Client Overview:

Objectives:

1. Automate the Defect Detection Process:
   Implement a Machine Vision AI system to automate the inspection of over 400 different metal parts, eliminating manual inspection processes.
2. Ensure High Precision in Defect Detection:
   Develop an AI solution capable of detecting defects as small as 0.5 mm, with a range up to 5 mm, across various metal parts to maintain consistent quality and prevent faulty parts from proceeding to packaging.
3. Customize Inspection Based on Part Categories:
   Design an AI system that can automatically identify each metal part and apply part-specific defect detection algorithms, ensuring the inspection system is optimized for different sizes, shapes, and tolerances of the metal parts.
4. Minimize False Positives and Negatives:
   Implement dynamic defect tolerance thresholds that adjust based on the part category, reducing the occurrence of false positives (incorrect defect detection) and false negatives (missed defects), to improve the overall reliability of the inspection process.
5. Increase Inspection Efficiency:
   Achieve faster inspection throughput by automating the process, enabling real-time defect detection on the production line.
6. Increase customer Satisfaction and reduce returns:
   Deliver the high quality products and minimize the costly reruns and recalls. 
7. Enable Scalability for Future Growth:
   Design the system to be scalable, allowing the client to add new part categories and defect types as their production expands, ensuring the inspection solution can grow with the company’s needs.

Challenges:

1. Diverse Range of Metal Parts:
   The company produced over 400 different types of metal parts, each varying in size, shape, and design. This presented a significant challenge because a single AI algorithm could not uniformly handle the wide variation in part characteristics.
2. Different Defect Types and Sizes:
   The nature and size of defects varied from part to part. The smallest defect the client was concerned about was as small as 0.5 mm, and the largest defects could reach up to 5 mm in size. Each part had a different tolerance threshold for defects, meaning that a defect considered critical in one part may not have been significant in another.
3. Inconsistent Relevancy of Defect Detection:
   Not all defects were relevant to all parts. For example, a small scratch might be irrelevant for some large components, while it could be grounds for rejection in smaller, more sensitive parts. This required a nuanced, part-specific approach to inspection that manual inspection teams struggled to manage, let alone automate using a standard AI system.
4. Part Identification:
   The first step of the inspection process required identifying the specific part under inspection to ensure that the correct defect detection parameters were applied. A one-size-fits-all approach to defect detection would result in either false positives or missed defects.

Solution:

To meet these challenges, we implemented a customized Machine Vision AI system with advanced part identification and defect detection algorithms, designed to address the variability in parts and defects.

1. Part Identification Module:
The AI system was built with an initial part identification stage. Using machine vision cameras and machine learning-based pattern recognition, the system accurately identified the specific part type under inspection. Each part had a unique visual signature that allowed the system to classify it and select the corresponding inspection parameters from a pre-configured database.

2. Defect Detection Engine:
After identifying the part, the AI system applied a tailored defect detection algorithm. We developed a unique algorithm for each category of metal parts based on their size, shape, and function. These algorithms detected corrosion, pits, scratches, marks, and cracks with high precision by using image processing techniques like edge detection, texture analysis, and pixel-level anomaly recognition.

3. Part-Specific Defect Tolerance Thresholds:
One of the key innovations was the ability to set part-specific defect tolerance thresholds. Based on the identified part, the AI system dynamically adjusted its sensitivity to different defects. For instance, it set a higher tolerance for large parts where small scratches were not critical, while for smaller, more precise parts, even minimal scratches were flagged for rejection. This dynamic thresholding ensured high accuracy and reduced false positives and negatives.

4. Real-Time Data Processing and Feedback:
The AI system was designed to inspect parts in real-time on the production line. High-speed line scan cameras captured images of each part as it moved through the inspection station, and the AI analyzed the images instantaneously. When a defect exceeding the threshold was detected, the system flagged the part for further manual inspection or automatic rejection. Detailed defect reports were generated for each part, documenting the type, location, and severity of the defect.

 

Implementation Process:

1. Discovery and Planning:
   We conducted an extensive discovery phase to understand the client's unique needs, reviewing the variety of metal parts and the types of defects they aimed to detect. We worked closely with the client’s quality control team to define the defect tolerances for each part category.
2. Imaging Device Selection and hardware provision:
   To ensure we captured the high-resolution images required for detecting small defects (as small as 0.5 mm), we selected a high-resolution, smaller pixel size line scan camera industrial camera that we integrated with our controller, Intelgic’s Live Vision software and lighting solutions to ensure optimal inspection across various metal surfaces.
3. AI Model and algo Development:
   We developed and trained an AI model for classifying parts and their categories. We developed a complex algorithm for defect relevancy check for each part category and flag it accordingly.
4. AI Training and Fine-Tuning:
   We continuously monitored the system during initial deployment to refine and fine-tune the AI model and algos. This phase included adjusting defect tolerance levels for each part category based on real-world inspection results.
5. System Integration and Testing:
   The Machine Vision AI system was fully integrated into the client’s existing production line, enabling seamless defect detection without slowing down operations. Extensive testing was performed to ensure the system met the client’s quality and performance expectations.

Results:

• Enhanced Inspection Accuracy:
  The AI system provided a significant improvement in inspection accuracy, reducing human error and detecting even the smallest defects, down to 0.5 mm, with high precision.
• Increased Efficiency:
  The automated system was capable of inspecting hundreds of parts per hour, far exceeding the throughput of the manual inspection process. This led to a marked improvement in production efficiency.
• Reduced False Positives/Negatives:
  By dynamically adjusting defect detection thresholds based on part category, the system reduced both false positives (flagging acceptable parts as defective) and false negatives (failing to detect actual defects).
• Cost Savings:
  By automating the inspection process, the client significantly reduced labor costs associated with manual inspection and decreased the number of defective parts that made it through to assembly, saving on rework and returns.

 

By implementing a custom-designed Machine Vision AI system, we successfully helped our client automate their metal parts inspection process, leading to greater efficiency, higher accuracy, and significant cost savings. The tailored approach to part identification and defect detection thresholds allowed the system to handle the variability in parts and defects, ensuring that only parts meeting the company's high-quality standards proceeded through the production line. The result was a more reliable and scalable inspection solution that can grow with the company’s future needs.