Method and system for monitoring clamp pressure of an injection molding machine mold

Abstract

The application discloses a kind of injection molding machine mold holding pressure monitoring method and system, it is related to injection molding machine pressure monitoring technical field, method includes: constructing historical sample set, the holding pressure sequence of each historical sample is executed adaptive empirical mode decomposition, grouping obtains multiple historical sample subsets;Set candidate boundary window length set, intercept the boundary area of each modal component, judge and filter distorted boundary area;Adaptive correction is carried out to distorted boundary area, reconstructs holding pressure sequence;Calculate reconstruction score, select the best boundary window length of each historical sample subset;Build and train exception determination model;Data of current mold process is obtained, constructs current input sample, determines the holding pressure state of current mold process based on trained exception determination model.The application can accurately distinguish global real anomaly and boundary distortion, realize the high reliability monitoring of mold holding pressure, reduce false anomaly alarm rate.

Description

Technical Field

[0001] This invention relates to the field of injection molding machine pressure monitoring technology, and in particular to a method and system for monitoring the mold holding pressure of an injection molding machine. Background Technology

[0002] Injection molding is a core process for the large-scale production of plastic products. In injection molding, after the molten plastic solidifies in the mold cavity, the ejector mechanism of the injection molding machine removes the molded plastic product from the mold core or cavity. This demolding process is commonly referred to as ejection in the industry. During the ejection stage, the temperature of the plastic product should be cooled to below its heat distortion temperature to prevent relaxation due to residual stress or warping and deformation caused by demolding forces. As the final stage of the molding cycle, the stability of the ejection process directly determines the yield rate of plastic products and the mold life. The ejection holding pressure is the thrust maintained by the ejector mechanism after reaching its endpoint. It is used to counteract the clamping force of the plastic product, the gripping force of the robotic arm, and the risk of internal leakage in the hydraulic system, preventing unexpected retraction of the ejector pins that could lead to whitening, deformation, or mold damage. Therefore, accurate monitoring of the ejection holding pressure, timely detection of abnormalities, and prompt early warning are crucial for ensuring the quality of injection molding production and equipment safety.

[0003] Existing mold holding pressure monitoring technologies typically use simple Butterworth filters, moving average filters, or Kalman filters to remove high-frequency noise from the pressure signal. They generally rely on fixed thresholds or simple threshold ranges to determine anomalies, failing to consider the interference of injection molding machine operating condition fluctuations (such as mold temperature drift and hydraulic oil viscosity changes) on the pressure signal, resulting in weak anti-interference capabilities. Furthermore, they ignore the local impacts on the pressure signal caused by the mechanical movements of "mold ejection" (ejection start) and "mold ejection" (ejection reset) during the mold ejection process. Such physical disturbances easily lead to severe boundary effects during mold holding pressure monitoring, causing distortion at the beginning and end of the pressure signal and generating numerous false anomalies. Consequently, traditional methods cannot distinguish between "true equipment pressure anomalies (such as hydraulic leakage or ejector pin jamming)" and "local boundary distortions," often resulting in overcorrection masking fault characteristics or misjudging false disturbances as anomalies. This leads to low monitoring accuracy and fails to meet the quality control requirements of high-precision injection molding production. Summary of the Invention

[0004] To improve the monitoring accuracy of mold holding pressure and solve the problems of overcorrection masking fault characteristics or misjudging abnormalities due to distortion at the beginning and end of the pressure signal caused by data boundary distortion, this invention provides a method and system for monitoring mold holding pressure in injection molding machines, the technical solution of which is as follows: In a first aspect, the present invention provides a method for monitoring the holding pressure of an injection molding machine mold ejection process. The steps include: acquiring holding pressure sequences, mold temperature sequences, and binary labels of mold ejection results from multiple historical ejection processes to construct a historical sample set; performing adaptive empirical mode decomposition on the holding pressure sequence of each historical sample to obtain the decomposition level, modal components, and residual terms for each historical sample; grouping the historical sample set based on the decomposition level to obtain multiple historical sample subsets; setting a set of candidate boundary window lengths, extracting the boundary regions of each modal component under each candidate boundary window length, acquiring the energy data of the boundary regions, calculating the energy deviation coefficient of each modal component under each candidate boundary window length, and judging and filtering distorted boundary regions under each candidate boundary window length; adaptively correcting the distorted boundary regions, and based on the correction... The modal components and residual terms after correction are used to reconstruct the holding pressure sequence of all historical samples under each candidate boundary window length. Based on the error before and after correction of the holding pressure sequence, the reconstruction score of each candidate boundary window length relative to each historical sample is calculated to select the optimal boundary window length for each historical sample subset. Then, the corrected modal components corresponding to each historical sample under its optimal boundary window length are superimposed to obtain the pure holding pressure sequence of each historical sample. Based on the historical sample set and the pure holding pressure sequence of each historical sample, a training sample set is constructed, an anomaly detection model is constructed, and training is performed based on the training sample set. The holding pressure sequence and mold temperature sequence of the current mold ejection process are obtained, and the input sample corresponding to the current mold ejection process is constructed. Based on the trained anomaly detection model, the holding pressure state of the current mold ejection process is determined.

[0005] Preferably, for the same specification product under the same injection molding machine model, the original mold holding pressure data and original mold temperature data of multiple complete historical mold ejection processes are collected synchronously at a fixed sampling frequency. After normalization processing, the holding pressure sequence and mold temperature sequence of multiple historical mold ejection processes are obtained. The combination of the holding pressure sequence and mold temperature sequence corresponding to each historical mold ejection process is taken as a historical sample. The historical mold ejection process corresponding to qualified products without any mold ejection-related defects is defined as a normal pressure sample and the label is set to 0. The historical mold ejection process corresponding to unqualified products with mold ejection-related defects is defined as an abnormal pressure sample and the label is set to 1. The set of all historical samples and their labels is taken as the historical sample set.

[0006] Preferably, any historical sample in the historical sample set is selected as the target sample. An iterative method of gradually adding modal components is used to perform adaptive empirical mode decomposition on the holding pressure sequence of the target sample. An adaptive cutoff threshold is preset, and the absolute value of the Pearson correlation coefficient between the newly added modal component and the mold temperature sequence after each decomposition is calculated sequentially. The absolute value is used as the cross-correlation corresponding to the newly added modal component. When the cross-correlation corresponding to the newly added modal component is greater than or equal to the adaptive cutoff threshold, the decomposition stops. The decomposition result of the layer above the newly added modal component is used as the final decomposition result to obtain the number of decomposition layers of the target sample, the modal components of each decomposition layer, and the residual term. Historical samples with the same number of decomposition layers in the historical sample set are divided into a group to obtain multiple historical sample subsets with different decomposition layers.

[0007] Preferably, a set of candidate boundary window lengths is constructed by presetting upper and lower limits and a gradual step size for the boundary window length. Then, based on each candidate boundary window length, the two ends and center of all modal components are truncated sequentially to obtain the starting boundary window, ending boundary window, and center reference window for each modal component under each candidate boundary window length. Based on the amplitude of each data point in each modal component, the energy values ​​of the starting boundary window, ending boundary window, and center reference window for each modal component under each candidate boundary window length are calculated. For any modal component under any candidate boundary window length, the ratio between the energy values ​​of the starting boundary window and the center reference window is calculated. The ratio is mapped using a natural logarithm function, and the absolute value of the mapped value is used as the energy deviation coefficient of the starting boundary window. Similarly, based on the energy values ​​of the ending boundary window and the center reference window, the energy deviation coefficient of the ending boundary window is obtained. A distortion judgment threshold is set, and starting or ending boundary windows with energy deviation coefficients greater than the distortion judgment threshold are judged as distorted boundary windows.

[0008] Preferably, for any distortion boundary window, the energy deviation coefficient of the distortion boundary window is back-mapped using the natural exponential function, and the back-mapped value is used as the adaptive smoothing factor of the distortion boundary window; based on the adaptive smoothing factor of each distortion boundary window, the exponential moving average method is used to correct each distortion boundary window; based on the corrected distortion boundary window, the corrected modal component is obtained to replace the corresponding uncorrected modal component; for modal components without distortion boundary windows, the original modal component is used as the corrected modal component; the corrected modal components of each historical sample under each candidate boundary window length are superimposed and reconstructed with the corresponding residual term to obtain the reconstructed hold pressure sequence of each historical sample under each candidate boundary window length.

[0009] Preferably, any historical sample is selected as the target sample, and any candidate boundary window length is selected as the target window length; the root mean square error between the reconstructed hold pressure sequence of the target sample and its original hold pressure sequence under the target window length is calculated, and the root mean square error is back-mapped using the natural exponential function, and the back-mapped value is used as the reconstruction score of the target window length relative to the target sample; any subset of historical samples is selected as the target subset, and the historical samples within the target subset are traversed to obtain the reconstruction score of the target window length relative to each historical sample within the target subset; the reconstruction scores of the target window length relative to each historical sample within the target subset are accumulated, and the ratio of the accumulated value to the number of historical samples within the target subset is used as the comprehensive score of the target window length relative to the target subset.

[0010] Preferably, the candidate boundary window length set is traversed to obtain the comprehensive score of each candidate boundary window length relative to the target subset, and the candidate boundary window length with the largest comprehensive score is selected as the optimal boundary window length of the target subset; each historical sample subset is traversed sequentially to obtain the optimal boundary window length of each historical sample subset, and the optimal boundary window lengths corresponding to the historical samples within the same historical sample subset are the same; the residual terms of the target samples are removed, and all corrected modal components corresponding to the target samples under their optimal boundary window lengths are extracted and superimposed to obtain the purity preservation pressure sequence of the target samples; all historical samples within each historical sample subset are traversed sequentially to obtain the purity preservation pressure sequence of each historical sample.

[0011] Preferably, based on the pure holding pressure sequence of each historical sample, the original holding pressure sequence of each historical sample in the historical sample set is replaced to obtain a training sample set, which is then divided into a training set and a validation set. An anomaly detection model is constructed using a long short-term memory model, including an input layer, a hidden layer, a fully connected layer, and an output layer. The anomaly detection model is trained based on the training set to learn the model parameters, and the model parameters are validated and optimized using the validation set. The two-dimensional time series matrix composed of the pure holding pressure sequence of each historical sample and the mold temperature sequence is used as input, and the corresponding binary classification label is used as output. The cross-entropy loss function is selected as the model loss function, and the model parameters are iteratively updated through the backpropagation algorithm until the model loss function converges or reaches the preset number of iterations.

[0012] Preferably, the following steps are taken: First, the holding pressure sequence and mold temperature sequence of the current mold ejection process of the injection molding machine are obtained to construct the current sample. Adaptive empirical mode decomposition (IDED) is performed on the current holding pressure sequence to obtain the modal components of the current sample. Second, based on the DTW algorithm, the DTW distance between the current holding pressure sequence and the holding pressure sequence of each historical sample in the historical sample set is calculated, and the historical sample with the smallest DTW distance is selected as the matching sample. Third, based on the optimal boundary window length corresponding to the matching sample, the same boundary distortion check and adaptive correction operation are performed on the modal components of the current sample to obtain the pure holding pressure sequence of the current sample. Fourth, the two-dimensional time series matrix composed of the pure holding pressure sequence of the current sample and the mold temperature sequence is used as the input sample. Based on the trained anomaly detection model, the anomaly probability of the current mold ejection process is obtained. Fifth, if the anomaly probability is greater than a preset anomaly detection threshold, it is determined that the mold holding pressure is abnormal during the current mold ejection process, and an anomaly alarm is triggered.

[0013] Secondly, the present invention provides an injection molding machine mold holding pressure monitoring system for implementing the above-mentioned injection molding machine mold holding pressure monitoring method, comprising: a processor, a memory, a communication interface, a data acquisition device, and an alarm device. The processor stores computer program instructions for implementing the above-mentioned injection molding machine mold holding pressure monitoring method, and the communication interface is communicatively connected to the data acquisition device and the alarm device.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention separates working condition interference through adaptive empirical mode decomposition with mold temperature reference, and combines the single-component boundary energy deviation coefficient index to screen out boundary regions with boundary distortion problems. It adopts a joint region reconstruction error minimization calibration to determine the optimal boundary window, and combines it with exponential weighted correction linked to the degree of distortion. This eliminates boundary distortion caused by endpoint effects and mechanical motion disturbances, while preventing excessive smoothing from masking real fault characteristics. The LSTM model that integrates pressure-temperature two-dimensional time series features further improves the anomaly identification accuracy, so as to accurately distinguish between global real anomalies and pseudo-anomalies caused by local boundary distortions. Ultimately, it achieves high-reliability monitoring of mold holding pressure, significantly reduces the false anomaly alarm rate, and ensures the yield rate of plastic products and mold life. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the implementation of the injection molding machine mold holding pressure monitoring method according to an embodiment of the present invention.

[0016] Figure 2 This is a structural block diagram of the injection molding machine mold holding pressure monitoring system according to an embodiment of the present invention. Detailed Implementation

[0017] The technical features of the present invention will be further described in detail below with reference to the accompanying drawings so that those skilled in the art can understand them.

[0018] A method for monitoring the mold holding pressure of an injection molding machine, the implementation process is as follows: Figure 1 As shown, the specific implementation steps are as follows: Step S1: Obtain the holding pressure sequence, mold temperature sequence, and binary labels of the mold release results from multiple historical mold release processes to construct a historical sample set.

[0019] Specifically, the execution process of this step is as follows: For the same product specification under the same injection molding machine model, the original mold holding pressure data and original mold temperature data of multiple complete historical mold ejection processes are collected synchronously at a fixed sampling frequency. After normalization, the holding pressure sequence and mold temperature sequence of multiple historical mold ejection processes are obtained; the holding pressure sequence and mold temperature sequence of all historical mold ejection processes have the same length. Since mold temperature directly determines the clamping force between the product and the mold, thus affecting the mold holding pressure, mold temperature data and mold holding pressure data are correlated and need to be collected synchronously. For products of the same specification using standardized injection molding processes, a unified baseline duration is set for the mold ejection stage, with the ejection process starting from the mold entry signal and ending with the mold retraction signal. Mold holding pressure data is collected via a hydraulic oil circuit pressure sensor, and mold temperature data is collected via a mold core thermocouple. For example, the sampling frequency can be set to 100Hz, and the number of historical mold ejection processes collected is no less than 1000. The maximum and minimum values ​​of the original mold holding pressure data and original mold temperature data for each historical mold ejection process are extracted, and Min-Max normalization is performed to eliminate dimensional differences.

[0020] Each historical mold ejection process is defined as a combination of the holding pressure sequence and the mold temperature sequence. The historical mold ejection process corresponding to a qualified product without any mold ejection-related defects is defined as a normal pressure sample with a label of 0. The historical mold ejection process corresponding to a non-qualified product with mold ejection-related defects is defined as an abnormal pressure sample with a label of 1. The set of all historical samples and their labels is defined as the historical sample set. The number of historical samples under the two types of labels in the historical sample set is the same. By linking product quality inspection data, a mapping relationship between historical mold-holding processes and product quality is established. Since the mold-holding process is the final process of product production, abnormal mold-holding pressure will directly lead to defects such as scratches, deformation, and whitening of the product. Therefore, if the product quality inspection is qualified and there are no mold-holding related defects, it means that the mold-holding pressure in the corresponding historical mold-holding process is normal and the label is set to 0. Conversely, if the product fails the quality inspection due to mold-holding related defects, it means that the mold-holding pressure in the corresponding historical mold-holding process is abnormal and the label is set to 1.

[0021] Step S2: Perform adaptive empirical mode decomposition on the hold pressure sequence of each historical sample to obtain the decomposition level, modal components and residual terms of each historical sample. Based on the decomposition level, group the historical sample set to obtain multiple historical sample subsets.

[0022] Specifically, the execution process of this step is as follows: Select any historical sample from the historical sample set as the target sample, and use an iterative method of gradually adding modal components to perform adaptive empirical mode decomposition on the hold pressure sequence of the target sample. The stepwise addition of modal components is carried out through an iterative method of gradually stripping away high-frequency components. Modal components are generated layer by layer in order of center frequency from high to low. The adaptive empirical mode decomposition process is a conventional technique, and this step will not be described in detail here.

[0023] An adaptive cutoff threshold is preset, and the absolute value of the Pearson correlation coefficient between the newly added modal component and the mold temperature sequence after each layer of decomposition is calculated sequentially. The absolute value is used as the cross-correlation corresponding to the newly added modal component. The adaptive cutoff threshold can be adjusted according to the plastic material characteristics of different injection molded products. For example, the adaptive cutoff threshold can be set to 0.6. The lower the cooling and setting temperature of the injection molded product, the greater the influence of mold temperature fluctuations when the product is demolded, and the stronger the correlation between the mold holding pressure and the mold temperature. At this time, a larger adaptive cutoff threshold needs to be set to avoid premature termination of the decomposition process, resulting in insufficient separation of temperature interference information. Cooling and setting in the injection molding process refers to the highest temperature state at which the plastic is cooled to maintain the shape of the product and does not undergo permanent deformation when demolded.

[0024] Decomposition stops when the cross-correlation corresponding to the newly added modal component is greater than or equal to the adaptive cutoff threshold. The decomposition result of the previous layer of the newly added modal component is selected as the final decomposition result. The decomposition layer number of the target sample, the modal components of each layer, and the residual term are obtained. The length of the modal component is consistent with the length of the maintenance stress sequence. Among them, the working condition interference caused by mold temperature fluctuation is a slowly changing trend component, mainly concentrated in the low-frequency modal components. Therefore, when a modal component strongly correlated with temperature appears for the first time, that is, when the cross-correlation corresponding to the newly added modal component is greater than or equal to the adaptive cutoff threshold, it indicates that all effective modal components weakly correlated with temperature have been completely separated. The newly added modal component and its corresponding residual term are both temperature interference terms, and there is no need to continue decomposition. The decomposition result of the layer above the newly added modal component is selected as the final decomposition result after removing the temperature interference.

[0025] Historical samples with the same number of decomposition levels are grouped together to obtain multiple subsets of historical samples with different decomposition levels.

[0026] Step S3: Set a set of candidate boundary window lengths, extract the boundary regions of each modal component under each candidate boundary window length, obtain the energy data of the boundary regions, calculate the energy deviation coefficient of each modal component under each candidate boundary window length, and determine and filter the distorted boundary regions under each candidate boundary window length.

[0027] The reasoning logic for this step is as follows: During the mold ejection process in injection molding, the pressure signal in the mold holding phase is affected by special physical conditions at the beginning and end. Specifically, the ejection movement at the beginning and the ejection movement at the end of the ejection mechanism cause pressure disturbances. At the same time, the empirical mode decomposition algorithm uses cubic spline interpolation to fit the upper and lower envelopes during the decomposition process. Due to the lack of external data points at both ends of the signal, the interpolation function exhibits uncertain divergence at the boundaries, i.e., the endpoint effect. The boundary disturbance phenomenon and endpoint effect of the mold ejection mechanism can easily lead to boundary distortion of the decomposed modal components. The holding pressure data in the distorted region cannot accurately reflect the actual holding pressure in the mold holding phase. Therefore, it is necessary to perform boundary distortion checks and corrections on each modal component.

[0028] Specifically, the execution process of this step is as follows: The upper and lower limits of the boundary window length and the gradual step size are preset to construct a set of candidate boundary window lengths. Based on each candidate boundary window length, the two ends and the center of all modal components are truncated in turn to obtain the starting boundary window, ending boundary window and center reference window of each modal component under each candidate boundary window length. Among them, due to the influence of equipment aging and operating condition fluctuations, the degree of distortion at both ends of the retention pressure sequence of different historical samples is different, and a fixed boundary window length cannot be adapted to all samples; therefore, it is necessary to construct a candidate boundary window length set based on the upper and lower limits of the boundary window length to provide a data foundation for subsequent selection of the optimal boundary window length. The value of the gradient step size can be adjusted according to the accuracy requirements. If the accuracy requirement is high, the gradient step size can be reduced and the number of candidate boundary window lengths can be increased. Conversely, if the accuracy requirement is low, the gradient step size can be increased and the number of candidate boundary window lengths can be reduced. For example, the value of the gradient step size can be set to 5 data points, which can prevent the number of candidate boundary window lengths from being too large due to the step size being too small, and also avoid the step size being too large due to the deviation from the true distortion range. The lower limit of the boundary window length can be set based on manual experience. For example, the lower limit of the boundary window length can be set to 10 data points. The upper limit of the boundary window length can be set to one-third of the length of the pressure sequence or modal component. When one-third of the length of the modal component is a decimal, it is rounded down to ensure that the three regions of the starting boundary window, ending boundary window and center reference window of the same modal component do not overlap and exist independently. For example, if the sequence length is 120 sampling points, the lower limit of the boundary window length is 10 and the upper limit is 40, and the gradient step size is 5 data points, then the set of candidate boundary window lengths is: .

[0029] Furthermore, the starting boundary window is truncated from the first data point of the modal component backward, and the ending boundary window is truncated from the last data point of the modal component forward. The central reference window is truncated from the midpoint of each data point in the modal component. The reference window is truncated from the midpoint, so that the central reference window is farthest from the beginning and end boundaries, which can accurately reflect the true steady-state pressure characteristics during the mold holding stage and provide a reliable reference standard for subsequent distortion testing. For example, let the length of the modal component be N, and the length of the candidate boundary window be L; if N is even, then the index of the center point is... If N is odd, then the index of the center point is... Meanwhile, if the candidate boundary window length is odd, the central reference window expands outwards from the center point to both sides. There are 10 sampling points; if the candidate boundary window length is even, the central reference window expands from the center point towards the starting direction. Each sampling point expands towards the end direction. One sampling point.

[0030] Based on the amplitude of each data point in each modal component, the energy values ​​of the starting boundary window, ending boundary window, and center reference window of each modal component are calculated for each candidate boundary window length. The energy value of a data point is the square of its amplitude. The energy values ​​of each data point within the initial boundary window are summed to obtain the energy value of the initial boundary window. Similarly, the energy values ​​of the ending boundary window and the center reference window are obtained.

[0031] For any modal component with any candidate boundary window length, calculate the ratio between the energy values ​​of the starting boundary window and the center reference window, map the ratio using the natural logarithm function, and use the absolute value of the mapped value as the energy deviation coefficient of the starting boundary window; similarly, based on the energy values ​​of the ending boundary window and the center reference window, obtain the energy deviation coefficient of the ending boundary window. For example, any candidate boundary window length is selected as the target window length, and any modal component is selected as the target component. The energy deviation coefficient of the initial boundary window of the target component under the target window length is... The calculation formula is as follows: In the formula, This represents the energy value of the initial boundary window of the target component within the target window length. This represents the energy value of the reference window at the center of the target component within the target window length. This represents a very small non-zero constant, preventing the denominator from being 0. For example, The value can be set to , Represent the natural logarithm function; The energy deviation coefficient is used to convert the degree of bidirectional energy deviation at both ends of the modal component into a quantitative index, thereby identifying local boundary distortion. Its value range is [value range missing]. If there is no boundary distortion, the energy value of the starting or ending boundary window is close to the energy value of the central reference window, and the energy deviation coefficient approaches 0. If boundary distortion is caused by mechanical impact and end effect, the energy level in the starting or ending boundary window will show abnormal accumulation or abnormal decay, the energy deviation coefficient will increase, and the larger the energy deviation coefficient is, the more severe the boundary distortion.

[0032] Set a distortion judgment threshold, and determine the starting or ending boundary window with an energy deviation coefficient greater than the distortion judgment threshold as a distortion boundary window; The process for obtaining the distortion threshold corresponding to the target component within the target window length is as follows: Calculate the energy value of each data point within the center reference window of the target component. Based on the formula for calculating the energy deviation coefficient, calculate the energy deviation coefficient between all pairs of data points within the center reference window. Then calculate the mean and standard deviation of all energy deviation coefficients. According to the three sigma criterion in statistics, the sum of the mean and three times the standard deviation is used as the distortion judgment threshold corresponding to the target component under the target window length.

[0033] Step S4: Adaptively correct the distorted boundary region, and reconstruct the hold pressure sequence of all historical samples under each candidate boundary window length based on the corrected modal components and residual terms.

[0034] Specifically, the execution process of this step is as follows: For any distortion boundary window, the energy deviation coefficient of the distortion boundary window is reverse-mapped using the natural exponential function, and the value of the reverse mapping is used as the adaptive smoothing factor of the distortion boundary window. The theoretical range of values ​​for the adaptive smoothing factor is as follows: The larger the value of the energy deviation coefficient, the more severe the boundary distortion, and the closer the value of the adaptive smoothing factor is to 0; conversely, the smaller the value of the energy deviation coefficient, the less severe the boundary distortion, and the closer the value of the adaptive smoothing factor is to 1.

[0035] Based on the adaptive smoothing factor of each distortion boundary window, the exponential moving average method is used to correct each distortion boundary window; If the distortion boundary window is the starting boundary window, the data points within the starting boundary window are corrected sequentially from back to front, and the correction formula is as follows: In the formula, This represents the corrected magnitude of the i-th data point from the back to the front within the initial boundary window. This represents the corrected magnitude of the (i+1)th data point from the back within the initial boundary window. This represents the magnitude of the data point before correction, starting from the i-th data point in the initial boundary window. This represents the magnitude of the (i+1)th data point from the back in the initial boundary window before correction. This represents the magnitude of the data point adjacent to the first data point from the back within the starting boundary window. This represents the adaptive smoothing factor corresponding to the distortion boundary window; If the distortion boundary window is the end boundary window, the data points within the end boundary window are corrected sequentially from front to back, using the following correction formula: In the formula, This represents the corrected magnitude of the j-th data point from the front to the back within the end boundary window. This represents the corrected magnitude of the (j+1)th data point from the front within the end boundary window. This represents the magnitude of the j-th data point from the front to the back within the end boundary window before correction. This represents the magnitude of the (j+1)th data point from the front within the end boundary window before correction. This represents the magnitude of the data point adjacent to the first data point from the front within the end boundary window. This represents the adaptive smoothing factor corresponding to the distortion boundary window; Adaptive smoothing factor The smaller the value, the higher the degree of distortion in the corresponding distortion boundary window. The smaller the weight coefficient of the current data point's amplitude before correction in the correction formula, the more adaptively stronger the correction force of adjacent data points on the current data point is needed to ensure a smooth transition in the distortion region; conversely, the larger the value of the adaptive smoothing factor, the higher the degree of distortion. The larger the value, the lower the degree of distortion of the corresponding distortion boundary window. The larger the weight coefficient of the amplitude before correction of the current data point in the correction formula, the smaller the proportion of correction of the current data point by the adjacent data points, thereby adaptively reducing the over-correction of the original effective signal.

[0036] Based on the corrected distortion boundary window, the corrected modal components are obtained to replace the corresponding uncorrected modal components; for modal components without a distortion boundary window, the original modal components are used as the corrected modal components. The modal components of each historical sample after correction under each candidate boundary window length are superimposed and reconstructed with the corresponding residual terms to obtain the hold pressure sequence of each historical sample after reconstruction under each candidate boundary window length.

[0037] Step S5: Based on the error before and after the maintenance pressure sequence correction, calculate the reconstruction score of each candidate boundary window length relative to each historical sample, select the optimal boundary window length for each historical sample subset, and then superimpose the corrected modal components corresponding to each historical sample under its optimal boundary window length to obtain the pure maintenance pressure sequence of each historical sample.

[0038] Specifically, the execution process of this step is as follows: Select any historical sample as the target sample, and select any candidate boundary window length as the target window length; calculate the root mean square error between the reconstructed hold pressure sequence of the target sample and its original hold pressure sequence under the target window length, use the natural exponential function to perform inverse mapping on the root mean square error, and use the inverse mapping value as the reconstruction score of the target window length relative to the target sample; Boundary distortion correction only slightly alters the waveform and amplitude of modal components. If the target window length is too wide relative to the target sample, the extracted boundary region will contain too much normal signal from the intermediate region. Including the intermediate normal signal segments in the correction range will cause the normal signal to be over-smoothed, resulting in a larger fitting error between the reconstructed hold pressure sequence and the original hold pressure sequence, i.e., a larger root mean square error and a lower reconstruction score. If the target window length is too narrow relative to the target sample, boundary distortion cannot be accurately corrected, leading to more severe signal distortion in the reconstructed hold pressure sequence. The root mean square error between the two is also larger, resulting in a lower reconstruction score. Therefore, the larger the reconstruction score of the target window length relative to the target sample, the better the target window length fits the target sample.

[0039] Select any subset of historical samples as the target subset, traverse the historical samples within the target subset, and obtain the reconstruction score of the target window length relative to each historical sample within the target subset; The reconstruction scores of the target window length relative to each historical sample in the target subset are accumulated, and the ratio of the accumulated value to the number of historical samples in the target subset is used as the comprehensive score of the target window length relative to the target subset. In addition, the candidate boundary window length set is traversed to obtain the comprehensive score of each candidate boundary window length relative to the target subset, and the candidate boundary window length with the largest comprehensive score is selected as the optimal boundary window length of the target subset; Iterate through each historical sample subset in turn to obtain the optimal boundary window length for each historical sample subset. The optimal boundary window length is the same for each historical sample within the same historical sample subset. Different historical sample subsets represent different degrees of temperature interference. Therefore, by selecting the optimal boundary window length for each historical sample subset, historical samples under similar temperature conditions can be placed within the same feature extraction range, thereby eliminating the differential influence of temperature interference, removing temperature condition interference and overall baseline drift, and making the extracted holding pressure features purer and closer to reality.

[0040] Remove the residual terms of the target sample, extract all corrected modal components corresponding to the target sample under its optimal boundary window length, and superimpose them to obtain the pure preservation pressure sequence of the target sample. By iterating through all historical samples within each historical sample subset, a pure preserving pressure sequence for each historical sample is obtained.

[0041] Step S6: Based on the historical sample set and the cleanliness preservation pressure sequence of each historical sample, construct a training sample set, construct an anomaly detection model, and train it based on the training sample set.

[0042] Specifically, the execution process of this step is as follows: Based on the pure hold-up pressure sequence of each historical sample, the original hold-up pressure sequence of each historical sample in the historical sample set is replaced to obtain the training sample set. The training sample set is then divided into a training set and a validation set. For example, it can be done according to... The training set and validation set are divided according to a preset ratio; An anomaly detection model is constructed using a long short-term memory model, comprising an input layer, a hidden layer, a fully connected layer, and an output layer. The fully connected layer uses the ReLU function as the activation function, and the output layer uses the Sigmoid function as the activation function. The anomaly detection model is trained based on the training set to learn the model's parameters, and the model parameters are validated and optimized using the validation set. The two-dimensional time series matrix composed of the pure holding pressure sequence and the mold temperature sequence of each historical sample is used as input, and the corresponding binary classification label is used as output. The cross-entropy loss function is selected as the model loss function, and the model parameters are iteratively updated through the backpropagation algorithm until the model loss function converges or reaches the preset number of iterations. The construction and training process of the long short-term memory model is a conventional technique, and this step will not be described in detail here.

[0043] Step S7: Obtain the holding pressure sequence and mold temperature sequence of the current mold ejection process, construct the input sample corresponding to the current mold ejection process, and determine the holding pressure state of the current mold ejection process based on the trained anomaly detection model.

[0044] Specifically, the execution process of this step is as follows: Obtain the holding pressure sequence and mold temperature sequence of the current mold ejection process of the injection molding machine, construct the current sample, perform adaptive empirical mode decomposition on the current holding pressure sequence, and obtain the modal components of the current sample. The process of obtaining the modal components is consistent with the execution process in steps S1 and S2. Based on the DTW algorithm, the DTW distance between the current hold pressure sequence and the hold pressure sequence of each historical sample in the historical sample set is calculated, and the historical sample with the smallest DTW distance is selected as the matching sample. Based on the optimal boundary window length corresponding to the matching sample, the same boundary distortion test and adaptive correction operation are performed on the modal components of the current sample through the execution process in steps S3, S4 and S5, so as to obtain the purity preservation pressure sequence of the current sample. Using the two-dimensional time series matrix composed of the current sample's pure holding pressure sequence and the mold temperature sequence as input samples, and based on the trained anomaly detection model, the anomaly probability of the current mold ejection process is obtained. If the probability of an anomaly is greater than the preset anomaly determination threshold, it is determined that the mold holding pressure is abnormal during the current mold support process, and an anomaly alarm is triggered. The anomaly determination threshold is set based on human experience and detection accuracy requirements. The higher the detection accuracy requirement, the smaller the set value of the anomaly determination threshold. For example, the anomaly determination threshold can be set to 0.5.

[0045] The flowchart provided in this embodiment is not intended to indicate that the operations of the method will be performed in any particular order, or that all operations of the method are included in every case. Furthermore, the method may include additional operations. Within the scope of the technical concept provided by the method in this embodiment, additional variations can be made to the above method.

[0046] It should be understood that in some embodiments, the components may be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods may be implemented using software or firmware stored in memory and executed by a suitable instruction execution system.

[0047] This invention also discloses an injection molding machine mold holding pressure monitoring system, used to implement the above-mentioned injection molding machine mold holding pressure monitoring method, the system structure is as follows: Figure 2 As shown, it includes: a processor, a memory, a communication interface, a data acquisition device, and an alarm device. The processor stores computer program instructions for implementing the above-mentioned injection molding machine mold holding pressure monitoring method. The communication interface is communicatively connected to the data acquisition device and the alarm device. The data acquisition device includes a hydraulic oil circuit pressure sensor for collecting mold holding pressure data and a mold core thermocouple sensor for collecting mold temperature; the alarm device includes an audible and visual alarm deployed on the injection molding machine and a communication module for sending alarm notification information to the monitoring platform.

[0048] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, status setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages ​​and procedural programming languages.

[0049] Computer program instructions can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server.

[0050] In the latter case, the remote computer can connect to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can connect to an external computer, such as via the Internet using an Internet service provider.

[0051] In some embodiments, in order to perform aspects of the present invention, electronic circuits including, for example, programmable logic circuits, field-programmable gate arrays (FPGAs) or programmable logic arrays (PLAs) can execute computer-readable program instructions to personalize the electronic circuits by utilizing state information of computer-readable program instructions.

[0052] The embodiments included in this invention are descriptions of preferred embodiments of the invention and are not limited to the precise structures already described above and shown in the accompanying drawings. Various modifications and changes can be made without departing from the scope of protection. All variations and improvements made by those skilled in the art to the technical solutions of this invention without departing from the design concept of this invention should fall within the scope of protection of this invention.

Claims

1. A method of monitoring clamp pressure of a mold on an injection molding machine, characterized by, include: Obtain the holding pressure sequence, mold temperature sequence, and binary labels of the mold release results from multiple historical mold release processes to construct a historical sample set; Adaptive empirical mode decomposition is performed on the hold-up stress sequence of each historical sample to obtain the decomposition level, modal components and residual terms of each historical sample. The historical sample set is grouped based on the decomposition level to obtain multiple historical sample subsets. Set a set of candidate boundary window lengths, extract the boundary regions of each modal component under each candidate boundary window length, obtain the energy data of the boundary regions, calculate the energy deviation coefficient of each modal component under each candidate boundary window length, and identify and filter the distorted boundary regions under each candidate boundary window length, including: The upper and lower limits of the boundary window length and the gradual step size are preset to construct a set of candidate boundary window lengths. Based on each candidate boundary window length, the two ends and the center of all modal components are truncated in turn to obtain the starting boundary window, ending boundary window and center reference window of each modal component under each candidate boundary window length. Based on the amplitude of each data point in each modal component, the energy values ​​of the starting boundary window, ending boundary window, and center reference window of each modal component are calculated for each candidate boundary window length. For any modal component with any candidate boundary window length, calculate the ratio between the energy values ​​of the starting boundary window and the center reference window, map the ratio using the natural logarithm function, and use the absolute value of the mapped value as the energy deviation coefficient of the starting boundary window; similarly, based on the energy values ​​of the ending boundary window and the center reference window, obtain the energy deviation coefficient of the ending boundary window. Set a distortion judgment threshold, and determine the starting or ending boundary window with an energy deviation coefficient greater than the distortion judgment threshold as a distortion boundary window; Adaptive correction is performed on the distorted boundary regions, and based on the corrected modal components and residual terms, the hold-up pressure sequence of all historical samples under each candidate boundary window length is reconstructed. Based on the error before and after the maintenance pressure sequence correction, the reconstruction score of each candidate boundary window length relative to each historical sample is calculated to select the optimal boundary window length for each historical sample subset. Then, the corrected modal components corresponding to each historical sample under its optimal boundary window length are superimposed to obtain the pure maintenance pressure sequence of each historical sample. Based on the historical sample set and the cleanliness preservation pressure sequence of each historical sample, a training sample set is constructed, an anomaly detection model is built, and the model is trained based on the training sample set. Obtain the holding pressure sequence and mold temperature sequence of the current mold ejection process, construct the input sample corresponding to the current mold ejection process, and determine the holding pressure state of the current mold ejection process based on the trained anomaly detection model.

2. The method for monitoring the mold holding pressure of an injection molding machine according to claim 1, characterized in that, The construction of the historical sample set includes: For the same product specification under the same model of injection molding machine, the original mold holding pressure data and original mold temperature data of multiple complete historical mold ejection processes are collected synchronously at a fixed sampling frequency. After normalization, the holding pressure sequence and mold temperature sequence of multiple historical mold ejection processes are obtained. Each historical mold ejection process is defined as a combination of the holding pressure sequence and the mold temperature sequence. The historical mold ejection process corresponding to a qualified product without any mold ejection-related defects is defined as a normal pressure sample with a label of 0. The historical mold ejection process corresponding to a non-qualified product with mold ejection-related defects is defined as an abnormal pressure sample with a label of 1. The set of all historical samples and their labels is defined as the historical sample set.

3. The method for monitoring the mold holding pressure of an injection molding machine according to claim 1, characterized in that, The grouping of historical sample sets based on the number of decomposition layers includes: Select any historical sample from the historical sample set as the target sample, and use an iterative method of gradually adding modal components to perform adaptive empirical mode decomposition on the hold pressure sequence of the target sample. An adaptive cutoff threshold is preset, and the absolute value of the Pearson correlation coefficient between the newly added modal component and the mold temperature sequence after each layer of decomposition is calculated sequentially. The absolute value is used as the cross-correlation corresponding to the newly added modal component. When the cross-correlation corresponding to the newly added modal component is greater than or equal to the adaptive cutoff threshold, the decomposition stops and the decomposition result of the previous layer of the newly added modal component is selected as the final decomposition result, and the decomposition layer number of the target sample, the modal components of each decomposition layer, and the residual term are obtained. Historical samples with the same number of decomposition levels are grouped together to obtain multiple subsets of historical samples with different decomposition levels.

4. The method for monitoring the mold holding pressure of an injection molding machine according to claim 1, characterized in that, The reconstruction of the hold-up pressure sequence of all historical samples for each candidate boundary window length includes: For any distortion boundary window, the energy deviation coefficient of the distortion boundary window is reverse-mapped using the natural exponential function, and the value of the reverse mapping is used as the adaptive smoothing factor of the distortion boundary window. Based on the adaptive smoothing factor of each distortion boundary window, the exponential moving average method is used to correct each distortion boundary window; Based on the corrected distortion boundary window, the corrected modal components are obtained to replace the corresponding uncorrected modal components; for modal components without a distortion boundary window, the original modal components are used as the corrected modal components. The modal components of each historical sample after correction under each candidate boundary window length are superimposed and reconstructed with the corresponding residual terms to obtain the hold pressure sequence of each historical sample after reconstruction under each candidate boundary window length.

5. The method for monitoring the mold holding pressure of an injection molding machine according to claim 1, characterized in that, The calculation of the reconstruction score for each candidate boundary window length relative to each historical sample includes: Select any historical sample as the target sample, and select any candidate boundary window length as the target window length; calculate the root mean square error between the reconstructed hold pressure sequence of the target sample and its original hold pressure sequence under the target window length, use the natural exponential function to perform inverse mapping on the root mean square error, and use the inverse mapping value as the reconstruction score of the target window length relative to the target sample; Select any subset of historical samples as the target subset, traverse the historical samples within the target subset, and obtain the reconstruction score of the target window length relative to each historical sample within the target subset; The reconstruction scores of the target window length relative to each historical sample in the target subset are accumulated, and the ratio of the accumulated value to the number of historical samples in the target subset is used as the comprehensive score of the target window length relative to the target subset.

6. The method for monitoring the mold holding pressure of an injection molding machine according to claim 5, characterized in that, Obtain the pure preservation pressure sequence of each historical sample, including: Traverse the set of candidate boundary window lengths to obtain the comprehensive score of each candidate boundary window length relative to the target subset, and select the candidate boundary window length with the largest comprehensive score as the optimal boundary window length for the target subset; Iterate through each historical sample subset in turn to obtain the optimal boundary window length for each historical sample subset. The optimal boundary window length is the same for each historical sample within the same historical sample subset. Remove the residual terms of the target sample, extract all corrected modal components corresponding to the target sample under its optimal boundary window length, and superimpose them to obtain the pure preservation pressure sequence of the target sample. By iterating through all historical samples within each historical sample subset, a pure preserving pressure sequence for each historical sample is obtained.

7. The method for monitoring the mold holding pressure of an injection molding machine according to any one of claims 1 to 6, characterized in that, The construction of the anomaly detection model and its training based on the training sample set includes: Based on the pure hold pressure sequence of each historical sample, the original hold pressure sequence of each historical sample in the historical sample set is replaced to obtain the training sample set, and the training sample set is divided into training set and validation set. An anomaly detection model is constructed using a long short-term memory model, which includes an input layer, a hidden layer, a fully connected layer, and an output layer. The anomaly detection model is trained on a training set to learn the model parameters, and the model parameters are validated and optimized using a validation set. The two-dimensional time series matrix composed of the pure holding pressure sequence and the mold temperature sequence of each historical sample is used as input, and the corresponding binary classification label is used as output. The cross-entropy loss function is selected as the model loss function, and the model parameters are iteratively updated through the backpropagation algorithm until the model loss function converges or the preset number of iterations is reached.

8. The method for monitoring the mold holding pressure of an injection molding machine according to any one of claims 1 to 6, characterized in that, The determination of the holding pressure state during the current mold-forming process includes: Obtain the holding pressure sequence and mold temperature sequence of the current mold ejection process of the injection molding machine, construct the current sample, and perform adaptive empirical mode decomposition on the current holding pressure sequence to obtain the modal components of the current sample; Based on the DTW algorithm, the DTW distance between the current hold pressure sequence and the hold pressure sequence of each historical sample in the historical sample set is calculated, and the historical sample with the smallest DTW distance is selected as the matching sample. Based on the optimal boundary window length corresponding to the matching sample, the same boundary distortion test and adaptive correction operation are performed on the modal components of the current sample to obtain the purity preservation pressure sequence of the current sample. Using the two-dimensional time series matrix composed of the current sample's pure holding pressure sequence and the mold temperature sequence as input samples, and based on the trained anomaly detection model, the anomaly probability of the current mold ejection process is obtained. If the abnormal probability is greater than the preset abnormal judgment threshold, it is determined that the mold holding pressure is abnormal during the current mold support process, and an abnormal alarm is triggered.

9. A mold holding pressure monitoring system for injection molding machines, characterized in that, include: The device includes a processor, a memory, a communication interface, a data acquisition device, and an alarm device. The processor stores computer program instructions for implementing the injection molding machine mold holding pressure monitoring method according to any one of claims 1 to 8. The communication interface is communicatively connected to the data acquisition device and the alarm device.

Images