A method for progressive dynamic adjustment of injection molding process parameters based on product quality prediction

Abstract

The application discloses a kind of based on product quality pre-judgment injection molding process parameter gradual dynamic adjustment method.It includes the following steps: define prophet-LSTM based defect function;Prophet-LSTM based hybrid model is constructed and trained;Obtain and introduce correction factor with preprocessed injection molding key state data;Set variation degree judgment condition;Using prophet-LSTM model based parameter pre-judgment and adjustment, complete injection molding process parameter gradual process adjustment.The application can analyze the influence law of periodic working condition factors on injection molding machine production process, fuse multidimensional factors to predict possible defects of injection molded products, and adjust injection molding process parameters in advance, which helps to avoid the generation of injection molded product defects and maintain high consistency of quality.

Description

Technical Field

[0001] This invention relates to the fields of machine learning and industrial automation, and specifically to a method for progressively and dynamically adjusting injection molding process parameters based on product quality prediction. Background Technology

[0002] Injection molding is typically a continuous process. Current injection molding processes rely on manual adjustments to process parameters based on the type and visually approximate size of defects in the molded products to control quality. However, the manual identification of defect types and sizes is inherently delayed, and the actual effect of these adjustments is also delayed. Therefore, adjustments during production can easily lead to unstable product quality, even resulting in a large number of defective products, and also impact production efficiency.

[0003] Therefore, optimizing the adjustment process and improving the reliability of the continuous injection molding process, identifying the changing trends of key product status data before defective products appear, and adjusting the injection molding process parameters in advance to avoid defects is also a trend in the intelligent development of injection molding equipment. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a method for progressively dynamic adjustment of injection molding process parameters based on product quality prediction. By real-time acquisition and recording of key state data related to defects in injection molded products, and by introducing a deep learning-based Prophet-LSTM algorithm model and correction factors, the method analyzes the influence of periodic operating conditions on the injection molding machine's production process, predicts defects in injection molded products, and pre-adjusts injection molding process parameters before defects occur to prevent their generation.

[0005] The technical solution of the present invention includes the following steps:

[0006] Step 1: Define the defect function based on prophet-LSTM;

[0007] Step 2: Construct and train a hybrid model based on prophet-LSTM;

[0008] Step 3: Set correction factors and revise the model based on the correction factors;

[0009] Step 4: Predict the injection molding process parameters based on the prophet-LSTM model, and adjust the injection molding process parameters according to the defect variation judgment conditions.

[0010] The definition of the prophet-LSTM defect function in step 1 is as follows:

[0011] y(t)=ω1P(t)+ω2L(t).

[0012] The construction of the prophet-LSTM hybrid model in step 2 includes the following steps:

[0013] 2.1) Establish a dataset {Y} with the number of days j as the unit. (1) Y (2) …Y (j-1) ,Y (j)}, Y (j) This is a time series of injection molding data obtained from the detection on day j;

[0014] From {Y (1) Y (2) …Y (j-1) ,Y (j) Data from the same time period were taken from the LSTM network to analyze the changing patterns of the periodic part and obtain the trend part of the prophet model.

[0015] 2.3) For Y (1) Y (2) …Y (j) The data is normalized.

[0016] 2.4) Input the normalized data from step 2.3) into the prophet-LSTM model for prediction training:

[0017] loss function The prophet model is trained with weights set to minimize p. t For the predicted value, y t The values ​​are the actual detected values; then, the Prophet model is used to predict the defect values ​​of the injection molded products in the i-th cycle within day j+1, and the result is denoted as P(i):{P 1,i (j+1) ,P 2,i (j+1) …P s,i (j+1)}, P s,i (j) This represents the predicted value of the prophet model for the s-th defect value;

[0018] After optimization using the adaptive estimation (Adam) algorithm, the LSTM model predicts the defect value of the injection molded product in the i-th cycle within day j+1, and the result is denoted as L(i):{L 1,i (j+1) ,L 2,i (j+1) …L s,i (j+1)}, L i,s(j) This represents the predicted value of the LSTM model for the s-th defect value;

[0019] 2.5) Combine the prediction results of the Prophet model and the LSTM model. Specifically, perform a weighted combined prediction on y(t)=ω1P(t)+ω2L(t), where y(t) is the overall prediction result, P(t) is the prediction result of the Prophet model, and L(t) is the prediction result of the LSTM model.

[0020] Under the constraint w1+w2=1, w1 or w2 is taken from 0 to 1 in steps of 0.1, and the values ​​of w1 and w2 are determined with the goal of minimizing the sum of squared residuals, thus completing the initial construction of the prophet-LSTM hybrid model.

[0021] The injection molding data time series in step 2.1):

[0022] a) The injection molding data time series includes time, process parameter detection values, and defect detection values, which are obtained through real-time monitoring of the injection molding process of the injection molding machine;

[0023] b) The injection molding data time series includes s major defect types and i injection cycles:

[0024] Based on injection molding experience and customer requirements, s main defect types were identified as the research subjects;

[0025] The number of injection cycles i is determined based on the injection process, where i∈(3,36).

[0026] When there are fewer than 3 data points, the data is too small to show a certain pattern, so the defect value of the previous injection cycle is returned as the prediction value. When there are more data points, the amount of calculation will increase, and data that is too far away from the current injection cycle will have a negative impact on the prediction results, resulting in unsatisfactory prediction results.

[0027] Step 3 specifically involves:

[0028] 3.1) Define the corrective factor:

[0029] ξ i-n =ξ0 / 2 n

[0030] Where i represents the number of corrections; n is initially 0; and ξ0 is the initial correction factor.

[0031] The weighting coefficients of the correction factors gradually increase from the beginning to the end, allowing the later correction factors to have a greater impact on the prediction results. The weights decay exponentially, meaning that as the distance from the predicted data increases, the weight of the correction factor decreases exponentially.

[0032] 3.2) Correct the model in step 2) based on the correction factor:

[0033]

[0034] Where Δ is a user-defined error value;

[0035] 3.3) Repeat step 3.2) until the average error is within the range (-Δ%, Δ%), then output the corrected model;

[0036] 6. The method for progressive dynamic adjustment of injection molding process parameters based on product quality prediction according to claim 5, characterized in that the correction factor in the i-th correction of step 3.2) is improved through the following analysis:

[0037] a) When the predicted value is greater than the actual detected value and the average difference exceeds Δ%, then n = n + 1;

[0038] b) When the predicted value is less than the actual detected value and the average difference exceeds Δ%, then n = n-1;

[0039] c) In all other cases, n remains unchanged.

[0040] Step 4 specifically involves:

[0041] 4.1) Using the Prophet-LSTM hybrid model obtained in step 3, predict the defect values ​​of the product in the current injection molding cycle, i.e., the nth cycle of day j. The result is denoted as Y(Y):{ 1,n (j) ,Y 2,n (j) …Y s,n (h)};

[0042] Among them, Y m,n (j) The predicted time series of the m-th defect in the n-th period on day j contains k consecutive predicted values;

[0043] Each defect in each injection molding cycle corresponds to k consecutive defect prediction values, specifically:

[0044] For the nth period of the mth type of defect, t i The defect prediction value at time t; i∈[0,k), and i∈Z; m∈[1,s], and m∈Z;

[0045] 4.2) Calculate the degree of variation δ for each defect based on the defect prediction value Y(n):

[0046]

[0047] In the formula, t is the nth period i The defect change value at any given time; For the m-th defect in the n-th period, t i The defect change value at any given time;

[0048] 4.3) Determine whether to initiate process parameter adjustment based on the predicted defect value;

[0049] a) When the predicted defect value to be adjusted does not exceed the conservative threshold, i.e., y(t) i ) <tv{tv1,tv2,…,tv s If the process parameters are not adjusted, it is directly determined that no adjustment is needed.

[0050] Where tv{tv1,tv2,…,tv s} represents the conservative threshold for the set s types of defects;

[0051] b) When the predicted defect value to be adjusted is greater than the conservative threshold and less than or equal to the acceptable threshold, i.e., tv{tv1,tv2,…,tv} s} <y(t i+1 ) <Tv{Tv1,Tv2,…,Tv s}, determine the degree of change in process parameters;

[0052] Where Tv{Tv1,Tv2,…,Tv s} represents the set of acceptable threshold values ​​for s types of injection molding defects;

[0053] When the degree of defect variation is less than the degree of variation threshold, i.e. No adjustment is made to the injection molding process parameters; conversely, based on y(t)... 修 Adjust the process parameters;

[0054]

[0055] In the formula, It is a threshold set for the degree of change; y(t) final The defect prediction value at time t is obtained by the prophet-LSTM model after correction in step 3; δ is the defect variability value at time t; η is the correction coefficient, which is greater than or equal to the defect variability.

[0056] c) When the predicted defect value to be adjusted is greater than the qualified threshold, skip the change judgment condition and restore all process parameters to their initial values.

[0057] Step 4.2) specifically refers to

[0058] 4.2.1) A combination of least squares algorithm and cubic fitting polynomial method was used to fit the injection molding data time series;

[0059] 4.2.2) Adjust the coefficients of the first-order term a1 and the second-order term a2 in the polynomial of step 4.2.1) to eliminate the influence of abrupt changes on the results and improve the accuracy of collecting data near the endpoints of discrete sequences; the polynomial terms a1 and a2 are adjusted as follows:

[0060]

[0061]

[0062] Where a3 is the cubic term in the polynomial of step 2.1); S represents the error limit;

[0063] After adjustment, the first derivative of the defect detection value and time relationship of the injection molding data time series is obtained using the difference theorem. The derivative of the defect detection value corresponding to the last moment in the injection molding data time series (i.e., at the right endpoint of the fitted curve) is used as the defect variability value of the next injection molding data time series.

[0064] At time i, t i The defect variation value is:

[0065]

[0066] In the formula, f(t) i f(t) i-1 ) represents t i ,t i-1 The defect detection value, h is the collection time interval, y(t) i+1 ) is t i+1 The defect prediction value at time t; i∈[0,k), and i∈Z. i ,t i-1 When the defect value at time t is only the defect prediction value and there is no actual recorded defect detection value, i The formula for calculating the defect variation value at time t is:

[0067]

[0068] In the formula, y(t) i-1 ) is t i-1 Defect prediction value at time t; y(t) i ) is t i Defect prediction value at any given time.

[0069] The variability determination method of this invention is based on a polynomial fitting method combining the difference theorem and the least squares method to solve for the variability value, which can determine the predicted value Y. jConservative corrections improve the accuracy of defect value prediction and the robustness of the injection molding feedback process parameter dynamic adjustment system, reduce the defect setting threshold, and improve the quality control level of injection molded products. It can also be used as one of the conditions for judging whether the injection molding process parameters need to be adjusted.

[0070] The correction factor of this invention decays exponentially, and the weighting coefficient of the correction factor gradually increases, so that the correction factors located later play a greater role in the prediction results, thereby realizing the gradual adjustment of process parameters.

[0071] The dynamic progressive adjustment method of the present invention can ensure the stability of the system with minimal adjustment operations, and provide a data basis for process parameter adjustment.

[0072] The beneficial effects of this invention are:

[0073] This invention can analyze the influence of periodic operating conditions on the injection molding machine production process, integrate multi-dimensional factors to predict potential defects in injection molded products, and adjust injection molding process parameters in advance, which helps to avoid the occurrence of defects in injection molded products and maintain a high degree of quality consistency. Attached Figure Description

[0074] Figure 1 A diagram defining the defect degree function value;

[0075] Figure 2 Flowchart for judging the degree of quality change;

[0076] Figure 3 For the detection data of underfill defects of the same product on the same injection molding machine;

[0077] Figure 4 This is a diagram illustrating the dynamic adjustment process of process parameters based on the Prophet-LSTM model.

[0078] Figure 5 The figures are the fitted curves before and after coefficient adjustment. Detailed Implementation Plan

[0079] Example 1:

[0080] Step 1: Define the overall defect function

[0081] y(t)=ω1P(t)+ω2L(t)

[0082] The 1.1 prophet model is defined as follows:

[0083] P(t)=g(t)+s(t)+h(t)+ε t

[0084] 1.1.1 The core part of the model, the trend component g(t), is used to analyze non-periodic changes in time series data and predict trend direction. It is a logistic function:

[0085]

[0086] In the formula, C(t) is the saturation value, (k+a(t)) T δ) is the growth rate at time t, (m+a(t) T γ is the bias parameter, and δ is the change in growth rate at the inflection point. In time series data, there are several inflection points, denoted as s. j j = 1, ..., S, where:

[0087]

[0088]

[0089] 1.1.2 The periodic component s(t) is represented by a Fourier series model:

[0090]

[0091] In the formula, T is the period length, N is the number of periods, and a n ,b n These are estimated parameters.

[0092] 1.1.3 Holidays and Emergencies Section h(t):

[0093] h(t)=Z(t)κ

[0094] In the formula, Z(t) = [1(t∈D1),…,1(t∈D1)] L )],k~Normal(0,v 2 () represents the impact of random factors during the window period on the predicted value.

[0095] 1.2 The LSTM output is defined as follows:

[0096] h t =o t ·tanh(c t-1 ·f t +i t ·tanh(W cx x t +W ch h t-1 +b c ))

[0097] Input gate:

[0098] i t=sigmoid(W ix x t +W ih h t-1 +b i )

[0099] Forgotten Gate:

[0100] f t =sigmoid(W fx x t +W fh h t-1 +b f )

[0101] Output gate:

[0102] o t =sigmoid(W ox x t +W oh h t-1 +b o )

[0103] Among them, memory units:

[0104] c t =c t-1 ·f t +i t ·tanh(W cx x t +W ch h t-1 +b c )

[0105] In the formula i t ,f t ,o t ,c t These represent the input gate, forget gate, output gate, and memory unit, respectively.

[0106] sigmoid is denoted as σ and tanh is the activation function.

[0107] W ix W ih These refer to the weight matrix of the input information to the input gate and the weight matrix of the previous output to the input gate, respectively.

[0108] W fx W fh These refer to the weight matrix of the input information to the forget gate and the weight matrix of the previous output to the forget gate, respectively.

[0109] W ox W ohThese refer to the weight matrix from the input information to the output gate and the weight matrix from the previous output to the output gate, respectively.

[0110] b i ,b f ,b o These refer to the bias values ​​of the input gate, output gate, and forget gate, respectively.

[0111] 1.3 The coefficient is defined as follows:

[0112] w1 + w2 = 1 and w1, w2 ∈ (0, 1)

[0113] The composition of the overall defect value is as follows Figure 1 As shown.

[0114] Step 2: Define defect value threshold and variability judgment conditions

[0115] 2.1 Quantify the degree of change and apply a conservative correction to the predicted defect value. The correction value is calculated as follows:

[0116]

[0117] In the formula, y(t) is the defect value at time t predicted by the algorithm in step 1, δ is the degree of change, and η is the correction coefficient, which is greater than the degree of change.

[0118] 2.2 The method of combining the least squares algorithm and the cubic fitting polynomial is used to fit the discrete defect detection values.

[0119] 2.3 Adjust the coefficients of the first and second terms to eliminate the influence of abrupt changes on the results, thereby improving the accuracy of collecting discrete sequence near the endpoints.

[0120] 2.4 Calculate the first derivative of the discrete data using the difference theorem, and take the derivative at the last discrete data point (i.e., the right endpoint of the fitted curve) as the defect variation value for the next discrete data point. The calculation is as follows:

[0121] The fitted polynomial terms a1 and a2 are adjusted as follows:

[0122]

[0123]

[0124] After adjustment, t i The defect variation value at time t is:

[0125]

[0126] Where S represents the error limit, t i Let f(t) represent the time at point i. i ), f(t) h-1) represents t i ,t i-1 The detected value, h is the collection time interval, Y(t) i+1 ) is t i+1 Defect prediction value at any given time.

[0127] Variation value δ i It is related to the polynomial coefficients a3 and the interval h.

[0128] 2.5 Using the degree of change δ i The predicted values ​​are corrected, and the corrected predicted values ​​are used as the basis for adjusting the injection molding process parameters.

[0129] y(t i+1 ) <tv{tv1,tv2,…,tv s}

[0130]

[0131] In the formula, Tv{Tv1,Tv2,…,Tv} s},tv{tv1,tv2,…,tv s} are the set of acceptable thresholds for s types of injection molding defects and the defect conservatism threshold, respectively. It is a threshold set for the degree of change. When the predicted value Y of the i-th type of defect... j and the obtained threshold δ j If the above requirements are not met, it is determined that such a defect requires pre-adjustment of process parameters.

[0132] The overall adjustment process is as follows Figure 2 As shown.

[0133] Step 3: Construct a hybrid model of prophet-LSTM

[0134] 3.1 Select several defect value data points from before the current injection molding cycle (assuming it is the i-th product cycle).

[0135]

[0136] 3.2 Preprocess the data. For Y... (1) Y (2) …Y (j) The data is normalized using standardization methods.

[0137] 3.3 Perform model decomposition and set model parameters. The overall step size is set to 1, and the LSTM model selects a 4-neuron structure for the injection-molded product in the current cycle, i.e., {y}. 1,i (j) ,y 2,i (j) …y s,i(j) Make predictions.

[0138] 3.4 Input data is fed into the model for prediction training. The Prophet model algorithm uses a loss function... The training weights are set with the minimum value as the objective, and then the prediction of the injection-molded product in the current cycle is performed. The result is denoted as P(i):{P i,1 (j) ,P i,2 (j) …P i,s (j) After optimization using the adaptive estimation (Adam) algorithm, the LSTM model predicts the current product in the cycle, and the result is denoted as L(i):{L i,1 (j) ,L i,2 (j) …L i,s (j)}

[0139] 3.5 Combine the results of the Prophet model and the LSTM model. According to the formula... The weighted comprehensive prediction results are calculated, and under the constraint of w1+w2=1, w1 or w2 is taken from 0 to 1 in a step size of 0.1, and the values ​​of w1 and w2 are determined with the goal of minimizing the sum of squared predicted residuals.

[0140] Step 4: Acquisition and Normalization of Key State Data

[0141] An injection molding machine was running at the same time for three consecutive days to produce the same product. The defect detection data of the under-injection products of the same cycle number produced on the three representative days were selected as key status data for analysis. The detection data for each day were obtained by visual inspection, which showed 8 data values ​​of injection under-injection defects. The detection cycle was equal to the molding cycle of the injection molding machine, and the data was normalized.

[0142]

[0143] During the production of the same product on the same injection molding machine, the initial process parameters at startup each day are basically the same: melt temperature (220℃), mold temperature (50℃), injection rate (26cm^3 / s), holding time (0.9s), holding pressure (50MPa), and cooling time (26s). Therefore, after the injection molding machine reaches its initial steady state, visual inspection of the product quality is performed, and defect value data is collected. Representative data are selected to illustrate data patterns, such as... Figure 3As shown, when the same injection molding machine produces the same product, the quality of the injection molded products changes in a certain pattern after the machine is started and stabilized each day. The data analysis in the figure shows that the defect value of the product fluctuates more in the initial stage. As the mold temperature, barrel temperature and other factors reach a certain equilibrium, the fluctuation of product defects becomes smaller. At the same time, the pattern of defect change is somewhat similar between different days.

[0144] Step 5: Adjust and predict parameters

[0145] 5.1 The trained Prophet-LSTM hybrid model is used to predict the defect value of the product in the current injection molding cycle. The result is denoted as...

[0146] 5.2 pairs By performing a fitting test, the degree of variation of each defect can be obtained.

[0147] 5.3 Determine whether the predicted defect value should trigger process parameter adjustment.

[0148] 5.3.1 If the predicted value of a defect does not exceed the conservative threshold, it is directly determined that no adjustment of injection molding process parameters will be made;

[0149] 5.3.2 When the predicted defect value is greater than the conservative threshold but less than the acceptable threshold, the defect variability is judged. If the variability is less than the variability threshold, no adjustment is made to the injection molding process parameters; otherwise, the process parameters are fine-tuned.

[0150] 5.3.3 When the predicted defect value is greater than the acceptable threshold, skip the change judgment condition and adjust the injection molding process parameters.

[0151] through Figure 4 The process predicts defects, yielding predicted values ​​Y(1,9) = 0.299, Y(2,9) = 0.324, and Y(3,9) = 0.358. Using the data sequence from the third day in the table as an example, the degree of change is calculated through least squares and cubic polynomial fitting. The final results are as follows... Figure 5 As shown.

[0152] From the fitted curve of the second day's data, a3 = -5.741E-4, we can obtain:

[0153]

[0154] Given a conservative threshold tv = 0.3, a pass threshold Tv = 0.35, and a variability threshold of:

[0155]

[0156] It is obtained according to the discrimination condition that tv: 0.3 < Y(2,9) = 0.324 < Tv: 0.35, and further judgment of the degree of change is required. It is obtained that Therefore, there is no need to adjust the process parameters.

[0157] Similarly, the result can be obtained:

[0158]

[0159] On the third day, the process parameters need to be adjusted. The defect normalization value after conservative correction is 0.325. That is, taking 0.325 as the defect value, the process parameters are slightly adjusted in advance.

[0160] Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Those with ordinary knowledge in the technical field to which the present invention pertains can make various modifications and refinements without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be determined by the claims.

Claims

1. A method for progressive dynamic adjustment of injection molding process parameters based on product quality prediction, characterized in that: Step 1: Define the defect function based on prophet-LSTM; Step 2: Construct and train a hybrid model based on prophet-LSTM; Step 3: Set correction factors and revise the model based on the correction factors; Step 4: Predict the injection molding process parameters based on the Prophet-LSTM model, and adjust the injection molding process parameters according to the defect variation judgment conditions; The definition of the prophet-LSTM defect function in step 1 is as follows: ； The construction of the prophet-LSTM hybrid model in step 2 includes the following steps: 2.1) Establish a system based on the number of days Data sets in units { }, For the first Time series of injection molding data obtained by Tianjian Testing; from{ Data from the same time period were taken from the LSTM network to analyze the changing patterns of the periodic part and obtain the trend part of the prophet model. 2.3) For The data is normalized. 2.4) Input the normalized data from step 2.3) into the prophet-LSTM model for prediction training: loss function The prophet model is trained with weights that minimize the minimum value, where... For predicted values, These are the actual detected values; then the prophet model is used to... Day The defect values ​​of injection molded products in each cycle are predicted, and the results are denoted as follows: , This indicates that the prophet model is for the th Predicted values ​​for each defect; After optimization using the adaptive estimation (Adam) algorithm, the LSTM model... Day The defect values ​​of injection molded products in each cycle are predicted, and the results are denoted as follows: , Indicates the LSTM model for the first... Predicted values ​​for each defect; 2.5) Combine the prediction results of the Prophet model and the LSTM model, specifically by... The weighted composite prediction results are as follows: For the overall prediction results, The prediction results of the Prophet model, The prediction results of the LSTM model; exist Under the constraints, or The values ​​are selected from 0 to 1 in steps of 0.1, with the objective of minimizing the sum of squared predicted residuals. , The value of is obtained, thus completing the initial construction of the prophet-LSTM hybrid model.

2. The method for progressive dynamic adjustment of injection molding process parameters based on product quality prediction according to claim 1, characterized in that, The injection molding data time series in step 2.1): a) The injection molding data time series includes time, process parameter detection values, and defect detection values, which are obtained through real-time monitoring of the injection molding process of the injection molding machine; b) Injection molding data time series includes One main type of defect and N injection molding cycles: Based on injection molding experience and customer requirements, determine The main types of defects are the subjects of study; The number of injection cycles N is determined based on the injection molding process. .

3. The method for progressive dynamic adjustment of injection molding process parameters based on product quality prediction according to claim 1, characterized in that, Step 3 specifically involves: 3.1) Define the corrective factor: in, The number of corrections is n; the initial value of n is 0. This is the initial correction factor; 3.2) Correct the model in step 2) based on the correction factor: in, This is a user-defined error value. 3.3) Repeat step 3.2) until the average error is within the range (- %, Output the corrected model between % and %.

4. The method for progressive dynamic adjustment of injection molding process parameters based on product quality prediction according to claim 3, characterized in that, The first step of step 3.2) The correction factor in the sub-correction was improved through the following analysis: a) When the predicted value is greater than the actual detected value, and the average difference exceeds... If %, then n = n + 1; b) When the predicted value is less than the actual detected value, and the average difference exceeds %, then n = n-1; c) In all other cases, n remains unchanged.

5. The method for progressive dynamic adjustment of injection molding process parameters based on product quality prediction according to claim 1, characterized in that, Step 4 specifically involves: 4.1) Use the Prophet-LSTM hybrid model obtained in step 3 to perform the current injection cycle, i.e., the [number]th [cycle]. The first day Defect values ​​for products within a given period are predicted, and the results are denoted as follows: ; in, For the first Heavenly The predicted time series of the m-th defect in each period contains k consecutive predicted values; Each defect in each injection molding cycle corresponds to k consecutive defect prediction values, specifically: { }, For the m-th type of defect, the first... Each cycle Defect prediction value at any given time; ,and ; ,and ; 4.2) Based on the defect prediction value Calculate the variation value of each defect. : In the formula, For the first Each cycle The defect change value at any given time; For the m-th defect in the first place Each cycle The defect change value at any given time; 4.3) Determine whether to initiate process parameter adjustment based on the predicted defect value; a) When the predicted defect value to be adjusted does not exceed the conservative threshold, i.e. It is directly determined that no process parameter adjustment is required; in, For setting Defect conservatization threshold; b) When the predicted defect value to be adjusted is greater than the conservative threshold and less than or equal to the acceptable threshold, i.e. To determine the degree of change in process parameters; in, For setting A set of acceptable thresholds for injection molding defects; When the degree of defect variation is less than the degree of variation threshold, i.e. No adjustment of injection molding process parameters is required; conversely, according to... Adjust the process parameters; In the formula, It is a set of thresholds for the degree of change; The results were obtained through predictions made using the prophet-LSTM model corrected in step 3. Defect prediction value at any given time; for The defect change value at any given time; It is a correction factor, which is greater than or equal to the degree of defect variation; c) When the predicted defect value to be adjusted is greater than the qualified threshold, skip the change judgment condition and restore all process parameters to their initial values.

6. The method for progressive dynamic adjustment of injection molding process parameters based on product quality prediction according to claim 5, characterized in that, Step 4.2) specifically refers to: 4.2.1) A combination of least squares algorithm and cubic fitting polynomial was used to fit the injection molding data time series; 4.2.2) For the linear terms in the polynomial of step 4.2.1) and quadratic terms Coefficient adjustment; polynomial terms Adjusted to: in, For step 2.1), the cubic term is the polynomial; S represents the error limit. After adjustment, the first derivative of the defect detection value versus time relationship in the injection molding data time series is calculated using the difference theorem. The derivative of the defect detection value at the last moment in the injection molding data time series is then used as the defect variation value for the next injection molding data time series. The defect variation value at time t is: In the formula, express Defect detection value at any given time. The time interval for data collection. yes Defect prediction value at any given time; ,and ;when When the defect value at a given time is only the defect prediction value and there is no actual recorded defect detection value, The formula for calculating the defect variation value at time t is: In the formula, yes Defect prediction value at any given time; yes Defect prediction value at any given time.

Images