Industrial
Visual Quality Inspection and Defect Detection
Unsupervised anomaly detection on DINOv2 features, YOLO detectors and vendor CV models.
Named Models
Kunlun Large ModelChina National Petroleum CorporationReleasednot in the notesRuns onnot in the notesClassvisionCNPC model on Huawei Pangu for pipeline porosity and micro-crack detection; vendor claims about 40% higher defect identification efficiency and 25% less manual work.Commercial use: proprietary terms; Jakarta hosting: unknown; Indonesian support: not stated
Pangu CVHuawei (Huawei Cloud)Releasednot in the notesRuns onnot in the notesClassvisionHuawei vision model for mines (cracks, belt damage, hot bearings, leaks, loose rock); won Mining Technology Excellence Awards 2025.Commercial use: proprietary terms; Jakarta hosting: unknown; Indonesian support: not stated
Pangu Mine ModelHuawei (Huawei Cloud)Released2023Runs onnot in the notesClassLLMHuawei mining model from 2023, used at Shandong Energy.Commercial use: proprietary terms; Jakarta hosting: unknown; Indonesian support: not stated
Dinomaly2Tsinghua University, Beijing Institute of Technology, Shanghai Jiao Tong University, Fudan UniversityReleasedOct 2025Runs onnot in the notesClassvision
Research model for industrial visual anomaly detection; image AUROC MVTec AD 99.9, VisA 99.3, Real-IAD 92.1. Licence unverified.Commercial use: not stated; Jakarta hosting: unknown; Indonesian support: not stated
Intel weld porosity detectionIntelReleasednot in the notesRuns onCPU serverClassvisionIntel OpenVINO classifier that detects porosity defects in welding video for industrial inspection.Commercial use: not stated; Jakarta hosting: unknown; Indonesian support: not stated
Optical Inspection and PCB Defect ClassificationNVIDIAReleasednot in the notesRuns onnot in the notesClassvisionNVIDIA models for automated optical inspection that classify printed circuit board defects.Commercial use: not stated; Jakarta hosting: unknown; Indonesian support: not stated
Visual ChangeNetNVIDIAReleasednot in the notesRuns onnot in the notesClassvision
NVIDIA TAO change detection models (classification and segmentation) that compare two images for defects or scene changes.Commercial use: not stated; Jakarta hosting: unknown; Indonesian support: not stated
PatchCoreAmazonReleasedJun 2021Runs onnot in the notesClassvision
Industrial visual anomaly detector that compares image patches with a memory bank of normal patch features, from the paper Towards Total Recall in Industrial Anomaly Detection (Roth et al., CVPR 2022).Commercial use: allowed; Jakarta hosting: unknown; Indonesian support: not stated
PaDiMCEA (French Alternative Energies and Atomic Energy Commission)ReleasedNov 2020Runs onnot in the notesClassvision
Patch distribution modelling framework for detecting and localising anomalies in images. Paper by Defard et al. (2020).Commercial use: allowed; Jakarta hosting: unknown; Indonesian support: not stated
EfficientADMVTec Software GmbHReleasedMar 2023Runs onnot in the notesClassvision
Visual anomaly detector built for millisecond-level latency. Paper by Batzner et al. (2023).Commercial use: allowed; Jakarta hosting: unknown; Indonesian support: not stated
Discriminatively trained reconstruction model for surface anomaly detection. Paper by Zavrtanik et al. (2021).Commercial use: allowed; Jakarta hosting: unknown; Indonesian support: not stated
CS-FlowLeibniz University HannoverReleasedOct 2021Runs onnot in the notesClassvisionCross-scale normalising flows for image-based defect detection. Paper by Rudolph et al. (2021).Commercial use: allowed; Jakarta hosting: unknown; Indonesian support: not stated
Fast surface defect detector that unifies unsupervised and supervised learning. Paper by Rolih et al. (2024).Commercial use: allowed; Jakarta hosting: unknown; Indonesian support: not stated
GLASSInstitute of Automation, Chinese Academy of SciencesReleasedJul 2024Runs onnot in the notesClassvisionUnified anomaly synthesis strategy for industrial anomaly detection and localisation. Paper by Chen et al. (2024).Commercial use: allowed; Jakarta hosting: unknown; Indonesian support: not statedWhere this comes from
Models the research names for this use case. Research, with sources on each model record.
Who Runs It
production, 5
PT Aneka Tambang Tbk (ANTAM), ferronickelCountry: IndonesiaConveyor belt tear and oversize detection; maintenance needs; reductant and dolomite composition monitoring (not stated)Vendor: not statedproductionDate9 Sep 2026AI is used per MIND IDEvidence: company statementConfidence: lowSource: en.antaranews.com
China National Petroleum Corp.Country: ChinaKunlun Large Model on Huawei PanguVendor: CNPC with HuaweiproductionDate24 Jun 2025Vendor claims about 40% higher pipeline-defect identification efficiency and 25% less manual workEvidence: company statementConfidence: lowSource: intelligentcio.com
Pertamina New & Renewable Energy (Pertamina NRE)Country: IndonesiaDrone plus AI inspection of a solar power plant in Riau for hot spots and panel anomalies (AI software not named)Vendor: Terra Drone Indonesia (drones); AI vendor not statedproductionDate24 Apr 2025
Inspection completed December 2024, to repeat periodically; none measuredEvidence: company statementConfidence: lowSource: listrikindonesia.com
PT Akebono Brake Astra Indonesia (Astra Otoparts group)Country: IndonesiaAI machine for final inspection of brake pads that flags likely cracks (shown at Hannover Messe 2023)Vendor: not statedproductionDate19 Apr 2023In use; scope and results not statedEvidence: company statementConfidence: lowSource: idxchannel.com
Shandong EnergyCountry: ChinaPangu Mine ModelVendor: HuaweiproductionDate2023Evidence: company statementConfidence: lowSource: mining-technology.comWhere this comes from
Documented uses by named organisations. The evidence column says who made the claim.
Candidate Models
COCO val
A general benchmark for this model class, not a test on this use case. Object detection and segmentation AP (mAP50-95). Higher is better. Showing 10 of 11.
RF-DETR 2XLRoboflow vendor-reported60.1
D-FINEUniversity of Science and Technology of China (USTC), Institute of Artificial Intelligence, Hefei Comprehensive National Science Center independent55
RF-DETR MRoboflow vendor-reported54.7
RF-DETR SRoboflow vendor-reported52.9
RF-DETR NRoboflow vendor-reported48050100
RF-DETR 2XLRoboflow, visionReleasedNov 2025runs on edgehas scoreslast 18 months
RF-DETR MRoboflow, visionReleasedNov 2025runs on edgehas scoreslast 18 months
RF-DETR NRoboflow, visionReleasedNov 2025runs on edgehas scoreslast 18 months
RF-DETR SRoboflow, visionReleasedNov 2025runs on edgehas scoreslast 18 months
SAM 3Meta, visionReleasedNov 2025open weightshas scoreslast 18 monthsShowing 10 of 654. All models of this class
Where this comes from
Derived, not researched: models in the directory whose class matches a class this use case uses, excluding family records and models shut down or deprecated. Ranked by how many fit facts they meet, then by release date. Each fit fact is read from the model record.
Links
Models 15
Dinomaly2Tsinghua University, Beijing Institute of Technology, Shanghai Jiao Tong University, Fudan University, vision
Kunlun Large ModelChina National Petroleum Corporation, vision
Pangu CVHuawei (Huawei Cloud), vision
CS-FlowLeibniz University Hannover, vision
EfficientADMVTec Software GmbH, vision
GLASSInstitute of Automation, Chinese Academy of Sciences, vision
Intel weld porosity detectionIntel, vision
Optical Inspection and PCB Defect ClassificationNVIDIA, vision
PaDiMCEA (French Alternative Energies and Atomic Energy Commission), vision
Pangu Mine ModelHuawei (Huawei Cloud), LLMDeployments 5
China National Petroleum Corp.Kunlun Large Model on Huawei Pangu, ChinaShandong EnergyPangu Mine Model, ChinaPertamina New & Renewable Energy (Pertamina NRE)Drone plus AI inspection of a solar power plant in Riau for hot spots and panel anomalies (AI software not named), IndonesiaPT Akebono Brake Astra Indonesia (Astra Otoparts group)AI machine for final inspection of brake pads that flags likely cracks (shown at Hannover Messe 2023), IndonesiaPT Aneka Tambang Tbk (ANTAM), ferronickelConveyor belt tear and oversize detection; maintenance needs; reductant and dolomite composition monitoring (not stated), Indonesia
Sources
- arxiv.org/pdf/2510.17611 https://arxiv.org/pdf/2510.17611
- Ultralytics YOLO26 | Ultralytics Docs docs.ultralytics.com https://docs.ultralytics.com/models/yolo26/
- Cross-industry use cases for Huawei’s Pangu Models 5.5 showcased – Intelligent CIO APAC intelligentcio.com https://www.intelligentcio.com/apac/2025/06/24/cross-industry-use-cases-for-huaweis-pangu-models-5-5-showcased/
- Mining Technology mining-technology.com https://www.mining-technology.com/newsletters/huaweis-pangu-5-5-ai-intelligent-mining-in-action
- FPT offers AI to Indonesia’s largest oil company Pertamina vir.com.vn https://vir.com.vn/fpt-offers-ai-to-indonesias-largest-oil-company-pertamina-124294.html
- MIND ID leverages AI to boost safety, mining operations en.antaranews.com https://en.antaranews.com/news/430657/mind-id-leverages-ai-to-boost-safety-mining-operations
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Visual Quality Inspection and Defect Detection", https://aurion.semai.id/usecases/ind-visual-inspection/, records of October 2026.