EuroWire January 2021

Technical Article

Improving the circularity of crosslinked cable insulations with neural network models By Abhay Bulsari, Nonlinear Solutions Oy, Turku, Finland; and Janne Harjuhahto and Mikko Lahti, Maillefer Extrusion Oy, Vantaa, Finland

Summary Non-circularity of the cross-section of extruded crosslinked cable insulations has become a significant issue for producers in the recent past. Impulse curing in a VCV line, patented three years ago, turned out to be an effective means of improving the cross-section circularity of extruded crosslinked polyethylene insulations. The vertical line precludes one major cause of non-circularity, but other factors remain, including the presence of seams and internal stresses. Impulse curing quickly creates a hard, cured surface layer, which further reduces distortion by stress relaxation. The lengths of the heating zones and their temperatures can be adjusted to affect the circularity. The effects of these variables are not simple or linear. Too short a time is not effective; neither is too long an impulse. In addition, there are strong cross effects of some of these variables. To determine optimal conditions, it becomes necessary to have a good quantitative understanding of the effects of these variables as well as their cross effects. Nonlinear modelling is therefore the most suitable approach. Neural networks are powerful tools for nonlinear modelling. In this work, nonlinear models in the form of neural networks were developed from experiments carried out by Maillefer Extrusion in its testing facility.

These models now help us determine good or near-optimal conditions for impulse curing of various sizes of conductors and insulations.

Introduction The vertical continuous vulcanisation process, which precludes one major cause of non-circularity, is now widely used for insulating medium and high voltage cables. The circularity of the cross- section of the cables is still not perfect, and has attracted some concern in the recent past. Maillefer developed and patented an improved process that leads to better circularity [1] . Figure 1 shows a schematic diagram of the process, which is referred to as Maillefer Round Technology. The vertical line starts from the extruder crosshead. The insulation is extruded onto the conductor in the crosshead first. Right below it is a splice box separating the high temperature equipment below it from the ambient. The equipment below is essentially a long vertical barrel with heating and cooling facilities in different zones. After the splice box, there is a zone for impulse heating to a high temperature such as 500°C, over a short length, followed by cooling to a lower temperature such as 200°C in the second zone.

The first zone causes quick curing of the surface, which will freeze the outer profile closer to a perfect circle. The second zone brings an end to the impulse curing by bringing down the temperature sufficiently. This part is shown in Figure 2 . After this, a longer zone for vulcanisation – the third zone – at a moderately high temperature of around 250°C is meant to allow the insulation to cure throughout its thickness. There could be more zones, which are extensions of the third zone. After that, the cable is cooled, taken up and wound. Neural networks Neural networks have gained popularity in the last few years as a machine learning mechanism – a part of artificial intelligence. Nothing in nature is very linear, and plastics technology in particular is full of nonlinearities. Conventional methods of empirical modelling are linear statistical techniques, which are not very efficient at treating nonlinearities, even when nonlinear terms are used. Neural networks are methods of

▼ ▼ Figure 1 : A schematic diagram of impulse curing in a VCV line

Core

Extrusion head Splice box

First zone Second zone

Vulcanising tube

Other heating zones

Core with insulation

Cooling tube

Take up

Cooling tube

▲ ▲ Figure 2 : Maillefer Round Technology for curing

61

www.read-eurowire.com

January 2021