Bruno Depraetere

Iterative optimization of the filling phase of wet clutches

This paper considers the control of wet clutches, and presents a two-level control strategy to learn and adapt the control signals during normal machine operation. With this approach it is possible to avoid the current practise of experimental calibrations, where regular recalibrations are needed to compensate for time-varying dynamics, e.g. due to wear and changes in oil temperature. On a low level, the developed controller determines the actuator signal by solving an optimal control problem before each engagement of the clutch. The models and constraints for this optimization problem are iteratively updated by a high-level controller, which consists of a recursive identification algorithm to model the system dynamics, and of an ILC-type algorithm to learn appropriate values for the constraints. The performance and robustness of this control scheme are validated on an experimental test setup.

Dynamic analysis and ℋ2 optimisation of a piezo-based tuned vibration absorber:

In this article, a piezo-based tuned vibration absorber is proposed and theoretically analysed. The proposed device consists of a proof mass, a piezoelectric actuator and a resilient element (spring). An equivalent mechanical model where the piezoelectric element is connected to a general circuit composed of a resistor, an inductor and a capacitor in series (RLC) is presented to illustrate the coupling of the electrical components with the mechanical systems. Based on the mechanical replacement model, a C or L circuit can be used to realise a piezo-based tuned vibration neutraliser, while with an RC or RL circuit the piezo-based tuned vibration absorber can be considered as a piezo-based tuned mass damper. For the C and L circuits, tuning strategies are derived to adjust the shunt capacitance and inductance to track the disturbance frequency. In the case of RC and RL shunt circuits, an ℋ2 optimisation criterion is used for tuning the piezo-based tuned mass damper. Closed-form expressions for the optimal tuning parameters are provided in this article.

Model-free and model-based time-optimal control of a badminton robot

In this research, time optimal control is considered for the hit motion of a badminton robot during a serve operation. For this task the racket always starts at rest in a given position and has to move to a target state, defined by a target position and a non-zero target velocity. The goal is to complete this motion in as little time as possible, yet without violating bounds on the actuator. To find controllers satisfying these requirements, a reinforcement learning approach is implemented, using a Natural Actor-Critic (NAC) reinforcement learning algorithm. This approach is experimentally shown to yield the desired robot motions after about 200 trials. Next to this model-free learning approach, the control signals obtained with a model-based optimization are also applied to the robot. The results achieved with both approaches are compared, and a thorough analysis is presented, highlighting the properties of each approach, as well as their advantages and drawbacks.

Identification of linear systems using output measurements with only two possible values

Classical identification cannot be applied when no output measurements are available. In many situations however, discrete information on the unmeasured outputs can still be obtained and used to identify the underlying dynamics. An example is a moving object where an optical sensor can detect whether or not is in the sensors line of sight but whose position is not measured. Using these discrete data sources to estimate a model for the underlying dynamics is equivalent to the estimation of the linear parameters of a Wiener system, which has a known but non-invertible static non-linearity with two output levels. Techniques are derived to perform this estimation, using sequential quadratic programming to minimize a least squares goal function. Simulations are used to validate the proposed approach, yielding good convergence of the linear model parameters to their targets and a high prediction accuracy for the unmeasured variable of the Wiener system.

A fast pick-and-place prototype robot: design and control

This article focuses on a newly built research platform (a robot). Using its vision system, it can identify round objects that are randomly scattered on a table. Then, using its gripper they are picked and placed inside a basket. The control system is tuned such that, the succession of operations runs fast and safe. In this paper we present how this has been achieved: going from concepts to design, validation in simulation and eventually experimental validation. Lessons learned can save time for other parties interested in building prototype robots.

Adaptive-passive control of structure-borne noise of rotating machinery using a pair of shunted inertial actuators

In this article, two piezo-based rotating inertial actuators are considered for the suppression of the structure-borne noise radiated from rotating machinery. Each inertial actuator comprises a piezoelectric stack element shunted with the Antoniou’s gyrator circuit. This type of electrical circuit can be used to emulate a variable inductance. By varying the shunt inductance it is possible to realise two tuneable vibration neutralisers to suppress tonal frequency vibrations of a slowly rotating machine. Also, reductions in the noise radiated from the machine housing can be achieved. First, a theoretical study is performed using a simplified lumped parameter model of the system at hand. The simplified model consists of a rotating shaft and two perpendicularly mounted shunted piezo-based rotating inertial actuators. Second, the shunted piezo-based rotating inertial actuators are tested on an experimental test bed comprising a rotating shaft mounted in a frame. The noise is radiated by a plate that is attached to the frame. The experimental results show that a reduction of 11 dB on the disturbance force transmitted from the rotating shaft through the bearing to the housing can be achieved. This also generates a reduction of 9 dB for the plate vibration and the radiated noise.

Self-tuneable velocity feedback for active isolation of random vibrations in subcritical two degree of freedom systems

It has been previously shown that skyhook damping can be used to actively reduce vibration transmission between masses in supercritical 2 degree of freedom (dof) systems. The method is based on measuring the absolute velocity of the clean body, multiplying it by a negative gain, and feeding the result back to a force actuator reacting between the clean and the dirty body. This approach results in a broadband vibration isolation. For subcritical 2 dof systems this is normally not possible due to control stability problems. These stability problems can be mitigated by including an appropriate amount of relative damping between the clean and the dirty body in addition to the absolute damping. This approach has been referred to as blended velocity feedback. In this paper the application of the blended velocity feedback on subcritical 2 dof systems is investigated using an auto-tuning controller. An algorithm to gradually change the relative and absolute feedback gains until the active isolation performance reaches its best by applying an optimal combination of the two gains is applied. There is only one such optimal combination which minimises the kinetic energy of the clean body, and consequently the performance surface has a global minimum. Furthermore there are no local minima so a trial and error algorithm could be applied. Although in the frequency domain finding the minimum of the performance surface is straightforward, in the time domain the determining the clean body mean squared velocity can take a considerable time per step of the algorithm, such that the convergence of the trial and error algorithm can be relatively slow. It is hypothesized that more sophisticated algorithms may speed-up the convergence but this would be at cost of using a model-based approach.

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.

A reference free iterative learning strategy for wet clutch control

This paper presents a new iterative learning strategy to control wet clutches. These are complex hydraulic systems that are commonly used in automatic transmissions of heavy duty vehicles, and their control aims at performing fast and smooth engagements. Learning is used to overcome the need for complex models and to maintain performance despite large variations in the system behavior. Classical iterative learning control techniques can however not be employed directly since reference trajectories corresponding to the performance requirements are unavailable. Instead, the presented iterative learning strategy translates the performance requirements directly into an objective function and constraints, hence constituting a numerical optimization problem. After each engagement, this problem is solved in order to find the control signal for the next engagement, using a piecewise linear model for the clutch. Learning is included by using the measured response data to update the models and constraints used by the optimization problem. The presented strategy is successfully validated on an experimental test bench containing wet clutches. The learning process is shown to converge towards the desired engagement quality, and a demonstration is given of the robustness with respect to changes in the operating conditions.

An ECMS-based powertrain control of a parallel hybrid electric forklift

In this paper we focus on the supervisory control problem of a parallel hybrid electric vehicle (HEV): minimize fuel consumption while ensuring self-sustaining State-of-Charge (SoC). We reapply the state of the art methodology by comparing optimal results of Dynamic Programming (DP) against a real-time control candidate. After careful selection, we opted for an Equivalent Consumption Minimization Strategy (ECMS) based approach for the following reasons: (i) results are quite remarkable with less than 5% fuel usage increase when compared to DP; (ii) simple and intuitive tuning of control parameters; (iii) readily usable for code generation (prototyping). Topics that distinguish this article from others in the literature include: (i) the usage of trapezoidal rule of integration implementing DP and ECMS; consequently, the offline simulation results are intended to be more precise and representative when compared against the more common, often used rectangular rule; (ii) a particular post-processing procedure of the recorded driving cycle data based on physical interpretation; it allows consistent offline simulations with quite high sampling period (in the order of seconds); (iii) tuning of control parameters in such a way that control system is robust towards new, unknown, unpredictable but closely resembling driving cycles. In particular, we focus on the supervisory control of a forklift truck. The real-time control is able to compute: (i) the power split (i.e. a balanced usage between an internal combustion engine and a supercapacitor); (ii) the drivetrain control (i.e. automatic gear shifting and clutching). Numerous numerical implementation issues are discussed along our presentation.