Search the digital library catalog

Abstract: This thesis explores the possibility of running neural networks on microcontrollers and how to optimize their performance using instruction set extensions. Microcontrollers are seen as too weak to run neural networks. We challenge this view and show that stripped-down neural networks can run and be useful for some applications. We used an open-source microcontroller called PULPino to run our neural network. The benefit of various instructions and optimizations for minimizing energy consumption to run deep learning algorithms was evaluated. Hardware loops, loop unrolling, and the dot-product unit were implemented and tested. We developed an FPGA-based testing system to evaluate our hardware. We also developed a deep learning library and a test neural network for our hardware. We wrote two versions of the deep learning library. One version is the reference code, and the other is the optimized code that uses the dot product unit. Using the testing system, we tested the performance of the two versions. The synthesis was run to determine the power and energy consumption. We also tried out various optimizations to see if the performance could be improved. Using instruction set extensions and algorithmic optimizations we reduced the clock cycle count by 72% for the convolutional layers and by 78% for fully-connected layers. This reduced power consumption by 73%. We compare our results with related research.
Keywords: deep learning, embedded system, instruction set, RISC-V
Published: 03.11.2020; Views: 498; Downloads: 97
Full text (2,69 MB)

Abstract: This master’s thesis presents the development of a machine-vision based localization unit developed at Robert Bosch GmbH, Corporate Sector Research and Advance Engineering in Renningen, Germany. The localization unit was developed primarily for position detection purposes with three degrees of freedom in highly versatile manufacturing systems but has an immense potential to be used anywhere where a precise, low-cost localization method on a two-dimensional surface is required. The complete product development cycle was carried out, from the components selection, schematic and optical system design, to the development of machine vision algorithms, four-layer Printed Circuit Board design and evaluation using an industrial robot. Thanks to the use of a patented two-dimensional code pattern, the localization unit can cover a surface area of 49 km2. The size and speed optimized, self-developed machine-vision algorithms running on a Cortex-M7 microcontroller allow achieving an accuracy of 100 µm and 60 Hz refresh rate.
Keywords: localization, machine-vision, code pattern, image sensor, embedded system
Published: 14.01.2020; Views: 398; Downloads: 0

Abstract: The master's thesis introduces the process of modernizing an older model of a passenger car. The mechanical, electrical and programming solutions were required to upgrade the existing car. Upgrades have been made to improve safety, comfort and performance. The safety upgrades contain the development and implementation of the brake temperature and the tire pressure monitoring system, respectively. The comfort upgrades include the car’s light system and accessory improvements, such as “follow me home” function and automatic window closing. The performance upgrades include installation of racing coilovers and the development of remote damping ratio adjustment system. For the central logic unit, a vehicle specific Android device was installed. It controls the local electronic control units made for each specific task and displays tire pressures and brake temperatures. The Android device communicates with the electromechanical car systems by using several serial buses. The communication protocols and their messages in the original car had to be reverse engineered to establish proper communication among the systems in the upgraded car.
Keywords: passenger car, serial buses, reverse engineering, adjustable suspension, sensors
Published: 02.08.2019; Views: 549; Downloads: 95
Full text (6,95 MB)

Abstract: V diplomskem delu smo razvili knjižnice visokonivojskih programskih funkcij za mikrokrmilniško razvojno ploščo STM32L4. Knjižnice vsebujejo funkcije za izbiro in nastavitev sistemske ure, za konfiguracijo in izbiro vhodnih/izhodnih priključkov, za realizacijo časovnih funkcij, za pretvorbo analognega signala v digitalnega in komunikacijo prek komunikacijskih vmesnikov UART, SPI ter serijskega vodila I2C. Za vsako funkcijo smo napisali kratek testni program in navodila za njeno uporabo.
Keywords: mikrokrmilnik, programski jezik C, UART, SPI, I2C
Published: 16.11.2017; Views: 1161; Downloads: 142
Full text (4,91 MB)

Abstract: Magistrsko delo obravnava metode preverjanja pravilnosti obnašanja sistemov, ki temeljijo na opisu sistema s procesno algebro. Podana je definicija procesne algebre in primeri opisov sistemov s procesi. Predstavljeno je ugotavljanje ekvivalence sledi, stroge, vejitvene in šibke opazovalne ekvivalence, ugotavljanje testne ekvivalence ter simbolično preverjanje modelov z izjavno vejitveno temporalno logiko ACTL. Vse obravnavane metode se med seboj odlično dopolnjujejo in skupaj tvorijo močno orodje za formalno verifikacijo sistemov. V magistrskem delu je opisana izvedba takšnega orodja z BDD-ji. Uporaba orodja je ponazorjena na primeru verifikacije komunikacijskega protokola BRP.
Keywords: formalne metode verifikacije, sistemi s sočasnostjo, procesne algebre, opazovalne ekvivalence, testne ekvivalence, simbolično preverjanje modelov, ACTL, BDD
Published: 04.08.2017; Views: 1668; Downloads: 61
Full text (1,10 MB)