Feature Extraction AI For Engineering Drawings

Challenge

Our client, a large metal parts manufacturer, provides custom metal components manufacturing services for a wide range of industries. Based on technical drawings sent by their customer, our client assesses part complexity and prepares quotes.

Looking for a way to increase quitting accuracy with the information from technical drawings and simplify metal part production process, our client approaches us to create an AI system for extracting price-driving features from engineering drawings.

The abundance of special symbols used in any technical drawing, like diameter symbols and leader arrows, variable drawing layout and a complex nature of multipage technical drawings makes this project a real challenge.

Solution

Our machine learning team has created an AI engineering drawing recognition system capable of extracting core information from the drawing: title block data, part measurements, tolerances and fits.

The solution is powered by multiple machine learning models, making it a versatile solution for processing engineering drawings of any layout and complexity.

Title Block Detection

Our solution is capable of extracting data from non-standard title blocks, including data from merged cells and cells with text overflow, processing title blocks compromised by the scanning process (e.g. low contrast image), and special symbols used to define angle projection, surface finish, and more.

Geometric dimensioning and tolerancing (GD&T)

Our solution is capable of detecting geometric dimensioning and tolerancing annotations independent of their positioning on the page, rotation and formatting. It captures a full range of geometric dimensioning and tolerancing symbols used in engineering drawings of metal parts, as well as interprets them and returns them in a standard data model.

Measures Extraction

Part measures are the baseline of any technical drawing, determining its overall size. Our system detects measures on a drawing and converts it into a traditional data format, allowing direct access to nominal size, tolerance, thread and chamfer information. Our system also detects which part of the drawing the measure refers to, ensuring precise data extraction and interpretation.

Hatching Detection

In engineering drawings, hatchings are used to indicate materials. Complex parts often consist of various materials, which need to be taken into consideration when preparing a quote.

Our technical drawing processing system is capable of detecting various types of hatchings and interpreting them.

Results

Our engineering drawing feature extraction system is in a soft launch stage: our client has started to automate technical drawing processing and quotation processes and is in the process of implementing the software into their digital ecosystem. We have received overwhelming positive feedback so far and continue to add new features to support a wider range of drawings.