1 April 2021 in Review of Scientific Instruments
The milling of highly flexible workpieces, such as thin-walled structures used in turbine blades, aerospace equipment, and jet engine compressors, requires vibration compensation to improve the quality of the workpiece surface. Vibration can be reduced by selecting appropriate cutting parameters. However, this approach reduces system productivity. This paper presents an active workpiece holder that controls the vibration of general computer numerical control machine tools. The proposed holder, which comprises a flexible guide mechanism, driver, and sensor, measures vibration and actively controls it using piezoactuators. A high-rigidity flexure mechanism was designed for the holder, and finite element method simulation and modal analysis were performed. Finally, the proposed system was fabricated, and experimental verification indicated that the system reduced vibration. The surface quality obtained using the controlled system was ∼50% better than that obtained using the uncontrolled system.
30 October 2025 in IEEE Access
A weighing system calculates the payload of an excavator bucket using sensor information. This technique has been employed to assess the work of the operators and to verify the termination point between the excavator and mobile construction equipment during collaborative operations. Certain commercial weighing devices, such as Trimble’s Loadrite, have measurement inaccuracies attributable to the restricted quantity of sensors. This research uses a deep learning model to mitigate measurement errors. Hyperparameters in deep learning are variables that must be predetermined prior to the learning process. To achieve optimal performance from a neural network, a user must engage in optimization, which is challenging due to several unknown variables. This work employs a genetic algorithm (GA) to optimize hyperparameters. The data was initially gathered during the loading and unloading activities. Subsequently, it was employed for a deep regression model. The data were converted to a structured format contingent upon the input shape of the network. A polynomial regression model was employed for the fully connected dense layer architecture. Strict criteria are necessary for evaluating deep learning models. A chosen model was refined through genetic algorithms and subsequently validated with loading and unloading tests using a different excavator.