Modeling and prediction for diesel performance based on deep neural network combined with virtual sample

Area of Science:

• Mechanical Engineering
• Computational Intelligence

Background:

• Accurate performance models are crucial for diesel engine condition monitoring and fault diagnosis.
• Traditional large-scale experimental methods for model construction are costly.

Purpose of the Study:

• To develop a cost-effective method for modeling marine diesel engine performance.
• To enhance the accuracy of performance models using deep neural networks and virtual sample generation.

Main Methods:

• Selected input factors: speed, power, lubricating oil temperature, and pressure.
• Utilized virtual sample generation technology to expand limited experimental data.
• Employed deep neural network (DNN) for performance model construction.

Main Results:

• Achieved an overall prediction accuracy exceeding 93%.
• Identified power as the key factor influencing brake specific fuel consumption (30% weighting).
• Determined speed as the primary factor affecting vibration (30%) and noise (30.5%).

Conclusions:

• The proposed DNN-based method with virtual samples effectively models marine diesel engine performance.
• The approach significantly reduces experimental costs while ensuring high prediction accuracy.
• Provides insights into the parametric influence on engine performance metrics.