Easy and low-budget automation of difficult visual inspection
~DX for manufacrurer that can introduce a high performance AI visual inspection system at low cost~
For those who have trouble considering the implementation of visual inspection system
- Incorporating in-line visual inspection is difficult due to budgetary constraints
- It is technically difficult to automatically identify defects and scratches on metal parts by visual inspection
- It is difficult to know how to proceed with the introduction of the system, and it is time-consuming
Less than 1% of manufacturing companies have adopted AI visual inspection
Mostly because of the high cost of introduction
※1 Based on our own research (Japan)
Generally, the cost of AI system introduction is 35~150 thousand US$, and the time required for introduction is 6 months to 1 year.
Leave it to ADFI for visual inspection!
ADFI solves the challenges of implementing AI visual inspection in terms of both cost and performance
※2 If you do not know how to check performance, please click here for a procedure manual.
⇒ ADFI makes it easy to introduce in-house visual inspection at a low budget!
Easy 3 step introduction at low cost
※2 If you do not know how to check performance, please click here for a procedure manual.
※3 For specifications of cameras and other equipment used for performance verification of ADFI, please refer to the introduction example.
Introduction Example
Examples of system configurations that can be realized at low cost
Examples of required parts and recommended specifications
| Item name/Model number | Recommended specification |
| ESP32-CAM / ESP32-cam-board | – |
| SD card | 32GB and more |
| （For ESP32）AC adapter AD-K50P300 | 5V,1A and more |
| LED bar light SOF-18W-30-SMD | 12V,0.1A and more |
| LED controller SCRGB1-3 | – |
| （For LED）AC adapter KSA-24W-120200HU | 12V,2A and more |
ADFI Inspection Performance
ADFI’s high performance has also been proven in a performance comparison with the world’s best anomaly detection method published in 2021 (Evaluation based on ROC-AUC※4 using MVTec dataset[*])
[*]: https://www.mvtec.com/company/research/datasets/mvtec-ad
CutPaste: Chun-Liang Li, et al.,” CX boutPaste: Self-Supervised Learning for Anomaly Detection and Localization”, CVPR, 2021
InTra: Jonathan Pirnay, et al., “Inpainting Transformer for Anomaly Detection”, ICIAP, 2021
DifferNet: Marco Rudolph, et al., “Same Same but DifferNet: Semi-Supervised Defect Detection With Normalizing Flows”, WACV, 2021
※4 ROC-AUC is the value of the area under the ROC curve (Area Under the Curve) when the ROC (Receiver Operating Characteristic Curve) curve
is created. The area value ranges from 0 to 1. The closer the value is to 1, the higher the discrimination performance.
⇒ Only ADFI can create a visual inspection system equipped with state-of-the-art AI algorithm at low cost and easily!
FAQ
If you have any questions or problems with ADFI settings, etc., we will support you by e-mail.
2. ADFI’s service level objective (SLO) is an uptime of 99.9% or higher. As an actual result, ADFI achieved 100% uptime during the first six months of its release (February to July 2022).
2. ADFI can detect objects that can be identified with the specified image size (800 x 800 pixels). If you want to detect objects smaller than that, please enlarge the image at pre-processing step in advance.
2. 100 times of sending/receiving a single inspection image (number of API usage) generated after the factory installation can be tried free of charge. If you are satisfied with the results of the trial, you can sign up for a paid plan and use the service without a limit of 100 times. The cost is based on the number of times images are sent and received (number of times API is used), so the running cost will be the same as the number of times used. For details, please refer to the Service Overview on our website.
2. Single board computers such as the Raspberry Pi can also be used.
1) Register a user
2) Create project
3) Creat dataset
4) Register images for training and create AI models
5) Register images for testing and set the decision threshold for the AI model
6) Increase the number of images for testing and validate the created AI model and decision thresholds to complete the AI model
※For details, please refer to Free trial instruction(PDF)
2. The following procedures can be used for on-site performance verification. Up to 100 free image inspection transmissions/receptions (API).
1) Copy the API code and implement it in your program
2) Execute the API programmatically
For more information, please refer to Service manual
2. Inspection images are immediately deleted as soon as they are determined, and no image data is stored on ADFI’s cloud. The security of image data is ensured by SSL encryption when sending and receiving images.
2. Invoices and receipts are issued by Stripe.For details, please refer to Stripe website.
Media Releases
