Ever feel like you’re playing ‘spot the difference’ with engineering drawings, meticulously comparing revisions to catch errors? It’s a tedious process, and mistakes can slip through. That’s why I’ve been exploring how AI can revolutionize drawing revision analysis, and I want to share my progress.

The Problem: Manual Revision Checks Are Time-Consuming and Error-Prone

In large engineering companies, drawing exchange happens within ISO 19650 compliant common data environments (CDEs) like ProjectWise. This means every drawing goes through multiple revisions, each requiring thorough checks by content checkers, technical checkers, and approvers. Each time an error is found, a new revision is created. This is a great process for quality, but:

It’s time consuming.
It relies on human eyes, which are prone to fatigue and oversight.
Identifying trends in errors across multiple projects is nearly impossible manually.

The Solution: AI-Powered Revision Analysis

My idea is to leverage AI to automate the error-spotting process and identify common mistakes across drawings. Here’s the approach:

Vision Processing with LLMs: Feed drawing revisions (as images) into a vision processing large language model (LLM). This LLM identifies and describes the elements within the drawing.
Difference Detection: Compare the LLM’s descriptions of two drawing revisions to pinpoint the changes. These differences are stored as text, along with a description of the drawing itself.
Unsupervised Machine Learning: Run a feature selection, unsupervised machine learning algorithm on the collected data from multiple drawings. This identifies correlations between specific drawing features and common errors.
Predictive Error Detection: Once the model is trained, it can analyze new drawings and flag potential errors based on the presence of error-prone features.

The Tech Stack & Workflow

This project involves a combination of technologies and a carefully designed workflow:

Vision Processing LLM (NVIDIA Describe Anything): This model provides the detailed image analysis needed to understand drawing content.
Common Data Environment (CDE) Integration: The system needs to seamlessly integrate with CDEs like ProjectWise or Autodesk Construction Cloud (ACC) to access drawing revisions. I’ve been working on incorporating ACC by creating specific tasks, breaking tasks down into briefs so they can be isolated and peer reviewed on their own. For example, tasks include how the user can initiate the workflow on the frontend, how to take the user through the authentication process, how to collect the files needed from ACC, how to collect the metadata needed from ACC, how to do both of these functions in bulk if needed, and then to integrate all of these processes into the larger process already in place on the project. All of these processes are documented in Miro process flow diagrams.
Feature Selection Algorithm: I will use this algorithm to correlate mistakes with features on drawings.

Challenges and Lessons Learned

This project is still under development, and there have been some interesting challenges:

User Story Specificity: It’s crucial to create very specific user tickets when working in a team. Vague instructions lead to unexpected results. I’ve learned to break down tasks into smaller, well-defined tickets with clear briefs.

The Future of Drawing Checks

This AI-powered approach promises to significantly improve the accuracy and efficiency of drawing revision analysis. By automating error detection and identifying common mistakes, we can:

Reduce errors and delays.
Improve drawing quality.
Free up engineers’ time for more complex tasks.

Could this technology help your firm?

If you’re interested in exploring how AI can streamline your engineering workflows, Construct Digitally can help. I specialize in building custom web apps, automating document workflows, and digitizing QA processes. Let’s discuss how to bring similar AI-powered solutions to your organization.

Loading...