SMART DATA ANALYSIS AND FORECASTING
Digitalization and automation entail extensive data acquisition in a wide variety of sectors such as industry, commerce and healthcare. As a result, large amounts of data must be stored securely and processed in a way that is benefi cial to the user in order to derive valuable information from them.
From an ICT perspective, the basis for knowledge generation is the digitized know-how of domain experts and the associated optimization of business and production processes. By applying statistical methods, modern data and visual analytics methods, as well as machine learning, the existing knowledge is analyzed in context with the recorded data.
This allows anomalies and patterns to be subsequently identified and additional information to be derived through correlations for error and cause analysis. Using methods from the fi eld of artificial intelligence, knowledge is generated and recommendations for action are formulated for experts (expert-in-the-loop).
RISC Software GmbH has already gained a lot of experience through research and development projects as well as numerous projects with industry in various areas of data management and analytics with small and large amounts of data. With this know-how, RISC Software GmbH supports its customers in preparing for new challenges by providing a better insight into their own data.
DATA SCIENCE PROCESS
1. Data Engineering
Machine, health, sensor or IT infrastructure data are generated on the most diverse levels and systems. In order to use these data, they must be intelligently combined, filtered and linked. Checking the correctness and completeness of the data is also relevant. Existing IT infrastructures must also be included, thereby making data security and quality indispensable.
- Development of data models
- Automation of data import
- Extraction of complex features from medical image data
- Use of modern Big Data technologies for large amounts of data
- Use of common industry standards e.g., OPC UA
- Validation and verifi cation of data
- Detection and correction of erroneous data
- Determination of training and evaluation data
- Data modeling and mapping of domain knowledge
- Model development
- Data preprocessing and validation
- Feature engineering
- Explorative data analysis and visualization
2. Data Analytics and Artificial Intelligence
In order to gain knowledge from collected and filtered data, mathematical, statistical and artificial intelligence methods, as well as visual analytics are used. By using selected algorithms, large amounts of data are examined for correlations and patterns. Models are created to answer various questions. Interrelationships, connections and laws in the data can contain important information, on the one hand, for short-term reactions in real-time operation and, on the other hand, for effective predictions for the future.
3. Data Presentation
The visual processing of data and analysis results is an important tool for gaining new insights. An essential factor for the interpretation of analysis results is the strong involvement of domain experts. In addition, intuitive perception and the ability to combine are essential factors in the visualization of the results. The interaction between human expertise and machine processing can contribute significantly to the support of the decision-making process.
- Visualization of models and analysis results
- Individual dashboards
- Web-based front-end
- REST-API for connection to external systems
- Connection to reporting tools
- Integration of solutions into existing systems
AI METHODS FRAMEWORK
DATA SCIENCE CIRCLE
AnnaLyze – Virtual Production Assistant
AnnaLyze primarily supports domain experts. It brings together the entire know-how from the production, processes, confi guration, orders and the tools of all machines. New knowledge is generated from this using data and visual analytics in order to derive cause-eff ect relationships. This makes it possible, for example, to reduce the failure probability of individual components or increase productivity by reducing unplanned machine downtimes.
Medical Image Processing, Modelling and Simulation
Although large amounts of medical image data are generated in the daily clinical routine, their use for automatic analysis methods is often diffi cult due to the lack of additional information (e.g., shape and position of the aorta, location of the tumor). Manual creation is very time-consuming and publicly accessible data sets are rare. In the MIMAS research area, research is therefore being carried out on automatic methods that enable the rapid generation of this additional information.
Ontology-based Data Infrastructure
CALUMMA is a new generation of data management software that combines data complexity, interface diversity and visualization with user-friendliness. Domain experts are supported by the modeling, integration, validation, processing and evaluation of their data. Due to the genericity of the complete system, the application can be adapted to diff erent scenarios and data.
Intelligent Twitter Analysis
This project focused on investigating the premise of whether positive emotions in tweets about listed companies are related to their share price performance. During the four-month observation period a verifi able correlation between share prices and emotions in the tweets was established. These findings were obtained, among other things, through the use of sentiment analyses and deep neural networks.
Intelligent Railway Switch
Fundamental knowledge that should enable the generation of a digital image showing the condtion of a railway switch was developed together with voestalpine Signaling Zeltweg GmbH in the research project iTPP 4.0. This should be able to reliably predict future wear and tear or defects of the points at any time. Machine learning algorithms derive decisions on planned maintenance from data from railway infrastructure sensors in a self-learning manner.