Scratch detection on aluminium extrusions
On aluminium extrusions, normal brushed texture can resemble a scratch and a real scratch can disappear when the light changes direction. The defect can blend into the normal pattern.
// Short answer
Aluminium extrusion surface defect detection is a camera check of each profile before pretreatment and coating. addaScan builds it on HIKROBOT line-scan cameras timed by an encoder, with deep-learning defect tools in VisionMaster that find scratches, dents, die lines and pick-up on the faces the cameras can see. One camera covers the main visible face; three cover the top and both sides. A flagged profile is marked or set aside, so it is not coated. Which marks are rejectable comes from your acceptance specification, and feasibility is proven on your own profiles first.
// At a glance
The station in three lines
- Faults caught
- Scratches, dents, die lines, pick-up and other visible surface marks on the faces in view
- Configuration
- One 4k line-scan camera on the main face (low) or three around the profile (high), encoder-timed, VisionMaster deep learning
- Line link
- Alarm, marker output and defect log to the line PLC; profile ID or rack position recorded with each fault
// Fit & limits
What this station will not do, and what it needs
What it will not do
- Hidden faces: inside hollow sections and faces turned away from the cameras are not inspected.
- Coating quality: film thickness, adhesion, colour and gloss after coating are not checked by this station.
- Alloy and mechanical properties: hardness, temper and internal defects are outside a camera check.
- Marks made after the camera: damage during racking, pretreatment or curing happens after the station.
What makes it work
- Samples: good profiles and profiles with each defect type, in each surface finish you run.
- An acceptance specification: which marks are rejectable on which face, agreed with your customers.
- Single, steady profiles: one profile at a time, not swinging, past the camera position.
- An encoder: on the conveyor, for image scale and fault position.
// On South African lines
Aluminium extrusions in South Africa, and why the surface is checked before coating
South Africa’s aluminium industry is represented by the Aluminium Federation of South Africa, whose members include semi-fabricators of extruded and rolled products, architectural and construction businesses, and surface finishers (AFSA). Extruded profiles for windows, doors and shopfronts are commonly finished by powder coating or anodising.
Surface marks made at the press or in handling travel with the profile into the coating line. Die lines are continuous longitudinal lines caused by irregularities or build-up on the bearing surfaces of the die; pick-ups are torn, comma-like spots caused by local material deposited on the die surface (aluminium-guide.com). Pick-up can occur together with die lines (aluminium-guide.com). Scratches and dents can also come from handling, stretching, sawing and racking.
Once a profile is pretreated and coated, removing a defect means stripping or scrapping it, and the coating, pretreatment and handling are spent. Catching the mark before coating is the cheaper point. Which marks are acceptable after coating depends on the finish and the customer, so the reject limits come from your own acceptance specification.
What the coating line reports, and what extrusion surface inspection checks
| What the line reports | What a station checks | Where it stops |
|---|---|---|
| “Scratches only show after we have powder coated the profile” | Every profile is imaged on its visible faces before pretreatment; scratches above the agreed limit are flagged before coating. | A scratch made after the camera, during racking or in the oven, is not seen. |
| “Die lines come back on every profile from one die” | Die lines are found and logged per profile, so a run of them from one die shows up as a trend, not a surprise at packing. | The station reports the symptom; the die still has to be corrected at the press. |
| “Pick-up spots and rough surface on the visible face” | Pick-up and rough patches are detected by deep-learning tools trained on your good and defective profiles. | Marks smaller than the agreed limit pass; the limit is set on samples, not promised in advance. |
| “Dents from handling at the saw and the stretcher” | Dents are found where the light shows their edges; low-angle line light makes shallow dents stand out. | Very shallow dents on curved faces may not show; that is established at the sample test. |
| “Profiles come in many shapes and colours of metal” | A recipe per profile family sets which faces are checked and the limits for each. | Faces the cameras cannot see, such as inside hollow sections, are not inspected. |
// The complete solution
Extrusion surface inspection station: hardware, software and installation
The cameras sit where profiles move singly and steadily towards the coating line, for example on the transfer to the pretreatment or racking area. An encoder on the conveyor times each line of the image to the profile’s travel. One line-scan camera covers the main visible face; three cover the top and both sides of most window and door sections.
[Schematic]Plan view, not to scale
Swipe sideways to see the whole drawing →
[1]: Times each image line to the profile’s travel, at any speed.
[2]: One camera on the main visible face, or three around the profile in the high case.
[3]: A high-power line light per camera, at an angle chosen for scratches and dents.
[4]: Marks or flags a defective profile so it is pulled before coating.
Hardware
| Component | HIKROBOT example | Role in this station |
|---|---|---|
| Camera (low) | HIKROBOT 4k GigE line-scan camera (MV-CL042-91GM) with a 25 mm 10 MP lens | Images the main visible face of each profile |
| Cameras (high) | 3 × HIKROBOT 4k GigE line-scan cameras (MV-CL042-91GM) with 25 mm 10 MP lenses, on a frame around the profile | Image the top and both sides |
| Lighting | HIKROBOT high-power line lights (MV-LTDS-H, 600 mm low, 1000 mm high), one per camera | Bright, even light along the scan line; angle chosen on samples |
| Processing | HIKROBOT VC3302P (low) or VC3704H (high) vision controller running HIKROBOT VisionMaster with deep learning | Builds the image from the line scans and runs the defect tools |
| Encoder | Local supply: incremental encoder on the conveyor | Keeps the image to scale at any conveyor speed |
| Marker and alarm output | Local supply: alarm and output to the line PLC; the marker device is yours or quoted separately | Flags the profile so it is pulled before coating |
Software and vision tools
| Tool or software | What it does here |
|---|---|
| HIKROBOT VisionMaster deep learning | Classification, segmentation and anomaly detection trained on your good and defective profiles, for marks that rule-based tools miss on textured metal. |
| VisionMaster defect tools | Region-based checks for scratches and spots where the surface is uniform enough for rules. |
| Profile recipe | Which faces are checked and the limits per profile family, chosen at changeover or from the PLC. |
| Defect log | Each fault with its image, position along the profile and profile ID or rack position, for trend reports per die. |
Installation and integration
01
Mounting
Cameras and lights on a rigid frame where profiles move singly and do not swing; a shroud keeps daylight and reflections out.
02
Lighting
Line lights at a low angle for scratches and dents; mill-finish and brushed surfaces are tested separately at the sample test.
03
Encoder and tracking
Encoder on the conveyor so the image keeps its scale through speed changes, and each fault is logged with its position along the profile.
04
PLC link
Alarm, fault code and marker output to the line PLC; recipe selection from the PLC or the operator screen.
05
Reject handling
Profiles are long and heavy, so a flagged profile is marked and pulled by the crew, or diverted where the conveyor allows, before racking. If the station is not ready, the alarm stays on so profiles do not pass unchecked unnoticed.
06
Recipes and operator screen
The screen shows the last profile’s image with each fault marked, so the crew can decide quickly on borderline marks.
// Indicative cost in South Africa
What this station costs, excluding VAT
Indicative budget for a typical application of this kind. Every project involves custom requirements (your products, line speed, layout, PLC and acceptance criteria), so the final price differs from case to case. A firm price follows the sample test.
- Low
- R 360 000
- one camera on the main face
- High
- R 1 340 000
- three cameras, frame, panel
- Basis
- Typical application of this kind, indicative
- South African rand, excluding VAT
- Prices as at 2 October 2026
- Travel (Gauteng): outside Gauteng quoted separately
// Included and excluded
The low figure is one line-scan camera on the main visible face with deep-learning defect tools, raising an alarm and marking flagged profiles. The high figure covers the top and both sides with three cameras on a frame around the profile and adds a panel and marker output. Both use line-scan cameras, which are outside the published general budget bands, so this page carries its own range. Prices are indicative, excluding VAT, at 2 October 2026 prices; a firm price follows the sample test on your profiles.
Hardware is priced from public list prices converted at the ECB reference rate, plus freight, clearance and margin; engineering at R 900 – R 1 300 per hour. How these ranges are built →
Swipe the table sideways to see the high figure and the basis →
| Cost block | Low | High | Basis |
|---|---|---|---|
| Vision hardware (landed, with margin) | R 144 513 | R 537 725 | Low: one 4k line-scan camera, lens, 600 mm line light, VC3302P controller. High: three 4k line-scan cameras and lenses, three 1000 mm line lights, VC3704H controller |
| Software licence | R 101 230 | R 120 036 | HIKROBOT VisionMaster with the deep-learning tier |
| Local materials and marker output | R 30 000 | R 115 000 | Low: encoder, mounts, shroud. High: encoder, frame around the profile, panel and HMI, marker output |
| Engineering (recipes, defect classes, PLC link) | R 54 000 | R 312 000 | 60 h at R 900 to 240 h at R 1 300 |
| Installation, FAT, SAT, training, documents | R 27 000 | R 124 800 | 30 h to 96 h |
| Project management | R 10 800 | R 52 000 | 12 h to 40 h |
| Travel (Gauteng) | R 2 000 | R 6 000 | Outside Gauteng quoted separately |
| Total (rounded) | R 360 000 | R 1 340 000 | Low: one camera on the main face. High: three cameras, frame, panel |
// Related technology
Which technologies this scenario uses, and why
The technologies behind this station, how much each matters here, and where each is explained in depth.
| Technology | Role | Why it matters in this scenario | Read more |
|---|---|---|---|
| Line scan imaging | Core | Long profiles moving past the camera are imaged line by line, so the whole length is covered at full resolution. | Web and line scan |
| Surface defect detection | Core | Scratches, dents, die lines and pick-up are surface defects found by contrast under the right light. | Surface defect detection |
| Deep learning (classification, segmentation, DL OCR) | Core | Textured, reflective metal is where deep-learning defect tools help most. | Rule-based vs deep learning |
| Lighting technique | Supporting | Low-angle line light makes scratches and shallow dents stand out on a reflective surface. | Machine vision lighting |
| PLC link and reject with confirmation | Supporting | Long profiles are marked and pulled rather than pushed off, so the alarm and marker link matters. | PLC and reject integration |
// Acceptance
How the station is proven before handover
Every station is accepted on your products, against numbers agreed in writing before the build. How we deliver →
01
Sample test
Your profiles in each surface finish, good and with each defect type; the result states which defects are reliably found on which faces, and whether one camera or three is needed.
02
FAT with challenge profiles
Before delivery, marked challenge profiles with each defect type must each be flagged, and good profiles must pass.
03
SAT on the line
On your line at production speed, false alarms and missed defects are counted over an agreed run against your acceptance specification.
// FAQ
Questions about aluminium extrusion surface inspection
Can a camera detect scratches, dents and die lines on aluminium extrusions before powder coating?
Often, yes, on the faces the cameras can see. Line-scan cameras with low-angle light make scratches, dents and die lines stand out on the metal, and deep-learning tools trained on your profiles separate real defects from normal texture. Which defects are reliably found, and from what size, is shown on your own profiles at the sample test.
What are the common surface defects on aluminium extrusions?
Die lines, which are continuous lines along the profile from the die bearing; pick-up, torn comma-like spots from material deposited on the die; blisters; streaks; and handling damage such as scratches and dents. Which of them matter depends on the finish and your customer’s specification.
Why inspect before coating rather than after?
Because a coated profile with a defect has to be stripped or scrapped, and the pretreatment, powder and oven time are already spent. Checking before coating also shows defects from the press as trends per die, while they can still be corrected.
How much does aluminium extrusion surface inspection cost in South Africa?
An indicative installed budget is R 360 000 – R 1 340 000 excluding VAT, at 2 October 2026 prices. The low figure is one line-scan camera on the main visible face with deep-learning tools and an alarm. The high figure is three cameras around the profile, a frame, panel and marker output. A firm price follows the sample test.
Does it work for anodised profiles too?
The check is the same before anodising: the surface is imaged before the profile is etched and anodised. The reject limits for anodised profiles can differ from those for powder-coated ones; they come from your specification.
Can it check the coating after the oven?
Not as part of this station. Visible coating faults after curing can be imaged with a separate camera, but film thickness, adhesion, colour and gloss need their own instruments.
Sources
- Aluminium Federation of South Africa: about AFSA and member segments (extruded products, architectural, surface finishing) (association site, read 2026-10-26)
- aluminium-guide.com: surface defects of aluminium profiles (die lines, pick-up, blisters, bearing streaks) (technical reference, read 2026-10-26)
- aluminium-guide.com: pick-up defect on aluminium profiles (technical reference, read 2026-10-26)
- HIKROBOT VisionMaster: deep-learning tools; defect detection finds small scratches and spots despite texture, colour and noise (vendor page, checked 2026-10-02)
- HIKROBOT Smart Product Catalog 2025Q4: VisionMaster defect tools including anomaly detection (p.62) (vendor catalogue, checked 2026-10-02)
- HIKROBOT MV-CL042-91GC 4k GigE line-scan camera: line and frame triggers (vendor product page, checked 2026-10-02)
- HIKROBOT light source catalogue: MV-LTDS-H high-power line-scan light (p.3) (vendor catalogue, checked 2026-10-02)
Related:
More line solutions in building materials and metals:
// Next step
Send us profiles, good and marked
Describe the line, the products and the fault. We test on your good and faulty samples and reply with a feasibility note and a budget.