Témakiírások
Integration of process mining, multitemporal event analysis, and reinforcement learning algorithms
témakiírás címe
Integration of process mining, multitemporal event analysis, and reinforcement learning algorithms
intézmény
doktori iskola
témakiíró
tudományág
témakiírás leírása
Goal: Predictive maintenance, anomaly detection, forecasting and missing event interpolation are common tasks in the development of operational excellence solutions for manufacturing. To put more emphasis stochastic nature of discrete events we intend to study how survival analysis and Markov models can be incorporated to sequence and process mining algorithms and how these tools can be used to increase process safety and reduce costs (e.g. energy consumption).
Key ideas: Since complex production systems are strongly interconnected, alarm signals generate complex multi-temporal patterns even in case of a simple failure. We propose a multi-temporal sequence mining based approach to extract these patterns and form alarm suppression rules to reduce the workload of operators. We demonstrate the applicability of the concept in a simulated chemical technology. The results illustrate that the analysis of the extracted sequences is useful for the detection of the root causes of the process failures.
The second main research direction of this subproject will be connected to the application of event analysis in learning control. Capturing all scenarios for an algorithm (used for control or fault detection) is difficult. Deep learning algorithms can help to generate compact feature space. We plan to investigate the broad applicability of this idea and work out algorithms as a hybrid of the classical event and neural network based solutions.
Innovative applications: We try to develop tools that can be used to support alarm management activities, early warning algorithms that can be used to increase (process) safety and security of the vertically integrated industry 4.0 environment. We are also planning to develop some demonstrative applications based on the creative utilisation of open source databases, like GDELT.
We developed tools for sequence and process mining and applied it in the optimization of logistic processes. More details: www.abonyilab.com
Key ideas: Since complex production systems are strongly interconnected, alarm signals generate complex multi-temporal patterns even in case of a simple failure. We propose a multi-temporal sequence mining based approach to extract these patterns and form alarm suppression rules to reduce the workload of operators. We demonstrate the applicability of the concept in a simulated chemical technology. The results illustrate that the analysis of the extracted sequences is useful for the detection of the root causes of the process failures.
The second main research direction of this subproject will be connected to the application of event analysis in learning control. Capturing all scenarios for an algorithm (used for control or fault detection) is difficult. Deep learning algorithms can help to generate compact feature space. We plan to investigate the broad applicability of this idea and work out algorithms as a hybrid of the classical event and neural network based solutions.
Innovative applications: We try to develop tools that can be used to support alarm management activities, early warning algorithms that can be used to increase (process) safety and security of the vertically integrated industry 4.0 environment. We are also planning to develop some demonstrative applications based on the creative utilisation of open source databases, like GDELT.
We developed tools for sequence and process mining and applied it in the optimization of logistic processes. More details: www.abonyilab.com
felvehető hallgatók száma
1 fő
helyszín
MTA – PE Komplex rendszerek Lendület kutatócsoport, Pannon Egyetem, Folyamatmérnöki Intézeti Tanszék
jelentkezési határidő
2017-06-30