Ensemble deep learning with multi-objective optimization for prognosis of rotating machinery.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Prognostics and Health Management (PHM)

Background:

• The Internet of Things and smart sensing generate vast amounts of data for PHM systems.
• Accurate Remaining Useful Life (RUL) prediction is crucial for mechanical component maintenance but remains challenging.
• Existing methods often struggle to balance prediction accuracy and model diversity.

Purpose of the Study:

• To propose a novel ensemble deep learning with multi-objective optimization (EDL-MO) method for RUL prediction.
• To improve RUL prediction accuracy by combining ensemble learning with diversity enhancement.
• To validate the EDL-MO method's effectiveness using experimental bearing data.

Main Methods:

• Developed a novel ensemble deep learning algorithm integrating accuracy and diversity.
• Employed evolutionary optimization to concurrently optimize diversity and prediction accuracy.
• Conducted bearing run-to-failure experiments, collecting vibration signals for RUL prediction.

Main Results:

• The EDL-MO method demonstrated superior performance in RUL prediction compared to existing algorithms.
• Introducing diversity into the ensemble learning process reduced uncorrelated errors and improved prediction.
• Experimental validation confirmed the effectiveness and superiority of the proposed EDL-MO approach.

Conclusions:

• The EDL-MO method offers a significant advancement in RUL prediction for rotating machinery.
• Combining ensemble deep learning with multi-objective optimization effectively addresses the challenge of accurate RUL estimation.
• This approach holds promise for enhancing predictive maintenance strategies in industrial applications.