Injection molding process parameter drift early warning method based on support vector machine

Abstract

The present application relates to the technical field of process monitoring, and particularly relates to a support vector machine-based injection molding process parameter drift early warning method. The method comprises the following steps: obtaining reference parameters of each process parameter and measured values of each process parameter in the current injection cycle, calculating standard deviation according to the reference parameters, obtaining positive cumulative amount and negative cumulative amount through double-sided recursion and determining a dominant direction; further calculating the mean vector norm of the standard deviation in a preset window to obtain a barycenter distance, and calculating a synchronous growth rate using the reference coupling degree and the dominant direction; splicing the positive cumulative amount, the negative cumulative amount, the barycenter distance and the synchronous growth rate into enhanced features, inputting the enhanced features into a preset support vector machine classification model for judgment, and issuing an early warning instruction in response to a drift state of the early warning result. The present application realizes automatic and real-time early warning of gradual drift of injection molding process parameters.

Description

Technical Field

[0001] This invention relates to the field of process monitoring technology, and in particular to a method for early warning of drift in injection molding process parameters based on support vector machines. Background Technology

[0002] Injection molding is a continuous batch production process in which thermoplastic plastic is heated and melted, injected into a mold cavity under high pressure, held under pressure and cooled, and then ejected to form the molded part. It is widely used in automotive parts, consumer electronics casings, medical devices, and other fields. Injection molding machine controllers operate with fixed process parameter formulas. Core parameters include melt temperature, injection pressure, injection speed, holding pressure, holding time, and mold temperature, and the controllers automatically record the measured values ​​of each parameter after each injection. During continuous batch production, the slow degradation of equipment conditions, such as screw wear, hydraulic oil temperature rise, and changes in mold thermal balance, can drive these parameters to continuously and monotonically deviate from their set values ​​over tens to hundreds of injection cycles, forming a gradual drift. This gradual drift has a cumulative impact on the dimensional accuracy and mechanical properties of the molded parts. Therefore, timely detection of this gradual drift in process parameters is of significant engineering importance.

[0003] Currently, the industry commonly uses a threshold alarm method based on fixed tolerance zones to monitor injection molding process parameters. This method sets upper and lower control limits for each process parameter. After each injection cycle, the injection molding machine controller compares the measured values ​​of each parameter with their corresponding control limits. An alarm is triggered if a parameter's measured value exceeds the control limit. The control limits are typically set offline based on process specifications and historical production experience. The alarm logic is simple and clear, easy to implement in existing injection molding machine controllers, and has good responsiveness to sudden parameter fluctuations, making it widely used in the injection molding industry.

[0004] However, the aforementioned threshold alarm methods have significant limitations in detecting gradual drift. Gradual drift involves extremely small offsets per injection, with measured parameter values ​​remaining within control limits for extended periods, preventing threshold alarms from being triggered. By the time the drift accumulates beyond the control limits, a large batch of defective products has often already been produced. Furthermore, in the injection molding process, when parameters are driven by the same physical cause, multi-parameter coordinated drift occurs. For example, a rise in hydraulic oil temperature causes both injection speed and holding pressure to decrease in the same direction. While the offset per injection for each individual parameter is even smaller, the combined direction of multiple parameters generates a strong drift signal. Existing threshold alarm methods monitor each parameter independently, failing to utilize the physical coupling between parameters, leading to long-term missed detections of coordinated drift and hindering proactive drift warnings. Summary of the Invention

[0005] To address the technical problem of existing threshold alarm methods failing to detect gradual drift and multi-parameter collaborative drift in injection molding processes, this invention provides an injection molding process parameter drift early warning method based on support vector machine, including: acquiring the baseline parameters of each process parameter and the measured values ​​of each process parameter in the current injection cycle; The measured values ​​of each process parameter are calculated based on the reference parameters to obtain the standard deviation of each process parameter; a two-sided recursion is performed based on the standard deviation to obtain the positive and negative cumulative values ​​of each process parameter; the dominant direction of each process parameter is determined based on the positive and negative cumulative values. Calculate the mean vector norm of the standard deviations of all process parameters within a preset window to obtain the centroid distance; obtain the benchmark coupling degree between each process parameter, which is a correlation coefficient characterizing the degree of linear correlation between the historical values ​​of two process parameters; calculate the synchronous growth rate based on the benchmark coupling degree and the dominant direction of each process parameter. The enhanced features are obtained by combining the positive cumulative amount, negative cumulative amount, center of gravity distance, and synchronous growth rate of various process parameters. The enhanced features are input into a preset support vector machine classification model for judgment to obtain a warning result; in response to the warning result being in a drift state, a warning command is issued.

[0006] This invention combines the positive and negative cumulative values ​​obtained from bilateral recursion, the centroid distance within a preset window, and the synchronous growth rate based on the reference coupling degree into an enhanced feature, and inputs it into a pre-trained classification model for judgment. It integrates the multi-parameter independent cumulative offset state with the overall offset amplitude and direction coordination into a unified judgment input, realizing automatic and real-time early warning of gradual drift of injection molding process parameters, and making up for the shortcomings of traditional threshold alarms in responding to gradual drift in a timely manner.

[0007] Preferably, the benchmark parameters include a benchmark mean and a benchmark standard deviation; the calculation of the measured values ​​of each process parameter to obtain the standard deviation of each process parameter includes: calculating the parameter difference between the measured value of any process parameter in the current injection and the corresponding benchmark mean; and taking the ratio of the parameter difference to the corresponding benchmark standard deviation as the standard deviation of the process parameter.

[0008] Preferably, the step of performing two-sided recursion based on the standard deviation to obtain the positive and negative cumulative values ​​of each process parameter includes: Obtain the historical positive and negative cumulative values ​​of any process parameter in the previous injection; The historical positive cumulative amount is added to the standard deviation and subtracted from the preset drift reference value to obtain the first recursive value. The first recursive value is compared with zero, and the maximum value between the first recursive value and zero is taken as the positive cumulative amount of the process parameter. The second recursive value is obtained by subtracting the standard deviation and drift reference value from the historical negative cumulative amount. The second recursive value is compared with zero, and the maximum value between the second recursive value and zero is taken as the negative cumulative amount of the process parameter.

[0009] The double-sided recursive method accumulates the extremely small gradual shift signal of each bet into a positive or negative cumulative amount. Even if the shift of each bet is far below the tolerance band, the cumulative amount will still form a monotonically increasing trend. This transforms the slow drift signal that traditional threshold alarms cannot capture into an identifiable cumulative statistic, improving the early detectability of gradual drift.

[0010] Preferably, determining the dominant direction of each process parameter based on the positive and negative cumulative amounts includes: Compare the magnitudes of the positive and negative cumulative values ​​of any process parameter; in response to a positive cumulative value being greater than or equal to a negative cumulative value, determine the dominant direction of the process parameter as a positive indicator value, wherein the positive indicator value is taken as... In response to a positive cumulative amount being less than a negative cumulative amount, the dominant direction of the process parameter is determined as a negative indication value, wherein the negative indication value is taken as... .

[0011] Preferably, the step of calculating the mean vector norm of the standard deviation of all process parameters within the preset window to obtain the centroid distance includes: taking the average of the standard deviations of any process parameter within the preset window under all injections corresponding to the preset window to obtain the average deviation of each process parameter; calculating the sum of squares of the average deviations of all process parameters; and using the square root of the sum of squares as the centroid distance.

[0012] The centroid distance compresses the overall magnitude of the multi-parameter deviation into a single scalar. When a single parameter shifts occasionally and significantly, this scalar changes only slightly, while when multiple parameters shift continuously and collaboratively, this scalar increases significantly. Therefore, the classification model can distinguish between occasional disturbances and true drift based on the centroid distance, thus reducing the false alarm rate.

[0013] Preferably, the calculation of the synchronous growth rate based on the baseline coupling degree and the dominant direction of each process parameter includes: Extract the average value of the dominant direction of each process parameter within the most recent preset direction window to obtain the stability factor, and combine all process parameters in pairs to obtain multiple parameter pairs; For any pair of parameters, if the sign of the product of the stability factors of the two process parameters is consistent with the sign of the corresponding reference coupling degree, the product of the absolute values ​​of the two stability factors is taken as the consistency score; otherwise, the consistency score is set to zero. Using the absolute value of the baseline coupling degree of each parameter pair as the weight, the consistency scores of each parameter pair are weighted, summed, and then divided by the sum of the absolute values ​​of all baseline coupling degrees. The resulting quotient is used as the synchronization growth rate. When the sum is less than the preset lower limit threshold, the equal weighted average of the consistency scores of each parameter pair is used instead.

[0014] The stability factor smooths out occasional reversals of the dominant direction of a single bet. Based on this, the parameter pairs are weighted and summed with the absolute value of the baseline coupling degree as the weight. The resulting synchronous growth rate tends to 1 during real cooperative drift and remains low during normal random fluctuations, providing a highly discriminative feature input for the classification model to distinguish between cooperative drift and occasional deviations.

[0015] Preferably, before obtaining the baseline parameters for each process parameter, the early warning method further includes: Collect historical data of each process parameter under normal injection number corresponding to the preset baseline injection number during the machine adjustment and stabilization phase; Calculate the Pearson correlation coefficient between the historical values ​​of any two process parameters; In response to the absolute value of the Pearson correlation coefficient being greater than or equal to a preset correlation threshold, the Pearson correlation coefficient is directly used as the benchmark coupling degree between the two corresponding process parameters. In response to the absolute value of the Pearson correlation coefficient being less than the correlation threshold, the baseline coupling degree between the two corresponding process parameters is forcibly set to zero.

[0016] After the physical coupling structure of the injection molding process is solidified in one go during the benchmark establishment stage, the noise interference introduced by the extremely weak correlation is eliminated, and there is no need to repeatedly calculate the coupling relationship in the online stage. Therefore, the direction consistency judgment of the synchronous growth rate has a stable and reliable calculation benchmark.

[0017] Preferably, before inputting the enhanced features into the preset support vector machine classification model, the support vector machine classification model is obtained through the following training steps: dividing the process parameters of the historical normal production stage into a normal sample set; superimposing linearly varied data of a single parameter on the normal sample set to construct a biased sample set; screening target parameter groups whose absolute value of the baseline coupling degree is greater than the strong coupling threshold, and controlling each parameter in the target parameter group to synchronously add linear offset data according to the sign direction of the baseline coupling degree to construct a collaborative sample set; extracting the enhanced features from the normal sample set, the biased sample set, and the collaborative sample set as training sets, and optimizing the initial model based on the training set to obtain the support vector machine classification model.

[0018] Preferably, the step of issuing a warning instruction in response to a drifting state in the warning result includes: storing the warning result of the current injection output into a preset continuous decision queue; reading multiple historical warning results in the continuous decision queue; and determining that a real drift event has occurred in the production process in response to a set number of consecutively confirmed warning results in the continuous decision queue being in a drifting state, and issuing a warning instruction.

[0019] The continuous judgment queue only triggers the warning command when the warning results of the consecutive preset number of confirmations are all in a drift state. Therefore, the false judgment caused by occasional fluctuations in a single bet is effectively filtered out. While ensuring the sensitivity of drift detection, the false alarm rate is reduced, making the warning command more credible to the operator.

[0020] Preferably, after issuing the warning command, the warning method further includes: Receive system reset commands sent from external sources; In response to the system reset command, all positive and negative cumulative values ​​of all process parameters are reset to zero. Clear the historical cache data of the standard deviation of all process parameters in the preset window so that the centroid distance can be recalculated in the next injection.

[0021] The system reset command resets all positive and negative cumulative values ​​to zero and clears the historical cache data of standard deviation within the preset window. The residual influence of historical drift data before parameter adjustment in subsequent injections is thus eliminated, and the monitoring benchmark is restored to normal production status in a timely manner, forming a closed-loop early warning management process.

[0022] The technical solution of the present invention has the following beneficial technical effects: This invention establishes a baseline coupling degree matrix reflecting the physical coupling structure of the process offline during the machine stabilization phase, and trains the support vector machine model using a specially constructed partial sample set and a collaborative sample set. This allows the decision boundary of the classification model to simultaneously cover two typical scenarios: single-parameter independent drift and multi-parameter collaborative gradual drift. This achieves low-cost model deployment without stopping the machine to collect real drift data. In the online phase, the injection molding machine controller directly provides the measured values ​​of each process parameter after each injection. Through two-sided recursion, centroid distance calculation, and synchronous growth rate extraction, the weak gradual drift signal is transformed into enhanced features that take into account the coordination of drift amplitude and direction and input into the classification model. Early warning decisions are made within milliseconds, making the drift detection timeliness significantly better than manual inspection and traditional threshold alarms. This significantly advances the early warning time point of gradual drift in the injection molding process from the point of accumulation beyond the tolerance zone to the early stage of drift, effectively reducing the risk of batch non-conforming products. Attached Figure Description

[0023] Figure 1This is a flowchart of an injection molding process parameter drift early warning method based on support vector machine according to the present invention.

[0024] Figure 2 This is a comparison curve of the drift characteristic amplitude with the number of injections in a collaborative gradual drift scenario between the method of this invention and the traditional threshold alarm method. Figure 3 This is a comparison curve of the false alarm rate between the method of this invention and the traditional threshold alarm method under different occasional fluctuation amplitude conditions. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0026] Reference Figure 1 A method for early warning of injection molding process parameter drift based on support vector machine includes steps S101 to S105, which are described in detail below: S101, obtain the baseline parameters of each process parameter and the measured values ​​of each process parameter in the current injection.

[0027] In one embodiment, taking a PP injection molding production line for an automotive bumper bracket as an example, the production line operates with a fixed process parameter formula. The injection molding machine controller automatically records the measured values ​​of six process parameters per injection cycle: melt temperature, injection pressure, injection speed, holding pressure, holding time, and mold temperature. The total number of process parameters is... The value is 6. During normal production, various process parameters fluctuate randomly within their tolerance zones. With the slow changes in equipment conditions such as screw wear and hydraulic oil temperature rise, multiple parameters may undergo coordinated gradual drift along the physical coupling law of the process. The offset per injection is extremely small, making it difficult for traditional threshold alarms to identify in a timely manner. Before formal production commences, an offline baseline parameter acquisition and classification model training must be completed to provide a characterization benchmark and decision model for subsequent online early warning.

[0028] Before obtaining the baseline parameters for each process parameter, the early warning method also includes: collecting historical records of each process parameter under normal injection times corresponding to the preset baseline injection times during the machine stabilization phase; calculating the Pearson correlation coefficient between the historical records of any two process parameters; in response to the absolute value of the Pearson correlation coefficient being greater than or equal to a preset correlation threshold, directly using the Pearson correlation coefficient as the baseline coupling degree between the corresponding two process parameters; in response to the absolute value of the Pearson correlation coefficient being less than the correlation threshold, forcibly setting the baseline coupling degree between the corresponding two process parameters to zero.

[0029] In actual engineering, continuous acquisition is performed after the equipment is set up. The historical value of a normal injection cycle is used. Considering that after the injection molding production line is stabilized, the fluctuations of various process parameters need to go through a sufficient number of injection cycles to fully reflect the normal distribution characteristics, the baseline injection cycle number is selected. Set to 300, with a value range of [100, 500]; If the value is too small, the normal fluctuation estimate will be insufficient, leading to a decrease in the benchmark standard deviation. Deviation from the true value If the values ​​are too large, the baseline establishment period will be too long, delaying production; for any two process parameters and The Pearson correlation coefficient between the historical values ​​of the two data points was calculated using the standard Pearson correlation coefficient formula, thus obtaining the baseline coupling degree. A baseline coupling degree matrix is ​​constructed from the baseline coupling degrees of all parameter pairs. This matrix is ​​calculated once during the baseline establishment phase and stored permanently, without being updated in subsequent production. Since parameter pairs with extremely small absolute values ​​of Pearson correlation coefficients do not have substantial physical coupling relationships, the corresponding baseline coupling degrees must be set to zero to avoid noise participating in subsequent feature calculations. The correlation threshold ranges from [0.01, 0.1], and in this embodiment, it is set to 0.05, forcibly setting the corresponding baseline coupling degree to zero to avoid introducing noise interference in subsequent feature calculations due to extremely weak correlations. Taking this production line as an example, injection pressure and holding pressure belong to the same hydraulic circuit, and their baseline coupling degrees are close to positive 1; melt temperature and injection speed have a negative baseline coupling degree due to the negative feedback relationship of thermal balance in the injection molding process; there is no obvious physical correlation between melt temperature and holding time, and the absolute value of the Pearson correlation coefficient is less than 0.05, so the baseline coupling degree is set to zero.

[0030] During collection While recording historical values ​​for each injection, each process parameter is also recorded. ,exist For each bet, the benchmark mean is calculated using the standard mean formula and the standard deviation formula, respectively. Standard deviation from the benchmark , base average Standard deviation from the benchmark Together constitute process parameters The baseline parameters are fixed and stored. Baseline mean Representative process parameters At the center level under normal production conditions, the baseline standard deviation Representative process parameters Normal random fluctuation range; all subsequent online measured values ​​are based on the baseline mean. and the benchmark standard deviation Calculations are performed based on this.

[0031] After establishing the baseline parameters and baseline coupling degree, the classification model also needs to be trained for online application. The training steps of the support vector machine classification model include: dividing the process parameters of the historical normal production stage into a normal sample set; superimposing linearly varied data of a single parameter on the normal sample set to construct a partial sample set; selecting target parameter groups whose absolute value of the baseline coupling degree is greater than the strong coupling threshold, and controlling each parameter in the target parameter group to synchronously add linear offset data according to the sign direction of the baseline coupling degree to construct a collaborative sample set; extracting the enhanced features from the normal sample set, partial sample set, and collaborative sample set as the training set, and optimizing the initial model based on the training set to obtain the support vector machine classification model.

[0032] Normal sample set baseline stage In each bet, from the first The window sample is constructed sequentially for each injection, taking into account that gradual drift in the injection molding process usually appears gradually over dozens of injections, and the preset window length is... Setting it to 20, with a value range of [10, 50], effectively smooths out positive and negative random fluctuations within the window while avoiding a window that is too long, leading to a delayed response to the start of the drift. The partial sample set, based on the normal sample set, selects a single process parameter each time, during the drift duration period. Within 50 injection cycles, the process parameter offset is increased linearly from zero to... Provided that the cumulative offset of the drifted samples from start to end reaches a distinguishable level, the drift duration period... The value range is [30, 100]. In this embodiment... A value of 50 corresponds to the duration of a typical gradual drift; based on the conventional process capability index. Lower tolerance zone width equal to Estimate, The corresponding tolerance band width of 10% is in the early drift range where traditional threshold alarms cannot be reliably triggered. It can form a distinguishable cumulative difference with normal samples and meets the verification requirements of this invention for early gradual drift recognition capability. After generating the drift sequence, window samples are constructed one by one and marked as drift states to obtain a single biased sample set.

[0033] Further screening is conducted to construct a collaborative sample set of target parameter groups whose absolute value of baseline coupling degree is greater than the strong coupling threshold. The strong coupling threshold ranges from [0.5, 0.9], and in this embodiment, it is set to 0.7. If the strong coupling threshold is too small, a large number of weakly coupled parameter pairs will be included in the target parameter group, and the decision boundary will be contaminated by parameter pairs with no substantial physical coupling. If the strong coupling threshold is too large, the number of target parameter groups will be insufficient, and the classification model's ability to identify collaborative drift will decrease. For each target parameter group obtained through screening, the parameters within the group are controlled within the drift duration period. Linear offset data is synchronously superimposed according to the sign direction of the reference coupling degree, with the offset increasing linearly from zero to... When the baseline coupling is positive, the two parameters shift in the same direction; when the baseline coupling is negative, the two parameters shift in opposite directions. After generating the cooperative drift sequence, window samples are constructed one after another and marked as drift states to obtain the cooperative sample set.

[0034] The enhancement features of the three types of samples were obtained using the same calculation process: standard deviation calculation, two-sided recursion, and extraction of centroid distance and synchronous growth rate, forming the training set. The support vector machine classification model was trained using the training set, with the RBF radial basis function as the kernel function and kernel parameters... and penalty coefficient The search range was determined using a 5-fold cross-validation grid search. , The highest F1 score is used as the selection criterion to prioritize the recall rate of drift samples. and There exists a mutually exclusive relationship in model training where one element gains at the expense of the other: When the RBF kernel is too large, its scope narrows and the decision boundary overly fits the training samples, requiring a smaller kernel size. Suppress overfitting; When the RBF kernel is smaller, its action domain widens and its decision boundary flattens, requiring a larger kernel. To enhance the penalty for misclassified samples and ensure the recall rate of drifted samples, this embodiment uses grid search to obtain... , As a balance between the two, 5-fold cross-validation calibration was performed on the training set samples of this production line. Typical values ​​are distributed in the interval [1, 50]. The typical value is distributed in the range of [0.01, 1]. The value taken in this embodiment is... and All fall within their respective typical intervals. After training, the support vector machine classification model determines the optimal classification hyperplane and obtains the decision function. The decision function uses the enhanced features as independent variables and the decision values ​​as independent variables. For the dependent variable, the decision value A value greater than zero corresponds to the normal state; decision value A value less than or equal to zero corresponds to a drift state. By storing the decision function in a fixed manner, we obtain the support vector machine classification model, which can be called sequentially in the online stage.

[0035] After production begins, the injection molding machine controller will automatically output the measured values ​​of each process parameter at the end of each injection cycle. , , This is the current bet number, and the measured value. Data is collected directly from the injection molding machine controller, requiring no additional sensors.

[0036] S102, calculate the measured values ​​of each process parameter based on the reference parameters to obtain the standard deviation of each process parameter; perform bilateral recursion based on the standard deviation to obtain the positive and negative cumulative amounts of each process parameter; determine the dominant direction of each process parameter based on the positive and negative cumulative amounts.

[0037] Each bet completed The measured value output by the injection molding machine controller The system immediately enters the online calculation process, calculates the measured values ​​based on the benchmark parameters, eliminates the dimensional differences between different process parameters, and obtains the standard deviation of each process parameter. On this basis, a two-sided recursion is performed on the standard deviation, and the tiny gradual shift signals are accumulated into a statistical quantity that can characterize the drift trend by successive iterations. Furthermore, based on the relationship between the positive and negative cumulative quantities, the current dominant drift direction of each parameter is extracted to provide directional input for subsequent feature calculations.

[0038] The baseline parameters include the baseline mean and the baseline standard deviation. Based on the baseline parameters, the measured values ​​are calculated to obtain the standard deviation of each process parameter, including: calculating the parameter difference between the measured value of any process parameter in the current injection and the corresponding baseline mean; and taking the ratio of the parameter difference to the corresponding baseline standard deviation as the standard deviation of the process parameter.

[0039] Among them, process parameters In the order of the bet standard deviation Satisfying the relation:

[0040] In the formula, For process parameters In the order of the bet The standard deviation; For process parameters In the order of the bet The measured value is directly output by the injection molding machine controller at the end of each injection. For fixed storage of process parameters The baseline mean; For fixed storage of process parameters The baseline standard deviation. The current measured value is converted into a relative deviation in the unit of normal fluctuation range, which eliminates the dimensional differences of different process parameters and makes the deviations of each parameter comparable.

[0041] For example, if the reference average of the melt temperature 230 degrees Celsius, reference standard deviation The measured melt temperature for a certain injection is 1.5. If the value is 231.8, then the parameter difference is... The corresponding standard deviation This indicates that the current melt temperature is 1.2 standard deviations above the baseline.

[0042] After obtaining the standard deviation of each process parameter in the current injection, a two-sided recursion is performed on each process parameter to accumulate the drift signal in each injection. The two-sided recursion based on the standard deviation yields the positive and negative cumulative values ​​for each process parameter, including: obtaining the historical positive and negative cumulative values ​​for any process parameter in the previous injection; adding the historical positive cumulative value to the standard deviation and subtracting a preset drift reference value to obtain a first recursive value; using the maximum value between the first recursive value and zero as the positive cumulative value of the process parameter; subtracting the standard deviation and the drift reference value from the historical negative cumulative value to obtain a second recursive value; and using the maximum value between the second recursive value and zero as the negative cumulative value of the process parameter.

[0043] process parameters In the order of the bet positive cumulative amount Satisfying the relation:

[0044] In the formula, For process parameters In the order of the bet The positive cumulative amount; For process parameters In the previous bet The historical positive cumulative amount is obtained by recursion from the previous bet; This is the standard deviation calculated in the preceding steps; The drift reference value is 0.5. This is the standard theoretical parameter of the two-sided cumulative sum control chart, which corresponds to the detection point most sensitive to the gradual drift of a baseline standard deviation in the standardized space. It can be adjusted within the range of [0.25, 1] ​​according to the actual drift rate of the production line.

[0045] process parameters In the order of the bet negative cumulative amount Satisfying the relation:

[0046] In the formula, For process parameters In the order of the bet The negative cumulative amount; For process parameters In the previous bet The historical negative cumulative amount is obtained by recursion from the previous bet; This is the standard deviation calculated in the preceding steps; The drift reference value is 0.5. (This is used when the bet number is...) hour, and Take the initial value as zero, that is , The initial value of zero corresponds to the initial state of the process during the stable adjustment phase, where there is no historical drift accumulation.

[0047] Positive cumulative amount The positive portion of the cumulative standard deviation per injection exceeding the drift reference value, and the negative cumulative amount. The negative portion exceeding the drift reference value is accumulated; during normal random fluctuations, the positive and negative standard deviations cancel each other out, and neither the positive nor negative cumulative amount can continue to increase; during gradual drift, the standard deviation continuously shifts in the same direction, and the cumulative amount in the corresponding direction increases monotonically. The weak drift signal is amplified with each bet, eventually forming a feature amplitude that can be recognized by the classification model. It should be noted that when the bet number... Less than the preset window length At this time, the positive and negative cumulative amounts are recursively accumulated to gather historical information, but subsequent enhanced feature extraction and early warning decisions are not performed; the number of injections reaches the preset window length. Only then does it enter the online early warning stage.

[0048] After obtaining the positive and negative cumulative values, the dominant drift direction of each process parameter is extracted by comparing their magnitudes. Determining the dominant direction of each process parameter based on the positive and negative cumulative values ​​includes: comparing the numerical values ​​of the positive and negative cumulative values ​​for any process parameter; determining the dominant direction of the process parameter as a positive indicator value in response to a positive cumulative value being greater than or equal to the negative cumulative value; and determining the dominant direction of the process parameter as a negative indicator value in response to a positive cumulative value being less than the negative cumulative value, where the negative indicator value is the opposite of the positive indicator value.

[0049] Specifically, the positive indication value is taken The negative indicator value is taken That is, the negative indicator value is the opposite of the positive indicator value. Process parameters. In the order of the bet The dominant direction is denoted as The current injection process parameters The cumulative drift tendency is compressed into a single directional sign: if Greater than or equal to This indicates that the positive offset is dominant, and the dominant direction is indicated by the positive indicator value. ;like Greater than This indicates that the negative offset is dominant, and the dominant direction is indicated by the negative indicator value. Taking this production line as an example, if the positive cumulative amount of injection pressure... It continues to increase and exceeds the negative cumulative amount. A positive indicator value in the dominant direction indicates a continuous positive shift in injection pressure; dominant direction This will serve as the input for determining whether the drift direction of multiple parameters conforms to the process coupling law.

[0050] S103, calculate the mean vector norm of the standard deviation of all process parameters within the preset window to obtain the centroid distance; obtain the benchmark coupling degree between each process parameter, which is the correlation coefficient characterizing the degree of linear correlation between the historical values ​​of two process parameters; calculate the synchronous growth rate based on the benchmark coupling degree and the dominant direction of each process parameter.

[0051] After obtaining the standard deviation and dominant direction of each process parameter in the current injection, the characteristics of multi-parameter collaborative drift are further extracted from the two dimensions of overall offset magnitude and directional consistency.

[0052] The mean vector norm of the standard deviation of all process parameters within the preset window is calculated to obtain the centroid distance. This includes: taking the average of the standard deviations of any process parameter within the preset window for all iterations corresponding to that preset window, and obtaining the average deviation of each process parameter; calculating the sum of squares of the average deviations of all process parameters; and using the square root of the sum of squares as the centroid distance.

[0053] Among them, the preset window covers the nearest Each injection, for any process parameter Average deviation Satisfying the relation:

[0054] In the formula, For process parameters In the order of the bet The average deviation The preset window length is set to 20, based on the drift detection sensitivity requirements. For process parameters In the order of the bet The standard deviation.

[0055] Furthermore, the distance from the center of gravity Satisfying the relation:

[0056] In the formula, For the number of times The distance between the centers of gravity; This represents the total number of process parameters for this production line. It is 6; For process parameters In the order of the bet The average deviation is calculated from the previous step in this step; The centroid distance is the sum of the squares of the average deviations of all process parameters. Taking the square root of the sum of squares gives the centroid distance. .

[0057] Center of gravity distance The Euclidean norm of the vector formed by the average deviations of various process parameters reflects the degree to which the overall deviation of multiple parameters deviates from the normal baseline level within the current preset window. During normal random fluctuations, the positive and negative values ​​of each parameter cancel each other out within the preset window, and the average deviations all tend to be close to zero. (The centroid distance is also mentioned.) When the parameters approach zero and multiple parameters gradually drift, the average deviation of each parameter continues to shift in the same direction, and the distance between the centers of gravity... Significantly increased.

[0058] For example, if the melt temperature and injection pressure are each consistently higher than the baseline by 0.5 standard deviations over the last 20 injections, then the average deviation of the melt temperature and injection pressure is... Each of them is 0.5, and the average deviation of the other four parameters is approximately zero. This effectively captures the coordinated offset amplitude of the two parameters; if only a single parameter deviates significantly in a certain injection due to occasional disturbances, while the average deviation of the remaining parameters is close to zero, then the centroid distance... The value is relatively small and insensitive to occasional deviations, demonstrating the ability to identify preferences for cooperative multi-parameter drift.

[0059] After calculating the centroid distance, the baseline coupling degree is further used to determine the synergy of the drift directions of each process parameter, and this characteristic is represented by the synchronous growth rate. The Pearson correlation coefficient is a well-known statistical method in the field of statistical process control used to characterize the degree of linear correlation between two variables, while the weighted average is a well-known data processing method that combines multiple sub-indicators into a single value based on their respective importance. The synchronous growth rate combines these two methods: first, the Pearson correlation coefficient is used to solidify the coupling relationship of each parameter pair at the baseline stage; then, a weighted average is calculated based on the directional consistency scores of each parameter pair, with the coupling strength as the weight. This transforms the qualitative judgment of whether the drift directions of multiple parameters conform to the physical coupling law of the process into a quantifiable scalar indicator. The reason for using coupling strength as a weight is that in injection molding production, equipment degradation phenomena such as screw wear leading to decreased plasticizing capacity and hydraulic oil temperature rise causing system pressure fluctuations, as the same physical root cause, simultaneously drive multiple process parameters affected by these degradation phenomena to shift synchronously along their respective coupling directions. The stronger the coupling, the higher the consistency of the shift direction under the same root cause, and the lower the probability of deviating from the coupling direction. Using coupling strength as a weighting coefficient can enable the synchronous growth rate to obtain a higher response amplitude during real coordinated drift, while maintaining a low value during normal random fluctuations. The synchronous growth rate is calculated based on the baseline coupling degree and the dominant direction of each process parameter, including: extracting the average value of the dominant direction of each process parameter within the most recent preset direction window to obtain the stability factor, and combining all process parameters in pairs to obtain multiple parameter pairs.

[0060] For any pair of parameters, in response to the consistency of the sign of the product of the stability factors of the two process parameters with the sign of the corresponding baseline coupling degree, the product of the absolute values ​​of the stability factors of the two process parameters is taken as the consistency score; the product of the consistency scores of all parameter pairs and the corresponding absolute values ​​of the baseline coupling degree is summed to obtain the comprehensive weighted score, and the comprehensive weighted score is divided by the sum of the absolute values ​​of the baseline coupling degrees of all parameter pairs, and the quotient is taken as the synchronization growth rate; when the sum of the absolute values ​​of the baseline coupling degrees of all parameter pairs is less than a preset lower threshold, the equal weighted average of the consistency scores of each parameter pair is used instead of the above weighted result as the synchronization growth rate.

[0061] It is worth noting that the dominant direction of a single injection may be reversed due to occasional disturbances. Before judging the consistency of direction, the average value of the dominant direction within a few recent injections is introduced as a stability factor to eliminate the impact of occasional reversals. For any process parameters... Stability factor Satisfying the relation:

[0062] In the formula, For process parameters In the order of the bet Stability factor; The recent direction window length is used for directional smoothing of the stability factor. The reference value is 5, which can be adjusted within the range of [3, 10] based on the frequency of fluctuations in the production line process. The preset direction window is the nearest. A window composed of individual entries, For the number of times medium process parameters The dominant direction, with a value that is a positive indicator value. Or negative indicator value The above formula represents the recent The dominant direction within each bet. Sum of each term and then divide by The stability factor is obtained. .

[0063] Stability factor The value range is [-1, 1], and the absolute value reflects the stability of the recent dominant direction: the closer the absolute value is to 1, the stronger the recent trend. The more consistent the dominant direction of each injection, the higher the directional stability; the closer the absolute value is to 0, the more frequently the direction reverses, and the better the process parameters. It may be in a state of random fluctuation rather than a true drift. Taking this production line as an example, if the injection pressure has been a positive indicator value in the dominant direction for the last 5 injections... Then the stability factor for This indicates that the injection pressure is continuously shifted positively and the direction is highly stable; if the dominant direction is at the positive indication value... With negative indicator value Alternating between them, stability factor A value close to 0 indicates that the process parameters are close to 0. There is no obvious unidirectional drift trend at present.

[0064] Furthermore, all The process parameters are combined in pairs to obtain A parameter pair for this production line A total of 15 parameter pairs are obtained by setting 6. For any parameter pair... , Consistency score Satisfying the relation:

[0065] In the formula, For parameter pairs In the order of the bet Consistency score; This is an indicator function; it takes the value 1 if the condition within the square brackets is met, and 0 otherwise. and These are the process parameters. and process parameters In the order of the bet The stability factor is calculated in the previous step of this procedure; For parameter pairs The baseline coupling degree is calculated once during the baseline establishment phase and then stored in a fixed manner; It is a symbolic function, and its value is... or .

[0066] The consistency score is calculated when the sign of the product of the stability factors of the two process parameters in the parameter pair is consistent with the sign of the baseline coupling degree. Take the product of the absolute values ​​of the two stability factors; otherwise, score the consistency score. Set to zero. Taking injection pressure and holding pressure as an example, the reference coupling degree between injection pressure and holding pressure. If the stability factor of a certain injection pressure is positive, for Stability factor of holding pressure for The product sign is positive, indicating a high degree of coupling with the reference. Sign consistency, consistency score This indicates that the injection pressure and holding pressure are drifting synchronously in the positive direction according to the physical laws of the process; if the stability factor of the injection pressure for Stability factor of holding pressure for The product sign is negative, indicating a positive base coupling degree. Inconsistent signs, consistency score A score of zero indicates that the injection pressure and holding pressure drift in opposite directions, and do not constitute coordinated drift. Consistency Score Using the absolute value of the stability factor as the credibility weight, the consistency score increases when any parameter direction is unstable. Automatic scaling effectively suppresses misjudgments caused by occasional directional flips.

[0067] Ultimately, the synchronous growth rate Satisfying the relation:

[0068] In the formula, For the number of times The synchronous growth rate; the numerator is the consistency score for all parameter pairs. absolute value of coupling degree with the corresponding reference The summation of the products is the overall weighted score; the denominator is the absolute value of the baseline coupling degree for all parameter pairs. The sum; in response to the denominator being less than a preset lower threshold, the consistency score of all parameter pairs is calculated. Summing and dividing by the parameter relative to the total number The obtained quotient is used as the synchronous growth rate, that is, the weighted average of the consistency scores of each parameter is used instead of the weighted calculation result to ensure the stability of the scheme. In this embodiment, the lower limit threshold is set to 0.01. In other embodiments, it can be based on the total number of parameters. The coupling degree with the reference is adjusted within the range of [0.001, 0.05]. When the denominator... When it is less than the preset lower threshold, according to calculate.

[0069] Synchronous growth rate The value range is [0, 1], and its physical meaning is: after weighting by historical coupling strength, what proportion of the parameters in the current injection shift synchronously with the drift direction in a stable manner that conforms to the process coupling law. During normal random fluctuations, the stability factor of each parameter... The absolute value is small and the direction is almost random, with a synchronous growth rate. It remains at a low level around 0.25; during true coordinated drift, the strongly coupled parameter pair continuously shifts synchronously according to the process rules, with highly stable direction and synchronous growth rate. It tends towards 1, forming a significant distinction from the normal state.

[0070] Center of gravity distance The overall magnitude of multi-parameter deviations is compressed into a single scalar, the synchronous growth rate. Then, it is determined whether the offset direction conforms to the process coupling law. The two are complementary and provide a basis for subsequent classification models to distinguish between single-parameter occasional offset and multi-parameter collaborative gradual drift.

[0071] S104, by splicing together the positive cumulative amount, negative cumulative amount, center of gravity distance and synchronous growth rate of each process parameter, the enhanced feature is obtained.

[0072] Current bet The positive cumulative amount, negative cumulative amount, centroid distance, and synchronous growth rate of each process parameter have been calculated. These statistics are then concatenated in a fixed order to construct an enhanced feature that can simultaneously characterize both single-parameter independent migration and multi-parameter cooperative drift. Enhanced Feature Satisfying the relation:

[0073] In the formula, For the number of times Enhanced features, with dimensions of For this production line It has 6 dimensions, for a total of 14 dimensions; For process parameters In the order of the bet The positive cumulative amount; For process parameters In the order of the bet The negative cumulative amount; and By parameter number Arranged in pairs in sequence, total One component; For the number of times The distance between the centers of gravity, arranged in the th order One location; For the number of times The synchronous growth rate is ranked in the first place. One position.

[0074] From the physical meaning of feature construction, enhanced features It also contains information at three levels: 2m positive and negative cumulative values ​​retain detailed information on the independent offset direction and magnitude of each process parameter; center of gravity distance. This compresses the overall magnitude of multi-parameter deviations within a preset window into a single scalar, making it particularly sensitive to the unidirectional accumulation of multi-parameter deviations; synchronous growth rate Further characterization is needed to determine whether the multi-parameter offset directions are synchronously offset in a stable manner that conforms to the process coupling structure. These three aspects complement each other, enabling the classification model to effectively distinguish between two typical scenarios: when a single parameter experiences occasional large offsets, the positive cumulative amount of the corresponding parameter... or negative cumulative amount Larger, but the distance between the center of gravity and the center of gravity is greater. Limited, synchronous growth rate Lower; during multi-parameter cooperative gradual drift, the positive cumulative amount of multiple parameters. or negative cumulative amount Continuing to increase, the distance between the centers of gravity Significant increase and synchronous growth rate It tends towards 1; during normal random fluctuations, the positive cumulative amount With negative cumulative amount Neither can be accumulated continuously, the distance between the centers of gravity If the value is close to zero, no drift decision will be triggered.

[0075] Enhanced features Each injection cycle of the injection molding production line Each process parameter data is compressed into A fixed-length vector of dimension allows the classification model to make a warning decision after each injection.

[0076] S105, the enhanced features are input into the preset support vector machine classification model for judgment to obtain the warning result; in response to the warning result being in a drift state, a warning command is issued.

[0077] In another embodiment, the current number of times Enhanced features After the calculation is complete, the data is immediately input into a pre-trained support vector machine (SVM) classification model for decision-making. The SVM classification model enhances the features. Calculate decision value The warning result is obtained: response to the decision value. If the value is greater than zero, the current bet is considered to be in a normal state, and no warning is output; this is in response to the decision value. If the value is less than or equal to zero, the current bet is considered to be in a drift state.

[0078] In response to a drifting state in the early warning result, a continuous confirmation mechanism is used to further determine whether to issue an early warning command. Issuing an early warning command includes: storing the early warning result of the current injection output into a preset continuous decision queue; reading multiple historical early warning results from the continuous decision queue; and, in response to the fact that all historical early warning results in the continuous decision queue for a preset number of consecutive confirmations are in a drifting state, determining that a real drift event has occurred in the production process and issuing an early warning command.

[0079] The design consideration for the preset confirmation count is that, with a reference value of 3, the warning command is only triggered when the warning results of 3 consecutive injections are all in a drift state. The injection number that was first judged to be in a drift state is recorded for operators to trace the start time of the drift. The preset confirmation count can be adjusted within the range of [2, 5] according to the actual false alarm rate requirements of the production line: the larger the preset confirmation count, the lower the false alarm rate, but the longer the warning delay; the smaller the preset confirmation count, the faster the response speed, but the slightly higher the false alarm rate. After receiving the warning, operators can check the positive cumulative amount of each process parameter. With negative cumulative amount The value is used to locate the parameter with the largest cumulative value, which helps to identify the root cause of the drift. For example, if the positive cumulative amount of injection pressure... The abnormal value is too high, indicating that the hydraulic circuit pressure remains excessively high.

[0080] After issuing the warning command and the operator completes the parameter adjustment, the warning method also includes: receiving the system reset command sent from the outside; in response to the system reset command, resetting all positive and negative cumulative values ​​of all process parameters to zero; clearing the historical cache data of the standard deviation of all process parameters in the preset window so that the center of gravity distance can start the cumulative calculation again in the next injection.

[0081] Positive cumulative amount and negative cumulative amount The reset process restarts the bilateral recursion from zero, eliminating the residual influence of historical drift data from before parameter adjustment in subsequent injections; it also clears the standard deviation within the preset window. Historical cached data, making the center of gravity distance Accumulation resumes from the reset point to prevent the pre-adjustment offset data from further increasing the center of gravity distance in the short term after reset. This interferes with the early warning decision. After completing the above reset operation, the injection molding production line returns to normal monitoring status and awaits the next injection measurement value. Input to begin a new drift warning cycle.

[0082] Reference Figure 2 In the context of coordinated gradual drift, a linear gradual offset is applied simultaneously to the injection pressure and holding pressure. The drift characteristic amplitude curve corresponding to this invention begins to deviate from the normal fluctuation baseline around injection count 500, exhibiting a continuous monotonically increasing trend with increasing injection count, rising to approximately 0.68 around injection count 2000, indicating enhanced characteristics. It effectively captured the cumulative effect of the cooperative drift signal per injection; the drift characteristic amplitude curve corresponding to the traditional threshold alarm method always fluctuated randomly around the normal fluctuation baseline of 0.3 to 0.4 throughout the entire observation interval, without forming an identifiable upward trend, indicating that the threshold logic of independent comparison of each parameter is not sensitive to gradual cooperative drift.

[0083] Reference Figure 3 By superimposing occasional single-entry fluctuations of varying magnitudes onto normal production data, the false alarm rates of the proposed method and the traditional threshold alarm method were statistically analyzed. When the amplitude of the occasional fluctuation increases from 0 to 1 baseline standard deviation, the false alarm rate curve of the traditional threshold alarm method rises sharply, accelerating to approximately 50% after the amplitude reaches 0.5; the false alarm rate curve of the proposed method rises more gently, remaining below approximately 25% even when the amplitude of the occasional fluctuation is 1. The confirmation mechanism of the continuous decision queue effectively filters out false alarms caused by occasional single-entry fluctuations, resulting in a significantly lower false alarm rate for the proposed method within the range of large occasional deviations compared to the traditional threshold alarm method.

[0084] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for early warning of injection molding process parameter drift based on support vector machine, characterized in that, include: Obtain the baseline parameters of each process parameter and the measured values ​​of each process parameter in the current injection; The standard deviation of each process parameter is obtained by calculating the measured values ​​of each process parameter based on the reference parameters. Based on the standard deviation, a two-sided recursion is performed to obtain the positive and negative cumulative values ​​of each process parameter; the dominant direction of each process parameter is determined based on the positive and negative cumulative values. Calculate the mean vector norm of the standard deviation of all process parameters within the preset window to obtain the centroid distance; obtain the benchmark coupling degree between each process parameter, which is the correlation coefficient characterizing the degree of linear correlation between the historical values ​​of two process parameters; The synchronization growth rate is calculated based on the baseline coupling degree and the dominant direction of each process parameter. This includes: extracting the average value of the dominant direction of each process parameter within the most recent preset direction window to obtain the stability factor, and combining all process parameters in pairs to obtain multiple parameter pairs; for any parameter pair, if the sign of the product of the stability factors of the two process parameters is consistent with the numerical sign of the corresponding baseline coupling degree, the product of the absolute values ​​of the two stability factors is used as the consistency score; otherwise, the consistency score is set to zero; using the absolute value of the baseline coupling degree of each parameter pair as the weight, the consistency scores of each parameter pair are weighted, and the sum is divided by the sum of the absolute values ​​of all baseline coupling degrees, and the quotient is used as the synchronization growth rate; when the sum is less than a preset lower threshold, the equally weighted average of the consistency scores of each parameter pair is used instead. The meaning of synchronous growth rate is: after weighting by historical coupling strength, what proportion of the parameters in the current injection shift synchronously with the drift direction in a stable manner that conforms to the process coupling law; The enhanced features are obtained by combining the positive cumulative amount, negative cumulative amount, center of gravity distance, and synchronous growth rate of various process parameters. The enhanced features are input into a pre-defined support vector machine classification model for judgment, and an early warning result is obtained; in response to the early warning result being in a drift state, an early warning command is issued. The support vector machine classification model is obtained through the following training steps: dividing the process parameters of the historical normal production stage into a normal sample set; superimposing linearly varied data of a single parameter on the normal sample set to construct a partial sample set; selecting target parameter groups whose absolute value of the baseline coupling degree is greater than the strong coupling threshold, and controlling each parameter in the target parameter group to synchronously add linear offset data according to the sign direction of the baseline coupling degree to construct a collaborative sample set; extracting the enhanced features from the normal sample set, partial sample set, and collaborative sample set as the training set, and optimizing the initial model based on the training set to obtain the support vector machine classification model.

2. The injection molding process parameter drift early warning method based on support vector machine according to claim 1, characterized in that, The benchmark parameters include the benchmark mean and the benchmark standard deviation; The calculation of the measured values ​​of each process parameter to obtain the standard deviation of each process parameter includes: calculating the parameter difference between the measured value of any process parameter in the current injection and the corresponding benchmark mean. The ratio of the parameter difference to the corresponding baseline standard deviation is used as the standard deviation of the process parameter.

3. The injection molding process parameter drift early warning method based on support vector machine according to claim 1, characterized in that, The positive and negative cumulative values ​​of each process parameter obtained by performing a two-sided recursion based on the standard deviation include: Obtain the historical positive and negative cumulative values ​​of any process parameter in the previous injection; The historical positive cumulative amount is added to the standard deviation and subtracted from the preset drift reference value to obtain the first recursive value. The first recursive value is compared with zero, and the maximum value between the first recursive value and zero is taken as the positive cumulative amount of the process parameter. The second recursive value is obtained by subtracting the standard deviation and drift reference value from the historical negative cumulative amount. The second recursive value is compared with zero, and the maximum value between the second recursive value and zero is taken as the negative cumulative amount of the process parameter.

4. The injection molding process parameter drift early warning method based on support vector machine according to claim 1, characterized in that, The dominant direction for determining each process parameter based on positive and negative cumulative amounts includes: Compare the magnitudes of the positive and negative cumulative values ​​of any process parameter; in response to a positive cumulative value being greater than or equal to a negative cumulative value, determine the dominant direction of the process parameter as a positive indicator value, wherein the positive indicator value is taken as... In response to a positive cumulative amount being less than a negative cumulative amount, the dominant direction of the process parameter is determined as a negative indication value, wherein the negative indication value is taken as... .

5. The injection molding process parameter drift early warning method based on support vector machine according to claim 1, characterized in that, The calculation of the mean vector norm of the standard deviation of all process parameters within the preset window to obtain the centroid distance includes: taking the average of the standard deviations of any process parameter within the preset window under all injections corresponding to the preset window to obtain the average deviation of each process parameter; calculating the sum of squares of the average deviations of all process parameters; and taking the square root of the sum of squares as the centroid distance.

6. The injection molding process parameter drift early warning method based on support vector machine according to claim 1, characterized in that, Before obtaining the baseline parameters for each process parameter, the early warning method further includes: Collect historical data of each process parameter under normal injection number corresponding to the preset baseline injection number during the machine adjustment and stabilization phase; Calculate the Pearson correlation coefficient between the historical values ​​of any two process parameters; In response to the absolute value of the Pearson correlation coefficient being greater than or equal to a preset correlation threshold, the Pearson correlation coefficient is directly used as the benchmark coupling degree between the two corresponding process parameters. In response to the absolute value of the Pearson correlation coefficient being less than the correlation threshold, the baseline coupling degree between the two corresponding process parameters is forcibly set to zero.

7. The injection molding process parameter drift early warning method based on support vector machine according to claim 1, characterized in that, The response to a drifting warning result includes issuing a warning command, which includes: Store the warning result of the current bet into a preset continuous judgment queue; Read multiple historical warning results from the continuous judgment queue; If all the warning results in the continuous decision queue for a preset number of consecutive confirmations are in a drift state, it is determined that a real drift event has occurred in the production process, and a warning instruction is issued.

8. The injection molding process parameter drift early warning method based on support vector machine according to claim 1, characterized in that, After issuing the warning command, the warning method further includes: Receive system reset commands sent from external sources; In response to the system reset command, all positive and negative cumulative values ​​of all process parameters are reset to zero. Clear the historical cache data of the standard deviation of all process parameters in the preset window so that the centroid distance can be recalculated in the next injection.

Images