The ROV Control System

This section describes the selection of the control strategy. It then presents the control system design, the sensor package and real-time monitoring system. The section concludes with an overview of the anticipated system architecture in terms of the computer hardware, operating system and system software.

4.1 Selection of control strategy

In selecting the control strategy for this ROV system a decision was made to limit the choice to those strategies which have been demonstrated on practical systems and which have been reported in the literature. A review of strategies which fit these criteria was presented in section three.

Although relatively cheap and simple to implement, Open-loop controllers provide no automatic compensation for environmental disturbances. Such controllers are therefore not suitable for vehicles used for marine survey or diver support operations, or for vehicles intended for AUV system development.

Closed-loop Two- and Three-Term Controllers include a feedback component enabling the controller to automatically compensate for environmental disturbances. It has been suggested, however, that three-term controllers may not be sufficiently robust to deal with the variations in system parameters which may be expected in a modular submersible vehicle in the marine environment. Several controllers will also be required for each degree-of-freedom if stable, high-performance, control is to be achieved across the whole operating range of the vehicle [Yoerger, et al., 1986]. Most authors seem to agree that more advanced controllers are better suited to the problem of controlling a submersible vehicle. PID controllers do, however, seem suitable candidates for control of ROV thrusters if problems of thruster non-linearity can be resolved. These problems might be eliminated, or at least significantly reduced, by the application of a simple 2- or 3-term controller to compensate for thruster non-linearity.

Linear-Quadratic (LQ) Control has been combined with the 'Smith-Predictor' Adaptive Controller to produce a robust controller for underwater vehicles which incorporate time delays. The strategy has been successfully demonstrated on the ARGO towed-vehicle at the Woods Hole Oceanographic Institute. However, there appear to be no examples of LQ control of ROV systems in which the effect of thruster dynamics produces more complex vehicle behaviour [Yoerger, et al., 1990]. As with other linear controllers, several would be required for each degree-of-freedom to be controlled. For this reason, the LQ control strategy has been rejected.

The marine environment is, by definition, 'unstructured'. In addition, the vehicle configuration will be changed as the sensor package is tailored for the mission requirements. An ideal controller for a submersible vehicle will be able to adapt to changes in vehicle configuration or operating environment. Of the so-called Adaptive Controllers, the 'Variable-Structure' or 'Sliding-Mode' Controller has been successfully used on several ROV/AUV systems. These systems have been widely

reported in the literature [for examples see: Yoerger et al., 1986; Christi et al., 1990; Healey & Lienard, 1993; and da Cunha et al., 1995].

Sliding-mode control has the additional attraction that a single controller can control the vehicle across it's entire operating range. This strategy appears to be the best suited of current control strategies for a vehicle AUV mission controller.

Whether the mission controller is a human ‘pilot’ or an ‘intelligent’ AUV controller, it is desirable to minimise the effect of external disturbances on the vehicle. This can be achieved by implementing cascaded control, with a lower-level controller to provide some compensation for external disturbances. Furthermore, when AUV controllers are employed, and especially when AUV controllers are being developed and tested, there will be occasions where control must be switched between manual and automatic. In these circumstances a ‘bumpless transfer’ between manual and automatic control is highly desirable.

Differences between the actual motion of the vehicle and that required by the controller at the moment control is switched have the same effect as step inputs applied to the controller. The effect may be severe leading to sudden, violent, movements of the vehicle. In extreme cases, effects may exceed the vehicle’s design characteristics. A rapid change after a large step change may upset or ‘bump’ vehicle operating conditions. Transient responses resulting from the bump may take a relatively long time to settle to the set-point. In circumstances where the vehicle is deployed in restricted spaces or close to divers in the water, bumpless transfer is thus essential. A lower-level controller, independent of the mission controller helps provide bumpless transfer on the vehicle.

Additional complexity in controllers normally results in higher costs. Thus it is normal practice to use the simplest controller which fulfils the requirements. This has been the approach used in selecting controller designs for this project. A 3-term PID controller has thus been chosen for thruster control to compensate for external disturbances. Implementing this controller by DDC offers a number of advantages. In particular, vehicle motion can be constantly monitored and regulated in three-dimensions, and controller parameters can be rapidly changed in software to optimise controller performance. This last is important given the iterative nature of ROV design.

As explained in chapter 2, thruster dynamics have a significant effect on overall vehicle dynamics. The inherent non-linearity degrades thruster performance especially at low speeds. Thruster performance can therefore be improved, and vehicle dynamics potentially simplified, by incorporating a third level of control to compensate for thruster non-linearity.

4.2 Design of the control system

As stated from the outset, the vehicle controller developed in this project is for a ROV system. The control system design thus incorporates the lower two levels of the control strategy. Both elements of the control system have been designed for implementation by DDC on an on-board computer. As stated above, one advantage of this is that controller parameters can be quickly and easily modified in software. Once a prototype vehicle becomes available for testing the estimated values for controller parameters used below

will be replaced by those obtained from testing the vehicle. The Zeigler-Nichols methods [Appendix 5] offer a practical approach for establishing these values.

The lowest level of the control system, compensation for thruster non-linearity, is considered first.

4.2.1 Thruster Non-linearity Compensation

Thruster non-linearity results from a number of sources. The main factors causing non-linearity in thruster response are static friction (‘stiction’) in the motor cog, gearing, o-ring water seal etc., and the problem of thrusters stalling at low currents. A closed-loop controller on each thruster to provide velocity compensation for the propeller will eliminate the non-linearity. This will improve thruster performance and reliability. It will also improve the reliability of controllers operating at a higher level in the control hierarchy. A simple and economical PI controller would appear to be suitable for this application.

Electrically the thruster is an armature-current controlled motor, in which the armature voltage is varied to change the speed. This provides a speed-controlled thruster similar to the motor described in PMT604 [Block 3, section 3.2]. This is a fairly common design for ROV thrusters. The thruster characteristics, provided by the manufacturer, are tabulated below:

Table 4.1 Thruster Characteristics

Rotor velocity and direction can be measured using an optical encoder. A number of suitable low-cost devices are available offering similar performance. The sensor selected is the HEDS-5640 optical encoder combined with the HCTL-2016 16-bit quadrature decoder/counter, both from Agilent Technologies [Appendix D]. The

Thruster Characteristics

Rated voltage 25VRated current 20AStall current 1.8AMax power (85% efficiency) 425WArmature resistance 0.24Armature inductance 1mHback EMF constant 0.092V (rad/sec)Torque constant 0.087 Nm/AMotor speed control 1000 rpm/VPropeller speed constant 125 rpm/VRotor inertia 0.00458 kgm-2

Gearing ratio 8:1

Electrical time-constant 4.2 msMechanical time-constant 600ms

maximum motor speed is 25000 rpm (approximately 417 revolutions per second) which corresponds to an input voltage of 25V. The 16-bit counter on the quadrature decoder offers a quantization interval of less than 1 rpm. The effects of quantization errors are thus avoided.

A PI controller to provide thruster compensation is shown below:

Figure 4.1: PI Thruster compensation

The general form of the PI controller is:

X

EKp

Ti

11

s

where Kp = controller gainTi = integral time

An alternative form, suitable for implementation in software is:

u n Kp e n IoT

Tie k

k

k n

( ) ( ) ( )

1

where u(n) = required outpute(n) = error signal (controller input)Kp = proportional gainIo = initial condition of integratorT = sampling periodTi = integral time

The closed-loop transfer function will be a second-order differential equation. To obtain an initial estimate for controller parameters the thruster transfer function has been assumed to be of the form:

G(s)s

K

1

is the thruster time-constant, and the gain K is given by KKb

1

where Kb is the

back-emf constant. Thus, using value from table 1 above, =600ms and K=10.87.

Assuming the sensor to act as a proportional device with unity gain, and neglecting the effect of the limiter, the closed-loop transfer function becomes:

G(s)s

s s

KpK

TiKKp KKp

Ti

1

12

The system must employ positive damping for a stable time-response. An underdamped system will provide a relatively fast-acting response. The system is intended to improve thruster non-linearity, so minimising overshoot and avoiding a ‘resonance’ peak at =n in the closed-loop frequency response are desirable. A damping factor =0.7 should provide an acceptable speed of response without excessive overshoot. Selecting the closed-loop undamped natural frequency n=1.4 rads-1 gives poles located at s = -1 j. From this the values for Kp and Ti can be calculated thus:

KpK

2 1

so Kp = 0.018

Ti 2 1

2

so, Ti = 0.167 s c.f. [PMT604, Block 3, Section

3.2]

substituting these values back into the closed-loop transfer function gives:

G(s)s

s s

2 32 22

The sampling rate needs to be at least five to ten times the closed-loop bandwidth. This would require a sampling interval of between 450 and 900ms. In fact, a sampling interval of 200ms has been selected since this is well within the capabilities of the on-board processor intended for the ROV. At maximum speed, this sampling interval corresponds to slightly more than 10 revolutions of the propeller.

It is recognised that the above model is a gross over-simplification. Firstly, the thruster cannot be represented as a linear first order system. The controller is being developed expressly to compensate for the thruster non-linearity. Secondly, the sensor gain is not unity, and the effect of the limiter has been neglected. Lastly, a digital to analogue converter (DAC) will need to be added to the loop, further modifying the closed-loop transfer function. Nevertheless, the model suffices to provide initial estimates for controller parameters.

An outline for the required software has been developed and is shown below:

4.2.2 Thruster Control

A human pilot controls a ROV by means of joysticks on a control panel. The pilot’s inputs correspond to required movement along the vehicle’s x, y or z axes. It seems reasonable to assume that an AUV controller will also provide x, y and z components for required vehicle movements. The ROV has been designed to have six thrusters mounted in pairs orthogonally aligned with the vehicle axes. A controller is therefore provided for each pair of thrusters.

DEFINE VARIABLES;controller settings

REAL gain = 0.018 ;proportional gainREAL TI = 0.167 ;integral time (seconds)REAL IO = 0 ;integrator initial conditionREAL T = 0.2 ;sampling period (seconds)REAL MAXV = 25000 ;maximum rotor speedREAL MINV = -25000REAL T1 = T/TI

;initial conditionsREAL Vact = 0 ;actual rotor speedREAL Vreq = 0 ;required rotor speedREAL error = 0 ;current error valueREAL Esum = 0 ;sum of samplesREAL UN = 0 ;initial output

REPEAT every 200msBEGIN

READ volt ;applied voltageVreq = volt * 1000READ Vact ;current speed

error = Vreq - VactEsum = Esum + error ;update sum of samples

UN = gain * (error + Io + (T1 * Esum))

IF UN > MAXV ;limit speedTHEN UN = MAXV

ENDIFIF UN < MINV

THEN UN = MINVENDIF

WRITE UN ;write outputEND

Motion can be measured using solid-state accelerometers acting as ‘rate-gyros’. Three such gyros, mounted orthogonally at the centre of mass and aligned with the vehicle axes, will therefore monitor the x, y and z components of the vehicle’s motion. The gyros can also function as the sensors for a basic inertial positioning system. The maximum ROV speed is specified as 2 knots (1.03 ms-1). For a positional error of 0.1m, the minimum sample interval is 51ms.

Several suitable accelerometers are available. Tri-axial units are also available but these cost significantly more than three equivalent single-axis models. The device chosen for the sensor is the Analog Devices ADXL105 high-accuracy ±1g to ±5g single-axis accelerometer, combined with the AD976 16-bit 105 sample/second A/D converter [Appendix D]. The ADXL105 incorporates an ‘uncommitted amplifier’ suitable for configuration as a low-pass filter to eliminate high frequency vibration.

The form of the controller is shown below:

Figure 4.2 PID Controller

The thruster assembly includes the non-linearity compensator described in section 4.2.1 above. The general form of the transfer function of a PID controller can be expressed as:

X

EK

TiTd

11

ss

where K = controller gainTi = integral timeTd = derivative time

An alternative form suitable for implementation in software is:

u n K e n IoT

Tie k

Td

Te n e n

k

k n

( ) ( ) ( ) ( ) (

1

1

where u(n) = required outpute(n) = error signal (controller input)K = proportional gainIo = initial condition of integratorT = sampling period

Ti = integral timeTd = derivative time

[PMT604, Block 2, p23]

Setting Ti = 4Td simplifies the tuning of the controller and is normal practice for PID controllers. Setting Io = 0 at time t=0 is also reasonable at this stage.

No data is available for the thruster response when attached to the ROV. The mechanical time-constant of the thruster is known to be 600ms, but this is now part of the control-loop described in section 4.2.1 above. The time-response of the loop to a unit-step is given by:

c te

tn t

d( ) sin

11 2

where d n 1 2

and

tan 1

21

The control-loop has been designed with a natural frequency n = 1.4rads-1, and a damping factor of =0.7, which will give a minimal overshoot of about 4.5% of the final value. If this overshoot is neglected, an approximate ‘time-constant’ for the closed-loop can be estimated, giving a value of 1.23 seconds.

For stable systems, an value for the integral time which is too large is almost always preferred than one which is too small. In this instance, a value for Ti of 5 seconds has been selected. This will be modified by the Zeigler-Nichols method [Appendix 5] when the prototype is available. The gain will also be determined by this method. For the present, the controller gain has been assigned a value of 2.

The complete controller is shown in figure 4.3 below:

2 11

51 25

ss. 0 018 1

1

0167.

.

s

Figure 4.3: Complete Control System

The sampling interval of 50ms has been chosen to enable the sensor to function as a simple inertial guidance module for a future AUV mission controller if this is required. The sampling frequency is thus far higher than the minimum actually required for the controller.

The outline for the required software shown below is for the x-axis PID controller. Controllers for the y and z axes will be essentially similar in form.

4.3 Signal Conditioning

The ROV system incorporates electrical, mechanical and hydraulic systems. The vehicle is therefore an electrically and mechanically noisy environment. Electrical noise will degrade signals and mechanical vibrations will effect sensors. The vehicle is a compact unit so signal paths are short. Shielded cables, which are also necessary

DEFINE VARIABLES;controller settings

REAL gain = 2 ;proportional gainREAL TI = 5 ;integral time (seconds)REAL TD = TI/4 ;derivative time (seconds)REAL IO = 0 ;integrator initial conditionREAL T = 0.05 ;sampling period (seconds)

REAL T1 = T/TIREAL T2 = TD/T

;initial conditionsREAL Xact = 0 ;actual motion in XREAL Xreq = 0 ;required motion in XREAL error = 0 ;current error valueREAL Elast = 0 ;last error valueREAL Esum = 0 ;sum of samplesREAL Ediff = 0REAL UN = 0 ;initial output

REPEAT every 50msBEGIN

READ XreqREAD Xact

error = Xreq - XactEsum = Esum + errorEdiff = error + ElastElast = error

UN = gain * (error + (T1 * Esum) + (T2 *Ediff))

WRITE UN ;write outputEND

for protection against the marine environment, will minimise the effect of electrical noise and reduce the signal conditioning requirements.

The optical encoder employed as the sensor for thruster non-linearity compensation provides a direct digital output corresponding to motor shaft speed. Pull-up resistors on the output pins enable each of the three channels to drive a single TTL load. These resistors can be mounted directly on the sensor. An analogue to digital converter and low-pass filter will be required on the motor drive voltage at the controller input. The low-pass filter will eliminate high-frequency noise. No further signal conditioning is required on this controller.

The sensors for the thruster controllers are accelerometers which provide an analogue voltage proportional to the acceleration along each of the vehicles axes. Vibration will also be measured by these sensors providing an unwanted output. A low-pass filter on the output will eliminate responses due to this vibration leaving only the required signal at the analogue to digital converter input, and also act as an anti-aliasing filter. A design for a suitable filter, which also provides 0g offset, using the uncommitted amplifier is provided in the ADXL105 data sheet [Appendix D]. The filter is a second-order Sallen-Key low-pass filter which has been re-drawn below as figure 4.4:

C2

C1

R1 R2

R4

R3

VR1

VDD

IN

OUT

Figure 4.4: Active low-pass filter with gain and offset

The cut-off frequency, c, and gain, K, are given by:

c cf

R C R C 21

1 1 2 2

and KRR

1 4

3

the component values given in the data sheet are R1 = R2 = R3 = 47kR4 = 100kC1 = C2 = 0.1FVR1 = 10k

giving a cut-off frequency fc = 33Hz and gain K 3. These appear to be reasonable values for a first estimate until measurements from a prototype unit are available. The controller input, from the pilot, is already in digital form, and should thus require no further signal conditioning.