OPHIR

DFA evaluation within a CAD environment.

The initial aim of the Ophir project was the development of techniques to facilitate DFA evaluation within a CAD environment. Therefore the research was broken down into four main areas. These were as follows:

• Development of techniques for optimum assembly sequence generation
• Further development of the manual DFA evaluation methodology for a software implementation
• Development of geometric reasoning techniques to automatically extract data from the CAD model
• Development of expert systems to faciltate decision making with respect to definition of product structure and sequence generation

A software implementation was developed in order to explore fully the potential of the research, to test ideas and to demonstrate the results of the research to industrial collaborators and a wider academic and industrial audience.

Software Demonstrator

The demonstrator is comprised of three main windows: a traditional CAD window, the product strucure window and the assembly sequence window. Within this environment expert systems and geometric reasoning algorithms are automatically invoked to validate and evaluate the design as the product is developed. DFA analyses are performed using a combination of data input from the designer and data extracted from the CAD model. Design activity is not restricted to any particular window at any particular time so that product development may proceed in a piecemeal manner as the design becomes more concrete.

Figure 1 - The Ophir Demonstrator

Development Environment

The Ophir demonstrator was developed using C++ and the MFC library around an ACIS solid modelling kernel. A "four-layer model" was used to store data about the product at four different levels. Each layer of the product model depends on, has access to or includes all data in the lower layers. They are labelled as follows:

• Component Model - Solid model together with attribute information such as surface finish, etc.
• Assembly Model - Position and orientation information for each component within the assembled product
• Component Interaction Model - Mating faces information incorporating joining processes, method of assembly
• Assembly Plan - Time sequence information i.e. sequence of assembly operations including non-assembly processes

This information was stored in a series of MS Access databases, to enable testing of the environment via manual data input to the database. This was necessary since development of interdependent elements was ongoing simultaneously. For example, geometric reasoning data required for DFA analyses was not fully developed prior to implementation of some DFA evaluations and so default values were provided in the database. In addition, access to materials and manufacturing process data was also provided in a series of databases.

Assembly Structure Definition

One of the main benefits gained from the application of DFA results from systematically reviewing functional requirements and replacing component clusters with single integrated pieces. Analysis of numerous DFA case histories indicates that invariably the success of proposed design solutions relies on adopting alternative fastening methods, different materials and/or manufacturing processes. Therefore it is essential that the designer is able to explicity consider the product structure in relation to the ease of assembly.

This window enables the user to interactively define and manipulate the product structure. Thus consideration and documentation of proposed component and subassembly partitions is achieved. This allows the exploration of differing levels of parallelism in the assembly sequence. Thus, the designer recognises the importance of a suitable assembly structure and by applying heuristics, is ensuring it is optimised in terms of ease of assemblability.

Sequence Construction

Most automated assembly sequence generation systems use sophisticated search algorithms, that take account of geometric constraints to create a collection of feasible sequences. Heuristics based upon prior knowledge, may then be imposed by the user to identify the most practical sequence. In this project, industrial investigations carried out across ten diverse companies (ranging from automotive to medical equipment) showed that, by contrast, human planners simultaneously consider geometric feasibility and 'best practice' to define a good assembly sequence. Furthermore, the assembly planners consulted invariably used experience of the product and assembly equipment in conjunction with geometric data from the CAD models to produce the best sequence. Therefore, in this project a manual sequence construction method was maintained, but with advice offered to the designer based on expert system evaluations.

In practical terms, the components added to the product structure are stored in a holding bay, prior to being placed in assembly order in the sequence construction window. This holding bay ensures that no parts are omitted from the sequence. Given the fact that all components may not yet be added or created, the user is able to develop the sequence in a non-sequential manner. Additional pre- and post-processes such as joining and other assembly operations are added as required to the sequence. The interface also allows the user to consider assemblability issues for the subassemblies defined in the Assembly Structure. The whole sequence or a subset of the sequence, dealing with a single subassembly can be considered at any one time and manipulated as appropriate.

Figure 2 - The Assembly Sequence Window

Sequence Validation

Sequence validation and optimisation occurs through the satisfaction of geometric ('hard') and process ('soft') constraints. Geometric constraints deal with the geometric feasibility of the assembly. Conversely, process constraints constitute suggestions for 'best practice' relating to processes and materials, and so whilst particular options may be feasible, they may not be recommended. By definition, a valid assembly sequence will not violate any geometric constraints and will satisfy as many of the process constraints as the user feels is acceptable after consideration of any potential conflicts.

Geometric constraints:

• Consistent geometry - there must be no interference between parts within the assembly other than where intended for a particular type of fit.
• Feasible trajectory - there must be no collisions as each part is brought into contact with the rest of the assembly, either between parts or between the gripping tools used during the insertion process.
• Feasible joining processes - each part in the assembly will be held in place either by surrounding parts or by some joining process, which must be validated in terms of access for the tools, required.
• Stability - every part must remain gravitationally and elastically stable at all times during assembly.

Process Constraints:

• Compatible materials - when different materials are placed adjacent to one another there will always be a detrimental reaction of some degree and hence, the assembly must be checked for incompatible adjacent materials. The application of surface coatings must also be considered at this point.
• Compatible joining processes - the joining process, once confirmed as being geometrically feasible, must be appropriate for the types of materials and other part characteristics.
• Compatible joint characteristics - there are many types of joint including kinematic, static, clearance and interference . The form of the joint must be suitable for the process and material involved.

By ensuring that a design complies with these constraints, it is possible to ensure creation of a "good" sequence with less errors. Evaluation of the designer's sequence is provided by the DFA analysis.

DFA and Sequence Evaluation

Analysis of numerous DFA case histories indicates that, invariably the success of revised design solutions relies on an understanding of the capabilities of manufacturing processes and the adoption of different materials and/or manufacturing processes. DFA analyses are invoked at the earliest possible stages of the design process, whether there is CAD data available or not. Inference from the components attributes, estimated size and complexity, and default data are used to approximate early DFA evaluation. As the design develops and more detail is accessible, so the DFA evaluations become more reliable and accurate. As a result of the research during this project, DFA support was categorised in three ways:

• Product Group Support - consideration of product families, variants and modular design
• Product Structure Support - support for partitioning of assemblies, part count reduction
• Component Detailed Design Support - consideration for component manufacture and handling, process capabilities

Geometric reasoning techniques are used to extract data from the CAD model. This may be for simple analyses, such as volume or size calculations, where the manufacture of a certain component may exceed the capabilities of its manufacturing process, or more complex geometric reasoning where the stability of a stack of parts may be called into question. A major factor affecting ease of assembly and component handling times is component symmetry - this provided the focus of geometric reasoning research during the Ophir project.

Expert System Support

As detailed data is not always available, any decision making process during the early design stages, must cope with indeterminacy. It is here that expert systems incorporating heuristics and expert knowledge with reasoning ability are best able to give guidance and suggestions on how to create a product that is easy to manufacture and assemble. There are many aspects of this environment which can benefit from expert system support and the major target of expert system support in the Ophir demonstrator was the generation of assembly sequences. During the sequence construction process the, so called, 'Expert Assembler' advises which part(s) to start with (the Starting Component Advisor) and which part(s) are appropriate for subsequent insertion to the assembly (the Next Component Advisor). These recommendations are made known to the user through the use of different coloured component hightlights in the assembly sequence 'holding bay'.

Figure 3 - The Holding Bay

The Starting Component Advisor provides suggestions on which part or parts should not be used as the base component(s). It is then easy to deduce which part(s) can be used as a starting component. Based on various attributes, a number of rules have been extracted from case studies and knowledge engineering exercises with experts in industry.

The Next Component Advisor highlights the best possible next part to add to the assembly. It provides suggestions based on component group information in the assembly structure, the assembly strategy preferred by designer (such as bottom up, inside out), and some general rules extracted from case studies and industrial experts.

Both advisors are transparent to the user and are continually updated as more information becomes available. The reasoning behind the advisors is accessible and secondary to the user's decision making. The advisors are working with incomplete data as only the designer is aware of data unavailable to the system.


Geometric Reasoning Support

Many aspects of DFA evaluation are dependent upon the geometric characteristics of the component parts in terms of manufacturing complexity and handling characteristics. Many current computer-based implementations of DFA rely on time consuming and subjective data input from the user. However, this may be reduced by the automatic extraction of data from the CAD model, once it is available. In addition, a successful assembly sequence is dependent upon its geometric feasibility (as defined by the geometric constraints) and therefore geometric reasoning may also be used for sequence validation purposes. For instance the degrees of freedom of a component will be determined by the order of assembly operations including the defined type and timing of joining processes. To analyse stability, the centre of gravity of the component relative to its position and orientation within the assembly must be evaluated. Most solid modellers are able to calculate simple properties such as volume and other mass properties but many other required aspects of the geometry are difficult to infer. For example the determination of mating faces, which affect the degrees of freedom, requires a complex algorithm.

One geometric characteristic which affects both part handling and ease of assembly is component symmetry. DFA recommends that a part should be either completely symmetric or distinctly asymmetric, parallel and perpendicular to its insertion direction, to minimise orientation times and avoid incorrect assembly. During the Ophir project, a geometric reasoning procedure was developed for detection of symmetry in a manner appropriate for identifying both exact and partial symmetries of a component from its b-rep solid model.

The methodology involves searching all the faces of an object looking for both "internal symmetry" and matching pairs. A naive implementation would be computationally very expensive involving many pairwise face comparisons, but some additional heuristics can be employed which tame this and also lead to some other desirable properties.

The published procedure uses face properties (including area, surface type and number of bounding edges) to quickly cut down the number of potential face matches and to allow a ranking / grouping which can be used to further reduce the number of face-face comparisons. The method actually conceived uses the loops of edges (outer and inner boundaries of faces) as the basic element for comparisons. This has the useful effect of, for example, allowing matching between same shape faces (i.e. similar outer loop) with different internal detail (inner loops). This led directly to consideration of how the concept of "inner" and "outer" loops can be generalised to deal with faces on curved surfaces (cylinder, sphere, torus).

Figure 4 - Definition of Loop Type Figure 5 - Symmetry Detection of Camera Model

This technique allows sufficiently rapid detection of symmetry to be practically useful and, additionally and quite importantly, the detection of partial symmetry : the symmetric elements and additional asymmetric features may be identified and considered independently. This is a valuable asset in extending the work to deal with other related functions, such as definition of insertion trajectories and evaluation of shape complexity.

OPHIR Project Site: If you want to visit the old ophir website, please click here