Presentation Outline

Design in General
State of the art
Guidelines for the current research work
The proposed methodology
An example case
Conclusions
Future Work

Design Issues

design representation schemes differ from problem to problem
variety of design knowledge formalisms
manipulation of both ‘hard’ and ‘soft’ design knowledge blocks
multi-dimensional solution spaces
discontinuity of design space
satisfactory but sub-optimal solutions

State of the art

Design representation and modeling (Deng et al., Otto et al., Wang et al., Lottaz et al., Suh et al.)

Design Structure Matrices (Browning, Yasine, Chen, Saridakis et al.)

Genetic Algorithms (Goldberg)

Fuzzy Logic, Artificial Neural Networks (Kosko)

Soft-Computing in Design (Rao et al., Vico et al., Tian-Li et al., Goldberg et al., Sasaki et al., Hsiao et al.)

Directives for The Current Research Work

Elaboration of systematic methodology for domain-independent parametric design.
Manipulation of available design knowledge of different formalisms.
Extraction of a globally optimal solution.
Simplification of the design problem under consideration.

Design Entities and Associative Relationships (1)

Every design problem may be expressed through a set of Design Parameters (DPs).

DPs may refer to physical attributes of the design object, to functions that the designed system should perform, or to performance metrics.

DPs may be either quantitative or qualitative.

Associative relationships among DPs may be expressed in terms of computational formulas, empirical rules, selection matrices, experiment values etc.

Design Entities and Associative Relationships (2)

Fixed Primary DPs: inputs for the design problem with their values remaining invariable during deign cycles.
Variable Primary DPs: their values change independently during design process.
Dependent DPs: their values are fully defined by their children DPs and the associative relationships.
Variable Dependent DPs: dependent DPs that are partially defined by their children DPs but their final value is user-defined.

Design Parameter Hierarchical Tree

Architecture of The Current Approach

Deployment of the proposed methodology in 3 steps:

1. Design problem statement through DPs,

associative relations and hierarchical trees.

2. Solution search with Genetic Algorithm

using certain optimization criterion.

3. Extraction of a simplified fuzzy structure

with neuro-fuzzy adaptation using elite

solutions from the previous GA deployment.

Solution Search with GAs

Genetic Algorithm (GA) uses a stochastic variation of values for both numerical and linguistic variable DPs.
GA may be deployed together with a numerical optimization method in order to extract globally optimal solutions.

Aspects to be considered before genetic optimization:

A set of DPs should be characterized as critical for the design outcome and identified as Performance Variables (PVs).
Definition of the optimization criterion.
Adjustment of the attributes related with a GA (population size, individual length etc.).

Simplification of the Design Problem

Initial design problem: DPs and their associative relationships

Neuro-Fuzzy adaptation for design parameter

Neuro-fuzzy adaptation for Performance Variables

Smplified DP tree for oscillating conveyor

Simplified flat structure of the design problem with fuzzy relationships

Results for the Example Case

GA run for initial problem

Solution search with Genetic Algorithm for the initial problem

Showcase of the fuzzy associative relationship among the Performance Variable "convspeed" and the three variable primary DPs

Solution search with Genetic Algorithm for the simplified problem

The genetic optimization was deployed in the example case of oscillating conveyors for 30 generations.

After 10 runs of GA, the average time needed for extracting the optimum solution was 94.2655 seconds.

The elite solutions during the genetic optimization were recorded in order to build a simplified flat fuzzy structure.

The re-design using the simplified structure, after 10 runs of GA of 30 generations each, resulted to a mean time for optimum solution extraction equal to 21.6540 seconds.

Conclusions

Different types of the design associative relationships can be modeled.

It is applicable to all parametric design problems in a formal and systematic, step-wise process.

It copes with both quantitative and qualitative design knowledge.

The design problem is successfully approximated with simpler fuzzy structures.

Design objectives may be integrated with a unified optimization criterion.

Future Work

Utilization of most efficient optimization criteria.
Automatic adjustment for choices concerning genetic algorithm’s features.
Retrieval and adaptation of successful past design cases.
Evaluation against large and complex design problems.

	Deployment of the proposed methodology in 3 steps:
	1. Design problem statement through DPs, associative relations and hierarchical trees.
	2. Solution search with Genetic Algorithm using certain optimization criterion.
	3. Extraction of a simplified fuzzy structure with neuro-fuzzy adaptation using elite solutions from the previous GA deployment.

Presentation Outline

Design in General

State of the art

Guidelines for the current research work

The proposed methodology

An example case

Conclusions

Future Work

Design Issues

design representation schemes differ from problem to problem

variety of design knowledge formalisms

manipulation of both ‘hard’ and ‘soft’ design knowledge blocks

multi-dimensional solution spaces

discontinuity of design space

satisfactory but sub-optimal solutions

State of the art

Design representation and modeling (Deng et al., Otto et al., Wang et al., Lottaz et al., Suh et al.)

Design Structure Matrices (Browning, Yasine, Chen, Saridakis et al.)

Genetic Algorithms (Goldberg)

Fuzzy Logic, Artificial Neural Networks (Kosko)

Soft-Computing in Design (Rao et al., Vico et al., Tian-Li et al., Goldberg et al., Sasaki et al., Hsiao et al.)

Directives for The Current Research Work

Elaboration of systematic methodology for domain-independent parametric design.

Manipulation of available design knowledge of different formalisms.

Extraction of a globally optimal solution.

Simplification of the design problem under consideration.

Design Entities and Associative Relationships (1)

Every design problem may be expressed through a set of Design Parameters (DPs).

DPs may refer to physical attributes of the design object, to functions that the designed system should perform, or to performance metrics.

DPs may be either quantitative or qualitative.

Associative relationships among DPs may be expressed in terms of computational formulas, empirical rules, selection matrices, experiment values etc.

Design Entities and Associative Relationships (2)

Fixed Primary DPs: inputs for the design problem with their values remaining invariable during deign cycles.

Variable Primary DPs: their values change independently during design process.

Dependent DPs: their values are fully defined by their children DPs and the associative relationships.

Variable Dependent DPs: dependent DPs that are partially defined by their children DPs but their final value is user-defined.

Design Parameter Hierarchical Tree

Design Parameter Hierarchical Tree

Architecture of The Current Approach

Deployment of the proposed methodology in 3 steps:

Solution Search with GAs

Genetic Algorithm (GA) uses a stochastic variation of values for both numerical and linguistic variable DPs.

GA may be deployed together with a numerical optimization method in order to extract globally optimal solutions.

Aspects to be considered before genetic optimization:

A set of DPs should be characterized as critical for the design outcome and identified as Performance Variables (PVs).

Definition of the optimization criterion.

Adjustment of the attributes related with a GA (population size, individual length etc.).

Simplification of the Design Problem

Results for the Example Case

The genetic optimization was deployed in the example case of oscillating conveyors for 30 generations.

After 10 runs of GA, the average time needed for extracting the optimum solution was 94.2655 seconds.

The elite solutions during the genetic optimization were recorded in order to build a simplified flat fuzzy structure.

The re-design using the simplified structure, after 10 runs of GA of 30 generations each, resulted to a mean time for optimum solution extraction equal to 21.6540 seconds.

Conclusions

Different types of the design associative relationships can be modeled.

It is applicable to all parametric design problems in a formal and systematic, step-wise process.

It copes with both quantitative and qualitative design knowledge.

The design problem is successfully approximated with simpler fuzzy structures.

Design objectives may be integrated with a unified optimization criterion.

Future Work

Utilization of most efficient optimization criteria.

Automatic adjustment for choices concerning genetic algorithm’s features.

Retrieval and adaptation of successful past design cases.

Evaluation against large and complex design problems.

End of Presentation

Thank you for your attention.