Thermal Behavior Modeling Based on BP Neural Network in Keras Framework for Motorized Machine Tool Spindles

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Thermal Management

Background:

• High-speed motorized spindles generate significant heat, impacting performance and lifespan.
• Accurate thermal behavior prediction is crucial for optimizing spindle operation and preventing failures.
• Existing thermal modeling methods may require extensive datasets or complex simulations.

Purpose of the Study:

• To develop and evaluate artificial neural network (ANN) models for predicting the thermal behavior of high-speed motorized spindles.
• To investigate the influence of network topology, learning parameters, and validation techniques on ANN model performance.
• To analyze the impact of various working conditions, including coolant type, on spindle thermal behavior.

Main Methods:

• Development and evaluation of multi-output regression ANNs using the Keras deep learning framework.
• Training and testing ANNs with a small input-output dataset, including data previously unseen by the network.
• Systematic analysis of network topology (hidden layers, neurons) and learning parameters.
• Simulation and analysis of the effect of working conditions on spindle thermal behavior.

Main Results:

• ANN models achieved high prediction accuracy for spindle thermal behavior, ranging from 95% to 98%.
• The developed ANN models accurately predicted spindle temperature under various working conditions using limited data.
• A significant influence of coolant type on spindle unit temperature was identified, with intensive water cooling showing a pronounced effect.

Conclusions:

• ANNs are effective tools for accurately predicting the thermal behavior of high-speed motorized spindles, even with small datasets.
• The findings highlight the critical role of coolant selection and application in managing spindle thermal performance.
• This approach offers a computationally efficient method for thermal analysis and optimization of motorized spindles.