Julian Stoev

Badminton Playing Robot - a Multidisciplinary Test Case in Mechatronics

Abstract We present a Mechatronics design approach and related technologies for a badminton playing robot, as a first stage of a multi-year project. The robot is using non-modified shuttles and rackets, which are detected and localized using purely visual information. The robot subsystems are presented: mechanical design, visual detection of the shuttle, shuttle trajectory estimation and interception, actuation, control hardware and software. The paper demonstrates the multidisciplinary nature of the Mechatronics.

Robust and two-level (nonlinear) predictive control of switched dynamical systems with unknown references for optimal wet-clutch engagement

Modeling and control of clutch engagement has been recognized as a challenging control problem, due to nonlinear and time-varying dynamics, that is, switching between two discontinuous dynamic phases: the fill and the slip. Furthermore, the reference trajectories for obtaining an optimal clutch engagement are not a priori known and may require adaptation to varying operating conditions. Two (nonlinear) model predictive control strategies are proposed based on the partial or full (non)linear identification of these two phases. First, a local linear model of the fill phase is identified and a robust model predictive control is designed to account for the consequent uncertainty in the slip phase. Second, (non)linear models of both the fill and the slip phases are identified and a two-level (nonlinear) model predictive control controller is proposed, where two (nonlinear) model predictive control controllers are designed for the two phases tracking references generated and continuously adapted by high-level iterative learning controllers. The robust and two-level (nonlinear) model predictive controls are validated on a real clutch. The results obtained from the real setup show that the proposed control strategies lead to an optimal engagement of the wet-clutch system.

Improving wet clutch engagement with reinforcement learning

A common approach when applying reinforcement learning to address control problems is that of first learning a policy based on an approximated model of the plant, whose behavior can be quickly and safely explored in simulation; and then implementing the obtained policy to control the actual plant. Here we follow this approach to learn to engage a transmission clutch, with the aim of obtaining a rapid and smooth engagement, with a small torque loss. Using an approximated model of a wet clutch, which simulates a portion of the whole engagement, we first learn an open loop control signal, which is then transferred on the actual wet clutch, and improved by further learning with a different reward function, based on the actual torque loss observed.

ENERGY OPTIMAL POINT-TO-POINT MOTION USING MODEL PREDICTIVE CONTROL

This paper discusses energy optimal point-to-point motion control for linear time-invariant (LTI) systems using energyoptimal Model Predictive Control (EOMPC). The developed EOMPC, which is based on time-optimal MPC, aims at performing energy-optimal point-to-point motions within a required motion time. Energy optimality is achieved by setting the object function of the EOMPC optimization problem equal to the system’s energy losses. The key issue is to utilize the strategy of the prediction horizon to ensure that the motion time is exactly equal to the required motion time. Application of EOMPC on a badminton robot shows the practical applicability of the developed method. In addition, an experimental comparison with time-optimal MPC is provided.

Robust predictive control design for optimal wet-clutch engagement

Modeling of hydraulic clutch transmissions is far from straightforward due to their nonlinear hybrid dynamics, i.e. switching between three dynamic phases. In this paper we identify a local linear model only for the constrained first phase, based on which a predictive controller is used to track a suitable engagement signal. The robustness of this controller in the latter two phases is guaranteed by making the constraints inactive and pre-tuning the control parameters based on its closed loop formulation and applying robust stability theorem. This controller is then implemented in real-time on a wet-clutch test setup and is shown to achieve optimal engagement.

Driver Modeling for Heavy Hybrid Vehicle Energy Management

The paper presents an approach for modeling and predicting the user intentions with application for optimization of the hybrid electrical vehicle. An auto-regressive moving-average model isdesigned to model and predict the driver behavior. The resulting model is converted to a Markov-chain model and used with stochastic dynamic programming, which optimizes the gear-shifting and the power split between the internal combustion engine and the electrical storage of a hybrid electrical vehicle. Verification of resulting energy efficiency is performed using real-life driving data from a heavy-duty industrial vehicle (forklift).

Design and Application of Signals for Nonlinear System Identification

Abstract This paper discusses the design, implementation and the advantages of three types of signals for nonlinear system analysis and identification. They belong to the class of multisine signals and are the random phase, positively skewed and crest factor optimised multisine signals. A straightforward routine to combine such a signal with the systems typical input signal is discussed. The advantages of using such signals is illustrated through the results obtained from identifying the dynamics of a mechanical wet-clutch system.

Identification of linear systems with binary outputs using short independent experiments

Abstract This paper considers the identification of linear systems based on binary measurements of the output. In contrast to existing techniques with strict requirements on the excitation signals, the identification is performed based on a sequence of short and independent measurements. The linear systems are represented using Finite Impulse Response (FIR) models, whose parameters are estimated by exploiting the known characteristics of the binary measurement. Two different methods are derived, both yielding convex parameter estimation problems that can be solved with standard software. The first achieves a high prediction accuracy but yields constrained optimization problems. A second alternative is therefore derived with a slightly worse performance but without constraints, such that solutions can be found more quickly. The identification procedure for both is illustrated on a simulation model.

Switched Nonlinear Predictive Control with Adaptive References for Engagement of Wet Clutches

Abstract This paper discusses the control of nonlinear systems where the reference trajectories are not known beforehand, e.g. the engagement of a wet clutch. The wet clutch is considered as a combination of two disjoint dynamic phases of linear fill and nonlinear slip. Subsequently, a (non)linear system identification is performed of these two phases. With these models at hand, we design separate control strategies. In this paper we show the added value of using a nonlinear model predictive controller (NMPC) to control the nonlinear slip, while a linear MPC suffices for the fill phase control. The reference trajectories for these two phases and the switching moment between them are learnt iteratively. Our results show that nonlinear MPC outperforms linear MPC for the slip phase, leading to an optimal engagement of a real wet clutch setup.

Switched predictive control design for optimal wet-clutch engagement

Abstract Identification and control of wet-clutches has been recognized as a challenging control problem, due to their non-linear time-varying dynamics and their discontinuous switching between filling and slipping phases. In this paper, our approach is based on a full (non)linear system identification of the filling and slipping phases, followed by two closed loop model predictive controllers (MPC) for the two phases and switching between these controllers at suitable moments for optimal torque transfer. Further, it is shown that an intelligent formulation of the switching function is beneficial to the robustness against time-varying model parameters. Finally, our control technique is validated on a real-life wet-clutch test-bench to obtain optimal engagement.