.webp)
A returned batch costs more than the product inside it. Rework eats margin, recall investigations eat time, and both point back to the same root problem: the inspection strategy missed something. Most manufacturers have invested in quality systems. Many are still shipping defects. The reason is almost always structural, not technological.
This guide explains the two-layer model that separates product inspection from process monitoring, and shows exactly where machine vision for quality control fits inside each.
Machine vision for quality control operates at two levels: product inspection, which catches defects in finished or in-process items, and process monitoring, which verifies that operators and assembly steps follow the correct sequence. Both are necessary. Confusing them is the most expensive mistake in AI vision deployment.
What is Machine Vision for Quality Control?
Machine vision for quality control is the use of cameras, edge AI, and image processing to automatically inspect products and monitor processes in manufacturing environments.
That definition covers two operationally separate problems. The first is product integrity: the finished item is wrong. The second is process integrity: the method used to make the item was wrong. Both produce defects. They require different instrumentation, different data streams, and different responses.
The performance gap between machine and human inspection is significant. AI vision quality control systems achieve 99.5% or higher defect detection accuracy, compared to 70 to 85% for trained human inspectors performing repetitive checks at line speed. The gap widens as throughput increases. At several thousand parts per minute, human inspection is not operationally viable.
The financial stakes match that performance gap. Manufacturing facilities lose an estimated $3.4 billion annually to defective products that escape human inspection. (iFactoryApp, 2026) The machine vision systems market reflects this pressure: valued at USD 20.4 billion in 2024, projected to reach USD 41.7 billion by 2030 at a 13% compound annual growth rate.
Quality managers working with AI vision systems use a specific vocabulary: defect escape rate, inline inspection, poka-yoke, first-time-right, rework cost. These terms appear throughout this article because they reflect how the problem is actually experienced on the floor.
The Two Failure Modes Machine Vision Must Solve
Most computer vision quality inspection vendors treat their product as a single answer to a single question: does this item have a defect? That framing misses half the problem.
Failure Mode 1: Defects escape the line undetected.
The product is wrong. A surface scratch, a dimensional deviation, a missing label, a compromised seal. Traditional inspection catches this after the fact, at sampling rates that understate the real escape rate. A batch with a 2% defect rate passes a 10-sample AQL inspection 82% of the time. (iFactoryApp, 2026) Most of those defects reach the customer.
Failure Mode 2: The process runs incorrectly.
The operator skips a torque step. The wrong component goes into the assembly. The kitting sequence is reversed. The product may pass visual inspection quality control and still fail in the field. This is process adherence failure, and it is invisible to product inspection systems.
Why solving only one is not enough: a line with complete product inspection and no process monitoring will still produce recalls driven by sequence errors. A line with complete process monitoring and no product inspection will still ship cosmetic or dimensional defects. The two failure modes operate in parallel. They require parallel solutions.
“The Jidoka Two-Layer Quality Model: Product Integrity (what comes off the line) and Process Integrity (how the line runs). Defect escape rates fall only when both layers are instrumented.” - Jidoka Technologies
This two-layer model is the framework the rest of this article builds from: the Product Integrity Layer and the Process Integrity Layer.
Comparison Table: Product Inspection vs Process Monitoring
How KOMPASS Handles Product Inspection
KOMPASS is Jidoka's Product Integrity Layer solution. It runs inline vision QC at up to 12,000 parts per minute with 99.9% defect detection accuracy, inspecting products from 360 degrees as they move through the production line.
The architecture runs a five-stage closed loop: SEE, THINK, ACT, LEARN, INTEGRATE.
- SEE: Multi-angle cameras capture the product in motion.
- THINK: Edge AI analyzes each frame in under 10ms, classifying surface condition, dimensional accuracy, label presence, and component completeness.
- ACT: The system routes or rejects defective products before they progress downstream.
- LEARN: The model refines continuously from production data, improving accuracy over time without manual retraining.
- INTEGRATE: Data feeds into quality dashboards and connects with existing manufacturing execution systems.
Traditional AI vision quality systems face a cold-start problem: they require large libraries of defect images before deployment, which early-stage lines and new product introductions do not have. KOMPASS addresses this through the DOJO self-training engine. DOJO trains on fewer than 10 good-product images. No defect samples are required, and no ML expertise is needed from the quality team.
The false-rejection rate matters as much as the detection rate. Falsely rejecting a good product creates its own rework cost and line disruption. KOMPASS reduces false rejections by 40% compared to traditional AI vision systems.
Industries where KOMPASS operates: automotive OEM component inspection at conveyor speed, FMCG packaging and label verification, pharmaceutical packaging integrity, and electronics soldering and assembly checks.
How Nagare Monitors Assembly Processes
Nagare is Jidoka's Process Integrity Layer solution. Where KOMPASS inspects the product, Nagare watches how the work is done. It tracks operator actions, component presence, and assembly sequence in real time using existing cameras. In most deployments, no new camera hardware is required.
The same SEE / THINK / ACT / LEARN / INTEGRATE architecture applies, with different inputs. Nagare's cameras capture process signals: what the operator is doing, whether the correct component is in hand, whether the station status matches the workflow step. The THINK stage validates those signals against SOPs and digital work instructions. If a deviation occurs, ACT triggers a poka-yoke alert or a station halt before the unit progresses to the next cell.
This distinction is critical. KOMPASS catches a defect after it exists. Nagare stops the process error before a defect is created.
The performance outcomes from Nagare deployments: 30% increase in process adherence, 35% reduction in rework, 25% productivity increase, and 30% reduction in process deviations.
“Nagare monitors every assembly step using existing cameras, ensuring 100% SOP adherence and delivering a 35% reduction in rework without requiring new hardware investment.” - Jidoka Technologies
Two additional capabilities distinguish Nagare on the plant floor.
AI Buddy Mode guides operators in real time and tracks individual skill levels. As operators progress, supervision requirements fall. New operators are brought up to proficiency faster. The result is a lower supervisor-to-operator ratio without a corresponding increase in error rates.
Edge AI deployment processes all data on-premises. No cloud dependency. This directly addresses the data-privacy and latency objections that block AI adoption in automotive and pharmaceutical manufacturing, where process data is proprietary and latency cannot be tolerated.
ROI timeline: Nagare deploys on existing camera infrastructure and delivers return on investment within 12 months.
When Do You Need KOMPASS, Nagare, or Both?
This is the decision that matters. The technology is clear. The harder question is which problem you are actually facing.
The Quality Problem Diagnostics Framework by Jidoka Technologies is a four-question self-assessment that routes manufacturers to the right solution. Work through these in order.
Q1: Where do your defects originate?
If defects are product defects, including dimensional deviations, surface damage, missing components, or label errors, the Product Integrity Layer is uncontrolled. KOMPASS is the solution. If defects trace back to process errors, including wrong assembly sequence, wrong component used, or skipped procedural steps, the Process Integrity Layer is uncontrolled. Nagare is the solution.
Q2: What is your defect escape rate?
If your sampling-based AQL process is passing batches that contain defects, the escape rate problem is a product inspection gap. KOMPASS closes it with 100% inline vision based quality system coverage. If your outgoing defect rate is low but rework and scrap costs remain high, the cost is being absorbed upstream. Nagare is the primary lever.
Q3: What does your current inspection infrastructure look like?
KOMPASS requires hardware integration: modular vision infrastructure connected with conveyors and rejection mechanisms. Nagare deploys existing CCTV cameras in most cases. If capital expenditure is constrained, Nagare is the faster first deployment. KOMPASS deployment runs 6 to 8 weeks from installation to production-ready.
Q4: What failure mode reaches the customer?
Field returns driven by product defects, scratches, dimensional failures, missing parts, point to KOMPASS. Field returns or compliance failures driven by assembly errors, missing process steps, or wrong-component issues point to Nagare. Both problems together require both systems.
Machine Vision for Quality Control Across Industries
The KOMPASS and Nagare split plays out differently by sector. Each industry below combines both layers, but the specific failure modes differ.
1. Automotive
KOMPASS handles OEM component inspection at 12,000 PPM on fast conveyors, catching surface defects, dimensional deviations, and assembly completeness before components leave the cell. Nagare enforces torque sequences and fastener installation steps at assembly stations, with poka-yoke alerts that prevent incorrect sub-assemblies from progressing to the next cell. An overtorqued fastener and a missing fastener are two different problems. One is KOMPASS territory. The other is Nagare's.
2. FMCG and Packaging
KOMPASS runs label verification, fill-level checks, and seal integrity at production line speed. A mislabelled SKU at 800 units per minute is caught before it ships. Nagare monitors kitting sequences in packing operations and verifies SKU accuracy in outbound warehouse flows. The product can look correct and still be the wrong item for the wrong order.
3. Pharmaceuticals
KOMPASS handles label and text verification for regulatory compliance: batch numbers, expiry dates, product codes. The margin for error in pharma label accuracy is effectively zero. Nagare monitors kit assembly steps to prevent wrong-drug-into-pack errors, supporting GMP documentation requirements. A label that passes visual inspection on the wrong product is the failure mode that ends careers.
4. Electronics
KOMPASS inspects soldering joints, PCB component placement, and connector alignment using multi-angle high-resolution imaging. Nagare monitors SMT assembly steps and operator handling sequences, catching sequence violations before a populated board moves to reflow.
According to the U.S. National Institute of Standards and Technology (NIST), defects account for approximately 10% of total manufacturing costs. That 10% splits between product integrity failures (KOMPASS scope) and process adherence failures (Nagare scope). Most manufacturers use instrument one. The cost of the other stays hidden in rework, scrap, and field returns.
Conclusion
Every quality manager has a version of this story: investments in inspection technology that still produce recalls, or rework costs that refuse to fall despite upstream automation. The reason is almost always the same. Product quality and process quality are two separate problems, and they require two separate answers.
KOMPASS monitors what leaves your line. Nagare monitors how it gets made. Deploying both is what Jidoka means by First-Time-Right production. If you know which problem is yours, the next step is a conversation. If you are not sure, the Quality Problem Diagnostics Framework in this article is the place to start.
Book a Jidoka demo at jidoka-tech.
Frequently Asked Questions
1. What is the difference between machine vision and computer vision in manufacturing?
Machine vision refers to industrial imaging systems designed for specific, controlled inspection tasks at production speed, typically using dedicated hardware and algorithms. Computer vision is the broader AI field covering image understanding across environments. In manufacturing quality control, machine vision is the applied subset: purpose-built, inline, and decision-immediate. The two terms are used interchangeably in vendor literature but describe different scopes.
2. How accurate is AI machine vision for quality inspection?
Current AI vision systems achieve 99.5% to 99.9% defect detection accuracy on trained production lines. Human visual inspection achieves 70 to 85% accuracy under consistent conditions, degrading further under fatigue, shift changes, and high throughput. The accuracy gap widens as production speed increases: at 12,000 parts per minute, human inspection is operationally impossible. AI vision accuracy is also consistent across all shifts, which human inspection is not.
3. Can machine vision systems be deployed without changing the production line?
Process monitoring systems such as Nagare are specifically designed to run on existing CCTV infrastructure, requiring no new camera hardware in most deployments. Product inspection systems such as KOMPASS require modular vision hardware integrated with conveyors and rejection mechanisms, but Jidoka's modular infrastructure connects with existing automation. Deployment timelines for KOMPASS run 6 to 8 weeks from installation to production-ready; Nagare deployments can be operational faster using existing cameras.
4. What is the ROI timeline for machine vision quality control?
ROI depends on current defect escape rates, rework costs, and inspection labor. Manufacturers implementing AI quality control report defect reductions of up to 50% and inspection cycle improvements of 30 to 50% (industry benchmarks, 2025 to 2026). Nagare is cited at a 12-month ROI timeline using existing camera infrastructure. KOMPASS ROI timelines vary by industry; pharmaceutical and automotive deployments with high recall-cost exposure typically recover investment within 6 to 18 months.
5. Does machine vision replace human quality inspectors?
Machine vision for quality control augments inspection by handling high-speed, repetitive detection tasks that exceed human capability. The practical result in most deployments is redeployment of inspection labor to higher-value quality activities, root-cause analysis, and exception handling. Nagare specifically includes an AI Buddy Mode that guides operators and tracks skill development, reinforcing human capability rather than eliminating it.
