Augmentation of Multimodal Time Series Data for Training Machinelearning Models

Simple SummaryContent extracted from patent full text and abstract with AI.

This invention introduces a system and method for generating synthetic multimodal time series data to improve the training of machine learning models in predicting industrial, time-dependent processes. By using generative neural network architectures, such as multimodal variational autoencoders combined with recurrent neural networks, it creates synthetic training data that reflect both process conditions (e.g., temperature, pressure) and key performance indicators (KPIs). This synthetic data augments real-world datasets, especially when real data is scarce or expensive to collect, enabling better generalization and robustness in predictive models.

Use CasesContent extracted from patent full text and abstract with AI.

• Augmenting small or imbalanced industrial datasets for machine learning applications.
• Predicting equipment degradation or failure in chemical plants and other industrial settings.
• Load forecasting for energy grids or other utilities with limited historical data.
• Battery health and discharge forecasting under varying conditions.
• Imputation of missing sensor modalities or KPIs in industrial process monitoring.
• Simulating rare or extreme operating conditions that are hard to observe in reality.
• Optimizing production processes by anticipating future performance based on simulated conditions.

BenefitsContent extracted from patent full text and abstract with AI.

• Mitigates overfitting in machine learning models by expanding the training dataset with realistic synthetic data.
• Enables data-driven modeling where real-world data collection is limited, costly, or slow.
• Handles multiple data modalities and can generate data even with missing modalities.
• Allows conditional data generation, providing flexibility to simulate specific scenarios or conditions.
• Improves generalization of predictive models to new or unseen data, especially in the presence of covariate shifts.
• Supports both regression and classification forecasting tasks for time series data.
• Enhances the robustness and performance of industrial monitoring, control, and maintenance prediction systems.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

The present invention relates to training predictive data-driven model for predicting an industrial time dependent process. A data driven generative model is introduced for modelling and generating complex sequential data comprising multiple modalities, by learning a joint time-dependent representation of the different modalities. The model may be configured to handle any combination of missing modalities, which enables conditional generation based on known modalities, providing a high degree of control over the properties of the generated sequences.