Using cutting temperature and chip characteristics with neural network BP and LSTM method to predicting tool life

The die steel NAK80 is used in specular optical molds, deep drawing forming dies, and cold extrusion dies in large quantities; high strength and hardness often induce tool wear during machining. This study established a tool wear prediction method for measuring, using the cutting temperature and chip chromaticity characteristic values to predict the tool life. The back propagation neural network (BP-LM) was compared with a long-short term memory (LSTM) model in the prediction method, and different characteristic signals were imported into the BP-LM and LSTM methods to predict the tool wear. In Taylor’s curve diagram, the repeatability accuracies of tool wear and cutting temperature are 2.83% and 9.29%, respectively. The BP-LM method was used for prediction in the comparison of prediction methods. When the input characteristic were temperature, chip chromaticity, and temperature and chip chromaticity, the MAPE percentage errors are 24.23%, 31.87%, and 19.88%, respectively. The error was reduced by 29% when the input characteristics were temperature and chip chromaticity. When the LSTM model was used for prediction, and the input characteristics were temperature, chip chromaticity, and temperature and chip chromaticity, the MAPE percentage errors are 30.33%, 28.55%, and 22.1%, respectively. The error was reduced by 25% when the input characteristics were temperature and chip chromaticity. Therefore, using the characteristic temperature and chip chromaticity in the BP-LM and LSTM prediction models resulted in good forecast accuracy, and a new model prediction form for tool life was provided.