State Classification in Injection Molding Cycles using Transformation of Acceleration Data into Images

In this paper, we present a method to distinguish the different states of an injection molding process which is an important basis for monitoring and subsequently optimizing the production process and its efficiency. For this purpose, a triaxial accelerometer is used, which can be easily and inexpensively retrofitted on the machine. The signals from the accelerometer are transformed into images using various algorithms known from the literature (especially for human activity recognition). Afterwards, these images are classified using Convolutional Neural Networks (CNNs). The classification results of different transformation methods and CNNs are combined by weighted majority voting to achieve higher robustness of the classification. The results show high accuracy and are promising for further developments in this area.