@

초록

The deployment of on-line tool condition monitoring systems and machine tool diagnostics is essential for achieving sustainable manufacturing systems. Accurate tool wear prediction enables automated tool changes by comparing real-time wear with the predefined tool life limit, thereby improving production efficiency and reducing production costs. This study introduces a multi-task learning model that simultaneously estimates tool wear states and cutting force. This approach enhances both monitoring efficiency and accuracy. Cutting force serves as a key indicator, reflecting tool condition, machining stability, and surface quality. This makes real-time monitoring the cutting force essential to optimize processes and prevent failures. The mean and variance of the cutting forces are designated as target variables and paired with tool wear data, forming a combined target set to improve the model's predictive performance. We collected the data from multiple sensors and the CNC system, with feature extraction techniques applied to derive meaningful information. To validate the multi-task learning approach, Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Temporal Convolutional Networks (TCNs) are utilized for model development. Performance metrics-accuracy, capacity (the number of parameters), and Floating Point Operations (FLOPs)-are compared between single-task and multi-task learning strategies to assess their effectiveness and efficiency. The study highlights the importance of correlation between the paired target variables in multi-task learning, showing its significant impact on model accuracy across different architectures. This research demonstrates the effectiveness of multi-task learning for machining processes and offers an optimal strategy for constructing such models.

키워드