Device and method for supporting optimal control of injection molding machine

Abstract

The present invention relates to a device and method for supporting optimal control of an injection molding machine to suppress the generation of hazardous gases in the injection molding machine. More specifically, by applying the Taguchi Method, gas collection data is used as a characteristic value, a signal-to-noise ratio (SNR) of the characteristic value is obtained according to a feature vector of control parameters, and a model for predicting the SNR using the feature vector is prepared. Among a plurality of control parameter data to be used in the process, optimal control parameter data is recommended according to the SNR predicted by the model, and is then applied to the process performance of the injection molding machine, thereby minimizing variations in gas generation and reducing the total amount of gas produced, such that the generated gas can be stably purified before discharge. This technology was developed as part of the Regional Intelligence Innovation Talent Training Program, supported by the Institute for Information & Communication Technology Planning & Evaluation and with funding from the Ministry of Science and ICT, Republic of Korea (RS-2024-00436765).

Description

Injection molding machine optimal control support device and support method

[0001] The present invention relates to an injection molding machine optimal control support device and support method for suppressing the generation of harmful gases in an injection molding machine.

[0002] A plastic injection molding machine is a device that performs an injection molding process including processes such as mold closing, filling, packing, cooling, and mold opening. In the filling process, the solid plastic raw material is melted and compressed into a cylinder and injected into the cavity of the mold through a nozzle. In the packing process, a constant pressure is maintained, and the molding conditions are determined according to the plastic raw material, the shape and size of the molded product, the structure of the nozzle and cylinder, etc., and the control parameters for obtaining the molding conditions are set so that the machine operates according to the values ​​of the set control parameters.

[0003] However, during the process of melting plastic raw materials inside the cylinder, harmful gases are generated and released to the outside, which not only worsens the working environment but also lowers the quality of the molded product. Therefore, a gas ejector is installed between the cylinder or the cylinder and the nozzle, and a gas collector / purifier is operated together to suck and purify the gas through the gas ejector.

[0004] However, the type and amount of gas generated vary depending on the raw materials and molding conditions, so there are cases where it cannot be processed by a gas collector / purifier. In addition, the replacement cycle of the filter used in gas purification also affects it, making it difficult to replace it at an appropriate time.

[0005] In the meantime, the temperature of the cylinder, which has the greatest influence on gas generation, was set to the temperature with the lowest gas generation intensity based on empirical data and operated, but since the type and amount of gas generated varied depending on various control parameters other than the cylinder temperature, there were limitations in suppressing gas generation.

[0006] Moreover, even if the injection molding machine is controlled according to the values ​​of the same control parameters, the actual molding conditions vary depending on various factors such as the ambient temperature and humidity, and the fluctuating flow rate and speed of the molten resin, and the amount of gas generated also varies accordingly. Therefore, it was not possible to provide a clear optimal control parameter value for this.

[0007] Meanwhile, although there is a technology to optimize molding conditions to minimize the defect rate of molded products, it does not provide optimal control parameter values ​​to suppress gas generation.

[0008] [Prior Art Literature]

[0009] [Patent Document]

[0010] (Patent Document 1) KR 10-2298636 B1 2021.08.31.

[0011] (Patent Document 2) KR 10-2311048 B1 2021.10.01.

[0012] Accordingly, the purpose of the present invention is to provide an optimal control support device and support method for an injection molding machine, which derives control parameters capable of suppressing gas generation and applies them to the process of an injection molding machine, and operates the machine to suppress gas generation while coping with fluctuations in the amount of gas generated during an actual process.

[0013] In order to achieve the above object, the present invention provides an injection molding machine (100) that melts a plastic raw material and injects it into a mold (1), captures, senses, purifies, and discharges gas generated from the molten resin to be injected, and performs a process according to control parameter data set by a controller (160) to match molding conditions, and an optimal control support device for suppressing gas emission, comprising: a data processing unit (12) that, when receiving a plurality of control parameter data applicable to the process, maps each control parameter data to a feature space to generate a plurality of feature vectors; an inference unit (14) that prepares a prediction model that predicts a signal-to-noise ratio (SNR) of a characteristic value used as gas capture data as a feature vector in order to use the Taguchi Method, and derives a signal-to-noise ratio according to each feature vector as a prediction model; It includes an application unit (15) that analyzes multiple predicted signal-to-noise ratios according to the Taguchi Method, determines control parameter data corresponding to one signal-to-noise ratio as an optimal control parameter, and applies the optimal control parameter data to process processing.

[0014] As an embodiment, the injection molding machine optimal control support device according to the present invention further includes a learning unit (13) that, when receiving a plurality of control parameter data applied to process execution and gas capture data measured while performing a plurality of processes with each control parameter data, generates learning data with a plurality of feature vectors generated for the plurality of control parameter data and the gas capture data through the data processing unit (12) and trains a prediction model that estimates a signal-to-noise ratio with the feature vector.

[0015] In order to achieve the above object, the present invention provides an injection molding machine (100) that melts a plastic raw material and injects it into a mold (1), captures and senses gas generated from the molten resin to be injected, purifies it, and then discharges it, and performs a process according to control parameter data set to match molding conditions by a controller (160), and an injection molding machine optimal control support device comprising a data processing unit (12), an inference unit (14), and an application unit (15) to suppress gas discharge, the method comprising: a data processing step (S30) in which, when receiving a plurality of control parameter data applicable to the process, the data processing unit (12) maps each control parameter data to a feature space to generate a plurality of feature vectors; The present invention comprises an inference step (S40) in which an inference unit (14) is prepared with a prediction model that predicts the signal-to-noise ratio (SNR) of a characteristic value used as gas capture data for using the Taguchi Method as a feature vector, and a prediction model that predicts the signal-to-noise ratio according to each feature vector; an application step (S50) in which an application unit (15) analyzes a plurality of signal-to-noise ratios predicted according to the Taguchi Method to determine control parameter data corresponding to one signal-to-noise ratio as an optimal control parameter, and applies the optimal control parameter data to process processing.

[0016] As an embodiment, the method for supporting optimal control of an injection molding machine according to the present invention further includes a learning unit (13) that receives a plurality of control parameter data applied to process execution and a plurality of gas capture data measured while performing a plurality of processes with each of the control parameter data, and receives a plurality of feature vectors generated for the plurality of control parameter data through a data processing unit (12), and a learning data generation step (S10) in which the learning unit (13) generates learning data with the plurality of feature vectors and the gas capture data; and a learning step (S20) in which a prediction model that estimates a signal-to-noise ratio with the feature vector is trained with the learning data.

[0017] The present invention analyzes the characteristics according to the control parameters by applying the Taguchi Method, defines gas capture data as the characteristics, and uses a prediction model that predicts the characteristics using the feature vector of the control parameters, thereby recommending optimal data that can effectively suppress gas generation for various control parameters to be used in an injection molding machine, thereby enabling the process to be performed, thereby minimizing fluctuations in the amount of gas generated and reducing the amount of gas generated as much as possible, thereby stably purifying the generated gas and then discharging it.

[0018] Figure 1 is a configuration diagram of an injection molding machine (100) connected to an injection molding machine optimal control support device (10) according to an embodiment of the present invention.

[0019] Figure 2 is a block diagram of an injection molding machine optimal control support device (10) according to an embodiment of the present invention.

[0020] Figure 3 is a configuration diagram of an injection molding machine (100) in which an injection molding machine optimal control support device (10) according to a modified embodiment of the present invention is implemented.

[0021] Figure 4 is a flowchart of a method for supporting optimal control of an injection molding machine according to an embodiment of the present invention.

[0022] Hereinafter, specific examples of embodiments of the present invention will be described with reference to the attached drawings so that those skilled in the art can easily implement the present invention. However, it is clear that the embodiments of the present invention can be implemented through various changes or modifications within the scope of the present invention, and thus, the present invention is not limited to the described embodiments.

[0023] Embodiments of the present invention will not be described in detail, as those skilled in the art can readily add and implement well-known components, circuits, functions, methods, and typical details. Furthermore, the injection molding machine that operates in conjunction with the embodiments of the present invention will be described briefly enough to facilitate a thorough understanding of the embodiments of the present invention.

[0024] Embodiments of the present invention may be implemented in a form of a combination of software and hardware, and the software and hardware forms may be described as parts, modules, or sub-parts, and may be implemented in the form of computer-readable program code implemented on a recording medium.

[0025] The connection between components may be a wired connection or a wireless connection to transmit data, signals, information, etc., and may also include indirect connections with other components in between.

[0026] To 'include' an element means to include, but not to exclude, other elements unless otherwise specifically stated.

[0027] Referring to FIG. 1, an injection molding machine optimal control support device (10) according to an embodiment of the present invention is configured to exchange data with a controller (160) of an injection molding machine (100) by wire or wirelessly, and as described in detail below, receives control parameter data of the injection molding machine (100) and derives optimal control parameters capable of suppressing gas generation using the Taguchi Method and transmits them to the injection molding machine (100), thereby enabling management so that only an appropriate level of gas that can be processed is generated in the injection molding machine (100).

[0028] First, in order to understand the description of an embodiment of the present invention, the hopper (110), cylinder (120), nozzle (130), gas discharger (140), gas capture / purifier (150), and controller (160) constituting the injection molding machine (100) will be briefly examined, and the control parameters analyzed by a statistical method in the present invention and the molding conditions related to the control parameters will be exemplified and examined, and the gas capture data will also be explained.

[0029] The hopper (110) generally supplies solid plastic raw materials manufactured in pellet form to the cylinder (120). Meanwhile, a dryer was installed in the past to dehumidify and dry the plastic raw materials, but as described below, if the present invention is optimally operated to remove gas and moisture, such a dryer can be dispensed with, thereby reducing equipment costs and power consumption.

[0030] The cylinder (120) can be moved forward and backward by a screw driver (122) installed at the rear end, and a rotatable screw (121) is installed inside, so that the screw (121) is moved forward and backward to adjust the position, and by rotating it, the plastic raw material supplied from the hopper (110) is pressurized and transported toward the header of the front end, and a heater and a temperature sensor are installed on the outer periphery to control the temperature and heat the plastic raw material to melt it and transport it to the header of the front end.

[0031] The screw (121) is configured to have a feeding zone, a compression zone, and a metering zone, and a heater and a temperature sensor are installed in each zone to control the temperature in each zone, so that the plastic raw material fed into the feeding zone is preheated and advanced, melted and compressed in the compression zone and transferred to the metering zone, and the molten resin is supplied to the nozzle side for injection from the metering zone.

[0032] The nozzle (130) is for injecting molten resin supplied from the cylinder (120) into the cavity (Cavity, 1a) of the mold (1). A heater and a temperature sensor may also be installed in the nozzle (130) to maintain the molten resin to be injected at a predetermined temperature.

[0033] The gas discharger (140) has a passage for discharging gas and moisture generated in the molten resin before the molten resin is injected through the nozzle (130). As shown in the drawing and described in detail in, for example, Patent No. 10-2298636, it can be installed between the cylinder (120) and the nozzle (130). In this case, a heater and a temperature sensor can be installed. However, this is not limited thereto, and for example, as disclosed in Patent No. 10-2379514, it can also be installed on the metering section side of the front end of the cylinder (120).

[0034] Although not shown in the drawing, a means for moving the injection molding machine (100) toward the mold is also included.

[0035] The gas capture / purifier (150) absorbs gas and moisture through the gas discharger (140), removes moisture using a water-oil filter, measures the oxygen concentration using a gas sensor to measure the amount of harmful gases, and discharges purified gas that has been captured and removed of harmful substances such as harmful gases and tar using a gas filter. Here, the amount of absorption can be adjusted to an appropriate amount.

[0036] The controller (160) sets the values ​​of control parameters to obtain molding conditions determined based on the plastic raw material, the size and shape of the molded product, the equipment characteristics of the injection molding machine, etc., and controls the operation of each component described above to perform the injection molding process according to the values ​​of the control parameters.

[0037] Here, molding conditions include, for example, the resin temperature of each part, injection pressure, injection speed, injection amount, injection capacity, holding pressure, metering distance, holding pressure changeover distance, injection time, holding pressure time, metering time, etc. And, control parameters include the temperature of each part, screw position, rotation speed and torque, and operating time, etc.

[0038] The optimal control support device for an injection molding machine according to an embodiment of the present invention is set for each injection molding process, and controls the process using the used control parameter data and the control parameter data for each process, and receives the gas capture data measured through the gas capture / purifier (150) as data for learning purposes when processing the process. Then, the gas capture data is defined as a characteristic value of the Taguchi method, and a prediction model for predicting the signal-to-noise ratio of the characteristic value is obtained.

[0039] Here, the control parameter data is data applied when repeatedly performing the injection molding process with the same plastic raw material, but is data applied by performing the process at least twice after setting the values ​​of each item to the same value, and repeating such multiple process performances while changing the values ​​of each item. If the injection molding machine (100) is used as a device that produces products using different plastic raw materials, the control parameter data is data applied for each plastic raw material. Of course, considering that gas generation differs depending on the size and shape of the molded product, the control parameter data may be data obtained for each molded product.

[0040] Gas capture data is data obtained from repeated process execution by applying the same control parameter data, and can be the amount of gas generated measured by a gas sensor.

[0041] These control parameter data and gas capture data can be obtained as data reflecting the maximum amount of information from the minimum number of process experiments by conducting process experiments using the orthogonal array table of the Taguchi Method.

[0042] In addition, the injection molding machine optimal control support device according to an embodiment of the present invention, when receiving a plurality of control parameter data for performing an actual process, predicts the signal-to-noise ratio according to each control parameter using a prediction model, and then selects the optimal control parameter data using the signal-to-noise ratio according to the Taguchi method, thereby applying the data to performing the process in the injection molding machine (100).

[0043] For this purpose, an injection molding machine optimal control support device (10) according to an embodiment of the present invention is described in detail.

[0044] Referring to the block diagram of FIG. 2, the injection molding machine optimal control support device (10) according to an embodiment of the present invention includes an interface unit (11), a data processing unit (12), a learning unit (13), an inference unit (14), and an application unit (15).

[0045] The above interface unit (11) is connected to the controller (160) of the injection molding machine (100) by wire or wirelessly to communicate data and perform linked operations. It receives data for learning a prediction model, receives a plurality of control parameter data to be applied to actual process execution, and is used to transmit the selected optimal control parameter data.

[0046] Here, the data transmitted for learning is data obtained through a process experiment in advance as described above, and includes multiple control parameter data applied to each process and gas capture data of gas generated when performing an injection molding process with each control parameter data.

[0047] The plurality of control parameter data for actual process execution is a plurality of control parameter data that can be applied to the process currently being performed by the controller (160). Among these plurality of control parameter data, the optimal control parameter data selected as described below is transmitted to the controller (160).

[0048] When performing an actual process, the control parameter data and gas capture data to be applied can be received and the prediction model can be retrained as described below.

[0049] The above data processing unit (12) is for generating a feature vector by mapping the control parameter data transmitted through the interface unit (11) to a feature space.

[0050] Feature vectors are created by embedding control parameter data into a high-dimensional feature space using a kernel function. As is well known, among various kernel functions, a feature vector can be created using a kernel function appropriate to the data characteristics to estimate the regression coefficients of the prediction model used as the regression analysis model in the embodiments of the present invention.

[0051] When data for learning is transmitted, a control parameter set consisting of multiple control parameter data is mapped to a feature space to generate multiple feature vectors, which are then transmitted to the learning unit (13) to be used to train a prediction model.

[0052] Even when multiple control parameter data for actual process execution are transmitted, a control parameter set composed of multiple control parameter data is mapped to a feature space to generate multiple feature vectors, which are then transmitted to the inference unit (14) to be used to derive a signal-to-noise ratio as a prediction model.

[0053] In addition, when the control parameter data and gas capture data applied while performing the actual process are transmitted, a feature vector is generated for learning purposes and transmitted to the learning unit (13), thereby allowing the prediction model to be re-learned.

[0054] Control parameters may include not only contributing parameters that contribute significantly to the characteristic value (gas capture data) but also noise parameters that contribute relatively little or insignificantly, and a feature vector reflecting these characteristics can be generated here.

[0055] Meanwhile, when the data processing unit (12) receives control parameter data classified by plastic raw material, whether for learning or relearning or for actual process execution, it classifies the control parameter data by plastic raw material and generates multiple feature vectors for each plastic raw material. In addition, when receiving control parameter data by molded product produced by the injection molding machine (100), it classifies the control parameter data by molded product and generates multiple feature vectors for each molded product.

[0056] The above learning unit (13) generates learning data using a plurality of feature vectors generated as a control parameter set among the data received for learning and a plurality of gas capture data among the data received for learning, and then trains a prediction model that estimates the signal-to-noise ratio (SNR) of the characteristic value (gas capture data in the present invention) according to the Taguchi Method as a feature vector.

[0057] The Taguchi method is a method of robust design against uncontrollable noise factors using controllable factors. It defines a signal-to-noise ratio (SNR) that reflects not only the target value of quality characteristics but also the variance (or deviation) of the characteristics, and determines each level of the control factor with a large signal-to-noise ratio.

[0058] In order to use this Taguchi method, the characteristic value representing quality in the Taguchi method is defined and used as gas capture data.

[0059] Since the lower the better the characteristic used for gas capture data, the signal-to-noise ratio (SNR) can be expressed as in the following mathematical expression 1.

[0060] [Mathematical Formula 1]

[0061]

[0062] Here, i is an index that distinguishes control parameter data, j and n are indices that distinguish processes performed with the same control parameter data and the number of processes, is the characteristic value (gas capture data) measured when performing the process with the i-th control parameter data, is the signal-to-noise ratio for the characteristic value obtained by performing the process n times with the i-th control parameter data.

[0063] This signal-to-noise ratio has a larger value as the dispersion or deviation of the characteristic value is smaller, and also has a larger value as the mean of the characteristic value is smaller, and shows robustness to noise control parameters among the control parameters or noise factors other than the noise control parameters.

[0064] The signal-to-noise ratio expressed in the above mathematical expression 1 can be calculated using characteristic values ​​(gas capture data) obtained for each control parameter data, and an actual vector composed of the signal-to-noise ratio of the actual measured characteristic values ​​can be generated.

[0065] In the present invention, rather than using the conventional Taguchi method of determining the optimal control parameter value based on the signal-to-noise ratio calculated based on experimentally applied control parameter data, a model for predicting the signal-to-noise ratio is created, and the optimal control parameter data is determined by predicting the signal-to-noise ratio based on the control parameter data to be actually applied.

[0066] The prediction model for predicting the signal-to-noise ratio of these characteristics can be a regression model, and the SVR (Support Vector Regression) model representing the regression model can be used.

[0067] Since the regression model and learning method used as a prediction model are well-known technologies, only the main contents are explained here.

[0068] The SVR model used to implement the present invention is for nonlinear regression, and as is well known, is expressed by the regression function of mathematical formula 2.

[0069] [Equation 2]

[0070]

[0071] Here, When learning as a signal-to-noise ratio, the actual vector is applied, is a weight vector, b is a constant representing the bias, is a kernel function.

[0072] Then, a learning process is performed to derive a weight vector and bias value b that satisfy the objective function and constraints as in the following mathematical expression 3 using the feature vector and the actual vector of the learning data, thereby completing the regression function.

[0073] [Equation 3]

[0074]

[0075] subject to.

[0076]

[0077] Here, is a precision parameter representing the radius of the tube of the regression function, C is a normalization constant, and the precision parameter And the constant C is preset. and represents the distance that the feature vector deviates from the tube of the regression function. represents the kernel function, and is expressed in mathematical expression 2. It corresponds to . is the above-mentioned feature vector, is the value of the above-mentioned real vector, and w is a slope variable for expressing the margin.

[0078] Next, multiple feature vectors from the training data are input into the regression function to obtain a predicted vector. Accuracy is then calculated based on the difference between the predicted vector and the ground truth vector. Of course, it's also possible to divide the data into training and test sets, using multiple feature vectors for testing. Here, accuracy can be calculated as the reciprocal of the difference between the predicted vector and the ground truth vector, as shown in Equation 4 below.

[0079] [Equation 4]

[0080]

[0081] Here, Eacc represents accuracy, is the value of the actual vector, is the value of the prediction vector, and i is the index of the training data.

[0082] If the accuracy derived in this way is below a preset threshold, the learning process of re-obtaining the weight vector and bias using the learning data is repeated. If it is above the threshold, learning is completed, and the regression function obtained at this time is selected as the prediction model.

[0083] In the present invention, the learning unit (13) is characterized in that it uses gas capture data as a characteristic value, defines a signal-to-noise ratio according to the network characteristics of the characteristic value, and obtains a prediction model of the signal-to-noise ratio.

[0084] In addition, when receiving learning-purpose data (multiple control parameter data and gas capture data) classified by plastic raw material, multiple feature vectors generated by the multiple control parameter data and the received gas capture data are classified by plastic raw material to generate learning data for each plastic raw material, and a prediction model to be used for each plastic raw material is generated.

[0085] In addition, when data for learning purposes (multiple control parameter data and gas capture data) classified by molded product produced by an injection molding machine (100) is received, the generated multiple feature vectors and the received gas capture data are classified by molded product to generate learning data for each molded product, and a prediction model to be used for each molded product is generated.

[0086] In addition, when additional control parameter data and gas capture data are received according to actual process processing, the generated feature vector and the received gas capture data are added to the learning data, and when the learning data increases by a certain amount, the prediction model is retrained to update the prediction model.

[0087] In this way, the prediction model generated in the learning unit (13) is transferred to the inference unit (14).

[0088] The above inference unit (14) is ready to use the prediction model generated in the above learning unit (13), and a prediction model classified by plastic raw material can be prepared, and a prediction model classified by molded product can be prepared.

[0089] And, when a plurality of control parameter data for actual process execution is transmitted through the interface unit (11), and a plurality of feature vectors generated by mapping the received plurality of control parameter data to a feature space are transmitted from the data processing unit (12), the inference unit (14) predicts the signal-to-noise ratio according to each feature vector using a prediction model.

[0090] Here, the multiple control parameter data may be different control parameter data applicable to the process and may include data that is not identical to the data used for learning purposes. Furthermore, the prediction vector is a predicted signal-to-noise ratio value, and multiple feature vectors are sequentially selected and generated as a prediction model to derive the prediction vector.

[0091] If the feature vector is a plastic raw material to be used in the process, a prediction vector is derived using a prediction model corresponding to the plastic raw material, and if the feature vector is a molded product to be produced in the process, a prediction vector is derived using a prediction model corresponding to the molded product.

[0092] The prediction vector containing the predicted value of each signal-to-noise ratio is transmitted to the application unit (15).

[0093] Meanwhile, although the data flow is not shown in Fig. 2, the inference unit (14) receives multiple control parameter data and gas capture data used for learning purposes, analyzes the effect of each control parameter on gas generation, and can infer control parameters to be adjusted according to changes in the amount of hazardous gas generated. For example, changes in gas capture data that appear when the value of each control parameter changes can be obtained through data analysis.

[0094] The above application unit (15) compares and analyzes multiple signal-to-noise ratios included in the prediction vector to determine the control parameter data corresponding to one signal-to-noise ratio as the optimal control parameter data. At this time, by transmitting the determined optimal control parameter data to the injection molding machine (100), the injection molding machine (100) can perform the process by applying the optimal control parameters. Of course, by transmitting the identification information of the determined optimal control parameter data, it is possible to determine which of the multiple control parameter data sent from the injection molding machine (100) it is.

[0095] Here, the optimal control parameter data can be determined as the control parameter data corresponding to the signal-to-noise ratio having the largest value among multiple signal-to-noise ratios.

[0096] In addition, the application unit (15) can be applied to control the gas intake amount by determining the gas intake amount according to the size of the signal-to-noise ratio corresponding to the optimal control parameter data and transmitting it to the injection molding machine (100) together with the optimal control parameter data. That is, the control is performed based on the gas intake amount determined here, and the gas intake amount can be adjusted according to the gas generation amount. For example, while controlling to inhale gas and moisture at the gas intake amount determined here, the amount of harmful gas generated measured by the gas sensor can be monitored, and the intake amount can be adjusted according to the amount of harmful gas generated.

[0097] In addition, the application unit (15) receives control parameters to be adjusted according to changes in the amount of gas generated from the inference unit (14) and transmits them to the injection molding machine (100), thereby monitoring the amount of harmful gas generated measured by the gas sensor and controlling the amount of gas and moisture intake.

[0098] By operating the application unit (15) in this way in connection with the injection molding machine (100), the process is performed by applying an optimal control parameter that has little fluctuation in the amount of gas generated and a small amount of gas generated, and control is possible according to the amount of gas generated. Through this, the entire amount of gas generated can be captured by the gas collection / purification device (150) and stably purified, and emission of harmful gases that are not purified can be fundamentally prevented.

[0099] FIG. 3 is a configuration diagram of an injection molding machine (100) in which an injection molding machine optimal control support device (10) according to a modified embodiment of the present invention is implemented in a controller (160).

[0100] Since the controller (160) of the injection molding machine (100) is equipped with a process control processor, the optimal control support device for the injection molding machine can be implemented as a program component in the processor. In this case, the interface unit (11) described above can be omitted and implemented as a program that receives and operates data processed by the process control program.

[0101] Meanwhile, the prediction model may not be generated by the learning unit (13), but may be generated in advance and then transplanted to the inference unit (14) when configuring the optimal control support device for the injection molding machine. In this case, it is also advisable to operate the learning unit (13) for re-learning, and the learning data used to generate the prediction model may be stored and used when re-learning in the learning unit (13).

[0102] Figure 4 is a flowchart of an injection molding machine optimal control support method comprising an injection molding machine optimal control support device.

[0103] The injection molding machine optimal control support method is performed by the data processing unit (12), learning unit (13), inference unit (14) and application unit (15) of the injection molding machine optimal control support device (10) described in detail above, so overlapping detailed descriptions are omitted and the main contents of each step are mainly described.

[0104] The injection molding machine optimal control support method may include a learning data generation step (S10) and a learning step (S20) for training a prediction model, a data processing step (S30), an inference step (S40) and an application step (S50) for deriving and providing optimal control parameter data to be performed in an actual process as a prediction model, and a learning data addition step (S60) and a retraining step (S70) for retraining the prediction model.

[0105] The above learning data generation step (S10) and learning step (S20) are steps performed when data for learning purposes are transmitted from an injection molding machine (100), and are performed by a data processing unit (12) and a learning unit (13). Here, the data for learning purposes includes multiple control parameter data applied experimentally or in actual process execution as described above, and multiple gas capture data for each control parameter data.

[0106] In the above learning data generation step (S10), the data processing unit (12) sequentially performs a step of generating multiple feature vectors with multiple control parameter data and passing them to the learning unit (13), and a step of the learning unit (13) generating learning data with multiple feature vectors and multiple gas capture data.

[0107] In the above learning step (S20), a prediction model that estimates the signal-to-noise ratio using a feature vector is trained using learning data and then passed to the inference step (S40) for use.

[0108] If the data for learning purposes is classified by plastic raw material, the learning data generation step (S10) generates learning data for each plastic raw material using multiple feature vectors and multiple gas capture data generated by classifying by plastic raw material, and the learning step (S20) generates a prediction model to be used for each plastic raw material.

[0109] If the data for learning purposes is classified by molded product, learning data for each molded product is generated using multiple feature vectors and multiple gas capture data generated by classifying each molded product in the learning data generation step (S10), and a prediction model to be used for each molded product is generated in the learning step (S20).

[0110] The above data processing step (S30), inference step (S40) and application step (S50) are performed by the data processing unit (12), inference unit (14) and application unit (15) when receiving multiple control parameter data applicable to the process to recommend optimal control parameter data.

[0111] In the above data processing step (S30), the data processing unit (12) maps each control parameter data to a feature space to generate multiple feature vectors.

[0112] In the above inference step (S40), the inference unit (14) derives the signal-to-noise ratio according to each feature vector as a prediction model using the prediction model prepared by the above learning step (S20). Here, the prediction model is a model that predicts the signal-to-noise ratio of the characteristic value required for using the Taguchi Method as a feature vector, and is trained to derive the signal-to-noise ratio as a characteristic value using the gas capture data as a characteristic value and the feature vector of the control parameter data, so the signal-to-noise ratio can be predicted using the feature vector.

[0113] Here, if the prediction model is prepared for each plastic raw material and the control parameter data is classified for each plastic raw material, the data processing step (S30) generates a feature vector by classifying the control parameter data for each plastic raw material, and the inference step (S40) predicts the signal-to-noise ratio as a feature vector using the prediction model corresponding to the plastic raw material to be used in the process.

[0114] If a prediction model is prepared for each molded product and control parameter data is classified for each molded product, the data processing step (S30) generates a feature vector by classifying the control parameter data for each molded product, and the inference step (S40) uses a prediction model corresponding to the molded product to be produced in the process to predict the signal-to-noise ratio as a feature vector.

[0115] In the above application step (S50), the application unit (15) analyzes multiple signal-to-noise ratios according to the Taguchi Method, determines control parameter data corresponding to one signal-to-noise ratio as an optimal control parameter, and applies the optimal control parameter data to the process processing of the injection molding machine (100). In one embodiment, the control parameter data corresponding to the signal-to-noise ratio having the maximum value among the multiple signal-to-noise ratios may be determined as the optimal control parameter.

[0116] Meanwhile, even if the process is performed by controlling the injection molding machine (100) with optimal control parameters, the amount of gas generated fluctuates depending on various factors, so it is recommended to process the process while coping with the fluctuations in the amount of gas generated.

[0117] To this end, the above-described inference step (S40) analyzes multiple control parameter data and gas capture data used for the learning purpose, and selects items of control parameters to be adjusted according to changes in the amount of harmful gas generated measured during process execution from among multiple items constituting the control parameters.

[0118] And, when performing the process, the items of the control parameters to be adjusted to reduce the amount of gas generated according to the amount of harmful gas generated are to use the items selected above.

[0119] In addition, the above application step (S50) determines a reference value for the gas intake amount according to the size of the signal-to-noise ratio corresponding to the optimal control parameter data, and transmits this reference value together with the optimal control parameter data to the injection molding machine (100) for use. Through this, the injection molding machine (100) can control the gas intake amount according to the gas generation amount by controlling the gas intake amount with the reference value.

[0120] The above learning data addition step (S60) and relearning step (S70) are steps performed when additional control parameter data and gas capture data are received according to actual process processing, and are performed by the data processing unit (12) and learning unit (13) as in learning.

[0121] In the above learning data addition step (S60), the learning unit (13) creates learning data to which the multiple feature vectors generated for the multiple control parameter data and the received gas capture data are added through the data processing unit (12).

[0122] The above retraining step (S70) is performed when a predetermined amount of additional training data is generated, thereby retraining the prediction model using the training data. Here, the retrained prediction model is used in the inference step (S40) performed when optimal control parameter data is subsequently derived.

[0123] Meanwhile, the prediction model is not generated by the learning unit (13), but can be generated in advance and then transplanted to the inference unit (14) when configuring the optimal control support device for the injection molding machine. In this case, the learning data generation step (S10) and the learning step (S20) are not performed, but it is good to operate the learning unit (13) for re-learning, and the learning data used for generating the prediction model is stored so that it can be used when re-learning in the learning data addition step (S60) and the re-learning step (S70).

[0124] [Explanation of symbols]

[0125] 1: Mold 1a: Cavity

[0126] 10: Injection molding machine optimal control support device

[0127] 11: Interface section 12: Data processing section 13: Learning section

[0128] 14: Inference section 15: Application section

[0129] 100: Injection molding machine

[0130] 110: Hopper

[0131] 120: Cylinder 121: Screw 122: Screw actuator

[0132] 130: Nozzle

[0133] 140: Gas discharger

[0134] 150: Gas capture / purifier

[0135] 160: Controller

Claims

1. In an injection molding machine (100) that melts plastic raw materials and injects them into a mold (1), captures and senses gas generated from the molten resin to be injected, purifies it, and then discharges it, and performs the process according to control parameter data set to match the molding conditions by a controller (160), in an injection molding machine optimal control support device for suppressing gas emission, When receiving multiple control parameter data applicable to a process, a data processing unit (12) that embeds each control parameter data into a feature space through a kernel function and generates multiple feature vectors to be input into a regression analysis model; A regression analysis model that predicts the signal-to-noise ratio (SNR) of the characteristic value used as gas capture data for using the Taguchi Method as a feature vector is prepared as a prediction model, and an inference unit (14) that derives the signal-to-noise ratio according to each feature vector as a prediction model; An application unit (15) that analyzes multiple predicted signal-to-noise ratios according to the Taguchi Method, determines control parameter data corresponding to one signal-to-noise ratio as an optimal control parameter, and applies the optimal control parameter data to process processing; Including Injection molding machine optimal control support device.

2. In paragraph 1, The above application part (15) The control parameter data corresponding to the largest signal-to-noise ratio among multiple signal-to-noise ratios is determined as the optimal control parameter. Injection molding machine optimal control support device.

3. In paragraph 1, When receiving a plurality of control parameter data applied to process execution and a plurality of gas capture data measured while performing a plurality of processes with each control parameter data, a learning unit (13) that generates learning data with a plurality of feature vectors generated for the plurality of control parameter data and the gas capture data through the data processing unit (12) and trains a prediction model that estimates a signal-to-noise ratio with the feature vector; Injection molding machine optimal control support device.

4. In paragraph 3, The above reasoning part (14) By analyzing the multiple control parameter data and gas capture data used in the above learning unit (13), the control parameters to be adjusted are selected according to the change in the amount of harmful gas generated measured during the process. The above application part (15) The selected control parameters are used as control parameters to adjust to reduce the amount of gas generated according to the amount of harmful gas generated during process processing. Injection molding machine optimal control support device.

5. In paragraph 1, The above signal-to-noise ratio is Using the signal-to-noise ratio according to the network characteristics (The lower the better) Injection molding machine optimal control support device.

6. In paragraph 1, The above application part (15) By determining the gas intake amount according to the size of the signal-to-noise ratio and including it in the optimal control parameter data, it can be applied to control the gas intake amount. Injection molding machine optimal control support device.

Images