Skip to main content
AI Automated Defect Classification AI Automated Defect Classification

Automated Defect Classification: Powered by Deep-Learning AI

The Role of Defect Classification in Industrial Automation

Deep-learning AI significantly enhances the speed and precision of automated defect classification (ADC) in industrial manufacturing. By enabling early defect detection and improving classification accuracy and consistency, AI-powered ADC allows factories to quickly identify and rectify defects, maximizing machine capacity and ensuring high product quality. This technology is highly sensitive and applicable to a wide range of industrial applications in both discrete and process manufacturing.

Robovision Platform AI Defect Classification

Built to Address Increasing Inspection Complexity

To boost yield, the ability to manage and classify defects efficiently is crucial. However, growing defect complexity and exploding product mix present major inspection challenges. Built to tackle tomorrow’s challenges, Robovision’s AI-powered ADC enables you to:

  • Customize inspection based on your specific product mix and processes
  • Speed up quality checks for real-time inline and offline inspection
  • Shorten Time to Resolution using efficient end-to-end workflows
  • Continuously improve accuracy
Comparison table image of AI-based Defect Classification

How it Works

Model Creation
The Human-in-the-loop (HITL) deep learning approach leverages both human and machine intelligence to create highly performant models.
Refining the Models
Putting defect classifiers in production holds an intrinsic opportunity for continuous refinement of the classifier model. This refinement can be triggered by several factors, including the lack of sufficient data outside the fab, the need to avoid downtime caused by minor SKU changes, the expansion of the model’s application across a wider range of SKUs or process steps, and the ongoing pursuit of improved purity and accuracy in the model’s performance.
Performance
During production, if a classification fails to meet the required level of certainty, it is manually classified by the operator. Labeled data is augmented and used to refine the relevant model. The control limit certainty can vary per product-process step, depending on the factory performance.

Real-World Applications

Automatic defect classification (ADC) for outgoing quality control

Defect classification is vital to determine if goods pass or fail inspection. Considering the quantity and severity of different defect types helps inform shipping decisions. The more granular the classification enabled by deep learning-based ADCs, the more valuable the insights are to suppliers and buyers.

Defect classification information can help:

  • Improve the supplier’s ability to self-identify and correct quality defects before outside inspection
  • Ensure more accurate inspection results that match quality tolerances and expectations
  • Reduce cases of “pending” results reported by the inspector due to unclear quality tolerances
Table image of defect detection in quality control

Spatial Signature Analysis

The spatial distribution of data (e.g., defects on a wafer surface) can be used to determine the potential source of problems in the manufacturing process. A spatial signature is defined as a population of defects originating from a single manufacturing problem. 


A significant proportion of systematic defects can manifest as spatial patterns (signatures) of failing chips on the silicon wafers. Combining wafer map pattern classification with a wafer’s equipment genealogy—such as the specific equipment that processes the wafer— assists engineers/technicians in pinpointing a root cause. 

Image of Wafer Map in Spatial Signature Analysis

Inconsistent Weld Bead Geometry

Inconsistent weld bead geometry means variations in bead width, height, alignment, and toe angle. These variations compromise product quality, increase rework, and lead to fatigue failures. Manual inspection is time-consuming and error-prone, which makes it unsuitable for high-throughput operations. By contrast, deep learning ADCs can automatically classify weld bead geometry defects, including overfill / underfill, misalignments, wrong toe angle, or inconsistent bead shape.

Robovision Platform AI Defect Classification in Weld Bead Geometry

AI ADC for Advanced Process Control

Advanced Process Control (APC) has been widely used in manufacturing to improve process and product performance. Key APC benefits include:

Faster root cause analysis

ADC categorizes defects and correlates them with process parameters. Engineers can pinpoint issues in the manufacturing line faster, perform corrective actions quicker, and ensure process optimization.

Improved process stability

ADC integrates with APC systems to provide real-time defect data. Dynamic adjustments can be made to processes, minimizing variations and boosting overall stability.

Yield improvement

Early detection and classification of defects allow manufacturers to take proactive measures, preventing yield loss and optimizing throughput.

Real-time defect identification

Reduce dependency on manual inspection and increase accuracy

Cost reduction

ADC reduces manual inspection costs and rework expenses, supporting efficient resource utilization.

Better decision-making and analytics

ADC provides trend analysis and predictive insights, allowing manufacturers to anticipate defects before they cause major disruptions.

Smooth Inline and Offline Integration

The Robovision AI-ADC supports both inline (or tool-centric) ADC and offline integration. This flexibility makes it applicable for multiple inspection applications, offering smooth integration with existing Manufacturing Execution Systems (MES). 

Leveraging Existing Imaging Tools and Data Sets

Imaging tools designed for other purposes can be used for defect classification. This approach means imaging datasets are already available, which speeds up time to resolution. By repurposing existing images through the Robovision AI-ADC solution, manufacturers can classify additional defects or implement excursion detection without investing in new equipment or major upgrades. This approach enables earlier identification of issues, leading to improved yields and increased ROI from existing equipment.

Image of Robovision Platform AI Platform

Talk to an expert about our vision AI software for your application