Automatic formation of rules for feature recognition in solid models

S.S. Dimov, E.B. Brousseau, R.M. Setchi

Manufacturing Engineering Centre, Cardiff University, Cardiff,  UK 

Presentation Outlines 

1. Problem definition
2. Objective of the research
3. Method proposed
4. Illustrative example
5. Testing

6. Conclusions

Problem definition

An information gap exists between:

1. Product design
• Computer-Aided Design (CAD) data - Low-level data - For example: B-Rep models (faces, edges, vertices)
2. Manufacturing applications

• Computer-Aided Manufacturing (CAM) data - High-level data - For example: Machining features (holes, pockets, ...)

Solution: application of automatic feature recognition techniques

Objective of the  research

• Numerous AFR techniques have been developed in the past 25 years. For example:
• Rule-based approach
• Neural-network based approach
• Main limitations of existing AFR systems → domain specific
• Objective of this research:  to address knowledge acquisition issues associated with the development of AFR systems that could be employed in different application domains.

Method proposed

• A new method for generating automatically feature recognition rules is proposed. The method comprises the following steps:
• Training data creation
• Adoption of a feature taxonomy
• Design of B-Rep feature models
• Coding of B-Rep feature models at two levels of abstraction → two training sets
• Rule formation
• Application of an inductive learning algorithm (RULES-5) on both sets of training data → formation of two sets of IF - THEN rules

Example - Training data creation

• Adoption of a taxonomy: machining features
• Design of B-Rep feature models
  
• Coding of B-Rep feature models at two levels of abstraction

  

Example - Rule formation

• Application of RULES-5 inductive learning algorithm
• Formation of two sets of IF - THEN rules

Testing

• Implementation approach:

• Solid models (STEP AP203)

• Learning and Feature Recognition modules (Java™)

• STEP file parsing

• Benchmarking part (machining features)         

  

Conclusions

• A method for automatic generation of feature recognition rules that is applicable in different application domains is proposed.
• Application of inductive learning techniques
• Coding of features at two levels of abstraction.
• Utilisation of inductive learning techniques for AFR has several advantages:
• Formal mechanism for rule definition + consistency of the generated rule sets
• The development of AFR systems for different application domains requires only representative training sets to be formed for each of them.
• Generalisation capabilities of inductive learning techniques → AFR systems could recognise features that are not present in the training sets.
• The knowledge base of such systems could be extended easily to cover new or user-defined features.