Michal Blaho (Humusoft)
MATLAB helps researchers, developers and engineers create new algorithms and devices. It is released twice a year with a number of new developments in the core module and individual extensions. During the presentation, you will see new capabilities for modeling, simulation, and design sharing, as well as new tools to increase your productivity and create better code and models. New features will focus on areas such as:
Lubor Zháňal (Humusoft)
3D FOUNDATION: MATLAB & Unreal Engine
Jaroslav Jirkovský (Humusoft)
In Simulink, dynamic systems can be modelled by different approaches: by identifying models from measured data, by means of classical (causal) modelling or by means of so-called physical modelling (acausal model). The causal approach to modelling describes the individual parts of the system by means of input-output relationships. The approach is suitable for modelling differential and difference equations describing a system, or if one wants to use standard approximations of systems in the form of transfer and state-space descriptions. The approach is also suitable for modeling control algorithms and filters. In contrast, physical modeling uses libraries of blocks that represent elements of real systems. From these elements, models of systems are constructed based on their physical structure. The connections of the blocks represent the bidirectional interaction of the elements and the flow of energy in the system. The balance equations of the system are derived automatically from the diagrams and the user does not have to worry about them.
The physical modelling in Simulink is provided by the Simscape superstructures. They allow modelling of mechanical, electrical, hydraulic and multiphysics systems. Prepared libraries of blocks are available, from elementary elements (e.g. spring, damper, resistor, capacitor, pipe, …) to complex components (e.g. 3D solids, electric motors, batteries, pumps, directional valves, …). If you do not find a suitable block in the libraries, you can create your own component using the Simscape Language, and combine the new element with other blocks from the libraries.
The latest physics modelling capabilities include 3D solid modelling with contact detection, modelling of variable and flexible solids (beams), modelling of fuel cells and batteries including aging effects, and modelling of fluid systems with thermal effects.
Michal Blaho (Humusoft)
Autonomous systems are becoming an increasingly common part of our lives. However, with the advent of autonomy comes challenges in algorithm development that need to be addressed. The development of autonomous systems is accelerated by simulation environments in which algorithms are designed and multiple scenarios are tested. In this talk, we will review the state-of-the-art for simulation and algorithm design for mobile, UAV or AUV applications. We will show how to use simulation in virtual environments, options for detection, localization, tracking scheduling and decision making. We will also show the possibilities of deploying the algorithms on target platforms.
Jaroslav Jirkovský (Humusoft)
Wireless communication systems are all around us. Standards such as WiFi, Buetooth, Lte and 5G are connecting an increasing number of devices. In MATLAB, it is possible to model and simulate these systems end-to-end, i.e. from signal generation according to the respective protocol, through its transmission and transmission to reception. Models can be created with varying levels of complexity depending on which aspect of transmission you need to focus on – for example, you can model transmission loss only as a statistically defined distortion (SNR) or create a sophisticated model of amplifiers and antennas including optimization of radiation characteristics, material effects, etc.
In MATLAB and Simulink, you can model the physical layer of a wireless communication system and add algorithms (such as OFDM, Massive MIMO, beamforming, …) for signal processing in these systems. It is also possible to create algorithms for Software Defined Radio (SDR), where the essential part of the signal processing is implemented in software in embedded processors or FPGAs, and by changing the program you can then change frequency bands or communication protocols. From the developed and tested algorithms, source code can then be automatically generated in C or VHDL and can be quickly moved to physical prototyping and actual implementation on the target device.
Jaroslav Jirkovský (Humusoft)
Artificial intelligence methods use data and create a program to perform a given task. At the core of the resulting program is a mathematical model that evaluates outputs based on input data. Currently, machine learning and deep learning (learning based on deep neural networks; a specific subset of machine learning) are widely used. The task of learning is to adjust the parameters of the model so that the evaluation of outputs is done with maximum accuracy and minimum erroneous results. The basic tasks of machine learning are classification, regression and cluster analysis. MATLAB provides features for the complete development of machine learning and deep learning based applications, from data preparation to model creation and learning to the implementation and deployment of the resulting algorithms in the form of server applications or intelligent embedded systems.
The use of artificial intelligence methods for solving classification and regression problems with signals and time series involves the following steps. Data preparation, feature selection and extraction, labeling of training data for learning purposes, actual learning by machine learning or deep learning methods, and then selection and optimization of the obtained models, if necessary. Different approaches can be used in all steps, some need to be done manually, others are already fully automated. MATLAB provides intuitive graphical applications for data preparation, automated methods for feature selection as well as the AutoML approach (automatic selection and simultaneous optimization of the machine learning model). In the field of deep learning, there are then special networks suitable for working with signals (LSTM) or various options for modifying signal data to work with convolutional neural networks (CNN).
Michal Blaho (Humusoft)
We describe systems and processes using models based on mathematics or physics. Depending on how much information we have, we can model the system parametrically, as a black-box or physically. All approaches have largely one thing in common. Part or all of the mathematical model needs to be fit to the actual data. MATLAB provides several superstructures for fitting data and dynamical systems. For classical data, built-in tools and specialized superstructures such as Curve fitting toolbox, Statistics and Machine Learning Toolbox or Deep Learning Toolbox are used. In the domain of dynamical systems, we try to represent the dynamics of the system by a model from measured process data. MATLAB in this area offers superstructures such as System Identification Toolbox, Deep Learning Toolbox or Simulink Design Optimization.
Martin Kožíšek (Humusoft)
In the lecture, the simulation tools COMSOL Multiphysics, COMSOL Server and COMSOL Compiler will be introduced on a specific task. Emphasis will be placed on demonstrating the workflow of creating a simulation from specifying parameters to processing the results. How can software make the simulation engineer's job easier? The presentation will conclude with references to a unique way of teaching laboratory exercises suitable during distance learning.
Martina Mudrová (Humusoft)
A brief overview of the current possibilities that MATLAB and its complementary services offer for teaching and education, not only in the field of higher education.
Tomáš Fridrich (Humusoft)
The development of autonomous driving systems is based on a methodology called data driven development. The development is based on the analysis of measured data. The data is collected from sensors, cameras, radars and lidars placed in test vehicles running in normal traffic. The measured data thus comes from millions of kilometres driven. The data collection itself is carried out on specially developed hardware for this purpose.
Post-processing is performed on the measured data to extract sensitive and interesting information, to detect surrounding objects, traffic signs, lanes and to anonymise the faces and number plates of surrounding cars. We then create individual driving scenarios on which the AI learns.
Matouš Lorenc (Humusoft)
Demonstration of the preparation and integration of a multiphysics FEM model into a block diagram of the control in the Simulink environment. Presentation of two possible approaches, the robust LiveLink for Simulink module for creating cosimulation blocks and the integration of a lightweight ROM model using a state-space block. Both methods will be presented in an understandable way on a heat exchanger model.
Alessandro Tarchini, Marco Rossi (MathWorks)
The main evening news from the MATLAB TV studio in Turin.
Petr Semotam (Siemens, s.r.o.)
Automatic identification of anomalies in casting curves resulting from the casting of passenger car engine blocks. The aim of the project is to quickly eliminate a potentially defective casting from further processing. The classification model is built using a special kind of recurrent network (RNN) LSTM Autoencoder neural network.
Patrik Kováčik (Žilinská univerzita v Žiline)
YouTube educational channel „Materials for machinists“. Solving interactive tasks in the ReEducate platform. Solution of the project – Programming of the application for strength analysis of beams.
Robert Grepl (MECHSOFT)
In this lecture, we will present the TRAMotion software under development, which solves kinematic analysis of the motion of a rail vehicle (tram) with a focus on the calculation of the motion envelope, which is essential for assessing the feasibility of the concept and optimizing the vehicle parameters. The computational core uses numerical algorithms to solve the kinematics and is implemented in MATLAB. The user interface is built on a custom technology based on HTML and JS.
Jiří Minarik, Petr Liškář (Eaton elektrotechnika s.r.o.)
An example of a virtual EV model configurable with one or more different motors, inverters and gearboxes simultaneously, which users can customize or use only parts of it to do their job. The presentation is followed by a demonstration of one possible approach to simulating the driving of a two-track vehicle in a virtual or realistic environment using GPS data, incorporating the effects of driving dynamics and the work of advanced stabilization systems.
Daniel Youssef (Garrett Motion Czech Republic s.r.o.)
Predictive control (MPC) is an advanced method of controller design based on a model of the evolution of the controlled system. The main benefits of MPC over standard control methods are the ability to directly incorporate: system constraints (inputs, outputs, states), fault information (external and internal) and their predictions to improve control quality (less fuel, smoother control – lifetime, etc.). In MPC, the control problem is converted into an optimization problem that must be solved at each discrete step. This leads to the need for a fast and reliable solver which can be run even on low power units used in the automotive industry.
In this talk, a generic NMPC Framework, built on a MATLAB/Simulink environment, will be rebuilt to allow easy and efficient deployment of the MPC controller on several demo applications.
Ivo Vodička, Ondřej Harvan (Digiteq Automotive s.r.o.)
In our talk, we will present the development of a test bed for automated testing of automotive control units, based on the dSpace Scalexio platform. HiL for component testing brings specifics based on customer requirements that need to be considered in the design. The project deals with the physical circuitry and software processing within the toolchain. For this we use programs from dSpace and MathWorks in combination with tools developed by Digiteq Automotive.
Milan Kertész (Schaeffler Kysuce spol s.r.o.)
The gradual mechatronization and systematization of products is now an obvious trend in our world. This is due to the increasing demands of product users on the quantity and quality of their functions. Products are thus becoming more and more complex, but above all also more intelligent. Their intelligence lies mainly in their ability to react to different situations in the desired way. Moreover, in the automotive sector, great emphasis is placed on the safety of products, even when they fail. In any case, the intelligent management of such systems, which are subject to stringent demands for accuracy, reliability and safety at all times, requires a thorough knowledge of the behaviour of the system in all these circumstances. Automotive controllers commonly use such multiparametric models in their control algorithms, which help them, given a perceived system state, to adapt the actionable interventions in the system to achieve the desired effect. We are talking about the so-called model-based design, which is implemented in the control software component as C code generated from the system model in Simulink.