In the presented effort, layered CFRP composites samples with differing thicknesses and cross-sections are manufactured and crushed under quasi-static loading conditions. Simulation of the crushes are conducted using traditional continuum mechanics damage models. Parameters are proposed to represent the post peak-stress material behavior including the residual strengths of the fiber and matrix, as well the ultimate strain for deletion of composite elements. This paper presents a systematic approach to identify optimal values for these post peak-stress parameters based on a methodology incorporating CAE models and numerical optimization. An adaptive meta-model based global optimization strategy, with the objective of matching the force-time characteristics of multiple crush experiments simultaneously, has been established to quantify the values of the CFRP’s post peak-stress degradation and erosion material model parameters through calibration. Using two separate test configurations for optimization, a set of values for those parameters are determined. This parameter set is shown to successfully predict the response of additional test cases, including matching of force-displacement curves and crushing modes. The resulting composite crush simulations show a good quantitative as well as qualitative agreement between simulations and experiments to a degree that is difficult to be achieved solely with previous engineering practice.

Due to strict CO2 emission limits, the optimal design of controllers for hybrid cars is an increasingly important topic for the automotive industry. Most current approaches to controller design rely solely on engineering knowledge. Utilizing technologies from computational intelligence is not yet common practice. In this work we evaluate how simple controllers can automatically be extracted from optimal control strategies computed by Dynamic Programming. We compare artificial neural network and decision tree based controllers in terms of performance (fuel consumption), stability, robustness, and interpretability, and we investigate the dependency on specific drive cycles used for generating optimal control. Our findings indicate that automatically derived controllers can result in a performance 1-2% below optimal fuel economy, but we observe a large variety in performance and in the controller structure for different drive cycles, thus, underlining the relevance of the correct choice of the drive cycles used for controller development. We also outline the impact of typical learning related issues like overfitting on the practical development process.

Although the integration of engineering data within the framework of product data management systems has been successful in the recent years, the holistic analysis (from a systems engineering perspective) of multi-disciplinary data or data based on different representations and tools is still not realized in practice. At the same time, the application of advanced data mining techniques to complete designs is very promising and bears a high potential for synergy between different teams in the development process. In this paper, we propose shape mining as a framework to combine and analyze data from engineering design across different tools and disciplines. In the first part of the paper, we introduce unstructured surface meshes as meta-design representations that enable us to apply sensitivity analysis, design concept retrieval and learning as well as methods for interaction analysis to heterogeneous engineering design data. We propose a new measure of relevance to evaluate the utility of a design concept. In the second part of the paper, we apply the formal methods to passenger car design. We combine data from different representations, design tools and methods for a holistic analysis of the resulting shapes. We visualize sensitivities and sensitive cluster centers (after feature reduction) on the car shape. Furthermore, we are able to identify conceptual design rules using tree induction and to create interaction graphs that illustrate the interrelation between spatially decoupled surface areas. Shape data mining in this paper is studied for a multi-criteria aerodynamic problem, i.e. drag force and rear lift, however, the extension to quality criteria from different disciplines is straightforward as long as the meta-design representation is still applicable. (C) 2014 The Authors. Published by Elsevier Ltd.

In many engineering domains like aerospace, vehicle or engine design the analysis of flow fields, acquired from computational fluid dynamics (CFD) simulations, can reveal important insights on the behavior of the simulated objects. However, the huge amount of flow data produced by each simulation complicates the data processing and limits the application of computational tools for flow analysis. Thus, an a priori transformation of the flow data into a compact low dimensional representation is desired. This paper introduces a new procedure for transforming flow field data into a compact streamline based representation. Wherein, streamlines with negligible information contribution are removed from the representation. The reduced set of streamlines defines the basis for a subsequent quantification of flow field distances. Experimental studies show that the distances calculated based on the compact representation well approximate the distances of the uncompressed flow field with a significant drop in memory consumption.

The choice of the representation in evolutionary design optimization defines the flexibility and constraints of the search process. Finding an adequate representation is a pre-requisite for the success of an optimization but requires extensive knowledge about the design at hand. Based on the results of an automotive part design optimization, the authors provide evidence that an adaptation of the representation based on sensitivity information leads to new outperforming designs. for retrieving reliable sensitivity estimates a robust variant of the mutual information has been introduced. the robust sensitivity measure provides valuable information for the setup of an improved representation.

Wide exploration of high-dimensional, multimodal design spaces is required for uncovering alternative solutions in the conceptual phase of design optimization tasks. We present a general framework for balancing exploration and exploitation during the course of the optimization that induces sequential exploitation of different optima in the search space by selecting on a solution’s fitness and a dynamic criterion termed interestingness. We use a fitness approximation model as a memory representing the parts of the search space that have been visited before. It guides the optimizer toward those areas that require additional sampling to be correctly modeled, and are hence termed interesting. Next to applying the prediction error of the model as a measure of interestingness, we consider the statistical variation in the predictions made by multiple parallel models as an alternative approach to quantify interestingness. On three artificial test functions we compare these setups running on a canonical ES to the same ES extended with either archive-based novelty, niching, or restarting, and to simply evaluating a Latin Hypercube set of sample points.

In the automotive industry Computational Fluid Dynamics (CFD) simulations have become an important technology to support the development process of a new automobile. During that process, individual simulations of the air flow produce a huge amount of information about the design characteristic, where mostly only a minority of information is used. At the same time knowledge about the relationship between design modifications and their aerodynamic consequences provides valuable insight into the entire aerodynamic system. In this work a computational framework is introduced, providing means to identify relevant interactions within the aerodynamic system based on existing design and flow data. For an efficient modeling, the raw flow field data is reduced to a set of relevant flow features or phenomena. Applying interaction graphs to the aerodynamic data set unveils interacting and redundant structures between design variations and observed changes of flow phenomena. The general framework is applied to an exemplary aerodynamic system representing a 2D contour of a passenger car.

In the conceptional phases of design optimization tasks it is required to find new innovative solutions to a given problem. Although evolutionary algorithms are suitable methods to this problem, the search of a wide range of the solution space in order to identify novel concepts is mainly driven by random processes and is therefore a demanding task, especially for high dimensional problems. To improve the exploration of the design space additional criteria are proposed in the presented work which do not evaluate solely the quality of a solution but give an estimation of the probability to find alternative optima. To realize these criteria, concepts of novelty and interestingness are employed. Experiments on test functions show that these novelty guided evolution strategies identify multiple optima and demonstrate a switching between states of exploration and exploitation. With this we are able to provide first steps towards an alternative search algorithm for multi-modal functions and the search during conceptual design phases.

In large and complex aerodynamic systems the overall performance of a design is mainly defined by interactions between design areas rather than by single design regions. Therefore it is necessary to identify these interactions in order to be able to understand and improve the designs. However, detecting and modeling those interactive effects between distant design areas is a very challenging task which usually requires a detailed understanding of the flow patterns and dedicated expert knowledge. In this paper we apply the information theoretic concept of interaction information to aerodynamic design data in order to detect and quantify interaction effects between distant design regions. Information graphs are suggested in order to provide the results to the aerodynamic engineer in a graphical form. In order to show the feasibility of this approach, the information theoretic quantities are applied to the data of a 2D wing assembly as well as to the 3D turbine blade design data.

Applying numerical optimisation methods in the field of aerodynamic design optimisation normally leads to a huge amount of heterogeneous design data. While most often only the most promising results are investigated and used to drive further optimisations, general methods for investigating the entire design dataset are rare. We propose methods that allow the extraction of comprehensible knowledge from aerodynamic design data represented by discrete unstructured surface meshes. The knowledge is prepared in a way that is usable for guiding further computational as well as manual design and optimisation processes. A displacement measure is suggested in order to investigate local differences between designs. This measure provides information on the amount and direction of surface modifications. Using the displacement data in conjunction with statistical methods or data mining techniques provides meaningful knowledge from the dataset at hand. The theoretical concepts have been applied to a data set of 3D turbine stator blade geometries. The results have been verified by means of modifying the turbine blade geometry using direct manipulation of free form deformation (DMFFD) techniques. The performance of the deformed blade design has been calculated by running computational fluid dynamic (CFD) simulations. It is shown that the suggested framework provides reasonable results which can directly be transformed into design modifications in order to guide the design process.

During CAD development and any kind of design optimisation over years a huge amount of geometries accumulate in a design department. To organize and structure these designs with respect to reusability, a hierarchical set of components on different scalings is extracted by the designers. This hierarchy allows to compose designs from several parts and to adapt the composition to the current task. Nevertheless, this hierarchy is imposed by humans and relies on their experiences. In the present paper a computational method is proposed for an unsupervised extraction of design components from a large repository of geometries. Methods known from the field of object and pattern recognition in images are transferred to the 3D design space to detect relevant features of geometries. The non-negative matrix factorization algorithm (NMF) is extended and tuned to the given task for an autonomous detection of design components. The results of the NMF additionally provide an overview on the distribution of these components in the design repository. The extracted components sum up in a parts-based representation which serves as a base for manual or computational design development or optimisation respectively.

We propose methods that allow the investigation of local modifications of aerodynamic design data represented by discrete unstructured surface meshes. A displacement measure is suggested to evaluate local differences between the shapes. The displacement measure provides information on the amount and direction of surface modifications. Using the displacement measure in conjunction with statistical methods or data mining techniques provides meaningfull knowledge from the data set for guiding further shape optimization processes.

Summary. To reduce the number of expensive tness function evaluations in evolutionary optimization, individual-based and generation-based strategies for metamodel management (evolution control) have been proposed. In this work, four individual-based frameworks for meta-model management are investigated. A feedforward neural network is employed to construct an approximation model of the tness function. Structure optimization of the neural network is used to reduce the approximation error. In an attempt to adapt the number of controlled individuals, adaptation mechanisms are suggested based on the model error, selection error, rank correlation, and tness correlation. Preliminary results indicated that the adaptation mechanisms do not work well as expected. Two of the frameworks are implemented in 3D blade design optimization. The results showed that individual-based meta-model management is promising, though further eorts are still needed to improve the performance of the evolutionary algorithms with meta-models for tness estimation.

Several heuristic methods have been suggested for improving the generalization capability in neural network learning, most of which are concerned with a single-objective (SO) learning tasks. In this work, we discuss generalization improvement in multi-objective learning (MO). As a case study, we investigate the generation of neural network classifiers based on the receiver operating characteristics (ROC) analysis using an evolutionary multi-objective optimization algorithm. We show on a few benchmark problems that for MO learning such as the ROC based classification, the generalization ability can be more efficiently improved within a multi-objective framework than within a single-objective one.

To reduce the number of expensive fitness function evaluations in evolutionary optimization, several individual-based and generation-based evolution control methods have been suggested. This paper compares four individual-based evolution control frameworks on three widely used test functions. Feedforward neural networks are employed for fitness estima- tion. Two conclusions can be drawn from our simulation results. First, the pre-selection strategy seems to be the most stable individual-based evolu- tion control method. Second, structure optimization of neural networks mostly improves the performance of all compared algorithms.