Multi-Objective Approach to Automated Fixture Synthesis Incorporating Deep Neural Network for Deformation Evaluation

In machining, the synthesis of a fixturing schema significantly impacts the accuracy of the final product. Moreover, A robust and automatic configuration of fixture elements can reduce production costs and eliminate the need for expert labor to perform the task. Given the multi-modal problem of fixture synthesis, this article presents a multi-objective approach to fixture synthesis in the discrete domain. The performance criteria are localization accuracy, detachment of locators, workpiece deformation, severity and dispersion of reaction loads, and the spacing between contact points. Optimization is performed via an improved Declining Neighborhood Simulated Annealing algorithm (DNSA). To achieve consistent performance across different inputs, the number of iterations is proportional to a Shanon entropy index reflecting the recurrence of folds/corners. Except for deformation, all other objectives are derived from the kinematic analysis of the workpiece-fixtures system. To reduce the computational burden, a Constitutive Deep Neural Network (CDNN) is proposed for real-time deformation evaluation. Both models incorporate the machining loads as quasi-static intervals. For the trade-off, a new strategy is applied based on Z-score quantification of objectives and a pre-calibration run of DNSA. Numerical examples from literature and industrial partner are provided for illustration and validation. These examples were implemented using our CAD-based tool (developed via NXOpen API and NX UI). The approach proved efficient in automating the selection of the robust fixture layout for a prismatic workpiece.