Multi-Task Data Imputation for Time-Series Forecasting in Turbomachinery Health Prognostics

Time-series forecasting is the core of prognostics and health management (PHM) turbomachinery. However, missing data often occurs due to several reasons, such as failure sensors. These partially irregular greatly affect quality time-series modeling prediction most models assume that are sampled uniformly over time. Meanwhile, training process requires a large number samples Due various it difficult obtain significant amount monitoring data, PHM turbomachinery has high timeliness accuracy requirements. To fix these problems, we propose multi-task Gaussian (MTGP)-based imputation method leverages knowledge transfer across multiple sensors even equipment. Thereafter, adopt long short-term memory (LSTM) neural networks build based on imputed data. In addition, model integrates methods denoising dimensionality reduction. The superiority this integrated framework, termed MT-LSTM, been verified in scenarios synthetic dataset real case.