machine vision inspection
The problem
Lines making steel coils, painted and coated parts, textiles, packaging, pharmaceuticals and food, where inspectors should spend their time on decisions instead of watching the line.
A typical situation
Our inspectors catch most defects, but not at full line speed and not on every shift. Customers return coils with surface marks, fabric rolls get downgraded, and a missing or unreadable code on a pack can hold up a whole shipment. When a complaint arrives, we struggle to trace it back to the line, the shift and the raw-material batch.
Outcomes
Indicative ranges, validated in your pilot. These figures come from comparable industry programmes, not from TashIT projects. Your pilot sets a baseline with you and measures the actual change.
Typically 20-30% fewer
Basis: Inline camera inspection replacing or supporting manual inspection for surface and packaging defects. Depends on defect types, current inspection coverage and whether root causes are fixed.
Typically 30-60% fewer
Basis: Full inspection at the line rather than sampling, with rejects or markers before dispatch. Depends on which defects drive complaints today.
Typically from sampled checks to every unit at the inspected station
Basis: Area-scan or line-scan cameras sized to line speed. Depends on line speed, required resolution and the smallest defect that matters.
Typically hours instead of days
Basis: Every defect stored with its image, position, batch, line and shift, alongside process data. Depends on the quality of batch records in MES or ERP.
From data to decision
With your quality team we build a defect catalogue: types such as scratches, dents, paint runs, pinholes, uneven coating, weave faults and print errors, with their sizes, severities and what counts as reject, downgrade or accept. Sample images are collected and labelled, including rare defects.
Cameras, lenses and lighting are selected for the material, line speed and smallest defect. Mounting, enclosures and triggering are designed with the line's PLC and maintenance access in mind.
Detection and classification models are trained on your images and validated against experienced inspectors on held-out samples. For codes and labels, OCR and code grading are combined with model checks.
Models run on an industrial edge computer beside the line, so inspection keeps pace with production and does not depend on the network. Results trigger alarms, markers or reject gates through the PLC.
Each defect is stored with its image and linked to batch records in MES or ERP. Dashboards show defect trends against process conditions, so engineers can remove causes rather than just sort output.
Data and integration
How the pilot works
Duration
Scope
Agreed in writing before work starts, with targets set against your baseline.
The pilot follows a short diagnostic and ends with a written results review. Scaling up is a separate decision, made on the measured results. How our engagements work
Solution blueprint
Line-scan cameras and a trained defect model that inspect every metre of coated or galvanised strip, grade each coil and link defects back to line conditions.
Status: Reference architecture and pilot plan, available for a paid proof of concept. Reference material, not a client project.
Where it applies
Keep exploring
Pilot 8–12 weeks
Sub-metering and forecast-based dispatch of grid, captive generation, solar and battery storage to cut energy cost per unit.
Pilot 8–12 weeks
Camera-based vehicle, load and label verification at gates and weighbridges, reconciled automatically with ERP and regulatory records.
Pilot 10–16 weeks
SKU-level forecasts, replenishment and allocation that reduce stock-outs, waste and working capital.
Tax stamps and unique identification marks on excisable and regulated goods create records that must agree with weighbridge, gate and ERP data. How gate vision and daily reconciliation turn a compliance obligation into a working control.
5 min read
Every topic
Practical articles on AI, automation and operational data: what works, what it needs and how to test it before you scale.
FAQs
Something else to ask?
Email your question with a little about your site and systems, and we will reply in plain language.
info@tashit.comIt depends on the defect. Common defects may need a few hundred labelled examples. Rare ones can be handled with anomaly detection trained mostly on good product, plus whatever examples you have. The diagnostic includes an image survey that tells you what is realistic.
No. The station is sized for your maximum line speed and processing runs on edge hardware at the line. If the system cannot keep up, the shadow-mode phase shows it before anything is connected to reject mechanisms.
Reading and verifying 2D codes, serial numbers and label content is part of the solution, and results can feed your serialisation system. Validation for regulated lines, for example under GMP, is planned with your quality team as part of the project scope.
Tell us about your site or process and the data you already have. We will reply with a diagnostic plan, the measures we would propose for a pilot and an honest view of feasibility.