Including Big and Unstructured Data

Including Big and Unstructured Data 

Decision support within various tasks in industrial businesses typically needs to consider both economical as well as technical aspects—with the latter often coming in extremely detailed and high-dimensional form (e.g. geometric data). Usually, the respective types of analyses also require the consideration of information that is only available in a semi-structured or unstructured form, e.g. service reports that sketch technical and geometric specifications in quality protocols or technical drawings. A comprehensive framework for BI in the manufacturing sector therefore needs to include both: an integrated presentation interface to connect structured and unstructured data as well as analytics of structured descriptions (meta data) to unstructured files. Given the sheer volume of the resulting data repositories and the need to also include Internet data for the reconciliation of decisions with customer and market trends, the potential of an inclusion of In-Memory and Big Data technologies (possibly Cloud-based) is salient.

Decision Support Based on Data from PDM and PLM Systems Many PDM and PLM systems are based on meta data centric file systems. They handle drawings, DMUs, or reports as files that are managed by their meta data. Diving deeper into the processes of the digital firm, increasingly often simulation and virtual reality methods are used for testing and enhancing product features. Those methods typically generate large quantities of data—with large portions being unstructured or semi-structured. Typical analytics regarding PDM data, simulation data, and data from the customer side are explorations on the degree of customer satisfaction.

Therefore, web analytics are increasingly used in social networks and use groups. Optical Methods in Manufacturing and Quality Monitoring The same applies to more and more widespread optical methods in manufacturing and quality monitoring. Process monitoring based on high resolution image processing leads to quickly increasing volumes of both structured and unstructured data, which both have to be integrated in decision support concepts to conduct lifecycle oriented root cause analysis e.g. for analyzing failure driven warranty cost in maturity stage management or engineering management.

An Integrated Framework 

Summarizing the previous insights, BI can unlock benefits in the manufacturing sector by bringing together technical and business-oriented information in a comprehensive decision support. This is on the one hand relevant at the operational and tactical level for the design and steering of processes. It is on the other hand of interest from a strategic perspective, as a holistic and in-depths view on these contents enables to uncover new decision options and a better understanding of the potentials and limitations of certain decisions. This requires changes to traditional BI architectures. A resulting architecture framework is depicted in Fig. 2.1. In more detail:
Data Support Layer 

To support the decisions mentioned above, data has to be extracted both out of business and technical systems. For example, product feature data can be found in DMUs, while process logic information can e.g. be gained from MES, SCM, or directly gathered from UC systems. Within the enhanced framework, the data sources therefore encompass IT systems from the complete product lifecycle. Due to the need to store additional data formats, structures, and models with distinctive use and access profiles, the data support layer is extended by additional data warehouses, esp. for product oriented and technical data (product DWHs) and for data on process logic as required by BPI applications (process DWHs). Some of the relevant data directly streams in from sensors on the shop floor. This, as well as unstructured data from inside and outside the enterprise, leads to real time and Big Data requirements that complement the data support layer.

Information Generation, Storage, and Distribution Layer  

The Information Generation, Storage, and distribution layer needs to include tools that are capable of analyzing the newly categories of data (polystructured data, process data)—leading to the need to build connections to NoSQL and Big Data components in the Data Support Layer. This goes along with the requirement to design pertinent data models.

Infrastructure Options 

As they open up or prohibit application options, new infrastructural options (e.g. In-Memory or Cloud Computing) for coping with the aggravated performance requirements and the increasing volatility of the solutions, need to be included in the framework.

As this overview indicates, an industry-specific BI architecture that incorporates and adapts various new trends in BI and consequent might unfold competitive advantages. However, as many of the discussed concepts are still evolving and so far only implemented selectively and rudimentarily, the architecture framework can only function as a starting point that cannot replace a comprehensive companyspecific evaluation. Furthermore, many open questions need to be addressed on the research side as well, e.g. questions on how to balance out trade-offs when choosing between Cloud-based Big Data services and in-house In-Memory solutions, or when comparing the use of established operational systems (like PLM or MES systems) and integrated DWH based solutions for various decision scenarios. Furthermore, a full-fledged integrated product DWH brings various challenges regarding the integrated data models, the data visualization, and the interplay with the process and “classical” business KPI DWHs. However, tackling these issues might be highly valuable—particularly for enterprises in turbulent, complex, and global environments. Here, the capability to come to a thorough understanding of the business and to respond in-time to unexpected challenges can have consequences for the sustained survival of the enterprise.