AI and image processing
Recognising patterns that others miss. That is what AI adds to machine vision and what Simac Masic has been doing for decades.
Image processing is the intelligence behind every inspection. It is the software that converts a raw camera image into a verdict: accepted or rejected, measurement A or measurement B, deviation of type X or type Y. At Simac Masic we develop that software entirely in-house and we have been doing so for more than thirty years.
We started with classical image processing algorithms. Today we combine that expertise with modern AI techniques, including deep learning and neural networks. Not because it is a trend, but because for certain inspection challenges it is simply the best approach. That combination, classical where it suffices, AI where it performs better, makes our systems reliable in the most demanding production environments.
Classic image processing
Classical image processing works on the basis of rules and models defined by engineers. The software knows exactly what it is looking for: a specific dimension, a colour threshold, an edge detection. This produces fast, transparent and repeatable results, indispensable in regulated environments where every decision must be traceable.
- Our classical algorithms have been deployed for:
- Edge detection and contour assessment
- Dimensional measurement with sub-pixel accuracy
- Colour and intensity analysis
- Pattern recognition and template matching
- Barcode and text recognition (OCR)
AI and deep learning
- For complex inspection challenges, where defects are variable, difficult to define or subtle in nature, AI offers demonstrable added value. Instead of programming rules, the model learns from examples. The result is an inspection algorithm that also recognises unexpected variants of a defect.
Simac Masic deploys AI when: - Defects are too variable for a classical rule-based approach
- Large quantities of training data are available
- Subjective assessment (such as cosmetic deviations) needs to be objectified
- Existing algorithms are reaching their limits
- Our deep learning models are trained on production data from clients.
- They are validated before going live and are designed to remain stable even when production parameters change.
Propietary software as foundation
We work with a proprietary software platform developed specifically for production environments. This offers advantages that standard vision software cannot provide:
- Offline working: adjust and test inspection programmes without stopping the production line
- Recipe management: quickly switch between product types, with all parameters centrally stored
- Audit trail: full traceability of inspection decisions, required in GMP environments
- Data export: structured output to SQL databases, MES systems and quality software
- Scalable: the same software runs on multiple lines or locations, with standardised inspection programmes
Validation and compliance
In the pharmaceutical and medical industry, every system must demonstrably do what it claims to do. Our software supports IQ/OQ/PQ validation processes and is configured for use in GMP-regulated environments. We think along on documentation, change control and software lifecycle management, so your validation does not run into delays.
A challenging inspection problem?
Is your current inspection hitting the limits of classical image processing? Or would you like to know whether AI could add value for your situation? Get in touch. We are happy to explore what is possible together.