In order to gain insight into business processes, multiple autonomous data sources residing in the manufacture enterprise need to integrate not only on storage and access methods but also capturing the meaning of data to get a coherent and meaningful data views for different applications requirements. This paper presents a semantic-based architecture for the integration of heterogeneous manufacturing data sources. The integration is realized on a semantic level by the explicit presentation of data semantics with ontology and relationships between ontologies. During applications usage, heterogeneous data sources which represent relations of relevance are dynamically organized in terms of their semantics. The paper discusses some major problems in the architecture: unified schema transformation, semi-automatic ontology generation and mediation.