Success Story – CYNEFY
Review of product labeling for automotive
Automated mass inspection of product labeling using anomaly detection based on image detection for the automotive sector
Keywords
Image & Anomaly Detection, Machine Learning, Supervised Learning, Product Compliance Management
Challenge
- In reference to the well-known CE marking, a UKCA marking will be mandatory for products sold in the UK market. By complying with these product certifications, the manufacturer and distributor confirm that the product meets the required health, safety, and environmental protection standards.
- Depending on the product class, the marking must be applied either on the product itself or its packaging.
- Manual inspection of the labeling on the product and/or packaging is time-consuming due to the need to gather relevant product and certification master data. Additionally, due to the lack of experience in the logistics environment of compliance management, it is prone to errors. This is particularly true when large quantities of components need to be inspected.
- Non-compliance with legal regulations results in products being unable to be sold and marketed.
Competencies and services
- Business Process Management & Business Analytics
- Agile project management
- RegTech
- Product Compliance Management Services
- Data Science and Information Systems Engineering
Solution
- Business Understanding: Process analysis to understand the validation logic for labeling requirements for products and packaging when applying UKCA and CE.
- Data Understanding, Data Pipeline, and Preparation: Reviewing the necessary data and relevant sources within and outside the customer organization for the use case, involving business and IT departments.
- Modeling a web-based image detection system to label images with appropriate tags, categories and metadata.
- Training the model with base data, testing, and retraining the proof of concept (PoC) to determine/increase detection accuracy and consistency.
- Deployment of the detection solution, including frontend and integration of the data pipeline.
- Monitoring and refinement: Continuously monitoring (sampling) and correcting the model as needed.
