The transformer health index (HI) theory is one of the most debated approaches in the power industry. It is widely endorsed in CIGRE brochures with various methodologies, approaches, and recommendations.
This article is not a technical discussion, instead an addressal to various keypoints shared by big-corp experts, mainly from the European utilites.
Some of the objections
- The health index does not allow you to identify the type or location of failure
- The health index is based on too few / too many indicators (sic)
- The health index is a black box from which you don’t understand the weights and the reasons why some indicators are chosen or not
- A health index cannot be generalized to all fleets but only for some types or even only transformers of a single manufacturer
We neither support nor embed any position here. However, we would like to share some of our experience while developming an automatic tool that would merge HI & artificial intelligence (AI), RONIN.
Our point of view
A true is not SUPPOSED to identify the points of failure or the type of failure
The origin of the HI concept, which comes precisely from the medical sector, assumes to be used for preliminary assessments and rapid screening on large populations to identify trends and wide recognizable factors based on anonymized data. Therefore, the purpose is not to find in a single person WHERE the illness is located
A personal remark: if the HI approach is considered valid, with the due limits of application, in the human medical sector, we wonder why it cannot be adopted for electrical machinery, which each of them, could not at all be considered as “unicum”.
Too many or too few indicators
As matter of general fact always use the KISS approach (Keep It Super Simple): the right move is to begin by collecting the largest possible amount of data, then aiming to “synthetize” them in some overall indicator which rely on a solid math basement underneath. The objection we discuss here, certainly had a foundation before the development of current ML algorithms.
Now it is possible, with the multivariate analysis methods, to be sure that HI integrates in its first layer, a large number of indicators and then by optimizing and reducing them while at same time maintaining the accuracy of the final indicator.
In specific case of Ronin, for instance, from the initial model of 63 indicators, an accuracy value of over 78% was guaranteed by reducing them to less than 14. So in general, there is always time to simplify but starting by too few ones is not the case.
Black Box Risk
This is a real risk: some consulting firms especially, may develop their HI without specific component detail, like when eating a hamburger whose origin of the meat is unknown. First, there is a critical element regarding the past data knowledge “history” of this company: if it has often operated with electric companies, perhaps its HI is too unbalanced on their types of machines and, applied to the industrial sector, could misrepresent some results. To remedy this, we could suggest following one of the different models endorsed by CIGRE in its Brochure 630 and 760. The added value of a standardization and exchange system such as that of the CIGRE is precisely this important sharing of information and making the black boxes a little less dark.
An HI for each single fleet or single manufacturer
Here again I believe there is a basic misunderstanding: it would be like saying that we cannot use the same health risk indices for all human beings, but we need to create specific ones for males, females, old, young, rich, poor.
It is obvious that trends and behaviors change according to the categories, but nevertheless the macro factors are corroborated and validated at every latitude. The challenge of digitization will be to foresee and act quickly and contextualize phenomena that are currently not yet considered, such as atmospheric phenomena and the long-term influence of climate on assets.
Consequently, the evolution of HIs will necessarily has to integrate more variables and factors into big data and not “regionalize” or sectionalize the approach
In conclusion HI: handle it with care but please do handle and use it!
That is, instead of limiting their use why we do not stress them instead to the limit and we open a debate for their improvement. Scientific development arises from even harsh discussion and contrasts, we must preserve this characteristic and we do not have to be afraid to talk about.