...
1 %
Defect detection accuracy
1 %
Reduction in waste
Computer Vision

How AutoTech Components Reduced Production Waste by 40% With Custom Computer Vision

Industry

Automotive Manufacturing

Size

200-500 Employees

Location

Detroit, MI

Background

AutoTech Components specializes in precision-machined engine parts for major automotive brands. Their reputation hinges on delivering flawless components with microscopic tolerances. For years, they relied on a team of human quality assurance (QA) inspectors to manually review parts on the assembly line. As production volumes scaled to meet new electric vehicle (EV) contracts, the manual inspection process became a critical bottleneck, unable to keep pace with the high-speed conveyor systems while maintaining their strict quality standards.

The challenges

As the company ramped up production, the limitations of manual inspection became apparent. The CEO and the Head of Production, Sarah Jenkins, recognized that human inspectors, while skilled, suffered from fatigue after long shifts, leading to inconsistent results. The “false acceptance” rate (bad parts getting through) began to creep up, risking contracts with Tier-1 OEMs. Furthermore, the inspection process was forcing the assembly line to run at 70% capacity, costing the company millions in potential revenue. They needed a non-invasive, high-speed automated solution that could integrate with their existing conveyor belts without requiring a factory redesign. 

                 


The computer vision model didn’t just match our human inspectors; it surpassed them. We are catching microscopic hairline fractures that were previously impossible to see with the naked eye at speed.”

Sarah Jenkins

Head of Production

 

The Solution

We designed and deployed a custom edge-based computer vision system tailored to AutoTech’s specific defect types.

We set up an image acquisition rig to capture 15,000 labeled images of both perfect and defective parts, training the model on specific defects like scratches, dents, and casting voids.
We utilize the YOLOv8 architecture for real-time object detection, fine-tuning the model to achieve inference speeds of 45 milliseconds per part.
The system was deployed on NVIDIA Jetson Orin devices directly on the factory floor, ensuring processing happened locally to maintain data security and eliminate latency.

The Result

AutoTech successfully moved from a manual, sampling-based inspection process to 100% automated inspection for every single part leaving the line.

The assembly line speed was increased to 100% capacity as the AI system could inspect parts faster than they could be manufactured.
By catching defects earlier in the process (before finishing steps were applied), the company reduced material waste by 40%.
We built a custom dashboard that visualizes defect trends in real-time, allowing engineers to pinpoint which upstream machines are causing specific issues.

Keep exploring more Services

Create your account

Seraphinite AcceleratorOptimized by Seraphinite Accelerator
Turns on site high speed to be attractive for people and search engines.