Trade Off Economics White Paper

Asset Performance Networks

TRADE-OFF ECONOMICS IN REFINERY AND

PLANT TURNAROUNDS

- TOWARD A DYNAMIC DECISION-MAKING MODEL -

By Jan A. Jackson, MBA, MIB, PMP, CCC (Senior Consultant – Project Consulting Services (2007)) Washington D.C. Office 3 Bethesda Metro Center Suite 925 Bethesda, MD 20814 Tel: +1 (240) 683-1001 Fax: +1 (240) 683-1009

www.ap-networks.com

Jan A. Jackson, MBA, MIB, PMP, CCC

Table of Contents

ABSTRACT KEYWORDS

I. INTRODUCTION .................................................... 2

A. Background of Study ................................................... 2

B. Purpose and Need ....................................................... 2

II. CASE STUDY – CREATING THE MODEL .................. 3

A. Available Data .............................................................. 3

B. Data Analysis ............................................................... 3

C. Modeling Data Relationships ...................................... 9

III. MAKING THE CASE .............................................. 10

IV. CONCLUSION ...................................................... 15

2

www.ap-networks.com


ABSTRACT

Plant turnaround economics are highly complex. Variables impacting trade-off considerations weave an intricate web of economic interdependencies. Decision-making in this dynamic environment however has often been based on little more than anecdotal information. This paper explores the fundamental relationships between major cost-contributing factors such as shift-patterns, labor productivity, turnaround duration, fixed costs, quality and opportunity costs. Once the basic interplay between these variables is established the paper then proposes a generic trade-off assessment model and presents a case study to illustrate –through the use of sensitivity analysis- the impact of specific trade-off decisions upon overall economic viability of the turnaround. The study further explains the concepts of dynamic decision-making, prioritization and selection of available economic and operational alternatives. Data collection and research is based on more than 500 actual turnaround events that have yielded a multitude of detailed observations.

KEYWORDS

Trade-off, sensitivity analysis, interdependency, opportunity cost, fixed, dynamic, turnaround, outage, productivity, shift-pattern, decision support, constraints

I. INTRODUCTION

A. Study Background

Within the process-related industries in general and petro-chemical areas in specific, individual plant effectiveness is gauged by and based upon a variety of metrics that measure financial, cost-, and technical aspects of the overall plant operations. Turnarounds (TAs) are considered critical events within the maintenance framework of plants. In recent years, major transformations in planning rigor for maintenance outages and changing attitudes toward turnaround preparedness have measurably improved the economic viability of refineries despite rising crude prices, escalating material costs, tightening labor markets, and stricter environmental regulations [1] [2]. Whereas pre-planning methodologies and risk management concepts for outages have improved, the actual understanding and implications of trade-off decisions and interdependencies between resource constraints during outage execution are still largely undefined in quantitative terms. B. Purpose and Need

This paper attempts to answer some of the underlying questions involved in effective decision-making for turnarounds. The relatively short execution timeframes of turnarounds largely inhibits corrective actions mid-performance. However, turnaround managers need tools to assess impacts that specific decisions may have on the financial outcome of the TA. With rising crude prices comes rising short-term volatility and increased difficulty to plan for outage economics and optimization of returns.

www.ap-networks.com


Therefore, linking relevant operational variables in an interdependent framework is

intended to produce a decision-making model that not only provides turnaround professionals with a more thorough understanding of such interdependencies but also supports managers and operations personnel in weighing concrete options and demonstrating the impact of specific changes upon the overall outcome. A simplified model will be established to show the effect of specific decisions, such as TA start date, duration, safety impacts or overtime structure upon the overall TA outcome. The final objective is to gain a better understanding of the dynamic environment that characterizes turnarounds. Given the significant number of topics addressed in this study, the available literature is rather expansive. Trade-off economics for refinery operations have been addressed by Gary, Kaiser, and Handwerk (2004) [3].

II. CASE STUDY – CREATING THE MODEL

A. Available Data

This study is largely based on a randomly selected sample of approximately 200 (n1) TAs from a proprietary turnaround database population of more than 500 (N) unique plant outages [4]. After screening for and eliminating turnarounds with inadequate levels of detailed information, the final data set numbered 178 (n2) turnarounds. Seventy-five (75) percent of these turnarounds occurred in North America with some data collected from chemical facilities. All turnarounds in sample n2 (or N for that matter) are categorized by more than 40 individual performance variables such as planned hours, actual hours, safety incidents, TA duration in days (planned and actual), and so on. Such wealth of specific data allows researchers to design studies as broad or as narrow as they wish.

The reality of data collection ‘in the field’ however is such, that not all variables may be

obtained for all TAs all of the time. In other words, specific gaps exist for individual TAs in some areas but not in others, whereas it might be the opposite case for another TA in the sample. In such cases, we screen TA sample n2 for availability of the specific variable under review (e.g. ‘Safety Incidents versus Total Actual Hours’). This process obviously yields smaller sample sizes given the more limited number of TAs featuring all required variables. The scope of this paper does not include exploration of basic methodologies on appropriate sample sizes and determination of standard errors but does caution that any data analysis techniques need to adhere to basic data sampling premises.

B. Data Analysis

TA SIZE AND LABOR PERFORMANCE Seventy-five (75) percent of turnarounds within sample n2 exceeded their labor-hour target.

On average, these turnarounds expended an additional twenty-three (23) percent of labor hours above and beyond the estimated hours. However, size does matter when it comes to assessing the probability of experiencing labor overruns, as the level of overruns increases with TA size. In the following chart Labor Performance Index (LPI) is defined as the actual labor hours divided by estimated hours (adjusted for approved scope change).

4

www.ap-networks.com


Figure (1) – TA Labor Hrs and Labor Performance Index

The causes for this worsening labor performance characteristic may not be easily discernable from the data itself, since we are not comparing final TA outcomes against a universally accepted performance baseline but company-provided estimates, which may be based on vastly different assumptions. One possible explanation however is that TA planners continue to apply outdated estimating standards that are not subjected to updates in regular feedback loops, thus producing a compounding effect of low estimates especially on larger TAs with higher complexity.

Larger TAs feature significant ‘scale-up’ issues (e.g. logistics) which need to be priced into

the estimate to avoid cost-related surprises. Applying the same estimating factors across all TA sizes may be one of the culprits responsible for missing cost targets on large TAs. The actual average size of TAs in sample n2 is approximately 231,000 labor hours. Labor hours include components such as field maintenance, TA support, and capital project hours. The definition of labor hours is critical for creating a common and comparable baseline variable for further analysis. WORK INTENSITY AND LABOR PERFORMANCE The difficulty in building realistic and dynamic costing models for turnarounds is increased by the number and complexity of individual data relationships. The existing data does indicate a potentially significant relationship between the hours worked per day during a turnaround and the Labor Performance Index. This relationship could simply be a derivative of the previously described correlation between LPI and TA size and thus not be causal at all, since larger TAs tend to post a higher number of hours per workday as well.

www.ap-networks.com


Figure (2) – Hrs/Day and Labor Performance Index

The Labor Performance Index is neutral (1.00) at around 3,900 hrs per day. In other words, if we assume a causal relationship, then the risk of overrunning labor budgets appears to increase steadily for turnarounds with higher labor intensity. This seems simplistic enough at first sight but warrants several more in-depth considerations. The above scatter plot displays a strongly linear relationship up through approximately 10,000 hrs per day (1.32 Labor Performance Index) but seems to follow a logarithmic pattern beyond. WORK INTENSITY AND SAFETY INCIDENTS The available data further suggest an accelerating relationship between the hours worked per shift/day (work intensity) and the number of safety incidents. Safety incidents are defined as the sum of recordable events and first aids. Among the variables collected for TAs in sample n2, ‘Safety Incidents’ constitutes one of the bigger challenges due to potential data reporting issues from refinery personnel. Frequently, the safety statistics for a specific TA are not known to the engineer charged with completing the TA data questionnaire, in which case many simply report 0 safety incidents.

This practice however, leads to obvious confusion with accurately reported 0-incident TAs. For the above chart only those turnarounds have been selected that reported at least one safety incident since many plants fail to report requested safety data. By using only turnarounds with at least 1 reported incident, these potential reporting failures are likely eliminated. Underlying trends however between work intensity and safety incidents can still be (or more so) elaborated with the remaining data.

6

www.ap-networks.com


Figure (3) – Hrs/Day and Safety Incidents

0

5

10

15

20

25

30

35

40

45

50

0 2,000 4,000 6,000 8,000 10,000 12,000 14,000

HOURS WORKED / DAY

SA

FE

TY

IN

CID

EN

TS

The resulting function is polynomial and thus characterized by an accelerating rate of safety incidents with increasing work intensity. Including safety considerations in trade-off decisions typically evokes strong reactions from plant stakeholders, especially, if one attempts to monetize safety incidents from a theoretical perspective. Nevertheless, the data are relatively clear about this relationship: the more hours are worked today, the higher the risk of safety incidents. LABOR PERFORMANCE AND SHIFT PATTERN

Much literature exists on the topic of fatiguing of craft personnel and loss of productivity due to extended overtime [5]. Available studies mostly agree on the longer-term, adverse effect of overtime, but can vary significantly on the degree of productivity decline across the measurement timeframe.

According to the selected sample, the 6-10s shift pattern offers by far the most predictable scenario among the three presented shift types in terms of outcome. Labor hour performance for 6-10s is tightly spread around the mean with only modest variability and small range among the individual values. The observed variances increase significantly from 6-10s to the 7-10s and 7-12s shift pattern. The collected sample of turnarounds however did not indicate a significant difference when comparing the mean of both 7-day shift types. In fact, the average Labor Performance value for 7-12s is slightly lower than the 7-10s mean. The following chart demonstrates the likelihood of labor overruns according to shift pattern.

www.ap-networks.com


Figure (4) – LPI, Average LPI by Shift Pattern

0.80

1.00

1.20

1.40

1.60

1.80

2.00

6-10s 7-10sSHIFT TYPE

LA

BO

R P

ER

FO

RM

AN

CE

7-12s

MEANMEAN

One possible conclusion may be that the number of days worked per week is a far better predictor of Labor Performance - as defined by actual hours versus budgeted hours - than the number of hours worked per day. The real importance of daily work hours manifests itself in calculation of actual labor dollars, due to additional overtime premiums that are expended on the extended work day schedules such as 12-hour workdays. The major loss of productivity seems to result from working seven days per week. TA COSTS AND LABOR HOURS

When normalized for 2007 dollars the TAs in sample n2 show strong correlation -as expected- between Total TA Costs and Labor Hours. Total TA Costs here do include equipment, direct and indirect labor as well as material. The related chart displays the tight fit of TAs in size below approx. 300,000 hours. The level of dispersion around the trend line increases significantly above that size. In other words, the ability to predict the cost outcome declines with increasing TA size as measured by labor hours. The ‘All-In Mechanical Rate’ equals $95.30 if all TAs in sample n2 are averaged. This rate however is the result of a slight but continuous increase from a base value of$76.20 for smaller TAs.

8

www.ap-networks.com


Figure (5) – TA Costs and Labor Hours

N.B. Only North American TAs from sample n2 included here (2007 normalized)

In reviewing the All-In Mechanical Rates in detail a more complex picture emerges. For one,

no simple algorithm may be discerned from the data if labor hours are grouped by size. In fact, the emerging pattern -though repeated- does not lend itself to be expressed in functional terms. In the following chart each data point represents a group of TAs of certain size (by actual labor hours).

Figure (6) – All-In Mechanical Rates per TA Size Group

70.00

75.00

80.00

85.00

90.00

95.00

100.00

105.00

110.00

115.00

120.00

<50K

<75k

<100

K

<150

K

<175

K

<200

K

<225

K

<250

K

<300

K

<325

K

<350

K

>375

K

<400

K

>425

K

<450

K47

5K

<500

K

>550

K

<600

K

>600

K

TA SIZE BINS (IN HRS)

AL

L-I

N M

EC

HA

NIC

AL

RA

TE

Cycle 1 Cycle 2 Cycle 3

SUPPORT AT $80

N.B. Only North American TAs from sample n2 included here (2007 normalized)

Scale economies may account for the repeated ‘step-up’ and ‘step-down’ progressions. A

support level exists at approximately $80.00 to which the curve returns in intervals for several specific size categories. It would be outside the charter of this study to determine the real causes for this rate pattern, but suffice it to say that further research needs to be conducted to isolate the critical few elements for creating this phenomenon.

www.ap-networks.com


INDIRECT LABOR AND TOTAL TA HOURS

Intuitively many industry participants may agree that larger TAs necessitate a higher input of planning as well as supervisory type of resources. Empirical data suggests that the proportion of indirect labor grows linearly to about 25-27% and a TA size of 400,000 hours, which then continues logarithmically and settles in on about 30%.

Figure (7) – Indirect and Total TA Hours

Since hourly rates for direct and indirect labor may diverge significantly the preceding

relationship could skew decisions on how to execute TA work. One option may clearly be to split a 400,000 hrs TA into smaller pieces to benefit from fewer required indirect hours and thus less labor costs. These types of relationships need to be addressed by and included in a reasonable trade-off study or model.

C. Modeling Data Relationships

Due to the complex interdependencies and existing collinear ties any empirical model for TA trade-offs still requires certain basic assumptions to fill existing data gaps. Productivity is certainly one of the major concerns during an outage as it impacts not only labor costs but also safety awareness and possibly quality (e.g. rework).

Several studies on the effect of overtime on labor productivity have been used in the past to either dispel or buttress claims of fatigue due to extended schedules. BLS Bulletin 917 is one of the first large scale efforts to measure industrial worker fatigue, but focuses however almost exclusively on the manufacturing environment during World War II [6]. Over the years, other trade groups and individuals, such as NECA, MCAA, Howerton, Thomas, Proctor and Gamble, and several more have either produced their own studies (more relevant to the contracting environment) or revised preceding work such as BLS Bulletin 917. The choice of studies in this simulation model represents no affirmative validation of any particular study’s underlying

10

www.ap-networks.com


applicability, but was made to simply demonstrate a larger point within trade-off economics. TA Managers may introduce other productivity patterns based on company history or industry references. For further discussion on the topic see Dr. Richard Larew [7].

Figure (8) – Published Productivity Studies by Shift Type

The preceding chart includes 6 distinct productivity patterns associated with a specific shift type in addition to the baseline case of 5-8s at 1.0. For this simulation a maximum possible duration of 80 CDs was chosen. Only few TAs in the industry would exceed this duration. The source of selected productivity patterns is NECA and Howerton [8+. The chart includes NECA’s 5-10s and 6-10s productivity lines, which are characterized by a significant initial drop in Period 1 (P1) only to regain some of the lost momentum by the end of P2. Once crested (after 3 weeks) the lines continue their descent through P3 and begin to stabilize approximately in the middle of P4 (after 9 weeks). The short-lived recapturing of momentum during P2 has frequently been explained by the increasing level of adjustment of personnel to the extended work schedule before true fatigue finally erodes such gains and productivity declines become inevitable.

The Howerton curves show largely linear declines through P1, P2, and P3. These curves however provide only continued data through the end of P3 (approximately 4 weeks). Consequently, any subsequent data points shown have been manually extended at the final observed value. Given the strongly linear declines of these curves doubts may be raised about the veracity of this straight-line extension. However, it is similarly doubtful that productivity declines would continue at the same rates unabatedly for example for the 7-12s as it already shows 50% productivity deterioration at the end of P3.

III. MAKING THE CASE

In order to create reasonable and workable scenarios a base case must be established to which all other alternatives can be compared. The base case TA duration is 46 CDs with planned daily hours (direct and immediate supervisor) at 8,000 (with 5-8s) for a total of 264,000 earned value hours (‘true’ scope). The number of earned value hours is important to gauge how many actual hours it would take in other scenarios to earn the same 264,000 hours. The number of planned personnel equals 1,000 and will be kept constant for the other shift scenarios in this

www.ap-networks.com


simulation. In other words, the TA Manager asks, “how many days and how much money does it take or save if I switch to a different work schedule with the same number of men?”

Table (1) – Base Case

A B C

SUM

1 Planned Daily Hours 8,000 264,000

2 Planned Cumulative Hrs 264,000

3 Planned Personnel/CD 1,000 Avg./WD

4 Planned TA Start Date 6-Apr-08

5 Planned TA Finish Date 21-May-08

6 Planned Duration in CDs 46

7 Labor Wage Rate $63.75 $14,386,284

8 Supervision Wage Rate $85.00 $3,258,288

9 Supervision Portion 14.52%

10 Planned Total Labor $17,644,572

11 Per Diem per CD/person $110.00 $5,060,000

12 TOTAL $22,704,572

The reason for choosing a 5-8s-work schedule as the base case is rooted solely in the

availability of productivity studies, which favor it as a baseline (1.0) to measure other shift alternatives against. Clearly, 5-8s as a work schedule is hardly ever chosen for critical TAs in the industry. The applicability of the 5-8s work schedule as baseline in this simulation however is not diminished since all other included (and more frequently occurring) alternatives can still be compared against each other on equal footing as they are expressed in terms of the 5-8s case.

Based on the daily productivity deterioration profiles each study provides, daily earned value hours can be calculated. Since daily productivity rates and actual hours worked (according to shift type) differ for all listed productivity curves, each produces a unique TA duration. The following figure depicts the various TA durations by shift type.

Figure (9) – Calculated Total TA Duration

12

www.ap-networks.com


Working crews an additional 2 hours per day (5-10s) over the base case results in a 2 CD savings. Nevertheless, the first significant time savings occurs, not unexpectedly, with a 6-10s work schedule and 8 CDs of reduced performance time compared to the base case. NECA and Howerton vary by only 1 CD in this scenario. A possibly surprising result is the equal duration for the 6-12s and 7-12s work schedule (33 CDs), from which one may conclude that earlier gains in the 7-12s work schedule –associated with the additional work day- are over time eroded by the more rapidly declining productivity rates as compared to 6-12s. Overall, 7-12s reduce the performance duration by 13 CDs to a total of 33 CDs. The following table records relevant summary data of this simulation. The lost profit opportunity cost in the table is based on the duration in days, plant output capacity for refined product of200,000 bbl/day and refinery margin per barrel of $4.00.

Table (2) – Summary Data by Shift Type

(5-8s) A- (5-10s) B-(6-10s) C-(6-10s) D-(6-12s) E-(7-10s) F-(7-12s)

Duration in CD 46 44 38 37 33 35 33

Labor Cost $17,644,572 $22,583,715 $24,515,261 $23,786,754 $28,584,207 $29,507,873 $36,323,758

Per Diem Costs 5,060,000 4,840,000 4,180,000 4,070,000 3,630,000 3,850,000 3,630,000

Lost Profit Opp. 15,158,000 12,826,000 5,830,000 4,664,000 0 2,332,000 0

Safety Costs 325,537 372,001 428,309 426,448 511,427 502,270 610,270

Total Costs $38,188,109 $40,621,717 $34,953,570 $32,947,203 $32,725,634 $36,192,143 $40,564,028 Total Costs (ex.

Opp.) $23,030,109 $27,795,717 $29,123,570 $28,283,203 $32,725,634 $33,860,143 $40,564,028

Safety Incidents 3.3 3.7 4.3 4.3 5.1 5.0 6.1

Actual Hours 264,000 310,000 320,000 310,000 336,000 350,000 396,000

There is naturally a clear trade-off dynamic between duration (days) and other performance or input variables such as labor hours, premium time, revenue, and safety. Economics however cannot offer one best solution that is easy to choose due to the variety of motivations, preferences, and shifting priorities. For such cases, simulation models are built to approximate expected decision constraints and allow for changing criteria weights. Time is one of the most dominant constraint components within any TA decision-making process.

The value of time within this context is typically determined by two prevailing factors: time-sensitive costs and lost revenue. This model includes per diem costs as the major time-sensitive component, but other costs could easily be added to adjust the model. One possible way to evaluate the efficacy of switching from one work schedule to another is the analysis of marginal costs for each reduced day. For instance, switching from the base case to a 5-10s work schedule reduces overall duration by 2 CDs but adds close to $4.8M to final TA costs (excl. opportunity costs). In other words, each day of schedule reduction was bought for approximately $2.4M in additional costs.

www.ap-networks.com


Figure (10) – Marginal Cost Analysis

N.B. Does not include Opportunity Costs for Lost Revenue

All other work schedules have significantly lower marginal costs for each reduced day, which

range from $583K (6-10s) to $1,348K (7-12s). The base case cost per calendar day equals $500K ($23M/46CDs). From this perspective, work schedules based on 6 days per week seem most economical if opportunity costs are not considered.

Figure (11) – Initial Simulation

N.B. Refinery Margin at $4.00/bbl refined product, daily output at 200,000 bbl and safety costs at $100,000 per event

The very essence of simulation is encapsulated in the “what-if” consideration. What

happens, if we change one or more variables in the model? Unlike the initial marginal cost analysis, this preceding chart now does include opportunity costs, which, if refinery margins are high, may easily eclipse other cost advantages based on duration or work schedule. The 7-12s and 6-12s do not include opportunity costs since both alternatives have produced the shortest work duration with 33CDs. Consequently, all other alternatives, including the base case must be assessed opportunity costs as a result of lost revenue. For purposes of this simulation, ‘Safety’ has been included to demonstrate the ability to model a variety of variables in trade-off economics. Safety incidents have been monetized to manipulate the model accordingly. Refineries can establish their own philosophy on how to quantify or value safety. The underlying function for ‘Safety Incidents’ has been derived from the previous figure on safety and hours

14

www.ap-networks.com


worked per day and used in calculating the number of expected events by shift type in this simulation. Once established the model produces specific results based on input variations. Changing refinery margins (opportunity costs) for example will have a significant effect on cost considerations. The 5-day work schedules become extremely expensive when compared to the other alternatives. In fact, opportunity costs are the major cost element for 5-day work schedules in this scenario.

Figure (12) – Simulation on Refinery Margin a.


During high-demand periods and above-average returns on refined products it is unlikely that such time-sensitive options should be chosen, even though their primary costs are typically lower than those associated with 6-day and 7-day work schedule. The refinery margin would have to be at approximately $5.83/bbl in the above scenario to achieve cost parity between NECA ‘s 5-10s work schedule and Howerton’s 7-12s at $40.9M. During times of over-supply in the market and low margins, time as a resource (or TA duration) becomes less critical.

Figure (13) – Simulation on Refinery Margin AND Safety


The preceding chart demonstrates the impact of opportunity costs as well other elements such as safety. Refinery margin has been set at $2.00/bbl and safety incidents assessed with $500,000 each to emphasize the importance of safety during this particular TA. The 7-12s work

www.ap-networks.com


schedule becomes the most expensive option in this case even though no opportunity costs are included due to its short duration of 33CDs. The 5-10s and 6-12s work schedules now seem cost competitive if safety costs are excluded. In this very case though, the decision could ultimately be made in favor of the 5-10s shift since its costs are lower once safety is considered. This however depends on the value and importance assigned to safety -philosophically- in the particular refinery. This research suggests an accelerating relationship between work intensity (daily hours) and the number of safety incidents.

IV. CONCLUSION

Important questions remain to be answered. The details and circumstances of turnarounds will always be unique therefore requiring adjustments to even the most sophisticated models on a case-by-case basis. Basic dependencies however hold true: what is known as the tenet of ‘triple-constraint’ associated with cost, time, and quality (or alternatively safety, revenue, or scope) describes the economic trade-off conundrum well. In reality, the decision-making process may be much more multi-faceted but will still adhere to the underlying principle of mutually exclusive or at least mutually conflicting choices. Simulation models help at a minimum in demonstrating the major consequences of adjustments to basic performance parameters. Endnotes

[1] “Maximizing Plant Productivity”, Debakey, Samman, Mohammad, Edmundson, and Blanchard, PTQ Revamps & Operations, 2001

[2] “Study measures effect on leading indicators on plant turnarounds”, Bobby Vichich, Oil+Gas Journal, May 21, 2007

[3] “Refining Economics”, Gary, Kaiser, and Handwerk , 2004 [4] Proprietary turnaround database of Asset Performance Networks, LLC, Bethesda, MD [5] “Calculating Loss of Productivity Due to Overtime Using Published Charts – Fact or Fiction”,

The Revay Report, Nov. 2001, Brunies and Emir *6+ “Hours of Work and Output”, BLS Bulletin 917, 1947 [7] “Are any Construction Overtime ‘Studies’ Reliable?”, 1995 AACE Transactions, Dr. Richard

Larew, PE *8+ “Scheduled Overtime Effect on Construction Projects,” Report C-2, The Business Roundtable,

Nov. 1980

Trade Off Economics White Paper

Documents

Transcript of Trade Off Economics White Paper