INGENIEUR
from the collected sensing data . Various measurement data can be stored in a database hosted on the IT platform or in the Cloud . A variety of online and offline measurement data and methods can be used for further analysis , e . g ., current and voltage measurements , temperature , furan content , water content in oil measurement , partial discharge detection , acoustic and vibration monitoring measurements , polarisation-based measurements , frequency response-based winding deformation and displacement detection . An illustration on the possible digitalisation usage is shown in Figure 4 .
With the advent of digital disruption , CI methodologies continue to evolve in many forms and have demonstrated their importance in preventative maintenance by their ability to assess and monitor the operational status of transformers [ 5 ]. CI opens up a new avenue to address complex real-world problems using nature-inspired computational methodologies .
CI defines the theory , design , application and development of biologically and linguistically motivated computational paradigms . These advanced approaches offer a useful means for transformer fault detection and diagnosis , as compared with conventional techniques . Nonetheless , this does not mean the conventional methods are ineffective , however , in many aspects , the conventional methods incur higher operating and management costs [ 6 ], along with some other deficiencies . On the other hand , CIbased measures aid the conventional methods by providing more precise and accurate detection and diagnosis results , leading to convenient and effective solutions .
Computational Intelligence : The Way Forward
There are three main pillars of CI , i . e ., databased , knowledge-based , and search-based methods . Data-based methods such as Artificial Neural Network ( ANN ), Support Vector Machines ( SVM ), Wavelet Neural Network ( WNN ), and Deep Learning ( DL ) make use of the data collected from the power transformer monitoring processes , to learn from the data , to identify and analyse the patterns , to make effective decisions in the pursuit of less preventive maintenance and efficiency .
Knowledge-based methods reason and make decisions based on the heuristic knowledge derived from data ( or provided by humans ) to solve complex problems , such as the Expert System ( ES ), Fuzzy Logic ( FL ), and Adaptive Neural Fuzzy Inference System ( ANFIS ).
Search-based methods exhibit powerful search capability in solving complex optimisation problems that often involve multiple conflicting objectives , such as the Genetic Algorithm ( GA ), Particle Swarm Optimisation ( PSO ), Gene Expression Programming ( GEP ) and Artificial Immune System ( AIS ).
When choosing the appropriate CI algorithm for a real-world power transformer diagnosis , one must consider several factors : ( 1 ) the model , ( 2 ) the category of the model ( 3 ) the strength and limitation of the model , ( 4 ) the tools used , ( 5 ) whether similar research has been carried out before , if yes , when ? ( 6 ) the corresponding applications with respect to the model - the what ? ( 7 ) the motivation of the research implementation - the why ?
These factors have been considered and explained in “ Computational Intelligence for Preventive Maintenance of Power Transformers ”, Applied Soft Computing , 2021 ( S . Y Wong and others ) [ 7 ].
An analysis has been published depicting the distribution of CI-based methods pertaining to the condition assessment of transformers from 2010-2021 , as presented in Figure 5 . It shows that the Artificial Neural Network ( ANN ) is the most widely used model in the literature . This is followed by DL , FL , and PSO . Moving forward there are clearly many exciting areas for CI researchers to drive innovation in the context of preventive maintenance for transformers .
Safe Operation
One of the main factors affecting the safe operation of transformers is internal short-circuit faults . Some transformers experience winding deformation and even insulation breakdown when hit by short-circuit currents below the specified strength . The cumulative effect of winding
10 VOL 89 JANUARY-MARCH 2022