Contributions à la commande de robots sous contraintes
TELEMACH: immersed robots for the teleoperated maintenance of TBMs
TELEMACH (TELE-operated MAintenance for TBMs Cutter-Head) is a research project proposed by Bouygues Travaux Publics in 2007 in response to a call for project of the Agence Nationale de la Recherche (ANR – France). This project lasted 30 months, from February 2008 to September 2010.
Topic and Context
TELEMACH finds its origin in the field of shield tunneling (see Fig. 1.1). There is a clear need to dig deeper and longer tunnels in urban areas, leading to major risks. Current working conditions are such that: • The cutter head must be inspected frequently, if possible daily; • In hyperbaric conditions, the intervention requires creating an air bubble in order to reach the cutter head tools. The realization of this air bubble takes time, and during this operation the front face (separation between the soil and the TBM) is not balanced optimally: an air bubble has an homogenous pressure profile whereas the front has an hydrostatic pressure profile, which is appropriately balanced during the excavation as the excavation room is full of mud and materials, see Fig. 1.2; • Maintenance operations are ensured by operators in harsh conditions: the tools replacement imposes the operators to work in a hyperbaric air bubble within a 3 4 Chapter 1. Context and Problems 1 2 3 4 Figure 1.1: Slurry TBM during an excavation phase: an air bubble is maintained on the rear part of the excavation room to control the pressure of the mud. During the maintenance phase, the air bubble (1) is extended to the whole excavation room (2) to let the operators access the tools on the cutter head (3). The pressure bulkhead (4) separates the hyperbaric area and the rear of the TBM at atmospheric pressure. The diameters retained for the project are 9 m and larger. narrow and dirty area only reachable by airlocks. The operators manipulate heavy tools (for example a disc cutter weighs 150 kg and is roughly an iron cube of 60 cm3 ) in a confined area. • Hyperbaric operations become very complex, long and expensive for pressures superior to 3 bars, which is common (equivalent to 30 m under the earth level). Up to this pressure, operators can be trained to intervene in the excavation room; nonetheless they must respect decompression cycles. Above 3 bars, it is mandatory to resort to professional divers (Le P´echon et al. 2000). Above 4 bars, exceptional measures of deep diving have to be taken: life in a hyperbaric caisson and displacement in a hyperbaric shuttle, helium based breathable gas . . . As an example, on the Westershelde site (Holland), teams of 9 divers remained 3 weeks under a pressure of 6.5 bars (see Fig. 1.2). TELEMACH is a feasibility study for replacing human interventions in the excavation room of TBMs by teleoperated maintenance. The missions to carry out (inspection, cleaning, tools replacement) would be performed by dedicated robotic systems: articulated arms for inspection, tools changer, heavy loads carrier, conveyor and automated airlock doors. The operator would act remotely through teleoperation and would be assisted in real time by a mobile camera, a virtual environment and force feedback. One of the particular aspects of the project concerns the performance of operations immersed in the mud (bentonite): it makes the air bubble unnecessary, which decreases the major risk of collapsing of the front face while reducing the idling time. Contributions to the Control of Constrained Robots 5 Figure 1.2: Left: the front face needs a hydrostatic profile pressure to be adequately balanced: using an air bubble constitutes a risk as the front face is not stable. Middle: Mobile airlocks used by divers for depressurization phases. Right: Operator during a depressurization phase.
TELEMACH consortium
TELEMACH was a feasibility project gathering various partners (see Fig. 1.3). The project was led by Bouygues Travaux Publics (BYTP) – one of the leading contractors in tunnel construction. The TBM design aspects were ensured by Herrenknecht (HK) – the world leader in the design and manufacturing of TBMs. The architectural development studies were led by Cybern´etix (CYX) – a leading SME specialized in remote operations for this industry. The Commissariat `a l’Energie Atomique et aux Energies Alternatives, Laboratoire d’Int´egration des Syst`emes et des Technologies (CEA-LIST) was responsible for the definition and development of the “hand-eye” support functions for the operator. The Institut des Syst`emes Intelligents et de Robotique (ISIR), a university based research laboratory brought its support for the initial design and control of the robotic manipulator. Figure 1.3: Consortium of TELEMACH and main protagonists during a TBM visit on A41 site (Geneva).
TELEMACH achievements
TELEMACH ended in September 2010. The feasibility was demonstrated through various achievements: • The TBM was modified and an original heavy load handling system was developed. A whole flow study (man/tools/equipment) led to a complete reorganization of the airlocks. Security constraints such as the capability to reach any area with a stretcher were taken into account. The equipment airlocks, as the whole logistic chain, were automated to avoid direct man interventions. All actuators were deported in the rear part of the TBM at atmospheric pressure. The Disc cutter Tool Changer (DTC), a parallel robot (Stewart platform), was developed to unmount/extract worn disc cutters and insert/mount new disc cutter (Fig. 1.4). A mobile platform enables to reach the disc cutter casings by moving along a vertical wall. The whole system is under patent application. Figure 1.4: Disc cutter Tool Changer. • New tools for Interactive Teleoperation were developed. A real time loop between a supervision system, the master arm and the slave arm was established with a satisfying transparency/stiffness compromise for the user. This framework enables to generate force feedback from a real time updated virtual environment. It opens the way to many applications developed along the project, for example: – Dynamic anti-collision (David et al. 2011): at each time step, a distance calculus between the robot and the environment is used to generate avoidance forces on the master arm proportional to the proximity to the environment. – Point of view optimization: a robot holding a camera holder is directly controlled by the scene tracking while avoiding the occultations thanks to an active anti-collision between the environment and a virtual cylinder representing the camera field of view. All the developments have been experimented on a full scale mock-up (see Fig. 1.5) with a Maestro1 articulated arm Figure 1.5: Full scale mockup of a 1/4 TBM cutter head. • Intervention in the mud was assessed. Intervention in the mud offers a major security advantage since pressure is balanced at the front (see Fig. 1.2). In this scope both touch-based recognition and ultrasound perception were assessed. On the one hand, peg-in-hole experiments ensured the precise localization of landmarks in the excavation room. On the other hand, the mud has been characterized (absorption, dispersion, sensitivity to pressure and temperature) and proper parameters were found to lead to satisfying observation of the tools. Other TELEMACH contributions are part of this thesis work and are described in the next sections.
Control problems met in TELEMACH Through
TELEMACH, general control problems are addressed. The 2 main topics of TELEMACH dealing with control are: • the task based design of a manipulator thanks to a genetic algorithm which fitness is a trajectory tracking; • the real time control of this manipulator in a cluttered environment.
From a design to a control problem
The kinematic design of robotic manipulators is often seen as one among numerous applications of engineering design. However, the workspace of the excavation chamber is cluttered, so usual design techniques are inefficient and time-consuming. Moreover, the complexity of the TBM environment and the tasks to be accomplished induce a high number of potential design solutions among which the best ones may, given their originality with respect to usual design problems solutions, probably not arise using 8 Chapter 1. Context and Problems classical design methods. The use of dedicated CAD tools may help to numerically discard some of the potential solutions, but checking each robot candidate with respect to a representative subset of tasks and environments still remains a complex and time consuming work. Instead, it is proposed to follow an approach where the design process is considered as a multi-objective optimization problem: tasks and constraints are formulated in terms of functions to optimize and constraints to satisfy. Such a formulation allows the automation of the design process in the preliminary phase. Given a family of automatically obtained solutions, the so-called classical design methods can then be used to converge towards a practical solution. The retained design process is an example of task-based design carried out thanks to an evolutionary process (e.g. Salle et al (2004)). Robot morphologies are generated thanks to a genetic algorithm which evaluation step (fitness function) aims to qualify the ability of a robot to carry out a maintenance mission in the TBM. So, a relevant trajectory has been defined in the simulation environment and the fitness function consists of a trajectory tracking. The objectives (indicators) retained in our problem are voluntarily simple and of a single dimension (no weighted sums representing a priori tradeoffs between different variables). The trajectory tracking quality but also intrinsic parameters, such as the number of DOFs, are evaluated. Each robot morphology must be rated through a fair evaluation, i.e. an evaluation which exploits its physical characteristics optimally. In that scope, the control law used in the simulation is an essential element: it is responsible for the robot behavior, and consequently its scores. As a result, some expected properties are identified at the control level. • Genericity in robots. Each morphology that can be described by the design process must be controllable by the control law without any specific restriction (geometry, degree of freedom, . . . ). The skills of each morphology must be exploited with equal chances. The genericity in robots demands a generic control law, the robot being redundant (the number of DOFs needed to obtain the desired motion is lower than the actual number of DOFs of the robot) or not. The terms of the control laws should be automatically computed without particular tuning; • Genericity in situations. The control law must take into account and manage appropriately specific conditions such as singular configurations, constrained areas, oscillating behaviors, . . . The genericity in situations requires the potential singularities to be automatically treated, and a correct behavior (no oscillations, smoothness) for all cases. If matrix singularities can be treated by the approximations of Damped Least Square Inversion (DLS, Wampler 1986 and Nakamura et al. 1986), the solution to a smooth and oscillations-free trajectory needs to be formalized and solved; • Representativity. The control law used in the fitness should produce realistic behaviors to get meaningful evaluations and meaningful results. In particular, collisions should not occur in the simulations, as they cannot occur in reality. Representativity induces the need to handle constraints at the top priority level (the robot should not collide and, if possible, track the trajectory). Most of these Contributions to the Control of Constrained Robots 9 constraints are expressed through inequalities, which induces a non linearity in the problem. As a consequence, simple inversion operators are not able to satisfy those constraints and the resort to iterative algorithms is needed2 to know which constraint can be taken as an equality (active constraint) and which cannot; • Coherency with indicators. Robots are evaluated though performance indicators (for example trajectory tracking error, effort transmission capabilities, . . . ). The control law must take these indicators into account. Coherency with indicators induces the ability to deal with multiple objectives (minimization of the trajectory tracking error, getting close to a reference configuration, etc.). These objectives does not have the same priority, thus the controller should satisfy a strict priority between them. As for the constraints avoidance, getting closer to the reference configuration should be carried out only if it does not impact the trajectory tracking error minimization. However, the main difference between the constraints and the objectives is that the first one usually prevents motions while the second one generates motions. The satisfaction of multiple objectives in presence of constraints can be treated by a sequence of Quadratic Programs (QP), but it may turn out time consuming, which goes against efficiency; • Efficiency. As a huge number of individuals are evaluated, the control law must not be time-consuming.
Introduction |
