SCC Susceptibility Modeling to Demonstrate New Guidelines for Machine Learning in Pipeline Integrity
L McAllister, M Chima,A Fadic, M Murray, S Middleton, M Al-Amin
International Pipeline Conference (2026)
Anton Fadic Data Scientist
I have a passion for bridging theory and first principles into real physical applications
About Me
Background & Expertise
I have a passion for computational physics and statistics. My main interest is to bridge these more theoretical fields into applications that become actionable insights by decision makers. Currently interested in Fracture Mechanics, Fitness-for-Service, Risk, Probabilistic Assessments, and Machine Learning applications with a focus on pipeline integrity. Also interested in Computational Fluid Mechanics, Reacting Flow, Statistical Mechanics and various mathematical topics.
API 579
CSA Z662
CFD
FEA
XGBoost
Neural Networks
Linux
Azure
Reaction Kinetics
Thermodynamics
Skills
Technical Expertise
Pipeline Integrity
- API 579
- CSA Z662
- DNV-RP-F101
- API 1176
- API 1163
- API 1160
- BS 7910
- B31.4
- B31.8 and B31.8S
- B31G
- PRCI MAT-8
- 49 CFR 192 and 195
Computational Physics
- CFD
- FEA
- Linear Elasticity
- Stress Intensity Factor
- Turbulent Flow Simulation
- Boundary Layer Analysis
- Numerical Methods
Machine Learning
- XGBoost
- Neural Networks
- Overfitting
- SHAP
- Data Leakeage
Software and Deployment
- Linux
- Azure
- git
- Kubernetes
- Docker
- C
- Python
- Ansys
- OpenFOAM
- Comsol Multiphysics
Process Engineering
- Reaction Kinetics
- Thermodynamics
- Computational Transport Phenomena
- Fluid Mechanics
Experience
Career Journey
Data Scientist
2019 – presentIrth Solutions / OneBridge Solutions
Implemented and automated Fitness-for-Service methodologies for piggable and hydrotested lines. Developed methodology towards best practices usage of ML in pipeline integrity. Investigated and developed probabilistic methodologies for quantitative risk assessments for cracks, including fatigue effects. Investigated and developed ILI-to-ILI alignment algorithms.
- Developed and deployed production quality Fitness-for-Service software used daily by 17 large pipeline operators
- Found and published the effects of autocorrelation in Machine Learning pointing towards the issue of data leakage under the random splitting assumption
- Speed up probabilistic assessments by 75-94%
- Studied the effects of corrosion clustering around elevation valleys, quantifying the effect with statistically rigorous methods
Visiting Researcher
2013 – 2014TU Darmstadt
Research focused on first principles approach modeling of catalytic reactors for industrial applications. Developed surrogate models for computationally expensive CFD simulations.
Publications
Selected Works
False Confidence in Machine Learning: Autocorrelation and its Consequences in Pipeline Integrity
A Fadic, M Murray
International Pipeline Conference (2026)
Probabilistic Crack Fatigue Analysis: Accelerating Rare Event Risk Evaluation With Sobol sampling
A Fadic, M Murray
International Pipeline Conference (2026)
Quantifying Corrosion Patterns: A Study of Corrosion Clustering Around Elevation Valleys With a Large Inline Inspection Dataset
A Fadic, M Murray
International Pipeline Conference, 88544, V02AT03A004 (2024)
Blog
Latest Writing
2026-10-03
Your ML Validation May Be Misleading You: Random Splits of Pipeline Data
Pipeline data is spatially autocorrelated. Random train-test splits can make machine-learning models look more reliable than they are.
draft
pipeline-integrity
machine-learning
spatial-autocorrelation
IPC
2026-09-20
The Cost of a Half Percent
When is a 0.5% discrepancy good enough?
draft
pipeline-integrity
uncertainty
radon
science
2026-09-20
Why grad school mattered: Blasius, boundary layers, and the value of understanding
Draft
draft
fluid-mechanics
boundary-layer
blasius
grad-school
engineering
fitness-for-service
Let's Work Together
Interested in collaborating on data science projects, process optimization, or research? I'd love to hear from you.