From data to action at scale:

How a Global Electronics Manufacturer Achieved Near-100% Automated Downtime Labeling and Instantly Scaled Across SMT Lines

A global electronics manufacturer partnered with Arch and Fuji to improve how downtime is identified and addressed across SMT operations. By leveraging high-frequency data from Fuji pick and-place machines, Arch implemented Automated Downtime Labeling (ADL) to structure machine signals, automate root cause identification, and enable faster operational response.

At the lighthouse site, automated downtime labeling increased from ~30% to near 100% in just a few weeks. This significantly reduced manual effort while giving teams immediate clarity into the causes of downtime, enabling faster, more consistent action on the shop floor.

Once optimized, the model was deployed to additional Fuji sites, delivering immediate performance gains. With no retraining required, sites quickly reached high levels of automation and insight, allowing the organization to standardize downtime analysis and accelerate time to value across its global operations.

Automated Downtime Labeling on SMT Lines _ Arch + Fuji Case Study

HIGHLIGHTS:

Increased automated downtime labeling to near 100% (from ~30%) within weeks, eliminating manual classification gaps

Achieved 95%+ automated root cause identification, significantly reducing reliance on manual input and accelerating issue resolution

Eliminated manual labeling for the majority of downtime events, improving operator efficiency
Scaled instantly across Fuji sites with no retraining, accelerating deployment timelines
Enabled faster identification and resolution of production issues

The Challenge

The manufacturer lacked a consistent, scalable way to interpret and act on downtime across Fuji SMT lines.

Manual, inconsistent labeling: Operators classified most downtime events, limiting reliability
Delayed insight: Issues were often identified downstream rather than in real time
Untapped machine data: Large volumes of Fuji machine data were available but not actionable
Operational inefficiency: Time was spent interpreting data instead of resolving issues
Limited scalability: of Improvements could not be easily replicated across sites

As a result, teams were slow to identify root causes and lacked a consistent basis for action across operations.

The Arch Approach

Arch partnered with Fuji and the customer to transform raw machine data into structured, real-time operational insight through Automated Downtime Labeling (ADL).

Structuring Fuji Machine Data

Processed high-frequency data streams from Fuji pick-and-place machines
Interpreted feeder signals, error codes, and machine states
Standardized downtime classification across lines and sites
Created a reusable model to support consistent deployment

Learning from Production

Rapidly improved as the system learned from live Fuji data and incorporated expert defined logic
Expanded coverage to capture nearly all downtime events across machine signals and operational knowledge
Increased both the accuracy and consistency of root cause identification by odifying expert decision-making into ADL

Enabling Faster Action

Converted machine signals into clear, actionable insights
Enabled teams to respond in real time rather than after inspection
Reduced reliance on dashboards by surfacing relevant actions directly

Key Outcomes

Near-100% Automated Labeling in Weeks

Increased automated labeling to near 100% (from ~30%) within weeks, closing visibility gaps across downtime
Achieved 95%+ automated root cause identification, accelerating issue resolution, and reducing manual effort
Expanded to 10–20+ meaningful downtime categories, enabling more precise analysis and targeted improvement actions
Gained near-complete visibility into downtime drivers, enabling more confident and timely decision-making

Rapid Learning Curve to Near-100% Automated Downtime Labeling

Automated downtime labeling increased from ~30% to near 100% within weeks as the system learned from Fuji machine data, rapidly improving visibility into true downtime causes

Refining Downtime Classification as Accuracy Improves

The number of unique downtime reasons initially expands as the system learns from Fuji machine data, then stabilizes and reduces as labeling accuracy approaches near 100%, indicating convergence on the most meaningful root causes.

Faster Time to Action

Shifted from delayed detection to real-time identification
Reduced time spent interpreting machine data
Enabled clear, immediate next steps for operators and engineers
Faster response to issues reduced production disruption and improved overall operational efficiency

Operator Efficiency Gains

Eliminated manual labeling for 95%+ of downtime events
Reduced repetitive data entry and labeling errors
Allowed operators to focus on production and problem-solving
Freed up operator time while improving the consistency and reliability of downtime data

Instant Scale Across Fuji SMT Lines and Sites

A second Fuji site started at ~20% automated coverage
Coverage increased rapidly after deployment with no retraining
A second site achieved strong performance quickly using the same model
The lighthouse model was reused without redefining downtime logic
Performance improvements were immediate across sites
Enabled rapid rollout of proven downtime intelligence, reducing deployment time and ensuring consistent performance across the network

Instant Performance Gains Across Additional Sites Without Retraining

Site 2 rapidly improved from low to high automated labeling coverage immediately after deployment without retraining, demonstrating fast, scalable impact across sites.

Strategic Lessons

Learn once, scale efficiently: Structured downtime intelligence enables rapid cross-site deployment
Visibility drives performance: Better insight into root causes leads to faster, more effective action
Machine data needs structure: Value comes from interpretation, not volume alone
AI supports operational decisions: Turning signals into actions is more impactful than dashboards
Consistency enables scale: Standardized logic creates alignment across sites