Artificial intelligence
Part recognition by OCR and Computer Vision – Industrie 4.0
Publiée le May 16, 2025
Publiée le May 16, 2025
In industrial and railway workshops, fast, reliable identification of spare parts is essential. Paper catalogs or manual databases slow technicians down: a search can take fifteen to twenty minutes, and identification errors cause delays and additional costs. The arrival ofAI-enhanced OCR (Optical Character Recognition ) and computer vision solutions is changing the game: they can recognize part codes, markings and even component geometry with great precision.
Traditional OCR, based on heuristic models, comes up against several obstacles: lighting variations, curved labels, reflective surfaces, damaged characters, non-standard fonts and irregular formats. NewOCR toolsusing artificial intelligence overcome these limitations by relying on neural networks trained on huge datasets. According to a study by Automation World, these new OCRs are capable of reading codes on curved or reflective surfaces, and interpreting text even when the characters are distorted. For example:
Cognex ViDi EL: a suite of deep learning tools that verifies dates and batches on pet food cans, delivering reliable results even under difficult conditions.
Matrox Design Assistant: software based on a pre-trained model capable of reading non-standard fonts on Ben & Jerry’s ice-cream lids, reducing the need for manual intervention.
These tools combine character recognition with full image analysis (edge detection, position tracking) to enhance robustness. They can be coupled with part libraries, to check consistency between the printed code and the part physically observed.
At the same time, startups have developed solutions that go beyond OCR and use computer vision to identify parts without visible codes. Notable initiatives include:
Partium – the result of the merger of Slyce (USA), Catchoom (Spain) and Humai (Austria). According to Design World, this company offers a part recognition system for industrial environments, capable of achieving a recognition rate of over 95% thanks to the combination of computer vision and a proprietary AI called Delta. The solution is available as a mobile application, software development kit (SDK) or warehouse kiosk. It reduces identification time, avoids errors and offers machine builders a new after-sales channel.
Synthavo – this startup offers an API for parts recognition. By taking a photo of a component, the AI identifies the model in a matter of seconds. The aim is to reduce search times and increase after-sales service revenues for equipment manufacturers.
Part-finder Kiosk – Partium has commercialized an in-store search kiosk that lets you place a part in a device and immediately obtain its ID and stock location. According to Design World, the kiosk uses a vision system to recognize parts and indicates exactly where to retrieve them.
PapAI (Datategy) – in addition to its data analysis functions, the PapAI platform offers part recognition modules incorporating machine learning, natural language processing and convolutional neural networks (CNN). An article by Datategy points out that AI enables accurate searches even when data is incomplete or confusing, thanks to features such as image recognition, contextual search and result prediction. AI improves user satisfaction and reduces misidentification.
Modern systems combine several building blocks:
Computer vision and CNN: models analyze the shape, texture and color of parts. They are able to differentiate between visually similar components and reduce errors due to ambiguous descriptions.
Natural Language Processing: to interpret textual queries (descriptions, incomplete serial numbers). NLP breaks down queries, corrects errors and searches for matches in the parts database.
Contextual search: algorithms combine image, text and context data (e.g. type of locomotive, assembly used) to refine results.
Predictive analytics: models forecast which components will be needed, based on consumption history and failure trends.
To implement such a solution, the company needs to prepare its data: collect and standardize its catalogs, standardize parts lists, and enrich part sheets with metadata (dimensions, photos, compatibilities). Model formation requires labeled data sets, including photos of parts from different angles and lighting. Models must be evaluated according to precision and recall metrics to guarantee their reliability.
Some railway companies have already implemented AI-based part search solutions:
Deutsche Bahn: the German rail operator has deployed a parts search engine, reducing identification time from 15-20 minutes to just a few seconds. More than 10,000 technicians use it, saving around 16,800 working days a year.
ÖBB (Austria): inspired by Deutsche Bahn’s experience, ÖBB has equipped around 800 agents with a visual search engine. Search times have been considerably reduced, and the company plans to extend the tool to other teams.
ConMoto Consulting: this consulting firm uses AI to optimize parts management. AI helps decide which parts to stock, by analyzing historical data, usage trends and consumption patterns. This improves accuracy and reduces errors.
These examples show that AI-based part recognition can save time, increase reliability and boost productivity. It fits into a broader continuum: predictive maintenance uses visual recognition to prepare interventions, while intelligent planning (article 3) will assign tasks taking available parts into account.
For a railway workshop, rapid parts recognition is strategic: when a locomotive is immobilized, every minute counts. An AI system coupled with a search kiosk or mobile application would enable technicians to photograph a part, obtain the part number and immediately see its availability. The system could also suggest alternative parts and automatically launch the order if stock is insufficient. Combined with predictive maintenance and load planning, this reduces downtime, optimizes inventory and improves customer satisfaction.