Aalto Gear Fault datasets for deep-learning based diagnosis.

Area of Science:

• Mechanical Engineering
• Data Science
• Artificial Intelligence

Background:

• Accurate system health state prediction using deep learning is hindered by insufficient and varied real-world data.
• Developing robust fault diagnosis models requires extensive datasets that capture diverse operational conditions and failure modes.

Purpose of the Study:

• To introduce two novel, extensive datasets: the Aalto Shim Dataset and Aalto Gear Fault Dataset.
• To provide valuable resources for advancing deep learning-based fault diagnosis in mechanical systems.
• To facilitate the development and testing of more generalized and robust intelligent fault diagnosis models.

Main Methods:

• Collected data under controlled laboratory conditions on a downsized azimuth thruster testbench.
• Included a wide range of gear faults, encompassing both synthetic and realistic failure modes.
• Utilized multiple sensors to capture data across various fault types, severities, and operating conditions.

Main Results:

• The datasets offer comprehensive data on gear faults under diverse scenarios.
• Methodologies for creating synthetic faults and replicating common gear failures are detailed.
• The collected data serves as a valuable resource for training and validating deep learning models.

Conclusions:

• The Aalto Shim and Aalto Gear Fault Datasets address the critical need for extensive data in deep learning fault diagnosis.
• These resources will significantly contribute to enhancing the generalization and robustness of intelligent diagnostic systems.
• The datasets enable researchers to develop and test advanced fault diagnosis models more effectively.