A digital twin emulator for production performance prediction and optimization using multi-scale 1DCNN ensemble and surrogate models

Production performance prediction and optimization play an important role in securing smooth production and maintaining great efficiency. Traditional methods suffer from tardy and inflexible adjustments, leading to a manufacturing system with low-level responsiveness and adaptability. To pave the way for this, a digital twin (DT) emulator is proposed and driven by the collaboration of continuous prediction and iterative optimization. An overall system architecture is first presented to provide a generic reference. A multi-scale one-dimensional convolutional neural network (1DCNN) ensemble model is designed for continuous prediction, which integrates multi-scale kernels and ensemble structure to boost performance. A surrogate model of multivariate adaptive regression spline (MARS) is utilized to explicitly fit the relationship between production system variables and serve as the fitness function of optimization for appropriate parameter value searching. An industrial company collaboration case of production and assembly lines is used to demonstrate the effectiveness and feasibility of the proposed approach.