WO2019171115A1

WO2019171115A1 - Method for controlling operations of mechanical device and method and device for determining reliability of data

Info

Publication number: WO2019171115A1
Application number: PCT/IB2018/051386
Authority: WO
Inventors: Yoshihisa Ijiri
Original assignee: Omron Corporation
Priority date: 2018-03-05
Filing date: 2018-03-05
Publication date: 2019-09-12
Also published as: JP7099531B2; JP2021516387A; EP3762792A1

Abstract

A method for controlling operations of a mechanical device and a method and device for determining the reliability of data are provided, wherein the method for determining the reliability of data comprises: receiving input data; determining, from a data area of training data, a data area corresponding to the input data; and determining the reliability of the output data on the basis of the data distribution feature of the data area corresponding to the input data. The present disclosure solves the problem that the reliability of the control parameter of the mechanical device be determined in the related art, having the beneficial effect that the reliability of the control parameter of the mechanical device can be determined.

Description

METHOD FOR CONTROLLING OPERATIONS OF MECHANICAL DEVICE AND METHOD AND DEVICE FOR DETERMINING

RELIABILITY OF DATA

Technical Field

The present disclosure relates to the field of automatic control in mechanical engineering, and in specifically, relates to a method for controlling the operations of the mechanical device and a method, a device and a system for determining the reliability of output data of the machine learning model and a method and a device for processing training data.

Background

In the mechanical control field, the technology that the optimal control is made according to various mechanical control parameters is developed in tremendous speed nowdays. For example, in the Document 1 (JP 2017-325), the technical solution that the acceleration and deceleration operations of each shaft of the mechanical device are generated by using the neural network. However, D1 merely discloses that the movement of each shaft is determined by the machine learning, but the reliability of the movement is unknown.

For the problem that the reliability of the control parameter of the mechanical device cannot be determined in the related art, no effective solution has been proposed yet.

Summary

^• The technical problem to be solved

A method for controlling operations of a mechanical device and a method, device and system for determining the reliability of data, and a method and device for processing training data are provided in the embodiments of the disclosure, so as to at least solve the technical problem that a user cannot determine the reliability of the control parameter of the mechanical device in the related art.

^• Means of solving the technical problem

According to one aspect of the embodiments of the disclosure, a method for controlling the operations of the mechanical device is provided, the method comprising: receiving input data, wherein the input data comprises at least one measured parameter of the mechanical device; determining a data area corresponding to the input data from a data area of training data; determining the reliability of the output data corresponding to the input data on the basis of the data distribution feature of the data area corresponding to the input data; and controlling the operations of the mechanical device according to the reliability of the output data.

In the method, on the basis of the distribution of the data area corresponding to the input data, the reliability of output data of the machine learning model which has been trained can be determined. Hence, the mechanical device can know the reliability of output data of the machine learning model, thereby performing a corresponding operation according to the reliability, such that the error rate of a determination or control that uses the output data can be reduced, and the robustness of the mechanical device can be improved.

In an embodiment of the disclosure, determining the data area corresponding to the input data from the data area of the training data comprises dividing the data area of the training data into a plurality of data areas, and determining the data area corresponding to the input data from the plurality of data areas; or determining a proximity area from the data area of the training data by proximity learning, and taking the proximity area as the data area corresponding to the input data. Hence, the data area corresponding to the input data can be determined quickly.

In an embodiment of the disclosure, controlling the operations of the mechanical device according to the reliability of the output data comprises: controlling a down dead point of the mechanical device according to the reliability of the output data, wherein the mechanical device is a press machine. Hence, the down dead point dynamic accuracy is improved.

In an embodiment of the disclosure, the reliability of the output data is a probability that a machine learning model, which has been trained by the training data, generates desired output data by using the input data.

In an embodiment of the disclosure, the input data is the data that is input into a 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.

In an embodiment of the disclosure, the data distribution feature comprises the amount or intensity of the training data within the data area corresponding to the input data.

In an embodiment of the disclosure, determining the reliability of the output data on the basis of the data distribution feature of the data area corresponding to the input data comprises: in a case where the amount or intensity of the training data within the data area corresponding to the input data is greater than a first preset threshold value, determining the reliability is high, and if not, determining the reliability is low, or in a case where the distances among part of the training data within the data area corresponding to the input data are all less than a second preset threshold value, determining the reliability is high, and if not, determining the reliability is low. Hence, by means of dividing the described data areas, the reliability of the output data corresponding to the input data can be easily known, and there is no need to calculate the accurate reliability of output data for each piece of input data using a complex algorithm, reducing the operation load of the system.

In an embodiment of the disclosure, controlling the operations of the mechanical device according to the reliability of the output data comprises: in a case where the reliability is high, controlling the operation of the mechanical device on the basis of the output data; and in a case where the reliability is low, setting the output data to preset data, and controlling the operation of the mechanical device on the basis of the set output data, wherein the preset data is the operation parameter applicable to the operation of the mechanical device or is the data instructing the mechanical device to terminate the operation. Hence, the mechanical device can be controlled more reliably, improving the robustness of the whole system.

In an embodiment of the disclosure, controlling the operations of the mechanical device: writing a log on the basis of the output data to track an operation history of the mechanical device; performing mechanical control or process control on the mechanical device or switching to human control on the basis of the output data, wherein the mechanical control comprises at least one of the following: stopping, slowing down and starting; and generating the training data on the basis of the output data so as to improve the control of the mechanical device by additional learning.

According to one aspect of the embodiments of the disclosure, a method for determining the reliability of data is provided, the method comprising: receiving input data; determining a data area corresponding to the input data from a data area of training data; and determining the reliability of the output data corresponding to the input data on the basis of the data distribution feature of the data area corresponding to the input data. Hence, an operator or an operating device can know the reliability of output data of the machine learning model, thereby performing a corresponding operation according to the reliability, such that the error rate of a determination or control that uses the output data can be reduced, and the robustness of the whole system can be improved.

According to another aspect of the embodiments of the disclosure, a method for processing training data is further provided, the method comprises: acquiring training data; dividing the acquired training data into a plurality of areas, wherein each of the training data falls within a corresponding data area; and generating a data distribution feature for each data area in the plurality of data areas, wherein the data distribution feature is used for determine the reliability of output data corresponding to input data falling within one of the plurality of data areas.

By means of dividing training data into a plurality of data areas, the distribution of the training data can be known in a training process for a machine learning model, such that the reliability of output data can be determined according to the distribution of the training data, when a machine leaning model which has been trained is used, thereby providing a basis for a subsequent operation.

In an embodiment of the disclosure, the data distribution feature of the each data area comprises the amount or intensity of the training data within the each data area.

In an embodiment of the disclosure, it is determined whether the amount or intensity of the training data within one data area in the plurality of data areas is greater than a first preset threshold value, so as to determine the reliability of the output data.

In an embodiment of the disclosure, the method further comprises determining whether the distances among part of the training data within one data area in the plurality of data areas are all less than a second threshold value, so as to determine the reliability of the output data.

According to still another aspect of the embodiments of the disclosure, a device for determining the reliability of data is further provided, the device comprising: an input unit configured to receive input data; a calculation unit configured to determine a data area corresponding to the input data from a data area of the training data; and an estimation unit configured to determine the reliability of the output data corresponding to the input data on the basis of the data distribution feature of the data area corresponding to the input data.

Hence, an operator or an operating device can know the reliability of output data of a machine learning model, thereby performing a corresponding operation according to the reliability, such that the error rate of a determination or control that uses the output data can be reduced, and the robustness of the whole system can be improved.

In an embodiment of the disclosure, the device further comprises a control unit, the control unit being configured to control the operation of a controlled object according to the reliability of the output data.

In an embodiment of the disclosure, the data distribution feature comprises the amount or intensity of the training data within the data area corresponding to the input data; and the estimation unit is further configured to: in a case where the amount or intensity of the training data within the data area corresponding to the input area is greater than a first preset threshold value, determine the reliability is high, and if not, determine the reliability is low; or in a case where the distances among part of training data within the data area corresponding to the input data are all less than a second threshold value, determine the reliability is high, and if not, determine the reliability is low.

According to yet another aspect of the embodiments of the disclosure, a device for processing training data is further provided, the device comprising: an acquiring unit configured to acquiring training data; a dividing unit configured to divide the acquired training data into a plurality of areas, wherein each training data falls within a corresponding data area; and a generating unit configured to generate a data distribution feature for each data area in the plurality of data areas, wherein the data distribution feature is used for determining reliability of output data corresponding to input data falling within one of the plurality of data areas.

By means of dividing training data, the distribution of the training data can be known in the training process of a machine learning model, such that the reliability of output data can be determined according to the distribution of the training data, when a machine leaning model which has been trained is used, thereby providing a basis for a subsequent operation.

According to yet another aspect of the embodiments of the disclosure, a system comprising the device for determining the reliability of data of any one of the described technical solutions and the device for processing training data of the described technical solution is provided in an embodiment of the disclosure.

According still another aspect of the disclosure, a computer program is further provided, wherein the computer program, when be executed by a processor, executes the method of any one of technical solutions. According still another aspect of the disclosure, a computer readable storage medium is further provided, wherein the computer readable storage medium storing a computer program, which, when be executed by a processor, executes the method of any one of the above technical solutions.

^• Technical effect

In the method for controlling the operations of the mechanical provided by the embodiments of the disclosure, the reliability of output data corresponding to input data is determined on the basis of the input data and the data distribution feature of the training data, and the operations of the mechanical device are controlled based on the reliability, thereby solving the problem that the reliability of the control parameters of the mechanical device cannot be determined in the related art, and having the beneficial effect that the reliability of the control parameters of the mechanical device can be determined.

In addition, for a device which performs control using output data of a machine learning model which has been trained, as the reliability of the output data, which is the basis of the control, can be known, and therefore, the a reliable control can be performed, and an operation going beyond a predetermined range will not occur, improving the efficiency and robustness of the control.

Brief Description of the Drawings

Fig. 1 is a flow chart of a method for controlling the operations of a press machine according to an embodiment of the disclosure;

Fig. 2 is a structure schematic diagram for illustrating a press machine control system to which the reliability of output data is applied according to an embodiment of the disclosure;

Fig. 3 is a schematic diagram for illustrating the transition of a control action of the system for controlling a press machine as shown in Fig. 2;

Fig. 4A is a three-dimensional schematic view corresponding to the measured parameter and target parameter of a system for controlling a press machine as shown in figure 2 according to an embodiment of the disclosure; and

Fig. 4B is a two-dimensional schematic view of the measured parameter and target parameter according to an embodiment of the disclosure;

Fig. 5 is a schematic diagram for illustrating the transition of a control action of the system for controlling a press machine as shown in Fig. 2;

Fig. 6 is a schematic view of an example of a PC (Personnel Computer) 600 as a part of the hardware configuration of a system for determining the reliability of data according to an embodiment of the disclosure;

Fig. 7 is a structural schematic diagram for illustrating a device for determining the reliability of data according to an embodiment of the disclosure;

Fig. 8 is a structural schematic diagram for illustrating a device for processing training data according to an embodiment of the disclosure; and

Fig. 9 is a structure schematic diagram for illustrating a system for determining the reliability of data according to an embodiment of the disclosure.

Detailed Description of the Embodiments

In order to allow a person skilled in the art to have a better understanding of the disclosure, the embodiments of the disclosure will be clearly and completely described in conjunction with the drawings of the disclosure in the following. Obviously, the described embodiments are only a part of the embodiments of the disclosure, but not all of the embodiments. On the basis of the embodiments of the disclosure, all of the other embodiments which can be obtained by a person skilled in the art without involving any inventive efforts fall within the range set forth by the disclosure.

Fig. 1 is a flow chart of a method for controlling the operations of a press machine according to an embodiment of the disclosure. As illustrated in Fig. 1 , the method comprises the following steps.

At Step S10, input data including a measured torque and a measured position is received.

A machine learning model receives input data input. The input data is the data that is input into a machine learning model obtained by machine training to determine the output data.

The process of training a machine learning model involves providing a machine learning algorithm with training data to learn from. The term machine learning model refers to model artifact that is created by the training process. To train a machine learning model, the user needs to specify the following: input training data source; name of the data attribute that contains the target to be predicted; required data transformation instructions; and training parameters to control the machine learning algorithm. During the training process, the correct learning algorithm is selected automatically or manually, based on the type of target the user specified in the training data source. Machine learning algorithms can be divided into 3 broad categories - supervised learning, unsupervised learning, and reinforcement learning. Supervised learning, which is employed in the disclosure, is useful in cases where a label is available for a certain dataset. Decision trees, Na^'ive bayes classification, ordinary least squares regression and logistic regression are the algorithms for the supervised learning.

Since the present disclosure does not emphasis on how to train a machine learning model, no further description will be provided.

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

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

The stored training data for training the machine learning model is divided into a plurality of data areas. All the training data may be divided according to different dividing strategies so as to obtain a plurality of data areas. For example, the training data may be randomly divided into n subsets which are mutually exclusive; alternatively, the space where the training data is located may be divided into n spatial subsets with the same size or different sizes. Of course, in other embodiments, other methods for dividing training data may be used. The description on how to divide training data in the embodiments of the present disclosure does not intend to limit the method for diving training data.

According to the input data, a data area corresponding to the input data is determined from a plurality of data areas. In most of cases, input data necessarily falls within one of a plurality of data areas divided according to training data. In a rare case where input data may fail to fall within any divided data area, the reliability of the output data corresponding to the input data can be directly determined to be low.

In one embodiment, training data may be divided into only one data area, that is, all training data is considered as one set, such that input data necessarily falls within this data area. When the reliability of output data is to be calculated, the reliability of the output data can be determined by calculating the amount or intensity of the training data which is proximate to the input data by proximity learning.

At step S14, the reliability of the output data corresponding to the input data is determined on the basis of the data distribution feature of the data area corresponding to the input data. The data distribution feature comprises the amount or intensity of the 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 according to amount or intensity of the training data in the data area corresponding to the input data. Particularly, the or intensity of training data in the data area where the input data falls within can be calculated with reference to a data distribution table so as to infer the reliability of output data. Alternatively, the reliability of output data can be inferred according to distances from training data which is proximate to the input data to part of proximate training data.

In a case where the amount or intensity of the training data in the data area corresponding to the input data is large, the reliability of output data is determined to be high; otherwise, the reliability of output data is determined to be low. Alternatively, in a case where distances among part of the training data in the data area corresponding to the input data are near, the reliability of output data is determined to be high; otherwise, the reliability of output data is determined to be low.

At step S16, the operations of a press machine is controlled according to the reliability of output data.

After the reliability of output data is determined, a corresponding process for the press machine can be performed according to the reliability of output data. For example, in a case where the reliability of output data is high, a down dead point of the press machine is controlled on the basis of the current output data. However, in a case where the reliability is low, which means the current output data may result in a high error rate of a subsequent process, the output data needs to be replaced with preset data, and a down dead point of the press machine is controlled on the basis of the preset data.

Various operations, not limited to the press machine can be controlled by using the reliability of output data, such as writing the log on the basis of the output data to track an operation history of the controlled object; performing mechanical control or process control on the controlled object or switching to human control on the basis of the output data, wherein the mechanical control comprises at least one of the following: stopping, slowing down and starting; generating the training data on the basis of the output data so as to improve the control of the controlled object by additional learning.

Of course, in many other scenarios, the reliability of output data also can be used. For example, when a user search for a keyword, a search engine can use the reliability of output data to determine which results are more suitable for the user and determine which advertisement is more suitable for the user. In a case of online shopping, an online shopping website can use the reliability of output data to recommend goods to a user for selection. To sum up, the reliability of output data is applicable to various fields, and application scenarios of reliability of output data are not specifically limited in the present disclosure.

Fig. 2 is a structure schematic diagram for illustrating a press machine control system to which the reliability of output data is applied according to an embodiment of the disclosure. Fig. 3 is a schematic diagram for illustrating the transition of a control action of the system for controlling a press machine as shown in Fig.2. As illustrated in Fig. 2, the system for controlling the press machine comprises a control device (PLC) 20, a torque detecting unit 21 , a position detecting unit 22, a servo driver 23, a servo motor 24, a press mechanism 25, and a training data saving unit 26, the control device 20 comprising an input data reliability calculating unit 202, a reliability inferring unit 204, and a control unit 206.

The torque detecting unit 21 , which may be a load cell, is configured to detect torque of the press machine to obtain actually-measured torque. The position detecting unit 22, which may be a position sensor, is configured to detect a position of the press machine to obtain an actually-measured position.

The torque detecting unit 21 and the position detecting unit 22 respectively send the control device 20 the obtained actually-measured torque and actually-measured position.

The control device 20 obtains training data from the training data saving unit 26 and divides the training data into a plurality of data areas, for example, divides the training data into a plurality of blocks as illustrated in Fig. 4A. Then, the control device 20 calculates a data distribution feature for each data area. In one embodiment, data areas can be pre-divided and a data distribution feature can be pre-calculated for each data area so as to improve the performance of the control device 20.

The input data reliability calculating unit 202 of the control device 20 determines, after receiving the actually-measured torque and actually-measured position, a data area corresponding to the actually-measured torque and actually-measured position in a plurality of data areas, that is, a corresponding block in the matrix, and further determines the data distribution feature of the determined data area, for example, calculates the intensity of the training data in the corresponding block.

The reliability inferring unit 204 determines the reliability of a target torque and a target position (i.e., output data, also called target parameters) according to the determined data distribution feature. Specifically, referring to Fig. 4A, X1 axis represents a measured torque, X2 axis represents a measured position, and Y axis represents a target parameter f(x1 , x2). The control unit 206 calculates, by using the machine learning model, the target parameter f(x1 , x2), i.e. the target torque and the target position, of a next period when a down dead point is reached. The reliability inferring unit 204 determines a corresponding area in the X1 X2 plane according to the inputted measured torque and measured position. In a case where the density of the training data in the determined 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 may use a preset torque parameter and position parameter (for example, default values for torque and position) instead of the target torque and target position from the control unit 206 to control the press machine. In one embodiment, an alarm may be provided when reliability is low. In another embodiment, actually-measured torque and an actually-measured position may be saved and learned so as to improve the machine learning model.

In the present embodiment, as shown in Fig. 4A, there are two measured parameters, i.e. a measured torque and a measured position. In other embodiments, there may be one or more than two measured parameter(s). In a case where there is only one measured parameter, as shown in Fig. 4B, X axis represents a measured parameter, Y axis represents a target parameter, and a plurality of interval segments on the X axis represent a plurality of divided data areas. The reliability of a target parameter f(x) can be determined according to the density of the training data in an interval segment corresponding to the inputted measured parameter.

When the reliability of the target torque and target position is low, the servo driver 23 uses a default torque parameter and position parameter to control the servo motor 24; otherwise, the servo driver 23 uses the calculated target torque and target position of the next period when the down dead point is reached to control the servo motor 24. The servo motor 24 further controls the press mechanism 25, such that the difference value between a down dead point of each period of the press mechanism 25 and an actually-measured position is 0 (see Fig. 3). The precision of the down dead point, having an effect on the precision of a workpiece, is an important technical index for a press machine. In the present embodiment, the precision of the down dead point is ensured according to the reliability of target torque and target position, thereby ensuring a high yield of products manufactured by the press machine.

It should be noted that the method for determining the reliability of output data can be applied to other scenes in addition to the press control system. In other embodiments, the reliability of output data can also be applied to a plurality of fields, such as automatic driving, healthcare, retail, aerospace, and traffic.

In an exemplary embodiment of the disclosure, the reliability of output data can be applied to an automatic driving system. The automatic driving system can includes a central processing unit, a braking system, an acceleration system, a steering system, a navigation system and a sensing system. The navigation system is used for receiving the data regarding geographical position information (for example, GPS data, the received data can be used to determine the current position of a vehicle), and for determining the overall driving line of the vehicle according to the current position of the vehicle and a target position set by a user. The sensing system includes more than one sensors which are configured to sense sensing information, such as obstacles in front of, behind, on the left and right sides of a vehicle, a traffic signal in front of the vehicle, and road signs in front of and on the right side of the vehicle, and to send the detected sensing information to the central processing unit. The central processing unit generates a control instruction according to the received sensing information, and determines whether the control instruction is reliable by using the method for determining reliability. In a case where the reliability of the control instruction is relatively high, the central processing unit uses the control instruction to control the braking system, the steering system, the acceleration system and so on, namely, using the reliable control instruction to control the various parts of the vehicle so as to control the direction and speed of the vehicle. It should be noted that although it is a vehicle that has been described in the present embodiment, but the vehicle can includes, but is not limited to, any type of vehicle, such as automobile, ship, airplane and train. In the present embodiment, the method for determining the reliability of data can allow the vehicle to operate more accurately according to the result derived by calculation.

In another embodiment, the method for determining reliability can be applied to the field of healthcare, for example, drug discovery, genetic testing, personalized healthcare, or precision surgery. In the present embodiment, surgery is taken as an example. In clinical surgery, real-time interactive quantitative analyses are usually needed to be performed on the three-dimensional volume, distance, angle, blood vessel diameter etc. of human organs by using images, so as to perform a full quantitative three-dimensional assessment before surgery. However, in practice, deviations sometimes may occur to the accuracy of such three-dimensional assessment. The method for determining the reliability of output data is applied to the three-dimensional assessment of the organs, such that it can be determined that the reliability of which part of three-dimensional data outputted by using image data is relatively high, and the reliability of which part is relatively low. In specifically, for the data of human organs that is obtained by using images (i.e. input data), if the density of the training data of the area in which the data falls is high, there are few different situations of output data, and the reliability of the output data is high. In another aspect, if the training data of the area in which the data falls is less, there are many different situations of output data, namely, the reliability of the output data is low. In a case where the reliability is low, the output of a machine (for example, surgery robot) may be outside expectations, and using the output as the action of the machine will be dangerous. In order to avoid such risk, in a case where the reliability of the outputted three-dimensional data is low, a doctor may be required to make a final determination, thereby generating precise three-dimensional data, so as to make the surgery to be more quick, precise and safe.

When realized in a form of a software functional unit and sold or used as an individual product, said device for determining the reliability of data, device for processing training data, and system for determining the reliability of data or part of them can be stored in a computer readable storage medium. On the basis of this understanding, the technical solution of the disclosure essentially can be, or the part of the technical solution which makes a contribution over the prior art or the whole technical solution or a part of the technical solution can be embodied in a form of a software product, and such computer software program is stored in a storage medium and comprises several instructions for enabling a computer device (which may be a PC computer, a server, or Internet equipment) to perform all the steps or part of the steps of the method according to each embodiments of the disclosure. Said storage medium includes various media capable of storing program codes, such as a USB flash disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or a compact disk, and may also includes a data flow that can be downloaded from a server or a cloud.

Fig. 5 is a flow chart of a method for processing training data and determining the reliability of data by using the training data according to an embodiment of the disclosure. As illustrated in Fig. 5, the method comprises the following steps.

At step S50, training data is obtained.

The training data is the data for training a machine learning model. Characteristics and amount of training data are major factors for determining how good the performance of a trained machine learning model is. Generally, all training data for training a machine learning model is stored in a memory device which has a sufficient capacity, such that all training data can be obtained from the memory device.

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

The stored training data for training a machine learning model is divided into a plurality of data areas. All the training data may be divided according to different dividing strategies so as to obtain a plurality of data areas. For example, the training data may be randomly divided into n subsets which are mutually exclusive; alternatively, the space where the training data is located may be divided into n spatial subsets with the same size or different sizes. Flere, the space includes a one-dimensional space, two-dimensional space, and three-dimensional space.

At step S54, a data distribution feature is calculated for each area.

A data distribution table is generated for each data area. A data distribution feature of each data area in the divided plurality of data areas can be calculated according to each data distribution table. A data distribution feature can be defined according to the following aspects: the central tendency of distribution which reflects the extent of each training data drawing close or aggregating towards the central value in a data area; the discretion extent of distribution which reflects the tendency of each training data distancing from the central value in a data area; the shape of distribution which reflects skewness and kurtosis of data distribution. In one embodiment, a data distribution feature may further include the amount or intensity of training data in a data area.

At step S56, it is determined whether the calculation of data distribution features of all data areas is completed.

Whether the calculation of data distribution features of all data areas is completed is determined, and if yes, step 58 is performed, if not, the process returns to step 56.

At step S58, reliability of each area is saved.

Reliability of each area is saved in a corresponding memory device having a sufficient capacity.

At step S60, input data is obtained.

The input data is the data that is input into a machine learning model obtained by machine training so as to determine the output data. In one embodiment, the input data may be sensor data obtained from a device in Internet of things (“IOT” in short).

At step S62, the data distribution feature of the data area corresponding to the input data is determined.

After input data is input, the input data necessarily falls within one data area of the plurality of data areas divided according to training data, and the data area corresponding to the input data thus can be determined. Then, the data distribution feature of the data area corresponding to the input data is determined according to a data distribution feature of each data area saved in step 28.

At step S64, the reliability of output data is inferred.

The reliability of the output data corresponding to the input data is determined according to the amount or intensity of the training data in the data area corresponding to the input data. In a case where the amount or intensity of the training data in the data area corresponding to the input data is large, the reliability of output data is determined to be high; otherwise, the reliability of output data is determined to be low. Alternatively, in a case where the distances among part of the training data in the data area corresponding to the input data are near, the reliability of output data is determined to be high; otherwise, the reliability of output data is determined to be low. In one embodiment relating to IOT, output data is a parameter, with respect to input data, for enabling a machine to properly operate.

At step 66, determination and control is performed according to the reliability of output data.

In a case where the reliability of output data is high, control can be performed directly on the basis of the current output data. However, in a case where the reliability is low, a machine is controlled by using a default parameter or a machine is terminated. Fig. 6 is a schematic view of an example of a PC (Personnel Computer) 600 as a part of the hardware configuration of a system for determining the reliability of data according to an embodiment of the disclosure. As shown in Fig. 6, the PC 600 can include a CPU 610 for performing overall control, a read only memory (ROM) 620 for storing system software, a random access memory (RAM) 630 for storing written-in/read-out data, a storage unit 640 for storing various programs and data, an input/output unit 650 being used as an input/output interface, and a communication unit 660 for implementing a communication function. Alternatively, the CPU 610 can be replaced by a processor, for example a microprocessor MCU or a Field-Programmable Gate Array FRGA. The input/output unit 650 can include various interfaces, such as an input/output interface (I/O interface), a universal serial bus (USB) port (can be included as one port of the ports of an I/O interface), and a network interface. It can be understood for a person skilled in the art that the structure shown in Fig. 6 is merely illustrative, and does not limit the hardware configuration of the system for determining the reliability of data. For example, the PC 600 can further include more or fewer components than those shown in Fig. 6, or have a configuration different from that shown in Fig. 6.

It should be noted that the described CPU 610 can include one or more processor(s), the one or more processor(s) and/or other data processing circuits in the disclosure can generally be referred to as“data processing circuit”. The data processing circuit can be wholly or partly embodied as software, hardware, firmware or any other combinations. In addition, the data processing circuit can be a single independent processing module, or wholly or partly integrated into any one of the other components in the PC 600.

The storage unit 640 can be used for storing software programs of application software and modules, as a program instruction/data storage device described in the disclosure later, the program instruction/data storage device corresponding to the method for determining the reliability of the data. The CPU 610 operates the software programs and modules stored in the storage unit 640 so as to implement the described method for determining the reliability of data. The storage unit 640 can include a non-volatile memory, such as one or more magnetic memory, flash memory or other non-volatile solid state memory. In some examples, the storage unit 640 can further include memories which are remotely provided with respect to the CPU 610, and these remote memories can be connected to the PC 600 by means of a network. The examples of the described network include, but are not limited to, Internet, Intranet, LAN, mobile communication network, and the combinations thereof.

The communication unit 660 is used for receiving or sending data through a network. The specific examples of the described network can include the wireless network provided by the communication provider of the PC 600. In an example, the communication unit 660 includes a network interface controller (NIC), and the NIC can be connected to other network devices by a base station so as to communicate with the Internet. In an example, the communication unit 660 can be a radio frequency (RF) module, which communicates with the Internet in a wireless manner.

Fig. 7 is a structural schematic diagram for illustrating a device for determining the reliability of data according to an embodiment of the disclosure. The device comprises: an input unit 70 configured to receive input data; a calculation unit 72 configured to determine, from a data area of the training data, a data area corresponding to the input data; an inferring unit 74 configured to determine the reliability of the output data corresponding to the input data on the basis of the data distribution feature of the data area corresponding to the input data.

Fig. 8 is a structure schematic diagram for illustrating a device for processing training data according to an embodiment of the disclosure. As illustrated in Fig. 8, the device comprises: an acquiring unit 80, configured to acquire training data; a dividing unit 82 configured to divide the acquired training data into a plurality of data areas, wherein each training data is divided to a corresponding data area; and a generating unit 84 configured to generate a data distribution feature for each data area in the plurality of data areas, so as to determine the reliability of the output data corresponding to the input data which falls within a corresponding data area in the plurality of data areas.

Fig. 9 is a structure schematic diagram for illustrating a system for determining the reliability of data according to an embodiment of the disclosure. As illustrated in Fig. 9, the system comprises the device for determining the reliability of data 90 and the device for processing training data 92. The device for determining the reliability of data 90 can be the device according to Fig. 7 and the device for processing training data 92 can be the device according to Fig. 8, and no further description is necessary.

In one embodiment, the device for processing training data 92 in the system as illustrated in Fig. 9 may further comprise: a learning unit 922 configured to learn output data corresponding to input data by machine learning so as to further ensure quality of the machine learning model; and a storing unit 924 for storing training data and a data distribution table generated according to training data. In one embodiment, the system in Fig. 9 may further comprise a control unit 94 configured to control a controlled object according to determined reliability of the output data.

The above are only the preferred embodiments of the present disclosure. For those skilled in the art, various improvement and modifications can be made without departing from the principle of the present disclosure, and such improvement and modification are intended to be included within the scope of protection of the present disclosure.

Reference signs:

70 input unit 72 calculation unit

74 inferring unit 80 acquiring unit

82 dividing unit 84 generating unit

90 device for determining the reliability of data

92 device for processing training data

922 learning unit 924 storing unit

600 PC 610 CPU

620 ROM 630 RAM

640 Storage unit 650 Input/output unit

660 Communication unit

20 control device 21 torque detecting unit

22 position detecting unit 23 servo driver

24 servo motor 25 press mechanism

26 training data saving unit 202 input data reliability calculating unit

204 reliability inferring unit 206 control unit.

Claims

1. A method for controlling operations of a mechanical device, wherein the method comprises:

receiving input data, wherein the input data comprises at least one measured parameter of the mechanical device;

determining a data area corresponding to the input data from a data area of training data;

determining reliability of output data corresponding to the input data on the basis of the data distribution feature of the data area corresponding to the input data; and controlling the operations of the mechanical device according to the reliability of the output data.

2. The method of claim 1 , wherein determining the data area corresponding to the input data from the data area of the training data comprises:

dividing the data area of the training data into a plurality of data areas, and determining the data area corresponding to the input data from the plurality of data areas; or

determining a proximity area from the data area of the training data by proximity learning, and taking the proximity area as the data area corresponding to the input data.

3. The method of claim 1 , wherein controlling the operations of the mechanical device according to the reliability of the output data comprises:

controlling a down dead point of the mechanical device according to the reliability of the output data, wherein the mechanical device is a press machine.

4. The method of claim 1 , wherein the reliability of the output data is a probability that a machine learning model, which has been trained by the training data, generates desired output data by using the input data.

5. The method of claim 1 , wherein the input data is the data that is input into a 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.

6. The method of any one of claims 1 to 5, wherein the data distribution feature comprises the amount or intensity of the training data within the data area corresponding to the input data.

7. The method of claim 6, wherein determining the reliability of the output data on the basis of the data distribution feature of the data area corresponding to the input data comprises:

in a case where the amount or intensity of the training data within the data area corresponding to the input data is greater than a first preset threshold value, determining the reliability is high, otherwise determining the reliability is low, or

in a case where the distances among part of the training data within the data area corresponding to the input data are all less than a second preset threshold value, determining the reliability is high, otherwise determining the reliability is low.

8. The method of claim 6, wherein controlling the operations of the mechanical device according to the reliability of the output data comprises:

in a case where the reliability is high, controlling the operations of the mechanical device on the basis of the output data; and

in a case where the reliability is low, setting the output data to preset data, and controlling the operation of the mechanical device on the basis of the set output data, wherein the preset data is an operation parameter applicable to the operation of the mechanical device or is data instructing the mechanical device to terminate the operation.

9. The method of claim 6, wherein controlling the operations of the mechanical device comprises at least one of the following:

writing a log on the basis of the output data to track an operation history of the mechanical device;

performing mechanical control or process control on the mechanical device or switching to human control on the basis of the output data, wherein the mechanical control comprises at least one of the following: stopping, slowing down and starting; and

generating the training data on the basis of the output data to improve the control of the mechanical device by additional learning.

10. A method for determining reliability of data, wherein the method comprises: receiving input data;

determining from a data area of training data, a data area corresponding to the input data; and

determining reliability of output data corresponding to the input data on the basis of the data distribution feature of the data area corresponding to the input data.

11. A method for processing training data, wherein the method comprises:

acquiring training data; dividing the acquired training data into a plurality of data areas, wherein each training data falls within a corresponding data area; and

generating a data distribution feature for each data area in the plurality of data areas, wherein the data distribution feature is used for determining reliability of output data corresponding to input data falling within one of the plurality of data areas.

12. The method of claim 11 , wherein the data distribution feature of the each data area comprises the amount or intensity of the training data within the each data area.

13. The method according to claim 12, wherein the method further comprises determining whether the amount or intensity of the training data within one of the plurality of data areas is greater than a first preset threshold value, so as to determine the reliability of the output data.

14. The method of claim 12, wherein the method further comprises determining whether the distances among part of the training data within one data area in the plurality of data areas are all less than a second threshold value, so as to determine the reliability of the output data.

15. A device for determining reliability of data, wherein the device comprises: an input unit configured to receive input data;

a calculation unit configured to determine, from a data area of training data, a data area corresponding to the input data; and

an estimation unit configured to determine the reliability of the output data corresponding to the input data on the basis of the data distribution feature of the data area corresponding to the input data.

16. The device of claim 15, wherein the device further comprises a control unit, the control unit being configured to control an operation of a controlled object according to the reliability of the output data.

17. The device of claims 15 or 16, wherein

the data distribution feature comprises the amount or intensity of the training data within the data area corresponding to the input data; and

the estimation unit is further configured to:

in a case where the amount or intensity of the training data within the data area corresponding to the input area is greater than a first preset threshold value, determine the reliability is high, otherwise determine the reliability is low; or

in a case where the distances among part of training data within the data area corresponding to the input data are all less than a second threshold value, determine the reliability is high, otherwise determine the reliability is low.

18. A device for processing training data, wherein the device comprises:

an acquiring unit configured to acquiring training data;

a dividing unit configured to divide the acquired training data into a plurality of areas, wherein each training data falls within a corresponding data area; and

a generating unit configured to generate a data distribution feature for each data area in the plurality of data areas, wherein the data distribution feature is used for determining reliability of output data corresponding to input data falling within one of the plurality of data areas.

19. A system for determining reliability of data, wherein the system comprises the device for determining the reliability of data of any one of claims 15 to 17 and the device for processing the training data of claim 18.

20. A computer program, wherein the computer program, when be executed by a processor, executes the method of any one of claims 1 to 14.

21. A computer readable storage medium storing a computer program, which, when be executed by a processor, executes the method of any one of claims 1 to 14.