Machine-Learning Drawing Title Identification Across Varying Layouts

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing construction management software requires manual entry and is prone to errors when assigning data attributes to electronic drawing files, particularly for the title, due to rigid and inflexible rules-based analyses that fail to accurately identify title information across varying drawing layouts.

Innovation Solution

A machine-learning model is employed to predict whether a given textual element is the title of a drawing by analyzing a set of data variables including spatial, linguistic, and relational information, reducing the need for manual verification and improving accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If manual entry is used to assign data attributes to electronic drawing files, then flexibility and adaptability to varying drawing layouts are improved, but productivity and accuracy are worsened due to time-consuming manual processes and human error

Engineering Contradiction:

Improveadaptability to varying drawing layoutsVSAvoidproductivity in assigning data attributes

Core Design Contradiction:

Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent replaces manual mechanical entry processes with an automated machine-learning-based system. The machine-learning model automatically analyzes drawing images, extracts textual elements, and assigns data attributes without human intervention, thereby maintaining adaptability to varying layouts while dramatically improving productivity

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system performs self-service by automatically analyzing and processing drawing files independently. The machine-learning model autonomously extracts information, identifies textual elements, and populates data attributes without requiring manual input, enabling the system to serve itself in the data extraction and assignment process

Inventive Principle:

Principle #25Self-service

2Ease of operation

If rules-based analysis is used to identify title information, then ease of operation is improved through automated processing, but measurement precision and reliability are worsened due to rigid rules failing to accurately identify titles across varying layouts

Engineering Contradiction:

Improveease of automated processingVSAvoidprecision in identifying title information

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The patent changes the underlying parameter of analysis from rigid rules to machine-learning models that can adapt to varying patterns. The machine-learning model analyzes multiple parameters including spatial position, textual content, and contextual relationships to accurately identify titles, overcoming the limitations of fixed rules while maintaining automated operation

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The system incorporates feedback mechanisms where the machine-learning model continuously learns from and adapts to variations in drawing layouts. The model processes feedback from the complex patterns in drawing images to refine its title identification accuracy, enabling both ease of operation and high measurement precision

Inventive Principle:

Principle #23Feedback

3Reliability

If manual verification is performed to correct errors in data attribute assignment, then reliability is improved by catching errors, but loss of time increases due to required manual review and correction

Engineering Contradiction:

Improvereliability of data attribute assignmentVSAvoidtime required for manual corrections

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent substitutes manual verification processes with automated machine-learning-based validation. The system automatically verifies data attribute assignments by analyzing the consistency and accuracy of extracted information, thereby maintaining high reliability while eliminating the time loss associated with manual review and correction

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS20250218208A1Machine-Learning-Based Identification of Drawing Attributes

Publication Date: 2025.07.03 PROCORE TECHNOLOGIES INC

AI summary

An example computing system is configured to: (i) access a drawing associated with a construction project; (ii) identify, in the drawing, a set of candidate textual elements that potentially represent a title of the drawing; (iii) for each candidate textual element, (a) determine a respective dataset comprising values for a set of data variables that are potentially predictive of whether the candidate textual element is the title of the drawing, and (b) input the respective dataset into a machine-learning model that functions to (1) evaluate the respective dataset and (2) output, based on the evaluation, a respective score indicating a likelihood that the candidate textual element represents the title of the drawing; and (iv) based on the respective scores for the candidate textual elements that are output by the machine-learning model, select one given candidate textual element as the title of the drawing.