Client Overview:

Objectives

1. Automate the Manual Inspection Process
   Replace the slow and inconsistent manual inspection process with an AI-driven machine vision system to enhance the speed and accuracy of defect detection.
2. Detect Fine Defects at Sub-Millimeter Levels:
   Implement a system capable of identifying defects as small as 0.2 mm, such as splits, hairline scratches, holes, extra threadings, and other manufacturing flaws, ensuring high-quality parts for painting and assembly.
3. Improve Overall Efficiency and Reduce Scrap:
   Reduce the amount of defective parts that proceed to the painting process, which leads to revenue loss. Ensure that defective parts are detected early, minimizing scrap and improving production efficiency.
4. Establish a Feedback System:
   Create a feedback loop to track the frequency and types of defects, enabling the client to identify root causes and implement corrective measures to reduce the occurrence of defects over time.

Challenges

1. Large Field of View (FOV):
   Inspecting large metal parts like doors and bonnets requires a wide FOV to capture all surfaces, including curved and angled sections. Traditional inspection methods were limited by the FOV of the cameras used, making it difficult to capture every surface effectively.
2. Complex Part Placement:
   The parts were inspected in a stationary section where operators manually positioned them for inspection. Ensuring accurate placement was critical, as misalignment could result in missed defects.
3. Surface Curvature and Multiple Angles:
   The large automotive parts were not flat, with curved surfaces and multiple angles, complicating the inspection process. A single camera setup was insufficient to capture all the areas prone to defects.
4. Backlighting for Hole Detection:
   Detecting smaller defects like holes required specialized backlighting to highlight imperfections that would otherwise be missed by front-facing lighting systems.
5. Lighting Conditions:
   Inspecting different types of defects required varied lighting conditions. For example, surface defects like scratches and splits required front lighting, while small holes needed backlighting for better visibility.

Solution

We designed a comprehensive AI-based machine vision system that leveraged multiple cameras and a two-phase lighting system to meet the client’s complex inspection requirements.

1. Multiple Cameras with Large FOV and High Resolution:
   We installed multiple high-resolution cameras with a large FOV and smaller pixel size to capture every part of the large automotive components, including the curved surfaces and angled sections. The cameras were strategically positioned to cover all critical areas of the parts.
2. Marked Inspection Area for Accurate Part Placement:
   To address the challenge of part positioning, we created a clearly marked inspection area where operators could easily station the parts in the correct position for inspection. This reduced the chances of misalignment and ensured that all surfaces were captured by the cameras.
3. Two-Phase Lighting System:
   We implemented a two-phase lighting system. In the first phase, front lighting was used to detect surface defects such as scratches, splits, and extra threading. In the second phase, backlighting was applied to detect smaller defects such as holes and fine surface irregularities. This combination ensured that no defect, regardless of size or type, was missed during inspection.
4. AI-Driven Defect Detection Algorithms:
   The system utilized advanced AI algorithms capable of detecting various types of defects—splits, hairline scratches, holes, and more. These algorithms were trained to detect even the smallest defects (0.2 mm and larger) and categorize them for reporting and feedback purposes.

Implementation Process

1. Discovery and Requirements Gathering:
   We began the project by conducting a detailed analysis of the client’s existing inspection process, including the types of defects they aimed to detect, the size of the parts, and the specific inspection challenges they faced. We also collaborated with the client’s quality control and production teams to define the necessary tolerances and inspection parameters.
2. Imaging and Camera Selection:
   Given the large size of the parts and the need to detect defects as small as 0.2 mm, we selected high-resolution industrial cameras with smaller pixel sizes and large FOV. Multiple cameras were positioned at different angles to ensure complete coverage of the curved and angled surfaces of the parts.
3. Lighting Design:
   To ensure that all defects were visible, we designed a two-phase lighting system with front lighting for surface defects and backlighting for holes and fine irregularities. This lighting combination was crucial for detecting all types of defects, regardless of size or orientation.
4. AI Model Training and Testing:
   The AI models were trained using a dataset of images that included both defective and defect-free parts. The models were optimized to detect various types of defects—splits, scratches, holes, and extra threadings—across different part surfaces. Extensive testing was performed to ensure the models were accurate and reliable in real-world production conditions.
5. System Integration:
   The machine vision system was fully integrated into the client’s production line. The system operated in the stationary inspection section, where operators brought parts for inspection. The marked inspection area guided operators to position parts correctly, ensuring consistent and reliable inspection results.
6. Feedback System Implementation:
   We incorporated a feedback system that captured data on detected defects, enabling the client to analyze trends and adjust their manufacturing processes to prevent recurring defects. This continuous improvement process helped the client enhance product quality over time.

Results:

1. Increased Inspection Speed
   The automated system significantly reduced inspection time by enabling real-time detection of defects, allowing hundreds of parts to be inspected per hour. This replaced the time-consuming manual process, which previously limited throughput and led to production bottlenecks.

2. Improved Accuracy and Defect Detection
   The system was able to detect even the smallest defects, as small as 0.2 mm, across large, curved, and multi-angled surfaces. With the high-resolution cameras and the two-phase lighting system, the system successfully identified defects such as hairline scratches, splits, holes, and extra threadings with exceptional accuracy, minimizing false positives and false negatives.

3. Consistent and Reliable Inspection
   The automation of the inspection process ensured consistent quality control across all parts, eliminating human error and variability in defect detection. The system’s AI algorithms performed at a high level of reliability, maintaining accuracy regardless of the operator or production conditions.

4. Scalable Solution for Future Growth:
   The system was designed to be scalable, enabling the client to easily adapt it for new parts or increasing production demands. The modular design and adaptability of the AI algorithms ensure that the solution can evolve alongside the client’s growing operational requirements.

5. Cost Savings and Increased Efficiency:
   The reduction in manual labor, combined with the prevention of defective parts from advancing in the production line, resulted in significant cost savings. The automated system also improved operational efficiency by streamlining the inspection process and enabling the client to meet production deadlines more effectively.

In summary, the AI-based machine vision system provided the client with a faster, more efficient, and highly accurate inspection process that not only reduced costs but also enhanced product quality, resulting in long-term operational improvements and financial benefits.