Dataset for statistically validated surrogate learning of residual vibration in a PI-controlled flexible manipulator

Description

This dataset contains numerical simulation data used to develop and evaluate surrogate-learning models for predicting RMS residual vibration in a PI-controlled single-link flexible manipulator. The input variables include trapezoidal motion-profile deceleration time and PI-control gains, and the output variable is the RMS residual-vibration response. Additional files report the extended regression benchmark results and held-out test predictions of the best-performing surrogate model.

Steps to reproduce

The dataset was generated through numerical simulations of a PI-controlled single-link flexible manipulator. The manipulator dynamics were modeled using a finite-element formulation previously validated against experimental modal analysis. Transient responses were computed in MATLAB using the Newmark time-integration method. For each simulation run, the manipulator was commanded to complete a 90-degree angular motion within 1 s using a trapezoidal velocity profile. The vibration response was then evaluated over a 4 s interval after motion completion. Three input variables were varied systematically: the deceleration time of the trapezoidal motion profile (tdec), the proportional control gain (Kp), and the integral control gain (Ki). The resulting output variable was the root mean square (RMS) residual-vibration response. A total of 1665 numerical samples were generated by sweeping the selected motion-profile and PI-control parameters over the considered design range. The complete dataset was used to train and test feedforward neural-network configurations and additional surrogate regression models. The data were split into 80% training and 20% testing subsets before normalization. The extended benchmark files report the performance of ten surrogate models, and the prediction file reports the held-out test predictions of the best-performing Gaussian-process regression model with a Matérn 3/2 kernel.

Institutions

• Dokuz Eylül University

  İzmir Province, Izmir

Categories

Licence