Feasibility study for detecting metal springs in mesh boxes

The project

The metal springs are emptied into mesh boxes to be hardened in a furnace. If any springs remain stuck in the mesh boxes during emptying, they undergo the hardening process again. This leads to a loss of quality and to additional costs, the aim is to prevent this.

The aim of this feasibility study was to test whether the required inspection characteristics can be reliably recognized using industrial image processing. In the tests carried out, the metal springs were detected 100% of the time, regardless of their position in the mesh box.

Special requirements

Three different types of mesh boxes are used: close-meshed, medium-meshed, and large-meshed. The system must therefore be able to ensure reliable error detection against variable backgrounds.

Results of the feasibility study:

• The technical viability has been proven.
• The necessary test equipment was determined
• The knowledge gained led to the optimum test concept and system price.
• This created a robust basis for a reliable ROI analysis and investment decision.

Deep Learning

With the help of classic industrial image processing, it is virtually impossible to detect all metal springs. This is because, in certain positions, the springs are very difficult to distinguish from the background due to the grid. Deep learning is used to solve this problem.

Deep learning is a method of machine learning and, put simply, artificial intelligence. The system is trained using provided images and is ideal for detecting anomalies on complex surfaces.

No faults detected

This image shows an empty mesh box in which no metal springs were detected. This box is therefore classified as IO.

Error on the ground

Das System hat festsitzende Metallfedern am Boden der Gitterbox erkannt.

Error in the corners

The system detected metal springs stuck in the corners of the mesh box.

Error on the side wall

The system detected metal springs stuck to the side walls of the mesh box.

Detection of curved outer walls

In this case, deformation of the outer walls is not generally a reason for exclusion, but it does lead to problems in detecting metal springs located directly beneath a bulge. Deep learning enables these deformations to be recognized automatically and the affected pallet cages to be excluded.