How AI Cameras Check Factory Parts for Flaws

This patent describes an automated system using a camera and artificial intelligence to inspect manufactured parts, identifying defect types, locations, and confidence levels, then communicating this data to a factory controller.

What it covers

The system uses a camera to capture images of a manufactured item (Claim 1). A 'node computing device' then analyzes these images using a specialized AI model, specifically a machine-learning based object detection model (Claim 1). This model is trained to spot various types of flaws, like a scratch or a dent, and can classify them into three or more categories (Claim 1). It outputs 'defect data' which includes the type of defect, its exact location, and how confident the AI is about its finding (Claim 1). This defect data is then sent to a 'programmable logic controller' (PLC), which is a computer that controls factory machinery (Claim 1). For example, if inspecting a car part, the system could take pictures, identify a specific crack, and then tell the PLC to reject that part. The system can also use a robotic arm to move the camera (Claim 3) or rotate the part being inspected (Claim 4) to get a full view, and can even stitch multiple images together for a complete picture (Claim 5). It can also confirm a defect by tracking it across multiple image frames before reporting it (Claim 7).

What it doesn't cover

• —Does not cover visual inspection systems that use traditional rule-based computer vision instead of a machine-learning based object detection model.
• —Does not cover systems that only classify defects into fewer than three categories or detect only a single type of defect (Claim 1).
• —Does not cover systems where the AI output lacks specific defect class, location, and a confidence level (Claim 1).
• —Does not cover systems where the AI output is not sent to a programmable logic controller (PLC) for factory automation (Claim 1).
• —Does not cover manual human inspection processes, as it specifies an automated AI system.

The clever bit

The clever part is combining a machine-learning object detection model that performs multiclass classification with a PLC, ensuring the AI not only spots flaws but also provides detailed defect data (class, location, confidence) directly to factory automation systems. This allows for precise, automated responses like rejecting a part or moving a camera for further inspection.

Why it matters

This patent is significant for improving quality control in manufacturing. By automating the inspection process with AI, factories can detect defects more consistently and quickly than human inspectors. This leads to higher product quality, reduced waste from faulty parts, and lower labor costs. The ability to precisely locate and classify defects also helps manufacturers understand and fix issues in their production lines.

Real-world examples

• 1.Automotive parts manufacturing quality control
• 2.Electronics assembly line inspection
• 3.Pharmaceutical packaging defect detection
• 4.Food processing quality checks
• 5.Metal fabrication flaw detection