Design and testing of an arc resistant power transformer with a reinforced turret

This paper presents the design and testing of a full-scale 510-kV, 415-MVA HVDC converter transformer prototype designed to withstand the resulting pressure of a 20 MJ low-impedance fault in oil occurring inside a bushing turret, which is one of the major causes of transformer fires. In addition, there is currently few studies and recommendation regarding this failure mode, which is the main motivation for initiating this study. In total, two full-scale units were designed and built: First consisting of a reinforced turret and a transformer tank designed for arc resistance using nonlinear static finite-element analysis with design pressures retrieved from explicit dynamic simulations, and a second reference unit, with a conventional turret design. Both units were then experimentally tested with a methodology based on injecting pressurized air into the water-filled units, with test equipment dimensioned and calibrated to achieve a linear injection of energy equivalent to a 23 MJ arc in oil in 100 ms. The reinforced turret and transformer tank was able to contain the arc energy and performed as predicted by the nonlinear static and explicit dynamic simulations. The energy levels used in tests are, to our knowledge, the highest reported for a low-impedance fault occurring inside a turret, demonstrating the feasibility of designing a transformer to safely withstand mentioned faults without rupturing. The suggested reinforced turret is a robust and maintenance free solution offering passive protection against internal arcing, without the need of additional relive devices. It will significantly reduce the risk of fires, improve overall safety of the substation, and has the potential to be optimized for different arc energy levels and voltage class for turrets, cable boxes and chimneys.


Introduction
The explosion of power transformers due to internal low impedance arcing is a concern for manufacturers and utility companies with a probability of transformer explosion failure leading to fire on the order of 0.1% per service year [1].Catastrophic consequences such as: oil leaks and toxic fires are a threat to personal safety, an environmental concern, and potential economic losses.A rupture resistant tank, that contains the arc energy by plastic deformation, has been proven to be a robust solution to mitigate the risk of fire if arcing occurs inside the transformer tank [2].However if high-energy arcing occurs in other oil-filled compartments such as bushing turrets, which is one of the major causes of transformer fires according to [1], one must account for the local peak pressures appearing at this fault location.These local pressure rises can, as will be shown in this paper by simulation and experimental data, be much higher and faster than for an equivalent arcing energy inside the tank [2], due to the smaller surrounding volume and stiffer geometry.At the same time there is currently limited guidance in [1] and [3] on how to design turrets for internal arcing.
This paper is a continuation of previous work on rupture resistant tanks, but unlike [2] instead focusing on internal arcing inside the turret and its failure mode.It presents the design and testing of a full-scale 510-kV, 415-MVA HVDC converter transformer prototype designed to withstand the resulting pressure of a 20 MJ lowimpedance fault in oil occurring inside the turret in 100 ms, as proposed by the IEEE guide for this transformer voltage class [3].The transformer handles the resulting peak pressures by relying on a reinforced turret design.In addition, an improved test setup compared to [2] was developed, which enabled experiments with higher power and shorter duration.
Experiments are important for increasing overall knowledge and confidence in current design calculation methods.To ensure that the latest knowledge was utilized when designing our test setup, a review of past investigations and experiments was conducted.
An investigation on circuit breakers concluded that arc energy dissipation can be broken down into; loss at contacts, radiated heat, oil heat to its boiling point, breakdown of oil into gas, heat of gas, expansion of gas and dissociation of the hydrogen formed [1].The authors also concluded that the average gas temperature is approximately 2,000 °K and the gas generation rate can be considered as a linear function of the energy at standard temperature and pressure.Its noteworthy to mention that more recent studies indicate gas temperature in the range of 1,500-2,100 °K [4].Other later arcing experiments [5] also includes more detailed measurements of the released arc energy, showing a small pulsating behavior in an overall linear trend.
The pressure difference between the gas and the surrounding liquid creates pressure waves in the tank until it stabilizes after a certain time.The authors of arcing tests on distribution transformers with high-speed photography concluded that the pressure against the tank was due to gas bubble expansion and the kinetic energy of oil movement [6,7].In addition, the impulse pressure immediately after the arc ignition produced a rupture of the distribution transformers, with a shape similar to that of the turrets studied in this paper.Another transformer arcing experiment showed a correlation between the gas bubble volume measured using a highspeed camera and pressure sensors on the tank [5].This means that the mechanical resistance of transformers depends on the gas bubble pressure rise and its duration; other aspects of the arc energy dissipation mentioned above are not relevant.
The energy generated by an arc in oil experiments is difficult to control, and there is a risk of damage to the laboratory equipment and overall security.In addition, producing a transformer rupture requires a high-power source, which limits the laboratory availability for energy levels upwards 20 MJ.The highest reported energy of an arc in oil experiment was performed in a on-load tap changer oil compartment with an arc energy of 10 MJ [8].Other techniques have been used to simulate the gas bubble pressure rise from an arc in oil; for example, a special powder combustion vessel has been used on a water filled 275 kV transformer up to an equivalent arc energy of 11.2 MJ [9].Another experimental approach is to suddenly discharge compressed gas in the transformer, it has been first used in 1959 [10].Subsequently, this experimental method showed good agreement with electrical arc in oil tests on pole-type distribution transformers [11].Finally, 210 MVA transformer tests with sudden discharge of compressed gas have produced, to our knowledge, the highest equivalent arc energy (up to about 30 MJ) and measured pressure rise (1,000 kPa) ever reported.
An important parameter which influences the pressure rise of transformers is the duration.For a while the effect of the arcing duration was misunderstood.Note that the power industry references and practices such as the proposed equation (2), which assumes isothermal expansion of the gas bubble generated in the oil, are based on a short-duration arc of 50 to 100 ms because of circuit breaker response time [1,12].However, this topic was investigated in 2020, and a numerical study [2] showed larger tank plastic work for shorter arc duration despite similar maximum pressures; thus, short-duration arcs are more severe for mechanical structures.Thus, an arcing test of 20 MJ in 50 ms is creating more structural damage than an arcing test of 20 MJ in 1 s.Therefore, the speed of gas bubble expansion is important when analyzing and comparing transformer mechanical resistance.
Over the years several authors focused on transformers tank strength under internal arcing by experiments [2, 5, 9, 10, 13-17] and numerical simulations [18][19][20][21].Despite the catastrophic consequences of an arc located in a turret is known, only a few studies are covering this topic.A first investigation of a turret failure suggests that retrofitting designs mitigate the risk of turret projections based on numerical simulations [22].Another numerical study analyzed the resistance of a tank and bushing turrets for an arc of 4,8 MJ located at the windings or at the tap changer.The results show the importance of the arc location where higher stress is observed on the tank in comparison with the bushing turrets [23].Moreover, a numerical model simulates an arc of 21 MJ in 59 ms at the entrance of a turret to reproduce a 415 MVA converter transformer accident.The results show a peak pressure in the turret in about 10 ms where the maximum stress appears at the joint between the turret and the tank [24].A bushing turret failure to a 570 MVA GSU transformer highlight the vulnerability of an arc located in this component [25].An arc in oil test on a turret prototype submitted to energy of 3,54 MJ in about 150 ms resulted in a peak pressure of about 1,75 MPa.A high pressure and temperature mixture of oil and gas was discharged through the turret pressure relief devices, but their reducing effect is not clear since no test comparison without devices are available.
The remainder of this paper presents the transformer design in section 2. and experimental testing of the mechanical resistance by sudden discharge of compressed gas to produce high energy and pressure in section 3. The paper is summarized and concluded in section 4.

Transformer design
The technical solution described in this paper relies on a reinforced turret design.An ideation process was followed, and the stated solution was selected from several ideas; including venting by pressure relief devices, rupture discs, and designing a turret with a greater possibility of expansion.The decision point during the selection phase was that the arcing pressure rise is too fast to be discharged, the considered system region is too small to absorb the energy via deformation, and the most efficient method is to reinforce the turret to contain the extreme pressure, which peaks within the first cycle, without rupturing.This passive protection solution relies on the turret handling this initial peak without the need for an additional relief device.This reinforced turret is designed to resist an arc energy level by assuming normal circuit breaker operation on the network to interrupt the power.Then, the transfer of pressure thru the lower turret flange into the transformer tank designed to absorb energy by plastic deformation.Ultimately resulting in a uniform residual pressure within the transformer which can later be vented by the tank pressure relief device.
The test objects included a turret, Resin Impregnated paper (RIP) bushing, mock-up active part and tank.The test facility limitations and manufacturing costs had to be considered when deciding on the size of the test tank; inner dimensions of 5 m length × 3 m width × 4 m height were used.Figure 1 compares the test setup with the HVDC converter transformer rated at 415 MVA, where it can be observed that while the dimensions of turret and tank height and width are equal, the length of the test tank is smaller.This will lead to a more conservative test setup because the residual pressure will be higher owing to the lower tank volume.
2.1.Turret design 2.1.1.Design requirement An arc within an oil-filled transformer non-tank component, such as a turret, generates a large amount of gas in a small volume, which creates a significant and rapid peak pressure increase.This event is different from an arc located in a tank; therefore, the proposed equation based on isothermal expansion of the gas bubble generated in oil cannot be directly applied [26].Consequently, the pressure needed to be withstood by the turret is determined by explicit dynamic simulations, which have been shown to be an appropriate method to simulate an arc in oil [22] and also provide good agreement with experimental testing of a transformer tank [2].These references provide a detailed description of the methodology applied here.The numerical model in figure 2 includes structural parts, a tank, and a turret.The last two are meshed by 2D shell elements and steel ISO 10025-2 grade S355J2 with nonlinear material properties are considered.The bushing and active part are meshed by 3D solid elements with linear steel material properties (Structural Steel in figure 2).The tank water is added to a Euler domain.Then, pressurized air was injected in 100 ms near the bottom of the bushing, which is the most likely arcing location.The amount of air was controlled to be equivalent to the volume created during an arc energy of 20 MJ in oil.The intention of the simulation was to replicate the loading conditions used in the experiment.Note that rupture is not considered for this explicit dynamic analysis, and it is assumed that the resulting pressure levels and distribution will be conservative for a design that ruptures.The rupture failure mode was assessed by the static nonlinear finite element analysis described in section 2.1.2.The numerical results in figure 3 show the expansion of the air bubble over time, displaying a significant fluid transfer between the turret and the tank.
The pressure was monitored by gauges initially included in the model where the average pressure near the turret cover reached a peak pressure of 7,942 kPa, as shown in figure 3.This peak pressure was used as the target design value for the reinforced turret to withstand.Note, the pressure peak is reached before the full expansion of the air bubble.In addition, a more uniform pressure distribution between the turret and tank is reached after approximately 200 ms.Finally, pressure propagation result of the explicit dynamic simulation is shown in figure 13 where it is compared with the experimental measurements.

Analysis methodology
The turret withstand capacity was assessed using static nonlinear finite element analysis in ANSYS ® Mechanical™, which has been used in the past for tank analysis [2,19 ].However, the design pressure to be withstood by the turret is 7,942 kPa, which is the result of the explicit dynamic simulation.
Additionally, this pressure load was applied only to the inner surfaces of the turret.The tank bottom was fixed to the ground and the model was split into two by a symmetry boundary condition.Most of the mesh, as shown in figure 4, is made of 3D eight-node solid shell elements with sizes ranging from 5 mm to 40 mm, for a total of over 80,000 elements.
The material properties include true stress-strain curves to consider the plastic deformation domain up to the corresponding ultimate stress.The turret and tank are made of steel ISO 10025-2 type S355J2 and weld E70C where the ultimate plastic strains are based on tension tests, as summarized in table 1.The fasteners are ISO 898 class 8.8 assume to have the same ultimate plastic strain as steel grade S355J2 because tension tests were not available.

Result of a 550 kV turret
The first turret model is taken from the 415 MVA HVDC converter transformer reference design and installed on a test tank.This conventional turret is further one denoted 550 kV turret, based on rated voltage of the bushing.The turret diameter is 1.1 m and with a reduction of 850 mm near the tank to the accommodate transportation clearance.The turret height was approximately 3 m.The M12 fasteners of the turret flanges were    modeled using 3D solid elements, and a preload force of 34.4 kN was applied to replicate the tightening torque in the first step of the simulation.In the subsequent steps, pressure was applied to the inner surfaces of the turret.The results in figure 5 shows the equivalent stress for a pressure load of 1.1 MPa.The result shows maximum stress on the fastener near the tank wall; therefore, this region was subjected to further analysis.A submodel was created, meshed by solid elements where the mesh size of the fastener is refined to 1 mm, as shown in figure 5.A strain-based rupture criterion was used to predict the failure when the equivalent plastic strain e P e exceeded the ultimate plastic strain u P e throughout the thickness of the part under consideration.This criterion, equation (1), has been validated by pressurized experiments on a tank wall [17] and recommended by a comparative numerical investigation with other criteria [18].The simulation results predicted a rupture when the turret arcing pressure reaches 1,025 kPa.

. Result of the reinforced turret
A model of a reinforced turret was designed with the aim of resisting the arcing pressure of 7,942 kPa, equivalent to an arc energy of 20 MJ in oil.As a constraint, the manufacturing process and overall dimensions, such as internal diameter and height, were kept the same.The reinforcements were achieved by a series of modifications to the 550 kV turret design, which comprised of larger fasteners and thicker turret shells and flanges.The final reinforced turret design was determined based on a sensitivity study of the main parameters, by performing multiple static nonlinear finite element analyses until the required bolt size was found to resist the arc pressure.
For simplification purposes, the model includes only the fasteners at the connection between the turret and tank, as observed from previous calculations to be the weakest point.These M36 fasteners were modeled by 3D solid elements with a preload force of 333 kN, a turret pressure load was applied to the subsequent steps.Figure 6 show results of displacement, stress, and strain under an internal pressure of 8 MPa.It is observed that the weakest point is located at the turret shell and a submodel analysis estimated a pressure of 8 MPa at the rupture.The reinforced turret is approximately eight times more resistant than to the existing 550 kV turret design.

Tank design
The ability of the reinforced turret design to contain extreme pressures relies on the arc energy being transferred into a transformer tank.The tank has a much larger volume and can absorb arc energy by plastic deformation.

Design requirement
The tank was designed with optimized flexibility and governing principles described in full detail in [2], dimensioned for a tank design pressure P d (kPa) according to equation (2), based on the conservative assumption of an isothermal expansion of the gas bubble generated by an arc energy of 20 MJ [26].where F is the dynamic amplification factor, k is the arc energy conversion factor (= 5.8 ×10 −4 m 3 /kJ), E is the arc energy to withstand (kJ), c is the tank expansion coefficient (m 3 /kPa), and P h is the hydrostatic pressure (kPa) at the middle height of the tank.

Analysis methodology
To determine the tank design pressure, a static nonlinear finite element analysis of the tank was performed, which included both material and geometrical nonlinearities as described in section 2.1.The tank and turret are made of steel ISO 10025-2 type S355J2 and all the welds of AWS E70C, with properties summarized in table 1.
The main step of the procedure consists of applying uniform pressure on all internal surfaces of the tank and turret, which was increased linearly in incremental steps while measuring the change in V , ∆ tank volume (m 3 ).The pressure was increased until P d was resolved using equation (2).Submodeling techniques were used to analyze the areas with the highest deformations to verify that the entire tank can withstand the design pressure.Furthermore, these locations on the tank were subjected to a pressure higher than P d until rupture occurred, to ensure a sufficiently high safety margin.

Result
The global model was analyzed, and the tank design pressure was established.The result can be seen in table 2.
A submodel analysis of the areas with the highest deformation confirmed a safe design at a design pressure equivalent to a fault of 20 MJ.The highest strain was obtained at tank cover weld, which is the first rupture point at an energy level >20 MJ with a good safety margin.The overall strain and stress levels at the tank walls and cover also displays safe values at design pressure as can be seen in figure 7.

Test plan and objectives
The objective of the tests is to replicate a 20 MJ arc in two different turret designs:  550 kV Turret Test: Testing of a conventional 550 kV turret subjected to an energy level of 20 MJ to produce a rupture as predicted by static nonlinear finite element analysis.
Reinforced Turret Test: The reinforced turret subjecting to an arc energy of 20 MJ will, according to the static nonlinear finite element analysis, handle the resulting pressures without ruptures, which will be proven by the experimental test.The success of the reinforced turret test was based on the absence of ruptures or leakages from the turret or from anywhere else in the transformer.The explicit dynamic simulation results were compared with pressure measurements.
The total test objects will include a turret, RIP bushing, mock-up active part, and a tank, which is required to capture the correct pressure propagation.Figure 8 compares the two test objects, and it can be observed that the main reinforcement consists of an increase in the size of all ISO 898 class 8.8 bolts from M12 to M36, and an increase in shell thickness from 6 to 8 mm.But as well an increase in flange thickness to handle the higher bolt  pretension and decrease the risk of leakage.One can also see the pressure sensors P1-P13 which are mounted in the same locations on both test objects.

Methodology
The experimental method used in our study reproduced the effect of an arc in oil by injecting the corresponding mechanical energy, which leads to a pressure rise.Only part of the electrical energy from a fault is transferred to mechanical energy owing to various dissipative mechanisms.Approximately 24% [1] of the electrical energy E elec (MJ) is converted to gas bubble enthalpy; mechanical energy E mech (MJ) which leads to a pressure rise within the transformer.
To reproduce an arc in oil with 20 MJ of electrical energy, the experimental setup needs to inject a gas equivalent to 4.86 MJ of mechanical energy.

Test setup
The arcing load was simulated by injecting highly pressurized air thru the turret into a water-filled test object.Figure 9 shows the experimental test setup, where the main components can be seen.The test equipment was dimensioned to achieve a linear injection of energy obtained by a pressure vessel with a volume V pv of 1.06 m 3 and a pressure rating of 210 bar.The pressure vessel was connected to the test object by an almost 10 m long pipe, which was designed with a reducing internal diameter along the flow direction.Nominal internal diameter was reduced in smooth steps from 189.2 mm down to 97.18 mm, resulting in the smallest cross-sectional area directly before injection location inside the turret.This method provides a choked flow with a constant velocity equal to the speed of sound in air.The internal pressure inside the pipe decreased along the injection trajectory, therefore pipes and components close to pressure vessel was dimensioned for a working pressure of 210 bar, while the pipe section after the mechanical valve had a pressure rating of 150 bar.
To ensure the highest health and safety standards the test was performed in a restricted area with remote control of all the equipment.The main test process consisted of pressurization of the vessel using a compressor, while having the gate valve closed.When the desired pressure had been reached, the gate valve was opened, and the air was then kept inside the vessel by two rupture discs.The pressure in the chamber between the two discs was measured and controlled and the injection was initiated by decreasing this pressure by venting air thru a solenoid valve, which created a pressure difference on the first rupture disc larger than the burst limit.The injection started and air flowed thru the 10 m long pipe into the test object.A pressure sensor in the pressure chamber registers the initiation and triggers a signal to stop the airflow after approximately 100 ms, which was performed by a mechanical valve.The later consisted of a 6-inch class 1500 swing check valve, which was modified with an electromagnet, allowing minimum flow restrictions in open position and a controllable and fast closing mechanism.Valve is closed by the air flow and closing time is estimated to be in the range of a couple of milliseconds.The test was monitored using seven pressure sensors on the tank and six were located on the turret.A high-speed camera documented the planned sequence of events during the tests.

Calibration of test equipment
The theoretical test parameters to reproduce 20 MJ of electrical energy in 100 ms were controlled by the main parameter ΔP, the pressure-drop in pressure vessel (Pa).Our experimental method assumes a test equipment containing an ideal gas undergoing an adiabatic process, with no heat exchange with the outside, and that all the energy leaving the pressure vessel enters the test object at the end of the injection system.Considering these assumptions, the change in internal energy can be express as ΔE (J) [27]: Where P 1 and P 2 are initial and final vessel pressures respectively (Pa); V 1 and V 2 are initial and final vessel volume respectively (m 3 ) and γ is the perfect gas constant for air (1.4).In our model, ΔE represents the mechanical energy leaving the pressure vessel, and is denoted ΔE mec (MJ).Additionally, both V 1 and V 2 can be substituted by V pv and equation (4) can then be written as: By rearranging (5) and substituting (P 2 -P 1 ) with ΔP one can express the corresponding pressure drop in the pressure vessel as a function of mechanical energy as: Finally, using (6) one can calculate the required pressure drop needed to inject 4.86 MJ as 1,834 kPa.The required pressure drop does not vary with P , 1 but the mass flow rate out of the vessel is a function of this initial pressure.Therefore, the required injection duration will be influenced by P .
1 The burst pressure of the rupture discs was 162 bars, and for this reason P 1 was maintained at approximately 165 bar during the entire calibration and both tests.
The required pressure drop according to (6) assumes, as mentioned, that all the mechanical energy leaving the pressure vessel enters the test object at the end of the injection system.This would be true if the pressure vessel could inject directly into the test object; however, in practice a 10 m long pipe was used with a nonnegligible volume (0.1 m 3 ) and friction; therefore, a small portion of the energy will not reach the test object.This effect was validated by explicit dynamic simulations of the test equipment where the initial pressure of the pressure vessel was released through the piping and closed after 100 ms.First, the simulation could not reproduce the pressure drop due to components of the piping system (valves) or friction in the piping.Thus, the piping diameter of the model is adjusted by trial and error until it fits the mechanical energy leaving the vessel measurement of a calibration test, the pressure drop recording is used with the calculation of (6).Second, from the numerical simulation results, the difference between the energy leaving the pressure vessel and that leaving the end of the piping is approximately 3%.Therefore, the targeted mechanical energy released from the vessel was increased to 5.01 MJ to achieve an injection of 4.86 MJ.The target test parameters are listed in table 3. The acceptance criteria for the calibration were specified as an acceptable accuracy of ±20% with respect to the of injection duration and arc energy injected.And with an acceptable reproducibility between trials of maximum 10%.

Calibration result
Over 30 calibration trials were performed to fine-tune the energy, duration, and repeatability.Incremental changes and improvements in the test equipment were required to obtain an acceptable behavior, and finally five consecutive trials were performed with injection duration and energy within allowable tolerances.The average results of these trials are summarized in table 3, where one can see that the calibration result predicted an energy of approximately 23 MJ to be injected in the test instead of initially targeted 20 MJ.Based on these five trials, calibration was concluded to be successful with an accuracy of injection duration of 4% and a precision of reproducibility within ±9%.The accuracy of the injected equivalent arc energy was 16% with a precision of reproducibility within ±9%.Both were within the accepted tolerance of ±20% and precision of reproducibility within ±10%.The final average injection capacity of the test equipment could be expressed as an average power of 225 MW during the injection duration.This is a fourfold increase in power compared to [2] (55 MW) which is due to the larger pressure vessel and internal pipe diameters, allowing for a higher mass flow rate.At the same time, a fast injection duration was achieved by using rupture discs for flow initiation and the modified swing check valve, which allowed for a controllable and fast closing mechanism.The force results in a substantial vertical displacement of the entire turret, as shown in figure 10, where initial and maximum vertical position are compared.Vertical movement has been limited by two steel supports, the injection pipe and additional lashings.None of these are used in service, and a real turret subjected to the same energy would be less restricted and cause additional damage to personal and nearby equipment.This is in line with observations from real arc failures, for example, the side turret described and analyzed in [22].
The high-speed recording was used to estimate the moment of rupture to occur within the first 10-20 ms of the arcing event.This is also supported by the pressure measured by the six sensors on the turret during the first 120 ms, as shown in figure 11.The average pressure at the turret top (P12 & P13) peaked at 10 ms and the level at this moment was well above the rupture pressure of 1,025 kPa predicted by the static nonlinear finite element analysis.
After 96 ms the data transfer from the pressure sensor located at the lower flange connection (P8) was stopped.Examination showed damage to the cable connection caused by evacuated water and turret displacement.Approximately 5 m 3 of fluid was evacuated during the explosion.Thus, such transformer rupture would result in catastrophic consequences, because warm oil sprayed with massive force and exposure to oxygen leads to a high potential for fires.

Reinforced turret
Test object was submitted to air injection equivalent to an arc energy in oil of 22.85 MJ in 101 ms.The turret successfully contains and transfers all injected energy to the tank, this last show walls and the transformer cover resist without leakages or ruptures anywhere in the system.Figure 12 shows the measured pressure from the six sensors on the turret during the first 120 ms.The maximum measured turret pressure occurs at the turret top cover (P13) after 10 ms and peaks at a value of 6,655 kPa, with a good safety margin to the rupture pressure estimated by the static nonlinear finite element analysis.This peak pressure, occurring after approximately 10 ms, which is later followed by decreasing pulsations, is in good agreement with real arc in oil experiments at lower energy levels [5].
Figure 13 compares the measurements at the top and middle of the turret with the explicit dynamic simulation during the first 60 ms.The first pressure peak appears in the middle of the turret at the arcing location, and one can later follow the pressure propagated with a peak pressure of 8,323 kPa appearing a couple of milliseconds later at the locations of sensors P12 & P13.The FEA peak is approximately 20% higher than the measurements, with a delay of about 4.5 ms.One reason for the slight delay was because injection system was not considered in the explicit dynamic simulation model.In general, the simulation results overestimate the measurement in the turret, which should be considered safe for the design.Finally, the peak of 6,655 kPa is more than six times higher than the peak pressure (1,000 kPa) measured during the test [2] with similar arc energy, where the injection location was inside the transformer tank.
Figure 14 shows the evolution of the average pressure measured by the six sensors in the turret and the seven sensors on the tank during the first 200 ms of the reinforced turret test.The average turret pressure reached a maximum of 2,563 kPa while the maximum average tank pressure measured in the test was 405 kPa.The tank pressure continued to rise for approximately 100 ms which was almost 10 times slower than the reinforced turret pressure rise (10 ms).Additionally, it is worth noting that the maximum average tank pressure was approximately 14 times lower than the average value measured near the reinforced turret cover (5,532 kPa).
Figure 15.shows the pressure evolution from seven sensors on the tank (P1-P7) as well as their average.This figure also shows the average tank pressure obtained from the explicit dynamic simulation, as well as the tank     design pressure calculated by (2).The maximum average tank pressure from the simulation was 369 kPa, whereas the maximum average tank pressure from the test was 405 kPa; a difference of approximately 10%.The tank design pressure of 475 kPa, calculated by nonlinear static analysis and (2), displayed a good safety margin and confirmed it as a safe method for designing tanks against internal arcing.As shown in figure 15 there are multiple peaks above the design pressure, with a maximum tank peak pressure of 867 kPa, observed on the back wall (P6) after 81 ms.However, these short duration peaks, lasting roughly 2-4 ms, are not a concern for tank integrity, which is proven by the experiment.This result is in good agreement with section 2.2.1in [1], indicating that the tank can withstand very high local peak pressures for shorter times, since it is the displacement of the tank walls causing the ruptures, not the pressure itself.According to the high-speed recording, the tank wall displacement reached its maximum after approximately 200 ms.This is well in line with the average tank pressure measurements in figure 15, which displayed a rather stable and uniform residual pressure of 275 kPa after 200 ms.The remaining overpressure was released through the Pressure Relief Device (PRD).Figure 16 shows the remaining deformation of the tank when the arcing and hydrostatic pressures were removed.The tank displacement was measured at multiple points along the tank height, with a maximum at the center of the tank wall with a permanent displacement of 87 mm, whereas the maximum displacement from the nonlinear static analysis was 109 mm.The high-speed video recording indicates a similar PRD activation time for both tests, and in the 550 kV turret test, it occurred after the rupture of the bolted connection.This is also in good agreement with a previous tank test [2] with a recorded PRD activation time of approximately 57 ms.The discharge of water was observed after about 300 ms, then the measurement show a slow relief of the pressure where after 5 s the average pressure in the tank was 225 kPa and after 10 s it was reduced to 180 kPa.Finally, it is safe to assume that the PRD had no impact on reducing the peak pressure inside the turret because it occurred already after 10 ms.

Conclusion
This study focusing on high-energy low-impedance faults occurring inside a transformer turret, which is one of the major causes of transformer fires [1].The governing design principle proposed is based on a reinforced turret that contains the extreme initial pressures and transfers the arc energy into a transformer tank designed to resist the load.The reinforced turret was dimensioned for a peak pressure obtained by an explicit dynamic analysis, and later verified by nonlinear static finite-element analysis and proven experimentally with, to our knowledge, the highest arc energy level ever performed in a transformer turret, equivalent to an electrical arc in oil of 23 MJ in 100 ms.Some of the main findings from simulations and experiments are: • Resulting peak pressure of 6,655 kPa for an arc inside the turret is roughly six times higher than an arcing location inside the tank.• Experimental tests of the original 550 kV turret and the reinforced turret, with the same arc energy levels, demonstrated the robustness of the new design.
• In general, the simulation results overestimate the measurement in the turret.While this demonstrates a safe and conservative design approach, it also indicates possibility for future improvements in the method.Finally, we concluded that suggested reinforced turret makes it is feasible to design a rupture resistant transformer that will significantly reduce the risk of fires and improve overall safety of the substation, while offering: • A robust and maintenance free solution offering passive protection without the need of additional relive devices.
• Potential to be optimized for different arc energy levels and voltage class for turrets, cable boxes and chimneys.
• Possibility for retrofit.Which assumes tank has already been designed to handle the arc energy and will require minor welding work at site to change mounting flange on transformer cover.

Figure 1 .
Figure 1.Experimental test model compared with HVDC converter transformer rated at 415 MVA.

Figure 2 .
Figure 2. Explicit dynamic model of the turret test.

Figure 3 .
Figure 3. (a) Air bubble evolution and (b) Pressure distribution in water over time.

Figure 4 .
Figure 4. Mesh of the tank with a turret model.

Figure 5 .
Figure 5. (a) Stress of the 550 kV turret,(b) submodel mesh and (c) fastener strain result in a rupture.

Figure 7 .
Figure 7. (a) Strain and (b) stress of the test object at tank design pressure.

Figure 8 .
Figure 8. Test setup including both test objects and part of gas injection equipment.

Figure 9 .
Figure 9. Schematic cross section view of test setup including test object and gas injection equipment.

3. 3 .
Test result and comparison with simulations 3.3.1.550 kV turret Test object was subjected to air injection equivalent to an arc energy in oil of 23.27 MJ during 105 ms.Turret rupture occurred at the lower flange connection as predicted by the static nonlinear finite element analysis.During failure, the internal pressure on the turret cover creates an upwards lifting force.The lower flange connection to the tank cover consists of 20 studs M12 (ISO 898 class 8.8) and figure 10 shows four of these studs ruptured.

Figure 10 .
Figure 10.(a) Ruptured M12 stud bolts from lower flange connection.Vertical position of the turret assembly; (b) Initial, (c) 30 ms after bolt ruptured and (d) at maximum displacement.

Figure 13 .
Figure 13.Reinforced turret test comparison between pressure measurements and simulation.

Figure 14 .
Figure 14.Reinforced turret test comparison between average turret and tank pressure measurements.

Figure 15 .
Figure 15.Reinforced turret test comparison between Tank pressure measurements and simulation.

Figure 16 .
Figure 16.Reinforced turret test tank deformation (a) before injection and (b) after, with arcing and hydrostatic pressure removed.

Table 2 .
Determination of tank design pressure.