Page 1 TECHNIQUES FOR ESTIMATION OF HOT SPOT TEMPERATURES IN TRANSFORMERS Vivian Ohis vivian.ohis@eng.monash.edu.au Tadeusz Czaszejko t.czaszejko@eng.monash.edu.au Department of Electrical and Computer Systems Engineering, Monash University, Clayton, Victoria 3800, Australia Centre for Electrical Power Engineering, Monash University, Clayton, Victoria 3800, Australia Abstract This paper contains a review of methods published by two standards organizations on the estimation of the hot spot temperature within a transformer, addressing some issues concerning the accuracy of their results. Included are summaries of articles on modelling techniques and the reasons offered by their authors regarding their effectiveness, focussing on their advantages and disadvantages. The paper also engages in some discussion on the current use of Artificial Intelligence (AI) systems on the dissolved gas analysis (DGA) results of transformer oils, and how similar AI systems are being adopted to estimate hotspot temperatures. Finally, some mention is made of the application of fibre temperature probes that have been recently developed. Such technology has been valuable to corroborate results and secure the validity of works done in this area. But there are limitations on their application, and these are discussed, highlighting the economic and safety implications. 1 INTRODUCTION Put into perspective, power transformers are probably the single most expensive asset within an electrical transmission and distribution network [1], and that alone justifies the requirement of developing appropriate reliable systems ensuring their availability and reliability. They are designed with a nominal life expectancy in years, evaluated for a rated load under ideal conditions as indicated by their nameplate. Excluding unusual circumstances unrelated to the normal functioning of a power transformer, failure is expected to occur as the windings’ paper insulation erodes past its limit to sustain appropriate structural and electrical stresses. The degradation is cumulative and adversely affected by the presence of both heat and oil contaminants, including gases and water [2]. Limited excessive loading beyond the transformer’s nameplate rating is permissible and often experienced during emergency conditions, but the trade-off is an accelerated demise of the transformer towards an endof-life condition. In Australia, the privatisation of the electricity utilities has met with increasing pressure by investors and shareholders to maximise Return-On-Investment (ROI). Utilities in the electricity industry must now reexamine operating strategies on long term, back up and failure procedures balancing the demands for both reliable and economic performance. Obviously, under utilised equipment, especially at the size of expense of power transformers, is undesirable. Ultimately the target will be to set transformers to run at “Full Warp Drive” as the normal mode of operation. However, the techniques used to predict a transformer’s loading limits are still too primitive to provide any guarantees. Any inadvertent abuse, where the transformer’s temperature thresholds are exceeded will compromise the performance of the transformer, a result caused by the application of stresses beyond the anticipated design limits, and possibly generating undetected incipient faults. Albeit the transformer may survive, avoiding premature failure, there will remain a high level of doubt concerning its state of reliability. An obvious resolution to this matter lies in the ability to accurately predict the safe limitations of power transformers and the associated loss-of-life trade-off. Such calculations will enable operational strategies that address economic and reliability issues without exceeding those thresholds that result in transformer damage or failure. 2 DYNAMIC LOADING “…A candle that burns twice as bright lasts half as long …”. So too transformers can run at over rated loadings within limits, consequently shortening their expected service life. A transformer’s loading capacity is related to the exposure of its insulation to heat, the highest temperature of which is referred to as the Hot Spot Temperature (HST). The HST’s effect on the windings’ paper insulation is used to quantify the limit of its temperature range over a calculated period of time. The problem, however, is that these limits are dynamic due to the changing transformer characteristics, and varying ambient climate conditions. Page 2 3 THE BENEFITS OF ACCURATE HOT SPOT ESTIMATION Development of an accurate method of estimating the dynamic loading limitations, and consumption of service life of a transformer, has economical and operational benefits. Economically, deferral of expenditure on expensive plant by “driving them harder”, has always been recognised as a wise business strategy. Conversely, but equally wise, being able to correlate the cost of depreciation on a transformer with usage in kilowatt hours (kWhs) allows for a more financially equitable method for assessing asset performance, thereby enabling the calculation of the most appropriate time for retiring older equipment. It also assists in planning for network growth and reducing expensive risks. Similarly, strategic maintenance and operational procedures are best formulated where the performance of existing plant has been accurately assessed. Unexpected failures are rarely welcome, especially with increasing pressure to minimise backup resources such as redundant transformer capacity. 4 CONDITION MONITORING Over recent years, condition monitoring of power transformers has become a far more affordable exercise; in actual fact, the true cost of not having adequate online monitoring is arguably much greater, especially if an avoidable incident should occur resulting in damage and injury. Several companies have already developed monitoring systems with simple algorithms that activate alarms and protection devices when monitored inputs have exceeded static thresholds. These thresholds are unquestionably conservative, perhaps through lack of confidence, which intuitively reduces any risk of damage or failure, but also under utilises the transformers’ true performance. 5 THE STANDARDS Several Standards have been developed to estimate the temperature of hotspots within a transformer as a function of current load and ambient climate conditions. Many electricity companies have used these as the basis of their operating procedures, developing tables and charts that highlight their exposure to risks within their network. As previously mentioned, there is often a policy among decision makers in this field to bias all errors on the side of caution. In this paper are presented two such Standards: IEC 354:1991 and IEEE Std. C57.91-1995. 5.1 International Electrotechnical Commission IEC 354:1991 “IEC: Loading guide for oil-immersed power transformers” The IEC Standard [3] provides a series of simplified equations that describe a mathematical model for the calculation of operating temperatures in a transformer. The assumptions listed in these standards include: a linear temperature rise in the oil from the bottom of the tank to the top, a parallel temperature rise in the windings, and an allowance for stray losses that is used to assess the HST. As this Standard points out, for large power transformers, the results for hotspot temperatures (based on temperature rise tests) may not be valid due to the significance and complexity of the contribution of flux leakage to the heating of the windings. Therefore, this method has a limited use, restricted at or below the transformers rated capacity. A further note in this Standard adds that corrections to account for load losses and oil viscosity, can be dismissed as either insignificant, or that the effects cancel each other. 5.2 The Institute of Electrical and Electronic Engineers IEEE Std C57.91-1995 “IEEE Guide for Loading Mineral-Oil-Immersed Transformers” In the IEEE [2] Standard there are two methods of calculating the HST. The first method, as with the IEC Standard, is a mathematical model. In the calculation there is the assumption that the oil temperature across the windings is equal to that of the top oil temperature. However, at the commencement of the Standard there is an admission on the validity of this assumption. After measurements were taken with recently available direct thermocouple and fibre optic devices, it was revealed to be incorrect. A paper written by Lesieutre et al. [4] also points out that the method does not adequately account for variations in the ambient temperature for which they [4] have suggested a modification and claimed to have verified. The second method in the IEEE Standard (Annex G) takes into account these observations and attempts to correct them, including thermal effects and liquid viscosity during overload conditions. 6 MODELLING While Mathematical Models, as those represented in IEEE [2] and IEC [3] Standards are based on the Page 3 simplified observation of the exponential cooling of lumped heated bodies, others such as Swift et al. [5] have suggested that a thermodynamic and heat transfer approach would be more appropriate. They explain that by incorporating the significance of the heat transfers between mediums such as the transformer oil and transformer tank-wall, a more practical and accurate estimation is probable. But even this model suggests the need for accurate parameter estimation and the assumption that the transformer is operating in an environment that supports nominal operation. 7 ARTIFICIAL INTELLIGENCE (AI) Owing to the complexity and variations in transformer design and operation, the expectation that a reliable and accurate “one size fits all” mathematical model is possible, is questionable. In any case, isolated experiments conducted using this method have met with the requirement for precision that is impractical, as discovered by Tylavski et al. [6]. Ultimately, such models are not capable of identifying and adjusting for the effects of aging, inaccurate inputs, and incipient faults. The identification of these difficulties has ignited particular interest in two categories of AI: Fuzzy Logic and Neural Networks. As a research tool, the Fuzzy Inference System (FIS) has been applied to a variety of similar projects where there has been a need to address the issue of inaccurate information and also apply heuristic reasoning [7]. Similarly, Artificial Neural Networks (ANN) has found implication where dynamic parameters are difficult to define mathematically [7] where training of the model can recognise trends and the dynamic nature of some of the parameters that might once had been believed to be constant. 8 FIBRE OPTIC TEMPERATURE PROBES Fibre Optic Temperature Probes are the obvious new technology, hailed as the easy, cost effective remedy for determining transformer HSTs. But seen objectively, the ultimate task of transformers is to provide reliable, efficient operation at the lowest possible cost. At the bare minimum, a transformer is little more than an iron core, wrapped with windings, bathed in a tank of oil. Anything else adds complexity and cost through design, maintenance and supervision. It is from this premise that the addition of measuring equipment may not always be the most advisable approach. Also, whilst for new transformers the task of embedding fibre optic cable into its windings is a seemingly easy and inexpensive task, it provides no reprieve for the vast majority of older transformers, many of which have operated undisturbed for the past few decades, and for economic reasons, may be wisely left that way. 9 DGA USING AI SYSTEMS A common method for identifying developing faults in power transformers is the Dissolved Gas Analysis (DGA) [8]. Analysis of ratios of specific dissolved gas concentrations, their generation rates, and the measure of total combustible gases are used as the attributes for classification. Thresholds are designed to partition the attributes into intervals. Specific combinations of these intervals are then used to identify the fault. However, in much the same way as determining hotspots, the results can differ, dependent on the thresholds used, coupled with the almost infinite variations and combinations of factors that influence the results. These are the classic characteristics that advocate the use of Fuzzy Sets (FS) and Artificial Neural Network (ANN) systems. 10 CONCLUSIONS A variety of models and techniques have been assessed over the years, with claims of improvement and higher dependability. Works done in this area by D.J. Tylavsky et al. [8] convincingly show that there is a balance between complexity and accuracy of results, and that in fact the major sources of error occurred from imprecision at the input. AI systems are customarily designed for these types of problems. Coupled with physical and mathematical models would assist in the development of a system that was both accurate and simple to implement. 11 REFERENCES [1] O. Roizman, V. Davydov, “Neuro-Fuzzy Computing for Large Power Transformers Monitoring and Diagnostics”, Fuzzy Information Processing Society, 1999. NAFIPS. 18th International Conference of the North American, 1999 Page(s): 248 -252 [2] IEEE Std. C57.91-1995 “IEEE Guide for Loading Mineral-Oil-Immersed Transformers” [3] IEC 354 1991-09 “Loading Guide for OilImmersed Power Transformers” Page 4 [4] B.C. Lesieutre, W.H. Hagman, J.L. Kirtley Jr., “An Improved Transformer Top-Oil Temperature Model for Use in An On-Line Monitoring And Diagnostic System”, IEEE Transactions on Power Delivery, Vol. 12, No. 1, January 1997 [5] G.Swift, T.S. Molinski, W. Lehn, “A Fundamental Approach to Transformer Thermal Modeling – Part I: Theory and Equivalent Circuit”, IEEE Transaction on Power Delivery, Vol. 16, No. 2, April 2001 [6] J.R. D.J. Tylavsky, Q. He, J. Si, G.A. McCulla, Hunt, “Transformer Top-Oil Temperature Modeling and Simulation”, IEEE Transactions on Industry Applications, Vol. 36, No. 5, Sept/Oct 2000 [7] M. Kezunovic, “Intelligent Systems in Protection Engineering”, Power System Technology, 2000. Proceedings. PowerCon 2000. International Conference on, Volume: 2, 2000 Page(s): 801 -806 vol.2 [8] Y.C. Huang, H.T. Yang, C.L. Huang, “Developing a New Transformer Fault Diagnosis System through Evolutionary Fuzzy Logic” IEEE Trans on Power Delivery, Vol. 12, No. 2, April 1997.