Search Results (3)

An engine-propeller cooperative control based on model predictive control (MPC), which takes a deep neural network (DNN) as the prediction model, is studied, and a two-layer optimization method is proposed to improve the economy and maneuverability of the COGAG system. The engine-propeller matching characteristic of the COGAG system is studied, and the economy of the COGAG system is analyzed. In the system planning layer, when the vessel speed command is given, the economic optimal point can be identified. In the local control layer, the DNN-MPC control for different dynamic processes is designed. Moreover, the DNN model has the ability to run in ultra-real time. Compared with parallel control based on PI and parallel power feedback control based on PID, the optimal control based on DNN-MPC can improve the maneuverability of the COGOG pattern by 31.82% and 16.67% in the process of accelerating from 1st to 8th gear and improve the maneuverability of the COGAG pattern by 50% and 23.08% in the process of accelerating from 1st to 10th gear. Moreover, DNN-MPC control can effectively avoid the overshoot of propeller speed caused by the change in pitch adjustment. It provides the theoretical basis for multi-objective optimization of the COGAG system. Full article

Gas turbine engines at sea, characterized by nonlinear behavior and parameter variations due to dynamic marine environments, pose challenges for precise speed control. The focus of this study was a COGAG system with four LM-2500 gas turbines. A third-order model with time delay was derived at three operating points using commissioning data to capture the engines’ inherent characteristics. The cascade controller design employs a real-coded genetic algorithm–PID (R-PID) controller, optimizing PID parameters for each model. Simulations revealed that the R-PID controllers, optimized for robustness, show Nyquist path stability, maintaining the furthest distance from the critical point (−1, j0). The smallest sensitivity function $Ms$ (maximum sensitivity) values and minimal changes in $Ms$ for uncertain plants confirm robustness against uncertainties. Comparing transient responses, the R-PID controller outperforms traditional methods like IMC and Sadeghi in total variation in control input, settling time, overshoot, and ITAE, despite a slightly slower rise time. However, controllers designed for specific operating points show decreased performance when applied beyond those points, with increased rise time, settling time, and overshoot, highlighting the need for operating-point-specific designs to ensure optimal performance. This research underscores the importance of tailored controller design for effective gas turbine engine management in marine applications. Full article

Recently, the research community has taken great interest in human activity recognition (HAR) due to its wide range of applications in different fields of life, including medicine, security, and gaming. The use of sensory data for HAR systems is most common because the sensory data are collected from a person’s wearable device sensors, thus overcoming the privacy issues being faced in data collection through video cameras. Numerous systems have been proposed to recognize some common activities of daily living (ADLs) using different machine learning, image processing, and deep learning techniques. However, the existing techniques are computationally expensive, limited to recognizing short-term activities, or require large datasets for training purposes. Since an ADL is made up of a sequence of smaller actions, recognizing them directly from raw sensory data is challenging. In this paper, we present a computationally efficient two-level hierarchical framework for recognizing long-term (composite) activities, which does not require a very large dataset for training purposes. First, the short-term (atomic) activities are recognized from raw sensory data, and the probabilistic atomic score of each atomic activity is calculated relative to the composite activities. In the second step, the optimal features are selected based on atomic scores for each composite activity and passed to the two classification algorithms: random forest (RF) and support vector machine (SVM) due to their well-documented effectiveness for human activity recognition. The proposed method was evaluated on the publicly available CogAge dataset that contains 890 instances of 7 composite and 9700 instances of 61 atomic activities. The data were collected from eight sensors of three wearable devices: a smartphone, a smartwatch, and smart glasses. The proposed method achieved the accuracy of 96.61% and 94.1% by random forest and SVM classifiers, respectively, which shows a remarkable increase in the classification accuracy of existing HAR systems for this dataset. Full article