Customer Stories
/
AI-Powered Corrosion Detection for Offshore Oil & Gas Inspections

Energy

AI-Powered Corrosion Detection for Offshore Oil & Gas Inspections

Automated detection of surface rust and coating failures using computer vision and drone data

99%

Detection accuracy for rust vs. non-rust surfaces

60%

Reduced manual inspection cost

100%

Coverage of hard-to-reach areas via drone automation

Customer Stories

Trained on a diverse, real-world dataset, the models were optimized for deployment on autonomous drones and seamlessly integrated into the client’s digital twin platform for real-time, remote inspections.

Energy

Industry

USA

Location

Computer Vision, Machine Learning, Model Development, Data Labeling, MLOps & API Deployment, AI Integration with Digital Twin Systems

Services

$400,000+

Budget

Challenge

Client required a fully automated solution capable of detecting corrosion on offshore structures, identifying coating failures, weld lines, and structural edges, and assessing the severity of degradation.

See what we can do for you

Outcomes We Deliver

Solution

We developed a high-precision computer vision system that automatically classifies surfaces as corroded or non-corroded, segments coating breakdowns and evaluates rust severity in real time, and accurately detects weld lines and structural edges.

Let’s talk about what’s possible

Tech Stack

To develop computer vision system that automatically classifies surfaces as corroded or non-corroded, Blackthorn AI applied a production-grade tech stack including:

Python

PyTorch

ESPNet

Google Cloud Platform

TensorBoard

Python

PyTorch

ESPNet

Google Cloud Platform

TensorBoard

Roadmap

Project duration

01-02 Weeks

Project Kickoff and Data Assessment

Validated business objectives, established scope, and audited image datasets to assess quality and coverage for corrosion types and offshore structures.

03–04 Weeks

Annotation Strategy Design

Defined labeling framework, created detailed annotation guidelines, and set up a peer-reviewed labeling workflow to ensure consistency and accuracy.

05–07 Weeks

Data Preparation and Model Research

Preprocessed image data, applied augmentation techniques to expand training diversity, and evaluated segmentation architectures suitable for edge and rust detection.

08–10 Weeks

Model Training and Benchmarking

Trained multiple semantic segmentation models (PSPNet, ESPNet, U-Net) to classify corrosion severity and coating damage; established initial performance benchmarks.

11–14 Weeks

Optimization and Scoring Model Development

Refined model performance through error analysis and hyperparameter tuning; developed classification head to generate numerical degradation scores.

15–16 Weeks

Infrastructure and API Deployment

Packaged the ML models into RESTful APIs and built batch prediction tools; deployed the solution on Google Cloud Platform for scalable access.

17–18 Weeks

Drone and Digital Twin Integration

Connected the vision system to autonomous drone infrastructure and integrated model outputs into the client’s digital twin platform for remote inspection workflows.

19–20 Weeks

Final Validation and Delivery

Ran final system validation under real-world conditions, prepared documentation, and delivered a production-ready corrosion detection pipeline with all deployment artifacts.

01-02 Weeks

Project Kickoff and Data Assessment

Validated business objectives, established scope, and audited image datasets to assess quality and coverage for corrosion types and offshore structures.

03–04 Weeks

Annotation Strategy Design

Defined labeling framework, created detailed annotation guidelines, and set up a peer-reviewed labeling workflow to ensure consistency and accuracy.

05–07 Weeks

Data Preparation and Model Research

Preprocessed image data, applied augmentation techniques to expand training diversity, and evaluated segmentation architectures suitable for edge and rust detection.

08–10 Weeks

Model Training and Benchmarking

Trained multiple semantic segmentation models (PSPNet, ESPNet, U-Net) to classify corrosion severity and coating damage; established initial performance benchmarks.

11–14 Weeks

Optimization and Scoring Model Development

Refined model performance through error analysis and hyperparameter tuning; developed classification head to generate numerical degradation scores.

15–16 Weeks

Infrastructure and API Deployment

Packaged the ML models into RESTful APIs and built batch prediction tools; deployed the solution on Google Cloud Platform for scalable access.

17–18 Weeks

Drone and Digital Twin Integration

Connected the vision system to autonomous drone infrastructure and integrated model outputs into the client’s digital twin platform for remote inspection workflows.

19–20 Weeks

Final Validation and Delivery

Ran final system validation under real-world conditions, prepared documentation, and delivered a production-ready corrosion detection pipeline with all deployment artifacts.

Team Size

1 x AI Solutions Architect

2 x Computer Vision Engineers

1 x MLOps Engineer

1 x Project Manager

5 team members from Blackthorn.ai

1 x AI Solutions Architect

2 x Computer Vision Engineers

1 x MLOps Engineer

1 x Project Manager

Delivering Impact

Beyond the values already highlighted, there’s even more to discover. Our commitment to innovation, client success, and impactful results sets us apart

Book a Meeting

99%

Detection accuracy

For rust vs. non-rust surface classification, enabling precise corrosion identification across diverse surface types and environmental conditions.

88.9%

Linear IoU

In segmenting weld lines, coating boundaries, and structural edges—critical for assessing integrity in offshore assets.

100%

Visual coverage

Of previously inaccessible or hazardous areas (e.g., underwater joints, pipe intersections) through autonomous drone integration.

60%

Reduced manual inspection costs

By eliminating the need for on-site personnel, scaffolding, and vessel-based access.

99%

Detection accuracy

For rust vs. non-rust surface classification, enabling precise corrosion identification across diverse surface types and environmental conditions.

88.9%

Linear IoU

In segmenting weld lines, coating boundaries, and structural edges—critical for assessing integrity in offshore assets.

100%

Visual coverage

Of previously inaccessible or hazardous areas (e.g., underwater joints, pipe intersections) through autonomous drone integration.

60%

Reduced manual inspection costs

By eliminating the need for on-site personnel, scaffolding, and vessel-based access.

AI-Powered Corrosion Detection for Offshore Oil & Gas Inspections

Trained on a diverse, real-world dataset, the models were optimized for deployment on autonomous drones and seamlessly integrated into the client’s digital twin platform for real-time, remote inspections.

Energy

USA

Computer Vision, Machine Learning, Model Development, Data Labeling, MLOps & API Deployment, AI Integration with Digital Twin Systems

$400,000+

Client required a fully automated solution capable of detecting corrosion on offshore structures, identifying coating failures, weld lines, and structural edges, and assessing the severity of degradation.

Solution

To develop computer vision system that automatically classifies surfaces as corroded or non-corroded, Blackthorn AI applied a production-grade tech stack including:

Project duration

01-02 Weeks

03–04 Weeks

05–07 Weeks

08–10 Weeks

11–14 Weeks

15–16 Weeks

17–18 Weeks

19–20 Weeks

01-02 Weeks

03–04 Weeks

05–07 Weeks

08–10 Weeks

11–14 Weeks

15–16 Weeks

17–18 Weeks

19–20 Weeks

Team Size

Delivering Impact

99%

88.9%

100%

60%

99%

88.9%

100%

60%

Discover More Related Customer Stories