Lieboud Van den Broeck

Time optimal MPC for mechatronic applications

Model predictive control (MPC) is an on-line control technique originally developed for slow processes which makes an assessment between input effort and output error while respecting constraints on inputs and outputs. Due to improved computing power and algorithms, MPC is nowadays also applied to mechatronic systems. For these systems, achieving minimal settling time is the main concern, while the input cost is usually of less importance. Hence, this paper presents a new type of MPC; time optimal MPC (TOMPC) which minimizes the settling time of the system. Theoretical considerations show that TOMPC is stabilizing. Simulations show the merits of this technique and indicate that it is applicable in real-time.

Embedded Optimization for Input Shaping

Traditional input shaping filters are linear mappings between reference input and system input. These filters are often unnecessarily conservative with respect to input and output bounds if multiple references with different amplitudes are applied. This conservatism is due to its offline design and linear mapping. This paper presents an online input prefilter design approach to overcome this conservatism. The resulting prefilters are called predictive prefilters because the online design is based on the model predictive control (MPC) framework. By theoretical considerations, simulation results and experimental results, it is shown that this new prefilter is at least as good as traditional prefilters, and can result in substantial gains in settling time. Tests show that a 30% decrease in settling time is possible in a common input shaping application.

Performant design of an input shaping prefilter via embedded optimization

Traditional input shaping filters are linear mappings between the reference input and the system input. These filters are often unnecessarily conservative with respect to input and output bounds if multiple references with different amplitudes are applied. This conservatism is due to the off-line computation of the prefilter. This paper presents an on-line input prefilter design approach to overcome this conservatism. The resulting prefilters are called predictive prefilters because the on-line design is based on the model predictive control (MPC) framework. By theoretical considerations, simulation results and experimental results, it is shown that this new prefilter is at least as good as traditional prefilters, and can result in substantial gains in settling time. Tests show that a 30 % decrease in settling time is possible in a classical application.

Experimental validation of time optimal MPC on a flexible motion system

This paper discusses the application and experimental validation of time-optimal model predictive control (TOMPC) on a flexible motion system, an overhead crane with fixed cable length. TOMPC realizes minimal settling times for point-to-point motions taking into account system constraints. Results of several different point-to-point motion experiments and of disturbance rejection experiments clearly demonstrate this time optimal behavior. The extension of the TOMPC algorithm to increase its feasibility range required for large point-to-point motions is discussed and validated.

Experimental validation of time optimal MPC on a linear drive system

Model Predictive Control (MPC) is a control technique capable of accounting for constraints on inputs, outputs and states, and traditionally makes a trade-off between output error and input cost. Originally developed for slow processes, MPC is nowadays also applied to faster systems such as mechatronic systems, thanks to increased computer power and more advanced algorithms. For these systems however, time optimality is often of the utmost importance, a feature that is not present in traditional MPC. This paper therefore presents and validates a new type of MPC, time optimal MPC (TOMPC), which minimizes the settling time. An experimental validation of TOMPC on a linear drive system with a sampling time of 5ms is performed and comparison with traditional MPC and linear feedback systems is given.

Model predictive control for time-optimal point-to-point motion control

Abstract This paper presents a time-optimal model predictive control approach, which is a new type of MPC approach for motion systems that aims at performing time-optimal point-to-point motions. It is shown how the structure of the underlying optimal control problem can be exploited such that it can be solved in real-time at a sampling rate of 200 Hz. Numerical and experimental validation on a linear motor drive system and comparison with traditional MPC and linear control techniques clearly demonstrate the advantages of time-optimal model predictive control.

A two-level optimization based learning control strategy for wet clutches

Abstract This paper proposes a two-level control strategy for wet clutches. On the low level, the control signal is calculated by solving a constrained optimal control problem. On the high level, the measured responses are used to update the system models and constraints that are used in the optimization for the next control signal. In this way a learning algorithm is obtained, which is able to optimize the control signal during normal operation, despite its complex and time-varying dynamic behavior, and without requiring long calibrations or complex models. The performance and robustness of this control scheme are validated on an experimental test setup.