| KEYWORD |
Elaborazione di test non distruttivi basati sull'intelligenza artificiale per l'identificazione dei difetti nelle saldature di condotte per petrolio e gas
Tesi esterna in azienda
Parole chiave ANOMALY DETECTION, INTELLIGENZA ARTIFICIALE, SEGMENTAZIONE SEMANTICA
Riferimenti TATIANA TOMMASI
Riferimenti esterni Francesco Cannarile, ENI
Roberta Bianchi, ENI
Gruppi di ricerca DAUIN - GR-23 - VANDAL - Visual and Multimodal Applied Learning Lab
Tipo tesi RICERCA CON AZIENDA
Descrizione Non-Destructive Testing (NDT) is a testing and analysis technique used by industry to evaluate the properties of a material, component, structure or system for characteristic differences or welding defects and discontinuities without causing damage to the original part.
Contractors involved in the construction of oil & gas facilities (both upstream and downstream) make available NDT operators who are responsible for visually identifying any indication, i.e., the response or evidence of any anomaly/defects from a non-destructive examination on piping welds from the reading of x-ray plates/films NDT images. In this light, this thesis aims at developing a computer vision model for piping weld indications identification relying on artificial intelligence which automatically detects and diagnoses possible indications from x-ray plates/films NDT images.
This intelligent system should be able to perform the following tasks:
1. Quality image check: checking of required Image Quality Indicators (IQIs)
2. Detection: detection of the presence of any indication (anomaly/defect);
3. Characterization: diagnosing the type of the indication (if any);
4. Size assessment: assessing the size of the indication (if any);
5. Compliance check: verifying if the size of the indication is greater than some fixed
default thresholds.
Vedi anche thesis_proposal+paper_letteratura.pdf
Conoscenze richieste The successful candidate is expected to have Good knowledge of Python programming and main libraries for data science (numpy, pandas, scipy, etc), computer vision (OpenCV, scikit-image, etc.) and at least one deep learning framework (PyTorch is preferred to Tensorflow).
Scadenza validita proposta 23/05/2024
PROPONI LA TUA CANDIDATURA