Springback Prediction Using Gated Recurrent Unit and Data Augmentation

Artificial intelligence (AI) has been widely used in manufacturing, healthcare, sports, finance and other fields to model nonlinearities and make reliable predictions. In manufacturing, AI has been applied to improve processes, reduce costs and increase reliability. A new manufacturing process enhanced by AI is single point incremental forming (SPIF), a technique that uses a computer numerical control (CNC) machine to incrementally feed a metal sheet or polymer blank. However, achieving the geometric accuracy of the process is still of primary challenge due to the impact of springback. One of the most common solutions is toolpath correction. In this paper, we proposed a mechanism to capture local geometry using a novel point series representation, which then forms a general global geometry information. Each point series can then be associated with a predicted springback value and learns using deep learning. In particular, this article proposes the use of data augmentation to solve the problem of insufficient data and enable deep learning models to achieve better performance. Intensive experimental results show that we achieved the best R² or “coefficient of determination” of 0.9228 compared to recent methods. We show that the proposed method provides a realistic solution to the current limitations of SPIF.