Advances in platform manoeuvring control - Toward full autonomy - - PDF document

UDT 2020 UDT Extended Abstract Template Presentation/Panel

Advances in platform manoeuvring control - Toward full autonomy

Nathan Thomas1, Chris Harris2

1Principal Engineer, Stirling Dynamics, Bristol , UK 2 Principal Engineer, Stirling Dynamics, Bristol , UK

Abstract — In this paper, the benefits and challenges of introducing increasingly complex algorithms to provide autonomy on underwater platforms will be discussed. The issues will be illuminated by a comparison between the existing approach to manned submarine platform manoeuvring control, current methods for controlling small and medium sized AUVs, and suitable approaches for the new generation

• f

XLUUVs.

1 Introduction

Stirling Dynamics provides technical engineering services and control system technology into the aerospace, marine, energy and training & simulation sectors. Since the early 1990s, Stirling has supported 10 worldwide navies in the design and test of a wide range

• f submarines, Stirling’s specialism lies in safety-critical

steering, diving control systems and state-of-the-art autopilot and hover solutions, which deliver enhanced control and performance. In this paper, the benefits and challenges of introducing increasingly complex algorithms to provide autonomy on underwater platforms will be discussed. The issues will be illuminated by a comparison between the existing approach to manned submarine platform manoeuvring control, current methods for controlling small and medium sized AUVs and the new generation of XLUUVs.

2 AUVs, XLUUVs and Submarines

Small and medium sized Autonomous Underwater Vehicles (AUV) have been in use for military purposes for several decades. Applications include mine countermeasures, seabed mapping and intelligence

• gathering. These vehicles are typically less than 5m in

length, less than 0.5m in diameter and displace no more than approximately 1 tonne In what follows, the term AUV shall be used to refer to small/medium sized AUVs

• nly.

More recently, two classes of much larger vehicles have emerged: Large Displacement Unmanned Underwater Vehicles (LDUUV) and Extra Large Unmanned Underwater Vehicles (XLUUVs). LDUUVs have diameters between 0.5m and 2m in length; XLUUVs have diameters of 2m and above. XLUUVs are intended to be much more capable than AUVs, with a far greater range, the ability to deploy AUVs as a payload and ultimately weapons. An example of the new generation of XLUUV is the Boeing Orca, which has a length of up to 26m, diameter of 2.5m and displaces roughly 50 tonnes. Stirling has extensive experience

• f

manned submarine platforms ranging from 50 to over 100m in length and from 1000 to more than 10000 tonnes

• displacement. On the face of it, the problem of

controlling flight style AUVs should have much in common with the automatic control of much larger manned submarines. Indeed, most AUVs of this type are torpedo shaped and have a similar control surface layout to submarines. Although this commonality is true to a certain extent, there are significant differences due to both operational and safety related factors. XLUUVs fit somewhere in between these two categories and will over time increasingly approach the capabilities of manned submarines. Consequently, many

• f the considerations that apply to platform control on

submarines become relevant to XLUUVs. Looking in the

• ther direction, the trend of increasing automation in

manned submarines means that some of the techniques that are necessary for XLUUV control have applicability to manned submarines.

3 Platform Control Comparison

In this section, the appropriate approach to platform manoeuvring control for the three vessel types will be compared by looking at various key factors that determine the control algorithm. 3.1 Performance Requirements Submarines are required to operate in close proximity to both the surface (at periscope depth) and the seabed. The automatic control system must be able to promptly and accurately change depth (or heading), and to be able to precisely keep depth all while subjected to environmental effects such as waves, currents and density changes. AUVs might not put the same demands on depth control, if there is not the same requirement to operate with high depth accuracy close to the surface. XLUUVs are more likely to need to operate close to the surface, while remaining concealed, for example to use diesel snort masts, perform visual identification or to communicate. The control performance requirements are a significant factor in determining both the complexity of any control algorithm and the fidelity and extent of the test environment used to validate it. For example, a simple Proportional Integral Derivative (PID) control algorithm that is tuned in-situ might be appropriate for a very basic AUV, whereas submarine control systems require robust control techniques and extensive testing using a detailed simulated test environment. XLUUVs

UDT 2020 UDT Extended Abstract Template Presentation/Panel probably require a similar level of complexity in both areas as submarines. 3.2 Vessel Geometry & Mass Distribution An accurate plant model is usually a pre-requisite for a high performance control system. When designing an automatic control system for a manned submarine, a high fidelity non-linear boat model will typically be available. Producing such a model is an expensive, time consuming, endeavour which requires extensive tank testing. However, the necessary resources are a small proportion

• f the overall program and this model is any case

necessary for other reasons, for example to obtain the Safe Manoeuvring Envelope (SME). For AUVs, the effort necessary to obtain a high fidelity model is probably not cost effective. Also, many AUVs are either modular and may change substantially in length as modules are added/removed or have variable payloads which cause the mass distribution to change significantly from any nominal condition. Thus both the hydrodynamic and inertial characteristics may vary considerably throughout the platform’s operational life. These two factors make adaptive control algorithms, which will adjust the control parameters to the plant dynamics, particularly attractive for AUV control applications. On a manned submarine, the geometry is mostly fixed although some vessels will attach/detach objects to the

• hull. The overall mass distribution of a manned

submarine generally does not change greatly when in its submerged state. It is necessary for the control system to be able to cope with moderate amounts of hydrostatic imbalance (Out-of-Trim). This, and minor variations in geometry, are best treated as uncertainty/external disturbances through standard robust control techniques rather than adaptive algorithms. XLUUVs can also have modular geometry so adaptive control is an option, provided the required performance can be achieved. Alternatively, the greater cost of the platform may make it viable to develop high fidelity models for each modular configuration with a different controller parameter set for each. 3.3 Hover Most AUVs with a hover function use vertical thrusters to achieve depth control at low speed. The hover function

• n submarines is achieved through ballast control, either

directly by the ballast pump or indirectly using a pressurised water tank. XLUUVs that are required to hover will likely do so in a similar fashion to submarines and so the algorithms used there are directly relevant. 3.4 Trim and Ballast Compensation AUVs typically use a passive buoyancy system whereby the vessel is initially ballasted to be positively buoyant and appropriate materials used to ensure that the net buoyancy does not change too greatly with depth pressure. Forward speed or vertical thrusters are required to keep the vehicle submerged, which means that serious failures lead to recovery by surfacing. Submarines use ballast compensation tanks along with fwd/aft trim tanks to make adjustments to the mass and centre of mass. It is usual for the trim and ballast compensation system to be manually operated with the help of an Out of Trim Estimator (OTE) along with look- up tables of the expected change in ballast required to compensate for a given depth change. XLUUVs include submarine style trim and ballast

• tanks. Fully automated control of trim and ballast

compensation is of course essential, rather than optional

• here. Stirling have developed FASC (Full Authority

Submarine Control [1],[2]), which provides integrated automatic control of trim and ballast compensation, hover and hydroplanes. While

• riginally

created for submarines, this has direct applicability to XLUUVs. 3.5 Guidance Both AUVs and XLUUVs require the ability to produce guidance commands to be used by the low level control

• algorithm. These commands may be generated from a

series of waypoints entered as part of the mission plan, or from a high level instruction (e.g. to track a target or loiter). For manned submarines, it is common for guidance commands to be entered manually into the autopilot at the appropriate time as a change to the depth or heading

• setpoint. The facility to perform continuous path

following, to automatically stalk a target while maintaining a set stand-off distance or perform other complex manoeuvres would potentially be a useful application of further autonomy. 3.6 Fault Tolerance Both AUVs and XLUUVs require the ability to detect and respond to faults to ensure that either the mission can continue or, in the worst case, that the vessel can be safely recovered. When operating in hostile territory, simply surfacing could lead to the loss of the vessel so it is vital for the control system to use all available means to maintain depth and heading control.An important example is a hydroplane failure event, such as a plane jam or runaway. If this should occur on e.g. an XLUUV having the common X-plane stern configuration , with 4 independently actuated planes, then the control system is required to adjust to use the remaining 3 planes to arrest any excursion and regain control in the vertical and lateral planes. To achieve fault tolerance to this specific failure case is not particularly difficult. The adaptive control algorithms discussed previously are ideally suited to this situation. However, the most general ‘reconfigurable control’ problem for XLUUVs which might involve making use of trim and ballast, speed control and any available manoeuvring thrusters would be much more challenging.

UDT 2020 UDT Extended Abstract Template Presentation/Panel The approach to fault tolerance in submarine platform control makes use of redundancy in critical elements such as sensors, computing hardware, actuators and control

• surfaces. There is emphasis on detecting faults and

alerting the operator. Reconfigurable control for plane failures is a useful feature and can reduce excursions but is not an essential requirement for the autopilot because the helmsman can provide the reversionary mode, through either manual or rate hydroplane/rudder control. Adherence to the SME helps ensure that the submarine is recoverable without automatic assistance in the event of a failure. 3.7 Safety Safety considerations are an important issue for manned

• submarines. The platform control software is a safety-

critical application which will typically have to meet a Safety Integrity level (SIL), the stringency depending on the vessel type. Certification of safety critical software is a time- consuming process and including complex mathematical algorithms can add considerably to the V&V burden. Stirling have developed a prototype Model Predictive Control (MPC) autopilot[2], that has some advantages

• ver traditional designs, and which can be used to

illustrate some of the issues. The MPC algorithm ultimately requires a large quadratic programming problem to be solved at each update step. Ideally, the quadratic programming solver would be an existing certified utility function. If this is not the case, then the effort required to implement and certify from scratch is considerable and very specialised mathematical knowledge is required to understand, document and verify the method. Open source code is not usually suitable for safety critical applications. Even if commercial source code of suitable provenance exists to solve the quadratic programming problem (e.g. that generated by the Simulink autocode function), it is still necessary for the code to meet coding standards consistent with the SIL. It may also be that a preferred, less common, language is imposed for safety critical work, such as ADA, which would further reduce the likelihood of existing utility functions being available. Timing issues are also of crucial importance in real- time software and it is essential that the control algorithm should complete its update step within a specified period. For the MPC algorithm in question, there is no formal guarantee that the algorithm will terminate at all, still less within a given interval. Reversionary software modes are therefore required to handle this eventuality. The use of hard to understand mathematical algorithms without guaranteed behaviour can, in itself, cause concerns for those responsible for software safety

• certification. A particularly important example is in the

use of ‘deep learning’ in safety critical applications but similar criticisms could be levelled at the adaptive control methods discussed earlier. While the problems identified above are not insurmountable, they do demonstrate that the decision to implement a more complex platform control algorithm on a submarine requires a clear benefit (e.g in performance

• r reduced manning) to offset the drawbacks.

Platform control for AUVs is unlikely to be considered a safety-critical system. For XLUUVs, the cost of the loss of a vessel, and the possibility that its large size could cause damage to manned craft, may lead to the platform control being designated a safety-critical function.However, since there is no alternative to automation on an XLUUV, the equation is different when it comes to deciding whether the difficulty in certifying a method outweighs the benefits.

4 Conclusions

This paper has compared the platform control problem for traditional AUVs, manned submarines and the new generation of XLUUVs. The areas where the techniques and approaches to submarine platform control are particularly relevant to XLUUVs have been highlighted. In addition, automated functions that are essential on manned XLUUVs have been identified that could be introduced to manned submarines to increase autonomy in future systems.

References

[1] R. Mansfield, D. Venn, Warship 2011: Naval Submarines & UUVs (2011) [2] N. Thomas, J. Green, Warship 2014: Naval Submarines & UUVs (2014)

Author/Speaker Biographies

Nathan Thomas holds the current position of Principal Engineer at Stirling Dynamics. He is a specialist in control design, modelling and simulation. His previous experience includes design and development

• f

submarine control systems and modelling of submarine

• dynamics. Nathan holds a degree in Mathematics and a

PhD in Electrical and Electronic Engineering. Chris Harris is a Principal Engineer, having worked for Stirling Dynamics the last 12 years. Chris currently works in the Marine Systems and Naval Architecture group specialising in the design of steering and diving control systems for in-service and future submarines