JP2021041820A - Unmanned flight body and computer program therefor - Google Patents

Abstract

To provide computer program which outputs crash probability with a view to use of an unmanned flight body during actual use.SOLUTION: A computer program receives, as input, at least a current value and a voltage value of one or more rotary wing motors provided on an unmanned flight body having a plurality of rotary wings, and causes a computer to work so as to output a crash probability of the unmanned flight body when a predetermined time t has lapsed via a learned model mentioned below on the basis of the above input. The learned model is generated by mechanical learning with use of teacher data. The teacher data includes at least: (A1) a current value and (A2) a voltage value of the rotary wing motor in flight record of the unmanned flight body; and (B) binary data as to whether the unmanned flight body has been crashed when the time t has lapsed.SELECTED DRAWING: Figure 1

Description

The present invention relates to an unmanned flying object, a flight controller and an electronic velocity controller used therein, and a computer program for driving them.

Recently, the development of unmanned aerial vehicles called drones has been active. A typical unmanned aerial vehicle is a multicopter equipped with a plurality of rotor blades. For example, an unmanned flying object having a plurality of rotor blades arranged radially through a shaft can be mentioned. In such an unmanned flying object, it flies by rotating a plurality of rotor blades at the same time in a well-balanced manner.

The balance of the plurality of rotor blades is usually controlled by a device called a flight controller provided in one unmanned projectile. The flight controller determines the rotation speed and rotation direction of each rotor, and the determined rotation speed and rotation direction are transmitted to each rotor. Each rotor is equipped with an electronic speed controller (ESC) that supplies power to the motor of one rotor. The commands for the rotation speed and rotation direction of each rotor specified by the flight controller are input to the electronic speed controller attached to the rotor, and the power supply to the rotor is determined so as to realize the above commands. .. In this way, the flight controller issues a rotation command to each rotor in consideration of the balance of the entire unmanned projectile, and the issued command is supplied to each rotor by the electronic velocity controller attached to each rotor. Converted to power.

In an unmanned flying object, the operation of each rotor may become unstable, resulting in an unexpected crash or loss of control. Various efforts have been made to prevent such falls in unmanned flying objects. For example, the accuracy of simulations that virtually verify flight conditions under a wide variety of conditions is quite high, and it is possible to acquire a large amount of virtual flight data without actually crashing an unmanned flying object. ..

However, in many simulations of unmanned flying object flight, it is necessary to input the flight path, meteorological conditions, etc. into the computer in advance. Therefore, it is difficult to apply the virtual flight data obtained by the simulation to actual use as it is. This is because, in actual use, an unintended difference from the conditions in the simulation occurs due to a sudden change in weather conditions, and as a result, the virtual flight data and the actual flight data may differ significantly. ..

In view of such circumstances, the present invention has a computer program that outputs a crash probability, an electronic speed controller that stores the program, and such an electronic speed controller, with a view to using an unmanned flying object in actual use. The subject is to provide an unmanned projectile with an electronic velocity controller.

As a result of diligent studies by the present inventors, the present invention having the following contents has been completed.
[1] At least the current value and the voltage value of one or more rotary wing motors provided in an unmanned flying object having a plurality of rotary wings are accepted as inputs, and based on the inputs, they are predetermined via the following learned model. It is a computer program for operating a computer so as to output the crash probability of the unmanned flying object when the time t elapses, and the trained model is characterized by being generated by machine learning using teacher data. The teacher data is based on the (A1) current value and (A2) voltage value of the rotary wing motor in the flight record of the unmanned flying object, and (B) the unmanned flying object when the time t has elapsed. The above computer program, which includes at least binary data, whether or not it has crashed.
[2] The flight record of the unmanned flying object includes a record obtained from at least one of a flight simulation of the unmanned flying object and an actual flight. The computer program of [1].
[3] Each rotary wing motor is provided with an electronic speed controller for controlling the rotary wing motor, and the input further includes at least one of the temperature of the rotary wing motor and the temperature of the electronic speed controller, and the teacher. The computer program of [1] or [2], wherein the data further includes (A3) at least one of the rotor motor temperature and the electronic speed controller temperature.
[4] The computer programs of [1] to [3], wherein the input further includes the rotation speed of the rotary wing motor, and the teacher data further includes the rotation speed of the rotary wing motor (A4).
[5] The input includes the current value and the voltage value of each of the plurality of rotary blade motors.
The teacher data includes (A1) current value and (A2) voltage value of the plurality of rotorcraft motors in the flight record of the unmanned flying object, and (B) the unmanned flying object when the time t has elapsed. Includes at least binary data on whether or not it has crashed,
Computer programs of [1] to [4].
[6] An electronic speed controller for controlling one rotary wing motor provided in an unmanned flying object having a plurality of rotary wings, characterized in that it stores the computer programs of [1] to [4]. , Electronic speed controller.
[7] A flight controller for an unmanned flying object having a plurality of rotary wings, and each rotary wing of the unmanned flying object has an electronic speed for controlling a rotary wing motor and the rotary wing motor, respectively. The flight controller includes a controller, the flight controller controls a plurality of electronic speed controllers, and stores the computer program of [5].
[8] An unmanned flying object having a plurality of rotor blades and an electronic velocity controller of [6] connected to each of the rotary blades.
[9] An unmanned flying object having a plurality of rotors, an electronic velocity controller connected to each of the rotors one by one, and a flight controller of [7].

According to the present invention, the teacher data used for generating the trained model can be generated in a wide variety by flight simulation. Therefore, a huge amount of data can be used as teacher data without actually flying (and crashing) the unmanned projectile, and as a result, it is expected that the accuracy of the trained model will be improved.

The program of the present invention using the trained model thus generated is when the "time t" elapses from that point when at least the "current and voltage values of the rotorcraft" at a certain point in time are input. The crash probability of the unmanned projectile is output. Therefore, if a high crash probability is output, it will be easier to take countermeasures such as taking evasive action by the time "time t" elapses, and it is expected to contribute to the prevention of crashes of unmanned flying objects during actual use. Will be done.

It is a schematic diagram of an example of an unmanned flying object of the present invention. It is the schematic of the computer program by this invention. It is a schematic diagram of an example of a flight simulation of an unmanned flying object. It is a conceptual diagram of an example of a PID controller. It is a schematic diagram of an example of a flight simulation of an unmanned flying object. It is a conceptual diagram of an example of an adaptive FIR controller. It is a schematic diagram of an example of a flight simulation of an unmanned flying object. It is a conceptual diagram of an example of a deep learning model. It is a schematic diagram of an example of a flight simulation of an unmanned flying object.

Hereinafter, the present invention will be described in detail with reference to the drawings as appropriate. The illustrated embodiments are not intended to limit the present invention, but are merely examples.

FIG. 1 is a schematic view of an example of an unmanned flying object of the present invention. The unmanned flying object of the present invention is configured to fly by remote control without a human getting in, and has at least two rotor blades. In the present invention, the power source of the rotary blade is assumed to be electric power. Typically, the power source is a storage battery (not shown) provided in an unmanned projectile.

The unmanned flying object of FIG. 1 has four shafts 300 extending radially from the center, a rotary wing 200 provided at the tip of each shaft 300, and an electronic velocity controller (ESC) 100 attached to each rotary wing 200. And a flight controller (flight controller) 400 provided at the center. As long as it has a plurality of rotor blades, the structure of the unmanned projectile is not particularly limited. The number of shafts and rotors is preferably 4-10.

The flight controller 400 calculates a duty ratio proportional to the voltage applied to the drive motor, which changes from moment to moment, as a command given to each electronic speed controller 100 as an example according to the traveling direction and speed desired as an unmanned flying object. To do. Each electronic speed controller 100 converts electric power according to a transmitted duty ratio instruction and supplies it to a drive motor (not shown) of each rotor 200. The communication between the flight controller 400 and the operator can be performed by remote control using radio waves such as radio waves.

Usually, the electronic speed controller 100 and the rotor blade 200 have a one-to-one correspondence. Specifically, one electronic speed controller 100 is provided to control one rotary blade 200. The connection between the electronic speed controller 100 and the flight controller 400 is not particularly limited, and may be wired or wireless. The connection between the electronic speed controller 100 and the rotary blade 200 is not particularly limited, and may be a power line or non-contact transmission. These connections are not depicted in the drawings.

According to the present invention, the electronic speed controller and / or the flight controller stores a computer program described in detail below. According to the present invention, a computer program is responsible for the following inputs and outputs.

As input, the current value and voltage value of the rotary blade motor are essential information. Preferably, one or more of the rotor speed, the rotor motor temperature and the electronic speed controller temperature can be further treated as inputs. Further, it may be further adopted as information for inputting parameters other than the above-listed information.

The above-mentioned current value, voltage value, and other parameters can be set or measured independently in each of a plurality of rotorcraft motors. However, in the present invention, the above-mentioned parameters in at least one rotorcraft motor are used as inputs. It is preferably used, preferably using the above parameters in a plurality of rotorcraft motors as inputs, and more preferably using the above parameters in all rotorcraft motors as inputs.

According to the present invention, the output of the computer program is the probability of the unmanned flying object crashing when a predetermined time t has elapsed. The time t is predetermined in the creation of the computer program. For example, when a program is created such that "the program outputs a crash probability after 5 seconds", the time t is "5 seconds". The time t is not particularly limited. The crash probability is generally expressed by a value between the crash probability of 0% and the crash probability of 100%.

According to the present invention, there is provided a computer program that outputs the crash probability of an unmanned flying object when the time t elapses from the input of each parameter as described above. These inputs and outputs are performed via the trained model described in detail below. The trained model is generated by machine learning using teacher data.

Here, as the teacher data, at least the following data is indispensably included.
(A1) Current value of rotorcraft motor at a certain time,
(A2) The voltage value of the rotorcraft motor at the time, and (B) binary data of whether or not the unmanned projectile has crashed when the time t has passed from the time.

By using the current value (A1) and voltage value (A2) as teacher data as described above, a crash due to a low battery (mainly related to the voltage value) and a crash of a motor factor due to a load such as wind (mainly a current value). (Related to), or the elements of the ESC factor crash (mainly related to the current value) due to arm short circuit due to burning of electronic components can be incorporated into the learning. Each of these can be judged individually by the threshold value, but even if both are within the normal range, it is assumed that a crash may occur due to multiple effects, and in such a case, it is difficult to set the threshold value of the individual parameter. The learning model is proposed as one of the solutions to judge such a complex case.

Further, in the binary data in (B), it is preferable that the non-crash is represented by 0 and the crash is represented by 1. By including such binary data as teacher data, if the learning model judges that the probability of a crash after time t seconds is low, the output approaches "0", and if it is judged to be high, the crash is "1". The output can be used as a "crash probability", such as approaching.

The teacher data is not limited to the above, and preferably includes one or more of the following data.
(A3) At least one of the rotor blade motor temperature and the electronic speed controller temperature at the time mentioned above.
(A4) The number of rotations of the rotary blade motor at the above time.
Data other than the above may be included as teacher data. For example, the position data of the unmanned flying object at the time, the data from the time to the past time, and the time difference of the data can be applied as the teacher data. is there.

These teacher data may be acquired from the actual flight data of the unmanned flying object, or may be acquired from the flight simulation of the unmanned flying object.

For example, it is assumed that the following data exists in one rotary wing motor of an unmanned flying object by flight simulation.

Time (seconds) Current value (A) Voltage value (V) Data No.
116.5 15.0 14.5 1
117.0 15.6 14.5 2
117.5 15.1 14.5 3
118.0 15.5 14.4 4
118.5 15.7 14.4 5
119.0 30.7 14.4 6
119.5 44.3 14.4 7
120.0 44.9 14.4 8
120.5 44.2 14.3 9
121.0 95.3 14.3 10
121.5 110.2 14.3 11
122.0 − − (Crash) 12

The above is the data of the final 5.5 seconds (116.5 to 122.0 seconds) when the aircraft crashed 122 seconds after the flight in a flight simulation. "Data No." is numbered for the convenience of the following description. More specifically, regarding the above data, the data No. Up to 5, the state where the aircraft is hovering, data No. In the state where the current value increased due to the wind in No. 6, the data No. In 10, the motor is locked due to the load of the wind.

Here, as an example, consider creating a computer program when the time t is "2 seconds". In "Data No. 1", the current value (A1) is 15.0 amperes and the voltage value (A2) is 14.5 volts. Then, the state of the unmanned projectile when "time t (2 seconds)" elapses is "non-crash (value = 0)". Therefore, the teacher data obtained from "Data No. 1" is "the current value (A1) is 15.0 amperes, the voltage value (A2) is 14.5 volts, and the crash state value (B) is 0".

Based on the same consideration, the teacher data obtained from "Data No. 2" is "Current value (A1) is 15.6 amperes, voltage value (A2) is 14.5 volts, and crash state value (B) is 0". Is. The teacher data obtained from "Data No. 3" is "the current value (A1) is 15.1 amperes, the voltage value (A2) is 14.5 volts, and the crash state value (B) is 0".
From the consideration so far, three teacher data have already been acquired.

Next, the teacher data obtained from "Data No. 8" will also be considered. It should be noted here that the state of the unmanned projectile when "time t (2 seconds)" elapses from the time of "data No. 8" is "crash (value = 1)". Therefore, the teacher data obtained from "Data No. 8" is "the current value (A1) is 44.9 amperes, the voltage value (A2) is 14.4 volts, and the crash state value (B) is 1." Similarly, the teacher data obtained from "Data No. 9" is "current value (A1) is 44.2 amperes, voltage value (A2) is 14.3 volts, and crash state value (B) is 1." ..

As in the above example, a huge amount of teacher data can be obtained from actual flight data and flight simulation of an unmanned flying object. By subjecting the obtained teacher data to machine learning, the above-mentioned trained model can be obtained. The specific mode of machine learning for generating a trained model from a large amount of teacher data is not particularly limited, and conventional techniques in machine learning can be appropriately adopted.

FIG. 2 is a schematic diagram of a computer program according to the present invention. The inference model represented by reference numeral 211 corresponds to the trained model described above. When at least the voltage and current of the rotary blade motor of the unmanned flying object are input, the crash probability of the unmanned flying object when the time t elapses is output as the crash determination 212 based on the inference model 211.

According to the present invention, the electronic speed controller (ESC) for controlling each rotor motor preferably stores the computer program described above. More preferably, a plurality of electronic speed controllers mounted on the unmanned projectile each store the above-mentioned computer program. Further, the above-mentioned computer program may be stored in a flight controller (flight controller) that controls a plurality of electronic speed controllers.

As described above, an unmanned projectile having an electronic speed controller or a flight controller that stores a computer program is also within the scope of the present invention.

Hereinafter, the implementation of the present invention will be described in a somewhat specific manner. First, an example of a flight model for obtaining the above-mentioned teacher data will be introduced.
FIG. 3 is a schematic diagram of an example of a flight simulation of an unmanned flying object. In this example, the deviation between the target position 21 at time t and the position at time t-1, the position at time t-1, the velocity at time t-1, and the acceleration at time t-1 are applied to each rotary blade of the unmanned projectile. A controller 1 that calculates the input voltage, a rotation speed conversion unit 11 that calculates the rotation speed of each rotary blade from the voltage, a force conversion unit 12 that calculates the force of each rotary blade from the rotation speed, and each rotation from the rotation speed. Torque conversion unit 13 that calculates the torque of the blades, current conversion unit 14 that calculates the current of each rotary blade from the torque, control device temperature conversion unit 15 that calculates the temperature of the motor control device from the voltage and current, and force conversion From the force calculated in Part 12, the dynamic model 2 that calculates the acceleration in the front-back direction, left-right direction, vertical direction of the unmanned projectile, and the roll angle, pitch angle, and yaw angle direction around each direction, and the acceleration. It has a speed conversion unit 16 that calculates the speed in six directions from the speed, and a position conversion unit 17 that calculates the position in the six directions from the speed. By inputting the target positions 21 in 6 directions for each time into the model of FIG. 1, the voltage data 31, the rotation speed data 32, the current data 33, the temperature data 34, the acceleration data 35, and the speed data for each time of the unmanned flying object are input. 36, position data 37 (hereinafter referred to as flight data) can be acquired. By giving various disturbances to the target position 21 and the output of each conversion unit 11 to 16, it is possible to simulate whether or not an unmanned projectile crashes, and if it crashes, acquire flight data before and after the crash. Become.

FIG. 4 is a conceptual diagram of an example of a PID controller. This can be adopted as the controller 1 in FIG. 3 described above. This PID controller inputs the deviation between the target position 21 at time t and the position at time t-1, and does not use the position itself at time t-1, the velocity at time t-1, or the acceleration at time t-1. It is composed of a P control unit 41 that performs proportional control using the input deviation, an I control unit 42 that performs integral control, and a D control unit 43 that performs differential control, and the sum of the outputs of the respective control units 41 to 43 is the voltage. It is configured to output as.

FIG. 5 is a schematic diagram of an example of a flight simulation of an unmanned flying object. In this example, the PID controller shown in FIG. 4 is adopted as the controller 1 in the flight simulation shown in FIG. In such a flight simulation, the flight and crash of an unmanned flying object are simulated by giving various disturbances to the target position 21 and the outputs of the conversion units 11 to 17. It is possible to obtain a large amount of various data such as current, voltage, and crash state for each time obtained at this time, and by using them as the above-mentioned teacher data, it is expected that the accuracy and accuracy of the trained model will be improved. ..

FIG. 6 is a conceptual diagram of an example of an adaptive FIR controller. This can be adopted as the controller 1 in FIG. 3 described above. This adaptive FIR filter takes the deviation between the target position 21 at time t and the position at time t-1 and the speed at time t-1 as inputs, and inputs the position itself at time t-1 and the speed and time at time t-1. The acceleration of t-1 is not used. The input deviations and velocities are input to the FIR filters 51 and 53, respectively, and the sum of the values calculated based on the filter coefficients of the respective filters is output as a voltage. The adaptive control units 52 and 54 calculate the filter coefficients of the FIR filters 51 and 53 at time t + 1 based on the deviation and the speed input at time t, and reflect them in the FIR filters 51 and 53 at time t + 1.

FIG. 7 is a schematic diagram of an example of a flight simulation of an unmanned flying object. In this example, the adaptive FIR controller shown in FIG. 6 is adopted as the controller 1 in the flight simulation shown in FIG. In such a flight simulation, the flight and crash of an unmanned flying object are simulated by giving various disturbances to the target position 21 and the outputs of the conversion units 11 to 17. It is possible to obtain a large amount of various data such as current, voltage, and crash state for each time obtained at this time, and by using them as the above-mentioned teacher data, it is expected that the accuracy and accuracy of the trained model will be improved. ..

FIG. 8 is a conceptual diagram of an example of a deep learning model. This can be adopted as the controller 1 in FIG. 3 described above. In the deep learning model, the outputs of the one-dimensional data conversion unit 61 and the one-dimensional data conversion unit 61 that collect the input deviations, positions, speeds, and accelerations as one-dimensional data are delayed by 1 to nth order (n is arbitrary). The nth-order lag element 62 that outputs data at time t + 1 to time t + n, the output of the one-dimensional data conversion unit 61 at time t, the output of the nth-order lag element 62 output at time t, and the voltage at time t-1. It is provided with an input data conversion unit 63 that collects data and data as one-dimensional input data to the deep learning network 64, and a deep learning network 64 that infers a voltage from the input data. The voltage inferred by the deep learning network 64 is fed back as data and becomes an input of the input data conversion unit 63 at time t + 1. For the learning of the deep learning network 64, the deviation, position, speed, acceleration, and voltage for each time acquired by simulating the PID control of FIG. 4 and the adaptive FIR filter control of FIG. 6 can also be used.

FIG. 9 is a schematic diagram of an example of a flight simulation of an unmanned flying object. In this example, the deep learning model shown in FIG. 8 is adopted as the controller 1 in the flight simulation shown in FIG. In such a flight simulation, the flight and crash of an unmanned flying object are simulated by giving various disturbances to the target position 21 and the outputs of the conversion units 11 to 17. It is possible to obtain a large amount of various data such as current, voltage, and crash state for each time obtained at this time, and by using them as the above-mentioned teacher data, it is expected that the accuracy and accuracy of the trained model will be improved. ..

Although some examples of flight simulations that can be used in the present invention have been introduced above, flight simulations other than these may be used, and actual flight data of an unmanned flying object may be used in whole or in part. May be good.

By using the computer program of the present invention that can generate a trained model by machine learning based on the teacher data obtained from such flight simulation and / or actual flight data and can be modeled as shown in FIG. 2 referred to above. From various input data including the current and voltage of the rotary wing motor that are observed every moment during the actual use of the unmanned flying object, the crash probability of the unmanned flying object when the time t elapses can be immediately obtained. If it is found early that the probability of a crash will increase, the chances and time for taking effective fall avoidance actions will increase, and as a result, the risk of an actual crash can be reduced.

100: Electronic speed controller 200: Rotor 300: Shaft 400: Flight controller

Claims (9)

At least the current value and voltage value of one or more rotor motors provided in an unmanned flying object having a plurality of rotor blades are accepted as inputs, and based on the inputs, a predetermined time t is performed via the following learned model. It is a computer program for making a computer function so as to output the crash probability of the unmanned flying object when the above has passed.
The trained model is characterized in that it is generated by machine learning using teacher data.
The teacher data includes (A1) current value and (A2) voltage value of the rotorcraft motor in the flight record of the unmanned flying object, and (B) the unmanned flying object crashes when the time t has elapsed. Includes at least binary data, whether or not
The above computer program. The computer program according to claim 1, wherein the flight record of the unmanned flying object includes a record obtained from at least one of a flight simulation of the unmanned flying object and an actual flight. Each rotorcraft motor is provided with an electronic speed controller for controlling the rotorcraft, the input further comprises at least one of the rotorcraft temperature and the electronic speed controller temperature, and the teacher data further includes. (A3) The computer program according to claim 1 or 2, which comprises at least one of a rotorcraft temperature and an electronic speed controller temperature. The computer program according to any one of claims 1 to 3, wherein the input further includes a rotation speed of the rotorcraft, and the teacher data further includes a rotation speed of the rotorcraft (A4). The input includes the current and voltage values of each of the plurality of rotorcraft motors.
The teacher data includes (A1) current value and (A2) voltage value of the plurality of rotorcraft motors in the flight record of the unmanned flying object, and (B) the unmanned flying object when the time t has elapsed. Includes at least binary data on whether or not it has crashed,
The computer program according to any one of claims 1 to 4. An electronic speed controller for controlling one rotary blade motor provided in an unmanned flying object having a plurality of rotary blades, characterized in that it stores the computer program according to any one of claims 1 to 4. Electronic speed controller. A flight controller for an unmanned projectile with multiple rotors,
Each rotary wing of the unmanned projectile is provided with a rotary wing motor and an electronic speed controller for controlling the rotary wing motor.
The flight controller controls a plurality of electronic speed controllers, and stores the computer program according to claim 5.
The flight controller. The unmanned flying object comprising a plurality of rotary blades and an electronic speed controller according to claim 6, one connected to each of the rotary blades. An unmanned flying object comprising a plurality of rotors, an electronic velocity controller connected to each of the rotors one by one, and a flight controller according to claim 7.

Images