Bart Meyers,Hans Vangheluwe,Pieter Lietaert,Geert Vanderhulst,Johan Van Noten,Michel Schaffers,Davy Maes,Klaas Gadeyne
Abstract:Abstract The development of artificial intelligence models for data driven decision making has a lot of potential for the manufacturing sector. Nevertheless, applications in industry are currently limited to the actionable insights one can discover from the available data and knowledge of a manufacturing system. We call the process to obtain such insights “ad hoc analysis”. Ad hoc analysis at system level is very complex in an industrial setting due to the inherent heterogeneity of data and existence of data silos, the lack of information and knowledge formalization, and the inability to meaningfully and efficiently reason about the data, information and knowledge. In this paper, we provide and outline a framework for ad hoc analysis in manufacturing based on knowledge graphs and influenced by the metamodelling paradigm. We derive its requirements and key elements from an analysis of several industry application cases. We show how manufacturing data, information and knowledge can be combined and made actionable using this framework. The framework supports workflows and tools for the data consumer (i.e., data scientist), and for the knowledge engineer. Furthermore, we show how the framework is integrated with existing data sources. Then, we discuss how we applied the framework to several application cases. We discuss how the framework contributes when applied, and what challenges still remain.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the challenges faced in conducting system - level ad hoc analysis in the manufacturing industry. Specifically, these challenges include:
1. **Data heterogeneity**: Data sources in manufacturing systems are diverse and in different formats, which complicates obtaining necessary data insights. For example, data in aspects such as product design, manufacturing execution, quality assurance, and sensor readings may be stored in different databases or data silos.
2. **Insufficient formalization of information and knowledge**: In enterprises, information and knowledge are usually scattered among different knowledge sources (such as domain experts, data scientists, machine manuals, research papers, AI / simulation models, etc.). This knowledge is often tacit or informal, causing data scientists to have difficulty understanding data when conducting ad hoc analysis, and being prone to analysis errors or inefficiency.
3. **Inability to effectively perform inductive reasoning**: Ad hoc analysis requires flexible experimentation with data and knowledge, but currently there are no effective tools to support statistical analysis, logical reasoning, and AI workflows (such as correlation analysis, feature modeling, training, and validation). In addition, the discovered insights need to be found and accessed in the future and explicitly linked to relevant concepts.
To address these challenges, the paper proposes a knowledge - graph - based framework, aiming to provide solutions in the following ways:
- **Unified data access point**: Provide a single access point to all data, information, and knowledge through the knowledge graph, simplifying the data acquisition process.
- **Formalize and structure knowledge**: Formalize data, information, and knowledge to ensure that they are well - formed and consistent, thereby improving interpretability and traceability.
- **Enhanced reasoning ability**: Support automatic reasoning and flexible workflows, improving the efficiency and accuracy of data analysis.
The paper also details the specific implementation methods and verification cases of this framework, demonstrating its value and challenges in practical applications.