Real-Time Thermal Spray Process Monitoring Using Convolution Neural Network Deep Learning Architectures

Thermal spray is essential for surface modification and coating of materials but is challenging to monitor in real-time due to high velocities, temperatures, and continuous torch or part movement. Existing static monitoring equipment cannot track the torch, making it difficult to ensure optimal process parameters. This study demonstrates that a specific convolutional neural network architecture, which divides the image into a grid and predicts bounding boxes and class probabilities for each cell, can accurately monitor High-Velocity-Oxy-Fuel thermal spray processes in real-time. The model is trained end-to-end to minimize errors in bounding box locations and class predictions, enabling one-pass object detection. This approach achieves up to 90% accuracy for all classes and processing speeds of up to 50 ms per frame on a modern smartphone, highlighting its potential as a practical, easily adoptable solution for real-time thermal spray process monitoring. This can have a significant impact on the thermal spray industry by enabling even small spray shops to achieve high-quality outcomes due to its simplicity and potential for wide adoption.