To support human decision making, Area 3 defines two objectives:

(1) Combine data-driven approaches with configuration management methods and simulation environments in order to provide a reliable, trustworthy (data) basis for decision making.

(2) Provide this objective basis for decision making to humans in such a way that it takes into account their cognitive capabilities (e.g., information filtering in stress situations) as well as the situation/context in which the decision has to be made (e.g., within production process versus design process) in order to ensure timely and optimal decisions.

This strategic project fosters these Area objectives by strategic research activities considering the following context:

• A secure data transmission module will be applied and extended that allows to transmit data from production site to assistive system.
• To ensure that end-user understand why the system made a particular decision, this project further focuses on state-of-the-art visualization tools (2D, 3D) and visual analytics methods that are used, e.g., to explain/explore decisions made by system, the applied model(s) respectively.
• The visual analytics tool can further be applied to support scheduling, performance monitoring, and anomaly detection for the manufacturing systems that might help the end-user in her decision-making process.
• The simulation of scheduling and re-scheduling after expected (predictive maintenance) and unexpected changes (e.g. down-time of machines due to failures) allows for a better resiliency of manufacturing processes. The (further) development of algorithms can therefore help in optimizing the design of production systems and schedules for shop floors in cases of stochastic failures.

• (Data) Analytics Methods Base & Computational (Data) Analytics

In order to build this reliable, trustworthy (data) basis for decision making we will create a collection of methods which allow us to collect data, facts, rules, engineering models, simulation models, etc. for a specific decision-making process within a specific application scenario. A specific challenge will be the integration of different methods into hybrid approaches which combine the advantages of the individual approaches, e.g. integration of model-based and data-based approaches. In order to prove the reliability and trustworthiness of the resulting data/facts basis it will be crucial to invest effort in the creation of training and test data sets which can be utilized as gold standards in order to benchmark the approaches and tools being developed.

• Decision Making Methods Base & Computation Decision Making

Decision support has to be personalized (to the individual human cognitive capabilities), contextualized (to the specific decision situation), and domain-specific in order to lead to timely and optimal decisions. Proven computational decision-making support mechanisms are visual analytics, (data-driven) recommender- and adaptable systems as well as simulations. Our aim is to synergize these methodologies in computational prototypes, enhancing decision-making support. Furthermore, we intend to implement specialized "industrial decision support" tools tailored to specific application domains. As with all human-machine environments, careful evaluation of the resulting methods and tools in real-world environments will be crucial to the success.

• Applying data transmission security in decision support assistance systems used in the manufacturing industry methods to protect company data
• Using visual analytics and data analytics methods to support transparency in decisions/models made/applied by/in decision-support systems in manufacturing industry
• New insights gained about the application possibilities or interlocking of data analytics and visual analytics

Decision support has to be personalized (to the individual human cognitive capabilities), contextualized (to the specific decision situation), and domain-specific in order to lead to timely and optimal decisions. Proven computational decision-making support mechanisms are visual analytics, (data-driven) recommender and adaptable systems. To contribute with regard to the later, we worked on a tool that should assist the users in analyzing their data by recommending the analytical methods to be used as next. For the recommendations, we observe the current analysis process and adapt the information space to what the user prefers and needs.

First, we worked on human-in-the-loop approaches for interactive data classification and comparison, by integrating active learning algorithms and similarity search methods with high-dimensional data analysis (see Figure 1). Second, we worked on novel concepts how eye tracking, as a novel user sensing modality, can be leveraged to detect user interest in visual data analysis, and support adaptive systems for data exploration. In a third line of research, we have developed concepts for user guidance in complex visual data exploration applications. A set of design guidelines was developed and analyzed.

In order to advance the field of Visual Analytics it is very important to collect and discuss the state-of-the-art in particular sub-fields. Together with collaborators from the University of Utah we surveyed existing work on multi-variate networks. In collaboration with US and UK colleagues we summarized the state-of-the-art on how to analyse interaction provenance data that is collected while users perform an interactive visual analysis. Besides these activities, we performed original research on the following topics: (1) guidance, (2) tabular data analysis techniques, and (3) onboarding. For the purpose of flexibly ranking tabular multi-variate data we continued the development of the Ordino visual analysis application, designed the novel Taggle visualization technique, and extended it with a support view that allows users to statistically confirm visual patterns. In cooperation with Prof. Aigner and his group at FH St. Pölten we designed and evaluated how to effectively onboard users to new visualization techniques.

Part of the project is related to the work in demonstrator project DP3. Classification and machine-learning are important methods for flexible production systems and adaptive scheduling. Here, data-driven approaches to optimize the configuration of production systems have been combined with simulation approaches used to determine the impact of changed configurations on the production system in the future. The new approach has been partially presented at the intermediate evaluation for the common research program of Pro2 Future and the Center for Digital Production.

The system consists of three components: (i) a classifier system capable of learning machine configurations given a particular product and the current state of the machine and tools, (ii) a scheduling and simulation system that is capable of re-organizing the production schedule if changes are required, and (iii) an integration component that links and controls the data-flows.

Web-based technologies are used to provide the technical connectivity. The scheduling component provides the means to rearrange the production schedule. However, this re-organization has again effects on the machine and tools usage.

Frequent Itemsets Mining is a fundamental mining model in Data Mining. It supports a vast range of application fields and can be employed as a key calculation phase in many other mining models such as Association Rules, Correlations, Classifications, etc. Many distributed parallel algorithms have been introduced to confront with very large-scale datasets of Big Data. However, the problems of running time and memory scalability still have not had adequate solutions for very large and “hard-to-mined” datasets. We proposed a distributed parallel algorithm named DP3 (Distributed PrePostPlus) which parallelizes the state-of-the-art algorithm PrePost+ and operates in Master-Slaves model. Slave machines mine and send local frequent itemsets and support counts to the Master for aggregations [1]. In the case of tremendous numbers of itemsets transferred between the Slaves and Master, the computational load at the Master, therefore, is extremely heavy if there is not the support from our complete FPO tree (Frequent Patterns Organization) which can provide optimal compactness for light data transfers and highly efficient aggregations with pruning ability. Processing phases of the Slaves and Master are designed for memory scalability and shared-memory parallel in Work-Pool model so as to utilize the computational power of multi-core CPUs. We conducted experiments on both synthetic and real datasets, and the empirical results have shown that our algorithm far outperforms the well-known PFP and other three recently high-performance ones Dist-Eclat, BigFIM, and MapFIM. Furthermore, a secure data connection framework has been developed, and it has been deployed at the pilot factory Vienna. To this end, a hardware component together with AVL and this component has been adapted to the context of data analytics in industrial settings.