Confidence Interval Estimation for Cutting Tool Wear Prediction in Turning Using Bootstrap-Based Artificial Neural...

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Cutting tool degradation poses a significant challenge in machining, leading to increased operational costs due to unpredictable wear.
• Optimal tool replacement is crucial for efficiency, but variability in degradation complicates timely decisions.

Purpose of the Study:

• To develop and validate a cutting tool degradation monitoring technique using bootstrap-based artificial neural networks.
• To accurately estimate flank wear (VB) and provide confidence intervals for predictive tool replacement.

Main Methods:

• Utilized input indicators from turning operations: RMS values of cutting force and torque, machining duration, and total machined length.
• Employed artificial neural networks with a specific architecture: two hidden layers (6 neurons with Tanh, 6 neurons with ReLu).
• Incorporated a bootstrap approach to generate confidence intervals for wear predictions.

Main Results:

• The proposed neural network model accurately tracks cutting tool degradation and detects end-of-life.
• The confidence interval effectively estimates prediction variability, aiding in optimal tool replacement strategies.
• Achieved best results with a two-hidden-layer network (Tanh and ReLu activation functions).

Conclusions:

• Bootstrap-based artificial neural networks offer a robust solution for monitoring cutting tool degradation.
• The developed technique enables timely and accurate tool replacement, optimizing machining operations and reducing costs.
• Confidence intervals enhance decision-making for predictive maintenance and tool management policies.