Isak Nielsen

An Ô (log N) Parallel Algorithm for Newton Step Computation in Model Predictive Control

The use of Model Predictive Control is steadily increasing in industry as more complicated problems can be addressed. Due to that online optimization is usually performed, the main bottleneck with Model Predictive Control is the relatively high computational complexity. Hence, much research has been performed to find efficient algorithms that solve the optimization problem. As parallel hardware is becoming more commonly available, the demand of efficient parallel solvers for Model Predictive Control has increased. In this paper, a tailored parallel algorithm that can adopt different levels of parallelism for solving the Newton step is presented. With sufficiently many processing units, it is capable of reducing the computational growth to logarithmic in the prediction horizon. Since the Newton step computation is where most computational effort is spent in both interior-point and active-set solvers, this new algorithm can significantly reduce the computational complexity of highly relevant solvers for Model Predictive Control.

Low-rank modifications of Riccati factorizations with applications to Model Predictive Control

In optimization algorithms used for on-line Model Predictive Control (MPC), the main computational effort is spent while solving linear systems of equations to obtain search directions. Hence, it is of greatest interest to solve them efficiently, which commonly is performed using Riccati recursions or generic sparsity exploiting algorithms. The focus in this work is efficient search direction computation for active-set methods. In these methods, the system of equations to be solved in each iteration is only changed by a low-rank modification of the previous one. This highly structured change of the system of equations from one iteration to the next one is an important ingredient in the performance of active-set solvers. It seems very appealing to try to make a structured update of the Riccati factorization, which has not been presented in the literature so far. The main objective of this paper is to present such an algorithm for how to update the Riccati factorization in a structured way in an active-set solver. The result of the work is that the computational complexity of the step direction computation can be significantly reduced for problems with bound constraints on the control signal. This in turn has important implications for the computational performance of active-set solvers used for linear, nonlinear as well as hybrid MPC.

Distributed primal–dual interior-point methods for solving tree-structured coupled convex problems using message-passing

In this paper, we propose a distributed algorithm for solving coupled problems with chordal sparsity or an inherent tree structure which relies on primal–dual interior-point methods. We achieve this by distributing the computations at each iteration, using message-passing. In comparison to existing distributed algorithms for solving such problems, this algorithm requires far fewer iterations to converge to a solution with high accuracy. Furthermore, it is possible to compute an upper-bound for the number of required iterations which, unlike existing methods, only depends on the coupling structure in the problem. We illustrate the performance of our proposed method using a set of numerical examples.

A parallel structure exploiting factorization algorithm with applications to Model Predictive Control

In Model Predictive Control (MPC) the control signal is computed by solving a constrained finite-time optimal control (CFTOC) problem at each sample in the control loop. The CFTOC problem can be solved by, e.g., interior-point or active-set methods, where the main computational effort in both methods is known to be the computation of the search direction, i.e., the Newton step. This is often done using generic sparsity exploiting algorithms or serial Riccati recursions, but as parallel hardware is becoming more commonly available the need for parallel algorithms for computing the Newton step is increasing. In this paper a tailored, non-iterative parallel algorithm for computing the Newton step using the Riccati recursion is presented. The algorithm exploits the special structure of the Karush-Kuhn-Tucker system for a CFTOC problem. As a result it is possible to obtain logarithmic complexity growth in the prediction horizon length, which can be used to reduce the computation time for popular state-of-the-art MPC algorithms when applied to what is today considered as challenging control problems.

Reduced memory footprint in multiparametric quadratic programming by exploiting low rank structure

In multiparametric programming an optimization problem which is dependent on a parameter vector is solved parametrically. In control, multiparametric quadratic programming (mp-QP) problems have become increasingly important since the optimization problem arising in Model Predictive Control (MPC) can be cast as an mp-QP problem, which is referred to as explicit MPC. One of the main limitations with mp-QP and explicit MPC is the amount of memory required to store the parametric solution and the critical regions. In this paper, a method for exploiting low rank structure in the parametric solution of an mp-QP problem in order to reduce the required memory is introduced. The method is based on ideas similar to what is done to exploit low rank modifications in generic QP solvers, but is here applied to mp-QP problems to save memory. The proposed method has been evaluated experimentally, and for some examples of relevant problems the relative memory reduction is an order of magnitude compared to storing the full parametric solution and critical regions.

Low-Rank Modifications of Riccati Factorizations for Model Predictive Control

In model predictive control (MPC), the control input is computed by solving a constrained finite-time optimal control (CFTOC) problem at each sample in the control loop. The main computational effort when solving the CFTOC problem using an active-set (AS) method is often spent on computing the search directions, which in MPC corresponds to solving unconstrained finite-time optimal control (UFTOC) problems. This is commonly performed using Riccati recursions or generic sparsity exploiting algorithms. In this paper, the focus is efficient search direction computations for AS type methods. The system of equations to be solved at each AS iteration is changed only by a low-rank modification of the previous one, and exploiting this structured change is important for the performance of AS-type solvers. In this paper, theory for how to exploit these low-rank changes by modifying the Riccati factorization between AS iterations in a structured way is presented. A numerical evaluation of the proposed algorithm shows that the computation time can be significantly reduced by modifying, instead of re-computing, the Riccati factorization. This speedup can be important for AS-type solvers used for linear, nonlinear, and hybrid MPC.

Depth and Temperature Control During Friction Stir Welding of 5 cm Thick Copper Canisters

Depth control is needed to repeatedly produce welds with minimum flash formation and hook defect, which disturb the temperature control and reduce the corrosion barrier, respectively. The need for depth control is mainly caused by different manufacturing techniques and heat treatments of the lids and tubes that lead to varying properties. The depth is measured using four different sensors; a laser sensor, two linear variable differential transformers (LVDT), and an axial position sensor. The actual depth is estimated from measurements of the shoulder footprint, and can then be compared with the depth sensors. The depth controller uses the axial force to manipulate the shoulder depth during the first two weld sequences. Thirteen welds were carried out in three different lids/rings (twelve short and one full circumferential) with an active depth controller. The laser sensor was used as feedback signal to the controller, and the desired shoulder depth was set to 2.2 mm. For comparison, eighteen short welds (also in three different lids/rings) were performed without any depth control. For the thirteen welds with active depth control, the shoulder depth measured by the laser sensor varied between 2.19 and 2.40 mm (0.21 mm span) at a point two degrees into the joint line, i.e. with a maximum control error of 0.20 mm. The shoulder footprint depth ranged between 2.52–2.86 mm (0.34 mm span). For the eighteen uncontrolled welds, the laser varied between 1.74 and 2.49 mm (0.75 mm span). The corresponding span for the footprint depth was 0.60 mm. Furthermore, macro samples from the thirteen depth controlled welds showed no signs of hook defect nor of joint line remains. The flash formations from the same welds were also small (0–1 mm). It was concluded that the LVDT sensor placed in the lid is best suited for feedback to the depth controller, partly because it is best at modelling the shoulder footprint depth out of the four depth sensors, but also since it does not suffer from extensive measurement noise like the laser sensor.

An O (log N) parallel algorithm for newton step computations with applications to moving horizon estimation

In Moving Horizon Estimation (MHE) the computed estimate is found by solving a constrained finite-time optimal estimation problem in real-time at each sample in a receding horizon fashion. The constrained estimation problem can be solved by, e.g., interior-point (IP) or active-set (AS) methods, where the main computational effort in both methods is known to be the computation of the search direction, i.e., the Newton step. This is often done using generic sparsity exploiting algorithms or serial Riccati recursions, but as parallel hardware is becoming more commonly available the need for parallel algorithms for computing the Newton step is increasing. In this paper a newly developed tailored, non-iterative parallel algorithm for computing the Newton step using the Riccati recursion for Model Predictive Control (MPC) is extended to MHE problems. The algorithm exploits the special structure of the Karush-Kuhn-Tucker system for the optimal estimation problem. As a result it is possible to obtain logarithmic complexity growth in the estimation horizon length, which can be used to reduce the computation time for IP and AS methods when applied to what is today considered as challenging estimation problems. Furthermore, promising numerical results have been obtained using an ANSI-C implementation of the proposed algorithm, which uses Message Passing Interface (MPI) together with InfiniBand and is executed on true parallel hardware. Beyond MHE, due to similarities in the problem structure, the algorithm can be applied to various forms of on-line and off-line smoothing problems.