TECH NEED

Seeking Automated Corrosion Grading and Dimensional Measurement for Construction Steel

KEY INFORMATION

TECHNOLOGY CATEGORY:
Infocomm - Video/Image Analysis & Computer Vision
Infocomm - Artificial Intelligence
TECHNOLOGY READINESS LEVEL (TRL):
LOCATION:
Singapore
ID NUMBER:
TN174440

Background/Description

In the construction industry, foundation steel solutions are used in a variety of applications ranging from waterfront to underground development projects. The application conditions typically lead to a deterioration in the steel condition following some duration in service. Particularly for temporary structures, such steel structural items (e.g., sheet piles, beams, channels) when retrieved from use may vary in condition, exhibiting signs of deterioration such as rust, pitting or dents. It is desired to be able to rapidly and effectively assess and characterise the steel items’ condition in order to determine their re-usability or disposal.

A Singapore-based steel solutions company is seeking solvers capable of developing an AI-enabled mobile app solution to capture, process and analyse the condition of steels, for the following tasks (in order of priority):

  1. Rust grading – Capture, process and analyse photos of physical structural steel items (e.g., sheet piles, beams, channels). Grade the rust condition in accordance with a pre-determined tiering
  2. Counting – To accurately count the number of physical steel items housed within a storage yard. Items are largely standardized parts but may comprise of several different types stored in a stack/pile
  3. Dimensional Measurement and Geometrical Analysis – To measure the external dimensions of the physical steel items, and identify the presence of defects such as warpage, deflections, pitting

The preferred collaboration mode is to co-develop a POC solution for evaluation followed by operational scale-up.

Technology Specification

Task #1 – Rust Grading [priority]

  1. To carry out image analysis of structural steel items via a mobile app
  2. The image processing and analysis may either be carried out on an Android/iOS mobile device (preferred) or via batch uploading to the cloud for cloud-based analysis
  3. Identify and characterise defects present based on a tiered criteria (4 to 5 grades) which will be provided by the Tech Seeker. (Typical conditions to be identified include rust conditions as well as deformities such as pitting, dents, cracks)
  4. Solution provider to advise on training and evaluation dataset requirements for AI model development
  5. Tech seeker will work with solution provider on data annotation for images dataset
  6. Expected application condition to be outdoors under natural lighting
  7. Target accuracy of 95%, operable with images captured at a maximum distance of 0.5m (target accuracy to be achievable following a mutually agreed period of AI model training)
  8. Measurement data to be transmitted and recorded to relevant storage means (e.g. RFID tags provided by the Tech Seeker)

Task #2 - Counting

  1. To carry out automated counting on physical construction steel items via a mobile app
  2. Processing and analysis may either be carried out on an Android/iOS mobile device (preferred) or via the cloud
  3. Measurement data to be transmitted and recorded to relevant storage means (e.g., RFID tags provided by the Tech Seeker)
  4. The app is not expected to be used for counting when items are partially or completely occluded
  5. Expected application condition to be outdoors under natural lighting
  6. Typical scenario may be to count the number of sheet piles stacked vertically in multiple columns 
  7. Target F1 score of 0.95, operable at a distance of 1.5m – 2m from items (target F1 score to be achievable following a mutually agreed period of AI model training)

Task #3 – Dimensional measurement & Geometrical Analysis

  1. To carry out automated dimensional measurements on physical construction steel items via an AR-based measurement app
  2. Besides measurement of physical dimensions, app to also identify (and quantify) defects such as warpage, deflections, torsional deviations etc.
  3. Expected viewpoints are expected to be cross-sectional (perpendicular) and orthogonal
  4. Target accuracy of 95%, operable at a distance of 1.5m – 2m from items (target accuracy to be achievable following a mutually agreed period of AI model training)
  5. Measurement data to be transmitted and recorded to relevant storage means (e.g. RFID tags provided by the Tech Seeker)

WHAT WE ARE NOT INTERESTED IN

  • Academic research or early-stage technologies.
  • Predictive Maintenance or condition monitoring sensors and technologies

Preferred Business Model

  • IP Acquisition
  • R&D Collaboration