• Journal of Korean Tunnelling and Underground Space Association
• /
• v.24 no.4
• /
• pp.327-339
• /
• 2022

The purpose of this study is to present the effective smoke control flow rate and mode for securing safety through quantitative risk assessment according to the smoke control flow rate and mode (supply or exhaust) of the platform when a train fire occurs at the subway platform. To this end, a fire outbreak scenario was created using a side platform with a central staircase as a model and fire analysis was performed for each scenario to compare and analyze fire propagation characteristics and ASET, evacuation analysis was performed to predict the number of deaths. In addition, a fire accident rate (F)/number of deaths (N) diagram (F/N diagram) was prepared for each scenario to compare and evaluate the risk according to the smoke control flow rate and mode. In the ASET analysis of harmful factors, carbon monoxide, temperature, and visible distance determined by performance-oriented design methods and standards for firefighting facilities, the effect of visible distance is the largest, In the case where the delay in entering the platform of the fire train was not taken into account, the ASET was analyzed to be about 800 seconds when the air flow rate was 4 × 833 m³/min. The estimated number of deaths varies greatly depending on the location of the vehicle of fire train, In the case of a fire occurring in a vehicle adjacent to the stairs, it is shown that the increase is up to three times that of the vehicle in the lead. In addition, when the smoke control flow rate increases, the number of fatalities decreases, and the reduction rate of the air supply method rather than the exhaust method increases. When the supply flow rate is 4 × 833 m³/min, the expected number of deaths is reduced to 13% compared to the case where ventilation is not performed. As a result of the risk assessment, it is found that the current social risk assessment criteria are satisfied when smoke control is performed, and the number of deaths is the flow rate 4 × 833 m³/min when smoke control is performed at 29.9 people in 10,000 year, It was analyzed that it decreased to 4.36 people.

• The Journal of the Acoustical Society of Korea
• /
• v.21 no.2
• /
• pp.199-210
• /
• 2002

In this paper, we carried out the study on speech recognition using the KM-Net topology design algorithm based on decision tree state-clustering to improve the performance of acoustic models in speech recognition. The Korean has many allophonic and grammatical rules compared to other languages, so we investigate the allophonic variations, which defined the Korean phonetics, and construct the phoneme question set for phonetic decision tree. The basic idea of the HM-Net topology design algorithm is that it has the basic structure of SSS (Successive State Splitting) algorithm and split again the states of the context-dependent acoustic models pre-constructed. That is, it have generated. the phonetic decision tree using the phoneme question sets each the state of models, and have iteratively trained the state sequence of the context-dependent acoustic models using the PDT-SSS (Phonetic Decision Tree-based SSS) algorithm. To verify the effectiveness of the above algorithm we carried out the speech recognition experiments for 452 words of center for Korean language Engineering (KLE452) and 200 sentences of air flight reservation task (YNU200). Experimental results show that the recognition accuracy has progressively improved according to the number of states variations after perform the splitting of states in the phoneme, word and continuous speech recognition experiments respectively. Through the experiments, we have got the average 71.5%, 99.2% of the phoneme, word recognition accuracy when the state number is 2,000, respectively and the average 91.6% of the continuous speech recognition accuracy when the state number is 800. Also we haute carried out the word recognition experiments using the HTK (HMM Too1kit) which is performed the state tying, compared to share the parameters of the HM-Net topology design algorithm. In word recognition experiments, the HM-Net topology design algorithm has an average of 4.0% higher recognition accuracy than the context-dependent acoustic models generated by the HTK implying the effectiveness of it.

• Journal of Intelligence and Information Systems
• /
• v.18 no.4
• /
• pp.129-143
• /
• 2012

Recently, the rapid progress of a number of standardized web technologies and the proliferation of web users in the world bring an explosive increase of producing and consuming information documents on the web. In addition, most companies have produced, shared, and managed a huge number of information documents that are needed to perform their businesses. They also have discretionally raked, stored and managed a number of web documents published on the web for their business. Along with this increase of information documents that should be managed in the companies, the need of a solution to locate information documents more accurately among a huge number of information sources have increased. In order to satisfy the need of accurate search, the market size of search engine solution market is becoming increasingly expended. The most important functionality among much functionality provided by search engine is to locate accurate information documents from a huge information sources. The major metric to evaluate the accuracy of search engine is relevance that consists of two measures, precision and recall. Precision is thought of as a measure of exactness, that is, what percentage of information considered as true answer are actually such, whereas recall is a measure of completeness, that is, what percentage of true answer are retrieved as such. These two measures can be used differently according to the applied domain. If we need to exhaustively search information such as patent documents and research papers, it is better to increase the recall. On the other hand, when the amount of information is small scale, it is better to increase precision. Most of existing web search engines typically uses a keyword search method that returns web documents including keywords which correspond to search words entered by a user. This method has a virtue of locating all web documents quickly, even though many search words are inputted. However, this method has a fundamental imitation of not considering search intention of a user, thereby retrieving irrelevant results as well as relevant ones. Thus, it takes additional time and effort to set relevant ones out from all results returned by a search engine. That is, keyword search method can increase recall, while it is difficult to locate web documents which a user actually want to find because it does not provide a means of understanding the intention of a user and reflecting it to a progress of searching information. Thus, this research suggests a new method of combining ontology-based search solution with core search functionalities provided by existing search engine solutions. The method enables a search engine to provide optimal search results by inferenceing the search intention of a user. To that end, we build an ontology which contains concepts and relationships among them in a specific domain. The ontology is used to inference synonyms of a set of search keywords inputted by a user, thereby making the search intention of the user reflected into the progress of searching information more actively compared to existing search engines. Based on the proposed method we implement a prototype search system and test the system in the patent domain where we experiment on searching relevant documents associated with a patent. The experiment shows that our system increases the both recall and precision in accuracy and augments the search productivity by using improved user interface that enables a user to interact with our search system effectively. In the future research, we will study a means of validating the better performance of our prototype system by comparing other search engine solution and will extend the applied domain into other domains for searching information such as portal.

• KIPS Transactions on Software and Data Engineering
• /
• v.2 no.8
• /
• pp.555-564
• /
• 2013

Innovative evolution in mobile devices along with recent spread of Tablet PCs and Smart Phones makes a new change not only in individual life but also in enterprise applications. Especially, in the case of medium-large mobile applications for large enterprises which generally takes more than 3 months of development periods, importance and complexity increase significantly. Generally Agile-methodology is used for a development process for the medium-large scale mobile applications, but some issues arise such as high dependency on skilled developers and lack of detail development directives. In this paper, S-MADP (Smart Mobile Application Development Process) is proposed to mitigate these issues. S-MADP is a service oriented development process extending a object-oriented development process, for medium-large scale mobile applications. S-MADP provides detail development directives for each activities during the entire process for defining services as server-based or client-based and providing the way of reuse of services. Also, in order to support various user interfaces, S-MADP provides detail UI development directives. To evaluate the performance of S-MADP, three mobile application development projects were conducted and the results were analyzed. The projects are 'TBS(TB Mobile Service) 3.0' in TB company, mobile app-store in TS company, and mobile groupware in TG group. As a result of the projects, S-MADP accounts for more detailed design information about 'Minimizing the use of resources', 'Service-based designing' and 'User interface optimized for mobile devices' which are needed to be largely considered for mobile application development environment when we compare with existing Agile-methodology. Therefore, it improves the usability, maintainability, efficiency of developed mobile applications. Through field tests, it is observed that S-MADP outperforms about 25% than a Agile-methodology in the aspect of the required man-month for developing a medium-large mobile application.

• Journal of Intelligence and Information Systems
• /
• v.25 no.1
• /
• pp.163-177
• /
• 2019

As smartphones are getting widely used, human activity recognition (HAR) tasks for recognizing personal activities of smartphone users with multimodal data have been actively studied recently. The research area is expanding from the recognition of the simple body movement of an individual user to the recognition of low-level behavior and high-level behavior. However, HAR tasks for recognizing interaction behavior with other people, such as whether the user is accompanying or communicating with someone else, have gotten less attention so far. And previous research for recognizing interaction behavior has usually depended on audio, Bluetooth, and Wi-Fi sensors, which are vulnerable to privacy issues and require much time to collect enough data. Whereas physical sensors including accelerometer, magnetic field and gyroscope sensors are less vulnerable to privacy issues and can collect a large amount of data within a short time. In this paper, a method for detecting accompanying status based on deep learning model by only using multimodal physical sensor data, such as an accelerometer, magnetic field and gyroscope, was proposed. The accompanying status was defined as a redefinition of a part of the user interaction behavior, including whether the user is accompanying with an acquaintance at a close distance and the user is actively communicating with the acquaintance. A framework based on convolutional neural networks (CNN) and long short-term memory (LSTM) recurrent networks for classifying accompanying and conversation was proposed. First, a data preprocessing method which consists of time synchronization of multimodal data from different physical sensors, data normalization and sequence data generation was introduced. We applied the nearest interpolation to synchronize the time of collected data from different sensors. Normalization was performed for each x, y, z axis value of the sensor data, and the sequence data was generated according to the sliding window method. Then, the sequence data became the input for CNN, where feature maps representing local dependencies of the original sequence are extracted. The CNN consisted of 3 convolutional layers and did not have a pooling layer to maintain the temporal information of the sequence data. Next, LSTM recurrent networks received the feature maps, learned long-term dependencies from them and extracted features. The LSTM recurrent networks consisted of two layers, each with 128 cells. Finally, the extracted features were used for classification by softmax classifier. The loss function of the model was cross entropy function and the weights of the model were randomly initialized on a normal distribution with an average of 0 and a standard deviation of 0.1. The model was trained using adaptive moment estimation (ADAM) optimization algorithm and the mini batch size was set to 128. We applied dropout to input values of the LSTM recurrent networks to prevent overfitting. The initial learning rate was set to 0.001, and it decreased exponentially by 0.99 at the end of each epoch training. An Android smartphone application was developed and released to collect data. We collected smartphone data for a total of 18 subjects. Using the data, the model classified accompanying and conversation by 98.74% and 98.83% accuracy each. Both the F1 score and accuracy of the model were higher than the F1 score and accuracy of the majority vote classifier, support vector machine, and deep recurrent neural network. In the future research, we will focus on more rigorous multimodal sensor data synchronization methods that minimize the time stamp differences. In addition, we will further study transfer learning method that enables transfer of trained models tailored to the training data to the evaluation data that follows a different distribution. It is expected that a model capable of exhibiting robust recognition performance against changes in data that is not considered in the model learning stage will be obtained.

• Journal of Broadcast Engineering
• /
• v.11 no.1 s.30
• /
• pp.3-14
• /
• 2006

Music summarization means a technique which automatically generates the most importantand representative a part or parts ill music content. The techniques of music summarization have been studied with two categories according to summary characteristics. The first one is that the repeated part is provided as music summary and the second provides the combined segments which consist of segments with different characteristics as music summary in music content In this paper, we propose and evaluate two kinds of music summarization techniques. The algorithm using multi-level vector quantization which provides a repeated part as music summary gives fixed-length music summary is evaluated by overlapping ration between hand-made repeated parts and automatically generated summary. As results, the overlapping ratios of conventional methods are 42.2% and 47.4%, but that of proposed method with fixed-length summary is 67.1%. Optimal length music summary is evaluated by the portion of overlapping between summary and repeated part which is different length according to music content and the result shows that automatically-generated summary expresses more effective part than fixed-length summary with optimal length. The cluster-based algorithm using 2-D similarity matrix and k-means algorithm provides the combined segments as music summary. In order to evaluate this algorithm, we use MOS test consisting of two questions(How many similar segments are in summarized music? How many segments are included in same structure?) and the results show good performance.

• Journal of Mushroom
• /
• v.12 no.4
• /
• pp.350-356
• /
• 2014

This study was conducted to evaluate the feeding value of the spent mushroom (Hypsizygus marmoreus) substrates (SMS) in laying hens (Hy-Line Brown). The fresh spent mushroom (Hypsizygus marmoreus) substrates collected from the Namkyung farm were fermented with Bacillus subtilis EJ3 for 2 weeks. A total of twenty-four laying hens were fed corn-soy based experimental diets containing 0% (T0), 5% (T1), 10% (T2) and 15%(T3) fermented SMS for 12 weeks. There were no significant differences among the treatments in egg production, egg weight, egg mass, feed conversion and viability during the experimental period. Feed intake was significantly lowered in T0 (118.3 g) than T1 (121.9 g), T2 (120.3 g) and T3 (122.4 g). There were no significant differences among the treatments eggshell breaking strength, thickness and haugh unit, whereas the yolk color of T1, T2 and T3 were significantly heavy than T0. The palatability of boiled meat was significantly better in the T3 laying hens than in the T0 laying hens. In conclusion, fermented SMS can be used as resource of feed in laying hen feed at 5.0-15% level without effect on performance and egg qualify.

• Journal of Internet Computing and Services
• /
• v.21 no.2
• /
• pp.139-148
• /
• 2020

Due to change in circumstances in the 2000s such as severe birthrate decline and shortened military service period, the armed forces of the Republic of Korea is currently turning to technologies and equipments from manpower, developing it to become high-tech, high-speed, and complex, resulting in an environment in which a single mistake could cause a mass mortality crisis.It is also evident that, considering aspects such as safety training curriculums and achievements of advanced countries and private education, hands-on training is a must in preventing suicides and accidents in the military, and establishing safety training centers is crucial for systematic and effective hands-on training.Soldiers who are joining the army as of now have experienced the Internet ever since they were born and easily use both virtual and augmented reality, and the current level of science and technology has developed to the point where most of the public safety experience centers are able to be replaced by virtual and augmented reality (VR/AR). Therefore, considering the aspects such as installation space, construction costs, maintenance costs, user characteristics, and education effects, other than for those trainings where real models and objects are more effective such as first aid training, it is with a strong recommendation that establishing military safety training facilities with VR/AR (Virtual and augmented reality) is a must in the coming future. We have derived the need for hands-on training by considering the development of virtual and augmented reality (VR/AR), analysis of operation status of the public safety experience centers, characteristics of military units, installation and maintenance costs, and proposed plan to establish safety training centers where effective training performance can be achieved at a lower cost than the public safety experience center. In addition, we suggested the scale of the required safety training center and the composition of the experience rooms considering the number of trainees and the environment of each military units. Given this analysis it will contribute to the prevention of military safety and suicide by building a safety training center in the future.

• Journal of the Korean Electrochemical Society
• /
• v.13 no.2
• /
• pp.116-122
• /
• 2010

A $Cu_3Si$ film electrode is obtained by Si deposition on a Cu foil using DC magnetron sputtering, which is followed by annealing at $800^{\circ}C$ for 10 h. The Si component in $Cu_3Si$ is inactive for lithiation at ambient temperature. The linear sweep thermammetry (LSTA) and galvano-static charge/discharge cycling, however, consistently illustrate that $Cu_3Si$ becomes active for the conversion-type lithiation reaction at elevated temperatures (> $85^{\circ}C$). The $Cu_3Si$ electrode that is short-circuited with Li metal for one week is converted to a mixture of $Li_{21}Si_5$ and metallic Cu, implying that the Li-Si alloy phase generated at 0.0 V (vs. Li/$Li^+$) at the quasi-equilibrium condition is the most Li-rich $Li_{21}Si_5$. However, the lithiation is not extended to this phase in the constant-current charging (transient or dynamic condition). Upon de-lithiation, the metallic Cu and Si react to be restored back to $Cu_3Si$. The $Cu_3Si$ electrode shows a better cycle performance than an amorphous Si electrode at $120^{\circ}C$, which can be ascribed to the favorable roles provided by the Cu component in $Cu_3Si$. The inactive element (Cu) plays as a buffer against the volume change of Si component, which can minimize the electrode failure by suppressing the detachment of Si from the Cu substrate.

• Journal of the Korean Society of Marine Environment & Safety
• /
• v.25 no.7
• /
• pp.906-913
• /
• 2019

To evaluate the removal performance of PO₄-P and NH₃-N, laboratory experiments were conducted by filling a container with oyster shells, pyrolyzed at 100℃ (POS100), 600℃ (POS600) and 800℃ (POS800), and passing artificial wastewaters through the container. The pH in the ef luent was found to increase due to CaO eluted from oyster shell. Removal amounts of PO₄-P of ~23.1 mg/kg, 16.1 mg/kg, and 15.9 mg/kg were obtained when POS100, POS600, and POS800, respectively, were used; therefore, the highest PO₄-P removal amount was obtained when POS100 was used. It is considered that Ca and dolomite in the oyster shells adsorbed and precipitated PO₄-P. Removal amounts of NH₃-N were of ~3.56 mg/kg, 5.72 mg/kg, and 3.97 mg/kg were obtained when POS100, POS600, and POS800, respectively, were used The low removal rate for NH₃-N is probably due to unstable nitrification, use of sealed containers, and the effect of NH₃-N being converted to NH₄⁺ upon increasing pH. Based on these results, pyrolyzed oyster shell is expected to promote changes in PO₄-P and NH₃-N concentrations through chemical reactions. These results can also be used for basic research in the development of wastewater treatment.