TWI874003B - Forming parameter decision-making model construction method and forming system with forming parameter decision-making model - Google Patents

Abstract

The invention discloses a method for building a forming parameter decision-making model. The steps include providing a plurality of data sets; establishing a regression model and an objective function based on the plurality of data sets; obtaining the sampling distribution of model parameter estimators of the regression model; and obtaining the Bayesian estimates of the model parameter. The invention also discloses a forming system with a forming parameter decision-making model.

Description

Forming parameter decision model construction method and forming system with forming parameter decision model

The present invention relates to a forming technology, in particular to a method for constructing a decision model that can assist in predicting forming parameters and a forming system having a forming parameter decision model.

Automated production technology has been widely used in the manufacturing industry. In recent years, with the vigorous development of machine learning technology, the intelligentization of production tools can further reduce the technical threshold of production operations and increase product yield. Many machine tool or forming machine manufacturers have invested resources in the hope of introducing artificial intelligence technology into machines or equipment to enhance the competitiveness of their products.

Chinese patent number CN 114741964 A discloses a method for inverse analysis of material parameters for sheet metal stamping. The method is based on dimensionality reduction analysis. According to the existing FLD forming images, the information generated by PCA dimensionality reduction is used to construct a simplified feature model suitable for analysis, and then a substitution model is established between the design parameters and the feature space to construct the positive response model required by the approximate Bayesian method. At the same time, the dimensionality reduction feature can be used as a summary statistic of the approximate Bayesian method, so that the posterior interval distribution of the design parameters is obtained by using the approximate Bayesian calculation and the adaptive nested point distribution method, so as to achieve the inverse analysis of the material parameters, thereby more comprehensively reflecting the forming quality of the parts.

Chinese patent number CN 116243667A1 discloses a digital production line product quality monitoring method and system, including collecting equipment data and production process data, and building a database; based on the data collection results, obtaining the historical data of key quality characteristics of key processes, conducting experimental design for the key quality characteristics of key processes, and establishing a transfer function; real-time monitoring of the feedback results of the transfer function, judging whether the analysis results of the transfer function meet the quality management requirements, if so, switching to reverse control based on the transfer function, if not, isolating the product, and if there is a risk, issuing a risk warning; based on the judgment results, displaying the prediction results of product quality. This invention can predict the source of quality problems in advance and improve the product quality monitoring effect through key processes and key quality characteristics; and improve the accuracy of product quality monitoring through real-time monitoring of the feedback results of the transfer function.

WIPO patent number WO 2022/051794 A1 discloses a method for optimizing the manufacture of a product consisting of two or more inputs, each input having a cost relative to the other inputs, and the product having one or more characteristics having an optimal value range. The method involves manufacturing a sample of the product, using corresponding input values of the two or more inputs. If the value of one of the characteristics of the product is not within its optimal range, a Bayesian optimization process that takes into account the input costs is performed to calculate an adjusted input value, taking into account the relative cost of the input. The process is then iterated so that the impact of the relative cost of the input on the calculation decreases in each iteration until the value of one or more characteristics is within its optimal range.

The above-mentioned prior art has proposed a solution for quality judgment in the product manufacturing process, which can further improve product quality. However, the method adopted by the above-mentioned prior art when predicting or estimating process parameters is relatively complex. For most punching and forging processes, the blank is usually formed in one pass. How to build a parameter prediction model and forming system in a simple way to best predict or estimate the forming parameters at the initial stage of feeding (i.e. before the blank is formed) has become one of the issues that people in this technical field are eager to solve.

The purpose of the present invention is to provide a forming parameter decision model construction method and a forming system with a forming parameter decision model. The construction method is relatively simple and can quickly obtain the optimal forming parameters before the blank is formed, thereby reducing the technical threshold of forming.

According to the above-mentioned purpose, the present invention provides a method for building a forming parameter decision model, the steps of which include: providing a plurality of data sets, each of which includes a forming result and a plurality of forming parameter data; establishing a regression model and a target function based on the plurality of data sets, the target function being the prediction equation of the regression model for a forming target; obtaining the sampling distribution of the estimated quantities of the plurality of model parameters of the regression model; and using the sampling distribution as the prior probability distribution of the model parameters, the forming target obeys the normal distribution, the expected value is the target function, the Bayesian estimation of the model parameters is the expected value of the post-probability distribution of the model parameters, and thus obtaining a forming parameter decision model.

According to the above-mentioned purpose, the present invention provides a forming system with a forming parameter decision model, comprising: a monitoring module for monitoring a feature of a blank before forming; a forming machine for forming the blank through at least one forming parameter; and a control platform electrically connecting the forming machine and the monitoring module, which comprises: a forming parameter decision model established by the method of claim 1; and a calculation module, according to an expected forming result and the data of the feature, estimating an optimal forming parameter through the forming parameter decision model, and transmitting the optimal forming parameter to the forming machine.

The forming parameter decision model establishment method of the present invention provides a relatively small number of data sets, reducing the time and cost spent in the early stage of establishing the forming parameter decision model. The forming system with the forming parameter decision model of the present invention can back-calculate an optimal forming parameter through the forming parameter decision model when an expected forming result is given, which helps to quickly obtain appropriate forming parameters in the trial stage, which can greatly shorten the trial time and cost, and reduce the technical threshold of the trial process and production.

1: Forming system with forming parameter decision model

11: Forming machine

12: Monitoring module

13: Control platform

131: Forming parameter decision model

132: Computation module

133: Human-machine interface

2: Blank material

S1~S4: Steps

Figure 1 is a flow chart of the steps of an embodiment of a forming parameter decision model construction method.

FIG2 is a block diagram of an embodiment of a forming system with a forming parameter decision model.

In order to make the above or other purposes, features and characteristics of the present invention more clearly understandable, the relevant embodiments of the present invention are described in detail with the help of drawings. The drawings are mainly simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner. Therefore, only the components related to the present invention are marked in the drawings, and the components shown are not drawn in the number, size ratio, etc. during implementation, and the layout of the components may be more complicated.

FIG1 is a flow chart of the steps of an embodiment of a forming parameter decision model construction method. Referring to FIG1, the first embodiment of the present invention discloses a forming parameter decision model construction method, the steps of which include: providing a plurality of data sets (S1); establishing a regression model and a target function based on the plurality of data sets (S2); obtaining a sampling distribution of the model parameter estimation quantity of the regression model (S3); and obtaining the Bayesian estimation of the model parameter (S4).

In the step of providing a plurality of data sets (S1), a plurality of data sets related to forming are provided, each of which includes a forming result y and data of a plurality _of forming parameters xi . In forming practice, there are many parameters related to forming, and the selection of forming parameters xi _is preferably one that will affect or contribute to the forming result y of the blank. In other words, it is advisable to select forming parameters xi _that have a higher forming correlation with the forming result y . The data of the forming result y and the plurality of forming parameters xi are obtained by performing forming tests on a plurality _of test pieces. Before the experiment, each test piece has a feature F, and the feature F of the plurality of test pieces has the same value. In one example, the feature F is a thickness of the test piece, and the thickness of the plurality of test pieces is the same as the thickness of the blank that the user wants to form. In one implementation, the number of data sets related to forming is 30.

Taking the forging forming of a metal blank 2 as an example, the forming result y can be a flatness of the blank 2 after forming, and the multiple forming parameters xi can be _selected as the forging pressure, forging speed and holding time of the forming blank 2. Through experiments or empirical judgment, the forging pressure, forging speed and holding time may indeed affect the flatness of the blank 2 after forming. _The data of the forming result y and the multiple forming parameters xi are obtained by forming tests on multiple metal specimens. In this example, the feature F of the metal specimen is the plate thickness before forming. The plate thickness of the multiple metal specimens is 1.8 mm. Some data obtained from the forming test are shown in Table 1.

，其中，X代表所有自變數的總集合。目標函數E(Y|X)包括至少一自變數X _i 和至少一對應的係數

。自變數X _i 是選自複數成形參數x _i ，惟自變數X _i 的項數不一定要等於成形參數x _i 的數目，在具有合理的解釋力下，自變數X _i 的項數可以少於成形參數x _i 的數目，以兼顧運算成本。以前述的鍛壓成形為例，一目標函數E(Y|X)的自變數X _i 可選擇為鍛壓量和鍛壓速度二項，但不以此為限。 In the step of establishing a regression model and a target function according to the plurality of data sets (S2), a regression model _MR and a target function E ( Y | X ₎ are established for the forming results y of the plurality of data sets against the plurality of forming parameters xi . The target function E ( Y | X ) is a prediction equation of the regression model _MR for a forming target Y.

, where X represents the total set of all independent variables. The objective function E ( Y | X ) includes at least one independent _variable Xi and at least one corresponding coefficient

. The independent variable Xi is selected from _the multiple forming parameters Xi , but the number of independent variables Xi _is not necessarily equal to the number of forming parameters Xi _. Under reasonable explanatory power, the number of independent variables Xi _{can be} less than the number of forming parameters Xi _to take into account the computational cost. Taking the forging forming as an example, the independent variable Xi of a target function E ( Y | X ) can be selected as the forging _pressure and the forging speed, but it is not limited to this.

及變異數

估計量的抽樣分配，並以此分配做為後續貝氏分析中模型參數β _i 與σ²之事前機率分配，分別給定為π _i (．)和π_σ(．)。 In the sampling allocation step (S3) of obtaining the estimated amount of model parameters of the regression model, the statistical principle of the regression model can be used to obtain the model parameters through the above-mentioned multiple data sets, and the coefficients

and variance

The sampling distribution of the estimated quantity is used as the ex ante probability distribution of the model parameters β _i and σ ² in the subsequent Bayesian analysis, which are given as π _i (.) and π _σ (.).

模型參數之聯合概似函數L(θ| y,x )×模型參數之聯合事前機率分配π(θ)。在π(θ)具共軛性的假設下，模型參數之事後機率分配π(θ| y ,x )會具有明確的統計分配，故模型參數之貝氏估計將為事後機率分配之期望值E(π(θ| y ,x ))，進而獲得一成形參數決策模型。 In the Bayesian estimation step (S4) of the model parameters, the prior probability distribution of all model parameters is obtained as π _i (．) and π _σ (．) through the statistical method of step (S3). Assuming that the shaping target Y follows a normal distribution with an expected value as the target function E ( Y | X ) and a variance of σ ² , that is, Y ~ N ( E ( Y | X ) , σ ² ), according to the Bayesian theorem, the subsequent probability distribution of the model parameters will be π( θ | y , x ) and satisfy π( θ | y , x )= L ( θ | y , x )π( θ )/[ʃ L ( θ | y , x )π( θ ) d θ ], where L ( θ | y , x ) is the probability distribution of the model parameters after the fact. ) is the joint likelihood function of the model parameters, and π( θ ) is the joint ex ante probability distribution of the model parameters. In other words, the ex post probability distribution of the model parameters π( θ | y , x )

The joint likelihood function of model parameters L ( θ | y , x )×the joint ex ante probability distribution of model parameters π( θ ). Under the assumption that π( θ ) is conjugated, the ex post probability distribution of model parameters π( θ | y , x ) will have a clear statistical distribution, so the Bayesian estimate of the model parameters will be the expected value of the ex post probability distribution E (π( θ | y , x )), thus obtaining a shaped parameter decision model.

According to the aforementioned steps (S1) to (S4), the forming parameter decision model of the blank corresponding to the feature F value f1 (such as the thickness of 1.8mm) can be obtained. Similarly, the forming parameter decision model of the blank corresponding to the feature F value f2 (such as the thickness of 2.0mm) can also be obtained according to the aforementioned steps.

。這將有助於在試作階段快速取得合適的成形參數，可大幅縮短試作的時間和成本、降低試作程序和生產的技術門檻。 Through the aforementioned forming parameter decision model, given an expected forming result y ^* , the optimal forming parameter can be deduced.

This will help to quickly obtain appropriate forming parameters during the trial stage, which can significantly shorten the trial time and cost, and lower the technical threshold of the trial process and production.

FIG2 is a block diagram of an embodiment of a forming system with a forming parameter decision model. Referring to FIG2, the second embodiment of the present invention discloses a forming system 1 with a forming parameter decision model, which includes a forming machine 11, a monitoring module 12 and a control platform 13.

The forming machine 11 may be a punching machine or a forging machine. Before forming a blank 2, at least one forming _parameter xi must be set for the forming machine 11. The forming machine 11 forms the blank 2 with _the forming parameter _xi . The forming parameter xi is, for example, the number of strokes per unit time, forging pressure, holding time, forming speed, and forming temperature.

The monitoring module 12 can monitor or measure online at least one characteristic F of the blank 2 before forming, such as online measurement of the thickness, diameter, weight and/or temperature of the blank 2.

。成形參數決策模型131的建置方法已於第一實施例說明，於此不再重複。人機界面133用以輸入和輸出資料，並輔助使用者操作具成形參數決策模型的成形系統1。運算模組132接收來自監測模組12所測得的特徵 F的數據，和使用者經由人機界面133設定一預期成形結果y ^*後，透過成形參數決策模型131以估算出一最適成形參數

，再將最適成形參數

傳送至成形機11以成形胚料2。 The control platform 13 is electrically connected to the monitoring module 12 and the forming machine 11 to receive and transmit data. The control platform 13 includes a forming parameter decision model 131, a calculation module 132 and a human-machine interface 133. The forming parameter decision model 131 is a probability model of forming parameters, by which the optimal forming parameters that meet a forming result y can be estimated.

The method for constructing the forming parameter decision model 131 has been described in the first embodiment and will not be repeated here. The human-machine interface 133 is used to input and output data and assist the user in operating the forming system 1 with the forming parameter decision model. The calculation module 132 receives the data of the feature F measured by the monitoring module 12, and after the user sets an expected forming result y ^* through the human-machine interface 133, the forming parameter decision model 131 is used to estimate an optimal forming parameter.

, and then the optimal forming parameters

The blank 2 is then transferred to the forming machine 11 to be formed.

，再將最適成形參數

傳送至成形機11對胚料2成形，即可成形出滿足預期成形結果y ^*的成品。 When using the forming system 1 with the forming parameter decision model, the user can first set an expected forming result y ^* through the human-machine interface 133, and the monitoring module 12 measures at least one feature F of the blank 2 before feeding. After the calculation module 132 receives the data of the feature F of the blank 2, it first determines whether the feature F of the blank 2 is qualified. If it is unqualified, the blank 2 can be repaired and adjusted and then return to the previous step, or the blank 2 can be discarded; if it is qualified, after the calculation module 132 receives the data of the feature F and the expected forming result y ^* , an optimal forming parameter can be estimated through the forming parameter decision model 131.

, and then the optimal forming parameters

The blank 2 is sent to the forming machine 11 to be formed into a finished product that meets the expected forming result y ^* .

The forming parameter decision model establishment method of the present invention provides a relatively small number of data sets, which reduces the time and cost spent in the early stage of establishing the forming parameter decision model. The forming system with a forming parameter decision model of the present invention can back-calculate the optimal forming parameters given an expected forming result through the forming parameter decision model, which helps to quickly obtain appropriate forming parameters in the trial stage, which can greatly shorten the trial time and cost and reduce the technical threshold of the trial process.

The above only records the preferred implementation methods or examples of the technical means adopted by the present invention to solve the problem, and is not intended to limit the scope of implementation of the present invention. That is, all equivalent changes and modifications that are consistent with the scope of the present invention's patent application or made in accordance with the scope of the present invention's patent are covered by the scope of the present invention's patent.

S1~S4: Steps

Claims (10)

A forming parameter decision model construction method includes the following steps: providing a plurality of data sets, each of which includes a forming result and a plurality of forming parameter data; establishing a regression model and a target function based on the plurality of data sets, the target function being the prediction equation of the regression model for a forming target; obtaining the sampling distribution of the estimated quantities of the plurality of model parameters of the regression model; and using the sampling distribution as the prior probability distribution of the model parameters, the forming target obeys a normal distribution, the expected value is the target function, and the Bayesian estimation of the model parameters is the expected value of the post-probability distribution of the model parameters, thereby obtaining a forming parameter decision model. A forming parameter decision model building method as described in claim 1, wherein the forming result and the data of the plurality of forming parameters are obtained by performing forming tests on a plurality of test pieces. A forming parameter decision model building method as described in claim 1, wherein the objective function includes an independent variable and a coefficient, and the independent variable is selected from the plurality of forming parameters. A forming parameter decision model construction method as described in claim 3, wherein the number of terms of the independent variable is plural, and the number of terms of the independent variable is less than the number of the forming parameters. A forming parameter decision model building method as described in claim 4, wherein the forming result is a flatness, the multiple forming parameters include forging pressure, forging speed and holding time, and the independent variable includes forging pressure and forging speed. As described in claim 1, in the step of obtaining the sampling distribution of the multiple model parameter estimates of the regression model, the sampling distribution of the coefficient and variance estimates of the regression model is obtained by using the multiple data sets. A forming system with a forming parameter decision model comprises: a monitoring module for monitoring a feature of a blank before forming; a forming machine for forming the blank through at least one forming parameter; and a control platform electrically connected to the forming machine and the monitoring module, which comprises: a forming parameter decision model established by the method of claim 1; and a calculation module, which estimates an optimal forming parameter through the forming parameter decision model according to an expected forming result and the data of the feature, and transmits the optimal forming parameter to the forming machine. A forming system with a forming parameter decision model as described in claim 7, wherein the forming machine is a punching machine or a forging machine. A forming system with a forming parameter decision model as described in claim 7, wherein the characteristic is selected from the thickness, diameter, weight and temperature of the blank. A forming system with a forming parameter decision model as described in claim 7, wherein the control platform further includes a human-machine interface for setting the expected forming result.