Bayesian neural networks modeling for tool wear prediction in milling Al 6061 T6 under MQL conditions

• Published in: International journal of advanced manufacturing technology 2024-11, Vol.135 (5-6), p.2777-2788
• Main Authors: Airao, Jay, Gupta, Abhishek, Nirala, Chandrakant K., Hsue, Albert Wen-Jeng
• Format: Article
• Language: English
• Subjects: Algorithms; Aluminum base alloys; Artificial intelligence; Bayesian analysis; CAE) and Design; Computer-Aided Engineering (CAD; Cutting wear; Deep learning; Engineering; Flood predictions; Industrial and Production Engineering; Machinability; Machine learning; Mechanical Engineering; Media Management; Milling (machining); Neural networks; Original Article; Real time; Signal quality; Surface properties; Surface roughness; Tool wear; Uncertainty
• ISSN: 0268-3768; 1433-3015
• DOI: 10.1007/s00170-024-14678-2

The integration of artificial intelligence, machine learning, and deep learning algorithms into machining processes has made them more intelligent, significantly reducing costs, improving production rates, and enhancing product quality by accurately predicting machining responses. In this study, a Bayesian neural network (BNN) is employed to predict tool wear during the milling of Al6061 T6 alloy, showcasing the novelty of BNN in handling uncertainty and providing reliable predictions. Milling experiments were conducted at three different spindle speeds under dry, flood, and minimum quantity lubrication (MQL) strategies. Machinability was evaluated by considering tool wear, milling forces, and surface quality. Unique to this study is the use of force and current signals as input to the BNN model, capturing real-time data to estimate tool wear. The signals were trained and tested to predict tool wear under varying cutting conditions. The results indicated that tool wear in dry conditions was primarily due to adhesion, leading to higher milling forces and poorer surface quality. In comparison, the wet and MQL conditions resulted in 11–21% and 9–13% less tool wear, respectively, than dry conditions, alongside improved surface roughness and reduced machining forces. The BNN model demonstrated its ability to avoid overfitting, providing highly accurate predictions with an error margin of 2–15% when compared to experimental results. Unlike conventional models, the BNN accounts for prediction uncertainty, making it more robust and reliable across different datasets. Thus, the proposed BNN model proves its effectiveness and generalizability in predicting tool wear under various machining conditions, setting a new benchmark for the application of artificial intelligence in machining processes.