Rapid G-Coding – Obtaining G-code Using AI

Abstract

The traditional process of creating G-code is time-intensive and requires expertise in CNC programming. Recently, several AI software have appeared. They communicate using prompts, and the user can also provide a drawing, which the AI processes. AI software, such as ChatGPT, Bing AI Copilot, became publicly available approximately two years ago. The question arises whether G-code can be obtained using these software. A review of the literature confirmed that no studies have addressed the application of these software tools for generating G-code. In this paper the input in AI software is a textual description of the part material and tool, the dimensions of the blank, and what details on the part need to be processed. In addition to the text, a drawing of the part with all necessary dimensions is given as input. The output is G-code. Testing of the code was performed: comparing with expert-created G-code to benchmark performance and running the generated G-code on virtual CNC machines. Based on the analysis of dozen parts and three different software tools, it was concluded that at the current stage of AI software development, the generated G-codes contain about 20% incorrect commands, requiring human review and correction.