Meet the SCCH at the HANNOVER Messe 2016
World stage for networked industry
We invite you visit us at H3, A44. We are glad do inform you about our topics.
The term Industry 4.0 (the smart factory) has become a buzzword. Often huge data streams (e.g., machine data,
process data, quality data) from diverse, heterogeneous data sources must be linked and analyzed in order to
provide a useful basis for decision support and recommended actions for humans. The application spectrum
ranges from the process industry and production, to energy management to the manufacture and maintenance
of machines and plants. Predictive maintenance is another important buzzword. If an industrial production system suffers unforeseen disruption of a machine and therefore downtime, then this is a worst-case scenario: production is delayed and enormous costs can ensue. Modern control systems are incapable of evaluating their own state so as to derive relevant information for the maintenance crew. The vision of machine diagnosis and prognosis is to close this gap. The goal is to predict a time (warning) when corresponding service measures can avoid possible damage or downtime (see figure). This improves the state of a component or machine and production continues according to plan. Data mining and machine learning methods enable us to create fault prediction models in order to find this early warning point and thus plan predictive maintenance strategies. The key is the combination of expert knowledge
and data-based fault prediction models. This increases plant availability with reduced use of resources.
The use of predictive maintenance strategies is promising in many areas:
- Increased plant availability because fault prediction promotes early detection of damages and reduced
- Reduced material and energy costs because maintenance is not bound to predefined schedules but
instead conducted as needed
- Improved planability of maintenance via state monitoring
- Heightened operational security by avoidance of
Chemometrics as a serice
The transformation of raw sensor data into processrelevant information via intelligent data analysis is
seen primarily in combination with Industry 4.0 (the smart factory) as an innovation engine. Applications
extend from process analysis to process optimization to process control. In process control, the factors to be
controlled are often not directly measurable, so that virtual sensors or soft sensors are used. Such sensors
convert other measurable data to the required factor, e.g., a critical quality attribute such as the concentration
of an unwanted byproduct in a chemical production process. Such virtual sensors achieve the required
precision only via analysis and modeling of historical process (and lab) data. Chemical production today
often employs such virtual sensors, e.g., to determine the concentrations of certain substances based on
spectroscopic measurements: sometimes elaborately calibrated chemometric models convert in real time the
(absorption) spectra recorded by a process spectrometer into the required concentration measures.
Nowadays in most of these Process Analytical Technology (PAT) applications, the chemometric models are
created by modeling experts and are manually applied on a process spectrometer. In the case of complex and dynamic processes and products, the required models often transcend static linear models. Therefore the models themselves become more complex, and especially the maintenance of the models, e.g., more frequent adaptation to the product matrix, becomes more costly. These challenges are difficult to meet with conventional solutions.
Members of the Austrian research network PAC are currently working in the realm of the research project
“imPACts” on methodological foundations and software tools that support more efficient handling of complex
chemometric models, from their development (calibration) to operationalization, real-time execution, and
maintenance. Specifically, they are working on a Computational (Chemometric) Model Environment that is
designed as a modern software service (Chemometrics as a Service). Via state-of-the-art Web service interfaces,
spectra can not only be evaluated and converted by complex models, but also used for recalibration and
online adaptation of models. An initial prototype of such Chemometrics as a Service was recently put into operation by research partners SCCH and RECENDT for an application of partner company MetaDynea; the prototype enables simpler
generation of new, more complex models and their computations.