This paper proposes a neural network predictive control (NNPC) method to address the challenge of controlling the output speed in a fixed displacement motor (FDM) system controlled by an electrohydraulic proportional variable displacement pump (EHP-VDP). The dynamic response of the system is divided into four segments, and mathematical models are derived for each segment. The system is modelled using over 400 input-output data pairs, with 75% of the data used for training and the remaining 25% for validation. The Levenberg-Marquardt algorithm (LMA) is employed to optimise the neural network, achieving optimal validation performance with a mean square error of 0.0029637 after 258 epochs. This real-time optimisation enables the output speed of the system to accurately track the reference speeds. Simulink simulations are conducted over a 25-second duration covering five reference speed intervals, demonstrating reliable tracking performance. The NNPC method shows competitive performance compared with backpropagation-neural network-proportional integral derivative (BP-NN-PID) and proportional integrative derivative (PID) controllers, achieving the root mean square error (RMSE), the mean absolute error (MAE), and the integral squared error (ISE) values of 6.72, 1.00, and 572.89, respectively. These results indicate that the NNPC offers advantages in fast-response scenarios while maintaining acceptable tracking accuracy.
This paper considers a multi-objective optimisation problem focusing on translational parallel manipulators. The proposed approach combines interval analysis with genetic algorithms to optimise the dynamic parameters of parallel manipulators. The objectives are to enhance the robot accuracy, while maximising the tolerance intervals of the parameters. This paper's contribution is to introduce an interval method to estimate the error bounds of a dynamic parallel manipulator within the desired workspace. A genetic algorithm is then applied to improve manipulator accuracy and reduce design costs. The resulting Pareto fronts illustrate the trade-off between each pair of objective functions.
This study examines springback in mild-steel sheet air V-bending through an integrated experimental-numerical-data-driven framework aimed at practical bend design and tool compensation. Bending tests were performed using a modular die set while systematically varying sheet thickness, punch radius, and target bending angle, and repeated measurements were used to ensure reliable springback evaluation. The experimental trends were crosschecked with nonlinear finite element simulations in MSC Marc Mentat (R), providing a physicsbased reference and confirming that the selected parameters govern the springback response over the investigated range. To enable fast prediction without repeated simulations, a feedforward artificial neural network (ANN) was trained on 106 experimental cases to map the forming parameters to springback. Because conventional ANN training can be sensitive to random initialisation and can become trapped in local minima, a genetic algorithm (GA) was employed to optimise the initial weights and biases prior to gradient-based learning. Compared with a standard ANN, the GA-optimised ANN delivered more stable convergence and improved generalisation, increasing test accuracy (R2 from 0.833 to 0.875) and reducing the mean absolute error from 0.195 degrees to 0.075 degrees (61.5% improvement). Overall, the proposed hybrid approach combines experimental reliability, FEM validation, and GA-enhanced learning to provide an efficient and robust springback prediction tool for sheet-metal forming applications.
This study investigates the mechanical performance and microstructural evolution of AA7075 aluminium alloy joints produced using the cold metal transfer process. Owing to its low heat input, cold metal transfer is well suited for welding heat-sensitive precipitation-hardened alloys prone to porosity, hot cracking, and strength loss. Response surface methodology with a central composite design was applied to quantify the effects of welding current (160 to 200 A), welding speed (100 to 140 mm/min), and wire feed speed (1000 to 2000 mm/min) on ultimate tensile strength. The resulting empirical model showed strong statistical significance and close agreement between predicted and experimental values. Microstructural analysis revealed the dissolution of dizinc magnesium precipitate, dendritic solidification, and inter-dendritic segregation in the fusion zone; partial over-ageing in the heat-affected zone; and the retention of the T6 structure in the base metal. Hardness decreased from 164-171 HV0.5 (the base metal) to 147-155 HV0.5 (the heat-affected zone) and 140-145 HV0.5 (the fusion zone). Validation using optimised parameters achieved an ultimate tensile strength of 366 MPa, matching the model prediction. Overall, the study confirms cold metal transfer as an effective technique for producing high-strength AA7075 welds with minimal thermal damage.
This paper presents a novel approach to the integration of biological inspiration into topological optimisation through the specification of input parameters derived from nature. The aim is to create biomimetic designs that may incorporate the multifunctionality inherent in natural systems. The presented methodology for modelling bioinspired structures using topological optimisation provides critical insights into translating natural forms and their load-bearing characteristics into topological optimisation design parameters (design space, boundary conditions, and loads). The proposed BioTop method is a structured framework that guides designers through a design-to-biology approach for conducting truly bioinspired topological optimisation by systematically defining the required inputs. An illustrative case study demonstrates the practical application of the BioTop method and its capacity to create bioinspired topologically optimised structures. While acknowledging current simplifications, this research lays a foundation for future research in bioinspired design by providing a systematic method for using nature as an informed input source for the standard topological optimisation software.
