RU78456U1 - AUTONOMOUS ADAPTIVE DEVICE FOR MOBILE OBJECT MANAGEMENT - Google Patents

Abstract

The utility model relates to the field of managing mobile objects and can be used to create systems for controlling the position of a material object in space in real time, for example, for semi-automatic or automatic parking of a car, for boarding an aircraft, in mooring systems for ships, in controlling closing and opening gates and doors, in production (loading and unloading), etc. Functionally, the mobile object control device contains three parts: neuroprocessor, consisting of consisting of a neuroprocessor, external read-only memory and external random access memory, switching, including a switching unit, and controller, including a controller with fuzzy logic, consisting of an analog-to-digital converter, digital-to-analog converter and built-in read-only memory and random access memory, and external permanent storage device, and the outputs of sensors installed on a mobile object, and the outputs of sensors, mechanisms, etc. the drives are connected to the input of the analog-to-digital converter of the controller, the inputs of the sensors. drive mechanisms are connected to the output of the digital-to-analog converter of the controller, and the inputs and outputs of the switching unit are connected to the inputs and outputs of the controller, the external read-only memory of the controller, the neuroprocessor and the external read-only memory of the neuroprocessor. The use of this device allows you to exclude the installation of a computer on a mobile object while providing automatic control of the object. The simultaneous use of a fuzzy controller making decisions, and a neuroprocessor training him, allows to increase accuracy and increase the reliability of the control device. Autonomous control device can be produced in series, which significantly reduces its cost.

Description

The utility model relates to the field of managing mobile objects and can be used to create systems for controlling the position of a material object in space in real time, for example, for semi-automatic or automatic parking of a car, for boarding an aircraft, in mooring systems for ships, in controlling closing and opening gates and doors, in production (loading and unloading), etc.

Currently, in automatic control systems for complex objects in conditions of great uncertainty and incompleteness of knowledge about the object and the external environment, several methods are used that focus on modeling based on expert knowledge and decision-making under conditions of fuzziness and uncertainty.

The well-known microcontroller Motorola 68HC08, using fuzzy logic, which can be used in a variety of control systems. The general structure of the microcontroller contains a phasing unit, a knowledge base, a decision unit, and a dephasing unit. The phasing unit converts the clear values measured at the output of the control object into fuzzy values described by linguistic variables in the knowledge base. The decision block uses fuzzy conditional rules embedded in the knowledge base to convert fuzzy input data to the required control actions of also a fuzzy nature. The dephasification unit converts fuzzy data from the output of the solution unit to a clear value, which is used to control the object [PC magazine WEEK / RE No. 37, 2004, p. 38].

However, fuzzy control systems are mainly applicable for a class of tasks in which the number of external parameters and control parameters is small and are not fully self-adjusting.

A known drilling process control system adopted for the prototype, comprising a downhole motor, a downhole pump for pumping flushing fluid, a telemetry system with sensors for monitoring parameters in the bottom, ground sensors for monitoring technological parameters, an information transfer unit and a control computer with software, a control database, which contains the design trajectory of the well, and a CAD database that contains design data on the design of the well, characteristics of the equipment, technologist the drilling process, geological and geophysical data, while the software is self-learning by accounting for CAD data (automatic design system) of the well and accounting for the previous

earlier, while drilling wells of the same cluster or field, the experience in making decisions was realized using artificial neural networks emulated on a neuroprocessor. The outputs from the computer are connected to the appropriate actuators [RF patent No. 2208153, IPC ЕВВ 44/00, 2001].

However, using only a trained neural network for control does not always allow achieving the desired result, since training the neural network and determining the required size of the training sample is a rather complicated process. Accordingly, a neural network is not always correctly trained. Also, a well-known system necessarily requires a computer.

When creating a utility model, the problem of creating an autonomous small-sized adaptive device for controlling a mobile object, previously trained by the operator, and the task of reducing the cost of manufacturing a control device were solved.

The problem is solved due to the fact that in a known device containing a neural processor with a neural network emulated on it, the database of the control object and environmental conditions, sensors installed on the mobile object, and sensors of actuators, according to the utility model, the device functionally contains three parts: neuroprocessor, consisting of a neuroprocessor, external permanent storage device and external random access memory, switching, including a switching unit and a controller, including a controller with fuzzy logic, consisting of an analog-to-digital converter, a digital-to-analog converter and built-in read-only memory and random access memory, and an external read-only memory, the outputs of sensors installed on a mobile object and the outputs of sensors of mechanisms drives are connected to the input of the analog-to-digital converter of the controller, sensor inputs. drive mechanisms are connected to the output of the digital-to-analog converter of the controller, and the inputs and outputs of the switching unit are connected to the inputs and outputs of the controller, the external read-only memory of the controller, the neuroprocessor and the external read-only memory of the neuroprocessor.

The use of this device allows you to exclude the installation of a computer on a mobile object while providing automatic control of the object. The simultaneous use of a fuzzy controller making decisions, and a neuroprocessor training him, allows to increase accuracy and increase the reliability of the control device.

Also, the inventive device can be produced in series, which significantly reduces its cost.

The utility model is illustrated by drawings, where in Fig.1 shows a General diagram

devices figure 2 - graphs of membership functions of the terms of linguistic variables:

a) input - “distance to the obstacle”; b) output - “steering angle”; figure 3 is a diagram of a car parking from various positions.

An autonomous adaptive control device for, for example, automatic parking of a car, according to its functions contains three parts: 1 - controller, making decisions, 2 - neuroprocessor, 3 - switching. The controller part 1 consists of a controller with fuzzy logic 4, including an analog-to-digital converter (ADC) 5, digital-to-analog converter (DAC) 6, built-in read-only memory (ROM) 7 and built-in random access memory (RAM) 8, from external read-only memory (ROM) of the controller 9 and an external random access memory (RAM) of the controller 10. The neuroprocessor part 2 contains a neuroprocessor 11, made, for example, based on the chip NM6403, external constant the detecting device (ROM) of the neuroprocessor 12 and the external random access memory (RAM) of the neuroprocessor 13. The switching part 3 includes a switching unit 14, made, for example, based on a programmable logic integrated circuit (FPGA), and a USB port 15 for connecting to a computer 16. The inputs and outputs of the switching unit 14 are connected to the inputs and outputs of the controller 4, the external read-only memory of the controller 9, the neuroprocessor 11 and the external read-only memory of the neuroprocessor 12. ADC 5 of controller 4 has inputs for connecting external analog sensors of distance to obstacles 17 installed on vehicle 18, the DAC 6 has outputs for communication with the steering mechanism of the steering wheel 19 and with the drive switch "forward-backward" 20, which have outputs to the ADC 5. Controller 4 has outputs to a digital steering angle sensor 21 and a digital motion direction sensor 22.

The programming and operation of the stand-alone device is as follows:

At the initial stage (at the factory), a reference device is made, through a USB port 15 it is connected to a computer 16 and a program programming the FPGA 14 itself is downloaded to the switching unit 14, which then distributes the subsequent programs to the remaining blocks. In the external ROM 9 of the controller 4 download a program containing fuzzy and linguistic variables and graphs of their membership functions (see figure 2). Graph 2a shows the graphs of fuzzy variables (terms) “Very close”, “Close”, “Near”, “Far” and “Very far”, presented in the form of triangular and trapezoidal fuzzy numbers, and constituting the input linguistic variable “Distance to obstacles” ". Graph 26 likewise shows the terms that form the output linguistic variable “Steering angle”. In the internal ROM 7 of the controller 4 load control rules in the form of fuzzy products.

At the second stage, the control device is undocked from the computer 16, installed on a car of a certain brand, and it is trained by an experienced driver. The driver, guided by his own experience, as well as the readings of the distance sensors 17, carries out a series of experimental parkings from various positions in

free space between two cars, and unsuccessful (non-optimal) parking excludes from data accumulation. Figure 3 presents two cases of parking, differing in the size of the zone of possible maneuvering 24, for example, the front bumper of the car 18, without collision with neighboring cars 25 located in front and behind the parking place. In this case, the initial position is the solid contour of the car located in the lower left corner 18. For the case with a larger maneuvering zone 24, the final position of the contour of the car 18 and the path to it are shown by a dashed line (lower right part of the figure), and for a case with a smaller zone maneuvering 24, respectively, with a dashed line (upper right part of the figure). It can be seen that in this case a greater number of “back and forth” movements are required for parking.

In this case, the data from each sensor 17 is converted into an ADC 5 and fed to the controller 4. At the same time, analog data from the steering mechanism of the steering wheel 19 and the drive switch 20 is also received there. Thus, the controller 4 fixes the dependence of the steering angle and direction of movement on the distance coming from each sensor 17 to two adjacent standing cars 25 for each parking from a certain initial position of the car 18. These data (dependencies) are fed to the built-in ROM 7 and then to the switching unit 14. Parallel to this, the percentage essa controller 4, based on the rules of fuzzy products recorded in the internal ROM 7, and on the basis of graphs of the accessory functions of linguistic variables entered in the external ROM 9, generates the recommended values of the rotation angles and direction of movement and sends these values along with the true (experimental) to the block switching 14, where they are remembered. To process all the information flows, the controller 4 uses its internal RAM 8, and to accelerate data processing, an additional external RAM 10 is used. At the same time, the software written into the external ROM of the neuroprocessor 12, together with the neuroprocessor 11 forms a multilayer artificial neural network of direct distribution. The resulting neural network is trained on the data coming from the switching unit 14, i.e. obtained as a result of a series of parking lots and containing the distance to obstacles, the recommended values of the steering angle and the actual values, using the latter as training goals. In the learning process, the neural network, after comparing the actual and recommended values, refines its weight coefficients. The updated (corrected) data through the switching unit 14 is supplied to the external ROM of the controller 9, where, in accordance with this, the graphs of the functions of the accessories of linguistic variables are adjusted (see the dashed line in Fig.2b).

At the third stage, the device is connected to computer 16 and the adjusted values from the external ROM of controller 9 through the switching unit 14 are transmitted via USB 15 to computer 16. For various makes and models of cars that differ in the gear ratio of the steering wheel and the dimensions, they train their reference samples, the data from which also entered into the computer 16, i.e. a consolidated database of specified accessory functions of output linguistic variables is created for a large number of cars of different makes and models. After that, all serial devices are connected to a computer and download programs similar to the initial stage, and in the external ROM controller 9

the specified membership functions of the output linguistic variables are written. After downloading the programs, the serial device is undocked and can be installed on any car. During operation, when parking the car, the controller 4, based on the incoming data from the sensors 17, generates the recommended values of the steering angle and direction of movement. In the case of direct control of the controller 4 by actuators 20, 22, the obtained values are translated into analog form in DAC 6 and supplied to these mechanisms, while the steering wheel itself rotates to the desired angle, and the gear itself switches from front to rear (automatic parking). Or the recommended values can come from sensors 21 and 22 on the information board, and the driver ensures the implementation of these recommendations (semi-automatic parking).

If necessary, the stand-alone device can be installed on a car of another model or brand that is not on the list, similar to the second stage, while the driver independently conducts additional training of the device, making a series of attempts to park the car.

Claims (1)

An autonomous adaptive device control device for a mobile object, containing a neural processor with a neural network emulated on it, a database of the control object and environmental conditions, sensors installed on the mobile object, and actuator sensors, characterized in that the device functionally contains three parts: neuroprocessor, consisting of a neuroprocessor, external read-only memory and external random access memory, switching, including a switching unit, and con a roller controller, including a controller with fuzzy logic, consisting of an analog-to-digital converter, a digital-to-analog converter and built-in read-only memory and random-access memory, and an external read-only memory, the outputs of sensors installed on a mobile object and the outputs of sensors of drive mechanisms connected to the input analog-to-digital converter of the controller, the inputs of the sensors of the drive mechanisms are connected to the output of the digital-to-analog converter of the control lera, and the inputs and outputs of the switching unit are connected to the inputs and outputs of the controller, the external read-only memory of the controller, the neuroprocessor and the external read-only memory of the neuroprocessor.