GAS TURBINE COMPRESSOR FOULING AND WASHING IN POWER AND AEROSPACE PROPULSION

This paper presents a well-researched subject area within academia, with a high degree of application in industry. Compressor fouling effect is one of the commonest degradations associated with gas turbine operations. The aim of this review is to broadly communicate some of the current knowledge while identifying some gaps in understanding, in an effort to present some industry/operational interest for academic research. Likewise, highlight some studies from academia that present the current state of research, with their corresponding methods (experimental, numerical, actual operations and analytical methods). The merits and limitations of the individual method and their approaches are discussed, thereby providing industry practitioners with a view to appreciating academic research outputs. The review shows opportunities for improving compressor washing effectiveness through computational fluid dynamics. This is presented in the form of addressing the factors influencing compressor washing efficiency. Pertinent questions from academic research and operational experiences are posed, on the basis of this review.


INTRODUCTION
Compressor fouling is one the commonest forms of degradation for gas turbine engines during operation. For stationary gas turbines, the operating environment can vary significantly. This could be industrial environments such as a refinery that can be around the proximity of the coast and subject to changes in seasonal conditions. Stationary gas turbines have found use in a desert climate such as a: Wintershall power plant at Nakhla site in the Libyan desert [1] and the 800MW Sentinel plant, Desert Hot Springs, California [2].
The type of environment also includes off-shore platforms and cement factories that bring about particles that are particularly problematic for the gas turbines engines that power such facility.
For jet engines that typically operate at high altitude clean sky, the exposure to airborne contaminants is relatively less. Nevertheless, at the lower altitude, the particles (especially larger) concentration is higher as shown in Alpert et al. [3]. For this engine application, some factors leading to the susceptibility of compressor fouling includes: • Location of the airport (e.g. around desert or coastal) • Number of take-offs/landings (short or long-haul) • Atmospheric and seasonal changes • Flight path/route The location of the airport is important and especially for the fact that the engine will operate at the highest power setting on ground, ingesting the most amount of air compared to any given individual segment of its entire mission. For example, a jet engine with a take-off mass flow of 1300kg/s (~2.3 times more than cruise mass flow) with about 30 seconds of take-off roll ingests about 39,000kg of air. During take-off, it is also typical for aircraft to depart facing the prevailing wind to achieve shorter take-off roll benefitted from higher relative speed and higher lift. Landing performance is also improved by descending in the prevailing wind; however, these two phases increases the vulnerability of airborne particle contamination or compressor fouling. Unlike stationary gas turbines, and with exception of helicopter engines, jet engines have no inlet air filters to mitigate this effect. While compressor fouling can be mitigated by on-wing compressor washing, particle ingestion can lead to compressor blade erosion that is a non-recoverable damage. Figure 1and  GTP-17-1147, Igie,5 In the desert location, the risk is sand and larger particle sizes, as well as sandstorm events which were reported at Queen Alia Airport in Amman, in 2015. The Hong Kong International airport is located on the coast, and while there is no well-documented evidence of Sodium Sulphate (Na2SO4) acid formation on turbine blades (caused by sodium in sea salt and Sulphur in jet fuel), the occurrence is possible and known for gas turbines operating in coastal environment as indicated in Khanna [6].
Igie et al. [7] indicate the impact of compressor fouling for short and long-haul aircraft with integrated engine models for their respective typical missions. This study which investigates different levels of assumed deterioration shows that the penalty of compressor fouling is worse during take-off than other flight segments (climb, cruise and descent). The additional total fuel consumed due to the rise in fuel flow to achieve the required thrust is shown to be less significant when compared with the accompanying rise in Turbine Entry Temperature (TET) that will reduce the turbine blade life. Both aircraft engines in their respective mission demonstrated similar magnitude in penalty to fouling, with the exception of the worst case simulated, for which the short-haul aircraft became more penalised. This relates to the take-off segment that constitutes a more significant portion of the flight duration compared to the long-haul. Syverud et al. [8] demonstrate the impact of compressor fouling with a single-spool turbojet engine. The experimental test carried out for the stand-alone machine (without airframe) involved accelerated fouling using atomised saltwater. This study conducted at different corrected shaft speed shows that the front stage of the compressor is the most fouled, based on the measurements of deposits on the stator blades. The rotors were significantly less fouled based on visual inspection, as reported. In addition to this, it is stated that finer deposits were observed around the annulus, while coarse particles concentrated at the hub. The observation of predominant fouling at the front stages and less or none at the rear stages is consistent with observations in Tarabrin  temperatures are the highest in the compressor. The referred study also indicates that amount of wetness and viscosity is also a determinant for deposition.
On-wing compressor washing is conducted with the engine still on the airframe. This typically occurs by using a starter motor to crank the engine at low rotational shaft speed. The procedure involves the use of spray nozzles on lances that are installed at the inlet of the intake as shown in Figure 3 . The installation varies in the arrangement, depending on the design of the engine/aircraft intake and varies with the vendor. The washing process usually takes several minutes. This involves the use of detergents and then several rinses with demineralised water until the collected liquid effluent is clear. The frequency of washing is at the discretion of the airline operator. Factors that influences the decision to wash is usually based on the number of flight cycles, the TGT margin, type of environment flown, physical observation of the fan, as well as during other routine maintenance related checks. A Lockheed service publication [13] propose compressor water wash after the final flight of the day in a situation whereby salt water from the sea has being ingested from wave splash or wind pickup that can occur in airports around the seashore. This publication also advises washing every 15 days in such environments when the exposure is not direct.

Figure 3 Example of on-wing compressor washing
On-condition monitoring is beneficial in keeping up-to-date with engine performance and health in service, using acquired measurement sensor data and subsequently calculated parameters. Nevertheless, the full potential and level of insight gained from machine data are just ever more exploited in current times, with GTP-17-1147, Igie, 7 the availability of more powerful computers, relatively cheaper sensors and developments in cloud computing.

STATIONARY GAS TURBINES
Unlike jet engines, stationary gas turbines are continuously predisposed to contaminants in a given fixed location. Some of the worst cases of compressor fouling are documented in this application, notwithstanding the widespread use of inlet air filters. Figure 5 shows the fouled rotor blades of a heavy-duty industrial gas turbine engine compressor of 17 stages, with predominant fouling in the front stages.
To mitigate or eradicate the effects of fouling, compressor blade washing off-line or on-line is possible in this application. Off-line compressor washing is implemented when there is an opportunity for engine shutdown. This can often take place accompanying scheduled maintenance checks, before the return to service from an outage or shut-downs related to peak operating plants. Off-line compressor cleaning can be categorised into abrasive blasting, hand cleaning and liquid injection washing. The liquid injection cleaning with installed nozzles involves crank washing, in which the starter motor is used in rotating the blades between 10-20% of the maximum rotational speed.
Abrasive blasting can involve the ingestion of rice, nutshells, walnut shells or synthetic resin particles.
Gordon [17] states that this method was discouraged, as the new turbine blades in the late 70s/early 80s came with a series of fine holes for cooling. The cooling air to these holes and its passages emanate from the compressor, which meant solid cleaners couldn't be used to avoid blockage. In addition to this, Boyce and Gonzalez [18] indicates that some of the abrasives tend to shatter and get into the bearings, seals and lubrication system. Further to this, compressor blades have become more sophisticated having fewer, thinner, larger 3-dimensional shaped airfoils with smaller clearances [19]. This also implies that they would be more sensitive to fouling and erosion.
GTP-17-1147, Igie, 9 For hand cleaning, this is a manual activity applicable to major maintenance or upgrades. The highest cleaning effectiveness can be obtained, due to direct contact with all set of rotor and stator blades but timeconsuming. With regards to the liquid injection approach mentioned, hand cleaning can be conducted for the IGV and first stage blades with a fair amount of access. Nevertheless, this can be an additional time for washing with liquid injection approach. Relatively smaller engines of the aero-derivative sizes may take as long as 2 hours to cool down before an off-line wash is implemented [20]. Larger engines take longer and such considerations have to be accounted for when implementing off-line wash if there is ever a time constraint.
Allowing the cooling of the hot-section component before a wash is the practice, to avoid thermal stressing in the turbine blade. This can occur when cold liquid is in contact with high-temperature metal surfaces.

Figure 5 Fouled rotor blades of heavy-duty engine compressor
On-line compressor washing involves washing the compressor blade during normal operation, at fullload or part-load. This technique typically involves the injection of atomised washing droplets into the compressor to dislodge fouled deposits on the blades. The nozzles are usually placed around the periphery of the intake. The philosophy behind on-line washing is that it is a proactive strategy to control the build-up of particles on the blades while the engine is in operation. This makes the approach particularly suited for base load continuous operation. The greatest benefits are achieved when the washing process is initiated timely, following engine commissioning, an overhaul or off-line wash. Figure 6 from Ref  The pump pressure for wash liquid injection can range between 45 to 90 bar for the high-pressure systems and below 10 bar for the low-pressure systems. Figure 7 shows a wash delivery system that can vary in size, depending on the size of the tank, related to the amount of liquid required for a given engine mass flow, the number of engines to be served, the size of the pump and type of application (e.g. off-shore platforms and space concerns). Wash skids or delivery systems are normally connected to the injection nozzles through a network of pipes. The pressure of the liquid in the pipe is regulated by the selected pump pressure, which should be determined based on the mass flow of engine and effective droplet sizes desired. Attention to pressure losses needs to be accounted for when considering elbows and bends for less convenient installations and when the wash system is not near to the engine. There are typically 2 tanks on the wash skid; for the surfactant liquid and demineralised water. In most cases, these tanks usually receive supplies from intermediate bulk containers, however, any mixing of liquid to desired quantities or ratios occurs in the wash system. At least one of these two tanks consist of a heating coil to maintain the temperature of the mixture before injection. These actions can be automated or performed manually at intervals of operating hours.

Figure 7 Compressor washing system
GTP-17-1147, Igie, 11 Gordon [17] indicates that the next stage of compressor wash system development points towards systems that are linked to the gas turbine performance; by initiating washing when certain amount power is lost, rather than a scheduled wash interval. As for the amount of liquid utilised for on-line compressor washing, it is usually no more than 2% liquid-to-air ratio, with 0.2-0.4% more common. Roumeliotis and Mathioudakis [22] highlights the effects of water injection in an experimental study of a single-stage axial compressor at low rotational speed, thereby avoiding evaporation in the stage. The study shows no significant implications in most aerodynamic performance (pressure rise, stall margin and flow pattern) of the compressor. Further to this, however, increase in water-to-air ratio increased the power consumed by the compressor, accompanied by decreases in compressor efficiency. Mechanical losses and acceleration of water were indicated as possible causes of the dominant losses.
Some washing system OEM use the same nozzle for both off-line and on-line washing by varying the injection pressure using lower pump pressure for the former and higher for the latter. This is essentially more liquid with larger droplet and less liquid with smaller droplets respectively. Other OEMs provide separate nozzles for both washing types. Some of the factors that determine the effectiveness of compressor washing is discussed subsequently.

Compressor Cleaners
Commercially available washing liquid type applicable to on-line compressor washing includes demineralised water, solvent-based cleaners and aqueous-based cleaners. These are described as follows: Demineralised water -The mechanism of washing is based on the injection of demineralised water and impact of droplets. It is non-toxic and not effective to clean oily and carbonaceous deposits especially at low temperatures. However, for some foulants, demineralised water alone is sufficient for cleaning. It is important to add that demineralised water is also used for rinsing after the use of detergents or surfactants. GTP-17-1147, Igie, 12

Solvent-based cleaners
Aqueous-based cleaners -These solutions contain a mix of surfactant with water. Surfactants are surface-active agents that lower the surface tension of water and oils or solid dirt. They contain hydrophobic and hydrophilic groups that ensure its spreading and wetting properties. Surfactants are effective on greasy fouled blades leaving behind molecules on the surface the help mitigate redeposition. They have non-toxic and non-inflammable characteristics, and also effective on inorganic deposits.
Quick evaporation or low flash point is a collective limitation for cleaner types, given the operating temperatures in the compressor. This problem is common from operational experience and evident when the engine compressor is stripped open for overhaul or upgrade. Figure 8 shows the effect of liquid evaporation on redeposition of foulants on the rear stages, in a study conducted by Syverud and Bakken [24]. In this case, the water-to-air ratio of 0.42% with 75µm droplet size was implemented. Igie et al. [25] address the impact of fouling on different stages individually, at a time, for the same level of input degradation. The rationale behind this is to understand the performance implications of redeposition. The findings from the simulation study shown in Figure 9 indicate that though repositions on the latter/back stages is not favourable, the impact on the overall engine performance can be a lot less bad, the farther away the redeposition is from the first stage.
This is mainly because the front stages of the compressor typically have higher loadings and pressure ratios compared to the back stages. The farther away high levels of fouling occur on a stage away from the first stage, lesser critical compression phases are adversely affected. This leads to a more dominant drop in the compressor efficiency in relation to the overall pressure ratio due to an increase in compressor discharge temperature as shown. Nevertheless, the power output reduction remains the most dominant yardstick in relation to the penalty on the machine. To address the problem of evaporation, the use of high-temperature carrier agents can be considered due to their better flash point.

Injection Spray Droplet Size
The spray droplet size emanating from the injection nozzle varies depending on the orifice diameter, its design spray angle/type, injection pressure, flow rate and liquid property (viscosity, surface tension and specific gravity). In general, if the droplet size is too large, there is a high possibility that the droplets get deflected towards the compressor casing due to centrifugal effects of the rotor blades. In addition to this, large droplets bring about the risk of blade erosion. The smaller droplets with lower inertia tend to flow along the main airflow streamline as show in Rocchi [26]. This is demonstrated in Figure 10 for 50µm, 100µm and 300µm droplet sizes. Nevertheless, droplets too small possess lower kinetic energy to penetrate the blade boundary layer to the effective. This is also accompanied by the higher possibility of evaporation already mentioned.
Bromley and Meher-Homji [27] indicates a droplet size range between 50 to 250µm, as an industry agreement for an efficient on-line washing. Other key consideration is the nozzle spray coverage area and spray distance from the nozzle tip to impact surface (compressor blade). Some knowledge of the potential droplet spray penetration distance is required in producing initial droplets that finally arrives at the compressor stages in liquid form. This is also influenced by the proximity of the nozzle tip to the compressor stages and air flow velocity in the compressor.
GTP-17-1147, Igie, 14 The latter effect makes compressor washing challenging to execute with a high level of accuracy, as most understanding of nozzle droplet characteristic is obtained from static air conditions, except through CFD analysis as shown subsequently.   depression and compressor isentropic efficiency is shown to be an effect of fouling in this study. The paper shows the use of demineralised water with water-to-air ratios by mass flow ranging from 0.4% to 3%. The influence of liquid quantity, droplet size and duration on the effectiveness of washing is shown in this study.
This includes evidence that more water at 1.7% water-to-air ratio was more effective than at 0.4%, and for the later, washing for 4 minutes did not lead to any significant performance benefit compared to 1 minute.
Increasing the droplet size from 25µm to 75µm led to an increase in performance of the compressor but not the case when increased to 200µm.
Experimental studies using actual machines are the closest forms of replicating actual fouling conditions. Nevertheless, the degrading nature of the fouling problem, especially with particle ingestion has meant that majority of the studies are wind tunnel cascade analysis. Most experimental studies conducted have provided typically insights on the aerodynamic performance due to fouling and compressor washing.

Numerical Methods
Numerical analysis using CFD is useful in obtaining approximate solutions to real problems. It also allows for investigation of existing design or processes where experimental measurement is impossible. For compressor fouling and washing, CFD is very useful in predicting particle/droplet trajectories within a compressor as well as improving Unlike experimental and actual engine operation, there is currently no numerical study that predicts the removal of fouling particles on compressor blades due to washing. This is a difficult and complex phenomenon to stimulate with a number limit factors arising from the fouling problem, even before the washing is investigated, as subsequently discussed. Studies on compressor washing currently focus on droplet particle tracking and coverage areas as discussed in Refs [26] [31]. There are a number of existing ways to account for fouling in compressors using CFD tools. This currently includes: • manually assigning/imposing surface roughness partially or entirely on the blades (without particle   that was simulated using the Eulerian frame of reference, the method applied here is the Lagrangian approach that considers each particle as a point without finite dimensions. The outcome of the particle ingestion study shows that particles of larger sizes generally have higher collection efficiencies. The rotor is seen to have higher collection efficiency than the stator due to its position in front of the stator. However, with respect to the number of particles at the stator inlet, the stator is relatively more fouled than the rotor. It is important to state that these findings are based on a trap model and do not include the effect of particles detachment. In an earlier study by Bouris et al. [44] that focuses on rotor and stator blade deposition rates, the leading edge is shown to have the highest deposition rate. The stator pressure side was found to be more susceptible to larger and heavier particles due to inertial impaction mechanism. As such, the authors suggest that the stator pressure side will also be more vulnerable to erosion, as the larger particles deviate from the flow streamlines leading to more impact. The deposition rate model implemented in this work takes into account the particle and blade surface material, as well as the energy balance at the impact point. Suman et al. This study shows the build-up of particles for isolated rotor and stator cases, also indicating a dominance in particle deposition on the pressure side for both rotor and stator respectively. This is dominant for the larger particles (1.5µm) for which the hit impact efficiency is the highest. For the smallest particle size (0.15µm) investigated, there is a more even concentration of particles on both sides of the blade, for isolated rotor and stator as shown in Figure 16. In another Eulerian-Lagrangian model, the emphasis is on a multi-stage highpressure compressor particle tracking [39]. This study shows the role of particle size, shape factors and bleed position on the trajectories of these particles. The paper shows particles centrifuged towards the casing but a more distributed radial profile for 40µm and 60µm particle sizes compared to 20µm. The first bleed is shown to extract more particles than the second bleed, with the smallest particle size of 20µm as the most extracted for spherical particles. In addition, a greater amount of particle extraction is observed for non-spherical particles at the first bleed.

Figure 16 Isolated rotor and stator -pressure and suction side particle depositions [47]
In a related study, on turbine blade particle deposition, El-Batsh [49] highlights the difficulty in attempting to account for the change in the geometrical shape of the blade applying user defined subroutines. and implemented when the deposited particle volume is greater than the fluid cell volume. Nevertheless, it is reported that obtaining convergence proved difficult due to the non-uniform surface of the blade. This also caused a change in the y + value that did not necessarily satisfy the criteria for near-wall modelling. Further to this, a converged solution was achieved by using the known deposited mass of particle and locations to create an added thickness of the blade (as shown on the right). This ensured that near wall modelling requirement  showed sign of degradation after 6 weeks, that of the engine washed weekly showed this signs after 9 weeks.
For the third engine in this plant washed daily, the signs of degradation were observed after 12 weeks. This amounted to a power reduction of 3.6%, 3.4% and 2% respectively, at the end of 9 months. The multi-shaft engines located in a tropical environment generally showed slightly higher reductions. This is 4% reduction in power for the unwashed engine in the 6 th week, which was similar to the magnitude of power loss with another neighbouring engine washed with detergent for the same period. The most optimistic case was the third engine washed daily with demineralised water that indicated 1.5% reduction in 11 weeks. The authors suggest that a daily wash with detergents may provide better results. Boyce and Gonzalez [18] also shows the results of tests comparing solvents and varied water wash frequencies.
GTP-17-1147, Igie, 25 A credible approach to identifying and quantifying fouling effects during operation (apart from complete engine decommission) is by: • comparing neighbouring engines in the same location (engine with and without washing, for similar corrected power setting) • comparing washed and unwashed periods for a given engine, for similar corrected power setting conducted using GTPRO software with an implanted degradation of 5% reduction in mass flow and 2.5% reduction in compressor component efficiency. This study shows that when applied to a wider variety of engines, the susceptibility index is more indicative of sensitivity and recommends that the net work ratio between the turbine and compressor is a yardstick for the susceptibility and sensitivity.
Further considerations on the impact of fouling on engines include how the engine is controlled/constrained; approximately constant TET or shaft power. For the case of maintaining constant TET and rotational speeds for single and multi-shaft engines, Tarabrin et al. [9] shows that multi-shaft engines are more sensitive than single-shaft. The effect of stage location and fouling is also presented in the early work of Zaba [57]. The study shows that the percentage reduction in mass flow and compressor efficiency due to fouling is dependent on the location of fouling in the compressor. A fouling influence coefficient that relates to the percentage change in mass flow divided by the percentage change in compressor efficiency is proposed.
This study indicates that an influence coefficient greater than one relates to a heavily fouled front stage fouling.
A coefficient less than one is an indication of rear stage fouling, while one is a uniformly fouled stage. As the mass flow is not an operational data, the compressor overall pressure ratio and compressor efficiency can be an alternative as can be inferred from Figure 9. The bar chart would infer an influence coefficient less than one from the third stage fouling, using mass flow or pressure ratio with compressor efficiency. The absence of change in compressor discharge temperature (CDT) in this figure is also reported in Aker and Saravanmuttoo marginal increase in exhaust gas temperature experienced during fouling in the gas turbine engine does not yield added overall advantage to the power plant efficiency, mainly due to flow capacity reduction. The overall thermal efficiency of the plant is seen to be reduced as the mass flow reduction in the gas turbine also amounts to less steam generated in an unfired HRSG.
Analytical methods have brought about a better understanding of compressor fouling fault, using approaches that are often less computationally intensive compared to numerical methods. Combining different faults and conditions is more convenient, leading to further understanding of engine off-design behaviour when working with operational data and creating models for diagnostic tools.

AREAS FOR RESEARCH EXPLORATION
This review has taken a perspective of focusing on the main areas that have furthered the understanding of compressor fouling and compressor washing, applicable to energy and aerospace applications of gas turbines. This has covered experimental, numerical, actual engine operations and analytical methods, citing their respective merits and limitations, and areas for development in some cases.
Given that the design of compressor blades is unlikely to change fundamentally to suit a conceivable novel configuration leading to reduced fouling, and that some level of fouling will occur in many operational GTP-17-1147, Igie, 28 circumstances even with high-efficiency air filters, finding effective ways of mitigating fouling when particles get past air filters, is important. From the review, some of the identified areas that also appear to consist of relatively few research contributions are as follows: 1. Compressor Washing using Numerical Methods -the use of numerical methods will enhance the knowledge and the prospects of improving washing effectiveness with regards to coverage areas of sprays. Investigations into varied nozzle positions, droplet sizes and injection velocities can be studied with respect to known vulnerable areas of deposition based on particle ingestion and deposition studies. The challenge of such proposed studies is the availability of geometries/dimensions of actual compressors and intakes, to ensure that simulated conditions and injection nozzle locations are indicative or applicable to real operations. An example of a related study is Fouflias [31] that considers this, however, the washing droplet investigation is not in relation to fouling studies.
2. Use of Diagnostics Tools -though this review does not cover this aspect comprehensively, collectively, limited work has been conducted in using diagnostic tools for predictions based on actual machine operational data (or realistic situations). The use of operational data is pertinent, given the widely diverse off-design engine behaviour inherent in such operations that is not embodied in simulation model based data. The level of accuracy in fouling degradation prediction and practicality of machine learning methods needs further investigation, given the influences on fault signatures as a result of changes in dominant fouled stage location and changes in power setting. Apart from the benefits of fault identification and quantification, that can inform maintenance practice, significant cost savings on washing frequency can be made with an on-condition based approach as opposed to washing at given intervals of time.

ACKNOWLEDGEMENT
The author owes gratitude to Mr Paul Lambart and colleagues of R-MC Power Recovery Ltd for their continuing technical support and extensive research collaborations with Cranfield University.