STATE DETECTION DEVICE AND STATE DETECTION PROCEDURE

Abstract

State determination device (1) which determines an operating state of an injection molding machine (2), wherein the state determination device (1) comprises the following: a data acquisition unit (30) that receives data relating to the injection molding machine (2); a specification data storage unit (52) that stores the respective specification data of a reference injection molding machine (2) and one or more other injection molding machines (2) that differ from the reference injection molding machine (2); a numerical conversion unit (32) that converts data obtained by the data acquisition unit (30) into scale data using a conversion formula defined for each data type, based on the specification data of the reference injection molding machine (2) and the specification data of the other injection molding machine (2) stored in the specification data storage unit (52); and a learning unit (100) that performs machine learning using the scaled data obtained by conversion performed by the numerical conversion unit (32) and creates a learning model.

Description

BACKGROUND OF THE INVENTION 1. Field of the invention

[0001] The present invention relates to a condition determination device and a condition determination method, and in particular relates to a condition determination device and a condition determination method to assist in maintenance work for injection molding machines. 2. Description of the related technology

[0002] Industrial machines, such as injection molding machines, are serviced regularly or when an anomaly occurs. During maintenance, service personnel use physical parameters that represent the machine's operating conditions, primarily recorded during operation, to determine whether a condition is normal or anomalous. They then perform maintenance work, such as replacing the part where the anomaly is occurring.

[0003] For example, a known maintenance procedure for a check ring in an injection cylinder of an injection molding machine involves regularly removing a screw from the injection cylinder to directly measure the dimensions of the check ring. However, this procedure requires temporarily stopping production, thus negatively impacting productivity.

[0004] Prior art techniques for solving such a problem are known in which a torque applied to a screw is detected, and a backflow phenomenon, in which resin flows to the rear of the screw, is detected in order to indirectly detect a wear measure of a check ring of an injection cylinder and diagnose an anomaly without temporarily halting production, such as by removing a screw from the injection cylinder. For example, JP H01-168 421 A discloses a technique in which a torque acting on a screw rotation direction is measured, and an anomaly is determined if the torque is outside an acceptable range. Furthermore, JP 2017-30 221 A and JP 2017-302 632 A disclose techniques for diagnosing anomalies by supervised learning on a load on a drive unit, resin pressure, and the like.Furthermore, JP 2017-134 786 A discloses a technique in which internal pieces of information from a variety of manufacturing machines are obtained and the internal pieces of information obtained from each manufacturing machine are compared to extract a difference and detect an anomaly based on the difference.

[0005] However, the technique disclosed in JP H01-168 421 A presents a problem in that an operation to set allowed ranges used to determine anomaly is needed for machines that have different specifications, such as a rated torque and inertia of a motor that is a drive unit of an injection molding machine, and a reduction ratio of reduction gears.

[0006] Furthermore, the techniques described above, disclosed in JP 2017-30 221 A and JP 2017-202 632 A, have a problem in that the discrepancy between measured values ​​obtained from machines with different component specifications (the drive units of injection molding machines) and numerical values ​​of training data fed into the machine learning system is too large to allow for accurate diagnostics. For example, measured values ​​such as a load on a drive unit and resin pressure obtained during operation of a large injection molding machine are large, while measured values ​​such as a load on a drive unit and resin pressure obtained during operation of a small injection molding machine are small.Therefore, a problem exists in that even if an anomaly degree is estimated by directly using measured values ​​obtained during the operation of a small injection molding machine as training data, based on a learning model performed by machine learning using measured values, such as a load on a drive unit and resin pressure obtained during the operation of a large injection molding machine, as training data, the anomaly degree cannot be estimated correctly due to the influence of differences between different specifications of injection molding machines.

[0007] Furthermore, the technology disclosed in JP 2017-134 786 A presents a problem in that there is a difference between a large injection molding machine and a small injection molding machine with respect to load on drive units and measured values ​​of resin pressure obtained during operation of the machines, and therefore the presence of anomaly cannot be correctly estimated, even, for example, by comparing measured values ​​used as internal information.

[0008] By applying machine learning techniques, various types of learning conditions are prepared, as many as combinations of machine parts, such as a motor, reduction gear, and a moving unit that make up an injection molding machine. Machine learning is performed when machine learning models are created, and therefore, these learning models are assigned to machines of different sizes and equipped with different types of equipment, which allows for improved diagnostic accuracy. Preparing the corresponding components, such as motors, reduction gears, and moving units—as many combinations of machine parts—incurs high costs. Additionally, manufacturing materials, such as resin and workpieces, must be prepared when operating the machines, resulting in high costs for the materials used to obtain the learning data.Furthermore, companies require a significant amount of time to obtain training data. This leads to the problem that training data cannot be collected efficiently. DE 10 2019 124 483 A1 describes a device and a method for improving state determination for injection molding machines. EP 3 678 831 B1 describes a system and method for normalizing the control and mold cycle execution of injection molding machines. CN 1 01 398 672 A describes a learning method for improving the positioning accuracy of the mold opening and closing mechanism of an injection molding machine. DE 10 2004 041 891 B3 describes an IT integration system for production machines. An object of the invention is to propose state determination devices and state determination methods for anomaly detection using machine learning on machines with different specifications.This problem is solved by a device according to the invention according to claim 1, a device according to the invention according to claim 2, a method according to the invention according to claim 7 and a method according to the invention according to claim 8. SUMMARY OF THE INVENTION

[0009] One object of the present invention is to provide a condition determination device and a condition determination method that can help in the maintenance of various injection molding machines without incurring high costs.

[0010] According to the present invention, even if physical time-series quantities to be input into machine learning are obtained from injection molding machines with different specifications, such as a different type of power machine, a state variable derived by numerically converting physical time-series quantities (current and velocity, for example) observed based on specification data stored in an injection molding machine into physical quantities on reference scales is input as training data into machine learning to estimate an anomaly level. This solves the problems described above.

[0011] In particular, estimation of anomaly degree by applying training data obtained through numerical conversion, performed in such a way as to incorporate differences in types of machines and components, is implemented on machine learning, even if types of injection molding machines differ from each other, especially even if sizes of machines differ as a small size or a large size, and even if components of injection molding machines, such as injection devices, mold clamping devices, injection cylinders, screws and power machines, differ from each other.

[0012] Furthermore, based on an anomaly level obtained as an output from machine learning, means are provided to stop or slow down the operation of a moving unit of a machine to ensure the safety of an operator when an anomaly level exceeds a predetermined threshold, to slow down a power machine driving the moving unit to allow the moving unit to be operated safely, to limit the torque of the power machine, and to display a message or symbol indicating an anomaly state on a display device.

[0013] A state determination device according to one aspect of the present invention determines an operating state of an injection molding machine and comprises: a data acquisition unit that receives data relating to the injection molding machine; a specification data storage unit that stores respective specification data of a reference injection molding machine and one or more further injection molding machines that differ from the reference injection molding machine; a numerical conversion unit that converts data obtained by the data acquisition unit into scaled data by means of a conversion formula that is defined for each data type, using the specification data of the reference injection molding machine and the specification data of the further injection molding machine that are stored in the specification data storage unit;and a learning unit that performs machine learning using the scaled data obtained through conversion performed by the numerical conversion unit and creates a learning model.

[0014] A state determination device according to a further aspect of the present invention determines an operating state of an injection molding machine and comprises: a data acquisition unit that receives data relating to the injection molding machine; a specification data storage unit that stores respective specification data of a reference injection molding machine and one or more further injection molding machines that differ from the reference injection molding machine; a numerical conversion unit that converts data obtained by the data acquisition unit into scaled data by means of a conversion formula that is defined for each data type, using the specification data of the reference injection molding machine and the specification data of the further injection molding machine that are stored in the specification data storage unit;a learning model unit, which is performed by machine learning based on scale data relating to the reference injection molding machine; and an estimation unit, which performs estimation using the learning model stored in the learning model storage unit, based on the scale data obtained by conversion performed by the numerical conversion unit.

[0015] The learning unit can perform at least one type of learning from supervised learning, unsupervised learning, and reinforcement learning.

[0016] The estimating unit can estimate a degree of anomaly with respect to an operating condition of the injection molding machine, and the condition detection device can display a warning message on a display device when a degree of anomaly estimated by the estimating unit exceeds a predetermined threshold.

[0017] The estimating unit can estimate a degree of anomaly with respect to an operating condition of the injection molding machine, and the condition detection device can display a warning symbol on a display device when a degree of anomaly estimated by the estimating unit exceeds a predetermined threshold.

[0018] The estimating unit can estimate a degree of anomaly with respect to an operating condition of the injection molding machine, and the condition determination device can issue a command to stop or slow down an operation and / or a command to limit the torque of a power machine for the injection molding machine if a degree of anomaly estimated by the estimating unit exceeds a predetermined threshold.

[0019] A state determination method according to a further aspect of the present invention is a method for determining an operating state of an injection molding machine and comprises: a data acquisition step for obtaining data relating to the injection molding machine; a numerical conversion step for converting the data obtained in the data acquisition step into scaled-down data by means of a conversion formula defined for each data type, using the specification values ​​obtained from specification data of a reference injection molding machine and one or more other injection molding machines that differ from the reference injection molding machine; and a learning step for performing machine learning using the scaled-down data obtained by conversion in the numerical conversion step and for creating a learning model.

[0020] A state determination method according to a further aspect of the present invention is a method for determining an operating state of an injection molding machine and comprises: a data acquisition unit for obtaining data relating to the injection molding machine; a numerical conversion step for converting the data obtained in the data acquisition step into scaled data by a conversion formula defined for each data type, using the specification values ​​obtained from specification data of a reference injection molding machine and one or more other injection molding machines that differ from the reference injection molding machine;and an estimation step for estimating a state of an injection molding machine based on the scale data obtained by conversion in the numerical conversion step, using a learning model obtained by machine learning performed based on the scale data relating to the reference injection molding machine.

[0021] According to the present invention, data obtained during learning and estimation are numerically converted into scaled data, and learning or estimation processing is performed without collecting training data from different types of injection molding machines to carry out machine learning. Accordingly, different states of injection molding machines can be estimated without incurring high machine learning costs. BRIEF DESCRIPTION OF THE DRAWINGS Fig. Figure 1 is a schematic hardware configuration diagram illustrating a state determination device according to one embodiment. Fig. Figure 2 is a schematic functional block diagram illustrating the state determination device according to a first embodiment. Fig. Figure 3 illustrates an example of specification data. Fig. Figure 4 illustrates an example of a numerical conversion of torque values. Fig. Figure 5 illustrates an example of numerical conversion of injection pressure. Fig. Figure 6 is a schematic functional block diagram illustrating a state determination device according to a second embodiment during estimation. Fig. Figure 7 illustrates an example of a display indicating an anomalous condition. DETAILED DESCRIPTION OF PREFERRED EXECUTION FORMS

[0022] Fig. Figure 1 is a schematic hardware configuration diagram showing main parts of a state determination device comprising a machine learning device according to an embodiment of the present invention.

[0023] A state-determining device 1 according to the present embodiment can, for example, be installed on a controller that controls an injection molding machine. Furthermore, the state-determining device 1 can be installed as a personal computer connected to a controller that controls an injection molding machine, or as a computer such as an edge computer, a cell computer, a host computer, or a cloud server connected to the controller via a wired / wireless network. The present embodiment provides a description of an example in which the state-determining device 1 is installed as a personal computer connected to a controller that controls an injection molding machine.

[0024] A CPU 11, present in the state determination device 1 according to the present embodiment, is a processor that controls the entire state determination device 1. The CPU 11 reads a system program stored in the ROM 12 via a bus 20 and controls the entire state determination device 1 according to the system program. Transient computational data, various data types entered by a user via an input device 71, and the like are temporarily stored in a RAM 13.

[0025] A non-volatile memory 14 consists, for example, of a memory, a semiconductor drive (SSD) or the like, which is backed up by a battery (not illustrated), and a memory state of it is therefore maintained even when the state determination device 1 is switched off.The non-volatile memory 14 stores a settings area containing setting information relating to the operation of the state determination device 1, data entered by the input device 71, various data types obtained from an injection molding machine 2 (for example, the type of machine, the mass and material of a mold, and a type of resin), time-series data of various types of physical quantities (for example, the temperature of a nozzle; the position, velocity, acceleration, current, voltage, and torque of a motor driving the nozzle; the temperature of a mold; and a flow rate, flow velocity, and resin pressure) detected by the injection molding machine 2 during a molding operation, data read out via an external storage device (not illustrated) and a network, and the like.The programs and various data types stored in non-volatile memory 14 can be loaded into RAM 13 when the programs and data are executed or used. A known analysis program for analyzing various data types and a system program, which includes, for example, a program for controlling communication with a learning device 100 (described later), are provisionally written to ROM 12.

[0026] Injection molding machine 2 is a machine for manufacturing a product that is shaped using resin, such as plastic. It melts resin, which is a material, and fills a mold with the molten resin (injects the molten resin into the mold) to form the product. Injection molding machine 2 consists of various components, such as a nozzle, a motor (power unit), a transmission mechanism, reduction gears, and a moving unit. The condition of each component is detected by a sensor or similar device, and the operation of each component is controlled by a controller. Examples of power units used in injection molding machine 2 include an electric motor, a hydraulic cylinder, a hydraulic motor, and an air motor.Furthermore, examples of the translation mechanism used in injection molding machine 2 include a ball screw, a gear, a pulley and a belt.

[0027] Each piece of data read into a memory, data obtained as a result of program execution or the like, data output by the machine learning device 100, which will be described later, and the like, are output via an interface 17 to be displayed on a display device 70. Furthermore, the input device 71, which consists of a keyboard, a pointing device, or the like, transmits a command, data, and the like, based on an operation by an operator, to the CPU 11 via an interface 18.

[0028] An interface 21 is an interface for connecting the state determination device 1 and the machine learning device 100. The machine learning device 100 has a processor 101 for controlling the entire machine learning device 100, a ROM 102 that stores a system program or the like, a RAM 103 for performing temporary storage during any processing related to machine learning, and a non-volatile memory 104 that is used for storing learning models or the like.

[0029] The machine learning device 100 is capable of observing any piece of information (various data types, such as the type of injection molding machine 2, the mass and material of a mold and a type of resin; and time series data of various types of physical quantities, such as the temperature of a nozzle; the position, speed, acceleration, current, voltage and torque of a motor driving the nozzle; the temperature of a mold; and a flow rate, flow velocity and resin pressure, for example) that can be obtained by the state determination device 1 via the interface 21.Furthermore, the state determination device 1 receives processing results that are output by the machine learning device 100 via interface 21, and stores and displays the received results, and transmits the received results to other devices via a network that is not illustrated, or the like.

[0030] Fig. Figure 2 is a schematic functional block diagram illustrating the state determination device 1 and the machine learning device 100 according to a first embodiment.

[0031] The state determination device 1 according to the present embodiment has components necessary for learning, which is carried out by the machine learning device 100 (learning mode). Functional blocks that are in Fig. 2 are illustrated, are implemented when the CPU 11 is in the state determination device 1 and the processor 101 of the machine learning device 100, which are in Fig. 1 are illustrated, each system program is executed and each operation of each unit of the state determination device 1 and each unit of the machine learning device 100 is controlled.

[0032] The state determination device 1 according to the present embodiment comprises a data acquisition unit 30, a numerical conversion unit 32, and a preprocessing unit 34. The machine learning device 100, which is present in the state determination device 1, comprises a learning unit 110. Furthermore, a learning data storage unit 50, which stores learning data used for machine learning performed by the machine learning device 100, and a specification data storage unit 52, which stores specification data of injection molding machines, are provided in the non-volatile memory 14. Additionally, a learning model storage unit 130, which stores learning models generated by machine learning performed by the learning unit 110, is provided in the non-volatile memory 14 of the machine learning device 100.

[0033] The data acquisition unit 30 receives various data types input from the injection molding machine 2, the input device 71, and the like. The data acquisition unit 30 receives various data types, such as the type of injection molding machine 2, the mass and material of a mold, and the type of resin; time-series data of various physical quantities, such as the temperature of a nozzle; the position, velocity, acceleration, current, voltage, and torque of a motor driving the nozzle; the temperature of a mold; and a flow rate, flow velocity, and resin pressure. It also receives various data types, such as information regarding maintenance work for the injection molding machine 2, which is entered, for example, by an operator, and stores these data pieces in the learning data storage unit 50.The data acquisition unit 30 can receive data from other devices via an external storage device (not illustrated) or the wired / wireless network.

[0034] The numerical conversion unit 32 numerically converts data relating to injection molding machine 2, which is present in the learning data storage unit 50, using specification data stored in the specification data storage unit 52. The numerical conversion unit 32 also converts data contained in the learning unit and obtained from injection molding machine 2 into data with a scale of reference types of the injection molding machine, using a conversion formula predefined for each data type, and specification data for each type of injection molding machine stored in the specification data storage unit 52.

[0035] Fig. Figure 3 illustrates an example of specification data stored in specification data storage unit 52.

[0036] Specification data is data in which various elements of machine performance are expressed in figures (specification values) and is provided as a specification table by a machine manufacturer. The specification data storage unit 52 stores identification information by which a type of injection molding machine can be identified, and specification data from each injection molding machine in a way that links the identification information to the specification data for a reference injection molding machine (machine type D in the example in Fig. 3) and other injection molding machines. Such specification data may include: a maximum torque, a rated torque, a maximum current, a rated current, a maximum speed, a time constant for the rise time, and a motor inertia; a reduction ratio, a belt width, the number of teeth on a pulley, a ball screw diameter, and a ball screw pitch on a speed reducer; and a maximum stroke, a maximum speed, a maximum thrust, a maximum pressure, an inertia, a machine efficiency, a maximum mold clamping force, and a screw diameter of a moving unit of an injection device or a mold clamping device provided in an injection molding machine.

[0037] The numerical conversion unit 32 converts time series data relating to the injection molding machine 2 and contained in the learning data stored in the learning data storage unit 50 into scale data based on specification data corresponding to the type of injection molding machine 2 and stored in the specification data storage unit 52, and specification data of a reference type of injection molding machine, using a conversion formula defined for each type of conversion object data.

[0038] A conversion formula for each data type is provisionally defined in the non-volatile memory 14 of the state determination device 1. In general, a conversion formula for each data type can be defined as a formula for multiplying data values ​​derived from a data piece physically related to the data, or a ratio in a range between the minimum and maximum values ​​of possible data values ​​derived from the data piece in a reference injection molding machine and an injection molding machine receiving the data as the conversion object, based on a ratio between the data and the data piece physically related to the data. For example, a conversion formula for the torque of a motor driving a moving unit of an injection molding machine can be formulated with formula (1) below, which gives the maximum motor torque maxT s(the minimum motor torque is 0) of a reference injection molding machine, the maximum motor torque maxT c (the minimum motor torque is 0) of an injection molding machine that has received conversion object data, a reduction ratio Rr s the reference injection molding machine and a reduction ratio Rr c the injection molding machine that received the conversion object data is used and defined. In formula (1), T denotes c a torque value, which is the conversion object data, and T s denotes a scale torque after conversion. The diagrams in Fig. Figure 4 illustrates torque values ​​before conversion and torque values ​​after conversion (scale torque values) obtained when converting a torque value, which is data from machine type A, as in Fig. Figure 3 illustrates how the injection molding machine can be obtained using formula (1) if a reference injection molding machine is of machine type D from specification data, as shown in Fig. 3 illustrates, shows. Ts=Tc×maxTsmaxTc×RrsRrc

[0039] Furthermore, for example, a conversion formula for the injection pressure of an injection molding machine can be used with formula (2) below, which gives the maximum injection pressure maxlp s (the minimum injection pressure is 0) of a reference injection molding machine, the maximum injection pressure maxlp c (the minimum injection pressure is 0) of an injection molding machine that has received conversion object data, a screw diameter Sd s the reference injection molding machine and a screw diameter Sd c the injection molding machine that received the conversion object data (a screw diameter refers to the injection pressure per square order) is defined. In formula (2) Ip denotesc an injection pressure, which is the conversion object data, and IP s denotes a scale injection pressure after conversion. The diagrams in Fig. Figure 5 illustrates the injection pressure before conversion and the injection pressure after conversion (injection pressure on the reference scale) obtained during the conversion of injection pressure, which is data from machine type B, as shown in Fig. Figure 3 illustrates how the injection molding machine can be obtained using formula (2) in the case where a reference injection molding machine is machine type D from specification data, as shown in Fig. Figure 3 illustrates this. Conversion formulas for other data types can be defined in a similar way to the description above, taking into account the configuration of the injection molding machine. Ips=Ipc×maxIpsmaxIpc×Sds2Sdc2

[0040] Preprocessing Unit 34 creates state data for use in machine learning by Machine Learning Device 100, based on scaled data obtained through conversion by Numerical Conversion Unit 32. Preprocessing Unit 34 also creates state data obtained by converting (for example, quantifying, normalizing, sampling) data from Numerical Conversion Unit 23 into a unit format suitable for use in Machine Learning Device 100.For example, when the machine learning device 100 performs unsupervised learning, the preprocessing unit 34 creates state data S that have a predetermined format during unsupervised learning; when the machine learning device 100 performs supervised learning, the preprocessing unit 34 creates a set of state data S and label data L that have a predetermined format during supervised learning; and when the machine learning device 100 performs reinforcement learning, the preprocessing unit 34 creates a set of state data S and destination data D that have a predetermined format during reinforcement learning.

[0041] The learning unit 110 performs machine learning using state data obtained from the preprocessing unit 34, based on scaled data obtained through conversion performed by the numerical conversion unit 32. The learning unit 110 creates learning models by performing machine learning using data obtained from the injection molding machine 2, employing a known machine learning method such as unsupervised learning, supervised learning, and reinforcement learning, and stores the created learning models in the learning model storage unit 130. Examples of unsupervised learning performed by the learning unit 110 include the autoencoder method and the K-means method.Examples of supervised learning include the multilayer perceptron method, the recurrent neural network method, the long-short-term memory method, and the convolutional neural network method. Examples of reinforcement learning include Q-learning.

[0042] The learning unit 110 performs supervised learning based on state data obtained by converting training data from the injection molding machine 2, which is in a normal operating state, through the numerical conversion unit 32 and the preprocessing unit 34. Thus, the learning unit 110 is able to create, for example, a distribution of training data obtained in a normal state (and converted into scaled data) as a training model. Using the training model thus created, an estimation unit 120, which will be described later, is able to estimate how much of the training data obtained from the injection molding machine 2 differs from the training data obtained in a normal operating state, and can thus calculate an anomaly degree as an estimation result.

[0043] Furthermore, learning unit 110 performs supervised learning using data obtained by applying a normal label to data obtained from injection molding machine 2 in a normal operating state and by applying an anomalous label to data obtained from injection molding machine 2 before and after an anomaly has occurred, which allows, for example, the creation of a discrimination boundary between normal data and anomalous data as a learning model.

[0044] Using the learning model that is thus created, the estimation unit 120, which will be described later, is able to estimate whether the learning data obtained from the injection molding machine 2 belong to normal data or anomalous data, and can thus calculate a characteristic value (normal / anomalous) as an estimation result and reliability of the characteristic value.

[0045] In the state determination device 1, which has the configuration described above, training data obtained from the injection molding machine 2 are converted into data on the scale of a reference injection molding machine, and the training unit 110 performs training using the data obtained through this conversion. Thus, the data used for training by the training unit 110 are based on scale data obtained through conversion by the numerical conversion unit 32, such that a training model created by the training unit 110 is used to estimate a data value that varies within a range of data that may originate from a reference injection molding machine.

[0046] Fig. Figure 6 is a schematic functional block diagram illustrating the state determination device 1 and the machine learning device 100 according to a second embodiment.

[0047] The state determination device 1 according to the present embodiment has the configuration necessary for estimation, which is performed by the machine learning device 100 (estimation mode). Functional blocks that are in Fig. Figure 6 illustrates that the following are implemented when the CPU 11 is in the state determination device 1 and the processor 101 is in the machine learning device 100, which is in Fig. 1 are illustrated, each system program is executed and each operation of each unit of the state determination device 1 and each unit of the machine learning device 100 is controlled.

[0048] The state determination device 1 according to the present embodiment comprises the data acquisition unit 30, the numerical conversion unit 32, the preprocessing unit 34, and the machine learning device 100, as is the case in the first embodiment. The machine learning device 100, which is included in the state determination device 1, comprises the estimation unit 120. Furthermore, a learning data storage unit 50, which stores learning data used for state estimation performed by the machine learning device 100, and the specification data storage unit 52, which stores specification data of injection molding machines, are provided in the non-volatile memory 14. The learning model storage unit 130, which stores learning models created by machine learning performed by the learning unit 110, is also provided in the non-volatile memory 14 of the machine learning device 100.

[0049] The data acquisition unit 30 and the numerical conversion unit 32 according to the present embodiment each have similar functions to the data acquisition unit 30 and the numerical conversion unit 23 according to the first embodiment.

[0050] The preprocessing unit 34, according to the present embodiment, creates data to be used for estimation performed by the machine learning device 100, based on data obtained by converting the training data stored in the training data storage unit 50 into scaled data by the numerical conversion unit 32. The preprocessing unit 34 creates state data obtained by converting (for example, quantifying, normalizing, sampling) the obtained data into data having a unit format to be used in the machine learning device 100. The preprocessing unit 34 creates state data S, having a predetermined format, during estimation performed by the machine learning device 100.

[0051] Estimation Unit 120 estimates the state of an injection molding machine using a learning model stored in the learning model storage unit 130, based on state data S generated by the preprocessing unit 34 using unlabeled data. In Estimation Unit 120, the state data S input from the preprocessing unit 34 is fed into the learning model (by which parameters are determined) created by the learning unit 110 to estimate and calculate the degree of anomaly with respect to the state of the injection molding machine and to estimate and calculate the class (for example, normal / anomalous) to which the injection molding machine belongs.Results obtained by estimation using the estimation unit 120 (for example, an anomaly degree with respect to a state of an injection molding machine and a class to which an operating state of the injection molding machine belongs) can be displayed on and output to the display device 70 and can be transmitted via a wired / wireless network to and output to a host computer, a cloud computer, and the like (not illustrated) for use.

[0052] Furthermore, if a result estimated by the estimation unit 120 is in a predetermined state (for example, if an anomaly level estimated by the estimation unit 120 exceeds a predetermined threshold, or if a class to which an operating state of an injection molding machine belongs and which is estimated by the estimation unit 120 is "anomalous"), the state determination device 1 can display an output on the display device 70, for example with a warning message or a warning symbol, as shown in Fig. 7 illustrates, performs, or can issue a command to stop or slow down an operation, a command to limit the torque of a motor, or the like to the injection molding machine.

[0053] In the state determination device 1, which has the configuration described above, training data obtained from the injection molding machine 2 are converted into data on the scale of a reference injection molding machine. The estimation unit 120 then performs an estimation of the state of the injection molding machine 2 using the data obtained through this conversion. A training model stored in the training model storage unit 130 is used to estimate a data value that varies within a range of data that can be obtained from a reference injection molding machine. However, even training data obtained from injection molding machines of different types from the reference injection molding machine is converted into scale data by the numerical conversion unit 32 before the estimation unit 120 performs the estimation. Accordingly, estimation processing can be carried out appropriately.

[0054] The embodiments according to the present invention have been described so far. However, the present invention is not limited to the examples of the embodiments described above and can be implemented by appropriately adding modifications in various aspects.

[0055] For example, the above embodiments provide the description in which the state detection device 1 and the machine learning device 100 are devices that have different CPUs (processors), but the machine learning device 100 can be implemented by the CPU 11 contained in the state detection device 1 and the system program stored in the ROM 12.

Claims

[1] State determination device (1) which determines an operating state of an injection molding machine (2), wherein the state determination device (1) comprises the following: a data acquisition unit (30) that receives data relating to the injection molding machine (2); a specification data storage unit (52) that stores the respective specification data of a reference injection molding machine (2) and one or more other injection molding machines (2) that differ from the reference injection molding machine (2); a numerical conversion unit (32) that converts data obtained by the data acquisition unit (30) into scale data using a conversion formula defined for each data type, based on the specification data of the reference injection molding machine (2) and the specification data of the other injection molding machine (2) stored in the specification data storage unit (52); and a learning unit (100) that performs machine learning using the scaled data obtained by conversion performed by the numerical conversion unit (32) and creates a learning model. [2] State determination device (1) which determines an operating state of an injection molding machine (2), wherein the state determination device comprises the following: a data acquisition unit (30) that receives data relating to the injection molding machine (2); a specification data storage unit (52) that stores the respective specification data of a reference injection molding machine (2) and one or more other injection molding machines (2) that differ from the reference injection molding machine (2); a numerical conversion unit (32) that converts data obtained by the data acquisition unit (30) into scale data using a conversion formula defined for each data type, based on the specification data of the reference injection molding machine (2) and the specification data of the other injection molding machine (2) stored in the specification data storage unit (52); and a learning model storage unit (130) that stores a learning model obtained through machine learning based on scale data relating to the reference injection molding machine (2); and an estimation unit (120) performs the estimation using the learning model stored in the learning model storage unit (130) based on the scale data obtained by conversion performed by the numerical conversion unit (32). [3] State determination device (1) according to claim 1, wherein the learning unit (100) performs at least one learning from supervised learning, unsupervised learning and reinforcement learning. [4] State determination device (1) according to claim 2, wherein the estimation unit (120) estimates a degree of anomaly with respect to an operating state of the injection molding machine (2), and The condition detection device (1) displays a warning message on a display device when an anomaly level estimated by the estimating unit exceeds a predetermined threshold. [5] State determination device (1) according to claim 2, wherein the estimation unit (120) estimates a degree of anomaly with respect to an operating state of the injection molding machine (1), and The condition detection device (1) displays a warning symbol on a display device when an anomaly level estimated by the estimation unit (120) exceeds a predetermined threshold. [6] State determination device according to claim 2, wherein the estimation unit (120) estimates a degree of anomaly with respect to an operating state of the injection molding machine (2), and The state determination device (1) issues a command to stop or slow down an operation and / or a command to limit a torque of a power unit for the injection molding machine (2) when an anomaly level estimated by the estimation unit (120) exceeds a predetermined threshold. [7] State determination method for determining an operating state of an injection molding machine (2), wherein the state determination method comprises the following: a data acquisition step to obtain data relating to the injection molding machine (2); a numerical conversion step for converting data obtained in the data acquisition step into scaled data by a conversion formula defined for each data type, using the specification values ​​retrieved from specification data of a reference injection molding machine (2) and one or more other injection molding machines (2) that differ from the reference injection molding machine (2); and a learning step to perform machine learning using the scaled data obtained by conversion in the numerical conversion step, and to create a learning model. [8] State determination method for determining an operating state of an injection molding machine (2), wherein the state determination method comprises the following: a data acquisition step to obtain data relating to the injection molding machine (2); a numerical conversion step for converting data obtained in the data acquisition step into scaled data by a conversion formula defined for each data type, using the specification values ​​retrieved from specification data of a reference injection molding machine (2) and one or more other injection molding machines (2) that differ from the reference injection molding machine (2); and an estimation step for estimating a state of an injection molding machine (2) based on the scale data obtained by conversion in the numerical conversion step, using a learning model obtained by machine learning performed on the scale data relating to the reference injection molding machine (2).

Images