Hydrogen explosion incident mitigation in steam reforming units through enhanced inspection and forecasting corrosion tools implementation

Journal Pre-proof Hydrogen explosion incident mitigation in steam reforming units through enhanced inspection and forecasting corrosion tools implementation N. Defteraios, C. Kyranoudis, Z. Nivolianitou, O. Aneziris PII:

S0950-4230(19)30076-2

DOI:

https://doi.org/10.1016/j.jlp.2019.104016

Reference:

JLPP 104016

To appear in:

Journal of Loss Prevention in the Process Industries

Received Date: 24 January 2019 Revised Date:

14 November 2019

Accepted Date: 19 November 2019

Please cite this article as: Defteraios, N., Kyranoudis, C., Nivolianitou, Z., Aneziris, O., Hydrogen explosion incident mitigation in steam reforming units through enhanced inspection and forecasting corrosion tools implementation, Journal of Loss Prevention in the Process Industries (2019), doi: https:// doi.org/10.1016/j.jlp.2019.104016. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 Published by Elsevier Ltd.

Hydrogen explosion incident mitigation in steam reforming units through enhanced inspection and forecasting corrosion tools implementation N. Defteraiosa, C.Kyranoudisa, Z. Nivolianitoub, O. Anezirisb, a

Chemical Engineering School, NTUA, Heroon Polytechneiou 9, Zografos 157 80, Athens, Greece Institute of Nuclear & Radiological Sciences & Technology, Energy & Safety, National Center for Scientific Research ‘Demokritos’, Aghia Paraskevi 15310, Greece b

ABSTRACT Hydrogen (H2) explosion effects recently examined, are confirming the devastating loss scenarios to humans, environment, assets, and associated business interruption. H2 production is a core process in refineries used in further process steps. Steam reforming of natural gas or a mix with naphtha or LPG is a common hydrogen production technique, where the latest technologies have adopted enhanced metallurgies to minimize explosion risk and the associated maintenance cost following plant degradation owing to corrosion effects. However, corrosion rates are still high in specific areas of piping and process equipment. The aim of this paper is to present a methodology based on semi-quantitative RBI modeling according to regulations by API and recent EN standards, adopting a family of linear regression forecasting models that depict the yearly corrosion rate (per corrosion loop) of a hydrogen production steam reforming unit; this is done under different operating conditions (e.g., temperature, pressure, and fluid speed), metallurgy and other related physicochemical variables. The model is based on the examination of both ultrasonic wall thinning measurements and the examination of quantitative crosslinking total corrosion effects along with the physicochemical properties prevailing in different plant corrosion loops. The outcome of the regression analysis is an expansive family of multivariable equations describing, with a defined accuracy, the yearly corrosion rate and associated lifespan forecast per corrosion loop, and per examined part. These equations were further utilized in a custom-made database that can be used as an additional loss prevention tool by the hydrogen production unit management team. Evaluation results regarding the tool efficiency are presented in the following of this paper. Keywords: gas steam reforming units, risk based inspection, corrosion prediction, condition monitoring locations, corrosion loops, wall thickness measurements, multivariable regression, multivariable prediction functions

1

1. Introduction Hydrogen is an emerging alternative fuel with high energetic potential and eco-friendly behaviour; yet its physicochemical properties, namely: a) the wide flammability range, b) the extremely fast burning rate and c) the considerably high amount of energy released, when it burns or explodes, could render it dangerous, if not handled with care. Knowledge gained from incident analysis and investigations helps industries to form a better safety management system which ensures a safer and healthier working environment in their facilities. Analysis by (Mirza et al., 2011) of the latest hydrogen-related incidents, has shown that most of the root causes in incidents, where H2 is involved, were ‘technical’. Even in cases where equipment failure was identified as the primary cause of the accidents, the more detailed study of full reports revealed organizational deficiencies that can be categorized either as inadequacies (e.g. wrong design or inappropriate selection of materials), or as insufficiencies (maintenance, inspection or training) (Nivolianitou et al., 2006; Krishnasamy et al., 2005). However, the overall serviceability of the associated hydrogen production and handling units relies heavily on phenomena such as metallurgy corrosion caused by a series of physicochemical and metallurgy related variables, while inspection frequency, is a key factor for the implementation of an effective loss prevention tool. Risk-based inspection and maintenance (RBIM) approach help in designing an alternative strategy to minimize the risk resulting from breakdowns or failures. Adopting a risk-based maintenance strategy is vital in the developing of cost-effective maintenance policies. As part of the RBIM approach, forecasting is taken into account in the associated qualitative or quantitative risk assessments in most different risk-based approaches reported in the literature (Krishnasamy 2005). In general, loss forecasting consists of the assessing of the safety state of an installation based on the available information and observations. Few studies have been performed in the past on comprehensive comparisons of different forecasting methods, where quantitative methods are further classified into two principal groups namely, on the one hand: time-series forecasting methods, such as (Zheng and Liu, 2009) : a) the time-series method, b) Markov chain method, c) the grey model, and d) the neural networks, and on the other hand, the causality forecasting methods, e.g.: a) scenario analysis, b) the regression method, and, c) the Bayesian networks (Zheng and Liu, 2009). Quantitative forecasting of multivariable corrosion is used in this study, while the methodology that has been adopted and is further investigated, is the multivariable linear regression (Montgomery et al., 2012). This technique takes into account a series of parameters that are able to affect positively or negatively the corrosion rate of an equipment or piping section; the total effect of this per section parameter synthesis is the wall thickness reduction of the item owing also to other corrosion effects traditionally reducing the wall thickness of steel alloy parts.

2

Despite the fact that there have been numerous efforts to predict corrosion rates, the complexity of the corrosion phenomenon does not allow the assessment of anticipated partial rates, or separate calculation of Hydrogen corrosion alone with a predetermined accuracy. In this light, this methodology approaches reversely the phenomenon and analyses statistically the possible effect caused by each variable, namely the temperature, the loop pressure, the metallurgy, the size & shape of the pipe and the type & nature of the fluid, of the fluid together with their synergetic corrosion effect; the outcome of this approach is a family of multivariable equations that are used in predicting the total corrosion effect measured as wall-thinning of similar areas of interest, the socalled process loops. This tool is further combined with a qualitative risk assessment associated with the physicochemical conditions and the metallurgy present in the areas of interest with a similar profile. All tools related to prediction determination have the common objective to prescribe an efficient scheme for inspection and maintenance scheduling of the installation together with the cost optimization without compromising safety performance. In the remainder of this article are presented: in section 2 the hydrogen incident data & associated corrosion mechanisms, in section 3 the risk-based inspection and maintenance techniques for equipment testing, in sections 4 and 5 the methodology followed in data gathering and evaluation, in section 6 analysis of data, where section 7 highlights the results of the study, and, last, section 8 discusses its findings. 2. The hydrogen incident data & associated corrosion mechanisms 2.1. Incident data and cause analysis Various hazards and risks analyses have been developed in the past related to hydrogen, compared to other traditional fuel sources of gasoline and natural gas (methane). The results show that, for flammability hazards, hydrogen has an increased flammability range, lower ignition energy, and a higher deflagration index. The probability of a fire or explosion is based on the flammability range, the auto-ignition temperature and the minimum ignition energy where hydrogen has a larger flammability zone and lower minimum ignition energy, thus the probability of a fire or explosion is higher in comparison with other fuels. Furthermore, hydrogen has an increased consequence due to the large value of the deflagration index, while gasoline and natural gas (methane) have a higher heat of combustion. Thus, based on physical properties alone, hydrogen poses an increased risk, primarily due to the increased probability of ignition (Crowl et al., 2007). Furthermore, learning from previous incidents is an old and effective technique in the process industry, where hydrogen incident reporting databases have been developed over the past years to collect incident information in the hydrogen industry where root causes have been analyzed. The latest studies have shown that most of these incident root causes, were nearly 33% ‘technical’ ones owing mostly to unsuitable metallurgy selected and equipment parts specs adopted in hydrogen units production, storage and transportation units, where associated ‘inspection and maintenance’, take a significant share that is nearly 23%. These incidents involved mostly fire and explosion events (see following Fig. 1 and 2) (Mirza et al., 2011). Please insert Figure Fig. 1. Analyzed causes of Hydrogen incidents. (Mirza et al., 2011). Please insert Figure 2 Fig. 2. Effects of Hydrogen incidents. (Mirza et al., 2011). In both technical and maintenance incident root causes, damage mechanisms owing to problems or failures of the process equipment are key factors for the analysis (API-571, 2011).

3

2.2. Hydrogen production units corrosion mechanisms and equipment susceptibility Damage mechanisms can range from corrosion to cracking, to heat damage, and everything in between. The following main types of damages are encountered in petrochemical equipment are local and general metal loss due to corrosion and/or erosion, surface connected cracking, subsurface cracking, micro fissuring/microvoid formation, and metallurgical changes (API-571, 2011). The most common physicochemical mechanisms found in the petrochemical industry behind the different corrosion types are: • High-Temperature Hydrogen Attack (HTHA), a mechanism that can affect equipment that is exposed to hydrogen at elevated temperatures (at least 204°C). • Corrosion Under Insulation (CUI), which occurs when moisture builds upon the surface of insulated equipment. • Sulfidation Corrosion, a type of corrosion that occurs at temperatures above 260°C due to sulfur compounds in crude. • Hydrogen Embrittlement, which happens when atomic hydrogen infuses into certain higher strength steels and causes them to become brittle. • CO2 Corrosion, which is a form of degradation that occurs when dissolved CO2 in condensate forms carbonic acid, which corrodes steels. • Wet H2S Damage, which can occur when atomic hydrogen from wet H2S corrosion reactions enters and weakens the steel. • Brittle Fracture, which is a sudden, very rapid fracture under stress where the material exhibits little or no evidence of ductility or plastic degradation before the fracture occurs API-571 (2003). Single or a mix of the above mechanisms may be present in the hydrogen production units and depending on the process loop properties and specs of the associated equipment. Substantial advances in the field of corrosion science have led to a clearer definition of the mechanisms of many forms of corrosion. However, rather than placing the mechanisms into distinct categories, the overlap between many of the forms has become greater. In this study, the general metal loss in thickness is examined where it represents a quite signification part of the plant deterioration resulting in incidents and associated operational interruption. In addition, equipment’s susceptibility to a damage mechanism is affected by several variables, including materials of construction, process fluids, operating conditions, external environment, etc. There are also many factors within an equipment integrity program that can contribute to piping integrity problems, including design issues, operating window compliance, management of change issues, etc. (Gysbers, 2012). A good understanding of the variety of damage mechanisms that exist is a must for any mechanical integrity program. A thorough damage mechanisms’ review is essential for creating an effective inspection strategy where once damage mechanisms and morphology are understood, then the associated inspection scheduling is arranged, with the highest probability of detecting, characterizing, and measuring potential damage. 3. The RBI Approach 3.1. Overview In refineries, petrochemical, and other processing plants, the enormous amount of piping is more complex in distribution, than different types of static and rotating equipment. In general, compared with these different types of equipment in these industries, more difficulty in inspection planning is encountered. However, under-inspection or over-inspection can occur due to the lack of jurisdictional requirements on the inspection interval and method for piping, or the inspection interval being based only on piping service classifications in the existing regulations. This can result in unacceptable risks, along with a costly loss of resources (Chang, 2005).

4

Risk-Based Inspection and Maintenance RBIM is a risk assessment ‘tool’ that has started to be applied in the early 00s, accommodating the need of the process industry to improve their knowledge of equipment (machinery, piping etc.) health status, minimizing inspection and maintenance costs, while being on the safe side of operational procedures. RBI addresses an area of risk management not completely addressed in other organizational risk management efforts such as process hazards analyses (PHA), IOWs or reliability centered maintenance (RCM) (API-580, 2002) The unexpected failures, the downtime associated with such failures, the loss of production and, the higher maintenance costs are significant problems in any process plant. RBIM approach helps in designing an alternative strategy to minimize the risk resulting from breakdowns or failures. Adopting a risk-based maintenance strategy is essential in developing cost-effective maintenance policies. The RBM methodology is comprised of four modules: identification of the scope, risk assessment, risk evaluation, and maintenance planning. Using this methodology, one is able to estimate risk caused by the unexpected failure as an equation of the probability and the consequence of failure. Critical equipment can be identified based on the level of risk and a preselected acceptable level of risk. Maintenance of equipment is risk-based prioritized, which helps in reducing the overall risk of the plant. Several RBIM tools have been developed in the past, where the most commonly found ones in the refinery and global petrochemical industry are the (RP) 580, Risk-Based Inspection (RBI) frame was issued in 2002, which is accompanied by the (RP) 581, Risk-Based Inspection Technology in 2016, providing guidance on developing a risk-based inspection (RBI) program for fixed equipment and piping. In addition, there is recent European effort to develop a similar and more integrated tool adopting a larger span of different industry occupancies that have been recently issued under the EN16991:2018, Risk-Based Inspection Frame (RBIF) standard. RBIM implementation presumes the systematically prescheduled performance of ‘ad-hoc’ inspections of static equipment and other machinery, which nowadays are performed faster and provide more and more reliable results without any need for shutting down units, reflecting significant business interruption. This is in combination with the latest approach to proactive maintenance requiring downtime minimization, especially in bottlenecked units. 3.2. RBI risk assessment An RBI study may use a qualitative, semi-quantitative and/or quantitative approach. A fundamental difference between these approaches is the amount and detail of input, calculations, and output. API 580 requires that an RBI assessment must systematically evaluate both Probability of Failure (POF) and the associated Consequence of Failure (COF), while the calculation of risk involves the quantitative determination of a POF combined with an estimate of the COF (API-581, 2016) POF is defined as a loss of containment from the pressure boundary resulting in leakage to the atmosphere and in full quantitative modeling is expressed as an equation of time by modifying a generic failure frequency by a damage adjustment factor (DAF). The DAF is determination is based on the applicable damage mechanisms, which are subject to the metallurgy involved, prevailing physicochemical conditions, nature of the fluid being present, and the shape/type and size of equipment parts exposed to these conditions. DAFs help in optimizing inspection scheduling and developing prediction methodologies. This approach is based on modeling the degradation behavior of safety-critical components by taking into account different operating and environmental conditions such as load, stroke, velocity, acceleration, temperature, etc. The approach involves formulating theoretical (mathematical) models to describe equipment degradation and damage processes over time. To this aim, a good understanding of the physical operations of components and potential failure modes of the system (e.g., crack propagation, wear, and corrosion) is required. This technique provides an opportunity to link data to the physical condition of an asset, while this may not be possible through a data-driven approach. Mathematical models are often expressed utilizing differential or partial differential equations (Shafiee and Animah, 2017).

5

COF refers to property, liability and associated operational interruption loss scenarios owing to release of hazardous fluids or gases from the examined processing equipment, and it is expressed as an effected impact area or in financial terms (API-580, 2002; API-581, 2016; Webb, 2017; James et al., 2018; Chang et al., 2005). A common 5X5 risk matrix is suggested per CML for qualitative, quantitative, or an intermediate semi-quantitative approach, and including the necessary risk ranking that includes categorization on consequence and probability. 3.3. Quality of risk assessment. Identification of the credible damage mechanisms and failure modes for equipment included in a risk analysis is essential to the quality and effectiveness of the risk analysis. Process conditions, materials, methods, and details of equipment part fabrication are considered for this process where it is often possible to have two or more damage mechanisms at work on the same piece of equipment or piping component at the same time (API-580, 2002). The data quality and quantity have a direct relation to the relative accuracy of the RBI analysis and the decision on what methodology to be implemented (qualitative, semi-qualitative, and quantitative). Although the data requirements are quite different for the various types of analysis, the quality of input data is equally important, no matter what exact approach to RBI is selected. 3.4. Inspection Strategy According to API 570, the inspection frequency should be scheduled at the remaining half-life or established on the basis of fluid content, depending on whichever is shorter, as suggested in Table 1. The results of an RBI study are used for the development of an overall inspection strategy in the examined unit. The determination of an acceptable residual risk level is a prerequisite for this purpose as if this level is violated; then additional measures are undertaken including more frequent inspections and other engineering risk control measures. Risk level refers always to the examined corrosion loops as different corrosion rates, and the associated risk level is involved per loop. To identify the defects, one should use different inspection methods for piping defects with various corrosion types, different piping locations, and operating conditions. Additional past historical data, the age of equipment, recordings of latest and older TMIs results per CML, associated corrosion predictions, as well as past incidents or findings/near misses are taken into account in this planning process. Particularly, historical data of TMIs results are critical as they help in assessing and crosschecking the anticipated corrosion rates (CR) per loop and per CML. This study and associated risk assessment is preliminary a function of time, but furthermore, a function of knowledge and experience obtained about the condition or damage state of the equipment. The benefit of this is that potential areas of concern related to damage mechanisms are pointed out allowing adjusting inspection frequency and methodologies selection (UT, RT, etc.). The associated rating of the inspections’ effectiveness helps in this process. Inspection effectiveness means the possibility and veracity of the inspection method, which relates to the inspector’s ability and fitness for the selected inspection method. The highest inspection effectiveness is always defined as 1.0 in a typical equation where inspection interval can be empirically expressed as per following Eq. (1) (Chang et al., 2005). NID = n C RL

(1)

where, NID = next inspection date (year), n = inspection effectiveness ratio (value range 0–1), C = confidence rating (value range 0–1, based on the risk ranking), RL = remaining life (year) Finally, the methodology of inspections and associated cost involved per different type and extent of implementation are critical factors for this study. 3.5. RBI incorporated in CMLs and TMIs frequency optimization 6

In general, the outcome of the RBI assessment has a direct impact on the frequency and type of inspections and associated maintenance of particular parts of equipment and piping belonging to pre-defined corrosion loops. The generic approach is that RBI focuses on the risk reduction maximization in highest risk components where damage mechanisms are able to cause increased corrosion rates, and as a result, the unexpected release of high-pressure flammable liquids and gases is more possible. RBI provides simultaneously a reduction of TMIs in comparison with the ordinary time-based inspection plan, where risk is much lower. In this way, RBI optimizes the level of TMIs while keeping overall unit risk at an acceptable level for all equipment and piping parts that examine. In regards to CMLs, as these create the frame where TMIs are performed, and the whole RBI assessment is based upon, it is quite challenging for the industry of hydrogen production where corrosion is a key reason of losses, these CMLs to be optimized and inspection frequencies to be balanced as cost involved in inspection and maintenance is quite high. As residual risk level is always uncompromised, the key in this continuous exercise is that the number and location of CMLs should be optimized but not systematically reduced as this would possibly increase the risk, especially in areas where there is no uniformity in corrosion. It is prerequisite that the primary focus of CML optimization should be to acquire accurate information for the RBI process, however by reducing the number of CML’s, and so reducing cost, should not be a primary objective of CML optimization, but it is often a welcomed side effect (Hatton and McConnell, 2016). Furthermore, API 570, defines decision on the type, number, and location of the CML’s should consider results from previous inspections, the corrosion mix, corrosion rate, and the potential consequence of loss of containment. In theory, a circuit subject to perfectly uniform corrosion could be adequately monitored with a single CML. With this approach, there might be some reduction in CMLs where there is some anticipated uniformity where RBI risk assessment is necessary to be implemented so that the frequency of TMIs to be scheduled (Hatton and McConnell, 2016). At any case, CMLs should be distributed appropriately over the piping system to provide adequate monitoring coverage of major components and nozzles (API-570, 2009). As far as TMLs are concerned these can be defined by adopting the following some requirements. These involve the need for in-depth knowledge of the corrosion pattern (e.g. presence of uniformity or not) and a sense of the anticipated and localized corrosion rates in different sections of the piping circuits in combination with the corrosion allowance that is defined per manufacturer’s specs (time (Gysbers, 2012; API-570, 2009). For example, in areas where there is uniform corrosion, the number of CMLs required to monitor a circuit will be fewer than those required to monitor circuits subject to more localized metal loss, or there is a higher potential for CUI (API-570, 2009). The following Fig. 3 gives an idea of the RBI risk assessment effect in CMLs’ optimization and associated inspection frequency. Please insert Figure 3 Fig. 3. The RBI approach on CML optimization 4. Methodology for data gathering 4.1. Equipment non-destructive testing Critical equipment in hydrogen units requires monitoring, surveillance, and diagnostics tools to assess component aging effects and update inspection plan if any changes have been made. Condition monitoring technology uses special devices and sensors to monitor the operation of critical components continually. The monitoring techniques include vibration analysis, temperature measurement, thickness test, acoustic analysis, etc., Intelligent are deployed for continuous supervision and health monitoring of equipment beyond their design lifetime while the acquired data will be used for maintenance planning (Shafiee and Animah, 2017)

7

Nondestructive Testing (NDT) represents a part of this state of art monitoring and consists of a variety of non-invasive inspection techniques used to evaluate material properties, components, or entire process units. The techniques can also be utilized to detect, characterize, or measure the presence of damage mechanisms (e.g., corrosion or cracks). The most commonly adopted by the petrochemical industry in regards to the identification of wall cracks and thickness identification is the Radiography (RT), and Ultrasonic Testing (UT) where it is various forms that use ‘short’ and ‘high’ frequency ultrasonic waves e.g. Advanced Ultrasonic Backscatter Technique (AUBT), Phased Array Ultrasonic Testing (PAUT), Long Range Ultrasonic Testing (LRUT), Internal Rotating Inspection System (IRIS), Time of Flight Diffraction (TOFD), and Dry-Coupled Ultrasonic Testing (DCUT). UT provides good outcome and are used alone or in combination with other methods (e.g. visual inspection, liquid penetrant examination, radiography, etc.) to provide more reliable results in process piping, equipment shells, or storage tanks, and prior to deciding on any shut-down for maintenance reasons (API-570, 2009; Honarvar et al., 2013; Chang et al., 2005; Kot, 2016, Boateng et al.,2013). 4.2. Condition monitoring locations (CMLs) and Thickness Monitoring Locations (TMLs) Condition Monitoring Locations (CMLs) are designated locations on pressure vessels and piping, where Thickness Measurement Inspections (TMIs) are conducted to monitor the presence and rate of damage and corrosion. Most CMLs consist of at least one examination point, usually a two to three-inch diameter circle on the surface of the equipment. The number of readings taken within these areas varies, and more sample data increases confidence in the estimated wall thickness condition. A CML is considered a place where multiple Thickness Monitoring Locations (TMLs) are located at the location pipe or fitting to allow for maximum data gathering value in assessing pipe condition and potential for localized detection (see Fig. 4). These refer to each of the four quadrants on pipe and fittings, with particular attention to the inside and outside radius of elbows and tees where corrosion/erosion could increase corrosion rates (Gysbers, 2012; API-570, 2009). Please insert Figure 4 Fig. 4. CMLs and TMLs on steel pipe and U-shape fitting. CMLs are categorized and prioritized per corrosion loop sharing similar physicochemical conditions (e.g. temperature, pressure, nature of fluid phase), type of fluid, and soluble or insoluble impurities involved; the actual metallurgy involving different carbon steel alloys and stainless steel in various process loops is also considered Boateng, Dr. Dagadu et al. (2014). Selected corrosion loops incorporate CMLs and TMLs on a pipe and associated fittings’ as well as tanks and other static equipment wall thickness, which have been qualitatively assessed to be associated with corrosion mechanisms dealing with physicochemical, electrochemical and hightemperature hydrogen attack reactions and possibly with stress cracking, where such information is available. TMIs are pre-scheduled measurements in a time frame that fulfills production demand restrictions. This time frame is on average one year or even more and depends on each loop risk classification defined after an associated risk assessment execution, and not always including the same CMLs though. 4.3. Inspection Isometric Drawings (ISOs) Condition monitoring locations (CMLs) are depicted in isometric drawings, where particular coding is given. Inspection monitoring locations incorporate all significant components of the piping circuits (e.g. all valves, elbows, tees, branches, etc.), all secondary piping for high consequence per RBI analysis piping circuits, secondary piping up to the block valve that is normally used per similar RBI analysis unit pipe (API-580, 2002). So, TMIs results are clearly depicted in these drawings providing useful data per loop which is further processed statistically in specialized software. The following Fig. 5 and 6 provide indicative examples of these recordings. Please insert Figure 5 8

Fig. 5. CMLs on ISOs Please insert Figure 6 Fig. 6. TMIs recordings on TMLs (A, B, C, and D) 4.4. General Versus Localized Corrosion Monitoring General corrosion can be defined as the relative uniform metal loss over the entire piping surface. If plotted, the loss would be characterized by a normal distribution of thicknesses. Hence, random sampling of CMLs over the piping and other static equipment is used to characterize file normal distribution (mean/standard deviation), and simple techniques are used to estimate the minimum thickness. Owing to cost optimization requirements related to inspection and preventive maintenance, minimal sampling is required, while better estimates would require additional random sampling. On the other hand, any CML coverage and inspection frequency should maximize the ability to detect the onset of localized corrosion. Data gathered, and a well thought out CML coverage strategy should be analyzed thoroughly to determine if the sampling CML coverage and frequency is detecting the onset of localized corrosion (Gysbers, 2012). This breakeven point analysis is based on a detailed risk assessment per corrosion loop, where factors such are corrosion mechanisms mix along with the involved metallurgy and physicochemical conditions prevailing in each loop and section/part of the examined equipment or pipe. Experience, deep knowledge of the behavior of different metallurgies under different fluids and conditions involved in combination with the adoption of advanced statistical analysis, are key tools for this assessment where associated corrosion prediction tools are extracted as an outcome of this analysis.

5. Methodology for evaluation 5.1. Corrosion rate (CR) The total corrosion rate of exposed metallic parts of piping and equipment owes to a mix of mechanisms that may differ between corrosion loops. It is determined by the difference between two wall thickness readings, divided by the time interval between the readings. The determination of the corrosion rate may include thickness data collected in more than two different time periods. These periods may include one, two or even more years, depending on the particular item of equipment, the process involved and the risk assessed in terms of wall thickness reduction expected; the last is caused by the presence of a mix of corrosion and erosion mechanisms in combination with the metallurgy present defining a corrosion loop. Short-term and long-term corrosion rates should be distinguished. In particular, short-term corrosion rates are typically determined by two yearly thickness readings, whereas long-term rates use the most recent reading and one taken earlier in the life of the equipment, usually from shortly after the construction of the unit under study. These different rates help distinguish recent corrosion mechanisms from those acting over a longer period of time (Gysbers, 2012). What makes the most sense in many units is the total impact of the corrosion mix both in the short and in the medium run. This is because of the difficulty in identifying an adequate number of wall thickness measurements over extensive time series making possible the distinguishing and predicting of corrosion rates per loop. This becomes more complicated in extensive units involving hundreds of meters of piping of various sizing and numerous equipment parts.

9

The total corrosion rate of piping and equipment exposed metallic parts owes to a mix of mechanisms that may differentiate per different corrosion loop. A quite significant part of this rate can be determined quantitatively by the difference between two readings of wall thickness divided by the respective time interval. By this approach, calculations can include data on wall thicknesses which is collected in different time periods. These periods may refer to one, two or even more years, depending on the equipment part, the process involved and the risk that it is assessed, in correlation with the anticipated wall thickness reduction per loop. The yearly corrosion rates, the lifespan and the next inspection intervals are all calculated to determine the limiting component. As a result, corrosion loops with high failure potential due to increased risk will require more frequent monitoring. The designated areas of piping and fittings in which periodic examinations were conducted in order to monitor the presence and the rate of wall thickness reduction were: a) elbows, b) T-connections, c) reducers, d) connections with instruments commonly found in examined units (e.g. pressure indicators, flow meters, etc.), and e) points of linear pipe parts. The yearly Corrosion Rate (CR) of a CML is determined over the period between two measurements as: Corrosion Rate - CR (mm/year) = [(mmlatest – mmprevious)/[years ( tlatest – tprevious)] (2) where, mm latest = the minimum TML wall thickness at a CML in mm measured during the most recent inspection. mm previous = the minimum TML thickness at the same CML in mm, either on the first thickness measurement at this point, or at the start of a new corrosion rate environment or as measured during a previous inspection. If there are two or more periods of measurements for the same CML, then the Yearly Corrosion Rate (CR) is calculated as a weighted average of the corrosion rates per examined period:


    =

∑     ∑  



(3)

where, CRJ = the sub-corrosion rate in the jth of n≥3 periods, tJ = the examined periods in years. The lifespan of the examined pipe or equipment part was estimated by the formula: Lifespan (4)

(years)

=

[(mmpresent

–

mmallowed)]

/

[CR

(mm/year)]

where, mmpresent = the wall thickness in mm of a CML given by the latest measurement (mmlatest), converted by applying a prediction tool, and, mmallowed = the wall thickness in mm at the same CML down to the corrosion allowance (CA). The Predicted Wall Thickness after a specific time span (number of years) that is less than the lifespan (in years), is given by the formula: Predicted Wall Thickness (mmpredicted) = [mmpresent - CR (mm/year)] / years (tpredicted – tpresent)

where, mmpredicted = the wall thickness in mm of a CML predicted for a future time point, and, mmpresent is as above. 5.2. The case study

10

(5)

The case study, from which thickness measurement inspections have been performed owing to corrosion mechanisms, refers to a 12–year-old ‘Hydrogen Production Steam Reforming Unit’ located in a refinery of the Hellenic State. The referred unit produces nominally 65,000 Nm3 of H2 per hour. Its design was based on various feeds such as natural gas and stabilized naphtha (less used) or even mixes of naphtha and LPG. The key stages of the process are the following: a) desulphurization, b) pre-reforming, c) steam reforming, d) conversion of carbon monoxide through a shift reaction, e) purification of the hydrogen stream in the absorbers, f) steam production using the energy of combustion, and g) the burning in the ovens. Natural gas represents 95% of the total current unit feedstock, which also contains carbon dioxide, helium, hydrogen sulfide, and several other compounds (impurities) in smaller percentages. Both carbon dioxide (CO2) and hydrogen sulfide (H2S) are acidic gases that can cause extensive corrosion damage to carbon and other alloy steels in the presence of water in unit sections before the desulphurization step (Sun et al., 2007). The concentration of such impurities may vary significantly depending on the natural gas feedstock provider. The same may apply to naphtha as a feedstock, where sulfur hydrocarbons may also exist in a variety of forms. Moreover, the aggressive behavior of hydrogen under high temperatures (HTHA) is expected to be present in several stages of the process, especially during the steam reformation. (API-571, 2011; API-941, 2008; Thomas, 2014). 6. Data analysis 6.1. Measurements Most of the data were obtained in the scheduled periodic shutdowns during which preventive and corrective maintenance is usually performed. Ultrasonic testing equipment specialized for steel piping TMI measurements was employed. Equipment was always calibrated before its use in the field to reduce systematic measurement errors. This took place on standard pipe pieces and fittings of the same type and specifications as those in the surveyed corrosion loop. The thinnest reading or an average of several measurement readings taken from the area of a specific CML were recorded and used to calculate the corrosion rates. This CML corresponds to a particular angle of the pipe or an associated fitting, and the thinnest reading refers to the minimum measurement identified in XX’ and YY’ cross-sectional axis of piping and fittings. The most common method of addressing corrosion is to specify the Corrosion Allowance (CA), which is given by the manufacturer of the steam-reforming unit. This is a supplementary metal thickness that is added to the minimum thickness and is considered as an essential condition for the pipe or fitting to resist the applied loads. CMLs at the examined unit are points located upon piping, process equipment shell, or internal parts (e.g., coils) and process equipment external surfaces that have been pre-selected by their manufacturer. In this process, API guidelines are followed in combination with the anticipated mix of corrosion mechanisms expected to be present, such as a combination of galvanic corrosion with HTHA uniform and pit corrosion (API-941, 2008). Records of all examinations at CMLs were initially written by hand on isometric drawings by the unit maintenance personnel who performed the wall thickness measurements. They were then transferred to a soft-copy format to allow easy identification of their exact position for future repetition of measurements, as required, and the accurate calculation of the corrosion rate (CR). Wall thickness readings were also transferred to Excel spreadsheets for further categorization, sorting, analysis and extraction of preliminary data source tables, which are finally used in the statistical analysis. Tables 2 and 3 provide indicative information on wall thickness measurements for part of a corrosion loop. 7. Statistical analysis

11

Following the grouping of TMI measurements by period, unit section, steel alloy type, fluid type and nature, size of pipe, and presence of fittings (e.g. T-connections, elbows, size reducers, taps), statistical analysis was performed using STATA 14 statistical software (Fang et al., 2015) to assess the accuracy and the presence of possible non-linearity in the model. Multivariable regressions were performed both separately for each sectional corrosion loop, where a similar mix of metallurgy is present, and globally including all process sections as one integrated unit. It should be noted that the design approach of the specific unit follows the generic rule of a standard metallurgy per section, thus simplifying regression analysis and modeling. The generic form of the simplest multiple linear regression model adopted for the analysis of the TMI measurements on CMLs is the following regression Eq. (6) (Givehchi et al., 2016; Montgomery et al., 2012; Zheng et al., 2009). i



=

0

+ 1$i1 + 2$i2 + 3$i3 … . .

k$ik + *i

(6)

for the ith of n measurements, where the predictors $ik are the examined variables, namely temperature (T), pressure (P), metallurgy (M), type of fluid, nature of the fluid, fluid speed, and size of the pipe, and u i is a random error. The unknown estimators 0, 1, 2,… k are calculated by the method of ordinary least squares (OLS). The coefficient of determination  + is a number between 0 and 1 that measures the explanatory power of the regression, that is, what part of the sample variation of the ,’s is explained by the sample variation in the X’s. The closer the approach to 1, the better the fit of the model to the experimental result (Montgomery et al., 2012). Given that the examined unit is approximately 13 years old, and was designed with state-of-theart technology requirements, the volume of wall thickness measurement data was quite limited; there were few repetitions within two or three different time intervals at the same condition monitoring locations of sectional corrosion loops. Unrepeated measurements were omitted from the statistical analysis, although they provide essential information about the variance of the nominal wall thickness and the actual one, as they confirm the accuracy of the models extracted. The regression method applied was based on the available data and the need for combining ‘ideally’ all involved variables together describing the corrosion conditions per loop. This contained the testing of a wide range of linear, exponential, logarithmic, power, and a combination of all the above models on all independent variables, ensuring the best fit of the derived model to the experimental measurements. Dummy variables taking 0 and 1 values were defined for distinguishing different corrosion loops involving variations in metallurgy and the nature of the fluid involved in the pipe network. The results of this methodology include a family of multivariable equations Yi describing the anticipated yearly corrosion rate (CR) in each corrosion loop, incorporating all corrosion mechanisms that were present in each case; their effect is depicted by the TMI values that can be used in both short (~1 year) and medium (~2-5 year) term predictions. In the long run, this approach does not provide a reliable fit to the measurements unless adequate repetition of regressions is performed periodically. 7. Results 7.1 The models In the previous section, we described the process that resulted in linear regression models providing acceptable accuracy in a series of corrosion loops representing nearly 50-60% of the unit piping involved. However, areas involving single (unrepeated) measurements were omitted from the regression, because the analysts considered that nominal wall thicknesses could not be treated as initial measurements. This limitation was dictated by the observation that significant tolerance was reported by the manufacturers of pipe and fittings themselves, reaching ±15% of the nominal dimensions. The same tolerance ratio applies to equipment parts as well. It has been foreseen in the present models that potential extension of measurements through the institution of additional CMLs can be incorporated in the modeling process, thus comprising additional corrosion loops.

12

The two models proposed are multivariable equations providing information about the annual corrosion rate Yi, per loop, and CML. These equations are correlated with Xik variables, namely temperature (T), pressure (P), fluid velocity (V), linear pipe and pipe fitting diameter (D), as well as with dummy variables for factors such as type of fluid (Ft) and the presence of fittings (FT). The metallurgy (M) involved in different corrosion loops depends on the type of fluid (F) at the sections examined (only certain alloys can be used for specific fluids) and is therefore omitted. Considering the above, the first linear approach is the following equation for all loops: (,.i)/1  = . 0 + . 123i + . j5ij + . z7iz + . q29i5:iq + . 2:i + . 32i + 29i; . 4=i + . 5?i@

(7)

where, PHi = dummy variable representing the fluid nature. Fij = variable representing the fluid type. Miz = dummy variable representing the metallurgies. FTiq = dummy variable representing fittings on piping. Ti = temperature (Co). Pi = pressure (bar). PSi = dummy variable representing the presence of piping systems. Di = pipe nominal diameter (inch). Vi = fluid speed (m/sec). The above Eq. (7) creates a series of linear nomographs depicting the predicted yearly corrosion rate per temperature, pressure, and pipe size variation (where applicable). An indicative linear nomograph is given in the following Fig. 7. Please insert Figure 7 Fig. 7. Corrosion rate – ‘Naphtha’ loop The second modified linear approach with logarithmic variables goes similarly, as in the following: (,.i)/1  = . 0 + . 123i + . j5ij + . z7iz + . q29i 5:iq + . 2 log(:i)K+ + . 3 log(2i)+ + . 429i(=i/Vi)0,1(8) Where, variables specifications are the same as in the simpler linear Eq. (7) ones. An application example of Eq. (8) is presented here. In the corrosion loop dealing with a mix of naphtha feed, the associated physicochemical parameters involved, and adopted dummy variables, and regression model fitted coefficients are shown in Table 4. Per this example, the prediction for the corrosion rate at this point is CR= 0,502727 mm/year. The adoption of the simpler Eq (7) for the same naphtha feed under the same conditions provides a CR= 0,51306mm/year. These two CR points are depicted in the above Fig 7. 7.2 The developed software The whole procedure of data collection, analysis, and model fitting culminated in the development of a software tool in Microsoft Access©. This is a user-friendly database designed for in-house use by the oil refinery maintenance personnel. It comprises three main parts; a) data entry, b) data processing and c) results’ extraction, as follows.

13

Data Entry: This is divided into two main sections. The first concerns the import of all detailed key elements that describe the examined equipment and the prevailing operating conditions. This is specific data related to the various process steps, to the equipment and piping features, and the physicochemical properties of the substances involved. The example includes coded information on metallurgy, dimensions, and type of equipment parts, nature of fluid involved, type, volume, and fluid speed, prevailing corrosion mechanisms, and nominal fluid physical conditions, such as temperature and pressure. In addition, PDFs of technical details and drawings of equipment, together with indicative information related to interesting inspection findings and maintenance actions, are also included. Two screenshots of the tool are presented, depicting data entry for the definition of conditions per loop (Figure 9) and measurements per CML (Figure 9). In general, data entry is relatively simple insofar as specialized personnel define the corrosion loops and the associated anticipated mix of corrosion mechanisms, add the examined lines, equipment, and all respective CMLs, and then ensure that wall thickness measurements are updated. Please insert Figure 8 Fig. 8. Data entry – Definition of conditions per loop Please insert Figure 9 Fig. 9. Data entry – TMI readings per CML Processing: This part includes all essential forecast and risk assessment equations, aiming at the determination of lifespan or the anticipated corrosion per predefined period of the examined equipment parts referring to particular CMLs; it also deals with the respective prioritization of risk. Namely, the family of multivariable equations with logarithmic variables is applied per Eq. 8, by adopting the implementation of Eq. (2,3,4 and 5) for the predictions. The simplified forecasting approach proposed by the multivariable equation is depicted in the graph of Fig. 10. Please insert Figure 10 Fig. 10. Typical multivariable equation graph per CML This results from the typical scenario and shows the curve, which is defined by at least two past measurements performed in different years t1 and t2 at the same CML. If more than two measurements are available, then Eq. 3 is applied. When applying the equation at the specific CML, the first step is to determine both the present value of the wall thickness and the value at a later stage (at the end of the desired prediction period), up to the corrosion allowance. This is achieved by extending the curve until the desired time (t) and reading off the respective value on the Wall Thickness (mm) axis of the graph; the software performs this calculation automatically. Similar approaches may be applied in newly established CMLs located in already examined corrosion loops, where there is either only one measurement available, or alternatively by taking the nominal thickness value deducting the tolerance provided by manufacturers; however, in this case prediction results involve a much higher error, which should be co-estimated by the unit operation engineers. The above prediction is additionally tested by applying a simple linear equation which is based solely on the experimental data per CML, provided that there are at least two measurements available for two different periods. The logic of the equation is based on the following standard model: Y(Xi)/year = αΧ + ω

(9) 14

where, α = the slope of the straight line that defines the corrosion rate. X = the time (in years). ω = is equal to the original actual thickness of the wall. This simple approach is based solely on experimental data. Eq. (9) is a straight line with a negative slope (α<0), which is defined by the first and the last measurement performed at different times t1 and tj; by extending the line down to the corrosion allowance or to any desired future time. There should not normally be large deviations from the prediction model developed in this study; otherwise, Eq. (8) should be re-examined. This is only a qualitative testing tool with the purpose of warning operators that the prediction model needs fine-tuning or the data needs further checking. Risk Analysis: This step incorporates an additional semi-quantitative risk analysis for each corrosion loop, assessing the likelihood and consequences of the failure mechanisms. The likelihood is evaluated upon the basis of various factors, such as the: a) age of the equipment, b) history of damage to components, c) corrosion rate, and, d) estimated number and combination of corrosion mechanisms. The consequences of an instantaneous or continuous energy release (e.g. explosion or fire) is similarly assessed by defining the type of liquid and its flammability or combustibility, its quantity, the presence of bottlenecks in the process and the possibility of anticipating a shutdown of the unit and total cessation of operations. The risk analysis incorporates an assessment of the present and the expected serviceability of equipment by answering the following questions: what material and equipment involved are expected to be worn or damaged, and what is the likelihood of such material suffering damage; and what are the consequences of the loss scenario described. This methodology is actually qualitative, incorporating, however, some quantitative parameters (e.g. the quantity of the fluid and its pressure and temperature), with the aim of distributing associated risk per corrosion loop, and per equipment part. The analysis poses questions and leads to more accurate results with respect to the quite general commonly applied qualitative analyses and therefore avoids an excessively conservative classification of risk. This level of assessment adopts an 8 x 8 matrix to display the risk level results. The results of the risk analysis are depicted in a family of reports presented either only onscreen or as Excel or PDF documents. These reports involve extensive lists of coded CMLs, sorted by equipment part and corrosion loops. Filters are also provided for selective access to data. Example screenshots of these reports are presented here for the measurement list per corrosion loop and corrosion prediction and confirmation values. 8. Discussion of results Hydrogen steam reforming units operate on a mixture of feedstock composed of natural gas, naphtha, LPG streams or a mixture of them, produced in other units of large oil refineries under particularly demanding conditions owing to severe corrosion of their piping and equipment. The metallurgy mix and wall thickness used for these components are prescribed precisely by the relevant international codes. In the present study, the corrosion loops have been selected upon specific operating design codes, also having average inlet stream quality synthesis.

15

Designers and operators take into account the corrosion rates cited in the literature or given by uniform experimental data (e.g., Nelson graphs), and they also consider the recommendations arising from root cause analysis of past incidents. Based on this bank of knowledge, manufacturers specify the anticipated range of operating conditions for each piece of equipment and piping, including key factors such as the operating pressure and temperature, the exact synthesis of the fluid present in the loops together with its speed, and the geometry of pipings and vessels. This information is applied to the metallurgy mix selected and the nominal wall thickness and corrosion allowance values defined for each loop. However, volatility in oil prices nowadays demands from oil refineries flexibility in the selection of the inlet stream mix per periodic demand that may vary quality-wise as well. This policy results in significant operational cost savings, as the cost of raw materials is directly reflected in the expected gross profit margins, which have been relatively narrow in recent years. This choice may also result in the selection of inlet streams away from those prescribed by the plant’s designers; they may also contain impurities, and their mix may have a composition that creates a stressful environment for the plant metallurgy mix, thus creating greater uncertainties in the expected lifespan of such hydrogen units. For the sake of illustration, one could pose the question: “what if we increase the ratio of naphtha to natural gas in the hydrogen steam reforming unit feed stream?” Experimental data obtained by performing tests within such units has shown that the consequence is increased temperature and pressure in the main vertical tube of the catalytic furnace and the respective process fluid and hydrogen streams; these conditions, by default, increase the hydrogen attack (HTHA) phenomenon. Similar results could be assumed for other corrosion mechanisms, making the prediction of the change in the corrosion rate highly uncertain in most cases. The equations proposed in this paper and the nomographs obtained from them provide refinery operators with a practical tool for assessing the expected corrosion rate upon changing specific conditions (e.g. temperature and pressure), which directly affect the synthesis of inlet streams for particular pipe and fitting sizes. In addition, they can be used as additional tools for predicting the average corrosion rate over short and medium periods of time. It is also worth mentioning that methodology errors are a summation of standard errors given by the ultrasonic instruments per se and the technology involved in wall thickness measurements, augmented by the measurement error; the latter is affected by both the human element and the process restriction factors involved in the calibration of the instruments, the selection of the monitoring locations, the repetition of measurements at the same points, the execution of measurements in all three pipe transverse axes, the nature of the fluid, its condition and speed traversing the pipes, and the cleanliness (lack of corrosion) of the external surface of the pipe. Analyzing, in particular, the errors related to the ultrasonic technique, the following factors can affect their ability to make thickness measurements accurately: Calibration: Since material thickness is calculated using a known velocity, any difference between the actual material velocity and that used for instrument calibration will result in a skewing of the ultrasonic thickness measurements. Therefore, instrument calibration should ideally be performed on a known thickness of material identical to that being tested. In reality, that is usually not possible. Instead, either the known velocity for carbon steel or a carbon steel calibration block is used to perform the calibration (Thompson and Chimenti, 1993). Couplant: A couplant is a material (usually liquid) that facilitates the transmission of ultrasonic energy from the transducer into the test specimen. When making measurements in the pulse-echo mode, it is essential that the couplant layer be as thin as possible, as its thickness will be included in the read-out of the material’s thickness. (Thompson and Chimenti, 1993). Surface Condition: The surface condition is an important factor to consider when any type of instrumentation is used, but particularly when using a digital thickness gauge. Thickness gauges are meant to be used where both surfaces are smooth, flat and parallel. If the surface under the transducer is rough, excess couplant can be trapped between the transducer and surface, resulting in erroneous readings. Furthermore, if the back wall is rough, the ultrasonic pulse can be distorted or scattered, again resulting in erroneous readings. In addition, any loose or flaking scale, rust, corrosion or dirt on the surface of the part must be removed, otherwise, it will interfere with the coupling of the sound energy from the transducer into the material. Finally, although it is possible to make measurements through thin (a few thousands of an inch) layers of tightly adhered paint, thick paint will attenuate the signal and may create false echoes. (Thompson and Chimenti, 1993).

16

Part Geometry: Curved surfaces of pipes make accurate measurement more difficult. The center of the transducer must be held steady and perpendicular to the pipe while the measurement is being made. This becomes more difficult when the transducer is larger in diameter. (Thompson and Chimenti, 1993). Transducer Characteristics: Single element transducers depend on the front and back surfaces of the test piece being parallel. When this condition is met, a wide range of thicknesses can be accurately measured. If the surfaces are not parallel or if they are rough or corroded, a dual element transducer should be used. A limitation of the dual element is that it has a limited thickness range over which it can operate linearly (Thompson and Chimenti, 1993). The total error created because of the above factors is usually significant, especially when measurements are performed by in-house technical personnel or when different instruments are used over time. Although the repetition of measurements in the present case study is quite limited, both regression analyses proved that the statistical measure of how close the data are to the fitted regression line. However, macroscopically both equations provide similar results, a fact that demonstrates the applicability of the linear nomographs in the medium run. Further study of the example in Section 7 shows that the annual corrosion rate at the NAP loop of a particular two-inch size fitting (namely, a 90o elbow) is approximately 0,502727 mm/year at fluid temperature 35oC, fluid pressure 4.7bar, and fluid speed 6m/sec. The scenario of temperature rise by 10oC, with all other variables remaining constant, will result in a corrosion rate increase up to 0,530437mm/year, almost 6% higher than in the milder conditions. This temperature increase may be the result of a possible change in the inlet feedstock mix and its enrichment with additional naphtha instead of natural gas. The operational cost gain owing to this variation should be weighed against the reduction in a particular unit’s loop life expectancy so that the total operational cost break-even point can be assessed. It must be admitted, however, that the error of the methodology is relatively large at the present stage, owing to a lack of sufficient statistical evidence. In this logic, interventions to inlet mixes can be assessed proactively in more detail so that refineries will be able to control costs and production schemes according to market demand, while simultaneously not compromising safety. 9. Conclusions The adoption of prediction methodologies contributes to the prevention of accidents and to the preparation for emergency response (Zheng and Liu, 2009). RBI is a systematic analysis, establishing and ranking the risk levels associated with the operation of each piping, where identifies 10–20% of items that cover 80–95% of the risk exposures of the equipment and can contribute in effective prediction of static equipment faulting by applying various models such as linear regression and other forecasting statistic methodologies (Zheng and Liu, 2009; Chang et al., 2005). A method for determining sectional corrosion rate prediction tools in a hydrogen steamreforming unit is presented in this paper. In addition, a software tool has been developed to assist plant operators and managers in the efficient running of their assets. This tool, composed of multiple equations, has been assessed by means of regression analyses of wall thickness measurements performed in pipes and pipe fittings of a unit over the past 12 years of operational life at pre-selected condition monitoring locations (CMLs). Although the number of measurements available is rather limited, leading to relatively high uncertainty in results at present, the regression models and nomographs that have been constructed can be upgraded in the future, when a more extensive database has been acquired. These equations and the associated software can assist oil refinery operators in assessing the possible effect on the examined unit performance of any small changes to the inlet fluid mix as a consequence of oil market speculation. Additionally, the software tool supports the implementation of the RBIM techniques in the standard operating procedures of oil refineries, aiming to optimize periodic maintenance intervals in specific corrosion loops and risk-prone sections of units. A significant effort in designing and developing RBIM-related software has focused so far on the risk assessment of corrosion loops; these rather general-purpose approaches give “rule of thumb” results for a class of similar industries, involving linear approaches that provide good results only in the short run. The proposed methodology goes one step further, as it examines detailed condition monitoring locations (CMLs) per loop and provides non-linear based predictions fitted to the experimental results with a given error.

17

The main limitation of the study is the high anticipated error in medium and even more so in long-term predictions, as well as the lack of provision of any prediction related to corrosion loops without or with limited repeated measurements. The resulting less accurate predictions are heavily associated with the limited number of experimental results, also impacted by inherent shortcomings of the ultrasonic scanning methodology. The examined study has adopted only 124 usable (for the modeling purposes) observations, which are considered quite limited for reaching good fitting; hence R2 equals 0.5; however, the modified equations with logarithmic variables show quite acceptable P-values of t-tests, which provides greater confidence in the reliability of this approach. So, it is self-explained that these disadvantages may be drastically mitigated by rerunning regression analyses at frequent intervals incorporating updated wall thickness measurements (observations), whenever possible. Furthermore, the in-house developed software has adopted an additional control equation that provides linear based predictions based on two past measurements per CML and extrapolating down to the corrosion allowance (CA). In this way, the regression model may be additionally tested by this simple linear model, when cross-check data is available. The proposed CML regression-based nonlinear prediction model can readily prove its usefulness when used as an additional Decision-Making Tool in defining more accurate inspection and maintenance intervals; this will lead to savings associated with Risk-Informed decisions and in direct cost reduction, if the inspection/maintenance run times can be relaxed or intensified with a certain degree of confidence. Indirect cost reduction is also expected when more in-depth knowledge of the total measurable effect of the corrosion mechanisms helps to prevent undesirable downtime and consequent interruption of production. This effect may be vital, especially for units such as the examined steam reforming unit that incorporate unique pieces of process equipment, leaving the whole refinery on hold in case of failure. This configuration may also be present in other sub-units of refineries or petrochemical complexes. The notion of overall Risk-Informed Decision Making in inspection and maintenance in refineries is promoted through the creation of the respective Risk Matrix, where several significant aspects have been aggregated. These include monitoring of cost, reliability, maintainability, and system availability; all factors should be treated simultaneously in the modeling of the decision maker’s preferences structure proposing suitable inspection intervals that consider the consequences involved (Ferreira et al., 2009). For this reason, Multi-Criteria Decision models should be pursued using the current study results as additional criteria for defining the trade-offs between earnings and plant downtime.

References Allevato, C., 2011. Utilizing acoustic emission testing to detect high-temperature hydrogen attack (HTHA) in Cr-Mo reformer reactors and piping during thermal gradients. Procedia Eng. 10, 3552–3560. https://www.sciencedirect.com/science/article/pii/S1877705811007739 API 571, 2011. Damage mechanisms affecting equipment in the Refinery Industry, American Petroleum Institute. https://www.api.org/Standards/ API 580, 2002. Recommended https://www.api.org/Standards/

Practice

for

API 581, 2016. Base Resource https://www.api.org/Standards/

Documentation,

Risk-Based Risk-Based

Inspection, Inspection,

American American

Petroleum

Institute.

Petroleum

Institute.

API 941, 2008. Steels for Hydrogen Service at Elevated Temperatures and Pressures in Petroleum Refineries and Petrochemical Plants, API Recommended Practice, American Petroleum Institute. https://www.api.org/Standards/ API-570, 2009. Piping Inspection Code, In-service Inspection, Rating, Repair, and Alteration of Piping Systems, American Petroleum Institute https://www.api.org/Standards/ Arunraj, N.S., Maiti, J., 2007. Risk-based maintenance—Techniques and applications. J. Hazard. Mater. 142, 653661. https://www.sciencedirect.com/science/article/pii/S0304389406007345 Arunraj, N.S., Maiti, J., 2010. Risk-based maintenance policy selection using AHP and goal programming. Saf. Sci. 48, 238-247. https://www.sciencedirect.com/science/article/pii/S0925753509001684 18

Bertolini, M., Bevilacqua, M., Ciarapica, F.E, Giacchetta, G., 2009. Development of Risk-Based Inspection and Maintenance procedures for an oil refinery. J. Loss Prev. Process Ind. 22, 244-253. https://www.sciencedirect.com/science/article/abs/pii/S0950423009000059 Bharadwaj, U.R., Silberschmidt, V.V., Winle, J.B., 2012. A risk-based approach to asset integrity management, Journal of Quality in Maintenance Engineering. 18(4) ,417-431. https://www.emeraldinsight.com/doi/abs/10.1108/13552511211281570 Boateng, A. Danso, K, Dagadu, C., 2013. Non-Destructive Evaluation of Corrosion on Insulated Pipe Using Tangential Radiographic Technique. IJSTR. 2(6), 7-13. http://www.ijstr.org/final-print/june2013/Non-destructive-Evaluation-OfCorrosion-On-Insulated-Pipe-Using-Tangential-Radiographic-Technique.pdf Boateng, A., Dagadu, C., Tikwa, A., Awuvey, D., Amoakohene, E., Kwaasi, E., 2014. Determination of corrosion rate and remaining life of pressure vessel using ultrasonic thickness testing technique. GJEDT. 3(2), 43-50. https://www.longdom.org/articles/determination-of-corrosion-rate-and-remaining-life-of-pressure-vessel-usingultrasonic-thickness-testing-technique.pdf Changa, M., Changa, R., Shua, C., Lin K., 2005. Application of risk-based inspection in refinery and processing J. Loss Prev. Process Ind. 18, 397-402. piping https://www.sciencedirect.com/science/article/abs/pii/S0950423005000951

.

Changa, M.K., Changa R.R., Shua C.M., Linb K.N., 2005. Application of risk based inspection in refinery and processing piping. J. Loss Prev. Process Ind. 18 ,397–402. https://www.sciencedirect.com/science/article/abs/pii/S0950423005000951 Fang, X., Zhou, R., Gebraeel, N., 2015. An adaptive functional regression-based prognostic model for applications with missing data. Reliab Eng Syst Safe, 266–274. https://www.sciencedirect.com/science/article/pii/S0951832014002063 Ferreira, R., Almeida A., Cavalcante C, 2009. A multi-criteria decision model to determine inspection intervals of condition monitoring based on delay time analysis. Reliab Eng Syst Safe., 94,905–912. https://www.sciencedirect.com/science/article/pii/S0951832008002512 Garverick L., 1994. Corrosion in the Petrochemical Industry, ASM https://www.asminternational.org/search/-/journal_content/56/10192/25871552/Publication

International.

Givehchi, S., Zohdirad, H., Ebadi, T., 2016. Utilization of regression technique to develop a predictive model for hazard radius from release of typical methane-rich natural gas. J. Loss Prev. Process Ind. 44, 24-30. https://www.sciencedirect.com/science/article/abs/pii/S0950423016302224 Gysbers, A. C., 2012. A Discussion on the Piping Thickness Management Process, Determining Corrosion Monitoring Locations. Inspectioneering Journal. https://inspectioneering.com/journal/2012-11-01/2995/a-discussion-on-thepiping-thi Hameed, A., Khan, F., 2014. A framework to estimate the risk-based shutdown interval for a processing plant. J. Loss Prev. Process Ind. 32(10),1016. https://www.sciencedirect.com/science/article/abs/pii/S0950423014001181 Hatami, S., Ghaderi-Ardakani, A., Niknejad-Khomami, M., Karimi-Malekabadi, F., Rasaei, M.R., Mohammadi, A.H., 2016. On the prediction of CO2 corrosion in petroleum industry. J. of Supercritical Fluids. 117,108–112. https://www.sciencedirect.com/science/article/pii/S0896844616301474 Hatton, G., McConnell M., 2016. CML Reduction - Are you planning to fail. Inspectioneering Journal. https://inspectioneering.com/journal/2016-12-29/5954/cml-reduction--are-you-planning-to-fail Honarvar, F., Salehi, F., V. Safavi, A. Mokhtari and Sinclair A. N., 2013. Ultrasonic monitoring of erosion/corrosion thinning rates in industrial piping systems. Ultrason. 53(7), 1251-1258. https://www.sciencedirect.com/science/article/pii/S0041624X1300070X James, W. and Widrig. J. R., 2018. A Lifetime of Integrity - Overcoming the Challenges in Managing the Life-Cycle of Aging Assets. Inspectioneering Journal. https://inspectioneering.com/journal/2018-06-28/7752/a-lifetime-ofintegrity---overcoming-the-challenges-in-managingKhan, F. I., Haddara, M. M., 2004. Risk-based maintenance (RBM), a quantitative approach for maintenance/inspection scheduling and planning. Institution of Chemical Engineers. Process Saf Environ. 82(B6), 398–411. https://www.sciencedirect.com/science/article/abs/pii/S0950423003000949

19

Kot, R., 2016. Hydrogen attack, detection, assessment, and evaluation. Intico, 84 Churchill Road North, Dry Creek, S.A. 5094 Australia. https://www.ndt.net/apcndt2001/papers/1154/1154.htm Krishnasamy, L., Khan F., Haddara, M., 2005. Development of a risk-based maintenance (RBM) strategy for a powergenerating plant. J. Loss Prev. Process Ind. 18, 69–81. https://www.researchgate.net/publication/223760785_Development_of_a_riskbased_maintenance_RBM_strategy_for_a_power-generating_plant Marlow, D.R., Beale, D.J., Mashford, J.S., 2011. Risk-based prioritization and its application to inspection of valves in the water sector. Reliab Eng Syst Safe. 100, 67-74. https://www.sciencedirect.com/science/article/pii/S0951832011002778 Mellor, P., 1987. Software Reliability Prediction: Derivation of Model Parameters from Failure Data. Qual Reliab Eng Int. 4(2),12-26. https://www.emeraldinsight.com/doi/abs/10.1108/eb002876 Mirza, N. R., Degenkolbe, S., Witt, W., 2011. Analysis of hydrogen incidents to support risk assessment. Int. J. Hydrogen Energy. 36, 12068 -12077. https://www.sciencedirect.com/science/article/pii/S0360319911015527 Monferini, A., Konstandinidou, M., Nivolianitou, Z., Leva, C., Kontogiannis, T., Kafka, P., Mckay, A., 2013. A compound methodology to assess the impact of human and organizational factors impact on the risk level of hazardous industrial plants. Reliab Eng Syst Safe. 119, 280–289. https://www.sciencedirect.com/science/article/pii/S0951832013001075 Montgomery, D.C., Peck, E.A., Vining, G.G., 2012. Introduction to linear regression analysis, 5th edition. ISBN: 978-0470-54281-1 pp.12-26, 67-103

.

Moura, M.C., Lins, I.D., Droguett, E.L., Soares, R. F., Pascua, l.R., 2015. A Multi-Objective Genetic Algorithm for determining efficient Risk-Based Inspection programs. Reliab Eng Syst Safe. https://www.sciencedirect.com/science/article/pii/S0951832014002348 Nibler, R., 1997. A regression model for matching parallel systems. Qual Reliab Eng Int. 14(2),176-185. https://www.emeraldinsight.com/doi/abs/10.1108/02656719710165437 Nivolianitou, Z., 1992. Safety Analysis of Naphtha Flash Separator Using the DYLAM Methodology. Qual Reliab Eng Int. 9(2). https://www.emeraldinsight.com/doi/abs/10.1108/02656719210008490 Nivolianitou, Z., 1998. Hazard Review of a Pesticides Formulation Plant and Consequence Assessment of Accident Scenarios in it. J. Loss Prev. Process Ind. 11, 123-134. https://www.researchgate.net/publication/223186370_Hazard_review_of_a_pesticides_formulation_plant_and_con sequence_assessment_of_accident_scenarios_in_it Papazoglou, I. A., Nivolianitou, Z., Aneziris, O., Christou, M., 1992. Probabilistic safety analysis in chemical installations. J. Loss Prev. Process Ind. 5(3), 181-191. https://www.sciencedirect.com/science/article/abs/pii/095042309280022Z Rasmekomen N., Parlikad A., 2016. Condition-based maintenance of multi-component systems with degradation state-rate interactions. Reliab Eng Syst Safe. 148, 1–10. Sepeda, A.L., 2008. A risk based maintenance approach (for facilities complying with the US OSHA PSM regulation). J. Loss Prev. Process Ind. 22(6), 680-684. https://www.sciencedirect.com/science/article/abs/pii/S0950423008000843 Shafiee, M., Animah, I., 2017. Life extension decision making of safety critical systems, An overview. J. Loss, Prev. Process Ind. 47,174-188. https://www.sciencedirect.com/science/article/abs/pii/S0950423016304508 Shankar, G., Sahani, V., 2003. Reliability analysis of a maintenance network with repair and preventive maintenance. Qual Reliab Eng Int. 20(2), 268-280. https://www.scribd.com/document/299668069/Reliability-analysis-of-amaintenance-network-with-repair-and-preventive-maintenance-pdf Shishesaz, M. R., Bajestani, M.N., Hashemi, S.J., Shekari, E., 2013. Comparison of API 510 pressure vessels inspection planning with API 581 risk-based inspection planning approaches. Int. J. Press. Vessels Pip. 111-112, 202-208. https://www.sciencedirect.com/science/article/pii/S0308016113001269 Singh, M., Markeset, T., 2009. A methodology for risk-based inspection planning of oil and gas pipes based on fuzzy logic framework. Eng. Fail. Anal.. https://www.sciencedirect.com/science/article/abs/pii/S1350630709000284

20

Sun, W. Nesic, S., 2007. A mechanistic model of H2S corrosion of mild steel. Institute for Corrosion and Multiphase Technology. Department of Chemical Engineering, Ohio University. N.07655. https://www.onepetro.org/conference-paper/NACE-07655 Thomas, C., 2014. Risk analysis for high-temperature hydrogen attack. Quest Integrity Group Limited Lower Hutt, New Zealand. US Chemical Safety and Hazard Investigation Board Report. 2010-01-I-WA. http://www.questintegrity.com/assets/PDFs/Technical-Papers-2014/Risk-Analysis-for-High-Temperature-HydrogenAttack-Asian-Nitrogen-Syngas-2014.pdf. Thompson, D.O., Chimenti, D.E., 1993. Ultrasonic Measurement of Pipe Thickness, Review of Progress in Quantitative Nondestructive Evaluation. Plenum Press, New York. 12, 1987-1994. https://pdfs.semanticscholar.org/5521/46d17d3486971bc2be30fdcc8f37a54d406a.pdf. Tien, S. W., Hwang, W.T., Tsai, C.H., 2007. Study of a risk-based piping inspection guideline system. ISA Transactions. 46,119-126. https://www.sciencedirect.com/science/article/abs/pii/S0019057806000073 Tixier, J., Dusserre G., Salvi O., Gaston D., 2002. Review of 62 risk analysis methodologies of industrial plants. J. Loss Prev. Process Ind. 15, 291–303. https://www.sciencedirect.com/science/article/abs/pii/S0950423002000086 Vianello, C., Milazzo, M., Guerrini, L., Mura A., Maschio G., 2016. A risk-based tool to support the inspection management in chemical plants. J. Loss Prev. Process Ind. 41,154-168. https://www.sciencedirect.com/science/article/abs/pii/S0950423016300602 Wang, Q., Gao J., 2012. Research and application of risk and condition based maintenance task optimization technology in an oil transfer station. J. Loss Prev. Process Ind. 25,21—1027. https://www.sciencedirect.com/science/article/abs/pii/S0950423012000915 Wei, L, Xiaolu, Pang X., Gao, K., 2016. Effect of small amount of H2S on the corrosion behavior of carbon steel in the dynamic supercritical CO2 environments. Corros. Sci. 103,132–144. https://www.sciencedirect.com/science/article/pii/S0010938X15301505 You, M.Y., Meng G., 2011. Updated proportional hazards model for equipment residual life prediction. Qual Reliab Eng Int. 28(7), 781-795. https://www.emeraldinsight.com/doi/abs/10.1108/02656711111150850 Zheng, X. and M. Liu, 2009. An overview of accident forecasting methodologies. J. Loss Prev. Process Ind. 22(4), 484-491. https://www.sciencedirect.com/science/article/abs/pii/S0950423009000552

21

Table 1 Recommended maximum inspection interval (API-570, 2009; Chang, Chang, et al. 2005) Type of circuit

Thickness measurements

Visual External

Class 1

5 years

5 years

Class 2

10 years

5 years

Class 3

10 years

10 years

Injection points

3 years

By piping class

Soil-to-air interfaces

NA*

By piping class

*NA, not applicable

Table 2 Indicative corrosion loop & variables of a corrosion loop PID code

Fluid

*

*

Temper ature o ( C)

Nature of fluid

76-PID 0021-31

NAP

35

G

76-PID0021-31

NAP

35

G

*

Pressur e (bar)

Line code

Metallu rgy code

Pipe dimensi on (inch)

Pipe Type

Nominal wall thicknes s (mm)

Corrosi on allowan ce (mm)

0,1

763100 1

C0031C

3

S-40

5,49

3

0,1

763100 1

C0031C

4

S-40

6,02

4

*PID=Process & Instrumentation Diagram, NAP= Naphtha, G=Gas

Table 3 TMI measurements vs. nominal thickness Monitoring location code (CML)

1.1

1.2

2

Line/Fitting type

R

R

90

T

Size (inch)

3

4

4

Nominal wall thickness (mm)

5,49

6,02

Measured wall thickness (mm)

5,40

∆ – Variance (mm)

0,09

*

4

5.1

5.2

6

7

L

90

R

R

90

45

90

4

4

4

4

4

4

4

4

6,02

6,02

6,02

6,02

6,02

6,02

6,02

6,02

6,02

6,00

5,90

6,00

5,70

5,76

6,10

6,00

6,00

5,90

5,60

0,02

0,12

0,02

0,32

0,26

-0,08

0,02

0,02

0,12

0,42

o

3

*

3.1

*

*

o

*R=Reducer, 90=Elbow 90 , 45=Elbow 45 , L=Pipeline 22

8

*

Table 4 Corrosion loop prediction model implementation example 0,1 Model (,. i)/1  = . 0 + . 123i + . j5ij + . z7iz + . q29i 5:iq + . 2 log(:i)K+ + . 3 log(2i)+ + . 429i(=i/Vi)

CML Parameters

Paramete r Variables

Values of Dummy Variables

Model Variables

L

23i

0

23i

n.a

Fluid Type

NAP

5ij

n.a

5ij

0,001

Metallurgy

C0031C

7iz

0*

7iz

n.a

Type of equipment

Piping

29i

1

90

5:iq

1

2inch

=i

n.a

Fluid Velocity

6m/sec

Vi

Fluid Temperature

35 C

o

4,7bar

Fluid Phase

Shape Diameter

Fluid Pressure

Paramete r Values

Lik

L′ik

Model Variable Estimators

Oi N

29i 5:iq

-0,03412

n.a

29i(=i/Vi)

0,22657

:i

n.a

log(:i)K+

0,21703

2i

n.a

log(2i)+

0,15317

23i = dummy variable representing the fluid nature. 5ij = variable representing the fluid type. 7iz = dummy variable representing the metallurgies. 5:iq = dummy variable representing fittings on piping. o :i = temperature (C ). 2i = pressure (bar). 29i = dummy variable representing the presence of piping systems. =i = pipe nominal diameter (inch). Vi = fluid speed (m/sec). o 90 = Elbow 90 NAP = Naphtha C0031C = type of carbon steel *collinearity with Fluid Type (NAP)

23

0,1

Model Constant Estimator O0 N

Predicted Corrosion

- 1,066377

0.502727

CR (mm/year)

Highlights • A family of multivariable functions was developed, providing a prediction of measurable corrosion yearly rate and associated remaining life of equipment parts of a hydrogen steam reforming unit. • A multivariable regression methodology has been adopted. • •

•

The application of these functions demonstrates the model’s capability to provide a comprehensive identification of failures and associated accidents involving hydrogen explosions. The outcome of this effort is to assist hydrogen and other high-risk units in the refinery and petrochemical sector in optimizing inspection and maintenance scheduling, based on the anticipated corrosion rate calculated by adopting the methodology, models and software proposed. Optimized inspection and maintenance without compromising the safety, lead to lower operational costs.

1 Author declaration 1. Conflict of Interest Potential conflict of interest exists: We wish to draw the attention of the Editor to the following facts, which may be considered as potential conflicts of interest, and to significant financial contributions to this work: The nature of potential conflict of interest is described below: X

No conflict of interest exists.

We wish to confirm that there are no known conflicts of interest associated with this publication and there has been no significant financial support for this work that could have influenced its outcome. 2. Funding Funding was received for this work. All of the sources of funding for the work described in this publication are acknowledged below: X

No funding was received for this work.

3. Intellectual Property X We confirm that we have given due consideration to the protection of intellectual property associated with this work and that there are no impediments to publication, including the timing of publication, with respect to intellectual property. In so doing we confirm that we have followed the regulations of our institutions concerning intellectual property. 4. Research Ethics X We further confirm that any aspect of the work covered in this manuscript that has involved human patients has been conducted with the ethical approval of all relevant bodies and that such approvals are acknowledged within the manuscript. IRB approval was obtained (required for studies and series of 3 or more cases) Written consent to publish potentially identifying information, such as details or the case and photographs, was obtained from the patient(s) or their legal guardian(s).

2 5. Authorship The International Committee of Medical Journal Editors (ICMJE) recommends that authorship be based on the following four criteria: 1. Substantial contributions to the conception or design of the work; or the acquisition, analysis, or interpretation of data for the work; AND 2. Drafting the work or revising it critically for important intellectual content; AND 3. Final approval of the version to be published; AND 4. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All those designated as authors should meet all four criteria for authorship, and all who meet the four criteria should be identified as authors. For more information on authorship, please see http://www.icmje.org/recommendations/browse/roles-andresponsibilities/defining-the-role-of-authors-and-contributors.html#two. X All listed authors meet the ICMJE criteria. We attest that all authors contributed significantly to the creation of this manuscript, each having fulfilled criteria as established by the ICMJE. One or more listed authors do(es) not meet the ICMJE criteria. We believe these individuals should be listed as authors because:

X

We confirm that the manuscript has been read and approved by all named authors.

X We confirm that the order of authors listed in the manuscript has been approved by all named authors.

6. Contact with the Editorial Office The Corresponding Author declared on the title page of the manuscript is: Dr. Zoe NIVOLIANITOU

X

This author submitted this manuscript using his/her account in EVISE. X

We understand that this Corresponding Author is the sole contact for the

3 Editorial process (including EVISE and direct communications with the office). He/she is responsible for communicating with the other authors about progress, submissions of revisions and final approval of proofs. We confirm that the email address shown below is accessible by the Corresponding Author, is the address to which Corresponding Author’s EVISE account is linked, and has been configured to accept email from the editorial office of American Journal of Ophthalmology Case Reports: [Insert email address you wish to use for communication with the journal here]

Someone other than the Corresponding Author declared above submitted this manuscript from his/her account in EVISE: [Insert name below]

X We understand that this author is the sole contact for the Editorial process (including EVISE and direct communications with the office). He/she is responsible for communicating with the other authors, including the Corresponding Author, about progress, submissions of revisions and final approval of proofs. We the undersigned agree with all of the above.

Author’s name (Fist, Last)

1. __N. Deftereos 2. C.Kyranoudis 3. Z. Nivolianitou 4. O.Aneziris On behalf of all authors:

Signature

_____________ _____________ _____________ _____________

Date

14.11.2019 14.11.2019 14.11.2019 14.11.2019