Industrial AI Training Data Transformation for Failure Detection

The patent creates synthetic copies of operational data by training a generative AI model on available real operational data and sensor data. The generative model learns the underlying distributions and relationships in the data, then generates synthetic training samples that replicate real operational conditions including failure scenarios. This copying approach enables multiplication of limited real data into comprehensive training datasets without requiring additional physical data collection from industrial systems.

The patent transforms data between different operational conditions by learning parameter mappings between matched data sets. By identifying corresponding relationships between data sets with similar operational conditions, the system can transform parameters and generate training data for scenarios that would be difficult or impossible to capture directly, thereby expanding training data coverage across various operational and failure modes.

The patent performs preliminary data preparation by collecting and preprocessing operational data and sensor data before AI module training. Matched data sets are identified and stored in advance, and the generative AI model is trained beforehand on this prepared data. This preliminary action creates a ready-to-use generative model that can quickly produce synthetic training data for various failure scenarios without requiring time-consuming data collection during actual AI module development or deployment.

Instead of collecting extensive real data for every possible failure scenario, the system creates synthetic copies through the generative model. Once the model is trained on available data, it can rapidly generate synthetic training samples for any failure mode, eliminating the need for time-consuming real-world data collection for each scenario while maintaining training comprehensiveness.