Human-robot interaction has received a lot of attention as collaborative robots became widely used in many industrial applications. This paper proposes a deep learning method for collision identification of collaborative robots. This method expands the idea of CollisionNet, which was proposed for collision detection, to identify locations of collisions. Collision identification is far more difficult compared to collision detection, because sensor data are highly correlated when collisions occur at close locations. To improve the identification accuracy, this paper proposes an auxiliary loss, which is called consistency loss. This auxiliary loss guides the training of a deep neural network to predict consistent predictions for each single collision event. In experiments, we demonstrate the effectiveness of the proposed method.

This paper presents the test and evaluation results of the fabric classification technology using deep learning of feedback vibration occurring on fabric surfaces. Ten fabrics composed of different materials were selected for the classification test. To build a database for the design of an artificial intelligence model, feedback vibration measurement equipment with functions of fixed tension, contact load, and contact velocity control, was constructed and feedback vibration data on each fabric surface were collected under the same measurement conditions. Then, training and validation datasets were created with the collected feedback vibration data, and a deep learning architecture with convolutional neural networks was designed in the consideration of data characteristics. A deep learning model development program was established and the fabric classification model was derived with the training and validation dataset. In the end, a test system including an embedded system with the developed model was constructed in order to test and evaluate the performance of the fabric classification model. Results of the fabric classification test were summarized and analyzed by means of the confusion matrix. Finally, the performance of the integrated system was confirmed to have an accuracy of more than ninety percent in fabric classification with the developed model.

In this paper, we propose an artificial intelligence (AI) service that plays a supportive role in robot assisted-surgery using deep learning algorithm that have recently been spotlighted in several fields. The proposed AI service is equipped with the ability to segment surgical tools and the ability to recognize the behavior of surgical tools. In addition, such AI service is opened using public web page to make them easier for surgeons to use. Models mounted on AI service are segmentation deep learning model and action recognition deep learning model. The segmentation deep learning model showed a final mIoU performance of 0.867 for seven surgical tools, and the action recognition deep learning model shows an accuracy of 86.96% for the opening and closing actions of all surgical tools.

This paper proposes a deep learning algorithm for estimating pig weight. The proposed algorithm estimates the weight of a pig using the point cloud obtained through a mobile device. The proposed model is based on the PointNet which is widely used in the point cloud data. Through the optimization of the PointNet, the proposed method not only improves the accuracy, but also reduces the computational complexity. The accuracy (82.4 %) of the proposed method was about 3 % higher than that of the conventional method (79.4 %). Also, the numbers of the trainable parameters for the PointNet and the proposed method were 3,1 4,7 1 and 150,5 4, respectively. That is, the proposed method used only 5 % of trainable parameters compared to the PointNet. The developed model makes it easier and faster to measure the weight of a pig than the conventional method.

This paper introduces a teleoperation control system of an industrial robot based on hand gestures using the convolutional neural network (CNN). The proposed system employs the gesture data obtained from an EMG sensor and considers a CNN-based deep learning method. Using the proposed CNN model, we develop a real-time teleoperation control system for the industrial robot. Finally, it is confirmed that the proposed system is reliable in real system since it can be applied to the teleoperation control of a real industrial robot.

Many companies try to analyze and utilize feedback on services. This can be used for improving service quality or marketing. Until now, most natural language processing studies have attempted to analyze emotions divided into positive, negative and neutral. However, in this work, specific negative reasons are extracted and classified. The dataset is a standard dataset from kaggle that uses tweet data for U.S. airline services. Tweets categorized as negative are labeled with 10 categories of negative reasons. The dataset was divided into train, validation, and test 8:1:1. The learning and classification process was largely divided into two stages. The first is to convert words and sentences into vector values. It is compared and analyzed using Doc2Vec and BERT (Bidirectional Encoder Representations from Transformers) models for embedding and vectorization. The second is to learn and classify sentences transformed into vectors by matching them with 10 negative reason classes. During this learning process, I converted the negative reason into a sentence and attached it to the back of the original text and made new data.I then used BERT's Next Sentence Prediction technique to allow further learning to be performed. This method was able to improve classification accuracy. For each dataset and classification method, metrics were computed, visualized, and compared.

In recent years, as ESS has become very popular, BMS related electric fires have occurring frequently. In this research we performed analysis of risk priority number (RPN) using Failure Mode and Effects Analysis (FMEA) technique to analyze the safety of BMS, which accounts for the most part in ESS electric fires. And the functional safety concept was used to eliminate or reduce failure modes that may cause overvoltage, overcurrent, and overheating for higher risk priorities of the analyzed BMS. The BMS hardware was redesigned so that the safety mechanism works for the high-priority risk modes, and the main firmware procedures were designed to control the battery against potential malfunctions. And we implemented the improved BMS hardware and experimentally verified that the safety mechanism works as designed. The test results have confirmed that the safety mechanism works normally and the battery can be controlled even if overvoltage, overcurrent and overheating occurs in the BMS, or major firmware procedures malfunction. Therefore we are confident that electric fires related to ESS can be prevented in advance by analyzing the safety of BMS in the way we used in this research to find high-priority risk factors and applying the concept of safety functions to the hardware and software design of BMS.