A Clean Graphical Interface
Set AI parameters manually or automatically and watch the results live on screen.
With the AI-Powered Real-Time Defect Detection System we developed at ARCA Yazılım, we remove the difficulties of manual quality control and turn your production line into an intelligent ecosystem.
This deep-learning platform brings the whole process together in a single solution — from capturing the camera image to building datasets, from model training to the real-time detection result reaching the operator.
Traditional inspection struggles to catch defects such as scratches or cracks continuously without slowing production. The human eye tires and loses focus — AI never does.
Human Error
Over long manual inspections, distraction and fatigue directly affect consistency.
Slowing Production
Inspecting products one by one by hand limits the speed and efficiency of the production line.
Defects Caught Too Late
Defects noticed at later stages lead to rework, material loss and expensive product returns.
Inspection That Cannot Scale
When the shift or the operator changes, the same defect may be judged differently.
The platform we developed integrates camera-based visual inspection into your production line with deep-learning technology. The system analyses continuously — without tiring, without losing focus and without slowing your production — processing large datasets in seconds to instantly spot patterns the human eye can miss.
“Quality control is not only about finding the defect, but finding it at the right time.”
The moment metal castings or components are presented to the camera, the system detects scratches, cracks and other deformations on them live, within milliseconds.
The metal casting or component enters the camera’s field of view on the production line; the system detects the part’s position automatically.
The deep-learning model processes the image, evaluating every region of the surface within milliseconds and extracting candidate defects.
Scratches, cracks and other deformations are classified live, each placed with a bounding box and a confidence value.
The detection result is monitored live on screen with its confidence values, and the part can be separated automatically where needed.
Every detection is logged; this data feeds quality reports and continuous improvement work.
Start the scan on the sample metal surface image below and see step by step how the AI model detects scratches, cracks and deformation.

How Does It Work?
When you press the button, the system scans the image just as it would on a real production line, analyses it with the AI model and marks each defect with its location and confidence value. This demo is a simplified simulation of the scan-analyse-detect flow in the real system.
Result Panel
Once you start the scan, the results will be summarised here.
Pick a sector from the tabs below and read how the system actually works in that production environment.
AutomotiveWheels and gearbox housings coming off the casting line pass in front of the camera at the quality control station. The system scans each part within seconds; the moment it detects a micro-crack it separates the part automatically and notifies the operator. At the end of the shift, the team sees a breakdown of every defect detected that day in a single report.
We designed a graphical interface that operators at every level can use comfortably, with no coding or AI expertise required.
The AI Defect Detection System is not just software but a complete ecosystem you can manage on your own. With a graphical interface (GUI) that operators at every level can use comfortably, and no need for coding or AI expertise, you can set AI parameters manually or automatically and monitor detection results live on real-time screens.
Set AI parameters manually or automatically and watch the results live on screen.
Build datasets from images of your own production line and label those images easily within the platform.
Train and optimise your own AI models for your production needs or for newly emerging defect types.
Push the best trained models straight into the live production environment in one click.
Try training your own model
mAP
—
precision
—
recall
—
The platform can be customised for different metal types, defect categories, camera infrastructures, lighting conditions and production speeds. The aim is for users to manage the system without needing AI or software expertise.
We do not offer a one-size-fits-all package; we configure the system together, around your business’s real production conditions.
While speeding up your quality control processes, this system delivers directly measurable value to your business.
A 97% accuracy rate delivers consistent, repeatable performance across your quality control processes.
Thanks to multi-threading optimisation it runs smoothly at high FPS values, with no lag at all.
You are not dependent on cloud systems or external services; all data and models belong entirely to you.
It lowers inspection times and labour costs, and prevents defective products reaching the customer along with expensive recalls.
Where do the cost savings concretely come from?
The 97% accuracy rate is based on project data; the 100% data security statement refers to the system being able to run without a mandatory dependency on external cloud services, and to ownership of the data and models remaining with the business.
The system has a flexible structure that adapts easily to different production lines. Its main areas of use:

Quality control of critical metal parts such as wheels, gearboxes and engine housings in spare part and component manufacturing.

Precise visual inspection of aerospace metal castings, where tolerances sit far below a millimetre.

Defect detection in the manufacture of industrial machine parts such as high-value machine housings and gear assemblies.

Uninterrupted inspection that does not slow production across laser cutting, welding and general metalworking processes.
We know the difficulties of industrial processes. We built our software with engineering precision around the needs of real production lines, not a laboratory environment.
Every production line is unique. Instead of standard packages, we offer a flexible structure that fits your metal types, your lighting conditions and your defect types exactly.
The AI world changes fast. By integrating the latest deep-learning algorithms into your system, we keep your investment current at all times.
As a local partner we offer fast technical support and on-site response, while making no compromise on international quality and security standards in our software architecture.
With this technological foundation that grows and scales alongside your production capabilities, take your product quality and customer satisfaction to the highest level. Get in touch with us for solutions tailored to your project and a system demo.