Power transformer management aims to obtain maximum profit whilst ensuring normal operation with acceptable risks [1]. Asset management involves several actions that can be taken during the life cycle of a transformer. The asset manager is responsible for deciding which ones of these actions are optimal. In fact, the manager has to answer questions such as: where and when an action must take place, which action is the best for each asset of the fleet, and what are the costs and the consequences of each action, etc. Actions that are decided upon must be provided with, for as long as possible, the largest amount of suitable information that aims to answer the above questions [2],[3].

In this context, risk assessment is a powerful and mandatory tool if an electric power company, for example, is seeking to implement a holistic asset-management system, including its power transformer fleet. This is because risk is a measure that integrates several pieces of information regarding condition, probability of failure, and the importance of a transformer in a power system. Therefore, an adequate calculation and interpretation of this measure is required to support the decision-making process.

It should be noted that evaluation of the power transformer’s condition is necessary to undertake risk assessment, but it is also necessary to schedule maintenance, especially when the company has adopted a scheme of reliability-based maintenance (RBM) and/or condition-based maintenance (CBM) [4].

In general, there are different test techniques that can be used for condition monitoring. There are also various methodologies to combine monitoring results in order to obtain a diagnosis. Thus, one of the main tasks for condition diagnosis consists of selecting the best-suited technique in terms of required input data, which is commonly acquired by power utilities.

Furthermore, as the scope of this paper is related to transformer risk assessment, the importance of each unit in a fleet must also be evaluated. In terms of this, there are a few reported methodologies in current literature. There are two different methods that are presently employed, one by an insurance company [5] and another by a U.S. power utility company [6]. This paper presents and analyzes these two methods. Additionally, analysis by risk matrices is introduced and the use of clustering techniques is proposed in order to support the decision-making process. The approach presented in this paper has been tested on a fleet of fourteen power transformers that are currently operational. Finally, the results obtained are discussed and some conclusions are presented.

A useful approach to manage a family of power transformers is to use the ranking method, which is based on risk. The International Organization for Standardization (ISO) defines risk as the likelihood of an adverse event and its consequences.

Therefore, in order to perform a ranking, it is necessary to estimate the Transformer Risk Index (TRI). TRI has been formally defined in several references, such as in [2,3,5,6]. While [2,3] lists the main criteria, i.e., pieces of information necessary to estimate TRI, [5,6] proposes practical approaches to combine these criteria. These approaches are based on weighted factor scoring models and will be described in subsequent sections.

TRI assessment is not an easy task since it does not only depend on unit data but also on the technical and economic characteristics of the power system where the transformer is installed. TRI is based on two components: 1) the Failure Probability Factor (FPF), which essentially depends on transformer unit condition (e.g., health condition), its calendar age, its effective age (estimated through operating history and operating environment), etc.; and 2) the Consequence Factor (CF), which relies on the fact that every transformer will reach a point of final failure, and the consequences of such a failure may be reasonably estimated.

Fig. 1 presents a methodological scheme for TRI assessment. It can be observed that input data presented in the first column in Fig. 1, correspond with those common criteria reported in Tables 1 and 4. These will be presented in the following sections.

Once CF and FPF have been estimated for each of the fleet’s transformer, there are two possibilities for the TRI analysis. One possibility is to multiply both variables, CF and FPF, the result being a nonlinear function. With the product obtained, i.e., the TRI value, a ranking of the units can be completed. The other possibility is to construct a risk matrix, as is shown in Fig. 2.

Figure 1 Source: Authors

Figure 2 Source: Authors

A risk matrix is useful for preliminary screening of the risk, especially when probabilities and consequences have been estimated in a qualitative manner. This is usually the case in management of power transformers. The idea of a risk matrix is simple: the probability dimension of a risk is envisioned as a continuum from 0 to 1. This continuum is broken into categories such as improbable, remote, occasional, and frequent. Likewise, examples of consequence categories are: negligible, marginal, critical, and catastrophic. The risk matrix has a fundamental subjectivity that suggests it should be used with caution and constructed with ratings supported by evidence, [7,8].

For instance, in the risk matrix shown in Fig. 2, sets of transformers belonging to a similar risk area have been identified, e.g., by using a clustering technique (in this paper, the well-known K-means technique will be applied to the case study). In fact, transformers in Set A represent the units with higher risk; therefore, actions to be considered about these transformers are quite similar, and priority must be given to this set.

The situations for the other risk sets are different. In particular, transformer units in Set B are either aged or have a deteriorated condition, but the impact in case of a final failure is lower than for the units in Set A. In this case, possible decisions could be: relocating, or refurbishing the units whenever possible.

Table 1

Source: Authors

On the other hand, units in set C are in good condition, but a failure may have expensive consequences. Therefore, the decision regarding this set may be, for example, to install a parallel unit or to acquire spare transformers that can be rapidly connected to the buses in the power system where the critical units are operating.

In order to evaluate the FPF sub-index, the main factors identified in references [2,3,5,6] are organized in Table 1 for the purpose of highlighting common criteria.

Although criteria for FPF calculation vary between references, some are similar or correlated. We can see that Table 1 identifies four common criteria (or four sets of aspects to be considered). These are: 1) The internal and external health index: this results from oil analysis and other diagnostic techniques that are useful to assess unit condition; 2) Manufacturer prestige: after-sales service, transformer technology, suitability for repair if a failure occurs, etc. are useful criteria to take into consideration; 3) Calendar age: this is an indirect indicator of the state of the unit and the risk of failure; and 4) Historical analysis of maintenance and findings: this must be performed by using the available data regarding the operation history of the unit, moreover this criteria could include the failure history.

In the following sub-sections, the two practical proposals given in references [5,6] will be presented. Then, a method to estimate the internal health index will be introduced.

CF relies on the fact that every transformer will reach a final failure point, and the consequences of such failure may be estimated. Post-failure consequences are different for each transformer, and they usually depend on the unit location in the power system as well as other characteristics. Table 4 organizes the main criteria that are considered in references [2,3,5,6] for CF calculation.

Table 4

Source: Authors

Table 5

Source: [5]

Two practical approaches to assess CF are introduced in the following subsections.

This work has focused on the formulation of a practical method to undertake risk assessment of a power transformer fleet. In order to do this, the main criteria to compute risk were identified and described. For FPF calculation these are: internal and external health index, manufacturer prestige, after sales services, calendar age, and historical maintenance and operation assessment. For CF calculation these are: company image, public safety, environmental impact and power system security.

Two ways to conduct the risk assessment were presented. These are: 1) the ranking method, which consists of ordering the units of the fleet in agreement with the TRI obtained (achieved by multiplying the CF and FPF) for each unit; and 2) by risk matrices.

This paper’s main contributions are: 1) the inclusion of an advanced technique for transformer health valuation into the practical methods for transformer risk estimation; 2) the introduction of the K-means clustering technique in order to analyze the risk matrices after CF and FPF calculation and normalization; and 3) a robust and practical methodology to perform transformer risk analysis.

We conducted a case study using a fourteen-transformer fleet. A risk assessment was undertaken for the fleet, and some of the main criteria identified for risk valuation were considered. Results show that just applying a ranking analysis could lead to different decisions being made, depending on the adopted method of analysis; this is because the weight given to the criteria varies between methodologies. We observed that for analysis performed over risk matrices and using the proposed clustering technique, both practical methods to estimate CF and FPF could lead to similar decision-making. This means that there is a loss of valuable information when the decision-making process is only based on TRI. This finding indicates that the use of risk matrices is advisable when a power utility wants to perform transformer management based on risk analysis. For instance, if a fleet’s transformers are grouped in clusters, specific actions such as relocation, unloading, refurbishing, maintaining, monitoring, etc., can be decided on depending on the risk level for each cluster. Thus, this clusterization may be a useful tool to improve the financial resource assignment of these electrical assets.

In addition, in the case study we analyzed, only four criteria of the above listed were considered due to the lack of real data. Therefore, we also recommend that the asset manager performs the risk assessment by considering all criteria listed in Fig. 1, as well as Tables 1 and 2. As such, we observed that more information (good information) means less uncertainty for the decision-making process.

Finally, we observed that practical methodologies for CF and FPF valuation are sensible, in terms of the weight given to each criterion by the experts. Therefore, future research may be conducted to develop expert systems, e.g., a fuzzy inference system, similar to the one employed to estimate the transformer health index, in order to compute both CF and FPF. Such expert system should consider findings reported in state-of-the-art, recommendations given by standards, appropriate expert-based professional judgment, utility experiences, among others, to suitably implement the membership functions, the fuzzy rule-bases, and the adequate defuzzification method.