Powered lower limb exoskeletons for rehabilitation and daily assistance

     Powered lower limb exoskeletons (LLE) and their appropriate control strategies (CS) are developed to assist gait for different categories of people, for rehabilitation applications, and/or daily living activities. In his presentation, Mohamed Bouri will highlight examples of exoskeletons developed at the Ecole Polytechnique Fédérale de Lausanne, and share the main drivers of these developments that are respectively related to 1) simplicity to ease the implementation and increase the accessibility of the technology to people with lower limb impairments, 2) well-defined needs and specifications to decrease time-to-evaluate the technology, and 3) development of appropriate control strategies to effectively assist gait and balance.

 Advanced signal processing techniques for fault diagnosis of electromechanical systems

     Most of the used failure detection and diagnosis techniques perform spectral analysis, such as Fourier or MUSIC techniques. Although these techniques exhibit good results in stationary conditions, they are not well- suited for a majority of electromechanical systems (i.e. electric machines and drives applications). Indeed, these applications environment is predominantly nonstationary due to transients or variable speed operations. In this context, the involved signals are usually nonstationary, embedded in noise, and can contain closely spaced frequencies. It is then obvious that failures detection and diagnosis in such applications are challenging tasks that need using advanced signal processing tools. This plenary intends therefore to focus on state-of-the-art research and development as well as future trends in electromechanical systems condition monitoring using advanced signal processing techniques. Attempts will also be made to highlight future issues so as to index some emerging techniques.

 Recent developements in deep learning and its applications

     Deep Learning (DL) is a subset of machine learning, based on artificial neural networks. DL is a process involving training models to learn from vast amounts of data using deep neural network architectures. In recent years, DL has become increasingly popular due to its ability to solve complex problems in various areas such as image and video recognition, natural language processing, autonomous vehicles and so on. Data can be learned more deeply through various approaches, which are supervised, unsupervised, semi-supervised, reinforcement, transfer and online learning. According to the used learning approach, various neural network architectures have been developed such as Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), Convolutional Auto-Encodeur (CAE), Generative Pre-trained Transformer (GPT) and so on. The objective of this keynote is to present the main concept of the DL learning and its recent developments as well possible applications in various areas.

 Digital communications using SDRs: understanding the basics

 Open source for robotics:ROS (Robot Operating System), overview and perspectives

 Software architectures for real-time embedded systems