Comparative Analysis of Machine Learning Techniques for Identifying Multiple Force Systems from Accelerometer...

Area of Science:

• Structural Engineering
• Mechanical Engineering
• Computational Mechanics

Background:

• Assessing structural integrity requires understanding applied forces, often unfeasible to measure directly.
• Solving inverse problems for force identification is crucial for predicting component lifespan.
• Machine learning (ML) offers powerful predictive capabilities for complex engineering challenges.

Purpose of the Study:

• To evaluate the efficacy of various ML models in identifying multi-force system parameters.
• To assess ML model performance based on prediction accuracy and processing time.
• To apply ML for predicting force position, frequency, magnitude, and quantity on structures.

Main Methods:

• A finite element method (FEM) computational model was developed and validated with experimental accelerometer data.
• Response surface methodology and Design of Experiment (DOE) were employed to create a comprehensive database.
• Six distinct ML models were trained and tested on the generated dataset for force parameter identification.

Main Results:

• k-Nearest Neighbors (k-NN) achieved a prediction error as low as 0.013%.
• Random Forests demonstrated a maximum prediction error of 0.2%.
• Both models showed high accuracy in predicting force parameters like position, frequency, and magnitude.

Conclusions:

• ML models, particularly k-NN and Random Forests, are highly effective for identifying parameters of multi-force systems.
• The proposed method offers an innovative approach to structural health monitoring and diagnostics.
• Accurate force parameter identification enhances the assessment of structural component useful life.