The article presents the result of using a neural network to recognize the types of forging parts manufactured with hot forging on Upsetting Forging Machines. The main feature of these machines is the presence in the machine of two sliders. This makes it possible to forge parts of various shapes in several operations. Number and type of operations depend on forgings configuration, which divided into more than 25 groups and subgroups. Forgings affiliation to one of the group is determined subjectively by the technologist, based on general recommendations and his own experience. But subjectivity in design does not always provide the best results. То solve forging geometry classification problems, custom neural network was designed using framework for working with convolutional neural network—TensorFlow with the Keras API to work with sequential models, allowing layers to be added sequentially. The synthesized model of neural network model consisted of the following layers: data normalization, convolutional layer, pooling, fully connected layer. To create a data set for training the neural network, five subgroups of axisymmetric of forgings were selected as the object of research. After the training stage with ten iterations, the synthesized neural network was able to recognize that the geometry of forged parts affiliation to a certain group with a confidence of 90–97%. The proposed approach makes it possible to move away from the subjective assessment of the geometry of forging parts, and to realize the complete digitalization of the design of the technological process.