Hydro­dy­namic opti­miza­tion of a cata­ma­ran house­boat for inland waterways

In recent years, the demand for house­boats has increased sig­nif­i­cantly, accom­pa­nied by a growing emphasis on envi­ron­men­tally friendly vessel designs due to the pressing need to address climate change. The company Rolly­boot has responded to this trend with its fully electric house­boat, the ​“Rolly­boot Evo­lu­tion”, pre­sent­ing a forward-looking solution. It is commonly known that hull geometry sig­nif­i­cantly affects wave resis­tance and overall hydro­dy­namic per­for­mance. However, like much of the house­boat industry, Rolly­boot has not yet con­sid­ered hull form opti­miza­tion to achieve addi­tional per­for­mance gains. A unique feature of Rolly­boot is its use of modular float struc­tures, which provides an ideal foun­da­tion for improv­ing effi­ciency with minimal design changes and resource expen­di­ture. In addition, the floaters are produced using the rota­tional molding process, which allows the result­ing plastic hollow bodies to be formed in any desired geometry.

This study inves­ti­gates the influ­ence of mod­i­fy­ing the bow shape on the vessel’s total resis­tance. A para­met­ric model was devel­oped and sub­jected to a RANS-based multi-objec­tive opti­miza­tion. While tra­di­tional opti­miza­tion methods are resource-inten­sive due to the high number of design vari­ables, Prin­ci­pal Com­po­nent Analysis (PCA) offers a promis­ing solution by reducing the design space while pre­serv­ing vari­abil­ity, poten­tially making the opti­miza­tion process faster and more cost-effec­tive. To prove this thesis, a com­par­a­tive opti­miza­tion was con­ducted using a para­me­ter-reduced Karhunen – Loève Expan­sion (KLE) model.

Approach and Objectives

The research utilizes a para­met­ric modeling approach realized within the CAESES envi­ron­ment for both CAD design and opti­miza­tion. The opti­miza­tion process inte­grates the NSGA-II algo­rithm along­side the Karhunen – Loève Expan­sion (KLE) method to reduce the dimen­sion­al­ity of the design space. Com­pu­ta­tional Fluid Dynamics (CFD) analyses are per­formed using the viscous RANS solver FS-Foam and the auto­mated mesh gen­er­a­tion tools provided by DNV’s Hydro­dy­nam­ics Department.

The study conducts a com­par­a­tive eval­u­a­tion between a con­ven­tional CAD-based opti­miza­tion model and a reduced-order KLE model, assess­ing both in terms of the attained hydro­dy­namic per­for­mance and com­pu­ta­tional effi­ciency. This approach is intended to demon­strate that it is possible to find a better shape for the house­boat cata­ma­ran hull form, and that a rea­son­able solution can be found faster and more cost-effec­tively by using prin­ci­pal para­me­ters from the Prin­ci­pal Com­po­nent Analysis.

Design Con­straints and Para­met­ric Model

The hull design is con­strained by the existing dimen­sions of the Rolly­boot platform, allowing only a 7 cm exten­sion at the bow to maintain com­pat­i­bil­ity with mooring and modular con­struc­tion. For pro­duc­tion reasons, only the bow is changed in this study. Addi­tion­ally, the new bow model must be suitable for both demi­hulls and use the same form, ruling out asym­met­ri­cal shapes.

The para­met­ric model was devel­oped using the CAD inter­face of CAESES 5, based on a general model struc­ture used at DNV. The focus was on mod­i­fy­ing the bow while keeping the middle and aft parts of the original Rolly­boot baseline model unchanged. The bow model’s length includes the original bow module, plus the afore­men­tioned 7 cm. The model is designed to explore the effects of bow shape on resis­tance, allowing for a bulbous bow and a flat vertical stem curve. However, the model excludes reverse bows due to geo­met­ric limitations.

The model consists of three main surfaces:

1. Pre-surface: Com­pris­ing three B‑Spline surfaces with 4x5 points, pro­vid­ing volume and a flat surface for platform fixing.
2. Smooth surface: Ensuring smooth tran­si­tions between the B‑Spline surfaces.
3. Blended surface: A fillet surface between the cross-section loft at the cut position and a defined point of the new bow model, ensuring smooth tran­si­tion to the baseline model.

The para­met­ric bow model features 23 free vari­ables, cat­e­go­rized into four groups based on their influ­ence on shape:

• Stem curve appear­ance (6 variables)
• Lower volume dis­tri­b­u­tion (6 variables)
• Middle region (5 variables)
• Upper part and tran­si­tion to middle section (6 variables)

Finally, the full demihull model is closed with a BRep by merging the blended bow model with the baseline model shape. Five possible geometry variants are pre­sented in the image below to illus­trate the wide range of model variations.

To reduce com­pu­ta­tional com­plex­ity, a Karhunen – Loève Expan­sion (KLE) model was created with the respec­tive tool avail­able in CAESES using 230 design samples. This reduced the number of para­me­ters from 23 to 7 while pre­serv­ing 96.2% of the model’s vari­abil­ity. The first prin­ci­pal para­me­ter alone accounts for over 30% of the geo­met­ric vari­a­tion, enabling effi­cient opti­miza­tion with sig­nif­i­cantly fewer variables.

Prepa­ra­tion Phase

During the prepa­ra­tion phase, key aspects such as the selec­tion of the CFD code, mesh quality assess­ment, a depth study, and a check of the hydro­sta­t­ics were sys­tem­at­i­cally evaluated.

To validate the plau­si­bil­ity of the RANS sim­u­la­tions, the computed calm water resis­tance was compared with real­is­tic propul­sion data. Accord­ing to the man­u­fac­turer, the Rolly­boot reaches a speed of 7.8 km/​h using a 6 kW electric motor. Since detailed pro­peller data was unavail­able, a Wagenin­gen B‑Series pro­peller was modeled using the online pro­peller tool from FRIEND­SHIP SYSTEMS, incor­po­rat­ing avail­able spec­i­fi­ca­tions from ePropulsion.

The result­ing pro­peller showed a thrust of 1097 N and an effi­ciency of 0.42 at the target speed. In com­par­i­son, the CFD sim­u­la­tion pre­dicted a resis­tance of approx­i­mately 800 N — about 27% lower. This dis­crep­ancy is attrib­uted to unmod­eled factors such as struc­tural gaps, hull openings, fouling, and wind resistance.

Addi­tion­ally, the sim­u­lated wave pattern closely matched the real waves at this speed, further sup­port­ing the cred­i­bil­ity of the CFD results.

Opti­miza­tion

To inves­ti­gate the hydro­dy­namic behavior of the cata­ma­ran house­boat, the fol­low­ing setup is used:

Two cases are con­sid­ered impor­tant for the use of the Rollyboot.

Case A covers the most common scenario of a family (or friend group) of four people trav­el­ing with the boat for a weekend or holiday trip at an average cruising velocity of 7 km/​h. Con­sid­er­ing the empty weight of the Rolly­boot Evo­lu­tion (2,940 kg), an average weight per person of 75 kg plus 25 kg of pro­vi­sions, and 100 kg for fresh water, the dis­place­ment for this case is assumed to be 3.5 t and evenly loaded.

Case B rep­re­sents a transit scenario for one person cruising at a higher speed of 10 km/​h. For this case, a dis­place­ment of 3.1 t and an aft trim of 1° is assumed.

Since reducing wave pro­duc­tion is a key objec­tive for the new model, wave pattern resis­tance and wave heights at dif­fer­ent wave cut posi­tions are mon­i­tored. The cuts at 0.2 m between the hulls and 1.6 m beside them are con­sid­ered most relevant for optimization.

The opti­miza­tion aims to minimize the hydro­dy­namic total resis­tance, the total wave pattern resis­tance at y = 1.6 m, and the maximum wave height at y = 0.2 m between the demi­hulls for both cases. The latter is par­tic­u­larly impor­tant, as initial sim­u­la­tions showed increased wave for­ma­tion between the demi­hulls, which must be min­i­mized to avoid wave impact on the platform. Min­i­miz­ing wave pattern resis­tance at the 1.6 m cut helps reduce wake wash around the boat.

As Case B is con­sid­ered less critical, a weight­ing of 30% for Case B and 70% for Case A is applied, reflect­ing the oper­a­tional profile. This results in a multi-objec­tive problem with three objec­tives. All values are pre­sented relative to the baseline model for clearer and more straight­for­ward comparison.

The center of buoyancy was con­strained to ±20 cm from the baseline for both cases. Each gen­er­ated design’s draught was deter­mined based on hydro­sta­t­ics and the spec­i­fied dis­place­ment. Two multi-objec­tive opti­miza­tions were con­ducted using the NSGA-II algo­rithm in CAESES: one with the standard CAD model (23 para­me­ters) and one with the reduced-order KLE model (7 prin­ci­pal para­me­ters). Both setups used a Sobol sequences for the initial sampling — 230 valid designs for the CAD model and 70 for the KLE model. Sim­u­la­tions were run with a coarser mesh on 16 proces­sor cores. 

Results

Results showed con­sis­tent improve­ments in total resis­tance for both cases compared to the baseline, with the best designs sig­nif­i­cantly out­per­form­ing it.

Key findings were:

• The NSGA II is stopped for the KLE approach when the objec­tives con­verged (556 valid designs), while for the con­ven­tional CAD approach no con­ver­gence could be iden­ti­fied and thus the algo­rithm stopped after 661 designs (time and resource limit).
• Great savings of resources can be realized when running the DoE phase of the opti­miza­tion (230 CAD designs compared to only 70 with KLE model, cap­tur­ing 96% of vari­abil­ity with 7 prin­ci­pal parameters).
• Addi­tional time and resources are needed to compare models when using NSGA II and RANS in combination.
• Good designs could be achieved for both the KLE and CAD model, with the best designs deliv­er­ing 30 – 40% improve­ment in total resistance.
• The best designs were deter­mined by low values for the 3 objec­tives, low change of XCB in case A and suit­abil­ity for the task.
• One of the CAD designs was the most con­vinc­ing and was proposed as final design to Rollyboot.

Best Design

The best design variant shows a reduc­tion in total resis­tance by 65.76 % of the baseline value for case A and by 55.92 % for case B. The improve­ment for the maximum wave height at y=0.2m is 44.76% for case A and 56.72 % for case B. The total wave pattern resis­tance at y=1.6m could be reduced the most, where the baseline values decreased to 21.16% for case A and 19.49% for case B. The figure below presents the resis­tance curves for the final design compared to the baseline Rolly­boot with a dis­place­ment of 3.5 tons.

The improved hull form will sig­nif­i­cantly extend the vessel’s range. The table below presents the results of an esti­ma­tion con­ducted for case A (3.5 tons dis­place­ment and 7 km/​h). While the increase in hours may seem modest, the poten­tial distance increases sig­nif­i­cantly from 35 km to 47 km. This applies under the examined con­di­tions and if the Rolly­boot Evo­lu­tion retains its 20,000 kW energy system.

The fol­low­ing figures show the geometry of the new bow compared to the original design as well as an overview of the results for the final design.

Finally, the last illus­tra­tion shows the newly designed module com­mis­sioned by Rollyboot.

Con­clu­sion and Outlook

The study suc­cess­fully iden­ti­fied an opti­mized hull form for the ​“Rolly­boot Evo­lu­tion”, achiev­ing up to 34% reduc­tion in total resis­tance at 7 km/​h for a 3.5‑ton load and 38% for the transit case. The opti­mized design also showed reduced wave height and wave pattern resis­tance, making it well-suited for fully electric oper­a­tion on inland water­ways. Both the standard CAD model and the reduced-order KLE model, using NSGA-II as opti­miza­tion method, deliv­ered promis­ing results. The KLE model shows strong poten­tial for sig­nif­i­cant effi­ciency gains, where further com­pu­ta­tional resources and time would allow these benefits to be demon­strated more fully. Still, the KLE model effec­tively reduced the number of CAD design para­me­ters from 23 to 7 prin­ci­pal para­me­ters while main­tain­ing 96.2% vari­abil­ity of the model.

Addi­tional research should include hydro­sta­t­ics, per­for­mance at lower speeds and shallow water, and poten­tial improve­ments through stern redesign, asym­met­ric hulls, or modular exten­sions. Other relevant aspects such as pro­peller-hull inter­ac­tion, tunnel clear­ance, and wave behavior between demi­hulls also warrant further explo­ration. Physical model testing is also sug­gested for validation.

The study estab­lishes a foun­da­tion for explor­ing the resource-saving poten­tial of applying a KLE model in NSGA-II and RANS-based opti­miza­tion, with promis­ing oppor­tu­ni­ties for val­i­da­tion through further com­par­isons between the KLE and CAD models under con­sis­tent con­di­tions. Enhanc­ing the KLE model quality and inves­ti­gat­ing single-objec­tive opti­miza­tion (e.g., using the Tangent Search method) could provide deeper insights.

About the Author

Fiona Schlote is a Project Engineer in the Ship Per­for­mance Center at DNV Maritime Advisory. She studied Naval Archi­tec­ture and Ocean Engi­neer­ing at Tech­nis­che Uni­ver­sität Berlin and com­pleted her master’s thesis in 2024 in col­lab­o­ra­tion with FRIEND­SHIP SYSTEMS, Rolly­boot, and DNV.