Welding is a manufacturing process that joins materials, usually metals or thermoplastics, using high heat to melt the parts and allow them to cool, resulting in fusion. Just how indispensable welding is is explained by the fact that almost all industrial and craft activities make use of this process.

Whilst this technology for series and large-scale applications has been and still is flourishing, for the welding of small components it is less developed in terms of automation and control techniques. If in the case of large welds that have a strong impact on safety, the controls are often defined by regulations, in the case of small welds that have little impact on safety but a heavy impact on product quality, quality control becomes complex. In fact, the traditional non-destructive inspections that are used for this type of process are not feasible in these cases, mainly due to the high costs associated with them. Typically, therefore, the control is carried out by visual inspection operators.  The objective of this project is to overcome this gap by developing a prototype bench for the fine-tuning of algorithms and the vision system that is capable of automatically assessing the quality of the welding process in the case of small dimensions (size less than 10 mm) by means of artificial intelligence algorithms applied to computer vision.

Intellimech, in the context of the TECH FAST Lombardia call for tenders, worked on the VIPS project -sistemi di Visione per Piccole Saldature, dealing with:

1. Development of the system hardware
2. Development of the vision algorithms
3. Development of the information architecture for the historicisation of weld images
4. Integration of the vision system for the automated inspection of welding defects.

The VIPS project set itself the goal of using the results of this research in an industrial context so that they could be trained efficiently, i.e. with the collection of a few training examples and with the help of a domain expert to guide the learning (human-in-the-loop) of modern neural networks for the localisation and classification of defects to be detected. The project led to the development of a prototype with non-specific hardware elements, i.e. colour camera, macro optics for welding magnification and illuminators, and new knowledge in terms of image analysis algorithms.

The project started in December 2021 and ends on 28 February 2023. The total budget of the project is € 215,507.78, 50% of which will be co-financed by the Regione Lombardia within the programme “INVESTIMENTI IN FAVORE DELLA CRESCITA E DELL’OCCUPAZIONE” through POR-FESR 2014-2020 funds.