Connected system for monitoring electrical power transformers using thermal imaging

1. Introduction

Electrical power transformers used in contemporary medium and low voltage distribution networks are critical components of the electrical infrastructure in developed nations. For this reason they are one of the main concerns of the industry and the sector’s regulators. These transformers are intended to be working without interruption and this involves to be able to handle peaks demand and prevent that potential temporary changes in the transformer workload could result in unacceptable voltage levels or cause any damage to the hardware system. In order to accomplish this requirement, power transformation systems are equipped with a variety of sensors and technologies that allow real-time monitoring of the system’s behavior, so that a prompt action can be taken in the event that the values of phase, intensity, energy, etc. provided by a power transformer deviate from those expected [1]. In addition, in recent years, some changes in society are threatening the resilience of the power system, which emphasizes the relevance of these monitoring technologies [2].

On the one hand, the introduction of disruptive developments such as electric cars, which demands large amounts of energy in a short time [3, 4, 5, 6], or the increased use of renewable production sources (of an intermittent character in the majority of cases [7]) are making the power transformation management more complicated and challenging [8], requiring more sophisticated monitoring systems.

On the other hand, electric power systems are currently experiencing a digitization process, which involves the integration of technologies associated with computing, wireless communication, Internet of Things (IoT) devices and a multitude of applications aimed at facilitating real-time and seamless operations [9, 10, 11]. These advances have enabled the development of more efficient semi-autonomous systems with reasonable operating costs, but have also brought additional concerns, including cyber security concerns, that should be addressed [12].

The management of electricity networks must pay special attention to potential overloads [13, 14], as they are one of the major problem sources affecting power transformers, whose failures represent approximately 12% of the total [15]. Transformer failures, in addition, are particularly dangerous as they can cause fires which, taking into account the urban location of transformers, can result in serious economic and personal damages [16].

The malfunctions affecting power transformers depend to a large extent on the transformer design, with important differences depending on the type of primary insulation [17]. In addition, the presence of certain components that are not present in all transformers may cause failures to be concentrated in some elements or others. Thus, for example, in transformers equipped with On-Load Tap Changer (OLTC), the percentage of failures associated with bushings and terminals is 14%, while in transformers not equipped with OLTC the percentage of failures of these components is more than double (33.3%) [18].

During the last years, several authors [19, 20, 21] as well as different international organizations have published a number of reports analyzing the probability of failure of the components of a power transformer. One of these organizations, the International Council on Large Electric Systems (CIGRE) has divided the types of failures into 6 categories based on the component causing the failure: winding, tap changer, core, tank, bushing, auxiliary. The probability of failure for each is shown in Fig. 1 [22].

OPEN IN VIEWER

OPEN IN VIEWER

The reason for failures depends to a large extent on the component in which they have occurred. Thus, for example, tank failures can occur due to corrosion, which in turn is linked to environmental conditions (high humidity, solar radiation, etc.) and can cause fissures in the tank walls, resulting in leaks. To avoid these failures, in addition to proper maintenance of the installations, a number of measurements should be carried out on a regular basis. This includes dissolved gas analysis [23], insulation moisture analysis, and controlling the operation temperatures [24]. The measurement of both the transformer temperature and the room temperature is therefore part of standard protocols for monitoring power transformers [25, 26, 27]. These are relevant variables that should be measured accurately and uninterruptedly and can also be used to forecast other variables such as reactive power or intensity [28].

In conventional approaches, temperature measurement is usually carried out by sensors that are integrated into the transformer equipment and are, to some extent, part of the transformer system. In this regards, a variety of sensors is available, and their characteristics and principles of operation are highly dependent on the type of transformer whose temperatures are monitored. In Liquid-Immersed power transformers, the temperature of the liquid (usually oil of different natures) is used as a estimation of the transformer temperature. In such cases, contact probes are used to measure the liquid temperature. The temperature of the transformer winding is also usually measured by contact sensors. This approach is possibly the simplest and most straightforward, however, it has the disadvantage that it can be affected by potential transformer problems, thus losing its ability to predict transformer malfunctions [29].

In this manuscript we describe a system based on thermal imaging to monitor the room and transformer temperature of electrical power transformers. Thermal imaging or thermography is a mature technology that allows us to capture the invisible heat emitted by objects whose absolute temperature is above zero [30]. It has been successfully applied for monitoring purposes in many fields, as civil structures [31, 32], machinery inspection [33] or in the nuclear industry [34]. In addition, this technology has been recently proposed to correct temperature problems in electrical substations [35].

The proposed approach has been implemented and evaluated in 14 transformation centers belonging to the Spanish power grid [36]. The measures obtained with it were validated against the standard sensor-based reference scheme. Compared to the latter, the proposed approach offers some advantages. The most important of these is the fact that the system is completely independent of the power transformer, and without physical contact with it, so it is not affected by potential transformer failures (high temperature peaks or too frequent temperature fluctuations may damage the sensors). It is also a more scalable system, capable of monitoring not only the temperature of the transformer, but also of different transformer components independently, or of adjacent equipment. This would be possible by only modifying the thermal image processing algorithm, without hardware changes. Finally, the proposed system based on thermal cameras can be used as a security system capable of detecting the presence of living beings (animals or people) near the transformer, based on the temperature changes detected.

The contributions of this work are both technical and scientific. On the technical level, this is the first time (to our knowledge) that electrical variables and temperatures obtained from sensors and temperature data extracted from thermal images have been analyzed together. This study, in fact, is part of a pilot project to evaluate the usefulness of this system in order to subsequently extend it to the whole ENEL network in Europe. On the scientific level, this work includes an extension of the work presented in [37], in which Granger causality is used to characterize the state of the transformer and its evolution. In this work, information exogenous to the transformer (variables extracted from thermal images) was included, demonstrating the relationship between this information and other electrical variables.

This work is part of the PASTORA project1 (ref. EXP – 00111351/ITC-20181102), which is being developed by ENEL and other international companies as an extension of the MONICA project.2 Both projects aim to develop an advanced monitoring system for the distribution networks and the electrical transformation system that allows operators to perform real-time monitoring and preventive control of these assets. These objectives include the use of artificial intelligence to enable self-management between the downstream elements of the network and the medium voltage substations, through intelligent electronic devices and automated control centers. Figure 2 shows a functional diagram of the PASTORA project, as well as the main information flows between the different subsystems and the MONICA platform. It includes the 5 parts in which the project objectives are divided and the companies of the consortium in responsible for implementing each of them.

OPEN IN VIEWER

The rest of the manuscript is organized as follows: Section 2 contains the details about the system implementation, including hardware description, communication protocol and image processing. Section 3 shows the comparison between the measures obtained by the proposed system and those obtained by the standard approach. Then, Section 4 discusses the advantages and limitations of each solution. Finally, Section 5 contains the conclusions and future work plans.

The proposed thermal imaging-based monitoring system was installed at 14 transformer stations of the Spanish power grid. They were configured to capture thermal data at a rate of one image per minute and data was stored and processed on a remote server as described in Sections 2.2 and 2.3. All equipment was configured in our laboratory prior to installation in the transformer centers. This configuration included the calibration of the thermal cameras, which was carried out by setting the factory calibration through the library provided by the manufacturer to configure and control the device. In our case, this factory calibration was adequate since we did not focus on absolute temperature values, but on temperature changes. Beside calibration, some environmental factors, as camera positioning, can affect the temperature measurements. This influence is small (less than 1% at distances between 0.5 and 2.5 meters) for the camera model used, however, in order to avoid differences between the operating temperatures assumed to be normal from one transformer substation to another, all the cameras were installed at a distance of approximately 2 meters from the surface of the transformer.

As mentioned in Section 2.4 the image processing carried out to extract two temperatures from each thermal image required a previous stage of segmentation of a reference image representing the temperature profile of each transformer. In order to build this image, 1440 thermal images taken over 24 consecutive hours were used, and three aggregation measures (mean, minimum and maximum) were considered. This way, the image is independent of the time of day and the whole method is able to deal with images captured at any time. Figure 6 compares the result of these three options. Note that all the three have a similar temperature profile and can be used indifferently for this purpose. For simplicity we selected the last one, where each pixel was calculated as the maximum number of pixels in the same position in all images.

Two segmentation approaches were evaluated. The first one, based on Otsu’s method for thresholds selection, carried out the segmentation into three step:

•

First, the image was downsampled using the SLIC superpixel method proposed by Achanta et al. [38]. The parameter k was set to 500.

•

Then, the resulted image was requantized using several thresholds (multiclass version) computed according to [39].

•

Finally, spatially separated regions of equal intensity are considered independently by applying the Maximally Stable Extremal Regions algorithm [40].

The intermediate results of these steps are shown in Fig. 8. Note how, after the last step (MSER application), four clearly differentiated regions appear. The two with the highest average intensity correspond to the transformer and were used to select the transformer pixels while the one with the lowest average intensity corresponds to the protective fence used to estimate the room temperature.

OPEN IN VIEWER

Figure 9.

Intensity profile of the reference image computed as the maximum.

The second segmentation approach proposed in this work modeled the reference image with a mixture of Gaussians and then identified each region with a Gaussian. After obtaining the Gaussian mixture parameters (see Eq. (2)) using a Expectation-Maximization algorithm, the MSER algorithm was also used to separate into regions areas of equal intensity located at distant positions. Given the shape of the intensity profile of the reference image (see Fig. 9), a model with 3 Gaussians was used to fit the image. This would ideally associate one component with pixels of higher intensity (located on the surface of the power transformer), another with pixels of lower intensity (representing the protection fence) and a third component with the rest of the pixels having an intermediate temperature. In practice, some pixels with a temperature similar to that of the protection fence were associated with the same component as the fence, as shown in Fig. 10. In order to isolate the pixels of the fence, we applied the MSER algorithm as in the alternative approach based on SLIC superpixels and Otsu’s methods. In our experiments, both approaches for image segmentation produced almost identical time series of temperatures.

OPEN IN VIEWER

Figure 10.

Result of segmenting the reference image computed as the maximum using a GMM model with 3 Gaussians.

OPEN IN VIEWER

Figure 11.

Comparison of transformer and room temperatures obtained by the proposed system and those obtained by sensor based approach.

OPEN IN VIEWER

Figure 12.

Transformer temperature obtained by the proposed system based on thermal cameras (red) and by sensors (blue). The former temperature has been shifted so that both have the same mean and in order to the correlation between them can be easily appreciated.

The evaluation of these systems was carried out by comparing the temperatures obtained by them with those obtained by the more standard approach, based on sensors. Figure 11 shows a comparison between the temperatures registered by the proposed system and those obtained by sensors. Note that, despite the offset probably due to differences between the calibration of the sensors and the thermal cameras, there is an almost perfect correlation between the temperatures recorded by the two systems. This correlation can be easily seen if we remove the offset, as is done in Fig. 12, where the temperature obtained from the thermal camera was shifted so that it has the same mean as the temperature obtained from sensors.

In order to further investigate the relationship between the temperatures obtained by the standard and the proposed monitoring systems, and even the relationship between temperatures and other operating variables, we performed an additional analysis based on G-causality. The results are shown in Figs 13 and 14. The former figure represents the F statistic for some variables of interest, including temperatures computed using the standard and proposed monitoring systems. It was computed using the expression (5). Figure 14 shows a circular graph with stronger connections among the same variables.

OPEN IN VIEWER

Figure 13.

F matrix based on conditional G-causality (see Eq. (5)) and containing a measure of the dependence of some electrical variables on others.

OPEN IN VIEWER

Figure 14.

Circular graph with the stronger relations between electrical variables.

4. Discussion

Temperature control plays a fundamental role in the ability of electrical power transformers to operate faultlessly [49]. Room and transformer temperatures are variables of great relevance in the analysis of transformer operation and has a clear relationship with other electrical variables. In [37], a methodology based on Granger causality is used to study the ability of some electrical variables to predict others. The results show that there is a clear relationship between the transformer temperature and other relevant variables such as current, active power, reactive power and voltage. The last experiment carried out in this work has shown that this relationship between the temperatures obtained by the proposed system and other electrical variables is also present. The relationship between transformer temperature and active power has also been corroborated in [50], which shows that it is possible to predict the transformer temperature in the near future from the active power of the three transformer lines together with the ambient temperature using long short-term memory (LSTM) networks. A similar approach was also demonstrated in [28]. In this case, a non-linear autoregressive neural network with exogenous inputs (NARX) was used to predict the temperature of the transformer as a function of past values of outputs and exogenous inputs. This scenario is also supported by the result of the analysis based on G-causality carried out in this work (figures 13 and 14). This analysis, similar to the one carried out in [37], but including the temperatures obtained by the thermal imaging-based system, meets two objectives: i) to confirm the dependence of some electrical variables on temperature and ii) to corroborate the relationship between the temperatures obtained by the proposed system and the sensor-based system.

The experiments carried out demonstrate that it is possible to obtain reliable information on the operating temperature of an electrical power transformer using a system based on thermal imaging. It could improve the temperature monitoring of electrical power transformers and, that way, reduce the probability of errors that affect the quality of the power supply and shorten the transformer lifetime.

After data acquisition, thermal images were analyzed in order to extract room and transformer temperatures. This analysis included the image segmentation by means for classical algorithms, however the use of these algorithms is not strictly necessarily. Given the fixed position of the thermal cameras and the limited number of locations where they have been installed, a manual segmentation would have been feasible. However, the automation of this task has the advantage of facilitating the extension of the proposed monitoring system to the entire power grid (manual segmentation can be tedious and time-consuming). In addition, it has all the advantages of eliminating the human factor, as subjectivity and error-proneness [51, 52]. In this work two proposals based on classical methods for automatic segmentation of the thermal images have been presented. Both provided a correct separation of the regions that allow us to calculate the transformer and room temperatures. Thus, there was no reason to opt for a manual segmentation.

Figures 11 and 12 show that the thermal camera-based system provides accurate temperatures, similar to the those obtained by the standard sensor-based system. A close examination of the figures shows, however, that there are some small differences, beyond the offset (possibly due to calibration differences). On the one hand, the data from the sensors have lower temporal resolution, which is closely related to sampling frequency of each system. Moreover, temperature peaks are slightly higher in sensor data probably due to sensor inertia. Larger differences at specific instants may be due to inhomogeneous heat distribution that causes heat to be detected earlier by one system than by another. On the other hand, differences in room temperature are more noticeable. These are due to the malfunctioning of the room temperature sensors, which were blocked for hours at a time. In any case, the proposed thermal imaging-based approach provides some interesting advantages:

•

It has been set up as an external system that is not coupled to the transformer, i.e. there is no physical contact between the devices of both systems and no data transfer through the electrical network, which guarantees that errors in the transformer will not affect the monitoring system.

•

Furthermore, the proposed approach is easily scalable to obtain a higher temperature granularity, i.e. to obtain not only the room and transformer temperatures but also those of other elements such as the connection terminals or the input and output lines. This would only require changing the image processing and eventually the camera orientation, but no hardware changes (additional cameras, etc.) would be necessary.

•

Finally, the proposed system based on thermal imaging can also serve as a security system since it allows monitoring the transformer environment. Thus, for example, it would be possible to detect temperature changes in devices close to the transformer but external to it, which could end up affecting the transformer’s operation. Furthermore, it would be possible to detect the presence of persons attempting to tamper with the transformer.

The comparison between the temperatures obtained by sensors and those obtained by the proposed system (Figs 11 and 12) also shows the differences between the sampling frequencies of each system, which is much higher in the case of the thermal imaging-based system. This difference may lead to lower reaction times of one system versus the other, however, it is important to note that it is not so much due to the technology itself, but rather, to the configuration of the system. The reaction time of the system to a failure resulting in overheating also depends on the element in which the failure occurs. Thus, an increase in temperature of an element that is not visible from the position of the thermal camera will not be detected by the camera-based system until the heat has been transferred to the air or other elements that can be captured by the camera. Similarly, a temperature sensor immersed in the oil will not be able to detect a temperature rise in the windings until that temperature rise has caused a temperature rise in the oil as well. For this reason, the safest and most reliable monitoring strategy would be one that includes sensors of different types, including thermal cameras.

Regardless of the system used for data acquisition, data must be permanently stored in a database that allows later analyses [24]. Ideally this permanent storage is carried out externally to the transformer, i.e. in locations other than the transformer whose operation is being monitored. A communication system is therefore required between the acquisition equipment and the servers external to the transformer. Possibly, the most direct application is to use the power grid itself to transmit the monitoring data, but this makes the monitoring system vulnerable to failures in the distribution network and its implementation entails substantial costs. In order for the monitoring system to be completely autonomous from the power grid, it is necessary that the data transmission is carried out independently from the grid. However, implementing an independent data network from scratch (whether wired or wireless), involves high costs. In the proposed approach, a commercial mobile phone network was used for this purpose. This allows significantly lower implementation costs by using already deployed infrastructure, while allowing data transmission to be carried out independently of the power grid. A similar approach was previously described in [53]. Unlike the approach proposed by Jalilian et al. the communication system proposed in this paper is unidirectional. The transmission is made only from the equipment installed next to the transformer to the external database. This facilitates the system to have a certain resilience since synchronous communication is not required and in case of connection error the data are temporarily stored in a buffer memory to be sent later in a new connection attempt.

The main limitations of this work can be divided into two aspects. On the one hand, the proposed system suffers from the limitations implicit in the collection equipment used, i.e., thermal cameras. Therefore, it is not possible to directly obtain other variables of interest, such as input or output line currents or voltages. Although, as mentioned above, it is theoretically possible to estimate these variables from the temperatures, this estimation would not be free of error and would possibly not be as accurate as the direct measurement by means of specific devices. In addition, thermal camera are substantially more expensive than the sensors used in the standard approach. However, despite the high value of the thermal cameras, their price is insignificant compared to that of the monitored equipment. Moreover, a malfunction of the equipment can result in incalculable damage (from hundreds of citizens deprived of a basic good such as electricity to a fire in an urban area with unpredictable consequences), which justifies any investment in the prevention of these events.

On the other hand, the use of commercial mobile networks for data transmission between the AE equipment and the SPS can be a concern when installing the monitoring system in power transformers placed in locations where there is little or no mobile network coverage. This can be stated at least for the climatic conditions of the locations in which our experiments were carried out, i.e., average temperatures between 9 (night) and 30 (day) degrees Celsius, as is typical in autumn on the southern coast of Spain. Although additional experiments in other regions with different climates would be desirable, the results shown can be, in our opinion, reliably extrapolated to other regions with similar or colder climates.

5. Conclusions

In this work we have described a system for monitoring electrical power transformers using thermal cameras. The proposed system allows the monitoring of transformer temperatures, which are critical variables to anticipate potential failures in the system operation. Using thermal cameras, we collected information on the temperature of the transformer and its surroundings at a rate of 1 image per minute. Subsequently, by appropriate thermal image processing, we obtained two temperature values from each image, representing the room and transformer temperature, and resulting in two temperature time series (after processing of a series of images).

This approach has been implemented and evaluated in 14 transformer stations belonging to the Spanish electricity system. The experiments carried out showed that the temperatures obtained by the proposed system are as good as those obtained by widely used sensor-based systems. In addition, the system based on thermal imaging provides some advantages such as greater reliability as it is a system totally independent of the transformation system and therefore not affected by the possible errors of the latter, as well as certain surveillance capabilities.

As future work, we plan to develop time series analysis methods to predict possible anomalies in transformer operation from the registered temperature series.