is divided by the difference between the maximum and minimum values, so that the output value falls within the range of 0 to 1. The maximum and minimum values ​​are determined by the qualified product sample data during the offline calibration stage and are fixed in the system parameters.

[0066] In step S6, if the edge trimming quality assessment value is less than a preset assessment threshold, a shutdown control command is generated based on the edge trimming quality assessment value to control the blanking process, the forming process, and the edge trimming process to stop running.

[0067] In one implementation, the system compares the edge trimming quality assessment value obtained in step S5 with a preset assessment threshold. If the assessment value is greater than or equal to the threshold, the edge trimming quality is deemed acceptable, and production continues. If the assessment value is less than the threshold, the edge trimming quality is deemed unacceptable, and the system immediately generates a shutdown control command. This command includes a shutdown timestamp, a shutdown reason code, and a process range identifier. The process range identifier specifies three equipment controllers: the blanking process, the forming process, and the edge trimming process. The system broadcasts the shutdown control command to the three controllers via an industrial Ethernet bus. Upon receiving the command, each controller sequentially stops the servo motors and hydraulic system according to a preset shutdown sequence, completing the shutdown of the entire production line.

[0068] It should be noted that the preset evaluation threshold is the numerical limit for determining whether the edge cutting quality is qualified. The threshold is set using the ROC curve analysis method. In the offline stage, 500 samples with known labels (300 qualified products and 200 defective products) are collected. The labels of qualified and defective products are determined by a coordinate measuring machine. Specifically, products with hinge mating surface dimensional deviation within ±0.05mm, flange contour within ±0.08mm, and edge burr height not exceeding 0.1mm are marked as qualified; otherwise, they are marked as defective. For each sample, a quality assessment value is calculated, and the quality assessment value is divided into 100 candidate thresholds ranging from 0 to 1 in increments of 0.01. For each candidate threshold, the corresponding true positive rate (the proportion of defective products correctly identified as defective, i.e., recall rate) and false positive rate (the proportion of qualified products mistakenly identified as defective) are calculated. An ROC curve is plotted with the false positive rate on the horizontal axis and the true positive rate on the vertical axis. The Youden index corresponding to each candidate threshold is calculated, where the Youden index is obtained by calculating the difference between the true positive rate and the false positive rate. The candidate threshold with the largest Youden index is selected as the final preset assessment threshold, which can maximize the distinction between qualified and defective products.

[0069] It should be further explained that after generating the shutdown control command, the system does not simply cut off the power, but stops sequentially according to the inter-process linkage logic. The specific parameters of the preset shutdown sequence are as follows: after receiving the command, the blanking process immediately stops the feed roller drive motor, and the stopping action is completed within 200 milliseconds; after a delay of 500 milliseconds, the forming process stops the main slide drive motor, so that the blank that has entered the forming cavity completes the current stamping cycle and stops at the dead point position, which takes about 800 milliseconds; after another delay of 500 milliseconds, the trimming process stops the tool feed servo motor, so that the workpiece that has entered the trimming station completes the current trimming cycle and stops at the retraction position, which takes about 1000 milliseconds. The physical basis for the above timing design is that the feeding must be stopped first to prevent the continued supply of new blanks, and the trimming must be stopped last to ensure that the cut workpiece is removed from the tool area and to avoid the tool and workpiece jamming. The total shutdown time is about 3000 milliseconds, that is, the entire line is safely stopped within 3 seconds.

[0070] It is worth noting that step S6 constitutes the final protection link in the entire collaborative control closed loop. When the upstream deviation, even after compensation and adjustment, still cannot restore the trimming quality to a qualified level, it indicates that the current production status has exceeded the adjustable range of the process, and the machine must be stopped for maintenance. By simultaneously stopping the blanking, forming, and trimming processes, the continued production of defective products can be interrupted immediately, avoiding the continued consumption of materials, energy, and tool life in subsequent processing, and minimizing the risk of batch scrap.

[0071] For example, the edge trimming quality assessment value is 0.62, and the preset assessment threshold is 0.80. The system determines that 0.62 is less than 0.80 and immediately generates a stop control command. The stop reason code in the command is "edge trimming quality not up to standard," and the process range identifier includes the three processes of blanking, forming, and edge trimming. After receiving the command, the three controllers stop their respective servo motors in sequence, and the entire line completes a safe stop within 3 seconds.

[0072] In summary, this invention utilizes deep neural networks to learn the temporal features and predict dimensional deviations of torque and vibration signals in the blanking process, breaking the rigid mode of traditional stamping production lines where each process relies on fixed parameters for independent operation. This enables real-time online quantitative perception of blanking dimensional deviations. By establishing a deviation resistance mapping relationship between dimensional deviations and die-closing pressure, the blanking deviation is transformed into the target die-closing pressure of the forming process via a feedforward approach, effectively activating the forming process's proactive compensation capability for upstream deviations. Furthermore, based on the identification of abnormal states in the forming process, the adjustment amount of the trimming cycle is dynamically calculated using a preset mechanical mapping relationship and tool wear compensation curve, constructing a transmission link from forming abnormalities to trimming cycle adaptation. Finally, an innovative multi-stage feature extraction and quality assessment mechanism for trimming signals is introduced, generating a stop control command when the quality assessment value is lower than a preset threshold, achieving synchronous safety interlocking of the blanking, forming, and trimming processes. This invention completely abandons the static processing logic of information fragmentation between processes and the gradual accumulation of deviations in traditional stamping production lines. It successfully constructs a multi-process collaborative control closed loop that seamlessly integrates signal perception, parameter compensation, cycle time adaptation, and quality interlocking. Ultimately, it achieves real-time suppression of deviations and rapid blocking of quality anomalies throughout the entire eyeglass case stamping process under high-speed continuous stamping conditions, significantly improving the precision and pass rate of finished products.

[0073] The second embodiment of the present invention provides a collaborative control system for stamping and forming of eyeglass cases based on multi-process fusion, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method described above.

[0074] It should be noted that the multi-process fusion-based collaborative control system for eyeglass case stamping provided in this embodiment of the invention is used to execute all process steps of the multi-process fusion-based collaborative control method for eyeglass case stamping in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0075] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0076] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A collaborative control method for stamping and forming of eyeglass cases based on multi-process fusion, characterized in that, include: Acquire equipment torque signals and mold vibration signals during the blanking process; Based on the equipment torque signal and the mold vibration signal, a blanking signal time series is constructed, and time series prediction is performed on the blanking signal time series to obtain the dimensional deviation of the blanking process; Adjust the mold closing pressure of the molding process according to the dimensional deviation to obtain the target mold closing pressure; The equipment torque signal and mold vibration signal of the molding process under the target mold closing pressure are obtained, and a molding signal time series is constructed. The state of the molding signal time series is identified to obtain the state identification result. When the state identification result is an abnormal state, the trimming cycle adjustment amount is determined according to the abnormal state and combined with the preset mechanical mapping relationship to adjust the running cycle of the trimming process. The equipment torque signal and mold vibration signal of the trimming process under the adjusted running cycle are obtained, and the trimming signal time series is constructed. The trimming signal time series is quality evaluated to obtain the trimming quality evaluation value. If the edge trimming quality assessment value is less than the preset assessment threshold, a shutdown control command is generated based on the edge trimming quality assessment value to control the blanking process, the forming process and the edge trimming process to stop running.

2. The collaborative control method for eyeglass case stamping based on multi-process fusion as described in claim 1, characterized in that, The step of constructing a material dropping signal time series based on the equipment torque signal and the mold vibration signal includes: The equipment torque signal and the mold vibration signal are time-aligned to obtain a synchronous torque sequence and a synchronous vibration sequence. The synchronous torque sequence and the synchronous vibration sequence are combined into a multivariate synchronous time sequence, which is used as the material dropping signal time sequence.

3. The collaborative control method for eyeglass case stamping based on multi-process fusion as described in claim 1, characterized in that, The step of performing time-series prediction on the blanking signal time series to obtain the dimensional deviation of the blanking process includes: The time series of the material dropping signal is subjected to time-series feature extraction to obtain a time-dependent feature vector; The time-dependent feature vector is regressed to obtain the size deviation.

4. The collaborative control method for eyeglass case stamping based on multi-process fusion as described in claim 1, characterized in that, The step of adjusting the mold closing pressure of the molding process according to the dimensional deviation to obtain the target mold closing pressure includes: Based on the dimensional deviation and the preset deviation resistance mapping relationship, the deformation resistance compensation gradient is determined; The pressure compensation amount is calculated based on the dimensional deviation and the deformation resistance compensation gradient. The pressure compensation amount is superimposed on the pre-acquired reference mold closing pressure to obtain the target mold closing pressure.

5. The collaborative control method for eyeglass case stamping based on multi-process fusion as described in claim 1, characterized in that, The process of acquiring the equipment torque signal and mold vibration signal under the target mold closing pressure during the molding process, and constructing a molding signal time series, includes: Real-time equipment torque and real-time mold vibration of the molding process are collected under the target mold closing pressure; The real-time equipment torque and the real-time mold vibration are time-aligned to obtain a synchronous equipment torque sequence and a synchronous mold vibration sequence. The torque sequence of the synchronization device and the vibration sequence of the synchronization mold are spliced ​​together to obtain the molding signal time sequence.

6. The collaborative control method for eyeglass case stamping based on multi-process fusion according to claim 1, characterized in that, The process of performing state identification on the formed signal time series to obtain state identification results includes: Extract the statistical features in the time domain and the spectral features in the frequency domain of the shaped signal time series to obtain the time domain feature vector and the frequency domain feature vector; The time-domain feature vector and the frequency-domain feature vector are fused to obtain a fused feature vector; The similarity between the fused feature vector and each template feature vector in the preset state type is calculated, and the state type corresponding to the template feature vector with the highest similarity is selected as the state recognition result of the molding process.

7. The collaborative control method for eyeglass case stamping based on multi-process fusion as described in claim 1, characterized in that, The step of determining the trimming cycle adjustment amount based on the abnormal state and a preset mechanical mapping relationship includes: Based on the abnormal state, obtain the corresponding material hardness change and cooling time delay; Based on the change in material hardness and the cooling time delay, the predicted value of cutting resistance is calculated according to a preset mechanical mapping relationship. The tool wear compensation value is matched with the tool wear compensation value through a preset tool wear compensation curve based on the predicted cutting resistance value; The feed rate of the current trimming process is corrected using the tool wear compensation value to obtain the target feed rate; The target feed rate is converted into the trimming cycle adjustment amount.

8. The collaborative control method for eyeglass case stamping based on multi-process fusion according to claim 1, characterized in that, The process of evaluating the quality of the trimmed signal time series to obtain a trimmed quality evaluation value includes: The cutting signal time series is divided into cutting stages, and features are extracted from the cutting signal time series corresponding to each cutting stage to obtain an initial feature vector; The initial feature vector is reduced in dimension and compressed to obtain the core feature vector; The deviation between the core feature vector and the preset quality standard is calculated and used as the edge cutting quality evaluation value.

9. A collaborative control system for stamping and forming of eyeglass cases based on multi-process fusion, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the method described in any one of claims 1 to 8.

Images