Hybrid Machine Learning – When Little Data is Available

The prediction of wear conditions in production are worth hard cash. In highly productive plants, which are common today, even smaller extensions of the production intervals lead to considerable cost reductions.
With the machine learning tools that are available today, it is relatively easy to develop predictions. These tools are partly even free, but especially for the deep learning algorithms a lot of data is needed. However, are usually not enough data sets available, because the processes are so efficient that only a low wear is present and certain wear patterns occur only rarely.
This is where Hybrid Machine Learning comes in. Since in engineering certain relations can be described well by physical clothing, but usually the boundary conditions and certain operating parameters are not known exactly, one can develop models that are based partly on physical descriptions partly on machine learning.