Computers in Industry 60 (2009) 604612

Contents lists available at ScienceDirect

Computers in Industry
journal homepage: www.elsevier.com/locate/compind

Development of an engine crankshaft in a framework of computer-aided innovation

A. Albers, N. Leon-Rovira *, H. Aguayo, T. Maier
t Institute of Product Development (IPEK), Universita Karlsruhe (TH), Center for Innovation in Design & Technology (CIDT), ITESM, Monterrey Campus, Mexico

A R T I C L E I N F O

A B S T R A C T

Article history: Available online 30 June 2009 Keywords: Genetic algorithms Splines Crankshaft imbalance TRIZ FEM

This paper describes the conceptual framework of a general strategy for developing an engine crankshaft based on computer-aided innovation, together with an introduction to the methodologies from which our strategy evolves. It begins with a description of two already popular disciplines, which have their roots in computer science and natural evolution: evolutionary design (ED) and genetic algorithms (GAs). A description of some optimization processes in the eld of mechanical design is also presented. We explain our approach to multi-objective optimization and show how tools like the Pareto diagram can help in identifying conicts. The concepts presented here are exemplied through the optimization of a combustion engine crankshaft. The main premise of the paper is the possibility to optimize the imbalance of a crankshaft using tools developed in this methodology. This study brings together techniques that have their origins in the elds of optimization and new tools for innovation. We reect on how computers can have an active role in the conceptual design process, and explain how TRIZ (Theory of Inventive Problem Solving) can enrich the discipline of ED. The aim of our research is to extend the search for solutions with GAs and present creative, innovative alternatives to the designer. Similarities between GAs and TRIZ regarding ideality and evolution are presented. We also explain how geometric optimization systems (size, shape, topology and topography) offer hints about the next generation of optimization tools. The role of splines in this context is found to be closely integrated with GAs in enabling this development on a computer-aided design and engineering (CAD&CAE) software interface, and in enabling integration with Java programming language for automation of the development. 2009 Elsevier B.V. All rights reserved.

1. Introduction The usefulness of computers in product design ranges from data management, drawing, analysis, and simulation to numerous other applications. Computers are programmed to do what humans intend and are capable of performing far beyond the abilities of the most skilled designer. In the world of mechanical design, software tools are used for a large number of applications, from modelling and optimization tasks to simulating the performance of a product. This allows the engineering designer to concentrate on activities related to software analysis, prototype testing, creativity and innovation. Computer-aided innovation uses software tools to provide extended support for the creative part of the design process. With this support, the designer can improve the performance of his or her concepts, letting computers take part not only in generating variants, but also in making judgments, by

* Corresponding author. Tel.: +52 81 81582012. E-mail address: noel.leon@itesm.mx (N. Leon-Rovira). 0166-3615/$ see front matter 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.compind.2009.05.017

simulation, of these variants. Thus, a designer can explore numerous creative solutions to problems (overcoming design xation or the limitations of conventional wisdom) by generating alternative solutions. Software tools can use knowledge from designers to generate new solutions based on many separate ideas and suggest entirely new design concepts. Methods for structural and topological optimization, based on evolutionary algorithms, are used to obtain optimal geometric solutions. They evolve into congurations that minimize the cost of trial and error. The case study is selected in view of the importance of motor crankshafts in the automotive industry [1] and the increased performance requirements for engines, which have increased the production of forged steel crankshafts worldwide. Imbalance, one of the quality parameters of crankshafts, has a signicant impact on the life of the entire system. In the forging process, given the variations of material composition, temperature, etc., the imbalance dispersion increases compared to casting crankshafts. It is of particular importance to reduce this variation right from the conceptual design of the crankshaft. The authors have written a series of articles related to this problem from various perspectives

A. Albers et al. / Computers in Industry 60 (2009) 604612

605

using a range of tools [2,3]. The work continues, with a general description of the project as a step forward towards improving the strategy. 1.1. General goal (general objective of the proposed research) The goals of the study are, rst, to construct a strategy for developing engine crankshafts based on a framework of computeraided innovation; and second, to lay the foundations for a research topic about computer-aided innovation that uses:  CAD and CAE tools as an interface to the designer,  genetic algorithms as the main optimization/innovation mechanism,  TRIZ to extend the breadth of the evolutionary operators, and  splines for shape construction. All of this takes place in an automated environment allowing computers to push designers to reach the boundaries of creativity by helping them overcome design xation when optimizing. A range of disciplines converge to form the foundation of this research topic in computer-aided innovation. Without any particular order, a brief description of them follows. 1.2. Evolutionary design A relatively new area of development called evolutionary design [4] has become an object of intensive research. Peter Bentley describes evolutionary design as a process capable of generating designs by changing shapes and topologies. Consequently, an intricate design can arise through a slow, gradual, mindless improvement process. Evolutionary design mimics the way nature behaves by using evolutionary algorithms that change the forms and topologies of the design object. Biological creatures far exceed the products created by humans in terms of complexity and performance. This idea has created a brand in design referred to as biomimetic (emulating biology). Evolutionary design has its roots in computer science, design, and evolutionary biology. It extends and combines CAD and analysis software, and borrows ideas from natural evolution. The use of evolutionary computation to generate designs has taken place in many different elds since the late 1980s [5]. Designers have optimized selected parts of their designs using evolutionary computation [6]. Although the eld of evolutionary design is showing some impressive results, computers are not fully autonomous. People are required to work out which function the design should perform, and how a computer should be applied to the problem. Evolutionary design can present characteristics that add value to the product, even by chance. If the creations of design concepts evolve generating novel designs, e.g. novel product shapes that achieve higher performance, this can be interpreted as being creative. Therefore, it can be said that computers can present creative behaviour [7]. 1.3. Genetic algorithms Genetic algorithms (GAs) are adaptive, heuristic search algorithms (stochastic search techniques) based on the ideas of evolutionary natural selection and genetics [8]. They are used to nd approximate solutions to optimization problems. The basic concept of GAs is aimed at simulating processes in natural systems that are necessary for the mechanics of evolution, specically, those processes that follow the principles proposed by Charles Darwin: the law of the strongest or the survival of the ttest. They represent an intelligent exploitation of a random search within a dened search area to nd a solution. Genetic algorithms

are typically used in computer simulation, from which a population of abstract representations (called chromosomes) of possible solutions (called individuals) are applied to a design problem that evolves towards the best solutions. Traditionally, the solutions are represented in binary code as strings of 0s and 1s, but different codes are also possible. The evolution begins with a population of random individuals and produces results in generations. In each generation, the tness of the entire population is evaluated, many individuals are chosen stochastically from the current population (based on their tness), modied (mutated or recombined), and form a new population. This new population is used in the next iteration of the algorithm. Genetic algorithms have been studied extensively and have been applied in many elds of engineering. Many real-world problems involving a search for optimal parameters may be hard to solve with traditional methods, but when GAs are used, the solution is more easily found. However, due to their outstanding performance in optimization, GAs have been wrongly regarded as a tool only for optimization. Genetic algorithms also show impressive results compared to other search engines, and for this reason, the authors perceive them as a potential tool for creativity enhancement. 1.4. Design optimization systems The evolution of product development tools has been characterized by various trends, and the analysis of these trends offers useful hints for predicting next generation systems. The optimization of products and processes has been studied since the spread of computers as an aid for seeking optimal forms and shapes of product geometry. In the mechanical eld, geometric parameterization is used to dene the kind of changes described by design variables. Particularly, two kinds of structural optimization are frequently used: topology optimization and shape optimization. Topology optimization tries to achieve the maximum ratio between volume and some geometrical parameters. It determines the optimal material distribution within a given design space. For example, it takes out the elements under low stress in geometry by modifying the apparent material density, considered a design variable in a FEM model. A basic FE model is created and analyzed in a design area with given boundary conditions. Commonly, the aims are to maximize stiffness or maximize the natural frequency of a product. The constraints of the design are the xations, material volume, and maximum displacement allowed. The design variables are the material density of the elements, which are counted usually in hundreds of thousands; this means a huge number of design variables. The goal, given a predened design domain in the 2D/3D space with structural boundary conditions and load denitions, is to distribute a given mass, which is a given percentage of the initial mass in the domain determined in such a way that a global measure takes a minimum (maximum) value. This type of topology variation is outside the scope of this paper but is also being analyzed by the authors and will be presented in future publications. Shape optimization consists of changing the external borders of a mechanical component. The geometry of the product is dened in terms of surfaces and curve parameters that dene the outer boundary of the product and allow more freedom to manipulate. Here, too, the topology remains unchanged. The shape of the structure is modied by the node locations of a product modelled with the nite element method (FEM). The aims are to minimize the stress or the volume or maximize the natural frequency. Constraints to the design include xations and restrictions for displacement of component borders. The design variables of the product for geometric models are length, angle, and radii measurements; and for FE models, node coordinates.

606

A. Albers et al. / Computers in Industry 60 (2009) 604612

To further develop optimization systems new concepts must be added to the previous paradigms. A third kind of parameterization is inferred by taking the characteristics of the FEM-based types mentioned above. To obtain a similar behaviour (shape and topology) within a CAD model, the geometry of the product is described in terms of splines. The splining approach extends these features, allowing the introduction of innovative concepts [9]. Evolutionary design takes both CAD and CAE and extends the analysis and detailed design capabilities for which they have been used until now. Moreover, the processing capabilities of computers continue to increase, which makes the resolution of more complex FEM models possible. 1.5. Evolutionary design optimization with the help of splines The use of evolutionary computation to optimize existing designs by parameterization was the rst type of evolutionary design to be widely used. A great variety of different engineering designs have been successfully optimized, e.g. antennas [10], turbine blades [11] and aircraft geometries [12]. Though the methods used by the developers of such systems vary, one of these types of evolutionary design, the splined shape approach [13], has the potential to be classied as generative or creative. The splining approach for optimizing designs using GAs is relatively new. Practitioners of evolutionary optimization using splines usually start the process with an existing design, substitute the current construction with splines because the number of degrees of freedom can be increased at will by adding control points, and then parameterize the control points of the splines that embody those parts of the design they feel need improvement. The control points are encoded as genes, and the alleles (values) from which the parameters are described are evolved by an evolutionary search algorithm, e.g. genetic algorithms. In the eld of mechanical engineering, this approach has been applied to the design of automotive components such as cams [14]. The designs are often judged by making an interface of the system to simulation software or analysis software, which is used to obtain a tness measure for each design. These tools derive to an extended optimization method, achieved by representing shapes with splined curves [15]. This eliminates non-parametric dependency on FEM and allows a concept variation of forms. Automatic variations in 3D shapes are produced and concept variations evaluated by analysis software are added. 2. Case study: crankshaft design The case study presented here involves developing an engine crankshaft, with emphasis on its imbalance and dynamic behaviour (i.e. eigen values). The objective of a balance analysis of engine crankshafts is to develop their geometry so that a specied previous imbalance is obtained on two correction planes [16]. These two planes lie where the outermost counterweights are located, and so the imbalance could be corrected at the manufacturing shop by specic drilling on those counterweights. This paper reports on advances made in work done by the authors [2], where the imbalance formulas are explained. A brief recapitulation follows. In a pivoted crankshaft turning with constant angular velocity, the sum of the moments produced by its mass differentials times their locations is equal to the bending moment acting on the crankshaft. The bending moment is equal to the result of the moment caused by the centrifugal force due to the crankshaft mass at the center of gravity, and the moment caused by the imbalance masses on the correction planes. A design practice is to take the sum of moments around the left correction plane (into the counterweight close to the ange) to eliminate the moment caused

by the left imbalance mass and nd the imbalance on the right plane: mR r Ry mR r Rz 1=2mg r gy Ixy l 1=2mg r gz Ixz l (1) (2)

where l is the distance between the two correction planes, mg is the mass of the crankshaft; rgy and rgz are coordinates y and z of the center of gravity. In the same way, moments around the right correction plane are taken to nd the imbalance on the left plane: mL r Ly mL r Lz 1=2mg r gy Ixy l 1=2mg r gz Ixz l (3) (4)

The mass properties of interest (mg, rgy, rgz and the inertia products Ixy and Ixz) of the crankshaft model can be obtained from CAD software by orienting the geometry with respect to a system of coordinates with the x-axis towards the ange end of the rotation axis and the z-axis in the proper vertical orientation. If a target for imbalance is specied (TmLrL and TmRrR), the difference between the target and the current imbalance is a set of vectors to be minimized by making modications to the crankshaft geometry. The tness function selected as objective is then a summation of the differences between the current imbalance response of the individual conguration and the imbalance target specied as a design requirement (goal programming). The algorithm will then try to minimize the absolute deviations from the differences. The simplest form of this method may be formulated as follows:
k X min j f i x T i j subject to x 2 X i1

(5)

where Ti denotes the target (TmLrL and TmRrR) or goal set by the designer for the ith objective function fi(x) (mLrL and mRrR), and X represents the feasible region of geometry. The criterion, then, is to minimize the sum of the absolute values of the differences between the target values and the actually achieved values of imbalance on the two external counterweights. 2.1. Splining of the counterweights of a crankshaft Starting from a sensitivity analysis, the authors decided to substitute the current prole of the outermost counterweights of the crankshaft. From a CAD shape consisting of lines and arcs, the geometry of the counterweights was substituted with spline curves. The crankshaft consists of nine counterweights, and numbers 1, 2, 8 and 9 are modied. The Y coordinates of the spline control points are controlled by the genetic algorithm. The modications of the control points produce changes in the crankshaft geometry that result in a balance response. Splines allow smooth shape changes via the coordinates of their control points. This is not possible with arcs because they have less degree of freedom. These smooth shapes are benecial for the material uency during the manufacturing process. Fig. 1 shows how the splines substitute the original prole of the crankshaft. The spline is inside the tolerances of the original counterweights prole but will not be limited during the development of the shapes. The splining of the shapes and their control points, codied to be interpreted by genetic algorithms, form the basis for an evolutionary designed shape. Three main genetic operators act on the genes of the geometry: selection, crossover, and mutation.

A. Albers et al. / Computers in Industry 60 (2009) 604612

607

3. From evolutionary design to computer-aided innovation As explained in the introduction, the aim of this research program is to extend the detailed design capabilities of computers and to merge them with conceptual activities during the product development process. Within this context, the use of genetic algorithms is put forward as a mechanism for mediating conicts. They are able to reveal the conict that may exist between parameters, shown by the different tness functions. When a designer is solving a multi-objective optimization problem (also called multi-criteria optimization, multi-performance or vector optimization problem), he or she tries to nd a vector of decision variables that satises the constraints and optimizes the objective functions. These functions form a mathematical description of the performance parameters, usually in conict with each other. Hence, the term optimize means to nd a solution that will give the values of all the objective functions an acceptable trade off to the designer [17]. To deploy the concepts explained, the Pareto diagram is used (used mainly in multi-objective optimization processes). This diagram shows a boundary that divides the region of feasible solutions from the region where restrictions or physical constraints prevent the solutions from acquiring certain values. 3.1. Multi-objective optimization, conicts, and the ideal solution Solving optimization problems with multiple, often conicting objectives, is generally a very difcult goal. Although singleobjective optimization problems may have a unique optimal solution, as a rule, multi-objective problems present an innumerable set of solutions. When evaluated, the solutions produce vectors with components that represent trade-offs. TRIZ deals with the concept of the Ideal at the moment of resolving trade-offs [18], as will be explained. However, the concept of the ideal point has been taken up also in other studies, as described by Coello [19]. These studies explain that attempts have been made to develop a methodology emulating the criteria that drive a human being to adopt a certain solution named the ideal point [17]. However, there is a set of optimal solutions lying on a line that prevents the functions from reaching the ideal at the same time. This notion is represented in a Pareto diagram. Given the constraints in the solution space, the ideal point lies in the unfeasible region (Fig. 2). To solve this, a decision must be made by exploring the achievable limits of each particular attribute of importance (parameter) to nd an ideal alternative. Some authors propose

Fig. 1. Counterweights sketched using splines (dark lines).

Crossover allows the geometrical characteristics of selected splines (compared from a tness function) to be merged in pairs and their properties to be extended to following generations. The crossover has different levels of inuence, the genotype and the phenotype (geometric coordinates of the splines control points), but its impact is on the level of the shape of the product. Moreover, the concept can be extended to the whole structure of the product and even to the functional structure. 2.2. General description of the optimization/innovation loop During the optimization loop, the parametric CAD software is automatically manipulated by the genetic algorithms thanks to an interface programmed in JAVA language. The JAVA interface allows the CAD software to run continually with the crankshaft model loaded in the computer memory, so that every time an individual is generated the geometry automatically adapts to the set of parameters. In other words, it is not necessary to restart the software to regenerate the CAD model during each evaluation. The equations of imbalance are introduced into the CAD software and automatically provide the value of the tness function, i.e. the difference between the current and target imbalance. This value is automatically updated every time the geometry is modied because of the interface linking the CAD model to the genetic algorithms that control the optimization process. The specic requirements to be satised by the strategies follow:  Approach the target of imbalance of a V6 engine crankshaft, without affecting either its weight or its manufacturability.  Develop interface programming that allows for integration of the different software: CAD for modelling and geometric evaluations, CAE for simulation analysis and evaluation, and genetic algorithms for optimization and the search for alternatives.  Find the innovation principles that complement the genetic operators (mutation, crossover, etc.) to be able to generate a list of extended operators to apply to the geometry.  Obtain new design concepts for the shape of the counterweights that help the designer develop a better crankshaft, in terms of functionality, more rapidly than with the use of a manual approach. With this outline as the guiding approach, a conceptual framework for extending the optimization paradigm to an innovation strategy to be applied in a general crankshaft development method follows.

Fig. 2. Pareto diagram and the concept of ideal.

608 Table 1 Comparison of genetic and inventive principles. GA operators Crossover, simple

A. Albers et al. / Computers in Industry 60 (2009) 604612

TRIZ principles Segmentation, combination Asymmetry Merging Nesting Inversion Another dimension Homogeneity Discarding and recovering Spheroidality Feedback Copying

Representation Divide two genotypes and combine alternate parts Break symmetrical genotypes Join genotypes (similar or not) Place part of a genotype inside another Turn around a genotype Create genotypes from different parameters Make a genotype homogeneous Break down and rebuild genotypes Join the beginning to the end to form a circle Change a gen in a genotype Return ttest genotypes Take a copy of ttest genotypes

Crossover, nesting

Mutation Selection

performing a set of mono-objective optimization tasks to reveal conicts [20]. TRIZ proposes the use of the inventive principles. 3.2. TRIZ and cataclysmic operators Traditional TRIZ theory encourages the attainment of an Ideal Final Result and TRIZ offers tools for identifying the underlying technical and physical contradictions in a technological system. TRIZ general solutions (i.e. inventive/separation principles, standard solutions, etc.) are proposed to overcome the conict and let the product evolve according to the laws of technical evolution [21]. This is a natural convergence direction to merge evolutionary design (based on laws of biological evolution) with TRIZ (based on laws of technical evolution) within a computer framework suitable for computer-aided innovation. Other authors also apply the TRIZ methodology to structural product optimization [20]. Still others are using GAs to generate creative forms and shapes [22], but the main difference with them lies in the merging of TRIZ inventive principles with genetic algorithms. In other words, genetic algorithms controlling splined shapes are the core interface to apply innovation operators from TRIZ to the development of a product in a CAD&CAE environment. In this work, the aim is to extend the paradigm of multiobjective optimization to reach the ideal point, in this case, the Ideal Final Result. GAs can be enriched by taking advantage of the inventive principles and extending the operators beyond basic mutation and crossover. Additionally, it is possible to have new operators or agents capable of modifying the way the algorithms operate on the CAD geometry [23]. In this way, CAD systems can develop new congurations and alternative modications to the geometry, and supply the designer with a set of alternative modications, dened automatically based on the selected principles, which may themselves be applied based on the concept of cataclysmic mutations [24]. Mutations following a pattern similar to a cataclysm are now being studied in the area of evolutionary algorithms as tools for nding innovation [25,26]. The extension of the solution can reach a level of detail that leads to two possibilities: either automatic development of solutions by the genetic algorithm with or without the use of cataclysmic operators or, on a more detailed level inspired by these suggestions, the designer can select the most suitable solution and implement it. Inventive principles suggest a series of recommendations to change the direction in which solutions are sought. The recommendations are based on the fact that they have provided solutions in previous analogous technical problems. These recommendations can be regarded as a knowledge database, which can be used to feed the cataclysmic transformation of genotypes in the course of evolution for optimization, allowing them to override the barriers of contradictions or constraints. The TRIZ inventive principles with a geometric interpretation can be added to form the extended cataclysmic operators. The level of

impact from the different operators can vary from a slow and steady accumulation of changes (the way an optimization algorithm normally performs) to a sudden disturbance in the nature of the system (a cataclysm). The most important effect is the creation of a leap in the phase transition. See Table 1. Some examples of how the agents may be applied to the case study are shown in Fig. 3. More suggestions can be enriched by means of guidelines provided by the inventive principles that can be associated with the genetic operators. As result, the algorithm should be capable of applying agents depending on the conict that is being faced. 4. Implementation using an example of a crankshaft (follow up) 4.1. Strategy and approach No. 1: multi-objective optimization of the imbalance The rst strategy involves optimizing the crankshafts imbalance as a multi-objective problem, the imbalance on each side of the crankshaft being one of the two objective functions. The

Fig. 3. Graphical representation of inventive agents (dark lines).

A. Albers et al. / Computers in Industry 60 (2009) 604612

609

Fig. 4. Integration of DAKOTA with CAD Software using JAVA.

proposed sequence of steps is given in Fig. 4, which shows parallel execution of the GA and the CAD software, running automatically. The GA selected for our case study is from DAKOTA software. DAKOTA is a parameter solver developed at the Sandia Laboratories [27] with the original goal of providing a common set of optimization tools for engineers who need to solve structural and design problems, including genetic algorithms. The parametric CAD software used in the project includes an interface programmed in JAVA, and some of the subroutines have been adapted to let this tness result be read and sent back to DAKOTA, where it is assigned to the corresponding individual. After the whole generation has been evaluated, tournament selection plus other genetic operators produce the next generation. We chose a multi-objective genetic algorithm (MOGA) from the DAKOTA toolkit to generate the individuals. The population consisted of 50 individuals randomly initialized; crossover rate was 0.8, mutation rate was 0.1, and the number of evaluations was stopped at 1000. The total number of independent variables (the coordinates of the control points of the splines) was 24, and there were eight control points in each of the splined counterweights: CW1, CW2, CW8 and CW9. The range of variation was set at 10 mm, which allowed a wide movement of the control points.

Fig. 5 shows a graph from the last generation with the value of the rst function (imbalance of CW1) on the x-axis and the value of the second function (imbalance of CW9) on the y-axis. The results from the rst attempts show that the imbalances from both sides of the crankshaft are in conict with each other, according to the perceived distribution in the diagram. Nevertheless, some points show an unusual movement towards the value of zero, which in our case is regarded as the ideal point. It should be noted that the graph does not show a dened frontier, which would have been expected as the balance reaches the target. As the target balance could not be attained inside the design constraints dened for the geometry of the counterweights, some trials were performed to allow the spline to override the geometry constraints. However, although improved results were obtained, the convergence was not yet satisfying. Results from previous attempts [2], where only two instead of four out of the nine counterweights had been splined, had not shown this behaviour. To give a visual impression of the way the algorithm is performing, some of the counterweights are shown in Fig. 6. The shapes present some notches that are not suitable for forging, but the direction of the solution is cataclysmic. An open-minded designer should be able to recognize that the paradigm is challenged and a new concept can be derived; in other words, a break with the notion that the counterweights have to be arcshaped and allow splined but smooth proles. The intention of these systems, as mentioned at the beginning of the paper, is to present challenging alternatives to the designer. Finally, these proposals are solution triggers that inspire him/her, but they do not substitute his/her role in selecting the most suitable solutions and implementing them properly. 4.2. Strategy and approach No. 2: integration of eigenvalue analysis as an additional objective function This section extends the work described in the previous section, where the objective functions consisted of the imbalance on both sides of the crankshaft. An evaluation of the eigenvalues of the crankshaft is added. This evaluation is performed on every geometrical conguration of the CAD model by external CAE software. This two-objective optimization problem has a nontrivial geometrical solution owing to the different nature of the objective functions, one being in the frequency domain (eigenvalues) and the other in the geometric domain (Imbalance). Further

Fig. 5. Distribution of imbalance on each side of crankshaft.

610

A. Albers et al. / Computers in Industry 60 (2009) 604612

Fig. 6. Representations of the rst attempt with the counterweight (dark line).

JAVA programming was performed to further link the CAE software to the strategy. The dynamic optimization of a crankshaft can be understood as an optimization problem with two distinct and, perhaps, conicting objectives. It also involves the unpleasant feature of crankshaft eigenvalues that cannot be explicitly written, but can only be calculated by simulation of the system model of the CAD geometry. In addition, constraints in the parameters of the design should be taken into consideration. The dynamic response of a crankshaft can be predicted on the basis of a nite element analysis of its structure. Even when calculating the eigenvectors for large structures is computationally expensive, the simplication of

the CAD model allows evaluations in an automated evolution process. The process is illustrated in Fig. 7. After an individual is assigned to the CAD geometry, an additional routine in the interface meshes the CAD geometry and exports it to an input le to be read by a FEM solver. The strategy allows this process to be automatic. Once the FEA result (the eigenvalue) is calculated, it is attached to the le containing the result from the imbalance and, together, is sent back to the GA to be attached to the corresponding individual and to enable DAKOTA to evolve the generation. The last generation for optimizing the multi-objective system is shown in Fig. 8 with the two objectives (eigenvalue and imbalance tness see previous section) in each axis. It can be observed in the graph that the direction of improvement of the eigenvalue is upwards while imbalance tness is towards the zero target. The conict of objectives can be appreciated, as in the previous section. In the graph, it can also be observed that the range of the balance tness is very broad. This is due to the very sensitive inuence of the counterweights on the geometry. On the other hand, the rst mode variation is not changing very much. The perception of the inuence of the geometry on the dynamic response of the crankshaft is not as signicant as the balance at the level of the system involved in the analysis. Nevertheless, the direction of improvement moves to a concentrated area in the upper-left corner of the graph, though the balance is still not reaching the target. We chose an individual distinguished by its position to the left as being the best (circled) and extracted the proles of its counterweights (see Fig. 9). From the images, it can be noted that from CW1 and CW2 the shapes are less wide and better shaped than in the rst approach. However, CW8 and CW9 still need improvement. This prole needs to be further smoothed to allow better material ow during the forging process. Even when the design of the counterweights is not totally nished, a designer should be able to recognize that the paradigm is challenged and a new concept can be derived; in other words, dropping the notion that the counterweights have to be arc-shaped and allowing splined but smooth proles. The intention of these systems is not only to create a totally automatic, optimized solution, but also to provide the designer with challenging alternatives. In reality, these proposals are solution triggers that inspire him/her, but they do not substitute his/her role in selecting the most suitable solutions and implementing them properly.

Fig. 7. Integration of DAKOTA with CAD and CAE software using JAVA.

A. Albers et al. / Computers in Industry 60 (2009) 604612

611

 Subsequent activities include the denition of additional tness functions not only in CAD but also in CAE simulation (forging simulation) for the control of strange shapes. Objective functions and restrictions will be added with the use of forging simulation and stress analysis during variations in geometry. The aim is to integrate different systems that run totally or partially automatically. 5. Conclusions A general strategy for developing the geometry of a crankshaft using CAI has been presented. Initially there was a description of the conceptual framework for its creation, including evolutionary design, genetic algorithms and TRIZ. Then an explanation of current optimization systems and of how different techniques, e.g. Pareto diagrams, can help to identify conicts was given. The aim is to resolve these conicts by innovation agents. Further development of the optimization systems (parametric and topology) can only be achieved by integrating them with innovation methods. The resulting systems have a parametric shape and an innovative topological conguration. Some features need to be added to the system to be able to work outside of the paradigm, and reach solutions that have not been considered before. The use and combination of optimization tools and innovation capabilities is intended to provide a means for automatically varying the shapes, forms and topologies derived from the evaluation made by CAE systems. Genetic algorithms are viewed as a mechanism for embedding the innovative principles in a CAD interface. For example, the parametric optimization capabilities of CAD systems can be extended from simple size variations to enhanced performance similar to what can be achieved by FEM, but without losing the parametric functionality. This is a step closer to building a computer-aided innovation environment that goes more deeply into the evolution of technical systems, as an analogy to biological evolution. The basics of a new concept of engineering design software and computer-aided innovation (CAI) are presented and the impact on global design performance and limits exploration is analyzed. In particular, the ability to introduce automatic changes in shapes and topologies in a 3D-CAD environment is linked to the concept of CAI, allowing for an exploration of a broader eld for possible solutions to a design problem. Experience and judgment can lead to good design. When evaluating a tness function, genetic algorithms rely only on the latter of these two characteristics (judgment), based on evaluation and comparison against certain criteria. The experience characteristic derived from the substantial knowledge of designers can be added to genetic algorithms by incorporating into the process inventive principles as cataclysmic genetic operators. Acknowledgment The authors acknowledge the support received from Tecnologico de Monterrey through Grant No. CAT043 to carry out the research reported in this paper. References
[1] Z.P. Mourelatos, A crankshaft system model for structural dynamic analysis of internal combustion engines, Computers & Structures 79 (2001) 20092027. [2] H. Aguayo-Tellez, N. Leon-Rovira, Computer aided innovation of crankshafts using genetic algorithms, in: Knowledge Enterprise: Intelligent Strategies in Product Design, Manufacturing and Management, Springer Boston, USA, 2006, pp. 471 476. [3] H. Aguayo-Tellez, N. Leon-Rovira, A. Albers, T. Maier, Comparison of strategies for the optimization/innovation of crankshaft balance, in: Trends in Computer Aided Innovation, Springer, USA, 2007, pp. 201210. [4] P. Bentley, Evolutionary Design by Computers, Morgan Kaufmann, USA, 1999.

Fig. 8. Distribution between imbalance and evaluation of natural frequency.

Fig. 9. Shapes of crankshaft after moga evaluation (new prole in dark line). (a) Best shape CW1; (b) counterweight 2; (c) counterweight 8; (d) counterweight 9.

Use of the JAVA interface allowed the genetic algorithm to be integrated into the CAD software, as described in the rst part of this paper, and the further integration with the CAE software, as described in the second part. The development of a Pareto frontier from the rst analysis and a Pareto with the additional eigenvalue objective were possible. The following important conclusions were extracted:  It is necessary to prevent development of sharp edges in the shape of the counterweight by adding extra control of the curvature and minimal radii.  Simulation of the forging process is a required next step for denition of the relationship between good shape-curvature and manufacturability. This is very important when a design proposed outside the initial shape restrictions needs to be justied in order not to affect formability.

612

A. Albers et al. / Computers in Industry 60 (2009) 604612 Albert Albers is a professor, the director of Product Development Institute at Karsruhe University (TH), Germany. His research work is focused on product development and design, innovation processes and drive trains for vehicles and machinery. The research in the eld of product development focuses on strategies, methods and tools to support the development process. A particular focus here are the integrated product development and computational shape optimization of highly stressed machine elements and complex machine systems. He is member of the Scientic Society of Machine Elements. Member of the Advisory Board of the EKV Karlsruhe. Board member of the University of Karlsruhe (TH). Noel Leon-Rovira is a full professor in the Center for Innovation in Design & Technology at Technological Institute of Monterrey, Mexico. His research work is oriented to Computer Aided Innovation, product concept development, Eco-Innovation, CAD&CAE and TRIZ. He has a BEng & MS degree in mechanical engineering and a PhD degree in Machine Design from Dresden Technical University, Germany. He was founder and Director of the Center for Studies of CAD&CAM at Higher Technological Institute of Holguin, Cuba. He has several years of engineering experience working in projects with industry. He is also a board member of the Working Group of Computer Aided Innovation from IFIP. Humberto Aguayo is a PhD candidate of Mechanical Engineering at Technological Institute of Monterrey, Mexico. He received his BS degree in Mechanical and Electrical Engineering from San Luis Potosi University, Mexico, MS degree in Manufacturing Engineering from Technological Institute of Monterrey, Mexico. He has several years of product design & development experience in industry. His research interests are in product development, optimization of design and manufacturing, and TRIZ in product/engineering design. Thomas Maier is a PhD candidate at the Faculty of Mechanical Engineer, Karsruhe University (TH), Germany. He is head of the research group of CAE/ Optimization at Product Development Institute, Karsruhe University (TH), Germany. He received his Diplom in Mechanical Engineer at the Faculty of Mechanical Engineer, Karsruhe University (TH), Germany. His research and teaching interests are related to model updating of dynamically loaded components and systems.

[5] C.A. Coello Coello, A comprehensive survey of evolutionary-based multi-objective optimization techniques, Knowledge and Information Systems 1 (1999) 129156. [6] D.A.V. Veldhuizen, G.B. Lamont, Multiobjective evolutionary algorithms: Analyzing the state-of-the-art, Evolutionary Computation 8 (2000) 125147. [7] J.R. Koza, M.A. Keane, M.J. Streeter, T.P. Adams, L.W. Jones, Invention and creativity in automated design by means of genetic programming, Articial Intelligence for Engineering Design Analysis and Manufacturing 18 (2004) 245269. [8] D.E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning, Addison-Wesley Longman Publishing Co, USA, 1989. [9] N. Leon-Rovira, J.M. Cueva de Leon, J. Gutierrez, D. Silva, Automatic Shape Variations in 3d CAD Environments, in: 1st IFIP-TC5 Working Conference on Computer Aided Innovation, Germany, (2005), pp. 200210. [10] S. Williams, Unnatural selectionmachines using genetic algorithms are better than humans at designing other machines, Technology Review: MITs Magazine of Innovation 108 (2005) 54. [11] M. Olhofer, Y. Jin, B. Sendhoff, Adaptive encoding for aerodynamic shape optimization using evolution strategies, in: Congress on Evolutionary Computation, Seoul, (2001), pp. 576583. [12] S. Obayashi, T. Tsukahara, T. Nakamura, Multiobjective genetic algorithm applied to aerodynamic design of cascade airfoils, IEEE Transactions on Industrial Electronics 47 (2000) 211216. [13] N. Leon-Rovira, J. Cueva, C. Villarreal, S. Hutron, G. Campero, in: N. Leon-Rovira (Ed.), IFIP International Federation for Information Processing, Volume 250, Trends in Computer Aided Innovation, Springer, Boston, 2007, pp. 179188. [14] J. Lampinen, Cam shape optimisation by genetic algorithm, Computer-Aided Design 35 (2003) 727737. [15] N. Leon-Rovira, D. Silva, J.M. Cueva de Leon, J. Gutierrez, Automatic Changes in Topology of Parts in 3D CAD systems, in: 1st IFIP TC5 Working Conference on Computer Aided Innovation, Ulm, Germany, 2005. [16] S. Rao, Mechanical Vibrations, Addison-Wesley, USA, 1990. [17] C.A. Coello Coello, An Empirical Study of Evolutionary Techniques for MultiObjective Optimization in Engineering Design, Tulane University, USA, 1996. [18] G. Altshuller, Creativity as an Exact Science: The Theory of the Solution of Inventive Problems, Gordon and Breach, New York, 1984. [19] C.A. Coello Coello, An updated survey of GA-based multi-objective optimization techniques, ACM Computing Surveys 32 (2000) 109143. [20] U. Cugini, G. Cascini, M. Ugolotti, in: N. Leon-Rovira (Ed.), IFIP International Federation for Information Processing, Volume 250, Trends in Computer Aided Innovation, Springer, Boston, 2007, pp. 189199. [21] G. Altshuller, L. Shulyak, 40 Principles: Triz Keys to Innovation, Technical Innovation Center, Inc., USA, 1997. [22] H. Liu, M. Tang, J.H. Frazer, Supporting creative design in a visual evolutionary computing environment, Advances in Engineering Software 35 (May) (2004) 261271. [23] N. Leon-Rovira, J. gutierrez, O. Martinez, C. Castillo, Optimization vs. innovation in a CAE environment, in: IFIP World Computer Congress 2004, France, 2004. [24] Y. Zhang, M. Liao, Z. Ren, Cataclysmic Genetic Algorithm for Reactive Power Dispatch Comparing with an Interior-Point Algorithm, M.H. Hamza, USA, 2005. [25] O. Dubrovsky, G. Levitin, M. Penn, A genetic algorithm with a compact solution encoding for the container ship stowage problem, Journal of Heuristics 8 (2002) 585599. [26] G. Levitin, J. Rubinovitz, B. Shnits, A genetic algorithm for robotic assembly line balancing, European Journal of Operational Research 168 (2006 Feb) 811825. [27] M. Eldred, A. Giunta, B.V. Bloemen, S. Wojtkiewicz, W. Hart, M. Alleva, DAKOTA, A Multilevel Parallel Object-Oriented Framework for Design Optimization, Parameter Estimation, Uncertainty Quantication, and Sensitivity Analysis. Reference Manual, Sandia Laboratories, USA, 2002.