As SMT manufacturers contend with growing demand and a shrinking supply of skilled labor, the question of how to preserve and scale factory expertise has become increasingly important. Veteran operators and engineers who once knew how to trace a component scrap issue to a misbehaving feeder or worn nozzle are retiring. At the same time, fewer replacements have the tenure or training needed to step in right away. In many factories, troubleshooting delays and inconsistent diagnostics are beginning to threaten uptime, quality, and delivery performance.
This challenge prompted a new study that examined whether artificial intelligence, specifically AI trained to interpret factory dashboards, can begin to fill the expertise gap. Conducted by Arch Systems, the study benchmarked multiple groups of “brains”, both human and artificial, to assess their ability to identify the root causes of manufacturing scrap using real FUJI pick-and-place dashboards. The aim was to measure diagnostic quality, clarity, and speed.
Study Overview: Comparing Four Approaches
The study centered on dashboards from FUJI’s NXT PnP systems, which are commonly used in SMT production to place resistors, capacitors, and chips. It simulated a familiar scenario: an alert signals excessive component attrition on a specific line, and the task is to determine the likely root cause and recommend actionable next steps.
Participants fell into four categories:
Factory SMEs with decades of hands-on SMT experience
PhD-level engineers with SMT-specific training
A general-purpose AI (ChatGPT) using a standard prompt
An AI system built by Arch, guided by expert reasoning
Each participant reviewed the dashboards and provided a brief root cause analysis with corrective guidance. A panel of SMT experts scored each response using a structured Effectiveness Index that evaluated relevance, diagnosis, accuracy, clarity, and actionability.
What the Results Reveal
As expected, the SME group performed well. They accurately diagnosed the issues and offered actionable steps, although their responses varied in clarity and often assumed that operators would already possess some tribal knowledge. The PhDs and ChatGPT both grasped the context but lacked precision. Their answers tended to be more generalized, with suggestions like full recalibration or generic repairs that did not reference specific part numbers, feeder IDs, or nozzle locations.
The most compelling results came from the fourth group: the ArchFX AI system. Equipped with both dashboard images and structured machine data, and guided by expert-informed reasoning patterns, this AI consistently matched or outperformed SMEs in terms of accuracy and clarity. It also delivered responses in a matter of seconds, which was more than ten times faster than the human experts, and at a significantly lower cost.
One of the most important findings was that high-quality AI guidance is possible even in environments without modern API integrations. By analyzing screenshots of dashboards alone, the AI system still delivered expert-level diagnostics. This is especially relevant for manufacturers working with legacy equipment or highly customized lines.
Why It Matters for SMT Professionals
SMT operations rely on tight process control, fast issue resolution, and minimal unplanned downtime. When experienced troubleshooters leave, it can slow down maintenance, reduce throughput, and increase scrap. These issues often don’t show up clearly in the KPIs. As this study shows, AI systems tailored for SMT applications can provide consistent, rapid, and usable diagnostics that support frontline teams in real time.
These findings are not about replacing human workers. Instead, they offer a path to scaling expert-level support across every shift, even on lines that are still transitioning to modern digital systems. With the right guidance models and access to the right data, AI can serve as a reliable diagnostic assistant that never tires, never makes assumptions, and always delivers results that are clear and actionable.
To explore the full methodology, comparisons, and results from the study, download the complete white paper.