Multiobjective Deep Belief Networks Ensemble for Remaining Useful Life Estimation in Prognostics.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Industrial Engineering

Background:

• Condition-based maintenance (CBM) is crucial for industrial safety and efficiency.
• Estimating remaining useful life (RUL) is key for CBM, with neural networks showing promise.
• Current neural network approaches for RUL estimation face limitations due to manual feature engineering and parameter tuning.

Purpose of the Study:

• To propose a novel multiobjective deep belief networks ensemble (MODBNE) method for improved RUL estimation.
• To address the performance limitations of existing neural network methods in RUL prediction.
• To enhance the reliability and efficiency of CBM in industrial settings.

Main Methods:

• Developed a multiobjective deep belief networks ensemble (MODBNE) integrating evolutionary algorithms with DBN training.
• Employed a multiobjective evolutionary algorithm to simultaneously optimize multiple DBNs for accuracy and diversity.
• Utilized a single-objective differential evolution algorithm to optimize ensemble weights for RUL estimation.

Main Results:

• Evaluated the MODBNE method on several benchmark prognostic datasets.
• Compared the proposed method against existing RUL estimation approaches.
• Experimental results demonstrated the superiority of the MODBNE method.

Conclusions:

• The proposed MODBNE method offers a significant advancement in RUL estimation.
• This approach enhances the effectiveness of condition-based maintenance strategies.
• The method shows potential for improving safety, efficiency, and reliability in industrial applications.