The system of classification and recognition of components/work pieces

About the project

We have developed a solution for TECHNODETAL, PJSC, a leader in the production of precision parts for aviation and the automotive industry. The system automates quality control, speeds up product identification and minimizes defects. Computer vision, machine learning, and integration with corporate systems have made the quality management process accurate and efficient.

Design — intuitiveness and functionality

The design team has paid special attention to creating an interface that simplifies interaction with the system provides operators with clear information.

- Prototype development
We started by analyzing the needs of Technodetal's production and IT departments through interviews and workshops. Based on the collected data, a prototype was created that defined the structure of the control panel and scenarios. interactions and requirements for mobile adaptation. This allowed us to lay the foundation for a convenient and an effective system.
- Creating a unique design
The control panel is designed in a minimalistic style with an emphasis on data readability. We used Clear visual elements such as graphs, heat maps, and notifications so that operators can quickly analyze the results. The color palette and fonts have been selected for long-term work without fatigue.
- Adaptability for all devices
Considering the need to work on mobile terminals, we have developed a fully adaptive interface. This It provides equally convenient access to the system from desktop computers, tablets and smartphones, which This is especially important for operators working directly on production lines.

Programming — from a concept to an industrial system

Our programmers have implemented a powerful solution combining advanced computer vision technologies., machine learning and seamless integration with enterprise systems.

- Project architecture
We have designed a hybrid "edge + cloud" architecture: local servers in production process real-time images, and the customer's centralized data center performs data analytics and storage. This ensures high processing speed and stability of the system. The architecture is designed taking into account scalability, allowing you to connect up to 50 production lines without modification.
- Development technologies
We used a modern technology stack to implement the project.:
- Computer vision: OpenCV and PyTorch for image processing and segmentation.
- Machine learning: Convolutional neural networks (ResNet-50, EfficientNet) for classifying parts and detection of defects. The models are trained on a sample of 30,000 images provided by the customer.
- Backend: FastAPI (Python) for high-performance recognition and integration services.
- Frontend: React with the Ant Design library to create an interactive dashboard with dynamic dashboards.
- Database: PostgreSQL with TimescaleDB extension for storing time series and events.
- Integration: REST API and connectors for SAP ERP and Technomes MES systems provide instant transfer data.
- Frontend: visualization and convenience
The control panel includes interactive dashboards with key metrics: classification accuracy, percentage marriage, shift statistics. Operators can manually check for "questionable" cases through a convenient the interface, and the reports are exported to PDF and Excel with one click.
- Backend: precision and reliability

Recognition algorithms provide classification of details with 99.3% accuracy and image processing time. less than 0.4 seconds. The defect detection module detects defects (cracks, scuff marks, geometry deviations) with using AI, and the reconciliation system automatically compares the data with ERP and MES. Data security It is guaranteed by encryption and role-based access.

Functionality — automation and quality control

The system includes several key modules, each of which solves specific production tasks:
- The part recognition module

- Classifies the part type according to the image.
- Identifies the article through barcodes, QR codes or visual identifiers.
- Marriage detection module

- Detects defects such as cracks, scuff marks or geometry deviations.
- Generates warnings in case of non-compliance with the reference parameters.
- Database reconciliation module

- Compares recognized parameters with data in SAP ERP and Technomes.
- Synchronizes information with a central database.
- Control and Reporting panel

- Displays metrics in real time: classification accuracy, rejection rate, shift productivity.
- Generates detailed reports with the ability to export.
- Integration adapters

- Provide seamless communication with SAP ERP and Technomes via the REST API.
- Transmit events and metadata for further analytics.

Implementation and training

- Deployment
We have installed 18 high-resolution cameras in three key production lines and deployed on-site servers in the customer's data center. The system was tested on 50,000 images per day, confirming stability and performance.

- Testing


- Functional testing covered all scenarios: recognition, reconciliation, reporting. - Load testing has confirmed the system's ability to handle intensive data flows. - Acceptance testing (UAT) with the OTC department has worked out complex cases, including atypical details. - Training and support

We conducted a two-day training for OTC operators and IT administrators, and provided detailed manuals. the user and the administrator. Within three weeks of the launch, "hot" support was provided, including during which the detection thresholds were optimized and minor defects were eliminated.

Result: Transformation of quality control

The project for PJSC Technodetal has become a real breakthrough in the automation of production processes. Final results:

- Accuracy: classifying parts with 99.3% accuracy, reducing defects by 28%
- Speed: The analysis time has been reduced to 0.4 seconds per part.
- Transparency: Reports and metrics are available in real time via dashboards.
- Integration: data is automatically synchronized with SAP ERP and MES, excluding manual input.
- Scalability: the system is ready to connect new lines without rebuilding.

We have created not just a software solution, but a strategic tool that minimizes human risk. This factor reduces costs and increases production efficiency. The system is ready for further expansion functionality, which guarantees long-term value for Technodetal.