Development of a risk analysis model for the installation of offshore wind farms

The installation of offshore wind farms represents a major driver of offshore wind energy costs due to the high level of associated risks. Substantial cost reduction potential can be realized if these risks are effectively identified and controlled, preventing project delays and financial damages. Therefore, we propose a new method to identify, analyse and evaluate installation risks for use in project planning. This paper provides an in-depth walkthrough of the steps involved in the development of a simulation-based quantitative risk analysis model. At the example of a turbine installation case study, project plan-related risks are identified and quantified by conducting a literature review and expert interviews. The corresponding risk likelihoods and consequences are modelled with probability distributions to simulate their impact on total project duration. Using Monte Carlo simulation, the distribution of results is statistically evaluated to derive the expected project delay caused by the analysed risks as well as reference values for optimistic and pessimistic scenarios. Regarding the underlying model assumptions, the most significant risks (e.g. supply chain failure) are identified and examined in a sensitivity analysis. The developed model can serve as a basis for developing more reliable schedule estimations and contribute to minimizing installation delays and costs.


Introduction
Achieving the ambitious goals of expanding offshore wind energy is only possible through substantial investments into the installation of new offshore wind farms (OWFs).Transport and installation (T&I) processes of OWFs are associated with high costs and risks, primarily because of the challenging environmental conditions at open sea.A wide variety of further project risks adds to the increased level of uncertainty for offshore installation projects, leading to delayed project completions and implicit cost overruns.Due to ongoing developments of offshore wind turbine (OWT) upsizing and greater distances to shore, installation risks are becoming more and more relevant.Minimizing these risks would significantly contribute to lowering the costs for OWF installation, reduce capital expenditures and ultimately decrease the levelized cost of offshore wind energy (LCOE) [1][2][3].Based on this problem definition, the first research question can be formulated: Which are the most relevant risks associated with the installation of an OWF and what is their influence on the project duration?
Numerous publications and industry standards call for appropriate methods to identify, analyse, and evaluate project risks that cause a schedule delay within the installation phase.These risks can then be considered in project planning, developing more reliable project schedules with a lower chance of overruns.Moreover, identified risks can be mitigated effectively by implementing adequate risk protection measures that reduce the risk's likelihood or impact on the project duration [4][5][6][7].The quantification of risks has already been addressed in various studies and models of offshore processes, originally with a limited application on the operation and maintenance (O&M) phase of OWF projects [8,9].More recently, simulation models as well as ready-to-use tools have also been developed for the T&I phase of OWFs.Yet, these models only consider weather-related risks by applying simulated weather conditions to the operational limits of each installation activity [10][11][12].Existing research reveals a gap in the assessment of risks during the installation of OWFs beyond weather conditions.Further installation risks originating from operational or logistical uncertainties, for example, have so far only been identified in qualitative studies or for aggregated project phases [13].Singular risk events, that may occur during specific installation activities, have not yet been appropriately represented in simulation models for an OWF installation.However, an analysis on the activity level is needed in order to accurately estimate and assess their influence on project duration and costs.While weather-related risks are a prominent risk factor, they are explicitly excluded in this model in order to address the identified research gap as a complementary approach to weather simulation models.This leads to the second research question: How can project plan-related installation risks, including their likelihoods and consequences, be appropriately modelled and considered in project planning?
Derived from these questions, the objective of the work presented in this paper is to develop a simulation-based quantitative risk analysis model to assess the impact of identified risks on the overall project schedule.For this purpose, section 2 provides characteristics of OWF installations regarding project and risk management before the methodology of the developed risk analysis model is explained in section 3. Section 4 guides through the identification and modelling of installation activities and associated risks as well as the validation of the simulation model.The results of the conducted simulation are discussed and evaluated in section 5 prior to summarizing the findings of this work and giving an outlook for further research in section 6.

Theoretical background on offshore installations
The installation of OWFs can be classified as a construction project with special considerations and additional challenges associated with its offshore nature.The project is characterized by a stable scope with requirements being clearly defined upfront [14].The involved components have long lead times in production and acquisition or are even engineered to order.Plus, the investment and financing required upfront is immensely high.Therefore, requirements and specifications are fixed from the beginning, and later changes are not anticipated [15].Building on that, a detailed planning of schedule and budget is carried out during project planning to cope with the increased difficulties of offshore logistics during T&I.This includes onshore delivery, assembly in port, chartering and mobilization of appropriate vessels, transit to site and finally offshore installation of all OWF components [16].
There is an increased level of risks associated with offshore logistics due to uncertain weather conditions and vessel availability.Thus, contingencies for schedule and cost overruns must be considered and planned before they actually occur.Moreover, there are rigorous health and safety regulations for offshore T&I processes, so the compliance with standards and recommended practices must be assured and planned [4,6].Furthermore, the project organization in OWF construction is inherently large and interdisciplinary.Apart from the multitude of companies involved in the supply chain of OWF components, there are several political, regulatory, environmental and societal actors that are relevant for the permitting process of the OWF [15].Due to the high levels of complexity and uncertainty, upfront planning is essential for successful and efficient project progression.The high level of risk can be lowered to a reasonably practical level by early and extensive risk planning and management, which minimizes or mitigates relevant risks and develops contingency capacities [17].
Several national and international institutions offer guidance and standards for the risk management of offshore installations and operations.ISO 29400:2020-05 [6], for instance, prescribes standardized processes for risk management, which shall also be examined in the approval process of an OWF.To demonstrate compliance with regulations, potential hazards are to be identified, analysed and evaluated using appropriate and recognized techniques [4].Based on an iterative risk assessment, identified hazards must be prevented or reduced to a residual risk level that meets the prescribed safety objectives [5].Moreover, contingency plans and emergency procedures shall be developed for damage containment in case of an incident [6].

Methodology
The methodology of this work is based on the standardized risk management process described in ISO 31000:2018-02 [18].Figure 1 shows the five-step approach of the suggested risk analysis model that enables a realistic and reliable project schedule estimation for the installation of an OWF.At first, the scope and limitations of the risk assessment at hand are established by selecting a representative sample of activities that will be examined for relevant risks.To be able to conduct an indepth risk assessment, the scope of this work is limited to the turbine installation phase (foundations are already installed).Considered as the most critical stage in the whole OWF installation process, the installation of OWTs contains a wide range of risk factors including technology, logistics, and planning risks.Moreover, even an ideal turbine installation phase spans a long period of time, so schedule delays caused by these risks may have a serious impact on project progress and completion [19].Thus, this selection of installation activities should serve exemplary as a proof of concept for the developed model.
The detailed activities are taken from a benchmarking project organized in 2020 by the National Renewable Energy Laboratory (NREL).In a case study featuring the installation of a representative OWF located in the British North Sea, actors from industry and research have compared their balanceof-system cost models for assessing weather downtimes during the installation process.The provided inputs contain standard installation processes including durations and resource requirements exemplary for modern OWFs.Thus, this reference project is considered suitable to develop a complete and realistic model of the installation process of an OWF [20].
For risk identification, a systematic review of relevant literature and existing risk assessments in the offshore wind industry is carried out.These sources are screened for conformities on construction risks that occur during the turbine installation phase and on general project risks that have a direct impact on the duration of installation.Additionally, several interviews with industry experts are conducted to validate that the most relevant risks for the selected installation activities have been considered.These expert interviews also serve to receive industry knowledge concerning the probabilities and consequences of the risks, because reliable data thereof is only scarcely available in official publications.Where possible, published data retrieved from standards, accident statistics or recognized methods like FMEA or Fault Tree Analysis is applied.In case no data is available, the industry representatives are asked to provide expert judgment or a subjective estimation of the needed characteristic.

Focus on OWT installation
The intermediate result from this risk identification process is then comprised of a compact risk register featuring a qualitative description and quantitative values for likelihood and consequences for each risk.Note that these variables depend heavily on the designated installation processes, utilized assets and equipment as well as OWF site criteria.From these input parameters, a quantitative model of the reference project is developed that can be analysed regarding durations and schedule delays of single activities and the complete project.The risk analysis of the developed model is then performed in a Monte Carlo simulation (MCS) using the software tool @RISK from Palisade.The quantified risk characteristics are represented as input variables with defined probability distributions according to the level of uncertainty.During the simulation, random sample values are taken out of these to calculate a discrete output value.This process is repeated over 10,000 iterations to derive a stochastically valid distribution of the obtained project durations [17,21].In a last step, the results of the MCS are then statistically evaluated regarding the range of total project duration and its sensitivity to the occurrence of each risk.Based on these outcomes, reliable schedule estimates and a significance ranking of the analysed risks can be created.

The developed risk analysis model
This section guides through the development stages of the risk analysis model for OWF installations as shown in Figure 1.Subsequently, the simulation model is validated by literature and industry experts.

Selection of installation activities
The mentioned case study provides project, component and resource parameters required to develop a complete and realistic model of the installation process of the OWF.Each stage of the installation process includes a project plan with all required activities and their associated durations.In the work at hand, the respective project plan for the turbine installation phase is selected and further investigated.The process steps featured in Table 1 describe the installation of 50 OWTs of 8 MW capacity (400 MW in total) to be built on monopile foundations by one wind turbine installation vessel (WTIV) operated from one installation port [11].Using single component installation, all OWTs and their subcomponents are to be transported and installed in the same manner, which constitutes the repetitive character of the turbine installation process.Putting the depicted activities into a linear sequence and adding up their respective default durations, a baseline project duration of 156 days is reached.However, this value only exists in an ideal scenario which does not consider uncertainty.In reality, diverse risks may lead to varying activity durations, delayed processes or project downtimes.To integrate the effect of risks on the duration of activities and the whole project, they need to be identified, analysed, and evaluated.

Identification of relevant risks
In the next step of the risk assessment process, the identification of risks related to the selected activities needs to be carried out.Since the scope of this work is limited to turbine installation, only risks that directly impact the presented processes shall be considered.The goal of this work is to analyse risks and their impacts for each specified activity of the installation process.This allows for more detailed assessment of singular risks, random events and possible systematic uncertainties associated with the project plan.Reviewing sources from theory and practice, the most relevant risks for the case study are discovered and validated against each other.Quantified data for risk frequencies and likelihoods are researched among accident statistics, incident reports or estimated through expert interviews.For risk consequences, optimistic, most likely, and pessimistic scenarios are explored to differentiate between possible severities of impact.In total, 15 individual risks are identified and validated as the most relevant risks for the OWT installation process at hand.In Table 2, these risks and their respective probabilities and consequences are listed in order of occurrence within the reference project plan.Apart from these concrete risk events there are also risks that originate in previous project phases but have a direct effect on the duration of the installation phase.Such risks mostly result from uncertainty or errors during the early phases of the project, when exact specifications and requirements are set for the complete project.Literature and business experts alike mention planning risks, consenting risks, design and engineering risks and poor project management as further reasons for delayed installation processes [22][23][24].These risks are aggregated under the term systematic uncertainty because they can't be mapped to specific installation activities.

Modelling of risk occurrences and consequences
The risk analysis model is structured in two basic input elements.The first element resembles the reference project plan sequence under the influence of systematic uncertainties regarding the baseline durations of the 20 selected activities.It is impossible to predict which exact activities are affected by these uncertainties.Furthermore, discrete consequences are difficult to quantify in terms of absolute schedule delays for individual operations.Therefore, systematic uncertainty is modelled by varying the baseline duration of each basic sequence activity.The minimal impact is assumed to be a shortage of duration by 10 %, referring to an overestimated baseline duration and learning effects.Most likely, the originally defined duration is accurate and no further uncertainties come true.Then, no alteration to the baseline duration is expected.If multiple of the mentioned risks occur at once, like missing process steps, underestimated task durations and later changes to processes, a maximum prolongation of 50 % is added to the original baseline duration.For each activity, the actual duration is determined randomly from a range of values modelled by a Program Evaluation and Review Technique (PERT) distribution.
The PERT distribution is a special form of probability distribution which requires the minimal, most likely, and maximal values of the duration as reference points and is especially suited for modelling expert judgments.Its probability density function prioritizes the value defined as most likely and gives less weight to the boundary values than a uniform or triangular distribution [25][26][27].The simulated activity durations are multiplied by the number of times the respective activities are repeated over the whole installation process.As a result, the obtained total duration includes the systematic uncertainty for all activities of the OWT installation project.
The second element models the concrete risk events for each activity and simulates their individual impacts.In order to analyse their impact on project duration, firstly their occurrence and secondly their consequences must be simulated.A quantitative probability of occurrence is assigned to each risk event as specified earlier.Based on the respective probability, the model simulates for each risk event randomly whether it occurs or not.Risk occurrence is simulated by a Bernoulli distribution, which either returns a value of one or zero (true or false).The value one stands for the risk occurring and is assigned the specific risk probability [28].At the example of the supply chain failure risk, Figure 2.a demonstrates how risk occurrences are simulated in the developed model.Based on a study regarding a similar OWT installation project, the probability of material shortage is set at 6 % [29].This probability is assigned to the value one, standing for the risk occurring.On the contrary, the value zero, which represents the case in which the risk does not take place, bears the complementary probability of 94 %.No values other than one or zero can be attained by this distribution.In case the risk does occur, its impact on the project duration is simulated in a second step.The consequence is again modelled by a modified PERT distribution using the minimal, most likely, and maximal values of the resulting schedule delay.Again, the example of the supply chain failure serves as visualization for the simulation of risk impact in Figure 2.b.The consequences of the risk are at least one day of waiting time for components and most likely three days of delay, represented by the maximum point of the graph.Reviewed risk assessments also mention the risk of supplier insolvency (a) which may cause the supply chain failure.While the likelihood of such a scenario is assumed to be low, the consequences can be more severe.Thus, the maximum delay associated with supply chain failure is estimated to be up to half a year, when substitutional suppliers must be found.The minimal and maximal impacts mark the boundaries of the probability distribution, although the figure is cut off at 20 days for better visibility.The actual risk impact is then simulated from within the described range of values.
Analogously, the risk occurrence and consequence are resimulated for each project activity mapped to the respective risk event to maintain independence from previous occurrences.After simulating each risk event individually, the simulated consequences in case of occurrence are aggregated to obtain the overall influence of one particular risk on the project schedule.Subsequently, the aggregated impacts of all occurred risk events are added to the simulated project duration regarding systematic uncertainty as indicated before.The simulation model output is then the total project duration considering systematic uncertainty and individual risk events for all activities of the OWT installation project.This process is repeated for 10,000 iterations within the MCS.The overall result is then expressed as a distribution of outcomes consisting of all obtained output values for the total project duration.

Validation of the developed model
The developed risk analysis model has been extensively validated using various approaches and techniques suggested by Sargent [30].To achieve internal validity, multiple tests were carried out.extreme conditions test, where all risk probabilities were set to zero, resulted in the model outcome equalling the original baseline duration of the reference OWT installation project.In a degenerate test, increased probability or consequence inputs lead to increased output values.In the MCS, stochastic errors are already minimized by the high number of trials.Nevertheless, several simulation runs with 100,000 iterations each were conducted to confirm that outcomes are replicable.
Apart from internal validation, simulation models can be externally validated against other models or the real system [31].The installation activities examined in this work were purposely selected from the NREL benchmarking project to enable cross-validation of the model inputs and scheduling logic [20].The computation of the default installation times per turbine and for the whole project without consideration of risks was validated this way.Throughout the different stages of model development, face validity was confirmed by conducting interviews with independent experts from the offshore wind industry.They approved the logical structure of the model and its applicability to existing and future OWF installation projects.Moreover, the modelling of relevant risks including their occurrences and consequences was demonstrated and considered appropriate and reasonable.

Discussion of results
When running the MCS of the model, the total duration of the OWT installation project is calculated for each iteration.Since some outcomes occur more frequently than others among 10,000 trials, the result of the simulation forms a distribution itself, which is plotted in Figure 3.In all iterations computed in the MCS, the simulated total project duration exceeds the original baseline of 156 days, indicating that a schedule delay is caused by the occurrence of the considered installation risks.The expected duration derived from the distribution of results amounts to 186 days (median/P50), which is equivalent to an expected schedule delay of one month or a relative overrun of 19 %.
However, no false certainty should arise that the expected value predicts the exact duration of the actual project.Instead, it is more appropriate to understand the simulation results as a range of possible outcomes.For instance, a striking observation is that 80 % of all simulated outputs fall into a range of 30 days, between 174 and 204 days.This interval implies that the probability of the project taking less than 174 days amounts to only 10 %, just as much as the probability that the project duration exceeds 204 days.The boundaries of this 80 % confidence interval represent the P10 and P90 values, respectively.It is also possible to broaden or narrow the interval, depending on whether more assurance or more precision on the range is desired.In risk management, especially the upper percentiles of the distribution are of high relevance, because risk treatment might differ between low and high impact scenarios.While project durations of up to 186 days (P50) can be considered in schedule planning by placing buffers between subsequent installation phases, it may no longer be reasonably practicable to prepare and plan for project durations over 273 days (P99), let alone the worst-case scenario of 425 days.Thus, project owners rather budget for expectable ranges of project duration and resort to other risk protection measures to hedge against high impact scenarios.This usually includes contingency plans and insurances against the worst ten, five, or one percent of possible outcomes (P90, P95 or P99), depending on risk aversion.Apart from the mere output durations, analysing the input parameters and their stepwise computation helps to understand where the variance of outcomes originates from.The three most significant risks in relation to the underlying model assumptions have been identified as supply chain failure, blade attachment failure, and systematic uncertainty.They share a common pattern regarding their influence on the total project duration.To visualize this in an example, Figure 4 shows the range of simulated impacts of the top ranked risk, supply chain failure.The most evident observation of the diagram is that most delays take on a value of zero days, meaning that no schedule delay is added to the project duration.This implies that the risk has not occurred at all, since its minimal impact in case of occurrence would be one day of downtime.In fact, in 45 % of all 10,000 simulation runs, the supply chain failure risk has not arisen over the course of the project.However, this also means that in 55 % of all iterations, the risk has occurred at least once, leading to a schedule delay.In these cases, the most frequently obtained value lies around three days of delay, which corresponds to the predefined "most likely" consequence.While the mean value for the simulated schedule delay equals 5 days, the longest downtime observed amounts to 55 days, which highlights the potential of this risk to affect the overall project duration.
Most of the analysed risks resemble a distribution of impacts similar to the one described above, partially to a more extreme extent.They share a most frequently obtained schedule delay of zero days, meaning that the risk most likely doesn't occur, and a highly skewed graph which features few very high values.This is due to their low probability of occurrence combined with very high impacts on the project schedule in case of happening.Thus, these risks predominantly increase the total project duration simulated in the model.General observations are that singular risk events occur rarely but can have high impacts sporadically.These consequences are amplified significantly when the corresponding risk event is associated with repetitive activities within the project plan, which is inherent in the case of OWF installations.

Conclusion
The developed risk analysis model represents a proof of concept of a method to identify, analyse and evaluate risks that are relevant during offshore wind farm installation.While these depend on the specific requirements and assumptions for each installation project, in the presented case the most significant risks were found to be supply chain failure, blade attachment failure and systematic uncertainty.The created model has demonstrated its suitability to represent and simulate installation risks, which can serve as a basis for developing more realistic and reliable schedule estimations during project planning.Project owners can use it to support decision-making regarding the selection and implementation of appropriate risk responses, prioritizing on the most important risks discovered by the model.This can contribute to minimizing installation delays and costs associated with delayed project completion.With the exploitation of cost reduction potentials, the cost of offshore wind energy will eventually decline, making the expansion of renewable energy generation more attainable.
Concerning the outlook for further research activities, four basic steps of improving or further developing the model can be identified, based on the experienced limitations of the current model.Firstly, the developed model is a highly simplified representation of the real system.It is based on assumptions that are subject to uncertainty themselves, which is especially true for the quantification of risk probability and consequences.More reliable assumptions or objective data would improve the risk modelling and the accuracy and quality of results.Thus, additional research using quantitative methods is needed to increase the available data set.Secondly, the current state of the model is limited to the turbine installation phase of one reference project.Further research is necessary to expand the model's applicability to other installation phases and OWF installation projects.Then, additional risk identification is required to consider individual OWF assets and changing environmental conditions of specific wind farm sites, which will become increasingly relevant considering the future developments of turbine upsizing, larger distances to shore, deeper waters, and floating substructures.Thirdly, weather-related risks remain crucial for offshore logistics and cannot be neglected under real circumstances.Therefore, a combination of approaches should be aspired to achieve a holistic risk assessment model able to represent all risks associated with OWF installation.This can be done by simulating weather conditions within the presented model or integrating this model into an existing weather simulation tool.Finally, additional research can aim towards establishing an economic model to examine the relationship between the simulated project durations and resulting installation costs.

Figure 1 .
Figure 1.Five-step approach of the suggested risk analysis model.
activity durations and requirements taken from an industry benchmark project Systematic literature review norms, standards, accident statistics or recognized quantitative methods Expert interviews industry representatives provide expert judgment or a subjective estimation Risk register qualitative description and quantitative values for likelihood and consequences developed for each risk Quantitative risk analysis risk occurences and schedule impacts simulated from inputs with defined probability distributions according to the level of uncertainty Research question 1 range of total project duration and its sensitivity to each risk Research question 2 practical and theoretical implications of the model

Figure 3 .
Figure 3. Distribution of results for the total project duration.

Figure 4 .
Figure 4. Simulated schedule delay of the "Supply chain failure" risk.

Table 1 .
Selected activities with associated durations and repetitions.

Table 2 .
Identified risks with quantified probabilities and consequences.