End inconsistent judgments
Aligns judgment criteria that varied by person into one model.
ECHO AI VISION
One camera handles product-type sorting, foreign matter · damage, and print inspection in a single pass. When the product changes, there is no re-setup; just retrain the AI.
- Inspect in one pass
- Type · foreign matter · damage
- Missing · smear · expiry reading
- Printed text
- Real-time analysis
- About 100 images/s
- New products just need photos
- Retrain
Key benefits
Keep using it as products change
AI sorts product types automatically on high-mix lines.
Catch the defects rule-based inspection misses
Finds compound defects like foreign matter, damage, and smeared print.
Trace defects back to the root cause
Inspection results link to process records to pinpoint the problem process.
Who uses it
Used together by quality, production, and process engineering
Quality managers
Review inspection results and pass/fail stats daily, and act on the product types and processes where defects cluster.
Production line staff
Inspection runs at line speed, and defect alarms link to the line for an immediate response.
Process engineering · IT staff
Collect and mark photos of the new product type, then retrain and deploy the AI.
Functions
Sort, find, read
01
Sort product types automatically
AI distinguishes products mixed in with similar types.
02
Find foreign matter · damage
Finds and marks spots with foreign matter, damage, or shape defects.
03
Read printed text
Catches missing and smeared print, and reads expiry dates to verify them.
04
Start even with few defect photos
When defect samples are scarce, synthetic defect images are generated for training.
05
Retrain for new product types
Collect photos, mark, train, deploy, monitor, and retrain in one flow.
06
Keep results as statistics
Keeps inspection results as history and generates pass/fail statistics automatically.
Screens
Check inspection results on screen
Screen 1 · line camera, conveyor, good · type · foreign matter judgment
Catch foreign matter · mix-ups on packing lines
Inspects every product at full line speed, instantly picking out products with foreign matter or a mixed-in product type.
- Foreign matter · damage
- Product mix-up
- At full line speed
Screen 2 · expiry date, LOT, manufacturing date, smear check
Expiry print check
Reads and verifies expiry dates and LOT numbers on packaging. Smeared or missing print is flagged for reprinting.
- Text reading
- Smear · missing
- Reprint alert
Screen 3 · collect photos, mark defects, train, deploy, monitor floor, synthetic defect images
When a new product launches
No re-setting inspection conditions from scratch. Collect photos of the new product and retrain the AI; if defect photos are scarce, fill in with synthetic defect images.
- Just collect photos
- Synthetic defects
- Deploy now
Use cases
Start by replacing visual inspection
Visual inspection
AI vision inspects cosmetic defects on appliance assembly lines.
Packing inspection
Identifies product types in packing and reads printed expiry dates.
Temperature tracking
Tracks the flame surface temperature in grinding with a camera.
Print quality
Catches missing and smeared print on pharmaceutical and food packaging.
Case studies
From appliances to food, it sees before the human eye
Large appliance manufacturing
One model for cosmetic inspection judgments
- Projects
- Visual inspection on the assembly line varied by inspector, and tracing which process a defect came from was difficult.
- Method
- Good and defect images were collected and labeled, scarce defect samples were filled with synthetic defect images, and a detection model was trained. Results are kept as history and linked to process records.
- What changed
- One standard for judgment. Find the line and time a defect occurred right away.
Food manufacturing · packing
Reads product type and expiry print in one pass
- Projects
- Rule-based readers missed smeared or distorted print, and inspection conditions had to be reset with every product change.
- Method
- A classification model identifies packaged product types, and AI character recognition reads expiry dates. New products are added by retraining with images, not by changing conditions.
- What changed
- The same camera catches mix-ups and print defects together. No line stop for product changeovers.
Abrasives manufacturing
A camera measures the temperature of flames you can't touch
- Projects
- Flame surfaces in grinding couldn't be measured by contact sensors, so temperature control was guesswork.
- Method
- AI finds and tracks the flame region in process video to estimate and record surface temperature.
- What changed
- Temperature is recorded as numbers, so process conditions and quality can be compared together.
PCB manufacturing
Connect semiconductor equipment without foreign libraries
- Projects
- Each line had tools from different makers, so aligning SECS/GEM communication took time, and non-standard commands had to be left unintegrated.
- Method
- Our own SECS/GEM receives equipment events, alarms, and production data, with non-standard commands tailored per tool for two-way control. Messages were tested in advance with a scenario editor.
- What changed
- Adding tools is a matter of configuration, not communication development. Even vision inspection was tied into one automated line.
Use-case scenarios
Link MES and LINKER, and the line runs itself
Add products one at a time as needed, and expand like this.
- AI VISION Judges defects.
- MES Inspection results are recorded to the work LOT.
- LINKER Sends a stop signal to the equipment.
- AI VISION Count pass and fail.
- LinkerFlow Show defect rates on the dashboard.
- Warns when the threshold is exceeded.
- AI VISION Collect defect images.
- Orchestrator Compiles causes alongside process records.
- Build the daily quality report.
Alongside existing systems
From camera installation to integration, we do it with you
Deployment model
Configuration and cost vary with scope, user count, and installation environment. Tell us your situation and we will recommend the right approach.
Contact salesInstallation environment
Installed on industrial PCs beside the line and on company servers. Camera and lighting installation are handled too.
Existing system integration
Links camera and equipment signals, and exchanges inspection results with MES via API.
Open standards
Trained models export to ONNX for use on other inspection equipment.
FAQ
FAQ
Can it learn with few defect samples?
Generates synthetic defect images to fill in scarce defect data.
Do I need to reconfigure when the product changes?
Retrains the model with new product images, without resetting conditions.
How fast is inspection?
Analyzes about 100 images per second in real time, as proposed. Configured to line speed and resolution.
How long does deployment take?
The standard schedule is 1 month of requirements analysis, 1–2 months of image collection and labeling, 1 month of trial, and 1.5–2 months of line integration.
Do you use open source?
No. Wavegram has specialized in AI vision and AI products since 2020. Because everything is built on our own technology, we can optimize for any customer's unique processes.