Date: 23 September 2019
Time: 12:30-4:30 PM

Deep learning has recently achieved significant breakthroughs in many different domains, including computer vision, language processing, genomics, and speech recognition; e.g., AlphaGo and AlphaZero have achieved super-human performance in complex games without human input. Despite these encouraging results, these techniques have seen little adoption by industry for PHM applications. There are several obstacles that need to be surmounted to enable the broad adoption of deep learning for PHM:

• Limited number of representative training samples, particularly for different types of faulty conditions and representative time-to-failure trajectories
• Appropriate benchmark datasets to compare the progress of newly developed algorithms
• Variability of operating and environmental conditions to appropriately transfer the learnt patterns between different operating conditions
• Heterogeneity of condition monitoring signals, system configurations, and operating conditions

This half-day workshop on the afternoon of 23 September will provide a forum for PHM researchers and practitioners to discuss the potential, applicability, benefits, challenges, and current obstacles of deep learning for PHM applications. The focus will be on theory and application of deep learning to anomaly detection, condition monitoring, diagnostics, and prognostics.

For more information, please contact Neil Eklund.

Agenda