ORE Catapult’s internal O&M simulation model COMPASS has been used for this study. The COMPASS tool is a Python-based model which is interfaced with Microsoft Excel for the input and output files. The COMPASS tool is intended primarily for obtaining reliable estimates of OPEX of offshore renewable energy farms for informing internal cost modelling projects. It takes key characteristics of an offshore wind farm, such as number of turbines, site capacity etc., and applies a series of O&M activities. Figure  shows a high-level structure of the COMPASS tool operating in a time-domain mode.

The “bottom-up” O&M reference activities are considered with respect to the Reference Designation System for Power Plants (RDS-PP) methodology of component assignment down to the subsystem designation. O&M activities are populated within these subsystems for wind turbine components. Activities for the wind farm assets are assigned at a system level (e.g. export cables) or where a clear designation could not be determined (e.g. seabed survey) at a wind farm level.

Each maintenance activity can either be planned (i.e. recurring scheduled measures which are organized in advance, along with the resources required to carry out an O&M task) or unplanned (i.e. the result of a turbine alarm or component failure) and has been assigned a rate (i.e. times per year per turbine for scheduled maintenance, or failures per year per turbine for corrective maintenance). These rates have been defined from a combination of publicly available academic sources (particularly and ) as well as from the experience of the team at ORE Catapult. Each turbine in COMPASS is split into subsystems, e.g. generator system or control and protection system. Each subsystem is then split into components. There is a set of planned and unplanned activities applied on turbine, subsystem and component levels. Type of activities, their rates (with exception of activities for which failure data were publicly available), their duration, personnel and vessel requirement, and urgency have been defined by the ORE Catapult team and are not available to the readers. The only exception is activities for which characteristics such as failure rate (for unplanned) or frequency (for planned) were publicly available. For the purposes of COMPASS, activities informed by the evaluation of condition monitoring systems or other relevant component data were not separately distinguished.

At each time step in the time-domain mode, the COMPASS tool checks whether maintenance is required and then checks for availability of personnel and vessels and suitable weather conditions (wave height and wind speed). In the case of unplanned maintenance, the Monte Carlo method is applied in order to model failures. The essence of this method is to generate a random number between 0 and 1 at each time step and based on this number determine whether the component has failed by comparing it to the reliability of the component (i.e. the probability of not failing, “R”). If this random number is below R, then the component has failed; if it is above, the component retains its operational state.

At the end of the simulation, the COMPASS tool sums up all the costs from all time steps associated with O&M activities and adds fixed costs (e.g. IT and onshore base costs). In addition to all costs, the tool calculates total energy produced by the project and calculates both time-based and energy-based availabilities, where time-based availability is the proportion of time that a farm was operational and energy-based availability is the proportion of energy that was generated out of the theoretical energy production (if the entire project was operating at full capacity 100 % of the time). In addition to these conventional outputs, the COMPASS tool calculates the overall risk rating associated with each activity and measures carbon emissions emitted by operation vessels which are summed up in the outputs of a simulation.

Each activity in COMPASS has a list of required personnel to perform that activity, which was defined by a user in the inputs. This personnel can be of different type, e.g. technician or a specialist. When an activity is due, a contractor gets created if it is not already available in the port. Each contractor gets assigned to a port, maintenance base or SOV. In case an activity requires a SOV, then only the contractors from that SOV will be considered to perform this activity, and if there is no personnel of the required type, a new contractor will be generated if it fits into the SOV accommodation. When SOV accommodation is full and there is no personnel of the required type, the activity will not be completed. When a SOV returns to port, it empties its accommodation, and the process of creating new contractors starts from scratch according to the activities that are due next. This way the code can guide a farm operator on the optimal type and the number of personnel to put on a SOV.

When activities accumulate they get scheduled by the software tool according to the urgency. If two assets have activities due with the same urgency, then the asset with more activities due is prioritized. When several activities are due on the same asset and have the same vessel requirement, they will be merged together until the vessel capacity (in terms of personnel) is reached.

When the task is complete and a personnel becomes available, it can either move to the next task or be picked up by a suitable vessel and return to the place where it was generated (SOV, port or OMB). Each vessel will pick up as many personnel as its capacity allows from the nearest assets and will update the carbon emissions accordingly.

For the duration of the SOV contract period, it is expected that there will be a fixed regular fee paid monthly or annually to the vessel owner. Contract duration varies from project to project, but typically it lasts several years. While the cost for the SOV will be distributed over the years, the fixed cost associated with building an OMB will occur before the farm goes live. Because of this difference in timing of the cash flow, net present value (NPV) needs to be taken into consideration using Eq. (), where Ct is the net cash flow at time t, r is the discount rate and t is the time of the cash flow (measured in years in this study). 1NPV=Ct(1+r)t

The hypothetical floating wind farm used for this study is located off the northeast (NE) coast of Scotland and consists of sixty-six 15 MW semi-submersible floating wind turbines located with an approximate distance of 2 km from each other (in latitudinal and longitudinal directions). The site is located within the boundaries of the NE8 seabed leasing area under the Scottish government's “sectoral marine plan for offshore wind energy” (“ScotWind”). This was highlighted as a site suitable for floating wind in the ORE Catapult Floating Wind Centre of Excellence’s cost reduction report . Characteristics of the site are summarized in Table , and the approximate farm layout is shown in Figs.  and . This is not intended to replicate how the real wind farm layout in that area will look, as no environmental measurements or wake assessments have been performed. However, it is sufficient for O&M simulation.

It was assumed the capacity of each offshore substation will be 500 MW. Therefore, two substations would be required for this site. These would be connected to the onshore substation in Peterhead by two export cables. Fraserburgh and Peterhead are the closest O&M ports to the considered wind farm; however, Peterhead has the capacity to play a role in assembly and manufacturing and is therefore selected as the main O&M port for this farm. It was assumed that all repairs and maintenance in this farm are to be carried out offshore. The use of CTVs may be unfeasible for this site, given that the vessels would have to travel approximately 2 h to get to the farm.

One of the inputs required to run COMPASS is weather data. There are two weather datasets that were used for this study: ERA5 and ERA-20C. Both are open-source global reanalysis data. ERA5 provides hourly data running until 2019, which can be retrieved from the Copernicus website , which gets updated yearly based on new observations. ERA-20C data were retrieved from . It is only available until 2010 and provides data for every 3 h interval. ERA5 data however have lower spatial resolution than the ERA-20C data; the respective resolution is 0.5 and 0.125 in both longitude and latitude. COMPASS uses 1 h intervals in its time-domain mode; therefore ERA-20C data were linearly interpolated to fill in 2 h gaps. Figure  shows the comparison between the two datasets. The data in the Fig.  were averaged over 25 years, and the 1990–2015 year span was used with ERA5 data and the 1985–2010 year span with ERA-20C. It can be seen that peaks and troughs of the two datasets align well even though two datasets were retrieved for different coordinates. ERA-20C tends to model lower wave height, resulting in 1.82 m mean wave height compared to 1.95 m for ERA5. ERA5 mean wave height matches better with other sources for the same location, and with how climate is changing it is important to have up-to-date data to model accurate results. Nevertheless, both datasets were used in this study to see how the choice of data affects the outputs.

Figure  shows the number of opportunities in January to finish an activity. Currently tasks in COMPASS are not interruptable, meaning that the software will look for a 12 h weather window if a 12 h long activity occurs; however making that change may increase the availability of the farm due to increased number of weather windows. The effect of energy availability on the OMB strategy results is demonstrated in Sect. .

There is a lack of actual reliability data from floating wind turbines due, in part, to limited deployment to date. Therefore, most failure rates are inferred from the oil and gas industry, reliability data from ships, and other research areas. There have also been studies estimating the floating wind component failure rates using Bayesian networks . The future-generation floating turbines are likely to be hybrid moorings having a chain-synthetic mooring-chain configuration for several reasons including reduced loads, better fatigue life and durability , reduced corrosion, and easier installation and maintenance due to the lighter weight of these moorings. The synthetic mooring failure rate was estimated from the DTOcean+ project and was combined from the “Polyester rope”, two “Connectors” and “Other” components' failure rates, assuming that the chain part of the mooring does not contribute to its failure rate (as a large proportion of the chain will lie on the seabed where the tension is significantly lower than in the hanging section of the mooring) . This resulted in the failure rate presented in Table , which is lower than that for a chain mooring. This was expected as synthetic ropes are resistant to corrosion and have a much greater fatigue life (chain mooring failure rate is 0.0025–0.00378; ). The anchor failure rate was taken from a reliability study of drag embedment anchors , which are most commonly used for semi-submersible platforms. Structural damage frequency was assumed to be the same as it is in the oil industry data for mobile platforms . The array cable failure rate was updated according to the latest research and the distance between turbines . Failure data for dynamic cables do not yet exist, and, whilst cable manufacturers claim the fatigue life of dynamic cables is over 25 years, there are failures which may occur accidentally due to collision, harsh weather and damage during installation and manufacturing. It was assumed that the dynamic cable failure rate is twice as high as for the array cable because of the harsher environment it is susceptible to.

In four out of the five components listed in Table , a complete replacement is assumed to be required after a failure occurs. The replacement costs are listed in Table  and are assumed to be the same as the original component costs. Full replacement is assumed for all dynamic cables and array cables in case of a failure. What makes it unfeasible to repair a section of a cable instead of a full replacement is the complex and time-consuming inspections of a failed cable to identify the failure point as well as long cure hours to connect the two ends of a cable and the risk of further damage during the repair process. The most common cable connection is the dry connection, which means a cable would be required to be lifted out of the sea for repair. The risks and the costs associated with the downtime required for array cable section repair outweigh the cost savings from repairing just this section. However in the case of an export cable, which is much longer, it can be cheaper to repair a part of the export cable than to replace the entire cable. Costs of anchors may range up to GBP 225 000 . However the drag-embedment anchors are usually the cheapest. The price for the drag-embedment anchor is assumed to be GBP 60 000 according to the Carbon Trust report , which is similar to what is given in the DTOcean report of GBP 67 900. Prices for full chain moorings according to the DTOcean vary between GBP 126 and GBP 233 and are GBP 600 m-1 according to the Carbon Trust report. According to the Carbon Trust the cost of the hybrid mooring would be GBP 2200 m-1. However this cost is likely to vary depending on the type of plastics used in the mooring and its width, which is likely to vary depending on the dynamics of the sea and water depth. Because a semi-submersible floating foundation is assumed, the length of the mooring is considered to be 6 times the water depth. The cost of replacing a dynamic cable was calculated using the coefficients provided in the Corewind report for a 33 kV dynamic cable resulting in GBP 385 m-1 assuming the length of the dynamic cable is (2× water depth (m) × 2.6). The maximum cable rating is 33 kV, for which the cost coefficients were provided, and is currently a common rating for offshore wind farm cables. However in the future we might see more 66 kV cables, which are more expensive but lead to the reduction of the cost of energy .

The annual survey is assumed to be carried out with an observation remotely operated vehicle (ROV) to visually check the condition of all moorings. For a special survey (i.e. once every 5 years) it is recommended where possible to raise moorings to the surface for a more detailed inspection . It is assumed that half of the special surveys will be carried out with a vessel onto which a mooring will be raised, and the other half with an observation ROV (i.e. subsea). Periodic inspections of moorings may involve checking an angle of the catenary mooring to estimate the changes in tension, which can also be done by an observation ROV. Anchors are expected to be inspected together with the moorings, and anchor failure rates are very low (Table ) and so do not require any additional planned inspection.

Dynamic cables require regular (every couple of years) visual inspection of bending stiffeners and hang-offs, any hardware attached to the cable free span, buoyancy modules and cables themselves (for marine growth), protection sleeves, and transition joints . All dynamic cable inspections are assumed to be done with an observation ROV to reduce the deployment of divers and decrease health and safety risks, because water depth in the site area reaches 100 m in some locations. Marine growth removal every 5 years was added as a regular activity to remove any additional weight which may increase the tension in hanging cables, especially in the top 10–15 m of the cable.

For the first strategy one SOV was used with one daughter craft and a capacity to accommodate 40 technicians. It was tested with additional simulations so that increasing the number of SOVs does not have a significant effect on the availability but can increase costs significantly. Depending on the dynamic positioning technology and the gangway technology, the limits for technician transfer can vary between 2.5 and 4.5 m. However 3–3.5 m is the most common limit . Some SOVs may have more daughter crafts; however for this study a medium-sized SOV was assumed based on the North Star SOV. For the second strategy a fleet with three CTVs was assumed. Increasing the fleet size did not have any significant effect on the availability of the farm. CTVs can be different in size and design and vary in their way of attachment to the turbine. This makes it hard to come up with one representative value for the capacity and wave limits. Although next-generation CTVs may be bigger in size and be able to sit 24 personnel, most common CTVs have a capacity to transit 12 technicians. CTVs which could be installed on the OMB are expected to be smaller, and hence the capacity of 12 was selected. Current-generation CTVs can operate in 1.2–2.5 m waves ; however according to experience due to sea sickness and safety concerns as well as steeper waves far offshore, most CTVs travel in 1.5–1.75 m waves. Future-generation CTVs promise to have better motion-compensating systems and designs to comfortably transfer personnel in 2.5 m waves. Due to this variability of wave limits, two options were selected for this study with current-generation CTVs and future-generation CTVs. In total six scenarios were simulated in this study that are summarized in Tables  and .

The fact that the Monte Carlo method is used in the software for modelling random failures means that each simulation will produce a different set of random numbers resulting in varying outputs from simulation to simulation. In order to get a good estimate of the range within which OPEX and other KPIs will lie, simulations need to run multiple times. The higher the number of simulations, the closer the output gets to the true mean of all KPIs. The level of confidence in results is determined by a convergence study. To perform the convergence study the 95 % confidence level was selected. The 95 % confidence interval bounds can be calculated using Student's t distribution, a type of distribution that is similar to normal distribution but is commonly used when a number of samples (simulations in this case) is low (<30). Confidence interval was calculated using Eq. (): 2CI95%=μ±SE, where SE is the standard error calculated using Eq. (). 3SE=t95%,n-1σnn

In Eq. (), μ is the mean value of the sample, t is the confidence level value which varies with the confidence level and number of samples (simulations in this case), σ is the standard deviation of the sample population, and n is the sample size. The value of t can be calculated numerically; however in this study the CONFIDENCE.T() function was used in Excel to calculate the 95 % confidence interval using built-in t values.

The convergence study presented in Fig.  shows how percentage error of the OPEX result mean converges with an increasing number of simulations. The error in the graph represents the proportion of SE of the mean of simulations. A larger number of simulations would lead to more accurate results. However they would require more computational time, especially because convergence slows down with the number of simulations. The standard error of 2 % of the mean was considered acceptable in this study. It can be seen that running 20 simulations is enough to bring the error below 2 %.

Table  summarizes the six types of simulations, each of which were run 20 times to minimize the error in the mean outputs. The table shows the results for the total OPEX of these simulations, which includes fixed costs, the costs of hiring the vessels and personnel costs as well as the average time and energy availability resulting from each scenario over all years. The cost associated with building the OMBs are not included in this table in order to compare OPEX only. The percentage error presented in the table was based on the convergence study; i.e. it is an error calculated using Eq. () based on the results of 20 simulations. Because fixed long-term contracts are assumed for SOVs and CTVs, the uncertainty in results arises mostly from the frequency and costs of maintenance activities generated with a Monte Carlo method. Therefore, when considering the difference between two strategies, the error effect on that difference will be small because, for example, if an upper bound is taken for the OMB scenarios, the upper bound should also be taken for the SOV scenarios. Previous studies (listed in Table ) have been used to benchmark the results of this study. Results lie between the NREL minimum and maximum estimations but are much lower than other estimations. Compared to the estimations presented in Table , COMPASS outputs are produced using much more detailed inputs and simulations. Cases with a SOV resulted in the highest availability due to higher weather limit, but drawback of this strategy is the higher OPEX due to significantly higher costs of hiring these vessels compared to CTVs. In can be seen in Table  and later in Fig.  that the 0.13 m difference in wave means of two weather datasets ERA5 and ERA-20C affects the difference between results, making the cases with ERA5 data more expensive. This difference is to do with the fact that vessels (e.g. ROV support vessel or jack-up) return to port more often with ERA5 data than with ERA-20C due to slightly worse sea conditions. However further improvements to the software should reduce that difference. There is also an effect of having lower time resolution in ERA-20C data, which is linearly interpolated to model hourly data. ERA5 is hence more restrictive to the vessel logistics because it allows for fluctuations in wave height within each 3 h time span. Energy availability however varies less than time availability between scenarios with the same logistical setup and different weather datasets. This is to do to the fact that ERA-20C predicts not only lower waves in that location but also lower winds. It should be noticed however that time availability results are more sensitive to switching from one dataset to another for cases with lower wave limits for vessels.

NPV formula shown in Eq. () was applied on both the regular SOV hiring fee and the energy produced throughout 25 years. It was assumed that after SOV contract expiration it is renewed with the same pricing. However in future work the rise/drop of the fee should be taken into account. For the OMB scenarios the cost of building the base was varied between GBP 0 and GBP 30 million in order to find the break-even points at which having the OMB would be cost effective. The discount rate was assumed to be 5.5 % estimated by for the first commercial-scale projects in the UK. Figure  shows the results of this analysis. For the cases with the OMB and low wave limit, even when the capital cost associated with the OMB is low, the lines stay above the SOV scenarios. The big difference between ERA5 and ERA-20C cases for OMB scenarios is to do with the fact that due to worse weather conditions vessels return to the port more often, and hence their usage is longer in scenarios with ERA5 data. With ERA-20C data, vessels can more often maintain several assets in a row without returning to port. Scenarios with high wave limit, i.e. 2.0 m for transfer and 2.5 m for transit, are the only ones that have a section of the line below SOV scenarios.

As was mentioned in Sect. , future work in COMPASS includes being able to suspend activities in COMPASS (i.e. break activities into multiple short activities). Splitting up activities can potentially increase time and energy availability. If a wind farm was installed at a different location with calmer weather conditions, the availability could also be higher. Figure  assumes that OPEX costs stay the same but energy availability increases for the scenario with the OMB, high wave limit and ERA5 data. In the bottom four lines more energy is produced at the same cost, which results in lower price per megawatt hour. Even for the case with the highest availability presented in the graph, the OMB would need to contribute less than GBP 32 million to the total CAPEX, which is unlikely if the OMB has a separate substructure. According to the report published by the ORE Catapult a separate substructure of this size would cost at least GBP 49.5 million. Appendix  shows how the results in Figs.  and would change if the discount rate were 4.2 %, which is estimated for floating offshore projects when the industry matures. Changing the discount rate only affects the OMB results. Decreasing the discount does not change the origins of the lines representing OMB scenarios but makes their slope more gradual, thus moving the break-even point to the right on each figure.

Spare part storage capacity is not considered in both scenarios; however in the SOV case modelling the additional return to port would not affect the costs but would change the carbon emissions and availability output. This is due to the assumption of a long-term SOV contract. However in the OMB case there would be an additional cost of helicopter or vessel delivery of a part if it is not available at the site.

Carbon emissions were measured in each simulation and averaged out for SOV and OMB scenarios. Table  provides the results from this study and benchmarks them over the most recent studies. The results align well with the sustainability report results published by . However they are expected to be higher as further changes to COMPASS are incorporated. The emissions resulting from stand-by of the vessels or dynamic positioning are not included here. However emissions due to transit are expected to be the highest in these three categories.