Real-world products and physics-based simulations are becoming interconnected. In particular, real-time capable dynamic simulation has made it possible for simulation models to run in parallel and simultaneously with operating machinery. This capability combined with state observer techniques such as Kalman filtering have enabled the synchronization between simulation and the real world. State estimator techniques can be applied to estimate unmeasured quantities, also referred as virtual sensing, or to enhance the quality of measured signals. Although synchronized models could be used in a number of ways, value creation and business model development are currently defining the most practical and beneficial use cases from a business perspective. The research reported here reveals the communication and collaboration methods that lead to economically relevant technology solutions. Two case examples are given that demonstrate the proposed methodology. The work benefited from the broad perspective of researchers from different backgrounds and the joint effort to drive the technology development towards business relevant cases.

Rotor bars are one of the most failure-critical components in induction machines. We present an approach for developing a rotor bar fault identification classifier for induction machines. The developed machine learning-based models are based on simulated electrical current and vibration velocity data and measured vibration acceleration data. We introduce an approach that combines sequential model-based optimization and the nested cross-validation procedure to provide a reliable estimation of the classifiers’ generalization performance. These methods have not been combined earlier in this context. Automation of selected parts of the modeling procedure is studied with the measured data. We compare the performance of logistic regression and CatBoost models using the fast Fourier-transformed signals or their extracted statistical features as the input data. We develop a technique to use domain knowledge to extract features from specific frequency ranges of the fast Fourier-transformed signals. While both approaches resulted in similar accuracy with simulated current and measured vibration acceleration data, the feature-based models were faster to develop and run. With measured vibration acceleration data, better accuracy was obtained with the raw fast Fourier-transformed signals. The results demonstrate that an accurate and fast broken rotor bar detection model can be developed with the presented approach.

This work focuses on the ontological representation of creep void analysis data to automate the training of the machine learning (ML) model detecting creep voids in scanning electron microscope images. Metallic high-temperature structures are subject to creep phenomenon that can lead to rupture and component failure when prolonged. ML models can be deployed to detect and obtain information about the density and location of creep voids using images as input data. However, due to the irregularities in the size and shape of creep voids and the associated uncertainty in ML models, the material engineer is required to inspect the detections and provide feedback on a regular basis. To automate the retraining of the ML model and to facilitate the close collaboration between the ML experts and the material engineers, domain ontologies providing common vocabularies can be utilized. We aligned the relevant concepts of the EMMO ontology, ML-Schema, and the PROV ontology to document creep void data for smooth information sharing and improving the quality of the ML detection process, hence resulting in better analysis and higher productivity compared to manual characterization. The SPARQL queries are used to gain valuable insights, such as the number and area of creep voids, condition of the metallic structure, and accuracy of the ML predictions.