From Signals to Remaining Useful Life: Multimodal Sensor Fusion for Fault Diagnosis and Prognostics-Methods, Pitfalls, and Reporting Standards.

This project studies how to combine different sensor signals to diagnose faults and predict the Remaining Useful Life (RUL) of complex aerospace structures and systems. It brings together data like vibration, temperature, and acoustic emissions and tackles three main problems: noisy sensors, misaligned data, and different sampling rates. The work gives step‑by‑step guidance on building fusion systems. It splits the approaches into three groups: * **Data‑level fusion** – raw sensor data are merged before any processing. * **Feature‑level fusion** – each sensor’s features are extracted first, then combined. * **Decision‑level fusion** – separate models make predictions, and their outputs are merged. A large part of the study points out common

mistakes. These include data leakage (using future information in training), using validation methods that don’t reflect real use, and the absence of standard metrics for reporting predictive‑maintenance results. Using simulations, the project creates reliable preprocessing pipelines and deep‑learning fusion models. It tests temporal convolutional networks and recurrent networks to learn the complex time‑varying patterns that indicate degradation. In the end, the research aims to set up standard reporting guidelines. Those guidelines should make prognostic models more reproducible, transparent, and trustworthy for safety‑critical aerospace tasks, helping engineers turn raw sensor data into clear health‑management decisions.

Message: Hi FE, I need help with “From Signals to Remaining Useful Life: Multimodal Sensor Fusion for Fault Diagnosis and Prognostics-Methods, Pitfalls, and Reporting Standards.” in “Aerospace Engineering”