Reverse Engineering of Domain Knowledge for Improving Configuration Management

As a product family evolves with the increasing number of customer specific members, the product configuration becomes extremely intricate. Configuration key-value settings are often incompletely documented, so their influence on the product structure and behaviour remains hidden. Since side effects and interdependencies of configuration settings are only partially known, the products can only be configured manually. In order to make the product variant management more efficient, we present an approach to reverse engineer the configuration knowledge from product configurations using data analysis techniques. We use correlation analysis to extract dependencies between configuration items. Our approach is conducted on an industrial product family with thousands of individually configured product instances. Each product configuration contains between 20 000 and 30 000 configuration parameters. Our goals in this case are (i) to accelerate the configuration process, (ii) to increase the costeffectiveness for quality assurance, and (iii) to extract and document the domain knowledge.