Enhanced classification of hydraulic testing of directional control valves with synthetic data generation

• DOI: 10.1007/s11740-023-01204-8
• Publication Date: 2023-04-26T16:02:20Z
• Authors: Christian Neunzig; Dennis Möllensiep; Melanie Hartmann; Bernd Kuhlenkötter; Matthias Möller; Jürgen Schulz
• Keywords: Oversampling; Data set; Feature (linguistics); Synthetic data

Abstract Production environments bring inherent system challenges that are reflected in the high-dimensional production data. The data is often nonstationary, not available sufficient size and quality, class imbalanced due to predominance of good parts. Data-driven manufacturing analytics requires quantity quality. In order predict quality characteristics, collected across processes industrial use case at Bosch Rexroth AG for purpose inferring results hydraulic final inspection using machine learning methods. Since high generation costly, synthetic methodologies offer a promising alternative improve prediction models thus generate safer, more accurate predictions companies. Among used, variational autoencoders compared generative adversarial networks minority oversampling technique methods best suited synthesize feature with highest importance from small sample set target variable.