Method for operating an injection molding machine, injection molding machine, and computer program product

Abstract

The invention relates to an injection molding machine and to a method for operating same. The injection molding machine has a plasticizing screw with a nonreturn valve for preventing a backflow of plasticized material during an injection process. At least one injection process is carried out, wherein at least one curve of at least one variable (IP, TM, RM, SP, SV, RV) which characterizes the injection process is detected. The detected curve is evaluated using an evaluation algorithm which is trained using a machine learning process in order to determine the closing behavior of the nonreturn valve, in particular the most probable closing time (S) and / or the backflow blocking efficiency.

Description

[0001] Method for operating an injection molding machine, injection molding machine and computer program product

[0002] This patent application claims priority from German patent application DE 10 2023 205 218.0, the contents of which are incorporated herein by reference.

[0003] The present invention relates to a method for operating an injection molding machine and to an injection molding machine. The present invention also relates to a computer program product for controlling and / or monitoring an injection molding machine.

[0004] When operating an injection molding machine, several processes are performed per injection molding cycle. In particular, in a metering process, the material to be injected is plasticized and conveyed by a rotating plasticizing screw toward the nozzle of the plasticizing unit, where it is metered. During the subsequent injection process, the plasticized material obtained in the metering process is injected into the injection mold via the nozzle by an axial feed movement of the plasticizing screw. In the subsequent holding pressure process, pressure is maintained to compensate for shrinkage of the material due to its cooling in the mold. To prevent backflow of the plasticized material along the plasticizing screw during the injection process and the holding pressure process, non-return valves are installed in the area of ​​the screw tip.During the dispensing process, the non-return valve is open to allow the plasticized material to be dispensed. At the start of the injection process, the non-return valve must be reliably closed to ensure reproducible process properties and thus the desired properties of the parts to be produced. Depending on the closing time of the non-return valve, more or less material flows back along the plasticizing screw, which is referred to as leakage.

[0005] Check valves have proven to be simple, proven non-return valves. When operating an injection molding machine, precise knowledge of the check valve's closing behavior is desirable. The plasticized material flowing behind the check valve exerts a torque on the plasticizing screw and thus on its rotary drive. Therefore, an attempt is made to derive information about the closing behavior from the torque of the rotary drive of the plasticizing screw by determining a local maximum of this torque. However, it has been found that in many applications, the torque curve is atypical and lacks a local maximum. In these cases, this approach does not allow any conclusions to be drawn about the closing behavior.

[0006] It is the object of the present invention to improve an operating method for an injection molding machine, in particular to provide an operating method in which reliable statements can be made about the closing behavior of the check valve.

[0007] This object is achieved by a method for operating an injection molding machine according to claim 1 and an injection molding machine according to claim 13. The object is also achieved by a computer program for monitoring and / or controlling an injection molding machine according to claim 14.

[0008] The injection molding machine to be operated according to the method has a plasticizing screw with a check valve to prevent backflow of plasticized material during an injection process. At least one injection process is performed, with at least one characteristic variable characteristic of the injection process being recorded. The at least one recorded characteristic of the at least one characteristic variable is evaluated using an evaluation algorithm trained by machine learning to determine the closing behavior of the check valve of the plasticizing screw.

[0009] The core of the method lies in the use of an evaluation algorithm trained using machine learning, preferably supervised machine learning. This enables a data-driven evaluation of at least one profile of the characteristic variable for the injection process. Machine learning makes it possible to determine the closing behavior from the at least one profile without a specific evaluation routine being specified, for example, without the evaluation algorithm being dependent on the existence and evaluation of a local maximum. With the aid of the evaluation algorithm, the at least one profile can be evaluated independently of its individual properties, in particular independently of the existence of local maxima, and a respective closing behavior of the check valve can be assigned.This allows for a reliable determination of the closing behavior, even with atypical curves of at least one characteristic variable for the injection process. A particular advantage of the method is that several curves of different characteristic variables for the injection process can be evaluated using the evaluation algorithm. This allows for a particularly precise determination of the closing behavior. The evaluation is not limited to specific variables and / or their typical characteristics, such as local minima and / or maxima.

[0010] The recorded curve is understood, in particular, to be a temporal curve of the at least one characteristic variable during the injection process. Preferably, the at least one curve is recorded at least for the duration of the injection process. The at least one curve can also be recorded beyond the injection process, for example, over at least parts of the holding pressure process.

[0011] The at least one recorded curve can be evaluated with the aid of the evaluation algorithm, for example using pattern recognition or image recognition methods. For example, the at least one recorded curve can be evaluated directly. It is also possible to extract characteristic characteristics, so-called features, from the at least one recorded curve and use them for evaluation with the aid of the evaluation algorithm. Suitable features include, among others, integrals, differentials, minima, maxima, median, standard deviation, entropy, sample entropy, longest time period above and / or below the average and / or one or more quantiles. Preferably, several features of the at least one curve, in particular several features from several curves with different characteristic variables, are used for the evaluation.

[0012] The evaluation algorithm can comprise suitable forms of machine learning, in particular supervised machine learning. The evaluation algorithm can in particular comprise one or more artificial neural networks, in particular one or more deep neural networks (DNN), and / or one or more suitable ensemble methods, in particular random forest and / or gradient boosted tree, and / or one or more support vector machines (SVM). The evaluation algorithm can in particular have a classifier, for example in the form of a suitable neural network and / or suitable ensemble methods (for example random forest or gradient boosted tree) and / or a support vector machine, for classifying the closing behavior into several categories. During the evaluation, correlations between different parameters and / or parameter curves and / or features extracted therefrom can in particular be determined.

[0013] The closing behavior is, in particular, a qualitative and / or quantitative measure of the closing of the check valve. The closing behavior indicates, for example, whether and / or to what extent the check valve is closed. The closing behavior indicates, in particular, at which point in the injection process and / or at which axial position along the injection path of the plasticizing screw the check valve is closed, in particular completely.

[0014] The closing behavior is a parameter associated with the respective injection process. The closing behavior is determined, for example, independently of other properties of the plasticizing unit, the plasticizing screw, and / or the check valve, in particular the wear condition of the check valve. The closing behavior can be and / or determined independently of the properties of the material being processed and related process parameters. In particular, the closing behavior can be determined independently of the process temperature.

[0015] The evaluation algorithm is trained in particular to assign the at least one determined profile of the at least one variable characteristic of the injection process to different closing behaviors, for example by classifying the closing behavior into two or more categories. Possible categories can, for example, cover one or more of the following case groups: check valve closed, not closed, not fully closed, . . .). Additionally or alternatively, the determination of the closing behavior can include the determination of further characteristic parameters, in particular a closing time and / or a non-return valve efficiency of the check valve. The non-return valve efficiency can, for example, quantify which portion of the injection path and / or which portion of the injection time is covered with the check valve closed.

[0016] Further characteristic parameters, in particular the closing time and the non-return valve efficiency, can be determined in particular using one or more suitable regression algorithms of the evaluation algorithm.

[0017] The material plasticized in the dosing process is, in particular, a plastic material. The plasticized material is, for example, in a melt-like state. The plasticized material is also referred to here and below as a melt, in particular as a plastic melt.

[0018] According to a preferred aspect of the method, the at least one characteristic variable for the injection process comprises one or more of the following variables: injection pressure, injection speed, position of the plasticizing screw, torque of a rotary drive of the plasticizing screw, rotation speed of the plasticizing screw, and / or drive torque of a linear drive of the plasticizing screw. These variables characterizing the injection process have proven particularly suitable for evaluation using the evaluation algorithm trained by machine learning.

[0019] The rotary drive of the plasticizing screw is, in particular, an electric drive. The rotary drive is used in particular to drive the plasticizing screw in rotation during the metering process and is also referred to as a metering drive. During the injection process, the angular position of the plasticizing screw can be maintained, for example, by means of position control. The rotary drive can, for example, apply a holding force to the plasticizing screw, for example, to prevent rotation due to backflow of melt. The torque of the rotary drive is also referred to as the metering torque. The torque of the rotary drive is, in particular, the holding torque required to hold the plasticizing screw. The torque can be measured, for example, using a torque sensor on the rotary drive.

[0020] The linear drive of the plasticizing screw, also called the injection drive, is preferably an electric drive. It can, for example, comprise an electric motor and a gear for converting the rotary motion into linear motion, such as a ball screw. The drive torque of the linear drive can, for example, be measured as the torque of the electric motor. The drive torque of the linear drive is also referred to as the injection torque, in particular the injection torque. The drive torque can, for example, be measured using a torque sensor on the linear drive.

[0021] The injection pressure is the pressure at which the plastic melt is injected into the mold by the axial movement of the plasticizing screw. The injection pressure can be measured, for example, by a pressure sensor on the plasticizing screw, particularly in the area of ​​the end facing away from the nozzle.

[0022] The position of the plasticizing screw is also referred to as the screw position. The position of the plasticizing screw is, in particular, its axial position. The axial position of the plasticizing screw changes during the injection process due to the feed movement required for this. The axial position of the plasticizing screw can be measured, for example, using a displacement sensor.

[0023] The injection speed is the axial speed of the plasticizing screw, particularly during the injection process and the holding pressure process. The injection speed can serve as a controlled variable during the injection process. For the evaluation of the closing behavior of the check valve, the injection speed curve during the holding pressure phase is particularly relevant. The injection speed can be determined, for example, as a change in the axial position using a displacement sensor. In the case of speed control, particularly during the injection process, the injection speed can also be read out as a controlled variable. The rotational speed of the plasticizing screw, also referred to as screw rotation, is the rotary speed of the plasticizing screw. During the dosing process, this is a controlled variable.The rotational speed during the injection phase is particularly relevant for evaluating the closing behavior of the check valve. The rotational speed of the plasticizing screw can be determined, for example, by monitoring the rotation of the rotary drive.

[0024] According to a preferred aspect of the method, at least two, in particular at least three, of the variables characteristic of the injection process are evaluated using the evaluation algorithm. The evaluation is particularly precise and not limited to the use of individual variables and their respective properties. The use of the torque of the rotary drive, the injection pressure, and the position, in particular the axial position, of the plasticizing screw has proven particularly suitable. Preferably, the torque of the rotary drive, the injection pressure, the position, in particular the axial position, of the plasticizing screw, and the speed, in particular the axial speed, of the plasticizing screw are used in the evaluation.Particularly preferably, at least the injection pressure, the in particular axial position of the plasticizing screw, the torque of the rotational drive, the rotational speed of the plasticizing screw, the drive torque of the translational drive, in particular the torque of the electric motor of the translational drive, and the injection speed are used in the evaluation.

[0025] In particular, at least one, preferably several, features can be extracted from the curves of the respective variables for evaluation.

[0026] In addition to the at least one curve of the at least one variable characteristic of the injection process, other process variables can also be taken into account in the evaluation, for example a duration of the dosing process and / or the size of a mass cushion and / or the clamping force, in particular the clamping force. Advantageously, the process variables to be evaluated, in particular the variables characteristic of the injection process to be evaluated, as well as their evaluation, in particular the extracted features, enable conclusions to be drawn about the closing behavior of the check valve, in particular the closing time and / or the non-return valve efficiency, regardless of other properties of the injection molding machine. The method can be used on multiple machines, in particular different machines, and / or for different process parameters, in particular regardless of the materials to be processed and / or process temperatures.The procedure, in particular the process variables and / or extracted features used for the evaluation, can in particular be standardized.

[0027] According to a preferred aspect of the method, the evaluation algorithm classifies the closing behavior of the check valve into two or more categories. This classification allows simple and precise statements to be made about the correct closing behavior. In particular, error conditions, such as a failure of the check valve to close, can be precisely determined. In particular, potentially defective production batches can be identified early and, if necessary, sorted out after further investigation. Unnecessary scrap is avoided.

[0028] According to a preferred aspect of the method, the classification is carried out into at least two categories: “check valve closed” and “check valve not closed”.

[0029] Preferably, the at least two categories may additionally include at least one of the subcategories “check valve not fully closed”, “check valve closed in time” and / or “check valve closed late”.

[0030] According to a preferred aspect of the method, the evaluation algorithm determines a closing time of the check valve and / or a non-return valve efficiency from the at least one curve of the at least one variable characteristic of the injection process. The closing time and / or the non-return valve efficiency can be determined, for example, using one or more suitable regression algorithms. Knowledge of the closing time and / or the non-return valve efficiency enables particularly practice-relevant statements about the closing behavior. In particular, it is possible to detect a late closing of the check valve. Based on the closing time and / or the non-return valve efficiency, it is possible, in particular, to infer a leakage volume, for example in order to adapt further processes of the same or subsequent injection molding cycles accordingly.

[0031] Preferably, the evaluation algorithm determines at least one non-return valve efficiency from the at least one profile of the at least one variable characteristic of the injection process, wherein the non-return valve efficiency quantifies which portion of an injection path and / or which portion of an injection time of the injection process was covered with the non-return valve closed.

[0032] The determination of the closing time and / or the check valve efficiency can be carried out, in particular, using a probability analysis. Using the evaluation algorithm, for example, a most probable closing time can be determined. Depending on the probability distribution, the validity and / or potential error range of the determined closing time can also be determined. The most probable closing time can be used to determine the most probable check valve efficiency.

[0033] According to a preferred aspect of the method, the at least one recorded curve of the at least one variable characteristic of the injection process is normalized and / or smoothed for evaluation using the evaluation algorithm. This enables simpler evaluation with the aid of the evaluation algorithm. In particular, the smoothing and / or normalization allows the extraction of comparable features for corresponding curves. The closing behavior of the check valve can be determined reliably and precisely regardless of the absolute values ​​in the process. In particular, the consideration of absolute values ​​is not necessary. The determination of the closing behavior is not impaired by outliers in the measured values. Particularly preferably, the at least one recorded curve is standardized in each case for evaluation. This allows good comparability of the recorded curves, especially for classification, regardless of the respective absolute values.Smoothing can be optional, for example in the case of very noisy signals.

[0034] According to a preferred aspect of the method, further processes of the respective injection molding cycle are adjusted depending on the closing behavior determined by the evaluation algorithm. In particular, the pressure and / or duration of a holding pressure process can be adjusted to account for different leakages due to different closing times of the check valve. The method allows for adaptive process adjustment.

[0035] According to a preferred aspect of the method, the pressure and / or duration of a holding pressure process of the respective injection molding cycle is adjusted. This enables particularly precise process adjustment, in particular to compensate for varying leakages due to different closing times of the check valve.

[0036] It is also possible to adapt subsequent injection molding cycles depending on the specific closing behavior, in particular setting parameters of the dosing, injection and / or holding pressure process of subsequent injection molding cycles.

[0037] Depending on the determined closing behavior, the setting parameters of subsequent injection molding cycles, in particular the dosing and / or injection process, can preferably be adjusted to improve the closing behavior of the check valve, in particular a closing time and / or a non-return valve's efficiency. The closing behavior can thereby be optimized, in particular iteratively.

[0038] According to a preferred aspect of the method, the closing behavior is determined for several injection molding cycles, in particular for all injection molding cycles. This allows for essentially seamless monitoring and early detection of possible error states and their potential impact on the parts to be manufactured. Particularly preferably, the determined closing behavior is compared with a target value and / or with predetermined closing behavior.

[0039] According to a preferred aspect of the method, a temporal progression of the closing behavior and / or one or more related parameters, for example, closing time, non-return valve efficiency, injection pressure, and / or injection speed, are determined over several injection molding cycles. From the progression of the closing behavior, for example, a trend can be determined indicating the extent to which the closing behavior is changing. This enables condition monitoring of the injection molding machine, for example, monitoring a successive change in the closing time, the non-return valve efficiency, superimposed changes in the injection pressure and / or the injection speed.

[0040] According to a preferred aspect of the method, the evaluation algorithm is trained using training data from at least one training injection molding machine, wherein the training injection molding machine has at least one sensor for detecting at least one profile of the at least one variable characteristic of the injection process for a plurality of training injection molding cycles to generate the training data, and wherein the training injection molding machine has an additional training sensor that at least indirectly measures the closing behavior of the check valve in the respective training injection molding cycle, wherein measurement data from the training sensor is used for, in particular, automatic labeling of the training data. The closing behavior of the check valve can be unambiguously determined using the measurement data from the training sensor.This knowledge enables simple labeling of the training data, particularly with regard to the closing state at the end of the injection process (fully closed, not closed, etc.), the closing times, the check valve efficiency, and / or other parameters that allow conclusions to be drawn about the closing behavior of the check valve. By training with the training data labeled in this way, the knowledge gained with the help of the training sensor can be used in the injection molding machine without the need for an additional training sensor. When operating the injection molding machine, especially a series machine, measurement data from the training sensor, if available at all, need not be taken into account for the evaluation.

[0041] With the help of the training injection molding machine, training data sets are generated, whereby the training data sets each contain the training data corresponding to the respective variables characteristic of the injection process and the label determined with the help of the training sensor.

[0042] The measurement data from the training sensor can, for example, be a temporal profile of a variable measured by the training sensor. The training sensor is arranged in particular in the area of ​​the check valve, in particular downstream of the check valve in the plasticizing cylinder. For example, the training sensor can comprise an ultrasonic sensor and / or a pressure sensor positioned downstream of the check valve to measure the state of the melt, in particular the melt pressure, downstream of the check valve. The state of the melt downstream of the check valve can be used to directly determine the closing behavior of the check valve.

[0043] According to a preferred aspect of the method, the training sensor of the training injection molding machine is a pressure sensor arranged behind the check valve for measuring the melt pressure behind the check valve. Measuring the melt pressure behind the check valve has proven particularly suitable for at least indirectly measuring the closing behavior of the check valve. The material flowing behind the check valve increases the melt pressure. When the check valve closes, the melt pressure decreases. The closing behavior can therefore be clearly determined by the onset of an abrupt drop in the melt pressure behind the check valve. The measurement behind the check valve is not influenced by other process conditions, so that the abrupt drop in the melt pressure is pronounced when the check valve closes.Atypical curves, in particular without an abrupt drop in melt pressure despite the closure of the check valve, do not occur. According to a preferred aspect of the method, the injection molding machine being operated does not have the training sensor. During operation of the injection molding machine, the closing behavior of the check valve is determined solely on the basis of the variables characteristic of the injection process. Sensors that at least indirectly measure the closing behavior of the check valve are expensive and require a lot of maintenance. In the injection molding machine to be operated, for example in a series machine, the training sensor is preferably omitted, which reduces setup and maintenance effort. Particularly preferably, the training injection molding machine is designed essentially the same as the injection molding machine to be operated, with the exception of the additional training sensor.

[0044] Preferably, the determined closing behavior, in particular a determined category, a closing time, and / or a non-return valve efficiency, can be output to a machine operator. The injection molding machine can have an output unit for this purpose, for example, in the form of a screen. Additionally or alternatively, measures derived from the determined closing behavior can also be displayed and presented to the operator for information and / or implementation.

[0045] The injection molding machine according to the invention comprises a plasticizing screw with a check valve for preventing backflow of plasticized material during an injection process, at least one sensor for detecting at least one profile, at least one variable characteristic of the injection process, and a control unit. The control unit is designed to evaluate the at least one detected profile of the at least one variable characteristic of the injection process using an evaluation algorithm trained by machine learning to determine a closing behavior of the check valve. The advantages of the injection molding machine correspond to the advantages discussed in connection with the method described above. The control unit can, in particular, be designed to implement one or more of the advantageous method features discussed above.

[0046] The injection molding machine comprises, in particular, a clamping unit and a plasticizing unit, which may be known per se. For example, the plasticizing unit may have a rotary drive for driving the plasticizing screw in rotation, particularly during a dosing process. The plasticizing unit may also have a translational drive for driving the plasticizing screw in translation, for example, for its axial displacement during the injection process.

[0047] The control unit controls, in particular, the injection process, preferably the entire injection molding cycle. The control unit serves, in particular, to monitor the closing behavior of the check valve. The control unit has, in particular, a data memory, a main memory, and a processor for storing and executing the evaluation algorithm. The evaluation algorithm can, for example, be stored in a data memory of the control unit and loaded into the main memory and processor for execution.

[0048] The at least one sensor comprises, in particular, a pressure sensor for detecting an injection pressure, a displacement sensor for detecting, in particular, an axial screw position and / or injection speed, a torque sensor for detecting a torque of a rotary drive of the plasticizing screw, and / or a sensor for detecting a drive torque of the linear drive of the plasticizing screw, in particular a torque sensor for detecting a torque of an electric motor of the linear drive. The injection molding machine preferably comprises several, in particular all, of the aforementioned sensors.

[0049] The computer program product according to the invention for controlling and / or monitoring an injection molding machine has commands which, when the program is executed by a computer, cause the computer to execute an evaluation algorithm trained using machine learning methods to determine the closing behavior of a check valve of a plasticizing screw of an injection molding machine based on an evaluation of at least one profile of at least one variable characteristic of an injection process. The computer program product can be executed, in particular, by means of a control unit of the injection molding machine. The computer program product has the advantages described with regard to the method. Advantageously, the computer program product can be designed to implement further of the optional method features described above.For example, the computer program can be designed to control and / or read one or more sensors of the injection molding machine to record the at least one curve to be evaluated.

[0050] The computer program product is stored, in particular, on computer-readable data storage devices, in particular data storage devices that can be connected to and / or inserted into computers, for example, in the form of a control unit of the injection molding machine. The computer program product can comprise the data storage device. The computer program product can also be stored on a data storage device of a computer, in particular, a control unit of the injection molding machine. Instructions of the computer program product can be loaded into a processing unit of the computer, for example, into a processor of the control unit, for execution.

[0051] According to a preferred aspect of the computer program product, the evaluation algorithm is trained using training data from at least one training injection molding machine, wherein the training injection molding machine has at least one sensor for detecting at least one profile of at least one variable characteristic of an injection process for a plurality of training injection molding cycles to generate training data, and wherein the training injection molding machine has an additional training sensor that at least indirectly measures the closing behavior of the check valve in the respective training injection molding cycle, wherein measurement data from the training sensor is used for, in particular, automatic, labeling of the training data. The advantages and optional features of training the evaluation algorithm, in particular with regard to generating the training data, correspond to the advantages and optional features described with regard to the method.

[0052] Further features, advantages, and details of the invention will become apparent from the following description of exemplary embodiments with reference to the drawings. They show:

[0053] Fig. 1 shows a schematic diagram of an injection molding machine, with only the components of the injection molding machine essential to the invention being shown, Fig. 2A - 2C show schematic time courses of various parameters characteristic of the injection process for different injection processes, which are evaluated to determine a closing behavior of a check valve of the plasticizing screw by means of an evaluation algorithm trained by machine learning,

[0054] Fig. 3 schematically shows a training injection molding machine for generating training data for training the evaluation algorithm, wherein the training injection molding machine has a training sensor that measures at least a direct closing behavior of the check valve, and

[0055] Fig. 4A - 4C schematically show time courses of measured values ​​of the training sensor of the training injection molding machine in Fig. 3 for the exemplary injection processes shown in Figs. 2A to 2C.

[0056] Corresponding parts are provided with the same reference numerals in Figs. 1 to 4. Details of the exemplary embodiments explained in more detail below may also constitute an invention in themselves or be part of a subject matter of the invention.

[0057] Fig. 1 schematically shows an injection molding machine 1. The injection molding machine 1 has a plasticizing unit 2. The plasticizing unit 2 can be moved to a fixed clamping plate 4 of a clamping unit (not shown) by means of a starting cylinder 3. The injection molding machine 1 has a control unit 5 for controlling the operation of the injection molding machine 1. The control unit 5 is connected to machine parts of the injection molding machine 1 via a data connection 6 for controlling the machine parts and / or for reading data from the machine parts, for example, for reading data from sensors. Other components of the injection molding machine 1 are not shown for the sake of clarity.

[0058] The plasticizing unit 2 comprises a plasticizing cylinder 7, a nozzle 8, and a plasticizing screw 9 that can be driven both rotationally and translationally within the plasticizing cylinder 7. A rotary drive 10 serves to drive the plasticizing screw 9 rotationally. The rotary drive 10 is an electric drive. The plasticizing screw 9 can be driven translationally with the aid of a translational drive 11. In the illustrated embodiment, the translational drive 11 comprises an electric motor whose rotational movement is converted into a translational movement for translational driving of the plasticizing screw 9 via a ball screw drive 12.

[0059] Plastic granules 14 are fed into the plasticizing cylinder 7 via a hopper 13. In the so-called dosing process, the plastic granules 14 are conveyed toward the nozzle 8 by means of the rotationally driven plasticizing screw 9, where they are plasticized. For this purpose, the plastic material is heated and melted via heating bands 15. The temperature is monitored by thermal sensors 16. During the dosing process, the rotary drive 10 is active to drive the plasticizing screw 9. The dosing process continues until a predetermined amount of plastic melt 17 is metered between the screw tip 18 of the plasticizing screw 9 and the nozzle 8.

[0060] After dosing, an injection process follows. For the injection process, the plasticizing unit 2 is moved to the clamping plate 4 using the start-up cylinder 3. The plastic melt 17 is injected into the injection mold (not shown) via the nozzle 8 by an axial feed movement of the plasticizing screw 9. The feed movement of the plasticizing screw 9 occurs over an injection path or injection time. For this purpose, the plasticizing screw 9 is moved axially using the translational drive 11. The axial movement of the plasticizing screw 9 is speed-controlled, for example. During the injection process, the rotary drive 10 is passive and exerts a holding force to prevent the plasticizing screw 9 from rotating.

[0061] The injection process is followed by a holding pressure process. During the holding pressure process, pressure is maintained by means of a pressure control of the plasticizing screw 9 to compensate for shrinkage of the plastic material cooling in the injection mold. The holding pressure process typically lasts until the so-called sealing point. Switching between the injection process and the holding pressure process, in particular between speed control and pressure control of the translational drive 11, occurs at the so-called switching point T.

[0062] The basic procedure for operating the injection molding machine 1 described above requires that the plastic melt can reach in front of the plasticizing screw 9 during the metering process. During the injection process, backflow of the plastic melt along the plasticizing screw 9 should be prevented. For this purpose, a known non-return valve 20 is arranged in the area of ​​the tip 18. The non-return valve 20 is designed here as a check valve. The check valve opens due to the pressure of the plastic melt conveyed towards the nozzle 8 during the metering process. During the injection process, a pressure difference arises between the plastic melt 17 in front of the check valve 20 and in areas behind the check valve 20, so that the check valve is pressed against the webs 21 of the plasticizing screw 9 and thus closes.The non-return valve 20, designed as a check valve, is structurally simple and does not require a separate control. However, the disadvantage is that the closing behavior of the check valve 20 is not always reproducible depending on the injection process. In particular, error conditions can occur in which the check valve 20 does not close, does not close completely, or does not close in a timely manner.

[0063] Until the check valve 20 closes, plastic melt 17 flows back behind the check valve 20, which is also referred to as leakage. Variations in the closing behavior of the check valve 20, particularly if it does not close or closes too late, change the leakage volume, which influences the properties and quality of the parts produced in the injection molding process. In particular, this can lead to non-reproducible properties and / or production downtimes.

[0064] The control unit 5 controls the injection molding machine 1 to carry out the injection molding process, in particular the dosing process, the injection process, and the holding pressure process. Furthermore, the control unit 5 monitors the closing behavior of the check valve 5. For this purpose, a temporal profile of at least one variable characteristic of the injection process is determined. Suitable characteristic variables include, in particular, an injection pressure IP, a torque TM of the translational drive 11, a torque RM of the rotational drive 10, an axial position SP of the plasticizing screw 9, a rotational speed RV of the plasticizing screw 9, and / or an injection speed SV. Various sensors are present on the injection molding machine 1 to detect these variables.

[0065] The injection pressure IP is the pressure at which the plastic melt 17 is injected into the mold. To measure the injection pressure IP, a pressure sensor 22 is arranged at the end of the plasticizing screw 9 facing away from the screw tip 18. The pressure sensor 22 can comprise one or more load cells.

[0066] During the injection process, the plasticizing screw 9 is held in a fixed angular position by the rotary drive 10 (position control). The plastic melt 17 flowing behind the check valve 20 exerts a rotational force on the plasticizing screw 9 via the webs 21. The torque RM of the rotary drive 10 is the holding torque that the rotary drive 10 applies to hold the plasticizing screw 9 in position. A torque sensor 23 is connected to the rotary drive 10 to detect the torque RM.

[0067] Despite the position control, rotational movements of the plasticizing screw 9 may occur. The rotational speed RV of the plasticizing screw 9 reflects these rotational movements.

[0068] The torque TM of the translational drive TM is the drive torque of the electric motor of the translational drive 11, which is required to axially displace the plasticizing screw 9 during the injection process. The torque TM of the translational drive 10 is determined using a torque sensor 24 of the translational drive 11.

[0069] The axial position SP of the plasticizing screw 9 is also called the screw position.

[0070] The screw position SP is determined by a position sensor 25. The injection speed SV is the axial speed of the plasticizing screw 9, particularly during the injection process. It can be detected, for example, as a change in the position of the plasticizing screw 9 using the position sensor 25.

[0071] Figures 2A to 2C schematically show the curves of the above-mentioned variables characteristic of the injection process over time t for different injection processes with different closing behaviors of the check valve 20. The respective variables are shown in standardized form. The respective curves are shown for the injection process and, beyond the switching time T, also for the holding pressure process.

[0072] Figs. 2B and 2C show exemplary injection processes in which the check valve 20 closes at a closing time S. In Fig. 2A, the check valve 20 does not close. This represents a fault condition in which component quality cannot be guaranteed.

[0073] The closing of the check valve 20 can usually be traced back to a local maximum of the torque RM of the rotary drive 10. Until the check valve 20 closes, plastic melt 17 penetrates behind the check valve and exerts a torque on the plasticizing screw 9. After the check valve 20 closes, this torque decreases. This is shown in Fig. 2C, where the torque R of the rotary drive 10 shows a clear local maximum at the closing time S. The closing of the check valve 20 is not always reflected in a maximum of the torque RM curve of the rotary drive 10. In the case shown in Fig. 2B, the check valve 20 closes without this being clearly deducible from the torque RM curve of the rotary drive. The same applies to the other variables shown in Fig. 2B that are characteristic of the injection process. In comparison with Fig.2A, no clear distinction can be made between the closing of the check valve 20 in Fig. 2B and the fault condition in which the check valve 20 does not close in Fig. 2A based on individual characteristics of the curves.

[0074] The control unit 5 has an evaluation algorithm 27 for determining the closing behavior of the check valve 20. The evaluation algorithm 27 is, for example, part of a computer program product stored in a data memory of the control unit 5. The computer program product has instructions that, when executed by the processor of the control unit 5, execute the evaluation algorithm 27. The computer program product can, for example, be loaded into the main memory of the control unit 5 for execution by a processor of the control unit 5.

[0075] The evaluation algorithm 27 was trained using supervised machine learning methods. The evaluation algorithm 27 comprises, in particular, at least one artificial neural network, in particular at least one deep neural network (DNN) and / or at least one suitable ensemble method, in particular random forest and / or gradient boosted tree, and / or at least one support vector machine (SVM).

[0076] The evaluation algorithm 27 is trained to assign the determined curves of the variables characteristic of the injection process to different closing behaviors by classifying the closing behavior into two or more categories. For example, the evaluation algorithm 27 classifies the curves into the categories: "check valve closed" (see Figs. 2B and 2C) or "check valve not closed" (see Fig. 2A). It is also possible to classify them into further subcategories. For example, the closing of the check valve 20 can be classified into the subcategory "closed in time" or "closed late". Further subcategories can relate to individual curves, for example, "typical curve of the torque of the rotary drive" (see Fig. 2C) or "atypical curve of the torque of the rotary drive" (see Fig. 2B).

[0077] The evaluation algorithm 27 evaluates the curves of at least one, in particular several, preferably all of the characteristic variables described above. The use of a machine learning-based evaluation algorithm 27 has the advantage that the evaluation is not restricted or dependent on fixed characteristic properties of individual curves (for example, on a determinable maximum of the torque RM of the rotary drive 10). The evaluation algorithm 27 can, in particular, use pattern recognition and / or image recognition methods to assign different characteristic curves to different closing behaviors, independent of specific evaluation parameters. For example, the evaluation algorithm 27 can extract one or more features characterizing the curve and use them for the evaluation.

[0078] Preferably, the evaluation algorithm 27 also determines further parameters of the closing behavior, in particular a most probable closing time S, using a suitable regression algorithm. By evaluating the curves, the evaluation algorithm 27 can determine the most probable closing time S without requiring any characteristic properties of the curves at the closing time S.

[0079] With the help of the evaluation algorithm 27, the closing behavior of the check valve 20 can be reliably determined. In particular, a fault condition can be detected early. If the check valve 20 does not close or closes too late, for example, the produced parts can be subjected to a closer inspection in order to identify and sort out defective parts at an early stage.

[0080] Based on the determined closing behavior, in particular the determined closing time S, the injection molding process can be adaptively adjusted, for example, by adjusting the holding pressure time and / or holding pressure level. This can be done for future injection molding cycles or within the same injection molding cycle. Variations in the closing behavior of the check valve 20 can thus be compensated to avoid production fluctuations and rejects.

[0081] The control unit 5 determines the closing behavior of the check valve 20, in particular a classified category of the closing behavior, the most probable closing time, and / or the non-return valve efficiency, for several closing cycles, preferably for all closing cycles. The determined closing behavior can, for example, be stored and / or evaluated over a longer period of time. For example, the current closing behavior can be compared with previous closing behaviors and / or a temporal profile of the closing behavior can be determined over several injection molding cycles. This allows a status of the injection molding machine 1, in particular of the plasticizing unit 2, to be monitored. For example, successive deviations in the probable closing time, the non-return valve efficiency, changes in pressure, and / or other variables can be detected early on.This allows for early maintenance measures to be initiated, which reduces waste in production and increases the longevity of the machine.

[0082] To train the evaluation algorithm 27, training data can be acquired during operation, for example, by recording the respective parameters characteristic of the injection process. The training data can then be labeled manually, for example, by an experienced operator and / or taking the part properties into account.

[0083] The evaluation algorithm 27 is preferably trained using training data generated and, in particular, automatically labeled using a special training injection molding machine. An exemplary training injection molding machine 30 is shown in Fig. 3. The training injection molding machine 30 essentially corresponds to the injection molding machine 1 in Fig. 1. Corresponding components bear the same reference numerals and will not be explained again.

[0084] The training injection molding machine 30 differs from the injection molding machine 1 essentially by an additional training sensor 31. The training sensor 31 at least indirectly measures the closing behavior of the check valve 20. In the illustrated embodiment, the training sensor 31 is a pressure sensor arranged behind the check valve 20 for measuring the melt pressure MP behind the check valve 20. The melt pressure MP depends directly on the plastic melt 17 flowing back behind the check valve 20. As soon as the check valve 20 is closed, no more plastic melt 17 flows in, so the melt pressure MP decreases. The training sensor 31 enables clear statements about the closing state of the check valve 20.

[0085] 4A to 4C, the melt pressure MP, which is detected with the aid of the training sensor 31, is plotted over time t for various injection processes. The injection processes shown in Figs. 4A to 4C correspond to those of the parameter curves shown in Figs. 2A to 2C. As can be seen from Figs. 4B and 4C, the melt pressure MP increases continuously up to the closing time S and drops abruptly after the check valve 20 closes. Independently of the curve of the other characteristic variables, the melt pressure MP forms a characteristic maximum at the closing time, which allows concrete statements to be made about the closing behavior and the closing time S. In the event that the check valve 20 does not close, the melt pressure MP also has a clear curve that does not have a local maximum up to the switching time T.

[0086] By evaluating the melt pressure MP, the closing time S, the check valve efficiency, and the closing behavior can be reliably determined independently of other variables. The findings obtained in this way can be used as labels for the training data. In particular, the training data obtained with the training injection molding machine 30 can be automatically fabled using the melt pressure MP.

[0087] For example, the training injection molding machine 30 is run through a plurality of injection molding cycles, and with the aid of its control unit 32, the respective course of the variables characteristic of the injection process, in particular the injection pressure IP, the torque TM of the translational drive 11, the torque RM of the rotational drive 10, the axial position of the plasticizing screw SP, the rotational speed RM of the plasticizing screw 9, and / or the injection speed SV, are recorded. In addition, the melt pressure MP is recorded in each case with the aid of the training sensor 31. The control unit 32 evaluates the melt pressure MP and determines the closing behavior and, if applicable, the closing time S and / or the non-return valve efficiency. These parameters are assigned to the training data as labels. Training data sets are generated from the training data and the associated label.

[0088] Using the generated training data sets, the evaluation algorithm can be trained using known machine learning methods. The trained evaluation algorithm 27 reliably determines the closing behavior without evaluating the melt pressure MP. In principle, it would be conceivable to use the training sensor 31 directly in the production machine, for example, in the injection molding machine 1, to monitor the closing behavior of the check valve 20. However, the additional sensor is expensive and requires a lot of maintenance. Training using the training injection molding machine allows, with the help of data-driven evaluation, to derive insights into the melt pressure MP from the other variables characteristic of the injection process, without the need for the training sensor 31.

[0089] In the example shown here, several variables characteristic of the injection process are recorded and evaluated. It is also possible to record and evaluate only individual variables.

[0090] In other embodiments of the training injection molding machine, an ultrasonic sensor can be arranged in addition to or instead of a pressure sensor arranged behind the check valve, which detects the closing of the check valve by means of ultrasound.

Claims

Patent claims 1. A method for operating an injection molding machine (1), wherein the injection molding machine (1) has a plasticizing screw (9) with a check valve (20) for preventing a backflow of plasticized material (17) during an injection process, comprising the steps Carrying out at least one injection process, wherein at least one profile of at least one variable (IP, TM, RM, SP, SV, RV) characteristic of the injection process is recorded, and evaluating the at least one recorded profile of the at least one variable (IP, TM, RM, SP, SV, RV) characteristic of the injection process by means of an evaluation algorithm (27) trained by machine learning for determining the closing behavior of the check valve (20) of the plasticizing screw (9).

2. Method according to claim 1, characterized in that the at least one variable characteristic of the injection process (IP, TM, RM, SP, SV, RV) comprises one or more of the following variables: injection pressure (IP), injection speed (SV), position of the plasticizing screw (SP), torque (RM) of a rotary drive (10) of the plasticizing screw (9), rotation speed (RV) of the plasticizing screw (9) and / or drive torque (TM) of a linear drive (11) of the plasticizing screw (9).

3. Method according to claim 2, characterized in that at least two of the variables (IP, TM, RM, SP, SV, RV) characteristic of the injection process are evaluated by means of the evaluation algorithm (27).

4. Method according to claim 2, characterized in that at least three of the variables (IP, TM, RM, SP, SV, RV) characteristic of the injection process are evaluated by means of the evaluation algorithm (27).

5. Method according to one of the preceding claims, characterized in that the evaluation algorithm (27) classifies the closing behavior of the check valve (20) into two or more categories.

6. Method according to claim 5, characterized in that the classification is carried out in at least two categories: “check valve closed” and “check valve not closed”.

7. Method according to one of the preceding claims, characterized in that the evaluation algorithm (27) determines a closing time (S) and / or a non-return valve efficiency of the check valve (20) from the at least one course of the at least one variable (IP, TM, RM, SP, SV, RV) characteristic of the injection process.

8. Method according to one of the preceding claims, characterized in that the evaluation algorithm (27) determines a non-return valve efficiency of the check valve (20) from the at least one course of the at least one variable characteristic of the injection process (IP, TM, RM, SP, SV, RV), wherein the non-return valve efficiency quantifies which portion of an injection path and / or which portion of an injection time of the injection process was covered with the non-return valve (20) closed.

9. Method according to one of the preceding claims, characterized in that the at least one recorded course of the at least one variable characteristic of the injection process (IP, TM, RM, SP, SV, RV) is normalized and / or smoothed for evaluation by means of the evaluation algorithm (27).

10. Method according to one of the preceding claims, characterized in that, depending on the closing behavior determined by the evaluation algorithm (27), further processes of a respective injection molding cycle are adapted.

11. Method according to claim 10, characterized in that an adjustment of pressure and / or duration of a holding pressure process of the respective injection molding cycle takes place.

12. Method according to one of the preceding claims, characterized in that the closing behavior is determined for several injection molding cycles.

13. Method according to claim 12, characterized in that a temporal course of the closing behavior and / or one or more parameters related thereto are determined over several injection molding cycles.

14. Method according to one of the preceding claims, characterized in that the evaluation algorithm (27) is trained by means of training data from at least one training injection molding machine (30), wherein the training injection molding machine (30) has at least one sensor (22, 23, 24, 25) for detecting at least one profile of the at least one variable (IP, TM, RM, SP, SV, RV) characteristic of an injection process for a plurality of training injection molding cycles for generating the training data, and wherein the training injection molding machine (30) has an additional training sensor (31) which at least indirectly measures the closing behavior of the check valve (20) in the respective training injection molding cycle, wherein measurement data (MP) from the training sensor (31) are used for, in particular, automatic, labeling of the training data.

15. The method according to claim 14, characterized in that the training sensor (31) of the training injection molding machine (30) is a pressure sensor arranged behind the check valve (20) for measuring the melt pressure (MP) behind the check valve (20).

16. Method according to one of claims 14 or 15, characterized in that the injection molding machine (1) operated thereby does not have the training sensor (31).

17. Injection molding machine (1), comprising a plasticizing screw (9) with a check valve (20) for preventing backflow of plasticized material (17) during an injection process, at least one sensor (22, 23, 24, 25) for detecting at least one profile of at least one variable (IP, TM, RM, SP, SV, RV) characteristic of the injection process, and a control unit (5) which is designed to evaluate the at least one detected profile of the at least one variable (IP, TM, RM, SP, SV, RV) characteristic of the injection process by means of an evaluation algorithm (27) trained by machine learning in order to determine a closing behavior of the check valve (20).

18. Computer program product for controlling and / or monitoring an injection molding machine, comprising commands which, when the program is executed by a computer, cause the computer to execute an evaluation algorithm (27) trained by means of machine learning methods for determining a closing behavior of a check valve (20) of a plasticizing screw (9) of an injection molding machine (1) based on an evaluation of at least one profile of at least one variable (IP, TM, RM, SP, SV, RV) characteristic of an injection process.

19. Computer program product according to claim 18, characterized in that the evaluation algorithm (27) is trained by means of training data from at least one training injection molding machine (30), wherein the training injection molding machine (30) has at least one sensor (22, 23, 24, 25) for detecting at least one profile of at least one variable (IP, TM, RM, SP, SV, RV) characteristic of an injection process for a plurality of training injection molding cycles for generating training data, and wherein the training injection molding machine (30) has an additional training sensor (31) which at least indirectly measures the closing behavior of the check valve (20) in the respective training injection molding cycle, wherein measurement data (MP) from the training sensor (31) are used for, in particular, automatic, labeling of the training data.

Images