AI-Powered Visual Inspection Delivers 95% Accuracy for Manufacturers

A manufacturing company’s outdated visual inspection system led to inconsistent defect detection and limited adaptability to increasing production demands. Implementing an AI-powered system boosted inspection accuracy to over 95%, significantly reduced training time for new operators, and enhanced defect detection capabilities.

Client:

Microtechnix is a Belgian company that specializes in developing image acquisition and analysis solutions for life science applications. Their equipment is highly sought after by professionals engaged in cell biology, microbiology, and pharmacological research. Renowned for its innovation and superior quality products, Microtechnix has established itself as a prominent figure within the life science imaging sector.

Problem Statement:

● Operator Dependence: Their current inspection system relied heavily on operators to assess photos against defect catalogs, leading to inconsistencies.

● Scalability Issues: Increasing production meant needing more inspectors, which was difficult and time-consuming.
● Knowledge Transfer Challenges: When employees left, their knowledge went with them, impacting quality control.
● Limited Training Time: The need for extensive training for new inspectors made it difficult to expand efficiently.
● Flawed Defect Reporting: Inaccurate and incomplete defect reports hinder defect tracking, analysis, and root cause identification.

Results:

☑️ AI boosted inspection accuracy (sensitivity, specificity, precision, accuracy > 95%), minimizing defect misses and false positives for consistent, reliable product quality.

☑️ Automated inspection scaled seamlessly, handling increased production volume without requiring additional operators, saving on hiring and training costs.

☑️ AI preserved and transferred knowledge, learning from labeled defect data, to maintain high quality standards even with employee turnover.

☑️ Reduced training time for new operators allowed faster onboarding of equipment and production expansion.
Automated defect reports with images and descriptions improved defect tracking, analysis, and root cause identification.

AI Solution:

Faced with limitations in their existing semi-automatic visual inspection system, which relied on operators comparing photos to defect catalogs, Microtechnix turned to Softengi, an AI development company. Together they created Ionbond, an innovative AI-powered visual inspection system that:

● Utilizes multi-layered AI: This enables it to identify and differentiate between various components and defects with greater accuracy, even for complex features like delamination (blank spots) and contamination (objects larger than 40 microns).

● Includes the PaDiM algorithm for anomaly detection: This method excels at identifying and pinpointing anomalies in images, crucial for catching subtle defects that might escape human inspectors.

● Integrates OpenCV: This library provides tools for image processing tasks like isolating and measuring damaged shapes, allowing for a more precise evaluation of defect severity.

Ionbond implementation involved:

Data collection and training: Operators collected and labeled images of good and defective products, creating a training dataset specifically tailored to identify the types of defects relevant to Microtechnix’s production process (e.g., damages, delamination, contamination).

Anomaly detection (Step 1): The AI first learned to differentiate between acceptable and unacceptable products, achieving a target precision of over 95%.

Object detection (Step 2): Once the basic functionality was established, the AI was further trained to identify the specific type of defect, allowing for more targeted corrective actions on the production line.

Continuous improvement: The system continuously learns and improves based on operator feedback and additional data, ensuring its effectiveness remains high as production processes evolve.

References:

1. AI-Based Visual Inspection Case Study

Industry: Manufacturing
Vendor: Softengi
Client: Microtechnix

Keywords: AI-powered visual inspection, AI in quality control, anomaly detection, AI-based quality inspection, Computer vision for manufacturing