|BOFFADOSSI ROBERTO||Cycle: XXXVII|
Section: Systems and Control
Advisor: FAGIANO LORENZO MARIO
Tutor: PIRODDI LUIGI
Major Research topic:
Energy-aware cyber-physical systems for sustainable manufacturing processes
Over the last few years, the theme of sustainability has become central on the global political agenda and on the industrial plans of most of the big companies. Production processes are one of the principal sources of energy consumption, pollution, and waste of natural resources. For this reason, the Green Manufacturing paradigm has been introduced with the aim of guiding the industrial innovation towards an increased environmental consciousness. Automation has a fundamental role in the “green revolution” of manufacturing and a large number of researches, both academic or industrial, are actually involved on this topic, in particular in the areas of modelling, data analysis, control, and optimization. To reach the sustainability goal it is necessary the implementation of information and communication technologies (ICT), such as internet-of-things (IoT), big data, artificial intelligence (AI), which involve the use of cyber-physical systems (CPS). Industry digitalization is considered as the main road to improve efficiency, integrating physical and virtual worlds. In the last five years the interest of the research community as constantly grown in this direction. The most promising and debated technology consists in the Digital Twin (DT), a living, intelligent, evolving, virtual replica of a physical entity or process. The main functionalities of a DT are the following: ;
- Monitoring: the DT replicates the physical object in the virtual world, thanks to the real-time cyber-physical synchronization and the so called ”closed-loop optimization”. It provides an-always-available and up-to-date representation, that can be used to analyse and store data, extract information, observe non measurable behaviour. This process includes the use of Virtual Sensors, Data Mining, and Machine Learning techniques. ;
- Control: the DT is not just an intelligent model, in fact, it can interact with the physical counterpart and can condition its behaviour. In smart manufacturing, where high level tasks need to be performed autonomously, the DT is the core of a possible AI controller, that deals with failure detection, decision-making, and predictive maintenance. ;
- Optimization: by means of the DT is possible to perform high fidelity simulations, what-if analysis and process planning, in order to optimize the performances of the physical entity. The updated model of the DT merges the benefits of both Data-driven and Knowledge-based methods. Furthermore it allows to operate the system analysis in the virtual world, saving costs and time. ;
- Desynchronization with the physical entity: the drift of the virtually expected energy performances from their real behaviour can be caused by model errors or by changes of the physical entity and of the environment. We propose to use Adaptive Learning Algorithms to deal with this problem. They can exploit the real-time data stream from the physical/virtual sensors to autonomously correct the DT. (Adaptive Machine Learning, Automated Modelling, Drift Detection Methods, Adaptive Windowing). ;
- Cost of development and poor reusability: the building of a DT needs a complex engineering process and field experts. Furthermore the obtained DT is uncapable to use directly pre-existing knowledge and to share/reuse the information extracted monitoring the physical entity. We propose to improve the Digital Shadow featuring the DT by exploiting the Knowledge-Base System (KBS) theory. (Inference Engine Strategies, Unsupervised Machine Learning, Ontology-based Models, Big Data Analytics) ;
- Low human independency and low interpretability: actually in DT frameworks, the human operator still has a central role, conflicting with the concept of smart manufacturing and intelligent manufacturing. The cause is the low interpretability of the AI decision-making process, which must be analysed and approved by a human operator. To solve this problem, a new generation of AI is emerging, called Explainable Artificial Intelligence (XAI), that merge Machine Learning with Automated Reasoning. This technology was recently implemented in the medical field, but it is new for industrial applications. We aim to improve the AI of the DT by means of a Knowledge Base (introduced in the previous step), in order to generate an XAI-based DT, that is able to perform an autonomous and interpretable decision-making process for energy optimization. ;
- Lack of standardization and practices: at the moment a shared definition of DT is still missing, as well as standard frameworks or data representation. That prevents the creation of a global infrastructure for DT sharing and the cooperation for a modular development of this technology. For this reasons the proposed research project intends to provide a general formulation of the Digital Twin for energy efficiency applications, and to validate the resulting CPS on a realistic setup. Thus, we will provide a complete and detailed report of the experimental results regarding the DT implementation. ;
These cookies are not optional. They are needed for the website to function.
In order for us to improve the website's functionality and structure, based on how the website is used.
In order for our website to perform as well as possible during your visit. If you refuse these cookies, some functionality will disappear from the website.
By sharing your interests and behavior as you visit our site, you increase the chance of seeing personalized content and offers.