• Smart Structures and Systems
• /
• v.31 no.4
• /
• pp.335-349
• /
• 2023

Bridges constantly undergo deterioration and damage, the most common ones being concrete damage and exposed rebar. Periodic inspection of bridges to identify damages can aid in their quick remediation. Likewise, identifying components can provide context for damage assessment and help gauge a bridge's state of interaction with its surroundings. Current inspection techniques rely on manual site visits, which can be time-consuming and costly. More recently, robotic inspection assisted by autonomous data analytics based on Computer Vision (CV) and Artificial Intelligence (AI) has been viewed as a suitable alternative to manual inspection because of its efficiency and accuracy. To aid research in this avenue, this study performs a comparative assessment of different architectures, loss functions, and ensembling strategies for the autonomous segmentation of bridge components and damages. The experiments lead to several interesting discoveries. Nested Reg-UNet architecture is found to outperform five other state-of-the-art architectures in both damage and component segmentation tasks. The architecture is built by combining a Nested UNet style dense configuration with a pretrained RegNet encoder. In terms of the mean Intersection over Union (mIoU) metric, the Nested Reg-UNet architecture provides an improvement of 2.86% on the damage segmentation task and 1.66% on the component segmentation task compared to the state-of-the-art UNet architecture. Furthermore, it is demonstrated that incorporating the Lovasz-Softmax loss function to counter class imbalance can boost performance by 3.44% in the component segmentation task over the most employed alternative, weighted Cross Entropy (wCE). Finally, weighted softmax ensembling is found to be quite effective when used synchronously with the Nested Reg-UNet architecture by providing mIoU improvement of 0.74% in the component segmentation task and 1.14% in the damage segmentation task over a single-architecture baseline. Overall, the best mIoU of 92.50% for the component segmentation task and 84.19% for the damage segmentation task validate the feasibility of these techniques for autonomous bridge component and damage segmentation using RGB images.

• Proceedings of the Korean Information Science Society Conference
• /
• 2012.06a
• /
• pp.170-172
• /
• 2012

실시간 시스템(real-time system)은 논리적 연산을 일정한 시간적 제약 하에서 수행하는 시스템이다. 시간적 제약을 충족하도록 주기적 태스크(periodic task)를 스케줄(schedule)할 때 일반적으로 태스크 오프셋(initial offset)이 0 이거나 고정된 것으로 가정한다. 그러나 오프셋에 약간의 유연성을 허용함으로써 태스크들의 평균 응답 시간을 줄일 수도 있다. 이 논문에서는 주기적 태스크의 오프셋을 주어진 허용 범위 안에서 선택하여 평균 응답 시간(response time)을 최적화할 수 있음을 보이고, 임의의 태스크 집합에 대하여 최적 오프셋이 존재하는 좁은 범위를 제시한다.

• Journal of Institute of Control, Robotics and Systems
• /
• v.18 no.4
• /
• pp.337-342
• /
• 2012

This paper presents performance evaluation of real-time mechanisms for real-time embedded linux. First, we presents process for implementing open-source real-time embedded linux namely RTAI and Xenomai. These are real-time extensions to linux kernel and we implemented real-time embedded linux over the latest linux kernel. Measurements of executions of real-time mechanisms for each distribution are performed to give a quantitative comparison. Performance evaluations are conducted in kernel space about repeatability of periodic task, response time of Semaphore, FIFO, Mailbox and Message queue in terms of inter-task communication for each distribution. These rules can be helpful for deciding which real-time linux extension should be used with respect to the requirements of the real-time applications.

• Journal of Korean Institute of Industrial Engineers
• /
• v.23 no.1
• /
• pp.1-22
• /
• 1997

Learning control refers to controllers that learn to improve their performance at executing a given task, based on experience performing this specific task. In a previous work, the authors presented a theory of indirect decentralized learning control based on use of indirect adaptive control concepts employing simultaneous identification and control. This paper extends these results to apply to the indirect repetitive control problem in which a periodic (i.e., repetitive) command is given to a control system. Decentralized indirect repetitive control algorithms are presented that have guaranteed convergence to zero tracking error under very general conditions. The original motivation of the repetitive control and learning control fields was learning in robots doing repetitive tasks such as on an assembly line. This paper starts with decentralized discrete time systems, and progresses to the robot application, modeling the robot as a time varying linear system in the neighborhood of the desired trajectory. Decentralized repetitive control is natural for this application because the feedback control for link rotations is normally implemented in a decentralized manner, treating each link as if it is independent of the other links.

• Proceedings of the KIEE Conference
• /
• 2004.05a
• /
• pp.59-61
• /
• 2004

Researches about scheduling distributed real-time systems have some weak points, not scheduling both sporadic and periodic tasks and messages or being unable to guaranteeing the end-to-end constraints due to omitting precedence relations between sporadic tasks. This paper describes the application model of sporadic tasks with precedence constraints in a distributed real-time system. It is shown that existing scheduling methods such as Rate Monotonic scheduling are not proper to be applied to the system having sporadic tasks with precedence constraints. So this paper proposes an end-to-end laxity-based priority assignment algorithm which considers the practical laxity of a task and allocates a proper priority to a task.

• Proceedings of the KIEE Conference
• /
• 2002.11c
• /
• pp.79-82
• /
• 2002

In this paper, we propose an adaptive repetitive control algorithm for the system the task of which is repetitive. The feedforward controller in the repetitive control system is modified by using the system parameter identifier in order to improve the convergence characteristics. The proposed algorithm is applied to the tracking control of a linear BLDC motor to which a periodic reference input is applied. It is illustrated by simulation results that the proposed adaptive repetitive control method yields better control performance than existing repetitive control even when modeling errors exist.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.15 no.8
• /
• pp.1820-1831
• /
• 2011

When both types of periodic and aperiodic tasks are required to run on a sensor node platform with limited energy resources, we propose an energy-efficient hybrid task scheduling technique that guarantees the deadlines of real-time tasks and provides non-real-time tasks with good average response time. The proposed hybrid task scheduling technique achieved better performance than existing EDF-based DVS scheduling techniques available in the literature, the FIFO-based TinyOS scheduling technique, and the task-clustering based non-preemptive real-time scheduling technique.

• The KIPS Transactions:PartA
• /
• v.17A no.6
• /
• pp.259-264
• /
• 2010

In case of scheduling tasks with shared resources in multiprocessor systems, Global Earliest Deadline First (GEDF) algorithm, equally applied Earliest Deadline First (EDF) which runs scheduling with deadline criterion, makes schedulability decline because GEDF typically does not have a specific process in order to handle tasks with shared resources. In this paper, we propose Earliest Deadline First with Partitioning (EDFP) for tasks with shared resources which partitions a task into two kinds of subtasks that include critical sections to access to shared resources, gives their own deadline respectively and manages them. As a result of simulations, EDFP shows better performance than GEDF for tasks with shared resources since system load goes up and the number of processor increases.

• Journal of KIISE:Computer Systems and Theory
• /
• v.27 no.5
• /
• pp.506-517
• /
• 2000

In distributed real-time system, a task activated by the completion of its preceding task can be modeled as a periodic task with activation jitter. An activation jitter of a task is defined asthe difference between the worst case and the best case response time of its preceding task. Becausethe existing approaches assume that the best case response time is much smaller than the actual one,the activation jitter and the worst case response time of lower priority tasks are overestimated. Thispaper proposes a new analysis technique to calculate the best case response time more precisely andto reduce the activation jitter bounds. The proposed technique obtains the best case response time byconsidering the relative phase between tasks. The precise analysis of the activation jitters can reducethe worst case response time of other tasks and increase the schedulability. The simulation resultsshow that the proposed analysis technique improves the accuracy of the best case and the worst caseresponse time up to 40% and 6%, respectively.

• Journal of Korea Multimedia Society
• /
• v.2 no.4
• /
• pp.378-389
• /
• 1999

Parallel tasks in distributed real-time systems can be divided into several subtasks and be executed in parallel according to their real-time attributes. But, it is difficult to gain the optimal solution which is to allocate a tasks deadline into the subtasks deadline while minimizing the subtasks deadline miss. Tn this Paper, we propose the algorithm that allocates deadlines into each subtask, according to the attributes of each subtask(i.e. using communication time and execution time to periodic tasks). Also, we suggest a processor mapping algorithm that considers the communication time among the processors and the effective duplication algorithm which is allocated to the identical processor for the purpose of improving the communication time between the subtasks. We can obtain a result that reduces IPC(Inter-Processor Communication) time and uses the idle processor through applying effective real-time attributes to FUTD(Fully connected, Unbounded Task Duplication) algorithms. As a result, we can improve the average processor utilization.