Idea
Platform automating defect detection model design to improve industrial quality and reduce manual effort.
Research Paper
Core Innovation
This paper presents AutoNAD, which jointly searches convolutional, transformer, and MLP architectures to capture local and global defect features. It introduces cross weight sharing to speed supernet training and a multi-level feature aggregation module for enhanced multi-scale learning. Latency-aware priors guide efficient architecture selection, balancing accuracy and runtime for industrial deployment.
Why It Matters
Industrial manufacturers face challenges detecting diverse surface defects accurately and efficiently, impacting product quality and operational costs. AutoNAD automates model design to handle defect variability and runtime constraints, enabling scalable, reliable defect detection that reduces manual trial-and-error and accelerates deployment in production lines.
Market Size (TAM)
$10–20B TAM for industrial AI and defect detection; $2–5B SAM from manufacturing and quality control sectors. Driven by automation demand and quality assurance needs.
Potential Customers & Pain Points
- Manufacturing plants–Need accurate fast defect detection
- Quality control teams–Require scalable adaptable inspection models
- Industrial AI solution providers–Seek efficient model design tools
- OEMs–Demand integration of reliable defect detection in production.
Business Model
Subscription-based SaaS platform offering automated defect detection model design and deployment tools, with tiered pricing based on usage and support levels.
Competitive Landscape
- Landing AI
- Cognex
- Instrumental
- Neurala
Implementation Challenges
- Integration complexity with existing manufacturing systems
- Data variability across industries requiring customization
- High initial investment for AI deployment
Validation Strategy
- Pilot deployments with manufacturing partners on diverse defect datasets
- Benchmarking against manual and existing automated defect detection models
- Measuring improvements in detection accuracy
- runtime efficiency
- and deployment speed
Research Paper Overview
Automated Neural Architecture Design for Industrial Defect Detection
Summary
Industrial surface defect detection (SDD) is critical for ensuring product quality and manufacturing reliability. AutoNAD is an automated neural architecture design framework that searches over convolutions, transformers, and MLPs to address intraclass difference and interclass similarity in SDD. It introduces cross weight sharing for efficient training, a multi-level feature aggregation module for multi-scale learning, and latency-aware priors for runtime efficiency. Validated on three industrial datasets and integrated into a defect imaging platform, AutoNAD reduces manual design costs and improves detection accuracy and efficiency.