Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)

Permanent URI for this collection

https://hdl.handle.net/10889/43

Browse

Now showing 1 - 20 of 2633

Open Access
3D subject-specific subject specific modeling, finite element analysis and evaluation of biomechanics of the knee complex after anterior cruciate ligament reconstruction
Ρισβάς, Κωνσταντίνος; Risvas, Konstantinos
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
The Anterior Cruciate Ligament (ACL) injury is one of the most common knee injuries during sports activities. Anterior Cruciate Ligament reconstruction (ACLR) is a surgical approach that has emerged as the golden standard treatment, where a graft is placed in the position of the torn ligament through tunnels that are drilled on the femur and tibia bones, close to the native ligament origin and insertion sites. Over the course of years, many ACLR techniques have been developed featuring a plethora of parameters like the number of the drilled tunnels, the tunnel positioning, the graft harvesting site, the graft pre-tensioning load and much more. To evaluate the results of each selected ACLR approach, follow up examination is required.The complex nature of these examinations and recent technological advances paved the way for computational biomechanics as an alternative and effective tool to the disposal of biomedical researchers. Finite Element Analysis (FEA) is a valuable modeling and simulation approach of computational biomechanics, that enables to capture the geometry and material effects on the response of the anatomical structures like the knee joint under various loading conditions. In this thesis, a workflow that generates subject-specific FEM models of the knee joint to assess knee biomechanics after ACLR is developed. The proposed pipeline includes modeling and simulation of all key steps of ACLR, from tunnel drilling to graft fixation. Furthermore, it provides simulation scenarios that represent daily life activities, such as gait, and clinical exams, such as the Lachman test, to evaluate the performance of the ACL reconstructed FEM knee models, laying the ground for performing "what if" scenarios in order to predict the effectiveness of the ACLR approach. In the confines of this work we evaluated the performance of Single and Double Channel, Anteromedial and Transtibial portal surgery techniques and the response of three different graft materials under various pre-tension loads and fixation angles in simulation scenarios representing the Lachman test and a gait trial. Despite the limitations regarding the FE models constraints and the inherent complex nature of FEM simulations, the present work provides the groundwork for developing an extended and fully featured ACLR modeling framework.
Open Access
3D γεωμετρία και όραση για αυτόνομα οχήματα
(2023-06-30) Τσακμακοπούλου, Δήμητρα; Tsakmakopoulou, Dimitra
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
Η εφαρμογή της υπολογιστικής όρασης στην οδήγηση είναι ένα από τα επιστημονικά πεδία που αναπτύσσονται με γοργό ρυθμό τα τελευταία χρόνια. Η ανάπτυξη αυτόνομων οχημάτων, ικανών να αντιληφθούν το περιβάλλον τους, να ξεχωρίσουν αντικείμενα, να λάβουν αποφάσεις για τις επόμενες κινήσεις τους και να προγραμματίσουν τη βέλτιστη και πιο ασφαλή διαδρομή, υπόσχεται μείωση των τροχαίων ατυχημάτων που μαστίζουν τη σύγχρονη κοινωνία. Εκτός από τον παράγοντα της ασφάλειας, τα αυτόνομα οχήματα μπορούν να συνεισφέρουν σημαντικά στη βελτίωση της ποιότητας ζωής των ανθρώπων. Το μεγαλύτερο μέρος της έρευνας που έχει πραγματοποιηθεί έως τώρα αναφέρεται στη μελέτη σεναρίων με έναν μοναδικό πράκτορα. Ωστόσο, πρόσφατα η δημοσίευση ενός συνόλου δεδομένων μεγάλης κλίμακας που εμπεριέχει δεδομένα για την ίδια χρονική στιγμή για ένα δεδομένο σενάριο από πολλαπλούς πράκτορες, έχει παρακινήσει πολλούς ερευνητές να εξετάσουν το ενδεχόμενο επικοινωνίας μεταξύ των αυτόνομων οχημάτων. Η παρούσα εργασία εστιάζει στα δεδομένα που μπορούν να ανταλλάξουν μεταξύ τους τα συνδεδεμένα αυτόνομα οχήματα σε πραγματικό χρόνο και μελετά μεθόδους που συγκεντρώνουν και αξιοποιούν την πληροφορία αυτή στοχεύοντας στην ανίχνευση αντικειμένων, συγκεκριμένα άλλων οχημάτων, στο περιβάλλον τους. Το σύνολο δεδομένων, που χρησιμοποιείται στη συγκεκριμένη εργασία, συλλέγεται από αισθητήρες LiDAR που είναι τοποθετημένοι στην οροφή των οχημάτων και επιστρέφουν 3D δεδομένα που έχουν τη μορφή νέφους σημείων. Οι υλοποιήσεις που μελετώνται βασίζονται σε βαθιά νευρωνικά δίκτυα, καθώς φαίνεται ότι είναι πλέον η βέλτιστη επιλογή για εργασίες σχετικές με την ανίχνευση δεδομένων. Συγκεκριμένα, οι αρχιτεκτονικές που αναπτύσσονται εκμεταλλεύονται τον αλγόριθμο Point Pillars, γνωστός για την ακρίβεια και ταχύτητα που παρέχει, και υλοποιούν μία από τις state-of-the-art μεθόδους για τη συγκέντρωση δεδομένων. Πλήθος πειραμάτων πραγματοποιείται για την εύρεση των κατάλληλων χαρακτηριστικών και παραμέτρων του μοντέλου που θα αποδώσουν τη μέγιστη ακρίβεια. Ακόμα, επειδή τα σύνολα δεδομένων στα οποία εκπαιδεύονται συνήθως τα μοντέλα αντιπροσωπεύουν πιο ιδανικές καταστάσεις από τη πραγματικότητα σημαντική είναι η αξιολόγηση τους σε δεδομένα με θόρυβο. Συνεπώς, εξετάζεται το ενδεχόμενο εισαγωγής θορύβου στα δεδομένα λόγω του GPS αισθητήρα που διαθέτει το κάθε όχημα και η αντοχή του μοντέλου σε δυσμενείς καιρικές συνθήκες, συγκεκριμένα στο φαινόμενο της ομίχλης.
Open Access
3Δ ανίχνευση αντικειμένων σε νέφη σημείων με χρήση βαθιάς γεωμετρικής μάθησης
Ρωμανέλης, Ιωάννης; Romanelis, Ioannis
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
Η 3D ανίχνευση αντικείμενων βρίσκει εφαρμογή σε προβλήματα που κυμαίνονται από την αυτόνομη οδήγηση έως τη ρομποτική όραση. Σε αυτά τα προβλήματα απαιτείται πολύ μεγάλη ακρίβεια, καθώς ένα λάθος μπορεί να έχει σοβαρές συνέπειες. Για το λόγο αυτό έχουν αναπτυχθεί πιο σύνθετοι αισθητήρες, όπως το LIDAR (Light Detection And Ranging), για την καλύτερη αναγνώριση του περιβάλλοντος. Στην εργασία αυτή μελετάται το πρόβλημα της 3D ανίχνευσης αντικειμένων σε σκηνές εξωτερικού χώρου, ως μέρος του γενικότερου προβλήματος της αυτόνομης οδήγησης. Για να δημιουργηθούν τα δεδομένα ένας LIDAR αισθητήρας τοποθετείται στην οροφή του οχήματος, το οποίος επιστρέφει ένα νέφος σημείων που περιγράφει το χώρο 360 μοίρες γύρω από αυτό. ́Οπως και στην 2D ανίχνευση δεδομένων έτσι και στην 3D οι αλγόριθμοι που έχουν ξεχωρίσει για αυτό το πρόβλημα είναι αυτοί βασισμένοι σε βαθιά νευρωνικά δίκτυα. Το πρόβλημα της γενίκευσης των δικτύων στα νέφη σημείων έγκειται στο πως θα δεχτούν τα νέα δεδομένα, καθώς σε αντίθεση με την εικόνα όπου τα εικονοστοιχεία είναι διατεταγμένα σε ένα πλέγμα, τα σημεία βρίσκονται στο συνεχή χώρο. Για να υπερκεραστεί αυτό το πρόβλημα γίνεται χρήση της αναπαράστασης των αραιών voxel, καθώς με αυτά μπορούν να δημιουργηθούν δίκτυα εκτελέσιμα σε πραγματικό χρόνο, διατηρώντας μικρές απαιτήσεις μνήμης. Το πρώτο δίκτυο που μελετάται αποτελεί επανυλοποίηση ενός από τα σημαντικότερα δίκτυα που έχουν προταθεί για το πρόβλημα της 3D ανίχνευσης. Το δίκτυο αυτό χρησιμοποιεί ανά σημείο πλήρως συνδεδεμένα επίπεδα για να εξάγει τα χαρακτηριστικά των voxel, το μέρος αυτό της αρχιτεκτονικής καλείται Voxel Feature Extractor (VFE). Τα voxel αυτά εισέρχονται σε μία σειρά από αραιά συνελικτικά επίπεδα για να εξαχθούν τα 3D χαρακτηριστικά, τα οποία γίνονται προβολή σε Bird’s Eye View (BEV) εικόνα και ένας 2D συνελικτικό δίκτυο εξάγει τις τελικές προβλέψεις. Η αρχιτεκτονική αυτή δοκιμάζεται για δεδομένα εισόδου σε καρτεσιανές αλλά και κυλινδρικές συντεταγμένες. Στη δεύτερη αρχιτεκτονική το δίκτυο διαμορφώνεται ώστε να αφαιρεθεί ο VFE, χρησιμοποιώντας απευθείας αραιές συνελίξεις στο κβαντισμένο νέφος. Για την αρχιτεκτονική αυτή γίνονται, επίσης, εκτεταμένα πειράματα ως προς τα δεδομένα εισόδου της αλλά και τον αλγόριθμο βελτιστοποίησής της. Το υλοποιημένο δίκτυο εκπαιδεύτηκε στο σύνολο δεδομένων KITTI. Για την αξιο- λόγηση του έγινε διαχωρισμός του συνόλου σε δύο υποσύνολα, ένα για εκπαίδευσης και ένα για αξιολόγηση των αποτελεσμάτων. Το τελικό δίκτυο καταφέρνει να εξάγει αποτελέσματα συγκρίσιμα με τις state-of-the-art μεθόδους.
Open Access
A containerized monitoring toolkit to measure performance in 5G/NFV networks
(2022-10-12) Σαμαράς, Κωνσταντίνος; Samaras, Konstantinos
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
Nowadays network connectivity touches almost every aspect of our daily lives which means the use of mobile data keeps growing thus creating the need for a more sophisticated mobile technology, 5G. The next generation of wireless technology, 5G is based on a cloud-native core (5GC), can utilize virtualization technologies and follow a Service-based architecture thus taking advantage of all their benefits. However, even on 5G networks, in order to fulfill the Quality of Service (QoS) requirements, it is crucial for network providers to monitor various points in the network and detect any faults. Testing network performance and collecting metrics helps a service provider to define network resource utilization, predict future traffic demands, identify abnormal traffic patterns, and offer suggestions for enhancing network performance. Network administrators traditionally use monitoring tools to measure and analyze the network, for instance a tool that uses client-server approach. Moreover, one of 5G’s key technology is virtualization and while hypervisor-based virtualization is the prevailing technology, the environment is altering recently, as a new form of virtualization that has existed for a long-time gains popularity, called containerization. Since enterprises grow more and more their adoption of Cloud Computing solutions, they also move towards containerized technologies and microservices concepts because of the wide range of benefits they offer, as better uptime, faster deployments, better hardware utilization and lower costs are only a few of them. The purpose of this thesis is to describe the basic concepts of these technologies, then create a containerized and orchestrated toolkit, with which we could monitor and make measurements throughout a 5G container-centric Cloud infrastructure to test the network performance and collect metrics. To achieve that, Docker and Kubernetes as the container technologies are used, along with a set of free open-source monitoring tools like Iperf3 and Ookla Speedtest, resulting in a lightweight and easily configurable implementation.
Open Access
A decentralized API registration and discovery system on the Ethereum blockchain utilized for smart cities
(2023-03-02) Παππάς, Θωμάς; Pappas, Thomas
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
The rapid emergence of the Internet of Things (IoT) technology has prospered the concept of smart city, which refers to an urban area that facilitates the coexistence of people, organizations, environment and technology. IoT plays a major role, as it enables the sensing of physical events that occur, along with actuating on various objects in a city, thus granting valuable information and services for the citizens. Usually, the architecture of a smart city IoT implementation can be conceptualized in five distinct layers. The realization of IoT in smart cities faces a significant obstacle in the form of interoperability challenges inside and between these five layers. Interoperability denotes the capacity of various systems to communicate and exchange data smoothly, without requiring manual intervention. The absence of interoperability between IoT devices and software systems can lead to information silos, wherein data cannot be accessed or utilized by other systems. This thesis focuses on the discoverability of Layer 3 platform systems by Layer 4 applications, without prior knowledge of them, in a complex city scenario. Layer 3 platforms abstract the sensor and networking layers from the end-user applications and are responsible for the registration, discovery and management of devices (sensors, actuators), security, semantics and communication management with other software systems. Generally, platforms expose RESTful northbound APIs, so applications can onboard and receive data and functionality. The proliferation of these APIs is a prominent issue and a common place for the accumulation of the API specifications of the platforms is necessary. After examining the importance of transparency of operations in a collaborative environment such as a smart city, the utilization of Peer-to-Peer systems is unfolded. Taking into consideration the advantages, but also the limitations of blockchain networks, the solution proposed is a decentralized API Directory that can be deployed in any Ethereum Virtual Machine (EVM) compatible public or private blockchain. This API Directory can be used as a common place for smart-city IoT platforms to publish their API specifications, formulated on the OpenAPI specification standard, and for end-user applications to discover them based on location and desired category. Finally, a smart parking use case is demonstrated as a proof of concept for the system.
Open Access
A low-power 4.5–7.5-GHz low noise amplifier–phase shifter design
Βερροιόπουλος, Δημήτριος-Στυλιανός; Verroiopoulos, Dimitris
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
This thesis presents a low-power RF receiver (Rx) front end that incorporates a low-power Low Noise Ampli er (LNA) and a passive reflection-type Phase Shifter (PS) capable of continuous phase shift, operating in the 4.5–7.5-GHz band. The LNA topology comprises of a cascode stage that increases ampli er gain, while keeping DC power consumption and noise figure (NF) low, followed by an emitter follower which has impedance matching capabilities. The PS consists of two cascaded 3-dB quadrature hybrid couplers with variable LC loads, including, but not limited to varactor diodes. The proposed reflection-type PS load design leads to 360° phase shift range and a maximum insertion loss variation of ±3:3 dB. The LNA consumes 9.6 mW and the overall Rx front end has a gain of 16:25 ± 3:75 dB and noise figure lower than 1.78 dB. The printed circuit board has an area of 21.76 sq cm = 9.42 cm × 2.31 cm.
Open Access
Accelerating VGG neural network in GPU and FPGA technologies
Χριστόπουλος, Στέφανος; Christopoulos, Stefanos
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
In recent years, Neural Networks have come to the forefront by eloquently solving long-standing problems. Deep Learning has made possible computational models, composed of multiple processing layers, to learn representations of data with multiple levels of abstraction. This in turn, has led to advances in Computer Vision, Speech Recognition, Machine Translation and many other fields such as Cancer Research. Due to their ability to solve real-world problems, Neural Networks have countless applications and are starting to become part of every-day life. Web searches, music/video recommendations, mobile phone camera processing and even semi-autonomous driving vehicles run with the help of Neural Networks. By nature, Neural Networks require little engineering effort. Instead, they mostly rely on computational power and vast amounts of data which appears to be a driving factor in their success in our modern digital world. Unfortunately, as Moore’s Law seems to be slowing down, it becomes increasingly important to write software that can utilize the underlying hardware to its maximum but also design hardware with software in mind. To this end, in the current thesis, I set to maximally use the available resources of the GPU for the purpose of training Deep Neural Networks. Furthermore, I seek to demonstrate the feasibility of running large Neural Networks on FPGAs that can be used as edge devices. With this goal in mind, I design a Convolution Accelerator for FPGA devices. I apply GPU software optimization strategies to develop custom CUDA kernels for various Neural Network layers and use them to accelerate the VGG16 Neural Network. By using the Winograd Convolution algorithm, optimizing Matrix Multiplication on GPUs and reducing memory bandwidth requirements by fusing layers, I achieve an 8% reduction in execution time over Nvidia’s cuDNN library during Inference and 30% during training of VGG16. In addition, my implementation surpasses Google’s Tensorflow performance by more than 40% during training. In addition, using Xilinx’s Vitis HLS tool, I implement 2 designs with different quantizations of the Winograd Convolution algorithm for the convolutional layers of VGG16 during Inference. Performance results show the FPGA design being 33% more energy efficient than the GPU while achieving sufficiently low execution times to be used as an edge device for Neural Network Inference.
Open Access
Advanced control strategies for collaborative, multiple satellite docking
(2023-02-20) Φούρλας, Φραγκίσκος; Fourlas, Fragkiskos
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
With the ever growing number of space debris in orbit, the need to prevent further space polution is becoming more and more apparent. Refueling, servicing, inspections and deorbiting of spacecraft are some examples of On-Orbit Servicing (OOS) operations that require precise navigation and docking in space. Using multiple, collaborating robots to handle these tasks can greatly increase the efficiency of the mission in terms of time and cost. This thesis will introduce a modern and efficient control architecture for satellites on collaborative docking missions. The proposed architecture uses a centralized scheme that combines state-of-the-art, ad-hoc implementations of algorithms and techniques to maximize robustness and flexibility. It is based on a Non-linear Model Predictive Controller (NMPC) for which, an efficient cost function and constraint sets are designed to ensure a safe and accurate docking. A simulation environment is also presented to validate and test the proposed control scheme. The controller is also tested in a real-world environment using a 2D floating satellite platform.
Open Access
Algorithms and hardware designs for detection in massive MIMO
Βορδώνης, Δημήτριος; Παλιούρας, Βασίλειος; Μπίρμπας, Μιχαήλ; Vodonis, Dimitrios
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
The increasing demand for higher data rates and for more connected devices has led to Massive MIMO Technology. Having a high number of antennas is the basic feature of Massive MIMO networks and this is what differentiates them from conventional MIMO systems where users are served from base-stations with a small number of antennas. Massive MIMO technology has many benefits such as high link reliability, high throughput rate, high spectral efficiency and for this reason it will play a crucial role at 5G networks in the future. However due to large number of antennas, Massive MIMO systems are known for their cost and for their high computational complexity. Maximum-Likelihood (ML) detector yields the optimal solution in MIMO detec- tion problem but its complexity is susceptible to high-order modulation schemes and to large-scale systems. Sphere Decoder (SD) provides a good bit error rate (BER) performance similar to ML detector with the advantage that the complexity of SD is polynomial in contrast with ML-detector’s complexity which rises exponentially with the number of transmit and receive antennas. SD intends to detect the transmitted sig- nal vector by searching only the candidate vectors which lie inside a hypersphere with radius R 0 around the received signal r in contrast with ML detector where all candidate vectors are examined. The method employed to calculate IR has a critical impact on the complexity of SD. A large initial value of sphere radius can lead to an exhaustive search between numerous possible transmitted symbols, while in contrast a very small radius may contain no lattice point inside the sphere and the search will have to be restarted with a new estimation of IR. In this thesis, an implementation of a SD in a Massive MIMO network is pre- sented. This thesis focuses on uplink communication where users transmit data to the base-station. Furthermore a comparison between SD and other detectors, proposed in the literature, is presented based on BER performance using MATLAB simulations. Various methods of calculating IR are evaluated in this thesis considering the compu- tational complexity, the latency, the required wordlength for hardware implementation and the number of visited nodes during detection, which implies execution time and power consumption of the method. Implementation results of these methods are also evaluated. An effective method for calculating IR is proposed. Finally, a hardware implementation of SD targeting FPGA designs is recommended.
Open Access
Algorithms and VLSI architectures for detection in massive MIMO systems for 5G
Κομποστιώτης, Δημήτριος; Kompostiotis, Dimitrios
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
The increasing demand for higher data rates and for more connected devices has led to the adoption of Massive MIMO Technologies for transmission in wireless channels. Having a high number of transmit and receive antennas is the basic feature of Massive MIMO networks and this is what differentiates them from conventional MIMO systems where users communicate using base stations with a small number of antennas. Massive MIMO technology has many benefits such as high link reliability, high throughput rate, high spectral efficiency and therefore it plays a crucial role at 5G networks. However due to the large number of antennas, Massive MIMO systems are known for their cost and for their high computational complexity. The complexity of optimum detectors such as MAP and ML increases fast with the number of antennas, which makes their implementation and application in networks practically impossible. Expectation Propagation (EP) algorithm provides a good bit error rate (BER) performance similar to linear algorithms when 𝑁 𝑅 ≫ 𝑁 𝑇 , where linear algorithms behave as the optimals. Also, EP seems to maintain this good behavior with low BERs compared to linear detectors, even when the number of transmission antennas is about the same as the number of receiving antennas. Moreover, the complexity of EP algorithm is polynomial in contrast with the complexity of MAP and ML detectors complexity, which rises exponentially with the number of transmit and receive antennas. EP is an iterative algorithm that attempts to minimize the Kullback Leibler divergence and approach the probability density function used for the MAP criterion via Gaussian distributions. Specifically, in the iterative section, an attempt is made to approximate the average value of the distribution, that we are interested in, because the average, or otherwise, the expected value of the distribution will be the estimate for the symbol that was sent. In this thesis, an implementation of an EP in a Massive MIMO network is presented. This thesis focuses on uplink communication where users transmit data to the base station. Furthermore a comparison between EP and other detectors, proposed in the literature, is presented based on BER performance using MATLAB simulations. Various methods of calculating the inverse of a matrix are used in this thesis considering the computational complexity, the latency and the required word length for hardware implementation. Implementation results of these methods are also evaluated. Finally, some hardware implementations of EP targeting FPGA designs are recommended.
Open Access
An application specific instruction set processor for cryptographic applications based on RISC-V architecture
Θεοδωρόπουλος, Νικόλαος; Theodoropoulos, Nikolaos
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
Nowadays, RISC-V instruction set architecture (ISA) is becoming increasingly popular in commercial CPU implementations. The benefits introduced by its open-source and simple design along with its ability to be customizable, make it a strong candidate, especially for inexpensive, low-power embedded devices. Implementations used in IoT applications can reap the benefits from an instruction set architecture like RISC-V, as energy efficiency is of paramount importance. Security and privacy of IoT networks have also been an ongoing debate and the most sought-after solutions are those achieving it with minimum performance requirements. There have been proposals that cryptographic hash functions and Blockchain technology could be parts of a potential solution to these problems. In this thesis, a RISC-V processor is presented and extended to make it suitable for cryptographic applications. Firstly, the RISC-V concept and architecture are introduced. Secondly, VexRiscv, the RISC-V core selected to host the cryptographic extensions, is studied. Subsequently, the cryptographic algorithm associated with this thesis and the technique used to accelerate its execution are thoroughly examined, with emphasis on details regarding the implementation. Finally, the design is implemented on a Field Programmable Gate Array, where its proper functioning is verified, initially via simulation and then through waveforms provided by an Integrated Logic Analyzer.
Open Access
An IoT edge-as-a-service (Eaas) distributed architecture & reference implementation
Λαμτζίδης, Οδυσσέας; Γιαλελής, Γιάννης; Γιαλελής, Γιάννης; Δενάζης, Σπύρος; Lamtzidis, Odysseas
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
Internet of Things (IoT) is an enabling technology for numerous domains worldwide, such as smart cities, manufacturing, logistics and critical infrastructure. On top of Internet of Things, the architectural paradigm shifts from cloud-centric to Edge-centric, offloading more and more functionality from the cloud to the Edge devices. Edge computing devices are transformed from simply aggregating data to performing data processing and decision making, accelerating the decentralization of the IoT domain. Given the considerable increase of the number of IoT devices and the size of the generated data, an increase of Edge devices with augmented functionality is expected. We propose an ”Edge as a service” scheme, where Edge devices will be able to procure unused resources and run services that are requested and consumed by IoT devices that belong to different stakeholders. In this work, we outline a high-level architecture of this scheme and give a reference implementation of a narrow part of the system, boasting high modularity and the extensive use of Open Source Technologies.
Open Access
Analysis of the LiDAR sensor for autonomous vehicle
(2023-03-01) Πολίτης, Παύλος; Politis, Pavlos
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
The constantly evolving automobile industry has reached the point of perfecting the design of the car with all those materials and alloys that protect the driver from accidents. However, motor vehicle accidents and crashes continue to cause millions of injuries and deaths annually on a global scale. Thus, it is demonstrated that in addition to the correct design for the safety of a car, errors resulting from the human factor must be reduced. Such factors include distraction, speeding on low speed marked roads and in bad weather conditions. Because of these, global automotive industry has been called upon to solve this problem by using technologies that give the car additional automated functions to avoid as many accidents as possible. Through the use of sensors and the combination of their basic characteristics, the optimal result for autonomous driving is obtained. Most basic of all is the lidar sensor, whose analysis we rely on because it tends to be the future in self-driving vehicles. This thesis will analyze in depth and breadth all the concepts related to the autonomous vehicle, while special emphasis will be placed on the lidar sensor. This sensor will be divided into categories and details will be given for each sub-category, followed by a comparison between the most dominant technologies and techniques. To make the work even more comprehensive, the sensors available on the market will be compared with each other and the features that make some patents stand out from all the rest. The title of the thesis is "Analysis of the LiDAR for autonomous vehicle" and for the scientific documentation is presented in the following chapters: Chapter 1: "What is an autonomous car" describes what characteristics make a vehicle autonomous. The levels given by SAE to separate and categorize a car's autonomous capabilities are also given. Chapter 2: "Autonomous Vehicle Sensors" lists the sensors used in the self-driving vehicle and the characteristics of each. Metrics and their requirements are given for a sensor to be useful and safe in the automotive environment. Chapter 3: “LiDAR” is the first chapter we focus on the LiDAR sensor which will be analyzed in depth in this thesis. Its uses in various fields are presented, one of which is autonomous driving, which interests us the most. A quick reference is also made to the first category that LiDARs can be divided into. Chapter 4: "The 4 fundamental choices in technologies and techniques of 3D LiDAR" presents with details and comparative tables the two techniques of LiDAR, the measurement process and the beam steering mechanism and then the two technologies of the 2 basic elements of the LiDAR sensor, the laser and the photodetector. Chapter 5: "Automotive and Industrial LiDAR systems" lists the patents and products of the leading LiDAR companies and finally a comparison table between them.
Open Access
Application specific instruction set processor based on RISC-V architectures, for state estimation algorithm on smart grids
Θεοδωρακόπουλος, Βασίλης; Theodorakopoulos, Vasilis
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
This work experimented with using a custom instruction set extension for performing the multiply and accumulate instruction which is critical for the Transient State Estimation algorithm. Custom modules were designed and implemented on a RISC-V processor core. The open- source nature of RISC-V makes it an ideal base to implement the extra hardware required to handle those specific workloads. The aim being to both speed up the computation and to reduce transmission data. The central strategy is to use a small core with specialized acceleration instead of a bigger core with excess features, in order to meet the project’s potential future of fitting a processor on the same die as other electronics. An attempt was also made to load the final design on the RISC-V core on PULPissimo . Emphasis is placed on System Verilog modules of multiply and accumulate, behavioral simulation of the designs and then on the interconnections of the designed modules and the core modules. It was found that the precision of the results are marginally within acceptable limits. Precision measurements were done against random data generated from C code. By measuring the cycle count of computations, it was possible to compare the speed of different implementations. In this case it was found that with the new modifications the time to compute the desired vector is in fact reduced in most implementations .It was found that with the timing paths introduced by the new modules the maximum frequency is reduced .
Open Access
Automated image analysis for mouse brain lesions assessment
Δαμίγος, Γεράσιμος; Damigos, Gerasimos
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
Παρά την ύπαρξη πληθώρας τεχνικών επεξεργασίας και ανάλυσης βιοϊατρικών δεδομένων, δεν έχει αναπτυχθεί ακόμη ένα αμερόληπτο, γρήγορο και αυτόματο πλαίσιο για ανάλυση εγκεφαλικών βλαβών σε τρωκτικά μετα απο την πραγματοποιήση ισχαιμικών εγκεφαλικών. Συγκεκριμένα, η ογκομετρική μέτρηση εγκεφαλικών πραγματοποιείται είτε χειροκίνητα (με οπτική εκτίμηση) από ειδικούς, μια μέθοδος η οποία είναι χρονοβόρα και ευαίσθητη στο ανθρώπινο λάθος, είτε χρησιμοποιώντας ημι-αυτοματοποιημενες προσεγγίσεις. Σε αυτή την εργασία αναπτύξαμε μια καινούργια μέθοδο ("Stroke Analyst") για εξ ολοκλήρου αυτοματοποιημένη ανάλυση του οιδήματος του εμφράκτου σε τομές εγκεφάλου βαμμένες σε 2% Triphenyltetrazolium chloride (2% TTC), βασισμένη σε τεχνικές εντοπισμού ακραίων τιμών (outlier detection) και μηχανικής μάθησης. Η μεθοδολογία μας είναι βασισμένη στον υπολογισμό ενός στατιστικού μοντέλου (Άτλαντα) όσον αφορά τη φωτεινότητα του εκάστοτε εικονοστοιχείο (pixel) σε κάθε τομή του εγκεφάλου. Το στατιστικό αυτό μοντέλο έχει δημιουργηθεί έπειτα από ελαστική ευθυγράμμιση υγιών τομών εγκεφάλου. Για τον εντοπισμό της περιοχής της βλάβης σε νέες εικόνες με ισχαιμικό εγκεφαλικό πραγματοποιείται πρώτα αυτόματη αναγνώριση ιστού και υποβάθρου και ευθυγράμμιση των τομών εγκεφάλου με τις αντίστοιχες του Άτλαντα. Από τη διαδικασία αυτή παράγεται ενας πιθανοτικός χάρτης ανωμαλιών, δηλαδή αποκλίνουσων (ακραίων) τιμών από τη φυσιολογική/υγιή βάση (Άτλαντα). Στη συνέχεια ακολουθεί επεξεργασία και φιλτράρισμα του χάρτη με μεθόδους μηχανικής μάθησης με στόχο τη μείωση των ψευδώς θετικών σημείων βελτιστοποιώντας τη μεθοδολογία αυτόματου εντοπισμού, αναγνώρισης και μέτρησης των περιοχών οι οποίες έχουν πληγεί από το εγκεφαλικό. Το παραγόμενο λογισμικό επικυρώθηκε σε 80 εργαστηριακά δεδομένα από διαφορετικούς χρήστες. Τέλος, η υλοποίηση όλων των βημάτων της μεθοδολογίας σε γραφικό περιβάλλον με δυνατότητα ρύθμισης παραμέτρων παρέχει τη δυνατότητα επέμβασης και βελτίωσης των επιμέρους βημάτων, ενώ η δημιουργία κατάλληλης διεπαφής χρήστη (user interface) ευελπιστούμε ότι θα διευκολύνει τη χρήση ακόμα και από τεχνικώς μη-εξειδικευμένους κλινικούς οδηγώντας στην ευρεία διάδοση της προτεινόμενης μεθοδολογίας.
Open Access
Automatic kidney segmentation, reconstruction, preoperative planning and 3D printing
Ζάγκου, Σπυριδούλα; Zagkou, Spyridoula
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
Renal cancer is the seventh most prevalent cancer among men and the tenth most frequent cancer among women, accounting for 5% and 3% of all adult malignancies, respectively. Κidney cancer is increasing dramatically in developing countries due to inadequate living conditions but and in developed countries due to bad lifestyles, smoking, obesity, and hypertension. For decades, radical nephrectomy (RN) was the standard method to address the problem of the high incidence of kidney cancer. However, the utilization of minimally invasive partial nephrectomy (PN), for the treatment of localized small renal masses has increased with the advent of laparoscopic and robotic-assisted procedures. In this framework, certain factors must be considered in surgical planning and decision-making of partial nephrectomies, such as the morphology and location of the tumor. Advanced technologies such as automatic image segmentation, image and surface reconstruction, and 3D printing have been developed to assess the tumor anatomy before surgery and its relationship to surrounding structures, such as the arteriovenous system, with the aim of preventing damage. Overall, it is obvious that 3D printed anatomical kidney models are very useful to urologists, surgeons, and researchers as a reference point for preoperative planning and intraoperative visualization for a more efficient treatment and a high standard of care. Furthermore, they can provide a lot of degrees of comfort in education, in patient counseling, and in delivering therapeutic methods customized to the needs of each individual patient. To this context, the fundamental objective of this thesis is to provide an analytical and general pipeline for the generation of a renal 3D printed model from CT images. In addition, there are proposed methods to enhance preoperative planning and help surgeons to prepare with increased accuracy the surgical procedure so that improve their performance.
Open Access
Autonomous aerial indoor navigation of quadcopter
Πολυκράτης, Δημήτριος; Μούρτζης, Δημήτριος; Κούσουλας, Νικόλαος; Polykratis, Dimitrios
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
This thesis addresses the design and development of a system for the autonomous navigation of a quadcopter vehicle in an indoor, GPS-denied, industrial environment, potentially performing logistics operations. To achieve autonomous navigation, precise localization of the vehicle was essential and it was performed with the combination of two methods. The first made use of a visual Simultaneous Localization and Mapping (SLAM) algorithm, that relies on data of an onboard stereo camera. The second method relies on images from an onboard, bottom-facing monocular camera. Additionally, by employing five ultrasonic proximity sensors, the quadcopter is capable of performing efficient obstacle avoidance. The system has been integrated in the Robot Operating Software (ROS) and implemented in the simulation environment of the Virtual Robot Experimentation Platform (V-REP). Finally, a series of experiments has been conducted to evaluate the developed system’s performance and identify gaps and requirements for future research.
Open Access
Autonomous driving and obstacle avoidance for car-like vehicles
(2022-11-22) Κατσανικάκης, Ανδρέας; Katsanikakis, Andreas
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
Autonomous navigation systems play a significant role in every aspect of the technological development humanity has known recently. A robot or a vehicle has the ability to determine its location and plan a path to some destination without human interference. This statement, by its own, indicates that autonomous technology can derive great scientific results since these systems can be incorporated in projects that a human could not achieve alone up until now. Automobiles, space, health, sports, food and many more industries invest more and more funds daily to develop autonomous systems for their own use. Especially in automobiles, autonomous vehicles can change the entirely operation of the whole road traffic network. The statement of a world that can travel autonomously is now a project that can become reality. The following thesis focuses on simulating a driving car scenario where a vehicle navigates autonomously inside a road scene. Obstacles appear throughout its course, and it is anticipated to avoid them while at the same time not diverging from the road boundaries. Trajectory tracking and collision avoidance techniques are developed and a main algorithm that completes those tasks is proposed. The correct functionality of the algorithm is firstly tested in Simulink/MATLAB software and then the simulation takes place in the robotic-based CoppeliaSim Edu software. The simulation is based mostly on kinematic theory. The scenario, the road scene, the model that has been adopted for the vehicle, and the obstacles that are used in the simulator try to approach a real case perspective. Results are presented in graphs and further discussion is made.
Open Access
Autonomous multi-agent exploration and mapping
(2023-03-01) Μαρκοστάμος, Γεώργιος; Markostamos, Georgios
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
Exploration and Mapping is a fundamental task in robotics as it provides important information about an unknown environment that can be crucial for many applications. Many areas are either too dangerous or infeasible for humans to explore. Therefore ground and aerial robots are often used to perform this task. In the past, the teleoperation of a single robot was the primary method of completing these types of missions. However, the improvements in the computational capability of mobile agents and the development of new algorithms have made it possible for them to operate autonomously and even collaborate to achieve a common goal. This thesis aims to describe the complete algorithmic process required to solve the Autonomous Multi-Agent Exploration and Mapping problem and develop a system capable of successfully performing this task. For this purpose, a distributed planning pipeline that generates and optimizes exploration paths using Genetic Algorithms in real-time is implemented. The pipeline is then integrated into a set of simulated UAVs that are part of a more extensive ROS-based system. This system consists of a high-level controller in the form of Behavior Trees that is responsible for the coordination of the agents and the safe execution of tasks and low-level controllers that enable them to navigate autonomously. To implement all the required functionality, a set of software components were developed, which are publicly available. The overall system is evaluated in simulation environments of different complexities. The experiments indicate that the proposed method can successfully explore and map an unknown environment and that the use of multiple agents is beneficial both in terms of computational efficiency and speed of exploration.
Open Access
Aνάπτυξη αρχιτεκτονικών και υλοποιήσεις σε υλικό του διακριτού μετασχηματισμού κυματιδίων για το πρότυπο JPEG XS
Πετρούτσος, Παναγιώτης; Θεοδωρίδης, Γιώργος; Κουφοπαύλου, Οδυσσέας; Petroutsos, Panagiotis
Τμήμα Ηλεκτρολ. Μηχαν. και Τεχνολ. Υπολογ. (ΔΕ)
Το θέμα το οποίο πραγματεύεται η συγκεκριμένη εργασία επικεντρώνεται στην μελέτη του προτύπου JPEG XS και στη ψηφιακή υλοποίηση μέρος του προτύπου. Αρχικά, στην εργασία εξηγείται η σημασία της συμπίεσης δεδομένων, παρουσιάζοντας τα πλεονεκτήματα που φέρει. Μια μορφή συμπίεσης δεδομένων αποτελεί το πρότυπο JPEG XS, το οποίο αναπτύχθηκε ειδικά για την κωδικοποίηση των δεδομένων της εικόνας. Στη συνέχεια, ακολουθεί αναφορά στο πρότυπο JPEG XS. Παρουσιάζονται τα στάδια ανάπτυξης του προτύπου, η αναγκαιότητα που αναπτύχθηκε καθώς οι μηχανισμοί και αλγόριθμοι που αξιοποιεί. Παράλληλα, η εργασία εστιάζει στη συνοπτική παρουσίαση του ευθύ διακριτού μετασχηματισμού κυματιδίων (Discrete Wavelet Transform). Ο μετασχηματισμός κυματιδίων αποτελεί ένα μηχανισμό που εκμεταλλεύεται το πρότυπο JPEG XS . Επίσης, στην εργασία εξηγείται ο τρόπος προσέγγισης και υλοποίησης μιας ψηφιακής σχεδίασης. Προτείνεται μια αρχιτεκτονική του μετασχηματισμού Discrete Wavelet Transform σε ψηφιακή υλοποίηση. Στόχος της υλοποίησης αποτελεί η επεξεργασία κάθε εικονοστοιχείου σε ένα κύκλο ρολογιού. Η σχεδίαση υλοποιήθηκε σε τεχνολογίες FPGA, υποστηρίζοντας διαφορετικές αναλύσεις εικόνων. Με την υλοποίηση μετρήθηκαν οι καταναλισκόμενοι πόροι υλικού και επιτεύξιμες συχνότητες λειτουργίας. Τέλος, η εργασία ολοκληρώνεται με την προσομοίωση της σχεδίασης, την αξιολόγηση των αποτελεσμάτων, την ανάλυση συμπερασμάτων καθώς και βελτιώσεις που μπορούν να επιτευχθούν.