A Neural Network Approach to Tool Wear Detection via Infrared Sensor Monitoring

The modern manufacturing sector encounters substantial financial setbacks due to unforeseen equipment downtime, hence mandating the implementation of more proactive maintenance strategies. The present study explores the potential of utilising AI-driven thermal imaging to monitor tool conditions in the context of predictive maintenance. In the context of the milling process, infrared camera technology was implemented to observe and assess tool wear and surface finishes. The analysis of key features derived from thermal imaging was conducted using two distinct methodologies: statistical analysis and polynomial feature extraction. Subsequently, a neural network was trained to categorise tools as either “fresh” or “worn”. This study provides a comparative examination of feature extraction techniques, focusing on the significant contributions of neural network and thermal imaging in enhancing predictive maintenance in the manufacturing sector. The study demonstrates that both statistical and polynomial feature extraction methods are effective for tool condition monitoring, with statistical features showing marginally higher success rates across various regions of interest, underscoring their reliability in predictive maintenance applications.