Introduction: The Information Overload in Design
Modern product development generates vast quantities of data, from initial requirements to detailed part geometries. While this documentation is essential for project continuity, it often leads to "information overload" during product reuse or adaptation. Designers struggle to identify relevant information within massive, outdated knowledge bases. Traditional databases excel at storage but lack the intelligence to "forget" obsolete data. In engineering, information is deeply interconnected; removing a single requirement can render entire subassemblies or functions irrelevant. Manually purging these dependencies is error-prone and inefficient.
Intentional Forgetting and the Cascading Operator
To address this, researchers have introduced Intentional Forgetting (IF)—a strategy to streamline knowledge bases by systematically removing unnecessary information. The core of this approach is the Cascading Forgetting operator, implemented within an ontological framework using Description Logic and OWL.
The operator relies on two primary metaproperties to govern the deletion process:
- Rigidity: A class is considered "rigid" if its properties are essential to the identity of its members. If an individual is removed from a rigid class, the system assumes it should be entirely deleted from the knowledge base.
- Dependency: This defines existential relationships. For instance, if a "subfunction" depends on a "function," the deletion of the parent function triggers the automatic removal of the dependent subfunction.
The logic is formalized into rules that guide SPARQL queries to update the knowledge base (ABox). These rules ensure that when a design component is rejected, its "shadow" in the data—all the related requirements and parts—is also cleared.
Evaluation: The Combustion Engine Test-Rig
The approach was tested using a real-world case study: the redesign of a test-rig for studying friction in combustion engines. The scenario modeled the transition from a first-generation rig using piezoelectric sensors to a second-generation version using torque measurement. This shift required the "forgetting" of all components and requirements specific to the piezoelectric setup.
Key Results and Insights
The evaluation, involving seven engineering experts, focused on two metrics: Precision (avoiding over-deletion) and Recall (ensuring all obsolete data is removed).
- Perfect Precision (1.0): Engineers agreed with every deletion performed by the operator. This is crucial, as engineers are naturally "forgetting-averse" and fear losing critical data. A "cautious" operator that never deletes the wrong thing is preferred over an aggressive one.
- Improving Recall: Initial recall was 0.48, meaning the operator missed about half of the expected deletions. However, after refining the underlying ontology model and adding missing metadata, recall improved to 0.70.
A significant insight was that the ontology served as a "common language" between IT experts and engineers, allowing them to map complex design changes without needing deep expertise in each other's technical domains.
Limitations and Future Directions
While successful, the current operator has limitations. It cannot yet handle "counting dependencies"—for example, deleting an assembly only if its part count drops below a certain threshold. Additionally, the manual entry of metaproperties remains a hurdle. Future work aims to automate metadata tagging and implement standardized Ontology Design Patterns (ODPs) to minimize the modeling errors that initially hampered recall.
Summary updated