DeepMachining: Online Prediction of Machining Errors of Lathe Machines

TL;DR

DeepMachining tackles online prediction of lathe machining errors by learning a pre-trained model over tool-life data and adapting to new tasks with two-shot, adapter-based fine-tuning. The architecture features a Dual Signal Encoder with Stem, Downsampling, and Dilated Inception, followed by a second Dilated Inception and a Projection Head to estimate the machining error $y$. Experiments on real factory data across multiple datasets show the method surpasses SVR, 1D-CNN, and 2D-CNN baselines in RMSE, MAE, and CORR, and demonstrate robust quick adaptation to tool/workpiece changes with only about 6.5% of parameters updated. The approach uses a compact model (~260k parameters) suitable for deployment on industrial hardware, and points to practical impact for CNC quality control and future expansion to broader tasks and process logs via LLMs.

Abstract

We describe DeepMachining, a deep learning-based AI system for online prediction of machining errors of lathe machine operations. We have built and evaluated DeepMachining based on manufacturing data from factories. Specifically, we first pretrain a deep learning model for a given lathe machine's operations to learn the salient features of machining states. Then, we fine-tune the pretrained model to adapt to specific machining tasks. We demonstrate that DeepMachining achieves high prediction accuracy for multiple tasks that involve different workpieces and cutting tools. To the best of our knowledge, this work is one of the first factory experiments using pre-trained deep-learning models to predict machining errors of lathe machines.

Figures (6)

• Figure 1: The Process of DeepMachining
• Figure 2: The experimental environment of the CNC lathe machine
• Figure 3: The Structure of The Core of DeepMachining
• Figure 4: Scatter plots for the actual machining errors vs. the ones estimated by different methods on our Pre-trained Dataset.
• Figure 5: Scatter plots for the actual machining errors vs. the ones estimated by different methods on our adapted Dataset.