The correlation between manufacturing parameters and the resulting surface topography is most often described with standardized profile surface texture parameters (R-parameters) which represents a strong simplification. The most common parameters are often neither function-oriented nor correlated with the manufacturing parameters. We use a convolutional neural network as a regression model to predict the manufacturing parameters from a measured topography dataset. As the training requires large amounts of data, we use stochastic surface models to generate artificial profiles. The prediction accuracy of the trained neural network is evaluated for a case study of an abrasive process using artificial profiles, measured profiles and frequency dependent representations of the training data. By comparing the best approach with a linear regression model between manufacturing parameters and R-parameters, the performance of the parameter identification compared to R-parameters can be demonstrated.