JP7099531B2 - Methods and devices for controlling the operation of mechanical devices and determining the reliability of data - Google Patents

Description

The present disclosure relates to the field of automated control in mechanical engineering, specifically, methods for controlling the operation of mechanical devices, methods for determining the reliability of output data of machine learning models, devices and systems, and training data. Regarding processing methods and equipment.

In the field of machine control, recently, a technique for performing optimum control based on various machine control parameters is being developed at an astonishing speed. For example, in Document 1 ( JP2017-68 325A ), there is a technical solution in which an acceleration / deceleration operation of each axis of a mechanical device is generated by using a neural network. However, Document 1 only discloses that the movement of each shaft is determined by machine learning, and the reliability of the movement is unknown.

No effective solution has been proposed for the problem that the reliability of the control parameter of the mechanical device cannot be determined in the related technique.

In embodiments of the present disclosure, a method of controlling the operation of a mechanical device, a method of determining the reliability of data, a device and a system, and a method and a device for processing training data are provided, and at least the mechanical device is conventionally used. It solves the technical problem that the user cannot judge the reliability of the control parameter of.

According to one embodiment of the present disclosure, a method of controlling the operation of a mechanical device is provided, wherein the method receives input data including at least one measurement parameter of the mechanical device and data of training data. Determining the data area corresponding to the input data from the area, determining the reliability of the output data corresponding to the input data based on the data distribution characteristics of the data area corresponding to the input data, and trusting the output data. Includes controlling the operation of mechanical devices based on gender.

In the method, the reliability of the output data of the trained machine learning model can be determined based on the distribution of the data area corresponding to the input data. Therefore, the mechanical device can know the reliability of the output data of the machine learning model, thereby performing the corresponding operation according to the reliability and reducing the error rate of the determination or control using the output data. It is possible to improve the robustness of the mechanical device.

In one embodiment of the present disclosure, determining the data area corresponding to the input data from the data area of the training data divides the data area of the training data into a plurality of data areas and corresponds to the input data from the plurality of data areas. It includes determining the data area to be used, or determining the proximity area from the data area of the training data by proximity learning, and setting the proximity area as the data area corresponding to the input data. Therefore, the data area corresponding to the input data can be quickly determined.

In one embodiment of the present disclosure, controlling the operation of a mechanical device according to the reliability of the output data comprises controlling the bottom dead point of the mechanical device according to the reliability of the output data. It is a press machine. Therefore, the dynamic accuracy of bottom dead center is improved.

In one embodiment of the disclosure, the reliability of the output data is the probability that the machine learning model trained by the training data will produce the desired output data by using the input data.

In one embodiment of the present disclosure, the input data is the data input to the machine learning model obtained by machine training to determine the output data, and the training data is the data for training the machine learning model. be.

In one embodiment of the present disclosure, the data distribution characteristic comprises the amount or intensity of training data in the data area corresponding to the input data.

In one embodiment of the present disclosure, determining the reliability of the output data based on the data distribution characteristics of the data area corresponding to the input data means that the amount or intensity of the training data in the data area corresponding to the input data. If it is larger than the first predetermined threshold, it is judged to be highly reliable, and if not, it is judged to be unreliable, or all the distances between the parts of the training data in the data area corresponding to the input data are all. If it is less than the second predetermined threshold, it is determined that the reliability is high, and if not, it is determined that the reliability is low. Therefore, by dividing the above-mentioned data area, the reliability of the output data corresponding to the input data can be easily known, and the accurate reliability of the output data for each input data is calculated using a complicated algorithm. It does not need to be done and reduces the operating load of the system.

In one embodiment of the present disclosure, controlling the operation of a mechanical device according to the reliability of the output data is more reliable than controlling the operation of the mechanical device based on the output data when the reliability is high. If low, it involves setting the output data to predetermined data and controlling the operation of the mechanical device based on the set output data, where the predetermined data is an operating parameter applicable to the operation of the mechanical device. Or, it is data instructing the mechanical device to end the operation. Therefore, it is possible to control the mechanical device more reliably and improve the robustness of the entire system.

In one embodiment of the present disclosure, controlling the operation of a machine is writing a log based on the output data, tracking the operation history of the machine, and controlling the machine or controlling the machine based on the output data. To perform process control or switch to artificial control (Human control), where mechanical control includes at least one of stop, deceleration and start, and to improve control of the mechanical device by additional learning. It is to generate training data based on the output data.

According to one embodiment of the present disclosure, a method for determining the reliability of data is provided, in which the method determines from the data area of training data the data area corresponding to the input data and the input data. Includes determining the reliability of the output data corresponding to the input data based on the data distribution characteristics of the corresponding data area. Therefore, the operator or operating device can know the reliability of the output data of the machine learning model, thereby performing the corresponding operation according to the reliability, and as a result, the determination or control using the output data. The error rate can be reduced and the robustness of the entire system can be improved.

According to another aspect of the embodiments of the present disclosure, a method of processing training data is further provided, wherein the method is to acquire training data and to divide the acquired training data into a plurality of areas, respectively. The training data is contained within the corresponding data area and generates a data distribution characteristic for each data area in the plurality of data areas, and the data distribution characteristic corresponds to the input data contained in one of the plurality of data areas. Includes being used to determine the reliability of output data.

By dividing the training data into multiple data areas, the distribution of the training data can be known in the training process of the machine learning model, so the reliability of the output data can be judged based on the distribution of the training data. When a trained machine learning model is used, it provides the basis for subsequent movements.

In one embodiment of the present disclosure, the data distribution characteristics of each data region include the amount or intensity of training data within each data region.

In one embodiment of the present disclosure, the reliability of the output data is determined by determining whether the amount or intensity of the training data in one data area in the plurality of data areas is larger than the first predetermined threshold value.

In one embodiment of the present disclosure, the method further determines if the distances between portions of training data in one data area within a plurality of data areas are all less than a second threshold of the output data. Includes determining reliability.

According to still another aspect of the embodiments of the present disclosure, an apparatus for determining the reliability of data is further provided, the apparatus from an input unit configured to receive input data and a data area of training data. It is configured to determine the reliability of the output data corresponding to the input data based on the data distribution characteristics of the data area corresponding to the input data and the calculation unit configured to determine the data area corresponding to the input data. Including the estimation part.

Therefore, the operator or operating device can know the reliability of the output data of the machine learning model, thereby performing the corresponding operation according to the reliability, and as a result, the determination or control using the output data. The error rate can be reduced and the robustness of the entire system can be improved.

In one embodiment of the present disclosure, the apparatus further includes a control unit, which is configured to control the operation of the controlled object according to the reliability of the output data.

In one embodiment of the present disclosure, the data distribution characteristic comprises the amount or intensity of training data in the data area corresponding to the input data, and the estimator further comprises the amount of training data in the data area corresponding to the input data . Alternatively, if the intensity is greater than the first predetermined threshold, it is determined to be reliable, otherwise it is determined to be unreliable, or between parts of the training data in the data area corresponding to the input data. If all the distances are less than the second threshold, it is determined that the reliability is high, and if not, it is determined that the reliability is low.

According to still another aspect of the embodiments of the present disclosure, an apparatus for processing training data is further provided, wherein the apparatus includes an acquisition unit configured to acquire training data and a plurality of acquired training data. Each training data is configured to be divided into regions of, and each training data is configured to generate a data distribution characteristic for each data region of a plurality of data regions and a division portion contained in the corresponding data region. Includes a generator used to determine the reliability of the output data corresponding to the input data contained in one of the plurality of data areas.

By dividing the training data, the distribution of the training data can be known in the training process of the machine learning model, and when the trained machine learning model is used, the reliability of the output data is determined based on the distribution of the training data. It can be determined, which provides the basis for subsequent actions.

According to still another aspect of the embodiments of the present disclosure, an apparatus for determining the reliability of data in any one of the above-mentioned technical solutions, an apparatus for processing training data for the above-mentioned technical solutions, and the like. Is provided in embodiments of the present disclosure.

According to yet another aspect of the present disclosure, a computer program is further provided, which, when executed by a processor, implements any one of the technical solutions.

According to yet another aspect of the present disclosure, a computer readable storage medium is further provided, said computer readable storage medium, which, when executed by a processor, is a computer that performs any one of the above technical solutions. Remember the program.

In the method of controlling the operation of the machine provided by the embodiments of the present disclosure, the reliability of the output data corresponding to the input data is determined based on the data distribution characteristics of the input data and the training data, and the operation of the machine device is , It is controlled based on reliability, thereby solving the problem that the reliability of the control parameter of the mechanical device cannot be determined in the past, and the reliability of the control parameter of the mechanical device can be improved. It has the effect of.

Furthermore, in a device that controls using the output data of a trained machine learning model, the reliability of the output data that is the basis of the control can be known, so that highly reliable control can be performed, which exceeds a predetermined range. No operation occurs, and control efficiency and robustness can be improved.

It is a flowchart of the method of controlling the operation of the press machine by embodiment of this disclosure. It is a structural schematic diagram which shows the press machine control system to which the reliability of the output data by one Embodiment of this disclosure is applied. It is a schematic diagram which shows the transition of the control operation of the control system of the press machine shown in FIG. It is a three-dimensional schematic diagram corresponding to the measured parameter and the target parameter of the system for controlling the press machine shown in FIG. 2 according to the embodiment of the present disclosure. It is a two-dimensional schematic diagram of the parameter measured and the target parameter by the embodiment of this disclosure. It is a schematic diagram which shows the transition of the control operation of the control system of the press machine shown in FIG. It is a schematic diagram of an example of a PC (Personnel Computer) 600 as a part of a hardware configuration of a system for determining the reliability of data according to an embodiment of the present disclosure. It is a structural schematic diagram which shows the apparatus which determines the reliability of the data by an embodiment of this disclosure. It is a structural schematic diagram which shows the apparatus which processes the training data by one Embodiment of this disclosure. It is a structural schematic diagram which shows the system which determines the reliability of the data by an embodiment of this disclosure.

In order for those skilled in the art to better understand the present disclosure, embodiments of the present disclosure will be described below clearly and completely in conjunction with the drawings of the present disclosure. Of course, the embodiments described are only a part of the embodiments of the present disclosure and not all of the embodiments. All other embodiments that can be obtained by one of ordinary skill in the art based on the embodiments of the present disclosure without any inventive effort are included within the scope of this disclosure.

FIG. 1 is a flowchart of a method for controlling the operation of a press machine according to the embodiment of the present disclosure. As shown in FIG. 1, this method includes the following steps.

In step S10, input data including the measured torque and the measured position is received.

The machine learning model receives the input data. The input data is the data input to the machine learning model acquired by machine training to determine the output data.

The training process of a machine learning model involves providing training data to train a machine learning algorithm. The term machine learning model refers to a model artifact created by the training process. To train a machine learning model, the user needs to specify the following: Training Data source inputs, names of data attributes including expected targets, required data transformation instructions, and training parameters to control machine learning algorithms. During the training process, the correct learning algorithm is automatically or manually selected based on the type of target specified by the user in the training data source.

Machine learning algorithms can be divided into three major categories: supervised learning, unsupervised learning, and reinforcement learning. The supervised learning adopted in this disclosure is useful when labels are available for a particular data set. Decision trees, naive Bayes classifiers, ordinary least squares regression, and logistic regression are algorithms for supervised learning.

This disclosure does not focus on how to train machine learning models and therefore does not provide further explanation.

In step S12, the data area corresponding to the input data is determined from the data area of the training data.

All training data for training the machine learning model is stored in a memory device having sufficient capacity. Training data is data for training a machine learning model.

The stored training data for training the machine learning model is divided into a plurality of data areas. All training data may be divided according to different division strategies to obtain multiple data areas. For example, training data can be randomly divided into n subsets that are mutually exclusive. Alternatively, the space in which the training data is arranged can be divided into n spatial subsets of the same size or different sizes. Of course, in other embodiments, other methods of partitioning the training data may be used. The description of how to divide the training data in the embodiments of the present disclosure is not intended to limit the method of dividing the training data.

Based on the input data, the data area corresponding to the input data is determined from the plurality of data areas. In most cases, the input data is necessarily contained in one of a plurality of data areas divided according to the training data. In the rare case where the input data is not included in the divided data area, it can be determined that the reliability of the output data corresponding to the input data is directly low.

In one embodiment, the training data can be divided into only one data area, i.e. all the training data are considered as one set, and the input data is necessarily contained within this data area. When calculating the reliability of the output data, the reliability of the output data can be determined by calculating the amount or intensity of the training data closest to the input data by proximity learning.

In step S14, the reliability of the output data corresponding to the input data is determined based on the data distribution characteristics of the data area corresponding to the input data.

The data distribution characteristics include the amount or intensity of training data in the data area corresponding to the input data. When the input data is input, the reliability of the output data corresponding to the input data can be determined based on the amount or strength of the training data in the data area corresponding to the input data. In particular, the amount or intensity of training data in the data area containing the input data can be calculated with reference to the data distribution table in order to estimate the reliability of the output data. Alternatively, the reliability of the output data can be estimated based on the distance from the training data closest to the input data to the portion of the proximity training data.

If the amount or intensity of the training data in the data area corresponding to the input data is large, the reliability of the output data is determined to be high, otherwise the reliability of the output data is determined to be low. Alternatively, if the distance between the training data portions of the data area corresponding to the input data is short, the reliability of the output data is determined to be high, and if not, the reliability of the output data is determined to be low.

In step S16, the operation of the press machine is controlled according to the reliability of the output data.

After the reliability of the output data is determined, the corresponding process can be performed on the press machine depending on the reliability of the output data. For example, if the output data is reliable, the bottom dead center of the press machine is controlled based on the current output data. However, if the reliability is low, that is, the current output data may increase the error rate of the subsequent process, it is necessary to replace the output data with the predetermined data, and the bottom dead center of the press machine is the predetermined data. It is controlled based on.

Not limited to press machines, various operations can be controlled by using the reliability of output data. For example, writing logs based on output data, tracking the operation history of controlled objects, and based on output data. Then perform machine control or process control on the controlled object or switch to artificial control, where the machine control includes at least one of stop, deceleration, and start, and the controlled object by additional learning. For example, to generate training data based on output data in order to improve control.

Of course, the reliability of the output data can also be used in many other scenarios. For example, when a user searches for a keyword, the search engine can use the reliability of the output data to determine which results are suitable for the user and which ads are suitable for the user. For online shopping, online shopping websites can use the reliability of the output data to recommend products to users for selection. In summary, the reliability of the output data can be applied in various fields, and the application scenario of the reliability of the output data is not particularly limited in this disclosure.

FIG. 2 is a schematic structural diagram showing a press machine control system to which the reliability of output data according to one embodiment of the present disclosure is applied. FIG. 3 is a schematic diagram showing a transition of the control operation of the control system of the press machine shown in FIG. As shown in FIG. 2, the system for controlling the press machine includes a control device (PLC) 20, a torque detection unit 21, a position detection unit 22, a servo driver 23, a servomotor 24, a press mechanism 25, and training data storage. Including the unit 26, the control device 20 includes an input data reliability calculation unit 202, a reliability estimation unit 204, and a control unit 206.

The torque detection unit 21, which can be a load cell, is configured to detect the torque of the press machine and obtain the actually measured torque. The position detection unit 22, which can be a position sensor, is configured to detect the position of the press machine and obtain the actually measured position.

The torque detection unit 21 and the position detection unit 22 transmit the acquired actually measured torque and the actually measured position to the control device 20, respectively.

The control device 20 acquires training data from the training data storage unit 26, divides the training data into a plurality of data areas, and divides the training data into a plurality of blocks, for example, as shown in FIG. 4A. Then, the control device 20 calculates the data distribution characteristic for each data area. In one embodiment, in order to improve the performance of the control device 20, the data area can be divided in advance, and the data distribution characteristic can be calculated in advance for each data area.

After receiving the actually measured torque and the actually measured position, the input data reliability calculation unit 202 of the control device 20 receives the actually measured torque and the actually measured position in a plurality of data areas. The data area corresponding to is, i.e., the corresponding block in the matrix, and further determines the data distribution characteristics of the determined data area, eg, calculates the intensity of the training data in the corresponding block.

The reliability estimation unit 204 determines the reliability of the target torque and the target position (that is, also referred to as output data and target parameter) based on the determined data distribution characteristic. Specifically, with reference to FIG. 4A, the X1 axis represents the measured torque, the X2 axis represents the measured position, and the Y axis represents the target parameter f (x1, x2). The control unit 206 calculates the target parameters f (x1, x2), that is, the target torque and the target position in the next period when the bottom dead center is reached, using the machine learning model. The reliability estimation unit 204 determines the corresponding region in the X1X2 plane according to the input measurement torque and measurement position. When the density of the training data in the determination area is less than a predetermined value, it is determined that the reliability of the calculated target parameter f (x1, x2) is low. In this case, the control unit 206 controls the press machine by using predetermined torque parameters and position parameters (for example, default values of torque and position) instead of the target torque and target position from the control unit 206. In one embodiment, alarms may be provided when reliability is low. In another embodiment, the actually measured torque and the actually measured position may be stored and trained to improve the machine learning model.

In this embodiment, as shown in FIG. 4A, there are two measured parameters, namely the measured torque and the measured position. In other embodiments, there may be one or more measured parameters. As shown in FIG. 4B, when there is only one measured parameter, the X-axis represents the measured parameter, the Y-axis represents the target parameter, and the plurality of spaced segments on the X-axis are divided into multiple parts. Represents a data area. The reliability of the target parameter f (x) can be determined based on the density of training data in the interval segment corresponding to the input measurement parameter.

If the target torque and target position are unreliable, the servo driver 23 controls the servomotor 24 using default torque and position parameters. Otherwise, the servo driver 23 controls the servomotor 24 using the calculated target torque and target position for the next period after reaching bottom dead center. The servomotor 24 further controls the press mechanism 25 so that the value of the difference between the bottom dead center of each cycle of the press mechanism 25 and the actually measured position becomes 0 (see FIG. 3). The accuracy of bottom dead center, which affects the accuracy of workpieces, is an important technical index for press machines. In the present embodiment, the accuracy of the bottom dead center is ensured according to the reliability of the target torque and the target position, thereby ensuring a high yield of the product manufactured by the press machine.

It should be noted that the method of determining the reliability of the output data can be applied to other scenes in addition to the press control system. In other embodiments, the reliability of the output data can also be applied to multiple areas such as autonomous driving, healthcare, retail, aerospace, and transportation.

In the exemplary embodiments of the present disclosure, the reliability of the output data can be applied to automated driving systems. Automated driving systems can include central processing devices, braking systems, acceleration systems, steering systems, navigation systems and sensing systems. The navigation system receives data about geolocation information (eg GPS data, the received data can be used to determine the current position of the vehicle), and the current vehicle position and user-configured. It is used to determine the overall driving line of the vehicle based on the target position. The sensing system includes one or more sensors configured to sense sensing information such as front, rear, left and right obstacles, traffic signals in front of the vehicle, and road signs on the front right side of the vehicle. Send information to the central processing unit. The central processing unit generates a control instruction based on the received sensing information, and determines whether the control instruction is reliable by using a method for determining reliability. When the control commands are relatively reliable, the central processing unit uses the control commands to control the braking system, steering system, acceleration system, etc., that is, the vehicle using the reliable control commands. It controls various parts of the vehicle and controls the direction and speed of the vehicle. Although it is a vehicle in this embodiment, it can include all kinds of vehicles such as automobiles, ships, airplanes, and trains. In the present embodiment, the method of determining the reliability of the data can operate the vehicle more accurately according to the result derived by the calculation.

In another embodiment, the method of determining reliability can be applied in the field of health care, such as drug discovery, genetic testing, personalized health care, or precision surgery. In this embodiment, surgery is taken as an example. In clinical surgery, it is necessary to perform real-time interactive quantitative analysis of 3D volume, distance, angle, blood vessel diameter, etc. of human organs using images in order to perform complete quantitative 3D evaluation before surgery. be. However, in reality, there may be a deviation in the accuracy of such three-dimensional evaluation. The method of determining the reliability of the output data is applied to the 3D evaluation of the organ, so that which part of the 3D data output using the image data is relatively reliable and which part is reliable. It is possible to determine whether the sex is relatively low. Specifically, for human organ data (ie, input data) acquired using images, if the training data in the area containing that data is dense, the output data will almost always be different. The output data is highly reliable. In another aspect, when the training data in the area containing the data is small, the output data is in many different situations, i.e., the output data is unreliable. If the reliability is low, the output of the machine (eg, surgical robot) can be unexpected and it is dangerous to use that output as the operation of the machine. To avoid such risks, if the output 3D data is unreliable, ask the doctor to make a final decision, thereby generating accurate 3D data and making the surgery more reliable. It can be done quickly, accurately and safely.

A device that determines the reliability of data, a device that processes training data, and a system that determines the reliability of data or a part thereof when realized in the form of a software functional unit and sold or used as an individual product. Can be stored on a computer-readable storage medium. Based on this understanding, the technical solutions disclosed are essentially or part of a technical solution that contributes to the prior art, or the entire technical solution or part of a technical solution is a software product. Such computer software programs can be embodied in the form of, stored in a storage medium, and in a computer device (which can be a PC computer, server, or Internet device) a method according to each embodiment of the present disclosure. Includes several instructions that make the entire step or part of a step executable. The storage medium includes various media capable of storing program codes such as a USB flash disk, read-only memory (ROM), random access memory (RAM), mobile hard disk, magnetic disk or compact disk, and is a server or a cloud. It can also include data flows that can be downloaded from.

FIG. 5 is a flowchart of a method of processing training data according to an embodiment of the present disclosure and a method of determining the reliability of data by using the training data. As shown in FIG. 5, the method comprises the following steps.

In step S50, data training is acquired.

Training data is data that trains a machine learning model. The characteristics and quantity of training data are key factors in determining how good the performance of a trained machine learning model is. In general, all training data for training a machine learning model is stored in a memory device having sufficient capacity, and therefore all training data can be obtained from the memory device.

In step S52, the training data is divided into a plurality of data areas.

Divide the stored training data for training a machine learning model into multiple data areas. All training data may be divided according to different division strategies to obtain multiple data areas. For example, the training data can be randomly divided into n subsets that are mutually exclusive. Alternatively, the space in which the training data is arranged can be divided into n spatial subsets of the same size or different sizes. Here, the space includes a one-dimensional space, a two-dimensional space, and a three-dimensional space.

In step S54, the data distribution characteristic is calculated for each region.

Generate a data distribution table for each data area. The data distribution characteristics of each data area in a plurality of divided data areas can be calculated according to each data distribution table. Data distribution characteristics can be defined by the following aspects: Central tendency of distribution that reflects the range (extent) of each training data that is close to or aggregates toward the center value of the data area; extend); The shape of the distribution that reflects the skewness and kurtosis of the data distribution. In one embodiment, the data distribution characteristic can further include the amount or intensity of training data in the data area.

In step S56, it is determined whether or not the calculation of the data distribution characteristics of all the data areas is completed.

It is determined whether or not the calculation of the data distribution characteristics of all the data areas is completed, and if it is completed, step 58 is executed. If not, the process is returned to step 56.

In step S58, the reliability of each area is saved.

Store the reliability of each area in a corresponding memory device with sufficient capacity.

In step S60, input data is acquired.

The input data is the data input to the machine learning model obtained by machine training to determine the output data. In one embodiment, the input data may be sensor data acquired from a device of the Internet of Things (abbreviated as "IOT").

In step S62, the data distribution characteristic of the data area corresponding to the input data is determined.

After the input data is input, the input data is necessarily included in one of the plurality of data areas divided according to the training data, and the data area corresponding to the input data can be determined. .. Then, the data distribution characteristic of the data area corresponding to the input data is determined based on the data distribution characteristic of each data area stored in step 28.

In step S64, the reliability of the output data is estimated.

The reliability of the output data corresponding to the input data is determined based on the amount or strength of the training data in the data area corresponding to the input data. When the amount or intensity of the training data in the data area corresponding to the input data is large, it is judged that the reliability of the output data is high. If not, it is determined that the reliability of the output data is low. Alternatively, when the distance between the parts of the training data in the data area corresponding to the input data is short, it is determined that the reliability of the output data is high. If not, it is determined that the reliability of the output data is low. In one embodiment of the IOT, the output data is a parameter with respect to the input data to allow the machine to operate properly.

In step S66 , determination and control are performed according to the reliability of the output data.

If the output data is reliable, it can be controlled directly based on the current output data. However, if the reliability is low, control the machine with the default parameters or stop the machine.

FIG. 6 is a schematic diagram of an example of a PC (Personnel Computer) 600 as a part of a hardware configuration of a system for determining the reliability of data according to the embodiment of the present disclosure. As shown in FIG. 6, the PC 600 has a CPU 610 for executing overall control, a read-only memory (ROM) 620 for storing system software, and a random access memory (RAM) for storing write / read data. ) 630, a storage unit 640 for storing various programs and data, an input / output unit 650 used as an input / output interface, and a communication unit 660 for realizing a communication function can be included. Alternatively, the CPU 610 can be replaced with a processor such as, for example, a microprocessor MCU or a field programmable gate array FRGA. The input / output unit 650 includes various interfaces such as an input / output interface (I / O interface), a universal serial bus (USB) port (which can be included as one of the ports of the I / O interface), and a network. Can include. Those skilled in the art can understand that the structure shown in FIG. 6 is merely an example and does not limit the hardware configuration of the system for determining the reliability of data. For example, the PC 600 may further include more or less components than those shown in FIG. 6, or may have a different configuration than that shown in FIG.

It should be noted that the CPU 610 described may include one or more processors, and the one or more processors and / or other data processing circuits in the present disclosure generally refer to "data processing circuits". Can be pointed to. The data processing circuit can be fully or partially embodied as software, hardware, firmware or any other combination. In addition, the data processing circuit can be a single independent processing module or can be integrated in whole or in part into any one of the other components within the PC600.

The storage unit 640 can be used to store application software and software programs of modules as a program command storage device / program data storage device described later in the present disclosure, and the program command storage device / program data storage device can be used as a data storage device. Corresponds to the method of determining the reliability of. The CPU 610 operates a software program and a module stored in the storage unit 640 to carry out the above-mentioned method for determining the reliability of data. The storage unit 640 can include one or more non-volatile memories such as magnetic memory, flash memory, or other non-volatile solid state memory. In some examples, the storage unit 640 may further include memory provided remotely to the CPU 610, which remote memory can be connected to the PC 600 via a network. Examples of such networks include, but are not limited to, the Internet, intranets, LANs, mobile communication networks, and combinations thereof.

The communication unit 660 is used to receive or transmit data via the network. Specific examples of the network may include a wireless network provided by the communication provider of the PC 600. In one example, the communication unit 660 includes a network interface controller (NIC), which can be connected to other network devices by a base station to communicate with the Internet. In one example, the communication unit 660 can be a radio frequency (RF) module that wirelessly communicates with the Internet.

FIG. 7 is a schematic structural diagram showing an apparatus for determining the reliability of data according to the embodiment of the present disclosure. The device includes an input unit 70 configured to receive input data. The reliability of the output data corresponding to the input data based on the calculation unit 72 configured to determine the data area corresponding to the input data from the data area of the training data and the data distribution characteristics of the data area corresponding to the input data. Includes an estimation unit 74 configured to determine.

FIG. 8 is a schematic structural diagram showing an apparatus for processing training data according to the embodiment of the present disclosure. As shown in FIG. 8, the apparatus is configured to divide the acquired training data into a plurality of data areas, where each training data is divided into corresponding data areas with a division unit 82. To generate data distribution characteristics for each data area within multiple data areas to determine the reliability of the output data corresponding to the input data contained within the corresponding data areas within multiple data areas. The generation unit 84, which is configured in the above, is included.

FIG. 9 is a schematic structural diagram showing a system for determining the reliability of data according to the embodiment of the present disclosure. As shown in FIG. 9, the system includes a device 90 for determining the reliability of data and a device 92 for processing training data. The device 90 for determining the reliability of the data can be the device according to FIG. 7, and the device 92 for processing the training data can be the device according to FIG. 8, and further explanation is not necessary.

In one embodiment, the device 92 that processes the training data in the system shown in FIG. 9 is configured to learn the output data corresponding to the input data by machine learning in order to further guarantee the quality of the machine learning model. Further includes a learning unit 922, a training data, and a storage unit 924 for storing a data distribution table generated based on the training data. In one embodiment, the system of FIG. 9 may further include a control unit configured to control the controlled object according to the determined reliability of the output data.

The above is only a preferred embodiment of the present disclosure. For those skilled in the art, various improvements and changes can be made without departing from the principles of the present disclosure, and such improvements and changes are intended to be included within the protection of the present disclosure.

70 Input unit 72 Calculation unit 74 Estimate unit 80 Acquisition unit 82 Division unit 84 Generation unit 90 Device for determining the reliability of data 92 Device for processing training data 922 Learning unit 924 Storage unit 600 PC
610 CPU
620 ROM
630 RAM
640 Storage unit 650 Input / output unit 660 Communication unit 20 Control device 21 Torque detection unit 22 Position detection unit 23 Servo driver 24 Servo motor 25 Press mechanism 26 Training data storage unit 202 Input data reliability calculation unit 204 Reliability estimation unit 206 Control unit

Claims (7)

A method of controlling the operation of a mechanical device using a control device equipped with a machine learning model obtained from the acquired training data.
The control device outputs output data for operating the mechanical device based on the input data input to the control device.
The control device outputs output data for operating the mechanical device based on the input data input to the control device, divides the acquired training data into a plurality of data areas, and divides the plurality of data. It calculates the data distribution characteristics for the region.
The method is
Receiving input data containing at least one measured parameter of the mechanical device and
Determining the data area corresponding to the input data from the data area of the training data,
Determining the reliability of the output data corresponding to the input data based on the data distribution characteristics of the data area corresponding to the input data.
Controlling the operation of the mechanical device according to the reliability of the output data,
Including
The data distribution characteristic comprises the amount of the training data in the data area corresponding to the input data.
Determining the reliability of the output data based on the data distribution characteristics of the data area corresponding to the input data is not possible.
If the amount of the training data in the data area corresponding to the input data is larger than the first predetermined threshold value, it is determined to be highly reliable, and if not, it is determined to be unreliable .
If the distances between the parts of the training data in the data area corresponding to the input data are all less than the second predetermined threshold value, it is determined that the reliability is high, and if not, it is determined that the reliability is low. ,
Controlling the operation of the mechanical device according to the reliability of the output data is not possible.
When reliability is high, controlling the operation of the mechanical device based on the output data, and
When the reliability is low, the predetermined data includes setting the output data to predetermined data and controlling the operation of the mechanical device based on the set output data, and the predetermined data is applicable to the operation of the mechanical device. It is an operation parameter, or it is data instructing the mechanical device to end the operation .
How to control the operation of mechanical devices, including. Controlling the operation of the mechanical device according to the reliability of the output data is not possible.
The method of controlling the operation of the mechanical device according to claim 1, wherein the mechanical device is a press machine, which comprises controlling the bottom dead center of the mechanical device according to the reliability of the output data. The input data is data input to a machine learning model obtained by machine training to determine the output data, and the training data is data for training the machine learning model. The method for controlling the operation of the mechanical device according to 1. Controlling the operation of the mechanical device
By writing a log based on the output data and tracking the operation history of the mechanical device,
Performing mechanical control or process control in the mechanical device based on the output data or switching to artificial control, the mechanical control includes at least one of stop, deceleration and start.
Generating the training data based on the output data to improve control of the mechanical device by additional learning.
The method of controlling the operation of the mechanical device according to claim 1, which comprises at least one of. It is a control device that controls the operation of a mechanical device equipped with a machine learning model obtained from the acquired training data .
The control device is
An input unit (70) configured to receive input data including at least one measured parameter of the mechanical device.
A calculation unit (72) configured to determine the data area corresponding to the input data from the data area of the training data, and
An acquisition unit (80) configured to acquire the training data, and
A division unit (82) that divides the acquired training data into a plurality of areas so that each training data is included in the corresponding data area.
For each of the data areas in the plurality of data areas, a generation unit (84) that generates the data distribution characteristics, and a generation unit (84).
An estimation unit (74) configured to determine the reliability of the output data corresponding to the input data based on the data distribution characteristics of the data area corresponding to the input data.
A control unit (206) that controls the operation of the controlled object according to the reliability of the output data,
Equipped with
The estimation unit (74)
If the amount of the training data in the data area corresponding to the input data is larger than the first predetermined threshold value, it is determined to be highly reliable, and if not, it is determined to be unreliable.
If all the distances between the parts of the training data in the data area corresponding to the input data are less than the second threshold value, it is determined that the reliability is high, and if not, it is determined that the reliability is low.
The control unit (206)
By controlling the operation of the mechanical device according to the reliability of the output data ,
When the reliability is high, the operation of the mechanical device is controlled based on the output data, and the operation is controlled .
When the reliability is low, the output data is set to predetermined data, and the operation of the mechanical device is controlled based on the set output data .
The predetermined data is an operation parameter applicable to the operation of the mechanical device, or is data instructing the mechanical device to end the operation .
A device that controls the operation of the mechanical device. A computer program that, when executed by a processor, performs the method according to any one of claims 1-4. A computer-readable storage medium that stores a computer program that performs the method according to claim 6, when executed by a processor.

Images