Frequently Asked Questions
- What is the business model?
Our business model consists of a one-time fee for the setup and development of the optimal AI model for your AVI machine, meeting the inference requirements (e.g. 8, 10, 12ms) and assuring the highest performance in terms of Detection Rate increase and False Reject Rate reduction.
- Who owns the data?
Data ownership may be with you. Nevertheless, an agreement needs to be made that INSPECTIFAI is allowed to store and process the data.
- What amount of data is required to train an AI model?
More important than the amount of data is “which data”. In order to develop a good working AI model, we need to capture images of defective products, good products, and ideally qualified True and False Rejects. A couple of 100 products per category is sufficient.
- How is INSPECTIFAI integrated into the AVI engine and how does it intervene in the decision-making process?
An edge device, which fulfills all hardware requirements for AI model inference, is connected to the vision unit of the inspection machine (physical connection + SW interface). The AI model is deployed on the edge device. As soon as the vision unit retrieves the image, it is sent to the edge device, the inference is performed, and the inspection decision is fed back to the inspection machine.
- How long does it take to implement INSPECTIFAI EMBRAICE?
In general, we calculate 5-6 months to perform all tasks related to the AI model development and the final implementation at the inspection machine. This includes the data acquisition, labeling of data, and subsequent model development as well as implementation, functional testings, installation qualification, and operational qualification.
- What makes the solution GxP-compliant?
We are developing software according to GxP guidelines, have set up an internal Quality Management System, maintain our SOPs, and provide the necessary documentation with our solution, e.g. risk assessments, functional specifications, IQ & OQ protocols.
- Which cloud provider does INSPECTIFAI work with and why?
Our solution is based on AWS, as it provides a high number of cloud services that we use to build the best AI models, monitor model health, and even maintain edge devices fully remotely. Furthermore, AWS provides significant cyber and data security measures and fulfills relevant certifications necessary for the regulated industry.
- Is the cloud approach mandatory?
For building high-performing AI models, we need the advantages of scalable infrastructures. However, after the deployment of the AI model to the local edge device, the cloud connection can be deactivated.
- Who is labeling the data?
INSPECTIFAI is labeling the data as data labeling is a tedious task. We exchange with vision experts regularly in order to discuss products/images and to develop labeling guidelines which are implemented into our labeling software application, specifically built for labeling images for visual inspection.
- How do you calculate the progress of the Detection Rate?
Using the same mechanism as illustrated above but with clear reference to the defect kits, which are typically used for machine qualification.
- How do you calculate the reduction of the False Reject Rate?
A first indication can be given after the model training, based on the submitted data. We have further developed a specific operational mode, the benchmark mode, which allows comparability of traditional vision results vs. AI-powered visual inspection during commercial batch production.
- What downtimes should we expect for our AVI?
Installation of the edge device incl. AI model and IQ /OQ can be typically done within 1 day. After that, the benchmark can be already performed.