19 March 2019 in IEEE Access
Transfer learning is a promising deep learning approach that can be used in applications that suffer from insufficient training data. Parameter transfer, which is a method of improving the accuracy and training speed by training the target network using the parameters of the source network, selects two conventional parameter repurposing methods such as parameter freezing and fine tuning depending on the amount of target data and network size. In this paper, we propose a novel method to increase performance in the target domain in an intermediate approach of both methods. The proposed method, selective parameter freezing (SPF) freezes only a portion of parameters not freeze or fine-tune all parameters within a layer by choosing output-sensitive parameters from the source network. Freezing only sensitive parameters while training is to reduce the amount of trainable parameter and protect informative parameters from overfitting to small number of target data. Using two sets source-target domain, 1) artificial faults with different fault size, 2) artificial faults – natural faults of rolling element bearing, the proposed SPF allows adaptation to the target domain by choosing best degree of freezing with various amount of target data and size of networks, compared to conventional approaches.
31 December 2021 in Journal of the Korean Society for Environmental Technology
The purpose of this study is to understand the current status and cause of the distribution of odor substances in the Yeosu National Industrial Complex and its surrounding areas. Samples were analyzed by collecting 90 samples at dawn, daytime and night from a total of 15 points, 9 points in the management area, 2 points in the boundary area, and 4 points in the affected area. The analysis items include 5 meteorological factors, complex odor, 22 types of the designated odors and other Volatile Organic Compounds (VOCs). At the MA3 (Fig. 1) point in the management area, the strict emission allowance standard for complex odors was exceeded three times, Ammonia, two types of sulfur compounds, two types of aldehydes, and six types of VOCs were detected, but they were within the standard. As a result of by calculating the odor contribution, the contribution of 74.9 % or more was found at the point where butyraldehyde was detected. Therefore, in order to prevent odors at the Yeosu Industrial Complex, measures to reduce aldehydes with high odor contribution should be prioritized, and it is necessary to simultaneously investigate through real-time harmful atmospheric measurement vehicles. Based on the contents presented in this study, if the system and policies reflecting the results of the odor occurrence survey of Yeosu National Industrial Complex and surrounding areas are supplemented, the odor environment of Yeosu National Industrial Complex will be sufficiently improved.