Machine learning algorithms are extremely efficient and oftentimes more accurate than their human counterparts or analytical methods. However, classic machine learning relies solely on data, and in presence of data quality issues, such as noise, insufficient diversity or size of data, it can lead to unreliable predictions or even overall inability to produce models for certain tasks.
To counter these problems, we combine machine learning methods with domain knowledge and known physics. Such approach allows to guide the learning process, ensuring their consistency with the physical world, as well as constraining their behavior to avoid unpredictable outcomes. The methods are applied to industrial settings in the context of predictive maintenance, such as increasing the lifetime of wind turbines, or detecting operational irregularities in maritime maneuvering machinery.