Book a Demo

SMTA International 2026 · Technical Paper

Solving SMT Component Attrition at Scale Using LLM-Based Agentic Analysis

How an agentic system isolates the root cause of placement attrition and recommends a corrective action, evaluated on 172 million placements.
Cameron Sobie, Ph.D. · Andrew Scheuermann, Ph.D. · Tim Burke, Ph.D.
Arch Systems · Presented at the SMTA International 2026 conference

Get the SMTAI 2026 Paper

ABOUT THE PAPER

Detecting attrition is solved. Explaining it is not.

Component attrition causes material loss, shortages and line stoppages, yet an alert rarely says why it is happening. Parts, feeders and nozzles are intertwined, so the worst-looking entity is often just a bystander to the real cause. The paper evaluates an LLM agent that automates the expert investigation, pairing deterministic analytical tools with expert-defined context to isolate the cause and recommend a corrective action.

172M

placements analyzed across five lines at two sites over six weeks

17–61%

of observed attrition identified as potentially avoidable (modeled upper bound)

~1 → 11

confirmed AOI defects per million, lowest to highest attrition quintile

INSIDE

What the paper covers.

Confounder-controlled diagnosis. Parts, feeders and nozzles are compared with matched peers, then confirmed, redirected or exonerated.

LLM plans, tools calculate. Every number in a finding comes from deterministic computation, not model arithmetic.

A worked example. An anonymized finding that traces three suspect parts to one failing nozzle.

Honest scope. A retrospective counterfactual study, with limits stated plainly.

From an attrition alert to an evidence-supported root cause.

Get the SMTAI 2026 Paper

Stay ahead of the trends

Get the latest news and content about AI in Manufacturing and ROI-driven processes.

Book a Demo icon