Metal Crack Length Prediction and Sensor Fault Self-Diagnosis Method Based on Deep Forest

Area of Science:

• Mechanical Engineering
• Materials Science
• Computational Mechanics

Background:

• Fatigue loading causes cracks in metal structures, impacting structural integrity and lifespan.
• Accurate prediction of crack length is essential for ensuring structural safety and performance.
• Crack length significantly influences local strain distribution within a structure.

Purpose of the Study:

• To develop an accurate method for predicting crack length in metal structures using strain data.
• To implement a Deep Forest model for optimizing data training and prediction accuracy.
• To propose a self-diagnostic method for strain sensors to enhance monitoring reliability.

Main Methods:

• Finite Element Analysis (FEA) was used to obtain strain data from compressive and tensile (CT) specimens under various loading conditions.
• A Deep Forest (DF) model was employed for optimizing the training of strain data for crack length prediction.
• Compensation was applied to dynamic strain data, and multi-dimensional input signals in the XY plane were utilized for prediction.

Main Results:

• The study successfully predicted crack length using multi-dimensional input signals in the XY plane.
• A novel self-diagnostic coefficient for strain sensors was proposed, based on the Pearson correlation coefficient.
• The combined DF model and self-diagnostic coefficient demonstrated enhanced intelligence in crack state monitoring.

Conclusions:

• The proposed method accurately predicts crack length in metal structures by leveraging FEA and DF modeling.
• The developed self-diagnostic strain sensor capability improves the reliability of crack monitoring systems.
• These advancements contribute to a higher level of intelligence in structural health monitoring for fatigue crack propagation.