JPWO2020158184A1 - Transport control device, transport control method, transport control program, learning device, learning method, learning program, and trained model - Google Patents

Abstract

In the transport control device, the transport control method, the transport control program, the learning device, the learning method, the learning program, and the trained model, it is possible to easily suppress the shaking when the object to be transported is stopped. The transport control device swingably suspends the transported object and transports the transported object from the start position to the target position. The drive unit is driven along a path including a start position and a target position. The trained model minimizes the sway when the transported object reaches the target position by inputting information indicating at least the length from the fulcrum of the swing to the center of gravity of the transported object, and the start position and the target position. Learning is performed to output the steady speed and the timing of acceleration / deceleration during transportation of the object to be transported. Based on the input length and the information representing the start position and the target position, the trained model outputs the steady speed and the acceleration / deceleration timing when the object to be transported is transported from the start position to the target position.

Description

The present disclosure relates to a transport control device, a transport control method, a transport control program, a learning device, a learning method, a learning program, and a trained model for transporting an object to be transported to a target position.

In a transport system that transports an object to be transported by a suspended crane, a series of suspensions of the object to be transported are suspended from the crane, the object to be transported is transported to a target position in a suspended state, and the object to be transported is lowered at the target position. The work is repeated. However, when the transported object is stopped at the target position, it swings according to the law of inertia. Therefore, in order to continue the work such as unloading the object to be transported from the crane, it is necessary to wait until the rocking stops or to forcibly stop the rocking, and as a result, the working time. Becomes longer.

In order to shorten the work time of such a series of operations, a method of suppressing the shaking of the transported object at the time of stopping has been proposed. For example, in Patent Document 1, by fuzzy inference using a rule map, an operating speed pattern including two-step acceleration and deceleration suitable for the rope length to be used is created, a load swing angle of a suspended load is measured, and a learning function is provided. A method has been proposed in which the control time for acceleration and deceleration is adjusted based on the load deflection angle. Further, Patent Document 2 proposes a method of measuring a rope state quantity such as a rope runout angle and outputting a speed reference for stopping the trolley at a target position by using model prediction control. Further, in Patent Document 3, the load swing angle of the suspended load is measured, and the parameters required for the speed control of the trolley by the learning model based on the crane state quantity such as the rope length and the weight of the suspended load and the load swing angle. A method for outputting is proposed.

Japanese Unexamined Patent Publication No. 7-144881 Japanese Unexamined Patent Publication No. 11-21077 Japanese Unexamined Patent Publication No. 2003-176093

However, in the methods described in Patent Documents 1 to 3, a means such as a sensor for measuring the load swing angle and the rope swing angle is separately required. Further, in the method using the rule map described in Patent Document 1, when a rope having a rope length not included in the rule map is used, an appropriate driving speed pattern may not be created. Further, since the method described in Patent Document 2 uses a model premised on the dynamics of a two-dimensional pendulum crane, it is necessary to analyze the model system. Further, since the method described in Patent Document 3 uses the equation of state with the crane as a simple pendulum, it is difficult to apply it to a complicated model other than the crane.

The present disclosure has been made in view of the above circumstances, and an object of the present disclosure is to make it possible to easily suppress shaking when the transported object is stopped.

The transport control device according to the present disclosure is a transport control device that swingably suspends the transported object and transports the transported object from a start position to a target position.

By inputting at least the length from the fulcrum of the swing to the center of gravity of the transported object, and the input of information indicating the start position and the target position, the transported object is to minimize the shaking when the transported object reaches the target position. A trained model that has been trained to output the steady speed and acceleration / deceleration timing during transportation,

It is equipped with an input unit that accepts input of information indicating the length, start position, and target position.

Based on the input length and the information representing the start position and the target position, the trained model outputs the steady speed and the acceleration / deceleration timing when the object to be transported is transported from the start position to the target position.

In the transport control device according to the present disclosure, the information representing the start position and the target position may include the three-dimensional coordinates of the start position and the target position.

Further, in the transport control device according to the present disclosure, the information representing the start position and the target position may include the distance between the start position and the target position.

Further, in the transport control device according to the present disclosure, the information indicating the start position and the target position includes the distance between the start position and the target position and the height difference between the start position and the target position. good.

Further, in the transport control device according to the present disclosure, the acceleration / deceleration timing may include an acceleration time or an acceleration end time.

Further, in the transport control device according to the present disclosure, the acceleration / deceleration timing may include an acceleration end time, a deceleration start time, and a deceleration end time.

Further, in the transfer control device according to the present disclosure, the acceleration / deceleration timings are the first acceleration end time, at least one second acceleration start time, at least one second acceleration end time, and at least one deceleration start time. And may include at least one deceleration end time.

Further, in the transport control device according to the present disclosure, a drive unit driven along a path including a start position and a target position, and a drive unit.

It may further include a control unit that controls the drive unit according to the steady speed and acceleration / deceleration timing output from the trained model.

Further, in the transport control device according to the present disclosure, the drive unit may include a hook for suspending the object to be transported.

Further, in the transport control device according to the present disclosure, the drive unit may include a suspension mechanism for suspending the transported object so as to be rotatable in the moving direction of the transported object.

The learning device according to the present disclosure is a learning device that learns a learning model for suspending an object to be transported so as to swingably and determining the timing of acceleration / deceleration when the object to be transported is transported from a start position to a target position. There,

Transport of the transported object to minimize shaking when the transported object reaches the target position with respect to at least the length from the fulcrum of the swing to the center of gravity of the transported object, as well as the information indicating the start position and the target position. An input unit that accepts input of multiple teacher data indicating the steady speed and acceleration / deceleration timing inside,

When the transported object reaches the target position by inputting at least the length from the fulcrum of the swing to the center of gravity of the transported object, and unknown information indicating the start position and the target position based on a plurality of teacher data. It is provided with a learning unit that learns a learning model so as to output a steady speed and acceleration / deceleration timing during transportation of an object to be transported in order to minimize shaking.

The trained model according to the present disclosure is a trained model for suspending the transported object so as to swingably and determining the acceleration / deceleration timing when the transported object is transported from the start position to the target position.

By inputting at least the length from the fulcrum of the swing to the center of gravity of the transported object, and the input of information indicating the start position and the target position, the transported object is to minimize the shaking when the transported object reaches the target position. Learning is done to output the steady speed and the timing of acceleration / deceleration during transportation.

The transport control method according to the present disclosure is a transport control method in which the transported object is swayably suspended and the transported object is transported from the start position to the target position.

By inputting at least the length from the fulcrum of the swing to the center of gravity of the transported object, and the input of information indicating the start position and the target position, the transported object is to minimize the shaking when the transported object reaches the target position. For a trained model that has been trained to output the steady speed and acceleration / deceleration timing during transportation, it accepts input of information indicating the length and start position and target position.

Based on the input length and the information representing the start position and the target position, the trained model outputs the steady speed and the acceleration / deceleration timing when the object to be transported is transported from the start position to the target position.

The learning method according to the present disclosure is a learning method for learning a learning model for swaying the object to be transported and determining the timing of acceleration / deceleration when the object to be transported is transported from the start position to the target position. There,

Transport of the transported object to minimize shaking when the transported object reaches the target position with respect to at least the length from the fulcrum of the swing to the center of gravity of the transported object, as well as the information indicating the start position and the target position. Accepts the input of multiple teacher data representing the steady speed and acceleration / deceleration timing in the middle,

When the transported object reaches the target position by inputting at least the length from the fulcrum of the swing to the center of gravity of the transported object, and unknown information indicating the start position and the target position based on a plurality of teacher data. The learning model is trained to output the steady velocity and the timing of acceleration / deceleration during transportation of the object to be transported in order to minimize the shaking.

It should be noted that the transfer control method and the learning method according to the present disclosure may be provided as a program for causing a computer to execute the method.

The other transport control device according to the present disclosure is a transport control device that swingably suspends the transported object and transports the transported object from the start position to the target position, and stores commands for the computer to execute the transfer control device. Memory to do, and

The processor comprises a processor configured to execute a stored instruction.

By inputting at least the length from the fulcrum of the swing to the center of gravity of the transported object, and the input of information indicating the start position and the target position, the transported object is to minimize the shaking when the transported object reaches the target position. For a trained model that has been trained to output the steady speed and acceleration / deceleration timing during transportation, it accepts input of information indicating the length and start position and target position.

Based on the input length and the information representing the start position and target position, the trained model executes the process of outputting the steady speed and acceleration / deceleration timing when transporting the object to be transported from the start position to the target position. ..

The other learning device according to the present disclosure is learning to learn a learning model for suspending the transported object so as to swingably and determining the acceleration / deceleration timing when the transported object is transported from the start position to the target position. A device that stores instructions for a computer to execute, and

The processor comprises a processor configured to execute a stored instruction.

Transport of the transported object to minimize shaking when the transported object reaches the target position with respect to at least the length from the fulcrum of the swing to the center of gravity of the transported object, as well as the information indicating the start position and the target position. Accepts the input of multiple teacher data representing the steady speed and acceleration / deceleration timing in the middle,

When the transported object reaches the target position by inputting at least the length from the fulcrum of the swing to the center of gravity of the transported object, and unknown information indicating the start position and the target position based on a plurality of teacher data. The process of learning the learning model is executed so as to output the steady speed and the timing of acceleration / deceleration during transportation of the object to be transported in order to minimize the shaking.

According to the present disclosure, it is possible to easily suppress the shaking of the transported object when it is stopped.

Overall configuration diagram of the transport device to which the transport control device according to the embodiment of the present disclosure is applied. Schematic block diagram showing the configuration of the control unit A diagram showing a speed pattern in which the timing of acceleration and deceleration is line-symmetrical. Diagram showing the configuration of a neural network Flowchart showing processing performed at the time of learning in this embodiment A flowchart showing the processing performed at the time of transport control in the present embodiment. Diagram showing teacher data used for learning neural networks Table showing learning evaluation results The figure which shows the output result when the combination of the length and the distance which is not in the teacher data is input The figure which shows the simulation result which carried the object to be conveyed using the steady-state velocity and acceleration time output from a trained model. Diagram showing transport paths with different heights between the start position and the target position A diagram showing a speed pattern in which the timing of acceleration and deceleration is non-axisymmetric. Diagram showing speed pattern for acceleration / deceleration in multiple stages The figure which shows the speed pattern which becomes S-shaped acceleration / deceleration Schematic block diagram showing other configurations of the control unit

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. FIG. 1 is an overall configuration diagram of a transport device to which the transport control device according to the embodiment of the present disclosure is applied. As shown in FIG. 1, the transport device 1 according to the present embodiment has a linear guide member 3 and a guide member connected to the upper ends of two columns 2A and 2B and two columns 2A and 2B, respectively. It includes a drive unit 4 that is moved along the 3, a moving mechanism 5 for moving the drive unit 4, and a control unit 6 that controls the drive of the drive unit 4. The drive unit 4 and the control unit 6 constitute the transfer control device of the present embodiment.

The moving mechanism 5 for moving the drive unit 4 includes a motor 10 and a ball screw 11 rotated by the motor 10. The motor 10 is attached to the guide member 3 on the front side in the transport direction of the transport start position Ps of the transported object 7 transported by the drive unit 4. The ball screw 11 extends along the guide member 3. One end of the ball screw 11 is connected to the motor 10, and the other end is rotatably supported by the bearing 12. The bearing 12 is attached to the guide member 3 at a position exceeding the target position Pe for transportation in the transportation direction of the object to be transported 7. As a result, when the ball screw 11 is rotated by rotating the motor 10, the drive unit 4 moves along the guide member 3 in the direction corresponding to the rotation direction. A hook 4a for suspending the object to be transported 7 is attached to the drive unit 4.

Further, an encoder 13 is connected to the motor 10. The encoder 13 detects the position of the drive unit 4 according to the rotation of the motor 10 and outputs the detection signal to the control unit 6.

The moving mechanism 5 for moving the drive unit 4 is not limited to the above configuration, and any configuration can be adopted as long as the drive unit 4 can be moved along the guide member 3.

In the present embodiment, the object to be transported 7 is, for example, an endoscopic flexible tube described in Japanese Patent Application Laid-Open No. 2011-183115. The object to be transported 7 is held and transported by the holding member 8. A hook 8a is attached to the holding member 8 on the side opposite to the side that holds the object to be transported 7. The object to be transported 7 held by the holding member 8 is swayably suspended from the drive unit 4 by hooking the hook 8a on the hook 4a of the drive unit 4. As a result, when the drive unit 4 is stopped after moving, the object to be transported 7 swings according to the law of inertia.

FIG. 2 is a schematic block diagram showing the configuration of the control unit. As shown in FIG. 2, the control unit 6 includes a CPU (Central Processing Unit) 21, a motion controller 22, a servo amplifier 23, a storage unit 24, a trained model 25, and a learning unit 26. Further, an input unit 27 including a keyboard and a mouse, and a display unit 28 including a liquid crystal monitor and the like are connected to the control unit 6. In the present embodiment, the input unit 27 also has a function such as a communication interface for inputting data to the control unit 6.

The steady speed and acceleration / deceleration timing of the drive unit 4 are input to the motion controller 22 from the learned model 25. Further, a position signal indicating the position of the drive unit 4 is input to the motion controller 22 via the servo amplifier 23. The motion controller 22 outputs a drive signal for driving the motor 10 to the servo amplifier 23 in response to an instruction from the CPU 21, an output of the learned model 25, and an input position signal. The trained model 25 will be described later.

The drive signal from the motion controller 22 and the detection signal from the encoder 13 are input to the servo amplifier 23. The servo amplifier 23 outputs a drive signal to the motor 10, whereby the motor 10 rotates to drive the drive unit 4. At this time, the drive unit 4 is accelerated / decelerated according to the acceleration / deceleration timing output by the learned model 25, and the rotation of the motor 10 is controlled so that the drive unit 4 has the steady speed output by the learned model 25. NS.

The storage unit 24 is composed of a semiconductor memory, and serves as a work area when the control unit 6 performs various controls.

Next, the trained model 25 will be described. In the present embodiment, the control unit 6 drives the drive unit 4 to move the drive unit 4 from the start position Ps to the target position Pe, so that the object to be transported is swayably suspended from the drive unit 4. 7 is conveyed from the start position Ps to the target position Pe. When the trained model 25 is input with information representing the length L0 from the fulcrum P0 of the swing of the transported object 7 to the center of gravity G0 of the transported object 7 shown in FIG. 1, the start position Ps, and the target position Pe. , To output the steady speed and acceleration / deceleration timing when the transported object 7 is transported from the start position Ps to the target position Pe in order to minimize the shaking when the transported object 7 reaches the target position Pe. Is being learned. In the present embodiment, the information representing the start position Ps and the target position Pe is the distance D0 between the start position Ps and the target position Pe shown in FIG. The fulcrum P0 of the swing is a point where the hook 4a and the hook 8a come into contact with each other when the object to be transported 7 is suspended from the drive unit 4. More specifically, the position of the fulcrum P0 of the swing can be specified by swinging the transported object 7 in a state where the transported object 7 is suspended from the drive unit 4.

In the present embodiment, a line-symmetrical velocity pattern is used to drive the drive unit 4 as shown in FIG. In the speed pattern shown in FIG. 3, the drive unit 4 is accelerated in the section z1, travels at a constant speed in the section z2 at a steady speed V0, and is decelerated in the section z3. In the speed pattern shown in FIG. 3, the section z1 and the section z3 have the same time length. Therefore, in the present embodiment, if the length of time in the section z1, that is, the acceleration time T1 from the start of acceleration to the stop of acceleration and the steady speed V0 in the section z2 are determined, the drive unit 4 Therefore, the object to be transported 7 can be moved from the start position Ps to the target position Pe. Therefore, in the present embodiment, the acceleration time T1 is used as the acceleration / deceleration timing output by the trained model 25. Therefore, in the present embodiment, the trained model 25 is constructed so as to output the steady velocity V0 and the acceleration time T1 by inputting the length L0 and the distance D0.

FIG. 4 is a diagram schematically showing the trained model 25. The trained model 25 includes a neural network having a three-layer structure of an input layer 31, a hidden layer 32, and an output layer 33. The layer of the neural network is not limited to three layers, and may be appropriately determined to be two layers or four layers or more in consideration of learning accuracy and learning load.

As shown in FIG. 4, the input layer 31 has two units u1-1 and u1-2, and the length L0 and the distance D0 are input to each of them. Further, the output layer 33 has two units u3-1 and u3-2, from which a steady speed V0 and an acceleration time T1 are output. The hidden layer 32 has, for example, 10 units u2-1 to u2-10. In this case, the number of connections between the input layer 31, the hidden layer 32, and the output layer 33 in the trained model 25 is 2 × 10 + 2 × 10 = 40.

The learning unit 26 performs the learning of the neural network. During training, the teacher data is used to determine the weight of the connections between each unit of each layer in the neural network. In the present embodiment, the transported object 7 is the target position with respect to the combination of the length L0 from the fulcrum P0 of the swing to the center of gravity of the transported object 7 and the distance D0 between the start position Ps and the target position Pe. A plurality of teacher data are prepared in which a combination of a steady velocity V0 and an acceleration time T1 during transportation of the object to be transported 7 for minimizing shaking when reaching Pe is defined. The neural network corresponds to the learning model of the present disclosure.

Here, in order to minimize the shaking when the transported object 7 reaches the target position Pe, for example, as described in Japanese Patent Application Laid-Open No. 2011-93333, the transported object 7 is moved from the start position Ps. The length of time for moving to the target position Pe may be n times the shaking cycle of the object to be transported 7 (n is a natural number). The shaking period C1 of the transported object 7 is represented by C1 = 2π√ (L0 / g). g is the gravitational acceleration. Therefore, as the teacher data, the length L0 and the various distances D0, and the length L0 and the distance D0 of the various objects to be transported 7 for moving the object to be transported 7 by the required time of n × C1. The data obtained by combining the steady velocity V0 and the acceleration time T1 corresponding to the above is used. Teacher data may be prepared by experimentally obtaining the steady-state velocity V0 and the acceleration time T1 with respect to the length L0 and the various distances D0 of the various objects to be transported 7. By preparing teacher data having a wide range of length L0 and distance D0, the performance of the trained model 25 can be improved.

The learning unit 26 learns the neural network and constructs the trained model 25 by updating the connection weight between the units in each layer of the neural network using a plurality of teacher data. As a result, when the unknown length L0 and the unknown distance D0 are input, the steady velocity V0 and the acceleration time T1 for minimizing the shaking of the transported object 7 are output from the trained model 25. Become.

After the learning is completed, it is preferable to evaluate the learning result of the trained model 25. In the evaluation of the trained model 25, the length L0 and the distance D0 included in a certain teacher data are input to the trained model 25, the steady velocity V0 and the acceleration time T1 output by the trained model 25, and the teacher data. This may be done by finding the correlation between the steady velocity V0 and the acceleration time T1. In this case, if the correlation is larger than a predetermined threshold value, it can be evaluated that the trained model 25 is properly trained. On the contrary, when the correlation is equal to or less than the threshold value, it can be evaluated that the trained model 25 is not properly trained. In this case, it is necessary to improve the accuracy of the trained model 25 by retraining the trained model 25.

On the other hand, when the trained model 25 outputs the steady velocity V0 and the acceleration time T1 for minimizing the shaking of the transported object 7 with respect to the combination of the unknown length L0 and the unknown distance D0, the unknown length is unknown. The steady velocity V0 and the acceleration time T1 with respect to the L0 and the unknown distance D0 may be used as new teacher data to further train the trained model 25.

Further, when a steady speed V0 and an acceleration time T1 capable of further suppressing the shaking of the transported object 7 are found for the combination of the length L0 and the distance D0 used as the teacher data, learning about the teacher data is found. Even if the trained model 25 is deleted from the trained model 25 and the trained model 25 is retrained using the new steady velocity V0 and the acceleration time T1 for the length L0 and the distance D0 used as the teacher data as the new teacher data. good.

The control unit 6 may be composed of a computer including a CPU, a semiconductor memory, a hard disk, and the like. In this case, the transfer control program and the learning program of the present disclosure are installed on the hard disk, and the drive of the drive unit 4 is controlled by the transfer control program.

Next, the processing performed in the present embodiment will be described. FIG. 5 is a flowchart showing the processing performed at the time of learning in the present embodiment. First, the learning unit 26 receives the input of the teacher data from the input unit 27 (step ST1), and uses the teacher data to learn the neural network which is the learning model (step ST2). As a result, the trained model 25 is constructed.

FIG. 6 is a flowchart showing a process performed at the time of transport control in the present embodiment. The control unit 6 receives inputs of the length L0 and the distance D0 for the object to be transported 7 from the input unit 27 (step ST11). Next, the trained model 25 outputs the steady velocity V0 and the acceleration time T1 by inputting the length L0 and the distance D0 (step ST12). Then, the control unit 6 conveys the object to be conveyed 7 according to the velocity pattern based on the steady velocity V0 and the acceleration time T1 (step ST13), and ends the process.

As described above, in the present embodiment, the start position Ps and the target position, which are at least the length L0 from the fulcrum P0 of the swing to the center of gravity G0 of the object 7 and the distance D0 between the start position Ps and the target position Pe. Based on the information representing Pe, the trained model 25 outputs the steady speed V0 when the object to be transported 7 is transported from the start position Ps to the target position Pe, and the acceleration time T1 which is the timing of acceleration / deceleration. Is. Therefore, if the drive unit 4 is driven by the steady speed V0 and the acceleration time T1 output by the trained model 25 and the object to be transported 7 is transported from the start position Ps to the target position Pe, the target position Pe is reached. The shaking of the object to be transported 7 can be minimized. Therefore, according to the present embodiment, it is possible to easily suppress the shaking of the transported object when it is stopped without separately providing a means such as a sensor.

Further, in the present embodiment, since the trained model 25 is used, the steady speed V0 suitable for unknown conditions is compared with the method using the rule map as in the method described in Patent Document 1. And the timing of acceleration / deceleration can be acquired.

Further, in the present embodiment, since the trained model 25 is used, even if the object to be transported 7 cannot be replaced by a simple model due to the occurrence of deflection such as an endoscope flexible tube. , It is possible to generate steady speed and acceleration / deceleration timing that minimizes shaking when stopped. Therefore, unlike the methods described in Patent Documents 2 and 3, it is not necessary to analyze the model system, and as a result, it is possible to easily suppress the shaking of the transported object when it is stopped.

Next, examples of the present disclosure will be described. In the embodiment of the present disclosure, in order to construct the trained model, the neural network having the three-layer structure shown in FIG. 4 was prepared as the training model. Further, as the teacher data, as shown in FIG. 7, various combinations of the length L0 and the distance D0, the steady speed V0, and the acceleration time T1 for minimizing the shaking of the transported object 7 reaching the target position Pe, and the steady speed V0 and the acceleration time T1 We prepared 14 teacher data corresponding to various combinations of. As shown in FIG. 3, the speed pattern is a line-symmetrical pattern in which the acceleration time and the deceleration time are the same.

As the neural network for constructing the trained model 25, a commercially available neural network fitting application was used. Specifically, a neural network was constructed using a Matlab Neural Fitting tool manufactured by MathWorks. The neural network fitting application is not limited to the above, and for example, Tensorflow's python or the like can be used.

Two types of algorithms used for learning were used, the Lebenberg-Marquardt method and the Bayesian normalized learning algorithm, in order to evaluate the learning performance. The Levenberg-Marquardt method is a network learning algorithm that updates the weight value according to the optimization of the Levenberg-Marquardt method, and has a feature of high calculation speed. In particular, the Lebenberg-Marquardt method is a least squares algorithm that finds a vector that minimizes the sum of squares function, assuming that the performance is the mean or sum of the square errors. The Bayes regularization algorithm is a network learning algorithm that updates the weight value according to the optimization of the Levenberg-Marquardt method, but it minimizes the linear combination of the squares of the weights in addition to the performance error. .. In the Bayes normalization learning algorithm, a regularization term is added to the evaluation function, and the linear combination is changed so that the generalization quality of the resulting network at the end of training is good (suppression of overfitting). NS.

FIG. 8 is a table showing the evaluation results of learning. As shown in FIG. 8, the evaluation value of the Bayesian normalization learning algorithm method was lower than that of the Levenberg-Marquardt method. Therefore, the Bayesian normalization learning algorithm was used for learning the neural network.

FIG. 9 shows the output result when a combination of length L0 and distance D0, which is not included in the teacher data, is input to the trained model 25. In FIG. 9, the round plot shows the steady velocity V0 and the acceleration time in each case of the length L0 = 1.65 m, 1.2 m, 0.9 m and the distance D0 = 1.1 m included in the teacher data. Indicates T1. The square plot shows the steady velocity V0 and the acceleration time T1 for each of the lengths L0 = 1.65m, 1.2m, 0.9m and the distance D0 = 0.3m included in the teacher data. The steady velocity output from the trained model 25 when the triangular plot is for length L0 = 1.65m, 1.2m, 0.9m and distance D0 = 0.6m, which are not included in the teacher data. V0 and acceleration time T1 are shown.

FIG. 10 is a diagram showing a simulation result when the object to be transported 7 is transported by using the steady speed V0 and the acceleration time T1 output from the trained model 25. The simulation results shown in FIG. 10 show the steady velocity V0 = 0.26 m / sec and the acceleration time T1 = 1 output from the trained model 25 when the length L0 = 1.65 m and the distance D0 = 0.6 m are input. .6 sec is used. In FIG. 10, the solid line shows the result of the transport simulation, and the broken line shows the swing period of the object to be transported 7 in the simulation. As shown in FIG. 10, the time required for transporting the transported object 7 based on the output result of the trained model 25 substantially coincides with the swing period of the transported object 7. Therefore, it is possible to suppress the shaking of the transported object 7 after the transported object 7 is transported to the target position Pe.

In the above embodiment, as the output of the trained model 25, the length of time until the acceleration is stopped may be used instead of the acceleration time T1.

Further, in the above embodiment, the length L0 and the distance D0 are used as the inputs of the trained model 25, but the input is not limited thereto. Instead of the distance D0, the three-dimensional coordinates of the start position Ps and the target position Pe may be used. In this case, since the length L0 and the three-dimensional coordinates (total of 6) of the start position Ps and the target position Pe are input to the trained model 25, the input layer 31 of the neural network has 7 units. Will have. The three-dimensional coordinates of the target position Pe based on the start position Ps may be input to the trained model 25. In this case, since the length L0 and the three-dimensional coordinates (three) of the target position Pe are input to the trained model 25, the input layer 31 of the neural network has four units. ..

Further, in the above embodiment, the height difference between the start position Ps and the target position Pe may be further used as an input to the trained model 25. In this case, the trained model 25 is constructed so as to output the steady speed V0 and the acceleration / deceleration timing by inputting the height difference in addition to the length L0 and the distance D0. Therefore, the input layer 31 of the neural network has three units. As a result, as shown in FIG. 11, even when the heights of the start position Ps and the target position Pe are different, the steady speed V0 capable of suppressing the shaking of the transported object 7 that has reached the target position Pe And the acceleration time T1 will be output from the trained model 25. When the three-dimensional coordinates of the start position Ps and the target position Pe or the three-dimensional coordinates of the target position Pe based on the start position Ps are used as the input of the trained model 25, the input to the trained model 25 is performed. The height difference can be included.

Further, in the above embodiment, as inputs to the trained model 25, in addition to the length L0 and the distance D0, the characteristics of the object to be transported 7, for example, the weight, length, projected area and rigidity of the object to be transported. Etc. may be further used. The projected area is the area when the object to be transported 7 is projected in the moving direction, and when the object to be transported 7 is cylindrical, the projected area is obtained by the product of the outer diameter and the length of the object to be transported 7. Can be done. Further, the Young's modulus of the object to be transported 7 may be used as the rigidity. In this case, the input layer 31 of the neural network has a number of units corresponding to the inputs. As a result, the object to be transported 7 can be transported at a steady speed V0 and the timing of acceleration / deceleration in consideration of the characteristics of the object to be transported 7, and as a result, the target position Pe is set according to the characteristics of the object to be transported 7. The shaking of the transported object 7 when it reaches can be minimized.

Further, in the above embodiment, as the velocity pattern of the object to be transported 7, a velocity pattern having line symmetry is used as shown in FIG. 3, but the velocity pattern is not limited to this. For example, as shown in FIG. 12, the present disclosure can be applied to speed patterns in which the acceleration time T1 and the deceleration time T2 are different. In this case, the trained model 25 is trained to output the steady velocity V0, the acceleration time T1 and the deceleration time T2 when the length L0 and the distance D0 are input. In this case, the output layer 33 of the neural network has three units. In this case, the trained model 25 may be constructed so as to output the acceleration end time t11, the deceleration start time t12, and the deceleration end time t13 from the start of movement instead of the deceleration time T2.

Further, as shown in FIG. 13, a speed pattern in which acceleration / deceleration is performed in multiple stages may be used. In this case, the trained model 25 is trained so as to output the times t1, t2, t3, t4, t5, t6, t7, t8, and t9, which are the timings of acceleration and deceleration from the start of movement. In addition, t1 is the first acceleration end time, t2 is the second acceleration start time, t3 is the second acceleration end time, t4, t6, t8 is the deceleration start time, and t5, t7, t9 is the deceleration end time. .. In this case, the output layer 33 of the neural network has 10 units including the steady velocity V0.

Further, in the above embodiment, a speed pattern in which acceleration / deceleration is performed at a constant acceleration is used, but the present invention is not limited to this. For example, as shown in FIG. 14, a speed pattern having an S-shaped acceleration / deceleration that reduces the acceleration at the start and stop of movement may be used. In this case, the trained model 25 may be trained so that the acceleration time T1 and the deceleration time T2 in the speed pattern that becomes the S-shaped acceleration / deceleration are output. The neural network may be trained so as to output either a constant acceleration or an S-shaped acceleration as an acceleration / deceleration pattern.

Further, in the above embodiment, the hook 4a is attached to the drive unit 4, the hook 8a of the holding member 8 is hooked on the hook 4a, and the object to be transported 7 is suspended from the drive unit 4, but the present invention is limited to this. It's not a thing. For example, any mechanism capable of rotatably suspending the object to be transported 7 in the moving direction can be used. In this case, by constructing the trained model 25 according to the mechanism to be used, the steady velocity V0 and the steady velocity V0 for minimizing the shaking when the transported object reaches the target position are determined according to the mechanism to be used. The acceleration / deceleration timing can be acquired. When a rotatably suspendable mechanism is used, the fulcrum of the swing is the rotation center of the mechanism.

Further, in the above embodiment, the control unit 6 includes the trained model 25 and the learning unit 26, but the present invention is not limited to this. As shown in FIG. 15, the setting unit 40 including the trained model 25 and the learning unit 26 may be provided separately from the control unit 6. In this case, by inputting the length L0 and the distance D0 from the input unit 41 of the setting unit 40, the steady speed V0 and the acceleration time T1 output by the trained model 25 are input to the control unit 6 and the drive unit 4 Will be controlled to drive.

Further, in the above embodiment, the trained model 25 and the learning unit 26 are provided in the control unit 6 or the setting unit 40, but the present invention is not limited thereto. The learning unit 26 may be provided separately from the trained model 25.

Further, in the above embodiment, the object to be transported 7 is transported along the linear guide member 3, but the transport path of the object to be transported 7 is not limited to a straight line but is curved. It may be a path including a straight line and a curved line. In this case, the neural network is trained using the teacher data in which various combinations of the length L0 and the distance D0 according to the path and various combinations of the steady velocity V0 and the acceleration / deceleration timing are associated with each other, and the trained model is trained. 25 will be constructed.

1 Transport device

2A, 2B prop

3 Guide member

4 Drive unit

4a hook

5 Movement mechanism

6 Control unit

7 Items to be transported

8 Holding member

8a hook

10 motor

11 ball screw

12 bearings

13 encoder

21 CPU

22 Motion controller

23 Servo amplifier

24 Memory

25 Trained model

26 Learning Department

27 Input section

28 Display

30 Neural network

31 Input layer

32 hidden layer

33 Output layer

40 Setting section

41 Input section

D0 distance

G0 Center of gravity of the object to be transported

L0 length

P0 swing fulcrum

Ps start position

Pe target position

T1 acceleration time

T2 deceleration time

t1-t9 time constant

t11, t12, t13 hours

u1-1, u1-2 ..., u3-1, u3-2 units

z1 to z3 section

V0 steady speed

Claims (16)

A transport control device that swingably suspends an object to be transported and transports the object to be transported from a start position to a target position.

At least, by inputting the length from the fulcrum of the swing to the center of gravity of the object to be transported, and the input of information indicating the start position and the target position, the shaking when the object to be transported reaches the target position is minimized. A trained model that has been trained to output the steady speed and acceleration / deceleration timing during transportation of the object to be transported, and

It is provided with an input unit that accepts input of information representing the length, the start position, and the target position.

Steady speed and acceleration / deceleration timing when the object to be transported is transported from the start position to the target position by the trained model based on the input length and information representing the start position and the target position. Transport control device that outputs.

The transport control device according to claim 1, wherein the information representing the start position and the target position includes the three-dimensional coordinates of the start position and the target position.

The transport control device according to claim 1, wherein the information representing the start position and the target position includes a distance between the start position and the target position.

The transport control device according to claim 1, wherein the information representing the start position and the target position includes a distance between the start position and the target position and a height difference between the start position and the target position.

The transport control device according to any one of claims 1 to 4, wherein the acceleration / deceleration timing includes an acceleration time or an acceleration end time.

The transport control device according to any one of claims 1 to 4, wherein the acceleration / deceleration timing includes an acceleration end time, a deceleration start time, and a deceleration end time.

The acceleration / deceleration timing includes a first acceleration end time, at least one second acceleration start time, the at least one second acceleration end time, at least one deceleration start time, and at least one deceleration end time. The transport control device according to any one of claims 1 to 4.

A drive unit driven along a path including the start position and the target position,

The transport control device according to any one of claims 1 to 7, further comprising a control unit that controls the drive unit according to the steady speed and the acceleration / deceleration timing output from the trained model.

The transport control device according to claim 8, wherein the drive unit includes a hook for suspending the object to be transported.

The transport control device according to claim 8, wherein the drive unit includes a suspending mechanism that rotatably suspends the transported object in the moving direction of the transported object.

A learning device that learns a learning model for suspending an object to be transported so as to swingably and determining the timing of acceleration / deceleration when the object to be transported is transported from a start position to a target position.

At least to minimize the swing when the transported object reaches the target position with respect to the length from the fulcrum of the swing to the center of gravity of the transported object, and the information representing the start position and the target position. An input unit that accepts input of a plurality of teacher data representing the steady speed and acceleration / deceleration timing during transportation of the object to be transported.

Based on the plurality of teacher data, at least the length from the fulcrum of the swing to the center of gravity of the object to be transported, and unknown information representing the start position and the target position are input to cause the object to be transported. A learning device including a learning unit that learns the learning model so as to output a steady speed and acceleration / deceleration timing during transportation of the object to be transported in order to minimize shaking when the target position is reached.



It is a learned model for swaying the object to be transported and determining the timing of acceleration / deceleration when the object to be transported is transported from the start position to the target position.

At least, by inputting the length from the fulcrum of the swing to the center of gravity of the object to be transported, and the input of information indicating the start position and the target position, the shaking when the object to be transported reaches the target position is minimized. A trained model that has been trained to output the steady speed and the timing of acceleration / deceleration during transportation of the object to be transported.

This is a transport control method in which the object to be transported is swayably suspended and the object to be transported is transported from a start position to a target position.

At least, by inputting the length from the fulcrum of the swing to the center of gravity of the object to be transported, and the input of information indicating the start position and the target position, the shaking when the object to be transported reaches the target position is minimized. For a trained model that has been trained to output the steady speed and acceleration / deceleration timing during transportation of the object to be transported, input of information representing the length, the start position, and the target position is accepted.

Steady speed and acceleration / deceleration timing when the object to be transported is transported from the start position to the target position by the trained model based on the input length and information representing the start position and the target position. Transport control method to output.

It is a learning method for learning a learning model for suspending an object to be transported so as to be swingable and determining the timing of acceleration / deceleration when the object to be transported is transported from a start position to a target position.

At least, in order to minimize the shaking when the transported object reaches the target position with respect to the length from the fulcrum of the swing to the center of gravity of the transported object and the information representing the starting position and the target position. Accepts the input of a plurality of teacher data representing the steady speed and the timing of acceleration / deceleration during the transportation of the object to be transported.

Based on the plurality of teacher data, at least the length from the fulcrum of the swing to the center of gravity of the object to be transported, and unknown information representing the start position and the target position are input to make the object to be transported said. A learning method for learning the learning model so as to output the steady speed and the timing of acceleration / deceleration during transportation of the object to be transported in order to minimize the shaking when the target position is reached.

It is a transport control program that causes a computer to execute a transport control method of suspending an object to be transported so as to be swingable and transporting the object to be transported from a start position to a target position.

At least, by inputting the length from the fulcrum of the swing to the center of gravity of the object to be transported, and the input of information indicating the start position and the target position, the shaking when the object to be transported reaches the target position is minimized. A procedure for accepting input of information representing the length, the start position, and the target position for a trained model that has been trained to output the steady speed and acceleration / deceleration timing during transportation of the object to be transported. ,

Based on the input information representing the length and the start position and the target position, the steady speed and acceleration / deceleration timing when the object to be transported is transported from the start position to the target position by the trained model. A transport control program that causes the computer to execute the procedure for outputting.

A learning program that causes a computer to execute a learning method for learning a learning model for swaying the object to be transported and determining the timing of acceleration / deceleration when the object to be transported is transported from a start position to a target position. And

At least, in order to minimize the shaking when the transported object reaches the target position with respect to the length from the fulcrum of the swing to the center of gravity of the transported object and the information representing the starting position and the target position. The procedure for accepting the input of a plurality of teacher data representing the steady speed and the timing of acceleration / deceleration during the transportation of the object to be transported, and

Based on the plurality of teacher data, at least the length from the fulcrum of the swing to the center of gravity of the object to be transported, and unknown information representing the start position and the target position are input to cause the object to be transported. A learning program that causes a computer to execute a procedure for learning the learning model so as to output a steady speed and acceleration / deceleration timing during transportation of the object to be transported in order to minimize shaking when the target position is reached.

Images