Manufacturing
Automation and computer vision for production lines. Deep learning, visual inspection and machine learning at the edge for zero-defect quality.
Computer Vision
We develop computer vision systems forautomatic identification of manufacturing defectswith deep learning.
Convolutional Neural Networks (CNN) architectureAttention U-Netfor semantic segmentation of defects.
Real-time pipelinefor analyzing moving products with PLC integration for automatic stops.
Areas of application:
- •Defect Detection: CNN, pattern recognition
- •Segmentation: Attention U-Net, semantic
- •Real-Time: pipeline, analytics in motion
- •PLC Integration: arrests, manual inspections
Defect Detection
Automatic defect detection with CNN neural networks
Semantic Segmentation
Attention U-Net architecture for precise localization
Real-Time Pipeline
Real-time analysis on moving products
PLC Integration
PLC interfacing for stops and inspections
Machine Learning on the Edge
We implement machine learning models directly on Beckhoff PLCs viaTwinCAT Machine Learning Server.
Export models from Python to TensorFlow formatONNXfor real-time inference without cloud latency.
TwinCAT Visionfor image management, post-processing and defect contour detection.
Areas of application:
- •TensorFlow: training Python models
- •ONNX Export: standard size, portable
- •TF3820: ML Server PLC inference
- •TF7100: Vision, post-processing
TensorFlow Training
Train models with Python TensorFlow and ML frameworks
ONNX Export
Export models in ONNX format for edge deployment
TwinCAT ML Server
Real-time inference on PLC with TF3820
TwinCAT Vision
Image management and post-processing with TF7100
Roll inspection
Vision systems forcontinuous inspection of rolls, coatings and surfacesmoving at production speed.
Collectionimage dataset, defect labelingand development of pattern recognition algorithms.
Defect classificationby type and severity with integration of quality systems for traceability.
Areas of application:
- •Continuous Inspection: rolls, coatings, surfaces
- •Dataset Creation: labeling, annotation
- •Defect Classification: typology, severity
- •Quality Integration: traceability, KPIs
Continuous Inspection
Continuous roll inspection at production speed
Dataset & Labeling
Dataset collection and structured defect labeling
Defect Classification
Classification by type and severity
Quality KPI
Integration of quality and traceability systems
Automation of Production Lines
We develop softwareComplete automation for manufacturing production lines.
From PLC programming tomotion control, from HMI to integration with MES systems.
OEE and KPI calculationreal-time production for continuous optimization.
Areas of application:
- •PLC Multi-Vendor: Siemens, Rockwell, Beckhoff
- •Motion Control: robot, handling, assi
- •MES Integration: orders, tracking, batch
- •OEE & KPI: real-time, analytics
PLC Multi-Vendor
Siemens, Rockwell, Beckhoff programming
Motion Control
Robot control, handling and movement
MES Integration
Order, tracking and batch record integration
OEE Real-Time
Real-time OEE and production KPI calculation
Vision System for Defects Detection on Rolls
Development of a computer vision system for detecting defects on rolls of moving material. The project included dataset collection, CNN algorithm development, real-time pipeline for analysis, HMI interface for visualization and interfacing to the PLC for automatic stops.
Challenges Faced
- •Detection of defects on material moving at high speed
- •Accurate labeling of datasets with different types of defects
- •Real-time inference pipeline with no noticeable latency
- •Integration with PLC for manual stops and inspections
Applied Skills
Technologies Used
Results Obtained
- Automatic defect detection with accuracy >98%
- Significant reduction in waste and more consistent quality control
- Complete automation of a previously manual process
- Improved traceability and production KPIs
Visual Inspection with TwinCAT Machine Learning
Development of a computer vision system for identifying manufacturing defects on screws. Attention U-Net neural network training for semantic segmentation, export in ONNX format and deployment on TwinCAT Machine Learning Server for real-time inference.
Challenges Faced
- •Precise semantic segmentation of microscopic defects
- •Model creation and training with limited dataset
- •Export in ONNX format for TwinCAT compatibility
- •Post-processing of defects detected with TwinCAT Vision
Applied Skills
Technologies Used
Results Obtained
- Real-time inference without dependency on cloud or external servers
- Native integration with PLC environment without gateway
- Advanced post-processing with defect edge detection
- Inference time < 50ms for full resolution image
Contact us
A quote, a particular request, or simply a coffee to meet us? Write to us and we will reply to a "nano-second".