R.L. Wood

Genetic algorithms in computer-aided design

Abstract This paper describes progress in research into the development of a computer-aided design (CAD) tool that aids designers in generating the form of a product by the use of evolutionary techniques. Genetic algorithm (GA) software has been developed and combined with a commercial CAD solid modelling system. The system creates objects that initially have the appearance of being random in form, but which can be subjected to a user-directed selective breeding programme which is also guided by pre-set internal, or environmental, factors. User scoring of each object, or an objective function, determines which objects are considered to be the ‘fittest’, and thus likely to become parents of the next generation. Through the co-operation of the user and the pre-set environmental factors, the forms on the screen progressively become more than an abstract collection of geometric primitives. It is believed that this can provide useful inspiration with regard to the aesthetics and functional characteristics of products, and the potential exists for this approach to be the basis of a new design methodology. Early work demonstrated that the software had the ability to evolve interesting shapes in line with a user’s particular criteria for rating objects. However, it was also obvious that the objects generated needed enhancement to convincingly represent some of the geometric complexities of real products. Using the CAD software’s blend function within the evolutionary process has provided that complexity, producing excellent results and greatly widening the field of application. In addition to simply creating secondary geometry through the smoothening of sharp edges or creation of curved fillets between adjoining solids, more significant and complex primary geometric forms have been generated by allowing relatively large blend radii. The current challenge is to combine the existing ability to predictably evolve simple geometric shapes, with the added complexity arising from the use of blends, to make the concept genuinely useful. The outcome of genetic manipulation needs to be predictable, to the extent that desirable features from objects are reproduced in subsequent generations. The key to this is the way the genetic shape defining data is stored and processed, especially the way the blend instructions are integrated into the existing genetic structure, and this is the major focus of continuing research.

Genetic algorithm behaviour in the solution of an inverse thermal field problem

Discusses the performance of a simple genetic algorithm (GA) applied to a one‐dimensional inverse thermal field problem. Builds on these results by considering changes in GA performance that result from the introduction of non‐complementary crossover, stochastic remainder sampling and a combination of the two. Shows that, in comparison to the simple GA, non‐complementary cross‐over provides more rapid convergence, while stochastic remainder sampling without replacement has the opposite effect. However, when both strategies are combined, they provide considerably better performance with greater diversity within the population.

Complexity of Traffic Interactions: Improving Behavioural Intelligence in Driving Simulation Scenarios

This paper introduces modelling concepts and techniques for improving behavioural intelligence and realism in driving simulation scenarios. Neural Driver Agents were developed to learn and successfully replicate human lane changing behaviour based on data collected from the TRL car simulator.

Shape modification using genetic algorithms

Abstract This paper describes research into the use of a genetic algorithm within a computer aided design tool. The genetic algorithm allows the evolution of shape aspects of designs against a subjective rating carried out by the designer. Shapes considered by the designer to contain pleasing characteristics are used to generate subsequent generations of different, but related, shapes. A genetic data structure made up of chromosomes is briefly described along with the methods used to evolve new generations of objects using reproduction methods involving crossover of chromosomes and the chance of mutation. Examples are given where the technique is used to generate ‘animal sculptures’, product concepts and a variety of concepts for seating.

An inverse thermal field problem based on noisy measurements: comparison of a genetic algorithm and the sequential function specification method

Draws a comparison between the use of a genetic algorithm and the sequential function specification method for the solution of a one‐dimensional linear inverse thermal field problem, based on the use of noisy measurements. In solving this problem aims to estimate the value of a single constant convective heat transfer coefficient. Documents the findings that both approaches can provide estimates within 1 per cent of the target solution and that the sensitivity and robustness of each approach to measurement location, time step size and measurement errors are markedly different.

Video Based Real-Time Pedestrian Detection on Zebra Cross

This paper proposed a novel algorithm for real-time pedestrian detection based on the road user trajectory recognition. Frame series captured from the video stream are used to compose the object trajectory presenting basic features including width, speed, and length. The triangular fuzzy numbers are used for the feature description and decision reference. The multi-variable decision tree is applied for the decision-making. The proposed framework and algorithm are implemented at the junction of Eshan, Dongfang Road, Shanghai for pedestrian alerting system in red light time. The accuracy over 93% is achieved with single decision iteration time less than 100ms.

A dynamic curation method for manufacturing-related knowledge

A method for the dynamic curation of manufacturing-related knowledge is proposed, based on the impact of successive paradigm introduction on the network structure within manufacturing companies. This draws together manufacturing system structure in terms of interacting component network types, the nature and consequences of knowledge silos and the underpinning dichotomous influence of language. The need and opportunities for an objective- rather than subjective paradigm-based view of manufacturing are identified, leading to a curation process in which paradigms and other knowledge specialisms are different viewpoints based on particular models of manufacturing processes and resources. The consequences of this are explored in terms of knowledge silo reduction, improved communication within component social- and information networks, increased operational resilience and better informed decision-making for future business.

Search Space Scaling in Genetic Algorithm-Based Inverse Analyses

Abstract Inverse problems arise at all levels of manufacturing organization through the need to estimate system parameters and optimize performance. In modelling the physics of various manufacturing processes, inverse analysis is primarily used to estimate boundary conditions at the part/machine interface. The quality of such estimates is influenced by the measurement process, the physics model, and the inverse analysis mechanism. Of the available mechanisms, genetic algorithms (GAs) are unusual because they do not rely on the use of sensitivity coefficients and they are easily integrated with direct models of process physics. Primary factors that influence GA performance are the topology of the search space and the way that this is sampled by the GA. Search space topology is dictated by comparison of the measured and calculated effects of the sought boundary conditions. Search space sampling is controlled by chosen chromosome structure and previous evolutionary progress. A method of dynamic search space scaling is discussed here, in which contractions of search space dimensions follow evolutionary progress and are triggered by changing population diversity within the GA. In comparison to evolution in a fixed search space, it is shown that dynamic search space scaling can be efficient in producing highly accurate estimates.

The Design of Experiments for Inverse Thermal Analysis

Abstract A procedure is demonstrated for the design of experiments within inverse thermal analyses, such as those used to estimate surface boundary conditions in quenching. The procedure involves computational modelling to support design decisions concerning thermocouple locations, sampling rate, and post-processing of measured data. The procedure also provides a framework into which detailed models of thermocouple operation can be inserted. The method is demonstrated via the estimation of a non-linear convective boundary condition in a quenching process in which the measurement process is modelled to allow the effects of measurement noise and other parameters to be assessed. Having carried out the inverse analysis using a combined genetic algorithm - finite element model, estimated boundary conditions are in good agreement with those specified in the measurement model.