Intelligent control: Intelligent controllers play an essential role in the design of more precise and energy-efficient smart systems, resulting in more safety, productivity, and sustainability. Combining modern nonlinear control techniques (sliding modes, adaptive methods and feedback linearization) with soft computing and machine learning algorithms (artificial neural networks, fuzzy logic, reinforcement learning and Gaussian process regression), it is possible to design control strategies capable of dealing with the uncertainties and external disturbances inherent to the unstructured world in which we live.
Autonomous systems: By interacting more closely with humans, smart systems will increasingly face unexpected situations that could not have been foreseen by their designers. In this way, autonomous systems must be able to make quick decisions and still respect the moral and ethical issues of our society. This exciting research area is at the fore front of technological development and also offers the possibility of dealing with a multidisciplinary scientific topic, ranging from biologically inspired questions to important philosophical aspects.
CaNeLis (2022 – 2025)
Our goal in this project is to develop a multisensor data fusion approach for the intelligent control of welding processes.
Weld properties can be measured using many different types of sensors, and each sensor type has its strengths and weaknesses for estimating attributes such as weld shape and quality. In fact, due to its inherent limitations, a single sensor type alone cannot provide a trustworthy estimate of weld properties. However, data fusion methods can be employed to combine the imperfect information (raw data) from different sensors so that the strengths of one type can compensate for the weaknesses of others. The adoption of a suitable method for sensor data fusion allows more valid information to be captured from the raw data, which can lead to a significant improvement in the overall accuracy and reliability of the weld. In this regard, data-driven machine learning methods can handle this task by means of the temporal and spatial correlation between the raw sensor data related to different physical quantities.