Recently, machine learning is widely used to solve optimization problems in various engineering fields. In this study, machine learning is applied to development of a control algorithm for a smart control device for reduction of seismic responses. For this purpose, Deep Q-network (DQN) out of reinforcement learning algorithms was employed to develop control algorithm. A single degree of freedom (SDOF) structure with a smart tuned mass damper (TMD) was used as an example structure. A smart TMD system was composed of MR (magnetorheological) damper instead of passive damper. Reward design of reinforcement learning mainly affects the control performance of the smart TMD. Various hyperparameters were investigated to optimize the control performance of DQN-based control algorithm. Usually, decrease of the time step for numerical simulation is desirable to increase the accuracy of simulation results. However, the numerical simulation results presented that decrease of the time step for reward calculation might decrease the control performance of DQN-based control algorithm. Therefore, a proper time step for reward calculation should be selected in a DQN training process.

자동 관개 시스템에서는 관수를 자동으로 개시하고 중지할 수 있는 기준값의 설정이 중요하다. 관수 기준값은 작물의 종류와 생육 시기, 토성, 용적 밀도 등에 따라 달라지는 포장 용수량의 토양 수분값으로 결정되기 때문에, 전문적인 지식과 분석 경험이 필요하여 현장 농업인이 직접 파악하는 것은 어렵다. 그래서 재배 작물의 명칭, 재배 지역 및 재배 토양의 토성 등을 조건 변수로 하여 적절한 토양 수분값을 데이터베이스로부터 추출하고, 작물의 종류 및 생육 시기별 토양수분 기준을 데이터베이스화하여 선택한 작물에 적합한 토양수분 장력값을 설정할 수 있는 알고리즘을 개발하였다. 이 알고리즘을 센서부, 제어부, 구동부로 구성되어 있는 시스템에 적용하여 토양 수분을 제어할 수 있는 시스템을 개발하였다. 실험구별로 수분 제어 기준값을 설정하여 측정한 수분값이 -33 kPa 실험구에서 부합률 97.3%, -25 kPa 실험구에서 부합률 96.6%의 결과를 나타내었다. 이 시스템을 이용하여 최근 농촌지역의 고령화와 노동인구 감소에 따른 생산성 감소를 억제하는데 기여할 것으로 사료된다.

PURPOSES : In a previous study, an error was detected in data pertaining to the direction and velocity of a roller. Hence, in this study, the correlation between these two variables and acceleration data is analyzed. Relevant algorithms are developed by adding variables to existing algorithms. METHODS : A tachometer and GPS are used to acquire the velocity, compaction direction of rollers, and number of compactions. Subsequently, data input to an accelerometer are compared and analyzed. RESULTS : Based on FFT analysis, it is discovered that the data are inaccurate when a forward reverse is performed. Using the GPS, the velocity of the roller is differentiated based on the number of pledges, and then added as a variable to the algorithm. Subsequently, it is evaluated and analyzed only with data during forward movement based on changes in the latitude and longitude. CONCLUSIONS : It is discovered that the acceleration data values from both the left and right rollers differ owing to their weight difference, as indicated by the asphalt gradient. Data changes based on asphalt gradients are analyzed using gyro sensors. If the correlation between the two sets of data is high, then the algorithm is advanced by introducing a cross spectrum after calibrating the acceleration value based on the gradient.

PURPOSES : This study aims to develop an algorithm to solve the user equilibrium traffic assignment problem using soft link capacity constraints. This model is used to relax the hard capacity constraints model. METHODS : In the traffic assignment model that imposes the hard capacity constraints, the well-known solution algorithms used are the augmented Lagrange multiplier method and the inner penalty function method. The major drawback of using the hard-capacity constraint model is the feasible solution issue. If the capacities in the network are not sufficient to absorb the flow from the diverged flows through the hard capacity constraints, it might result in no solution; whereas, using a soft capacity constraint model guarantees a feasible solution because the soft capacity constraint model uses the penalization of constraint violation in the objective function. In this study, the gradient projection (GP) algorithm was adapted. RESULTS : Two numerical experiments were conducted to demonstrate the features of the soft capacity constraint model and the computational performance of the solution algorithm. The results revealed that imposing the soft link capacity constraints can ensure convergence. CONCLUSIONS : The proposed model can be easily extended by considering other traffic assignment models, for e.g., non-additive traffic equilibrium problem, stochastic traffic equilibrium model, and, elastic demand traffic equilibrium problem. Furthermore, the model can exist regardless of the sufficient capacity for each O-D pair to cater to their demands.

PURPOSES : The driver's ability to make a commitment has resulted in excessive force and a lack of commitment. To solve this problem, we are developing an algorithm that analyzes resolution in real-time by introducing IoT and informs drivers of the completion of compaction. METHODS : Real-time compaction was analyzed by installing accelerometers on the rollers. To evaluate the algorithms, we conducted an apparent density test. RESULTS : The algorithm data and apparent density test data showed similar trends. This means that the proposed algorithms are sufficiently reliable. However, a lack of data samples and the fact that only data prior to completion of the commitment were analyzed may indicate a lack of reliability. CONCLUSIONS : In subsequent studies, the number of samples will be increased and the data after completion of the commitment analyzed to increase reliability. Introducing a tachometer will prevent the TVL from falling sharply when the direction of the rollers' progress changes. In addition, it is also planned to upgrade the algorithms by researching cases in which the algorithms did not produce satisfactory results owing to problems such as temperature and speed.

Control performance of a smart tuned mass damper (TMD) mainly depends on control algorithms. A lot of control strategies have been proposed for semi-active control devices. Recently, machine learning begins to be applied to development of vibration control algorithm. In this study, a reinforcement learning among machine learning techniques was employed to develop a semi-active control algorithm for a smart TMD. The smart TMD was composed of magnetorheological damper in this study. For this purpose, an 11-story building structure with a smart TMD was selected to construct a reinforcement learning environment. A time history analysis of the example structure subject to earthquake excitation was conducted in the reinforcement learning procedure. Deep Q-network (DQN) among various reinforcement learning algorithms was used to make a learning agent. The command voltage sent to the MR damper is determined by the action produced by the DQN. Parametric studies on hyper-parameters of DQN were performed by numerical simulations. After appropriate training iteration of the DQN model with proper hyper-parameters, the DQN model for control of seismic responses of the example structure with smart TMD was developed. The developed DQN model can effectively control smart TMD to reduce seismic responses of the example structure.

본 연구는 화재진압 및 피난활동을 지원하는 딥러닝 기반의 알고리즘 개발에 관한 기초 연구로 선박 화재 시 연기감지기가 작동하기 전에 검출된 연기 데이터를 분석 및 활용하여 원격지까지 연기가 확산 되기 전에 연기 확산거리를 예측하는 것이 목적이다. 다음과 같은 절차에 따라 제안 알고리즘을 검토하였다. 첫 번째 단계로, 딥러닝 기반 객체 검출 알고리즘인 YOLO(You Only Look Once)모델에 화재시뮬레이션을 통하여 얻은 연기 영상을 적용하여 학습을 진행하였다. 학습된 YOLO모델의 mAP(mean Average Precision)은 98.71%로 측정되었으며, 9 FPS(Frames Per Second)의 처리 속도로 연기를 검출하였다. 두 번째 단계로 YOLO로부터 연기 형상이 추출된 경계 상자의 좌표값을 통해 연기 확산거리를 추정하였으며 이를 시계열 예측 알고리즘인 LSTM(Long Short-Term Memory)에 적용하여 학습을 진행하였다. 그 결과, 화재시뮬레이션으로부터 얻은 Fast 화재의 연기영상에서 경계 상자의 좌표값으로부터 추정한 화재발생~30초까지의 연기 확산거리 데이터를 LSTM 학습모델에 입력하여 31초~90초까지의 연기 확산거리 데이터를 예측하였다. 그리고 추정한 연기 확산거리와 예측한 연기 확산거리의 평균제곱근 오차는 2.74로 나타났다.

PURPOSES : This study verifies the appropriateness of the observed traffic volume using car navigation traffic volume data. METHODS : In this study, we developed an annual average daily traffic (AADT) estimation model that can verify the total amount of traffic by using navigation traffic volume data. In addition, a method to verify the appropriateness of the observed traffic volume was developed using time-based navigation traffic volume data that can check the characteristics of traffic volume at each point. RESULTS : As a result of the analysis of this study, it was found that 674 of the 697 short-duration survey spots of the freeways were appropriate and that 23 spots needed to be revised. CONCLUSIONS : As a result of the analysis of this study, it was found that there was a strong positive correlation between the observed traffic volume and the car navigation traffic volume. Thus, the appropriateness of the observed traffic was determined by verifying the total amount of observed traffic and the observed traffic volume by time.

In the study, a shape finding procedure for the tensegrity system model inspired by the movement pattern of animal backbone was presented. The proposed system is allowing a dynamic movement by introducing the concept of “saddle” for the variable tensegrity structure. Mathematical process and an algorithm for movable tensegrity to specified points were established. Several examples have applied with in established shape finding analysis procedure. The final tensegrity structures were determined well to a object shape.

Spot welding is a representative process in automotive welding and the application of intelligent systems is accelerating. In particular, in the case of welding electrode management, the timing of electrode wear and dressing was determined by continuous spot welding evaluation, however there is concerned that errors in welding equipment or processes may work in a complex manner. In this study, a dynamic resistance waveform sensing and image measurement system that greatly affects the nugget formation, which is important to the quality of spot welding, was fabricated and used. Based on the experimental data of the galvanized steel sheet, an electrode life prediction algorithm for electrode wear was derived through CNN(Convolutional Neural Network) model of machine learning training.