Nigel Ball

Genetic algorithm representations for laminate layups

Abstract The design of laminate layups using a genetic algorithm (GA) search-based function optimizer has been investigated using both a generate-and-test evaluation function and the layup-synthesis rule-base from a knowledge based design system. (The small rule-base from the laminate design was translated into a procedural evaluation function which could be called by the GA.) A simple coding of plies to genes is shown to be applicable in each case, but only when coupled with a penalty function which constrains genetic search to permissible layup candidates. Two experiments have been performed: a comparison of the GA search algorithm with a random/greedy search algorithm and a comparison of the GA with a rule based design system. Comparative tests with the rule based laminate design system are presented for four different load types. These demonstrate the robustness and accuracy of the system for the design of small scale asymmetric laminates.

Machine learning in configuration design

New methods of configuration analysis have recently emerged that are based on development trends characteristic of many technical systems. It has been found that though the development of any system aims to increase a combination of the performance, reliability and economy, actual design changes are frequently kept to a minimum to reduce the risk of failure. However, a strategy of risk reduction commits the designer to an existing configuration and an approved set of components and materials. Therefore, it is important to analyze the configurations, components, and materials of past designs so that good aspects may be reused and poor ones changed. A good configuration produces the required performance and reliability with maximum economy. These three evaluation criteria form the core of a configuration optimization tool called KATE, where known configurations are optimized producing a set of ranked trial solutions. The authors suggest that this solution set contains valuable design knowledge that can be reused. This paper briefly introduces a generic method of configuration evaluation and then describes the use of a self-organizing neural network, the Kohonen Feature Map, to analyze solution sets by performing an initial data reduction step, producing archetype solutions, and supporting qualitative clustering.

Design of laminate composite layups using genetic algorithms

Fibre-reinforced laminate composites have attractively high stiffnesses and strengths and low densities. However, designing with laminate composites is more difficult than designing with metals because (a) it involves the design of the material itself, and its manufacturing route, at the same time, (b) laminates are highly anisotropic, and (c) they have complex failure modes. The failure modes and anisotropy combine to make design details unintuitively important and small detailed-design oversights have been responsible for most failures in composite structures. Design is ‘the process of converting an idea into information from which a product can be made’. Thus the central role of information processing in any design activity implies that software should be able to help. Here we show three different ways in which laminate stacking sequences can be designed.

Managing Conceptual Design Objects

The set of conceptual entities generated during the early stages of a design form an integral part of a product’s final definition (whether accepted or rejected) because they provide an implicit record of the rationale that led to the final configuration. Many of these alternative concepts are created from a non-spatial perspective and cannot be adequately captured using conventional CAD systems which still require a geometric framework. This paper describes a product data model that supports the capture and characterization of alternative concepts from both a product and process viewpoint by using non-geometric objects to structure a multilayered project entity. A simple management tool that uses this model is presented and demonstrated in the context of an undergraduate design project to build a semiautonomous guided vehicle.

Neural networks for power systems alarm handling

A multi-layered architecture of self-organizing neural networks is being developed as part of an intelligent alarm processor to analyse a stream of power grid fault messages and provide a suggested diagnosis of the fault location. Feedback concerning the accuracy of the diagnosis is provided by an object-oriented grid simulator which acts as an external supervisor to the learning system. The utilization of artificial neural networks within this environment should result in a powerful generic alarm processor which will not require extensive training by a human expert to produce accurate results.

A Framework for Design Object Evolution

This paper presents current work on Product Data Modelling in the Cambridge Engineering Design Centre (EDC) that offers a novel approach to circumventing some of the known problems with the Object Oriented paradigm in the design domain. A data driven approach to object based design is described that allows the designer to build class prototypes during the design process and capture these prototypes onto a catalogue. Catalogue class entries can be reused in an evolving product configuration through a process of selection and specialization with new characteristics. New classes generated during the design can be instantiated as part of the developing product design object and also written back onto the catalog as new prototypes. Catalogues implicitly cluster design objects into abstraction hierarchies that are maintained and developed by the designer rather than a computer programmer. The paper illustrates the technique with an industrial case study and discusses how the approach is being used to develop applications within and without the EDC.

Self-organising neural networks for adaptive control

Self-organizing neural networks have been implemented in a wide range of application areas such as speech processing, image processing, optimization and robotics. Recent variations to the basic model proposed by the authors enable it to order state space using a subset of the input vector and to apply a local adaptation procedure that does not rely on a predefined test duration limit. Both these variations have been incorporated into a new feature map architecture that forms an integral part of an Hybrid Learning System (HLS) based on a genetic-based classifier system. Problems are represented within HLS as objects characterized by environmental features. Objects controlled by the system have preset targets set against a subset of their features. The systems objective is to achieve these targets by evolving a behavioural repertoire that efficiently explores and exploits the problem environment. Feature maps encode two types of knowledge within HLS — long-term memory traces of useful regularities within the environment and the classifier performance data calibrated against an objects feature states and targets. Self-organization of these networks constitutes non-genetic-based (experience-driven) learning within HLS.This paper presents a description of the HLS architecture and an analysis of the modified feature map implementing associative memory. Initial results are presented that demonstrate the behaviour of the system on a simple control task.

Towards Mechanical Design Object Reuse

The Common Product Data Model (CPDM) developed at the Cambridge Engineering Design Centre (EDC) captures structured descriptions of both the designed artefact and the process of its development, from multiple, interlinked points of view. The links allow each fragment to serve as context for others. We present a search mechanism for the physical, shape-feature aspect of this representation, and explain how to extend it to other aspects. It does not require component classification or a controlled naming scheme but proceeds directly from the product description. It retrieves similar objects as well as exact matches. This allows design objects to be reused in new situations, as required in the selection of standard components from catalogues. It also allows us to investigate the meaning of the labels (classes, names with functional content…) which may occur in CPDM descriptions, which can then be fed back to improve retrieval.

A NOVEL FEATURE MAP ARCHITECTURE FOR THE REPRESENTATION OF CLASSIFIER CONDITION SETS

A new feature map architecture has been developed as part of a Hybrid Learning System (HLS) based on a genetic-based classifier system. Feature maps encode two types of knowledge within HLS - long term memory traces of useful regularities within the environment and classifier performance data calibrated against an objects feature states and targets. Self-organization of these maps constitutes non genetic-Based (experience-driven) learning within HLS. This paper describes the modified feature map and presents results from a simple control task demonstrating how the modified maps calibrate system behaviour under a variety of parameter settings.