http://scholars.ntou.edu.tw/handle/123456789/215 Mon, 10 Aug 2026 19:44:59 GMT 2026-08-10T19:44:59Z https://scholars.ntou.edu.tw:443/retrieve/97/資訊工程學系.png http://scholars.ntou.edu.tw/handle/123456789/215 http://scholars.ntou.edu.tw/handle/123456789/26807 標題: Remaining Useful Life Prediction of Turbo Engine Based on CNN-BiLSTM Model With Feature Enhancement Approaches 作者: Li, Shih-Yu; Wu, Shih-Ping; Tam, Lap-Mou; Cheng, Shyi-Chyi 摘要: This article proposes a novel method for predicting the remaining useful life (RUL) of turbine engines by integrating convolutional neural networks (CNNs) and bidirectional long short-term memory (BiLSTM) frameworks with chaotic mapping (CM) and fractional-order (FO) filters (called CBCF). The method follows a sequence of CM, FO filter, CNN, and BiLSTM. CM is used for preprocessing to extract meaningful features from engine data, while the FO filter enhances image details produced by CM, contributing to more accurate RUL predictions. The enhanced images are then used as inputs to the CNN-BiLSTM model. CNN is employed to automatically extract and learn hierarchical features from the engine data through convolutional layers, capturing spatial patterns and local dependencies for more effective feature representation. BiLSTM is then used to process the time-sequenced data, integrating forward and backward network information to capture both past and future dependencies, thereby improving prediction accuracy. The commercial modular aero-propulsion system simulation (C-MAPSS) dataset, which is a benchmark dataset in this field, is utilized to evaluate the feasibility and effectiveness of the proposed method. The C-MAPSS dataset provides comprehensive data on engine degradation and has been widely used for validating RUL prediction models. Experimental results show that the integration of CM and FO filters with the CNN-BiLSTM model enhances performance, significantly reducing the root-mean-square error (RMSE) and demonstrating the potential of this approach for accurate RUL predictions. Thu, 01 Jan 2026 00:00:00 GMT http://scholars.ntou.edu.tw/handle/123456789/26807 2026-01-01T00:00:00Z http://scholars.ntou.edu.tw/handle/123456789/26796 標題: Regional exploitation signals inferred from a data-limited per-recruit assessment of Trichiurus japonicus in Taiwanese waters 作者: Lin, Chih-Yu; Lu, Yu-Heng; Xu, Wen-Qi; Wang, Sheng-Ping; Kitakado, Toshihide 摘要: Trichiurus japonicus is a heavily exploited coastal fishery species in the Northwest Pacific Ocean. Although Taiwan contributes less than 2% of regional landings, localized assessment is important because exploitation pressure reflects cumulative fishing mortality across jurisdictions and biological parameters may vary geographically. The lack of species-specific catch data and long-term abundance indices has prevented formal stock assessment in Taiwanese waters. To evaluate relative exploitation status under severe data constraints, agestructured and length-structured per-recruit analyses were conducted using length-frequency data collected from Taiwanese coastal fisheries between 2019 and 2023. Natural mortality was estimated using six empirical methods, and total mortality was estimated using seven catch-curve approaches, yielding forty-two fishing mortality scenarios. Length compositions were consistently dominated by small to medium individuals, and reconstructed age compositions were strongly truncated, with age-0 and age-1 fish comprising most of the catch. Current fishing mortality ranged from 0.749 to 1.483 year(-1) and exceeded conservative biological reference points, including F0.1 (0.203-0.731 year(-1)), in most scenarios. It also frequently approached or exceeded FMAX (0.385-1.887 year(-1)) and spawning potential ratio (SPR)-based thresholds. Current SPR ranged from 0.128 to 0.456 and was generally below the 40% target threshold and often near or below the 25% limit threshold. Although available data preclude a complete stock assessment, consistent signals across model structures and selectivity assumptions indicate elevated exploitation that reduces survival to older ages and limits reproductive output. These findings demonstrate that per-recruit analyses can provide useful indicators of relative exploitation under data-limited conditions and highlight the importance of incorporating localized biological indicators into future regional assessments of this transboundary resource. Thu, 01 Jan 2026 00:00:00 GMT http://scholars.ntou.edu.tw/handle/123456789/26796 2026-01-01T00:00:00Z http://scholars.ntou.edu.tw/handle/123456789/26779 標題: Constructing Chatbots using Generative AI and Domain Knowledge 作者: Lin, Chih-Ying; Lin, Han-Yi; Ilang, Yan-Cih; Ma, Hang-Pin 摘要: Generative AI powered by large language models has made it possible for chatbots to generate human-like responses; however, major challenges remain in intent recognition, knowledge enrichment, dynamic functionality, and knowledge integration. This paper proposes three design patterns to address these issues: domain adaptation for the retrieval of external information, Intent Bridge for dynamic intent execution, and Knowledge Weaver for the seamless integration of external knowledge. Together, these patterns bridge the gap between theoretical and practical deployment, providing a framework for enhancing chatbot performance in complex tasks. This framework provides developers with systematic methods to improve accuracy, functionality, and contextual relevance in generative AI-powered chatbot systems. Thu, 01 Jan 2026 00:00:00 GMT http://scholars.ntou.edu.tw/handle/123456789/26779 2026-01-01T00:00:00Z http://scholars.ntou.edu.tw/handle/123456789/26743 標題: Efficient Data Collection for Underwater Wireless Sensor Networks 作者: Xiao, Kai-Lun; Chao, Chih-Min 摘要: The inherent high-latency feature of acoustic transmission within an underwater wireless sensor network (UWSN) enables multiple transmitters at varying distances from the receiver to transmit simultaneously. Many existing medium access control (MAC) protocols utilize this high latency feature to reduce data collection time. However, most MAC protocols use contention-based mechanisms and face transmission collisions as the number of nodes increases. In this article, we propose a contention-free pipelined scheduling (CFPS) protocol to reduce data collection time in a single-hop network. The primary concept behind CFPS involves partitioning nodes into subclusters and developing a collision-free transmission scheduling table for different subcluster heads. The CFPS protocol is expected to reduce data collection time and improve network throughput when compared to the three well-designed data collection protocols, namely, a delay-aware probability-based MAC protocol for underwater acoustic sensor networks, random handshake MAC protocol based on Nash Equilibrium for UWSNs, and leveraging the near far effect for improved spatial-reuse scheduling in underwater acoustic networks. Thu, 01 Jan 2026 00:00:00 GMT http://scholars.ntou.edu.tw/handle/123456789/26743 2026-01-01T00:00:00Z