Since the 1950s, artificial intelligence and artificial neural networks have been researched. In the polymer processing industry the call for machine learning and the use of artificial intelligence becomes loud at least since the efforts to introduce Industry 4.0. The aim of this work is to use an artificial neural network, more specifically a convolutional neural network (CNN), to assess the surface condition of components produced by injection foam molding and to provide recommendations for robust process parameter settings. In a parameter study, simple plates with different surface quality were produced and photographed while varying injection speed, blowing agent content and mold temperature. In a first step, a CNN was trained to assign the individual images of the component surfaces to the respective process settings. The process is then iteratively optimized by means of relative parameter adaption until a predefined ideal is achieved. By automatic input generation via photographing the latest produced component and integrating the CNN into the machine control, a fully autonomous self-optimizing process is realized.