小型无人直升机的系统设计(中英文翻译)-USB迷|专注于互联网分享

2024年10月25日发(作者：管忆然)

SYSTEM DESIGNING FOR A SMALL-SCALE

AUTONOMOUS HELICOPTER

This paper presents the design of a relative low-cost and more compatible autonomous

helicopter system using HIROBO 50 scale as an experimental platform. Because of the limit

of helicopter payload, we choose the MP2128 Autopilot and a number of sensors to build the

system and the weight of instrumentation is about 500 g, much less than the payload

capability of model helicopter. Thus it is feasible to design the binocular stereo-camera

system to achieve full autonomous flight and the whole weight (include power)of

instrumentation is about 1500 g. After getting the model of the helicopter using the subspace

model identification (SMI) algorithms, we present the structure of fuzzy PID controller.

Keywords: Helicopter; MP2128; MOESP; Fuzzy PID controller.

1. Introduction

Over the past decade, small-scale helicopters are increasingly popular platforms for

unmanned aerial vehicles (UAVs). As helicopters have unique flight capability (for example:

Low-speed flight, hovering flight, taking off and landing vertically and their agility, etc), it

can offer a useful platform for a number of special flight missions such as surveillance, rescue,

security monitoring, photography, etc. There are many autonomous helicopters which have

been developed for aerial applications differently [Amidi et al., 1998;Conway, 1995]. Roberts

et al. presented a small autonomous helicopter which requires noground-to-helicopter

communications unless in the event of an emergency [Roberts et al.,2001]. Normally, the

onboard instrumentations are designed differently for different missions in a way.

In this paper, we presents a relative low-cost and more compatible autonomous

helicopter system using HIROBO 50 scale radio-controlled helicopter equipped with a

number of more compatible onboard instrumentations. The initial aim of this project is to

develop an unmanned helicopter system which can fly autonomously. The further goal of this

research is to achieve soft-landing on a moving target autonomously.

In order to fulfill the flight mission, Instrumentations onboard of the helicopter are

necessary to measure the flying data of helicopter and control its velocity, position and

attitude, as well as to communicate with the ground control software system. But the payload

of HIROBO 50 scale radio-controlled helicopter is about 2 kg. This makes the system

designing onboard more significant.

第 1页共 27 页

This paper is organized as follows: In Sec. 2, we introduce the model helicopter and

MP2128-UAV. In Sec. 3, we present the configuration of sensors. In Sec. 4, the structure of

control system of the unmanned helicopter is introduced. In Sec. 5, we describe the

communication strategy of system. Finally, in Sec 6, we calculate the whole payload of

instrumentation and draw some conclusion.

2. Configuration of the System

Yamaha R50 model helicopter is the perfect choice for many research groups because of

its adequate payload (about 20 kg) and reliable operation. But it is rather more expensive for

our research group. The helicopter we chosen is a relative low-cost, radio-controlled

helicopter— HIROBO 50 scale helicopter as flight platform which equipped with autopilot

component— MicroPilot Autopilot MP2128-UAV and a number of sensors.

2.1. The helicopter

HIROBO 50 scale model helicopter, shown in Fig. 1, was chosen as an experimental

platform [Chen et al., 2006]. This model helicopter is a commercially available small-size

helicopter. As with other small-size helicopter, HIROBO 50 has two blades of the main rotor

which generate the needed to lift the helicopter. Because of the small size and relative fast

rotor speed, it is fitted with a control rotor to add damping in order to lower the dynamics of

the system. The control rotor also reduces the power needed by the actuators to control the

helicopter [HIROBO Limned, 2003]. Its parameter is as follows: Helicopter type: HIROBO

Shuttle SCEADUE volution 50, rotor diameter: 1350mm, gross weight: 3.23 kg, gear ratio:

8.7:1:4.71Engine type: OS 50 class engine Payload: about 2 kg.

Fig. 1. HIROBO 50 scale model helicopter.

第 2页共 27 页

Normally, there are five servos which act as inputs to pilot the model helicopter in the

small-scale autonomous helicopter:

• Elevator (longitudinal cyclic pitch)

• Aileron (lateral cyclic pitch)

• Collective (main rotor blade pitch)

• Rudder (tail rotor blade pitch)

• Engine throttle.

Accordingly, the outputs which fulfil to control the helicopter’s behavior are pitch

control, roll control, up/down control, yaw control and engine revolutions per minute control.

In HIROBO 50 scale helicopter, the throttle servo and collective servo are mixed and there are

four inputs actually. The servos receive signals from Micro Pilot Autopilot which is mounted

in the model helicopter when it flies autonomously, or from R/C when manually.

2.3. MP-2128 UAV-FB

The autopilot system, shown in Fig. 2, is produced by MicroPilot Corporation. It is

consisted of MP-2128g [MicroPilot Corporation, 2006], ultrasonic sensor, wireless video

camera system, and 2.4 GHz standard range data-link. The MP-2128g is the main component

of all. It is designed for fully autonomous operation and can provide flight speed, flight

altitude, and GPS navigation. There are PIC(pilot in command mode) and CIC (computer in

command mode) which can be switched by Channel 5. There are 12 feedback loops that can

be selected by control system to fly UAV. All feedback loops gains of PID and flight

parameters are adjustable in flight. Significantly, the MP 2128g core is only 28g and this is the

sound reason for us to choose it.

There are two methods that can be used to adjust settings of the fields on MisroPilot

Autopilot: The HORIZONmp ground control software [MicroPilot Corporation, 2006] and

HyperTerminal included with all version of Windows via the standard RS232 serial link.

When equipped with the HORIZONmp ground control software running in portable computer

shown in Fig.3, the autopilot system provides flight mission creation, flight parameter

adjustment, flight monitoring as well as an extensive internal data logging that can be used to

analyze flights.

第 3页共 27 页

Fig. 2. MicroPilot autopilot system.

Fig. 3 The HORIZONmp running in portable computer.

3. Configuration of Sensors

3.1. MP2128g board

In order to measure the helicopter’s attitude, position and velocity, certain sensors are

needed. GPS, gyros, airspeed pressure transducer and altitude pressure transducer are

intergraded on the MP2128g board as shown in Fig. 4 [MicroPilot Corporation, 2006].

The Gyro provides roll rate and yaw rate of the helicopter. MP 2128 board includes an

integrated GPS receiver using the Trimble TSIP protocol. The GPS provides position,

velocities of the helicopter.

Since the altitude pressure transducer cannot detect the relative height, an AGL sensor is

required for autonomous runway takeoff and landing. The AGL is an ultrasonic altimeter that

provides altitude information up to 16 feet above the ground. The AGL board is connected to

the P2 connector on MicroPilot Autopilot.

第 4页共 27 页

For helicopter has the capability of hover, Compass is needed to provide the azimuth

position. The compass module is a three-dimensional compass that can compensate for pitch

and roll. Use the compass module in applications where GPS headings are inaccurate or

unreliable, such as [MicroPilot Corporation, 2006].

• In a hovering aircraft which cannot use the GPS for direction when hovering

• In slow moving aircraft, like a blimp, in which GPS headings are unreliable

• For dead reckoning if the GPS is lost

• When operating the autopilot in strong winds.

The electronic compass has a double sided connector which connects to the expansion

connector(P3) on the autopilot board with an expansion cable.

Fig. 4 MP2128g board.

3.2. Vision system

The helicopter will be equipped with binocular stereo-camera system later based on PC

104 and PC 104-Plus cards shown in Fig. 5 and consists of:

• a PC104-Plus Profive-CPU-P5 motherboard with an Intel Pentium M1.6G processor

• a PC104-Plus Profive Ethernet board

• a PC104-Plus Profive VGA card

• a Tri-M PC-104 power supply

• a set of radio transmitter/receiver communicates with MP2128 UAV

• Binary-cameras to orient the moving target.

The vision processing software runs under the on-board real-time operating system, RT-

Linux, and uses a custom streaming video driver for the frame-grabber. The flow chart of

第 5页共 27 页

stereo-image processing was shown in Fig. 6.

The vision system provides the surface information of the landing position. Furthermore, it

can identify the moving target and track it. The vision system was described in [Zhu et al.,

2006] in detail.

Fig. 5 The stereo vision based on PC 104-plus.

第 6页共 27 页

Capture Scene

Is target?

Image data ruduction

Shape filtering

Feature segmentation

Is target?

Color thresholding

Is target?

Size filtering

Target’s distance

estimation

Tracking

Fig. 6 Flow chart of stereo-image processing.

3.3. Force sensors

Since the attitude of moving target is uncertain, the force sensors are needed for

helicopter soft-landing. Here, we want to choose the conductive rubber as force sensor. When

the conductive rubber is pressured, its resistance will change. The relation between

conductive rubber resistance and pressure is shown in Fig. 7 [Jin et al.,1997]. According to

the circuit like Fig. 8 [Tianet al., 2004], we can get the relation between pressure and output

voltage as Eq. (1):

u = f(Rp) = g(P). (1)

The A/DC board mounted on the helicopter then sample the signal and transmit it to

force value. There are four conductive rubber force sensors which be mounted on the model

第 7页共 27 页

helicopter(shown in Fig. 9)

Fig. 7 The relation between conductive rubber resistance and pressure.

Fig. 8 Measurement of the conductive rubber resistance.

第 8页共 27 页

Fig. 9 The landing gear set with force sensors

In Fig. 9, the landing gear supplied with model helicopter is made of aluminium tube.

It can be certain of relative relation between the attitude of the model helicopter and the

landing area according to the four force sensors for the landing area is not always the same

level of horizontal surface as in Table 1.

Table 1. The situation of force sensors

In Table 1, “1” means this force sensor has an output value, and “0” has no output value.

The control strategy in these nine situations are different each other. We can change some of

the four inputs to adjust the attitude and velocities of helicopter in order to achieve soft-

landing properly. For example, in situation 1, the model helicopter’s attitude is pitching

backward and rolling left side, so inputs of lateral control and pitch control must adjust to

第 9页共 27 页

maintain the helicopter’s behavior.

4. Control System

Because most multivariable control methods are model-based, and the dynamic model

for a particular small-scale helicopter which is simple enough to be practical for controller, is

not readily available, the identification of the small-scale helicopter‘s dynamic model is a

necessary part of any model-based control design.

4.1. Small-scale helicopter identification using MOESP

As it is described in [Hashimoto et al., 2000],a problem of importance for the flight

control of autonomous helicopter is its inherent qualities: The dynamics of the helicopter are

essentially unstable, there are nonlinear variations in dynamics with air speed. Moreover, the

helicopter has six degrees of freedom in its motions which is multi-input multi-output system

and its flight modes are cross-coupled. Usually, the helicopter can be modeled as a linear

system around trim points and there are several examples of the application of system

identification techniques to the modeling of small-scale helicopter [Tischler et al., 1996;

Morris et al., 1994].

Subspace Model identification (SMI) algorithms are a group of methods that identify a

MIMO state-space system using numerically robust computation tools such as QR

factorization and SVD (singular value decomposition).Subspace methods rely on the ideas of

stochastic realization and have been developed by many researchers [Chiuso, 2001; Akaike,

1974;Van Overschee, 1996]. In contrast to the classical approaches of maximum likelihood or

PEM methods, the subspace method avoids the use of canonical forms. The MOESP

(multivariable output error state space) identification method has already been applied to

some modeling the identification of a test satellite [Adachi, 1999] and BO 105 helicopter

[Verhaegen, 1995].

We had applied the MOESP method in the model identification simulation of a small-

scale helicopter dynamics using MATLAB to find a model of the helicopter in hover using the

collective pitch and rudder inputs as Eq. (2):

第 10页共 27 页

In order to test this algorithm’s validity, we use the sequence consisting of the Gaussian,

white noise signal with 4000 data points as the inputs of the system model obtained by

MOESP and PEM. We compare the outputs error using this method with that of the PEM in

Eq. (3) and find that the result of using MOESP method is more accurate than that of the PEM

method. Moreover, it does not need any priori information about the identifying model:

= 1.5062e − 022, er

= 1.7125e − 023, er

= 9.4457e − 018;

= 153.1331, er

= 2.4398e + 004, er

= 3.9928e + 007.

Here, the Index k is the length of data points,y

(i = 1, 2, 3) are the outputs error of the

real system to the result of using MOESP and,Y

−y'

(i = 1, 2, 3) are the outputs error of the

real system to the result of using PEM..

4.2. Fuzzy PID controller

There are many different types of controller for an unmanned helicopter [Kadmiry et al.,

2001;Jensen et al., 2005; Shim et al., 1998; Phillipset al., 1996]. To control the helicopter

第 11页共 27 页

different advanced control methods could be used which have their own advantages and

disadvantages and the most significant for this project will be discussed and a conclusion will

be made. The robust controller [Clausen et al., 2001] is considered a good controller if there

are uncertainties on the parameters and disturbances in a MIMO system to be dealt with apart

from the complexity computation of matrix. Optimal control is mostly used to optimize a

given controller, often MIMO, to give the control profile a power optimal, time optimal, jerk

optimal, or a combined optimal profile, and can be extended to a robust optimal controller

[Jensen et al., 2005].

A classical fuzzy logic controller is a knowledge-based system and is easy to construct

because it has no need of a model of the system. Often it is used in control of nonlinear

systems and systems where the parameters are hard to determine. But they are hard to tune,

and the accuracy of outputs usually cannot be guaranteed if the number of rules are small. On

the other hand, the number of rules increases exponentially with the number of inputs and it

makes this classic fuzzy logic controller becoming impractical.

Neural network controller has its primary advantage with highly nonlinear systems. It

can “learn” a system’s transfer function from training data and function. It is not to be suitable

for unmanned helicopter control, since it has no explicit expression for the inputs and outputs

and the necessary parameters for training will not be at hand.

Fig. 10 Vertical climb manoeuvrers using (A) nonlinear,(B) fuzzy and (C) robust control for

the (1) nominalmodel and for uncertainties in (2) system weight.

第 12页共 27 页

Fig. 11 The block diagram of Fuzzy PID controller for small-scale unmanned helicopter.

Shim and Koo et al. had test that the performance of the controllers designed by using

linear robust multi-variable control, fuzzy logic control with evolutionary tuning, and

nonlinear tracking control in regard to disturbance rejection, uncertainties in system parameter

and tracking accuracy [Shim et al., 1998], and obtained the conclusion shown in Fig. 10 that

the robust and fuzzy controllers are capable of handling uncertainties and disturbances and the

nonlinear control covers a substantially wider range of flight envelopes, but requires accurate

knowledge about the system for vertical climb. Taking into account of the capability of

payload of the unmanned helicopter and the complexity of computation, we develop our

control system using Fuzzy PID controller shown in Fig. 11. According to the paper [Phillips

et al.,1996], the fuzzy logic controller is composed of four sections: the longitudinal cyclic

control, later cyclic control, rudder control, and collective control. In order to decrease the

steady state errors between the real outputs and the desired outputs, we design the PID

controller after fuzzy logic controller. This control system takes advantage of the merits of

these two controllers.

5. Communication

The communication system is shown in Fig. 12[MicroPilot Corporation, 2006]. The pilot

can control the helicopter by RC transmitter/receiver manually and can also switch it to fly

autonomously by pushing a select channel. The portable computer running the HORIZONmp

on the ground communicate with the MP2128-UAV, which mounted on the helicopter by RF

model which frequency is 2.4GHz, data rate is 9600 bps. The outdoor/RFline-of-sight rang of

radio model is up to 16 km,indoor/urban range is up to 180m and receiver sensitivity at

9600bps is −105 dBm.

第 13页共 27 页

Fig. 12 Communication system of small-scale autonomous helicopter.

There is aCOM connector which can be used to copy flight data of the helicopter such as

position, velocities,attitude, altitude and landing force from MP2128-UAV to portable

computer for analyzing and resolving problems.

6. Conclusion

Due to our desire to design a small-scale helicopter which can achieve soft-landing on a

moving target, we present the configuration of the unmanned helicopter system which has a

large number of sensors located on it in order to measure the attitude, velocity and position of

the helicopter and other certain the small-scale autonomous helicopter has

limited payload, we choose the MP2128-UAV,HORIZONmp ground control software and

some sensors to design autopilot system. The whole weight (include power) of

instrumentation is about 1500 g (the whole weight of MP2128-UAV, compass, AGL, and

force sensors is within 500 g, stereo vision system based on PC 104-plus is about 1000 g),

less than the payload capability of HIROBO 50 scale model helicopter. Now,we are using the

XTENDER Software Development Kit [MicroPilot Corporation, 2006] to develop our own

flight mission to run with MicroPilot autopilot code and ground control software.

第 14页共 27 页

小型无人直升机的系统设计

本文介绍了设计一个相对低成本和一个更兼容的无人直升机系统，该系统以

HIROBO50作为实验平台。由于直升机的有效载荷限制，我们选择MP2128自动驾驶仪和

一些传感器建立系统，仪器的重量约为500克，远小于直升机模型的负载能力。因

此，设计双目立体照相机系统实现完全无人飞行的可行性，仪器的整个重量（包括电

源）约为1500克。该直升机模型使用子模型辨识（SMI）的算法，我们的讨论处理模

糊PID控制器的结构。

关键词：直升机，MP2128，MOESP，模糊PID控制器。

1.引言

在过去十年中，小型直升机日益流行采用无人驾驶飞行。由于直升机独特的飞行

能力（例如：低速飞行，盘旋飞行，起飞和着陆得垂直性和他们的敏捷性等），所以

它可以提供一些执行特别飞行任务一个有用的平台，如监视，救援，安全监测，摄影

等。许多无人直升机发展应用并不相同[ Amidi等，1998年; 康威，1995 ] 。罗伯

茨等人提出了一种小型无人直升机不需要与地面进行通讯，除非到遇紧急情况[罗伯茨

等人，2001 ] 。通常情况下，仪表板因为执行不同的任务而设计不同。

在本论文中，我们提出了一个相对低成本和更自主直升机系统，在HIROBO 50级

无线电遥控直升机上它将配备一些更兼容仪表板。该项目最初的目的是是开发一种无

人驾驶可自主飞行的直升机系统，进一步研究的目标是实现对移动目标的自主软着

陆。

为了实现这个飞行任务，在直升机仪表盘有必要测量飞行数据并且控制它的速

度、方向和姿态，以及与地面控制软件系统的通信。但HIROBO 50级无线电遥控直升

机的有效载荷约为2 kg。这使随身携带系统在设计上更为重要。

本文安排如下：在第二部分中，我们介绍了直升机模型和MP2128无人机。在第

三部分中，我们介绍传感器的构造。在第四部分中，介绍无人驾驶直升机控制系统的

的结构。在第五部分中，我们描述了通信战略系统。最后，在第六部分我们计算了整

个仪器的有效载荷并得出一些结论。

第 15页共 27 页

2.系统的配置

对于许多研究团体来说，雅马哈R50直升机模型是最佳选择，因为它足够的有效

载荷（约20公斤）和可靠的操作。但对于为我们的研究小组来说，它非常的昂贵，因

此是我们选择的是一个相对低成本直升机，无线电遥控直升HIROBO 50级直升机飞行

平台，它配备有自动驾驶仪—微小的自动驾驶仪MP2128无人机和一些传感器。

2.1直升机

HIROBO 50直升机模型如图2-1所示，如图所示模型被选定作为实验平台[陈等

人。， 2006 ] 。该模型直升机是商用小型直升机。与其他小型直升机一样，HIROBO

50有两个叶片的主旋翼产生直升机飞行的动力。由于体积小，相对快速的转子速度，

它装有控制转子添加阻尼来降低系统的动态性能。控制转子还减少了驱动器控制直升

机的功率需求，[ HIROBO Limned ，2003 ] 。它的参数如下：直升机类型：HIROBO

Shuttle SCEADUE volution 50；转子直径：1350毫米；总重：3.23 kg；传动比：

8.7:1:4.71；引擎类型：OS50级引擎；有效载荷：约2公斤。

图2-1 HIROBO 50直升机模型

2.2舵机

在通常情况下，在小型无人直升机中，有五个舵机用作直升机模型飞行的输入：

•升降机（纵向循环间距）

•副翼（横向循环间距）

•整体的（主旋翼间距）

•舵（尾旋翼间距）

第 16页共 27 页

•引擎节气门

因此，输出满足控制直升机行为，向下倾斜，左右摇摆，向上/向下，偏流以及

发动机每分钟的旋转率的控制。在HIROBO 50型直升机中，油门舵机和主舵机混合在一

起，它们实际上有四个输入。当它自主飞行，或手动的装在R /C上时舵机从微型飞行

员那里接收信号，自动驾驶仪是安装在直升机模型上。

2.3. MP-2128 UAV-FB

自动驾驶系统如图2-2所示,是由MicroPilot公司制造的。它是由MP-

2128g[ MicroPilot公司， 2006 ] ，超声波传感器，无线摄像机系统，和2.4 GHz的

标准范围内的数据连接组成。MP-2128g是最主要的组成部分。它是专为完全独立的操

作而设计的并且可提供的飞行速度，飞行高度，和GPS导航。PIC（驾驶命令模式）和

CIC （计算机命令模式）可以切换5个频道。飞行无人机控制系统可挑选出12个反馈回

路。在飞行中所有反馈循环都PID控制和飞行参数都可以调整。值得注意的是，MP

2128g 的核心仅仅有28g，这是我们要选择它的原因。

这有两种方法可以用来调整设置MisroPilot自动驾驶仪的领域：HORIZONmp地面

控制软件[ MicroPilot公司，2006 ]和终端机包括Windows的所有版本通过RS232标准

进行串行链接。如图2-3所示，在便携式计算机中配置HORIZONmp地面控制软件。自动

驾驶系统提供创造性的飞行任务，调整飞行参数，飞行监测以及能够用作分析飞行器

的大量的内部数据记录。

图2-2. Micro Pilot自动驾驶系统

第 17页共 27 页

图2-3. The HORIZONmp运行在笔记本电脑中

3 传感器配置

3.1. MP2128g 板

为了衡量，这架直升机的姿势、位置、速度，无疑会需要很多传感器，GPS 、陀

螺仪、空速压力传感器和海拔压力传感器都集成在MP2128g板上，如图3-1所示

[ MicroPilot，公司，2006 ] 。

图3-1 MP2128g板

陀螺仪为直升机提供辊速度和偏航率。MP 2128板包括一个使用Trimble TSIP协议

的集成GPS接收器。全球定位系统为直升机提供位置，速度导航。

由于海拔压力传感器无法检测的相对高度，AGL传感器都需要自主的跑道起飞和降

落，AGL是一个超声波测高仪，它可提供海拔信息高达地面以上16英尺。AGL板连接到

的MicroPilot自动驾驶仪的P2连接器上。

因为直升机有盘旋能力，罗盘是需要提供的方位信息。罗盘模块是一个三维的指

南针，可以弥补斜度和角度。当全球定位系统的不准确或不可靠的时候使用指南针模

块，如[ MicroPilot公司，2006 ] 。

第 18页共 27 页

•在盘旋的飞机摇摆不定的时候不能使用时GPS的方向导航。

•在移动缓慢的飞机，就像飞艇，其中全球定位系统的是不可靠的。

•对于失事的推算，如果失去了全球定位系统。

•当操作自动驾驶仪的时候遭遇强风。

电子罗盘具有双面连接器能够连接扩展接口（P3），P3在自动驾驶仪板上并且有扩

大电缆。

3.2 视觉系统

这种直升机将配备双目立体相机系统后，基于PC 104和PC 104+ 如图3-2所示。

由以下几部分组成：

•一个配备英特尔奔腾M 1.6G处理器的PC104+ Profive-CPU-P5母版

•一个PC104+ Profive 以太网板

•一个PC104+ Profive VGA显卡

• 一个Tri-M PC-104电源供应器

•一套能够与MP2128无人机沟通的无线电发射机/接收机

•能够确定移动的目标方向的二进制相机。

视觉处理软件在机载实时操作系统下运行，应用RT - Linux ，以及使用自定义的

视频流驱动程序。在图3-3中显示立体图像处理的流程。

视觉系统提供表面着陆的位置信息。另外，它可以识别移动目标并且进行跟踪。

视觉系统的描述在[朱等人，2006 ]详细说明。

图3-2 在基于PC 104-plus的立体图

第 19页共 27 页

Capture Scene

Is target?

Image data ruduction

Shape filtering

Feature segmentation

Is target?

Color thresholding

Is target?

Size filtering

Target’s distance

estimation

Tracking

图3-3 立体图像处理流程图

3.3 力传感器

由于目标的运动姿势是不确定的，直升机软着陆的需要力传感器。我们要选择的

导电的橡胶的作为力传感器，当导电的橡胶受到压力作用，它的电阻将会改变，导电

橡胶阻值和压力的关系如图3-4所示 [金等人， 1997 ] 。根据电路图3-5所示

[Tian

et al

., 2004]，我们可以得到压力和输出电压之间的关系（ 1 ）：

(

) =

(

)

(1)

第 20页共 27 页

图3-4 导电橡胶阻力和压力之间的关系

图3-5 测量电阻的导电橡胶

A /DC板安装在直升机上，然后抽样信号并传输这些抽样信号转换成有效值。有四

个导电橡胶力传感器安装在模型直升机（如图3-6所示）

第 21页共 27 页

图3-6 起落架放置在力传感器上

在图3-6中，模型直升机的起落架是由铝管制成的。

从表1可以看出，四个力传感器的着陆区并不总是在同一级别的横向表面，因此我

们发现，模型直升机姿势和着陆区有相关的关系。

在表1中，“ 1 ”是指力传感器的输出值， “ 0 ”没有输出值。控制策略在九种

情况下是不同的。我们可以改变四个输入去调整直升机的姿态和速度，以便实现软着

陆。例如，在情况1，该模型直升机的姿势是相反的俯仰和左侧滚动，所以输入的横向

控制和斜度控制必须调整，以保持直升机的正常飞行。

表3-1 力传感器的各种情况

第 22页共 27 页

4 控制系统

因为大多数多变量控制方法是基于模型和动态模型的，一个特殊的小型直升机是

简单可行的控制器，是没有现成的，小规模直升机的动力学模型的建立是任何基于模

型的控制设计的一个必不可少的一部分。

4.1 用MOESP识别小型直升机

正如在[Hashimoto

等人

, 2000]论文中描述，无人直升机的固有的品质在飞行控

制中是非常重要的：动力学直升机本质上是不稳定的，随着空气速度的变化，动力学

呈现非线性的变化。此外，直升机在自身监测中有六个自由度，它的运动系统是多输

入多输出，它的飞行模式是交叉耦合的。通常情况下，直升机模型可以定义为多个小

点周围的线性系统，有几个例子是系统辨识手法建模小规模直升机的应用。

子空间模型辨识（SMI）算法是一组方法，使用强大的数值计算工具来确定MIMO的

状态空间系统，如的QR因数分解和SVD （奇异值分解）。这种子方法依靠随机的认识，

并且被许多研究人员发展。与传统的最大似然法或质子交换膜方法相反，子空间方法

避免使用规范的形式。该MOESP （多变量输出错误状态空间）识别方法已应用于一些

建模，比如确定了试验卫星和 BO 105 直升机。

我们已经采用了MOESP法用模型辨识模拟了一个小规模的直升机动力学问题。利用

MATLAB找到一个模型直升机悬停使用集体斜度和方向舵输入，如等式2所示：

为了检验该算法的有效性，我们使用高频信号包括：高斯信号、白噪声信号与4000的

数据点作为系统模型的输入。我们用这个方法比较输出错误在等式3中，发现用MOESP

方法的结果比PEM方法的结果更精确。另外，它不需要任何模型建立的信息。

第 23页共 27 页

= 1

5062e

−

022

= 1

7125e

−

023,er

= 9

4457e

−

018;

= 153

1331

= 2

4398e + 004, er

= 3

9928e + 007.

在这指数K是转辙器数据的长度，

(

= 1

3) 是使用MOESP实时系统

的结果输出错误，

−y'

(

= 1

3) 是使用PEM真正系统的结果的错误的输出。

4.2模糊PID控制器

有许多不同类型的控制器专门针对无人驾驶直升机，为了控制这些直升机，运用

了不同的拥有自己的优点和缺点的先进控制方法，经过讨论，最终将会选择针对项目

最有用的方法。鲁棒控制器被认为是一个很好的控制器，它可以处理MIMO系统中不确

定性的参数和干扰的原理的复杂矩阵计算。最优控制主要用于优化某一控制器，往往

使用MIMO技术，使配置控制电源最优化，时间最优，成绩最佳，或结合最优配置轮

廓，并可以延长成为一个强有力的最优控制器。

经典的模糊逻辑控制器是一个以基础知识的系统，因为它不需要系统模型，因

第 24页共 27 页

此，它很易于建模。它经常用来控制非线性系统以及参数的难以确定的系统。但是，

当一些规则很小的时候它们很难调整，而且产出的准确性通常不能得到保证。另一方

面，随着输入的增加，规则的数目以指数次幂增加成倍的增加，这使经典模糊逻辑控

制器变得不切实际。

神经网络控制器在高度非线性系统有其主要的优势。它可以“学习”系统从训练

数据和功能中转移功能。它不适合于无人控制的直升机，因为它没有明确表示的输入

和输出以及必要的在手中的训练参数。

Shim 和 Koo等人已经测试利用线性鲁棒多变量控制设计的控制器的性能，模糊逻

辑控制与进化调整和非线性跟踪控制方面的干扰，不确定性的系统参数和跟踪精度，

并得到了如图4-1显示的结论，健全和模糊控制器能够处理不确定性、动荡的、非线性

控制涵盖了大量广泛的飞行信息，但需要准确了解系统的垂直的爬行能力。考虑到无

人驾驶直升机得有效载荷的能力和的复杂性的计算，我们的控制系统采用模糊PID控制

器，它的显示图如图4-2所示。模糊逻辑控制器是由四个部分组成的：纵向循环控

制，后翼循环控制，舵机控制和集体控制。为了减少稳态误差之间的实际输出和所期

望的输出，我们设计了模糊PID逻辑控制器。这种控制系统采用了这两个控制器各自的

优点。

图4-1 垂直的爬行移动使用了(A) 非线性,(B)模糊 (C) 强大的控制 (1) 无象征和不确定 (2)

系统重量

第 25页共 27 页

图4-2 小规模无人驾驶机模糊PID控制器的图解

5.通信

通信系统如图5-1所示：

图5-1 小型无人直升机的通信系统

飞行员可以通过RC发送器/接收器手动控制直升机，也可手动自主选择频道切换到

自动飞行。便携式计算机运行HORIZONmp软件与 MP2128-UAV无人机实现了实时沟通，

射频频率是2.4GHz频段安装在直升机模型上，数据传输速率是9600bps。户外/ RFline

的视线范围内的无线电模型可达16公里，室内/城市范围可达180m和接收灵敏度在

9600bps是-105dBm。可以通过互联网连接器将直升机的飞行数据复制在笔记本电脑

上，如位置，速度，姿势，高度和着陆点，以便便携式计算机的分析和解决问题。

第 26页共 27 页

6.结论

由于我们希望设计一个小规模的直升机，该直升机能实现软着陆的目标，我们介

绍了配置有大量的传感器来测量飞行姿势，速度和位置及其他某些信息的无人直升机

系统的构造。因为小型无人直升机的有效载荷是有限，所以我们选择MP2128-

UAV,HORIZONmp地面控制中心的软件和一些传感器设计自动驾驶仪系统。整个仪器的重

量（包括电源）的是1500克（MP2128无人机的整个重量，指南针，AGL ，力传感器在

500克以内，基于PC机104的立体视觉系统约1000克），小于HIROBO50模型直升机的有

效载荷的能力。现在，我们使用的是XTENDER软件开发工具包[ MicroPilot公司，

2006 ]开发MicroPilot自动驾驶仪的代码和地面控制软件来实现我们自己的飞行任

务。

第 27页共 27 页