Kosmas Knödler

Evolutionary Search for Smooth Maps in Motor Control Unit Calibration

We study the combinatorial optimization task of choosing the smoothest map from a given family of maps, which is motivated from motor control unit calibration. The problem is of a particular interest because of its characteristics: it is NP-hard, it has a direct and important industrial application, it is easy-to-state and it shares some properties of the wellknown Ising spin glass model. Moreover, it is appropriate for the application of randomized algorithms: for local search heuristics because of its strong 2-dimensional local structure, and for Genetic Algorithms since there is a very natural and direct encoding which results in a variable alphabet. We present the problem from two points of view, an abstract view with a very simple definition of smoothness and the real-world application. We run local search, Genetic and Memetic Algorithms. We compare the direct encoding with unary and binary codings, and we try a 2-dimensional encoding. For a simple smoothness criterion, the Memetic Algorithm clearly performs best. However, if the smoothness citerion is more complex, the local search needs many function evaluations. Therefore we prefer the pure Genetic Algorithm for the application.

Modellbasierte Online-Optimierung moderner Verbrennungsmotoren

Der erste Teil dieses zweiteiligen Beitrags stellte die Basis des mbminimize-Algorithmus vor, der von der Universitat Tubingen in Zusammenarbeit mit der BMW Group zur modellbasierten Online-Optimierung von Verbrennungsmotoren entwickelt wurde. Der vorliegende Artikel beschreibt die Erweiterungen hinsichtlich der Behandlung von Motorlimits wie zum Beispiel Motorklopfen. Den Schwerpunkt stellen Modelle fur Motorlimits dar, die es erlauben, den Suchraum sukzessive und kontrollierbar einzuschranken.

Using Memetic Algorithms for Optimal Calibration of Automotive Internal Combustion Engines

Many combinatorial optimization problems occur in the calibration of modern automotive combustion engines. In this contribution, simple hill-climbing algorithms (HCs) for three special problems are incorporated in Memetic Algorithms (MAs) using specific crossover and mutation operators. First, the k-exchange algorithm as a well known technique of D-optimal design of experiments (DOE) is improved. Second, a (near-)optimum test bed measurement scheduling (TBS) as a variant of the traveling salesman problem (TSP) is calculated, and third, the final design of look-up tables (LTD) for electronic control units is optimized. It is shown that in all cases MAs that work on locally optimal solutions calculated by the corresponding HCs significantly improve former results using Genetic Algorithms (GAs). The algorithms have been successfully applied at BMW Group Munich.

Optimal Energy Efficiency, Vehicle Stability and Safety on the OpEneR EV with Electrified Front and Rear Axles

This paper relates to the European publicly funded project OpEneR. A central field of research and development in this co-operative project, is energetically optimal operation of the twin electric axle drivetrain of the fully electrified OpEneR prototype cars. Advanced operation strategies for the two 50 kW e-machines are being developed, which consider both energetically optimal propulsion and energetically optimal braking of the vehicle. Both e-machines are used in such a way, that the maximum possible amount of energy can be electrically recovered. However, during electrical recuperation the impact of the brake force distribution on the vehicle’s stability, e.g. during braking on split–μ surfaces, and thus safety, has to be carefully considered. In this work, an advanced co-simulation approach which allows the virtual evaluation of different electrical recuperative braking, frictional braking and related control strategies is described.

Genetic Algorithms Solve Combinatorial Optimisation Problems in the Calibration of Combustion Engines

Several combinatorial optimisation problems occur during the calibration of combustion engines. In this work, it is shown that three particular process steps benefit from genetic algorithms: First, the D-optimal experimental design is improved by the use of an appropriate crossover operator. Thereby the heuristics DETMAX or k-exchange perform a local search. The second problem concerns the optimal test bed scheduling for a more efficient and thus less expensive execution of measurements. This higher dimensional variant of the Travelling Salesman Problem (TSP) is solved by a hybrid genetic algorithm using adjacency coded individuals and a 2-opt heuristic as a local search. Finally, well-defined look-up tables, that lead to smooth maps, are composed from multiple valued look-up tables. Again a genetic algorithm finds better solutions than local search heuristics.

On the Efficient Construction of Rectangular Grids from Given Data Points

Many combinatorial optimization problems provide their data in an input space with a given dimension. Genetic algorithms for those problems can benefit by using this natural dimension for the encoding of the individuals rather than a traditional one-dimensional bit string. This is true in particular if each data point of the problem corresponds to a bit or a group of bits of the chromosome. We develop different methods for constructing a rectangular grid of near-optimal dimension for given data points, providing a natural encoding of the individuals. Our algorithms are tested with some large TSP instances.

Controlling model trust with compactly supported smooth RBF

Building models for any kind of complex process is an important tool of todays applied computer science. There are many situations where the trust in the model varies over the input space, and where the amount of trust or confidence should significantly affect the behaviour of the model and the resulting decisions (this applies when the model is used within some decision process, e.g., in a control or optimization task). In this paper, we will focus on special one-sided situations where overestimating the true process is considered critical, while underestimating is tolerable (or conversely). We introduce a new type of radial basis function, the confidence term, with the following properties: (a) it is smooth, i.e., infinitely differentiable and (b) compactly supported. We show how one-sided trust control can be achieved for any kind of model by a simple multiplication with the confidence term. To demonstrate the power and flexibility of our approach, two quite different applications are presented, both of which are practically relevant. One is model-based optimization with constraints, where we have to be careful not to narrow the search space too quickly, until we can trust the constraint model. This requires imposing a low confidence on the constraint model until enough data is available. In the other application, active learning with multiple point queries, we need to achieve the opposite and impose a high value of trust in regions that have been already explored.

Model-based online optimisation of modern internal combustion engines

This two-part article presents the model-based optimisation algorithm ”mbminimize“. It was developed in a corporate project of the University of Tubingen and the BMW Group for the purpose of optimising internal combustion engines online on the engine test bed. The first part concentrates on the basic algorithmic design, as well as on modelling, experimental design and active learning. The second part will discuss strategies for dealing with limits such as knocking.