Energy, Ultrasonic Testing
Perspectives: AI Doesn’t Replace NDT Expertise—But It Can Multiply It
Title image: 2025 Battle of the AI Agents winners Mehdi Matinfar (left) and Gary Orlove (center) impressed judges with an AI tool that streamlines the creation of calibration blocks for phased array ultrasonic testing (PAUT). The two-day challenge was led by Dr. Gleb Tsipursky (right).
Phased array ultrasonic testing (PAUT) relies on calibration blocks that match inspection goals, geometry, and applicable project requirements—yet developing those blocks can be slow, iterative, and inconsistent across teams. At ASNT 2025, I built an AI agent that helps structure requirements, propose calibration block layouts, and generate machinist-ready drawing packages more quickly. The key lesson is practical: AI does not solve nondestructive testing (NDT) problems on its own. When guided by human intent, constraints, and mandatory review gates, AI agents can reduce friction and help professionals focus on verification, performance, and safety.
From Competition to Practical Tool
At ASNT’s annual conference last year in Orlando, Florida, I led a team that competed in (and ultimately won) the new ASNT Battle of the AI Agents competition by demonstrating an AI agent built to support a practical need in our industry: designing and documenting PAUT calibration blocks within minutes.
Many organizations are curious about AI in NDT, but they hesitate for good reasons—wrong assumptions, missing traceability, and outputs that look confident but are not technically validated. My takeaway from building and demonstrating this tool is simple: AI alone cannot solve NDT problems. People solve NDT problems. But when we pair human creativity with disciplined constraints, an AI agent can become a powerful assistant that speeds up repeatable work and strengthens consistency.
The Hidden Bottleneck
PAUT technology continues to advance, but one bottleneck remains surprisingly manual: moving from an inspection objective to a calibration block that is fit for purpose and fabricable. The work is essential—but much of the cycle time is consumed by repetitive translation and documentation.
Common friction points include:
Slow iteration cycles and back-and-forth revisions
Inconsistent outcomes between teams for similar inspection needs
Incomplete machinist packages (dimensions, tolerances, feature callouts, notes)
Requirements fragmented across codes, customer specifications, internal practices, and lessons learned
The AI Agent Approach
The agent I built is not intended to “decide” inspection engineering. It supports the workflow so experts spend more time on verification and performance confirmation, and less time on repetitive drafting and packaging.
It focuses on four functions:
Requirements capture: Prompt for geometry, material, thickness and outside/inside diameter ranges, objectives, scan constraints, and fabrication limits.
First-pass concepts: Propose one or more block layouts and reflector arrangements to accelerate the first iteration.
Manufacturable deliverables: Generate clear drawings and a feature/reflector table with assumptions and notes.
Review gates: Force confirmation of reflector purpose, beam-path/coverage intent, and qualified reviewer approval before release.
Why This Worked
AI becomes risky when it silently fills in unknowns. It becomes useful when it is required to operate inside constraints.
NDT is a constraint-driven discipline: acceptance criteria, equipment limitations, geometry/access limitations, and traceability expectations. This project worked because the agent was designed with explicit inputs, controlled assumptions, standardized outputs, and mandatory human review. The value was not a model making decisions—it was a workflow that captures intent and makes expert review faster and more consistent.
How AI Changed My Perspective—and What Comes Next
Building this agent changed how I think about “AI in NDT.” Instead of treating AI as a general tool, I now treat it as an engineered assistant with guardrails. The project clarified that high-value technical work must remain human-led (inspection intent, validation, and field practicality), while repeatable work (structured inputs, documentation, packaging, and revision hygiene) is where agents can reliably help.
Looking forward, I believe agents can help the industry in three near-term ways:
Standardized technical packages that reduce miscommunication between engineering, inspection, and fabrication.
Faster iteration with better traceability by capturing inputs, assumptions, revisions, and reviewer decisions in a structured format.
Better training support when combined with hands-on learning—helping inspectors understand the “why” behind setups, not just the output.
For my own work, the next step is to keep the approach conservative and audit-friendly: improve assumption documentation, expand proven templates, and ensure every output is designed to be reviewed—not blindly accepted.
Practical Impact
Even when used conservatively, this approach can help teams:
Move from requirements to a reviewable first draft faster.
Improve consistency of drawing packages and reflector documentation.
Reduce rework caused by unclear notes or late-discovered fabrication constraints.
Free Level III and senior staff time for verification and optimization.
The goal is not automation. The goal is removing friction while keeping responsibility where it belongs.
Limitations and Responsibility
An AI agent should not be treated as a final authority on code compliance, a substitute for qualified technical review, or a replacement for validating beam paths, coverage, and calibration response.
If AI makes a workflow faster, it must also make the workflow more controlled—not less. Versioning, traceability of assumptions, and documented reviewer sign-off are non-negotiable.
Conclusion
Winning the 2025 ASNT Battle of the AI Agents was exciting, but the more important win is what it signals: our industry is ready for practical AI—used responsibly.
AI does not replace NDT professionals. But professionals who apply AI agents thoughtfully—inside constraints, with verification—can improve speed, consistency, and traceability while keeping safety and technical excellence at the center.
About the Author
Mehdi Matinfar, BS, is an engineer, project manager, Principal Surveyor, and an ASNT NDT Level III with over 23 years of experience in inspection and NDT. For the past 13 years, he has been with DNV, where he certifies oil and gas equipment to DNV rules and industry standards, including API, ASME, ASTM, and ISO. As DNV’s corporate Level III, he has helped develop a hands-on NDT training program that improves inspection quality and consistency among US regional surveyors.
Disclaimer: The views and opinions expressed in this article are those of the author and do not necessarily reflect the official position of ASNT. Articles in this series have not been peer-reviewed and are published for informational and professional development purposes only. Accuracy of technical content is the responsibility of the author.
About Perspectives
Perspectives is a digital series from Materials Evaluation featuring first-person insights from NDT practitioners, engineers, and specialists across the field. Articles in this series reflect the views of the author and have not been peer-reviewed. To submit a piece for consideration, contact the editorial team.
