Frame Assembly Check: 110 to 38 Minutes With AR and AI
Large frame assemblies, hydraulic support assemblies and rear assembly frames are built from many small welded parts and brackets. They carry many inspection items and complex assembly relationships, and they are poorly suited to being moved into a metrology room. Traditionally an inspector works through drawings item by item, spending 110 minutes per unit on drawing-based inspection, with incomplete records and high structural rework cost. Deviations are often found only at a later process.
Two task families share one inspector
Frame inspection covers structural questions and content questions at once. On one side are bracket offset, hole mismatch and missing accessories; on the other are labels, model, quantity, OCR characters, surface flaws and SOP steps. In manual inspection the two families interleave, attention switches repeatedly, and the result is slow and easy to miss. Any single tool covers only half.
AR owns space: a bracket offset shows as displacement
The tablet overlays 3D CAD on the workpiece in real time. A bracket offset appears as the CAD contour displaced from the physical one, a hole mismatch as a position deviation, a missing accessory as an empty slot in the overlay. Frame structures have rich external features, so markerless registration holds reliably in a shop environment.
AI owns content: OCR, quantity, SOP and surface flaws
AI handles nameplate and label characters, material model and quantity, SOP step conformity and surface appearance, judging each independently and then consolidating. The design point is confidence plus human review: low-confidence results are flagged for confirmation and are not released directly. That choice decides whether the AI stage is an aid or a new risk.
The glasses are for guidance and first-person evidence
AR smart glasses carry the workflow and the first-person record. Steps are prompted so items are not skipped, and first-person images are stored automatically in place of paper notes. Tablet and glasses share data, so structural comparison and process recording sit in one system instead of two disconnected tools. The point is not another device; it is fixing a record that people used to copy by hand.
Where the four improvements come from
After implementation, drawing-based inspection time per unit fell from 110 minutes to 38 minutes, a 65 percent reduction. Metrology prescreening dropped from about 42 hours a week to about 19 hours, a 55 percent reduction. Inspection record completeness rose to 96 percent, and monthly structural rework cost fell by about 63 percent, with deviations found 1 to 2 processes earlier. These figures are subject to site acceptance results.
What to confirm before rolling out
First, integration: confirm the data interfaces and field mappings to the existing MES, QMS and ERP, so results enter the existing management flow. Second, acceptance: the tablet and glasses are direct to operate and can be learned quickly, but the guided steps must match the shop process, or guidance itself becomes a burden. Where text confirmation dominates, AI value is highest.
Questions from the floor
Where does the 110 to 38 minute saving come from?
Real-time CAD overlay replaces drawing-by-drawing checking, so structural problems are visible at a glance, while AI confirms labels, model, quantity and SOP in parallel.
What happens when AI is wrong?
Low-confidence results are not released and go to human confirmation; high-confidence results are recorded, and anomalies are marked for repair and rechecked.
How do results enter the quality system?
Results sync to MES, QMS and ERP, and records, anomaly marks and repair status travel with the workpiece, forming a path from discovery to closure.
Can it run without the glasses?
Structural comparison works on the tablet alone. The glasses provide workflow guidance and first-person records; without them, record completeness and skip prevention are weaker.