Search in the TU Delft Repository Collections
TU Delft Repository
210,286
Records
238,802
Files
81,387
Contributors
Recently Added Records
Influence of Hammer Design on Pile Driving Noise
The Sound of Impact
Master thesis
(2026)
-
K.S. Broersma, J. Jovanova, Andrei B. Faragau, Ike van Giffen, Rutger Voeten, Josué Labaki
Underwater noise from monopile impact driving can harm marine mammals and is restricted in several European countries. Changing the hammer blow can reduce this noise at its source. Near-field models can predict this noise, but they often prescribe the pile-head force or represent the hammer by lumped masses. They thus leave out the stress waves in the ram and the anvil that shape this force. This thesis examines how the hammer design influences the underwater noise, while accounting for the hammer efficiency: the fraction of the impact energy delivered to the pile.
A transient, axisymmetric finite element model coupled the deformable ram and anvil of an IHC S-4000 hammer to the pile and the water. The pile-head force therefore followed from contact. The model was verified and benchmarked against GRLWEAP and scaling laws, but not validated against measurements. The results are relative comparisons of single blows for one pile. Thirty variations of the ram, the anvil and their coupling, including simplified models of the MENCK Noise Reduction Unit (MNRU) and IQIP's PULSE, were compared at equal impact energy. Each variation was evaluated by its hammer efficiency, the fraction of the pile energy that entered the water and the depth-averaged sound exposure level (SEL) at 10 m from the pile axis. These quantities show whether a variation changes the noise through the energy delivered to the pile or through the fraction entering the water.
Doubling the ram length lowered the SEL by 4.9 dB at almost unchanged hammer efficiency. Shifting cross-sectional area from the impact end to the top of the ram lowered the SEL by 5.0 dB, although more energy entered the pile. The fraction of that energy that entered the water decreased by almost a factor of four. The doubled-length ram achieved a similar reduction to the MNRU model, without its loss of hammer efficiency but with twice the ram mass. The PULSE model gave a larger reduction, whose size depends on an uncalibrated chamber stiffness. Differences in underwater sound exposure were largely explained by the pile-head force spectrum. A more gradual force puts less of its spectrum where the pile radiates most strongly. The hammer design therefore changes the noise mainly through how gradually the pile-head force develops. The noise of a design is thus better assessed from its force spectrum than from its hammer efficiency. Future work should measure the differences between configurations and compare them at equal penetration per blow. ...
A transient, axisymmetric finite element model coupled the deformable ram and anvil of an IHC S-4000 hammer to the pile and the water. The pile-head force therefore followed from contact. The model was verified and benchmarked against GRLWEAP and scaling laws, but not validated against measurements. The results are relative comparisons of single blows for one pile. Thirty variations of the ram, the anvil and their coupling, including simplified models of the MENCK Noise Reduction Unit (MNRU) and IQIP's PULSE, were compared at equal impact energy. Each variation was evaluated by its hammer efficiency, the fraction of the pile energy that entered the water and the depth-averaged sound exposure level (SEL) at 10 m from the pile axis. These quantities show whether a variation changes the noise through the energy delivered to the pile or through the fraction entering the water.
Doubling the ram length lowered the SEL by 4.9 dB at almost unchanged hammer efficiency. Shifting cross-sectional area from the impact end to the top of the ram lowered the SEL by 5.0 dB, although more energy entered the pile. The fraction of that energy that entered the water decreased by almost a factor of four. The doubled-length ram achieved a similar reduction to the MNRU model, without its loss of hammer efficiency but with twice the ram mass. The PULSE model gave a larger reduction, whose size depends on an uncalibrated chamber stiffness. Differences in underwater sound exposure were largely explained by the pile-head force spectrum. A more gradual force puts less of its spectrum where the pile radiates most strongly. The hammer design therefore changes the noise mainly through how gradually the pile-head force develops. The noise of a design is thus better assessed from its force spectrum than from its hammer efficiency. Future work should measure the differences between configurations and compare them at equal penetration per blow. ...
Underwater noise from monopile impact driving can harm marine mammals and is restricted in several European countries. Changing the hammer blow can reduce this noise at its source. Near-field models can predict this noise, but they often prescribe the pile-head force or represent the hammer by lumped masses. They thus leave out the stress waves in the ram and the anvil that shape this force. This thesis examines how the hammer design influences the underwater noise, while accounting for the hammer efficiency: the fraction of the impact energy delivered to the pile.
A transient, axisymmetric finite element model coupled the deformable ram and anvil of an IHC S-4000 hammer to the pile and the water. The pile-head force therefore followed from contact. The model was verified and benchmarked against GRLWEAP and scaling laws, but not validated against measurements. The results are relative comparisons of single blows for one pile. Thirty variations of the ram, the anvil and their coupling, including simplified models of the MENCK Noise Reduction Unit (MNRU) and IQIP's PULSE, were compared at equal impact energy. Each variation was evaluated by its hammer efficiency, the fraction of the pile energy that entered the water and the depth-averaged sound exposure level (SEL) at 10 m from the pile axis. These quantities show whether a variation changes the noise through the energy delivered to the pile or through the fraction entering the water.
Doubling the ram length lowered the SEL by 4.9 dB at almost unchanged hammer efficiency. Shifting cross-sectional area from the impact end to the top of the ram lowered the SEL by 5.0 dB, although more energy entered the pile. The fraction of that energy that entered the water decreased by almost a factor of four. The doubled-length ram achieved a similar reduction to the MNRU model, without its loss of hammer efficiency but with twice the ram mass. The PULSE model gave a larger reduction, whose size depends on an uncalibrated chamber stiffness. Differences in underwater sound exposure were largely explained by the pile-head force spectrum. A more gradual force puts less of its spectrum where the pile radiates most strongly. The hammer design therefore changes the noise mainly through how gradually the pile-head force develops. The noise of a design is thus better assessed from its force spectrum than from its hammer efficiency. Future work should measure the differences between configurations and compare them at equal penetration per blow.
A transient, axisymmetric finite element model coupled the deformable ram and anvil of an IHC S-4000 hammer to the pile and the water. The pile-head force therefore followed from contact. The model was verified and benchmarked against GRLWEAP and scaling laws, but not validated against measurements. The results are relative comparisons of single blows for one pile. Thirty variations of the ram, the anvil and their coupling, including simplified models of the MENCK Noise Reduction Unit (MNRU) and IQIP's PULSE, were compared at equal impact energy. Each variation was evaluated by its hammer efficiency, the fraction of the pile energy that entered the water and the depth-averaged sound exposure level (SEL) at 10 m from the pile axis. These quantities show whether a variation changes the noise through the energy delivered to the pile or through the fraction entering the water.
Doubling the ram length lowered the SEL by 4.9 dB at almost unchanged hammer efficiency. Shifting cross-sectional area from the impact end to the top of the ram lowered the SEL by 5.0 dB, although more energy entered the pile. The fraction of that energy that entered the water decreased by almost a factor of four. The doubled-length ram achieved a similar reduction to the MNRU model, without its loss of hammer efficiency but with twice the ram mass. The PULSE model gave a larger reduction, whose size depends on an uncalibrated chamber stiffness. Differences in underwater sound exposure were largely explained by the pile-head force spectrum. A more gradual force puts less of its spectrum where the pile radiates most strongly. The hammer design therefore changes the noise mainly through how gradually the pile-head force develops. The noise of a design is thus better assessed from its force spectrum than from its hammer efficiency. Future work should measure the differences between configurations and compare them at equal penetration per blow.
Estimating Ship Structural Response Using Sparse Sensor Measurements
Transient (Slamming) Effects
Master thesis
(2026)
-
A.C. Gaard, J.H. den Besten, A. Grammatikopoulos, P.R. Wellens, N. Yue, R.L.G. Slange, Gabriele Bufalari, Robin van Dijke
Slamming can accelerate fatigue damage accumulation in ship structures, but the developing cracks are not always easily detectable. Sparse sensor measurements can be used to estimate the structural response of ships, perform a fatigue assessment, and identify areas requiring preventive maintenance. To estimate irregular linear wave loading and response, measured strains are often multiplied by an inverted modal basis, resulting in a set of amplitude estimates that scales the external loading and structural response at all finite elements. The modal basis itself is commonly constructed as a set of particular solutions to the frequency domain equation of motion (distortion modes), but higher-order vibrations are excited by slamming loads and are best represented by the homogeneous solution (wet modes). To address global effects, this work proposes a method for estimating the global high-frequency loading and response by means of using wet modes, a truncated set of complex eigenvectors obtained using structural mass, stiffness, damping, and frequency-dependent radiation coefficients. To address local hydroelastic slamming effects, this work proposes a method for estimating the local high-frequency response by means of using dry modes, a truncated set of real eigenvectors obtained using structural mass and stiffness matrices of an orthotropically stiffened panel. The two methods are investigated using one-dimensional finite element models: a scale segmented barge and an orthotropically stiffened plate, which is represented by a beam. Results show that a truncated basis of wet modes accurately estimates high-frequency strains and pressures at unmeasured locations, while low-frequency rigid body motions introduce inestimable effects. The basis of distortion modes is found to introduce significant loading estimation error, despite estimating strains near the measured locations with relatively good accuracy. A truncated basis of dry modes accurately estimates the hydroelastic response of a panel at unmeasured locations. Three sensors and three modes are required in both cases, but separation of high- and low-frequency measurements by means of a discrete filter is required for accurate high-frequency estimates in the global model. Models are verified against analytical, numerical, and experimental results from literature.
...
Slamming can accelerate fatigue damage accumulation in ship structures, but the developing cracks are not always easily detectable. Sparse sensor measurements can be used to estimate the structural response of ships, perform a fatigue assessment, and identify areas requiring preventive maintenance. To estimate irregular linear wave loading and response, measured strains are often multiplied by an inverted modal basis, resulting in a set of amplitude estimates that scales the external loading and structural response at all finite elements. The modal basis itself is commonly constructed as a set of particular solutions to the frequency domain equation of motion (distortion modes), but higher-order vibrations are excited by slamming loads and are best represented by the homogeneous solution (wet modes). To address global effects, this work proposes a method for estimating the global high-frequency loading and response by means of using wet modes, a truncated set of complex eigenvectors obtained using structural mass, stiffness, damping, and frequency-dependent radiation coefficients. To address local hydroelastic slamming effects, this work proposes a method for estimating the local high-frequency response by means of using dry modes, a truncated set of real eigenvectors obtained using structural mass and stiffness matrices of an orthotropically stiffened panel. The two methods are investigated using one-dimensional finite element models: a scale segmented barge and an orthotropically stiffened plate, which is represented by a beam. Results show that a truncated basis of wet modes accurately estimates high-frequency strains and pressures at unmeasured locations, while low-frequency rigid body motions introduce inestimable effects. The basis of distortion modes is found to introduce significant loading estimation error, despite estimating strains near the measured locations with relatively good accuracy. A truncated basis of dry modes accurately estimates the hydroelastic response of a panel at unmeasured locations. Three sensors and three modes are required in both cases, but separation of high- and low-frequency measurements by means of a discrete filter is required for accurate high-frequency estimates in the global model. Models are verified against analytical, numerical, and experimental results from literature.
Accelerating sustainability transitions requires moving innovations from experimental testbeds into mainstream practice. Across the Netherlands, Living Labs and Field Labs have emerged as environments in which technological, organisational and social innovations can be tested in real-world settings. However, successful experimentation within these testing environments does not necessarily result in development or adoption beyond the pilot. This is reflected in the ‘pilot paradox’, where projects can achieve their experimental objectives while remaining limited in their ability to develop beyond the pilot.
Stakeholder involvement is important in this process because stakeholders can provide knowledge, resources and access to the contexts in which innovations need to be implemented. At the same time, previous research shows that identifying, attracting and maintaining stakeholder involvement can be difficult. This thesis therefore investigates the challenges that projects themselves encounter when organising stakeholder involvement to support developments beyond the pilot.
The aim of this thesis is to understand how stakeholder involvement is organised in Dutch Living Labs and Field Labs and to identify the challenges that constrain stakeholder involvement in supporting scaling beyond the pilot. The central research question asks: What challenges do projects within Dutch Living Labs and Field Labs encounter when involving stakeholders to support scaling beyond the pilot?
A qualitative multiple-case study was conducted on three innovation projects in a Dutch Living Lab and Field Lab. Two cases took place at The Green Village in Delft and one case took place at Energie-Lab Zuid Oost in Amsterdam. Data were collected through project documentation, participant observations and 15 semi-structured interviews with actors from governmental institutions, Living and Field Labs, industry, and residents. The data were analysed using abductive thematic coding.
The findings show that stakeholder involvement was dynamic rather than continuous. Several challenges in involving stakeholders to support scaling beyond the pilot were identified within the case studies. These challenges occur on both sides of the involvement process, as both project teams and stakeholders face challenges that can limit involvement. 6 categories were identified: (1) limited resources and capacity; (2) identifying and reaching relevant stakeholders; (3) perceived relevance and willingness to participate; (4) meaningful participation and influence; (5) institutional and commercial constraints; and (6) project purpose and development stage.
The main implication is that deliberate choices need to be made within projects about the relevance of stakeholders at different stages, how they can realistically participate, and what opportunities they have to contribute. The study recommends more flexible approaches to involvement, explicit attention to representation and power, and participation methods that are adapted to stakeholder capacities and project needs. The main limitations are the retrospective design, the focus on three cases within two experimental environments, and the emphasis on actors who were already connected to the projects. Future research should therefore also investigate the perspectives of stakeholders who were not involved but could have been relevant and examine how meaningful stakeholder involvement can be organised under the practical and institutional constraints of real-world experimentation.
...
Stakeholder involvement is important in this process because stakeholders can provide knowledge, resources and access to the contexts in which innovations need to be implemented. At the same time, previous research shows that identifying, attracting and maintaining stakeholder involvement can be difficult. This thesis therefore investigates the challenges that projects themselves encounter when organising stakeholder involvement to support developments beyond the pilot.
The aim of this thesis is to understand how stakeholder involvement is organised in Dutch Living Labs and Field Labs and to identify the challenges that constrain stakeholder involvement in supporting scaling beyond the pilot. The central research question asks: What challenges do projects within Dutch Living Labs and Field Labs encounter when involving stakeholders to support scaling beyond the pilot?
A qualitative multiple-case study was conducted on three innovation projects in a Dutch Living Lab and Field Lab. Two cases took place at The Green Village in Delft and one case took place at Energie-Lab Zuid Oost in Amsterdam. Data were collected through project documentation, participant observations and 15 semi-structured interviews with actors from governmental institutions, Living and Field Labs, industry, and residents. The data were analysed using abductive thematic coding.
The findings show that stakeholder involvement was dynamic rather than continuous. Several challenges in involving stakeholders to support scaling beyond the pilot were identified within the case studies. These challenges occur on both sides of the involvement process, as both project teams and stakeholders face challenges that can limit involvement. 6 categories were identified: (1) limited resources and capacity; (2) identifying and reaching relevant stakeholders; (3) perceived relevance and willingness to participate; (4) meaningful participation and influence; (5) institutional and commercial constraints; and (6) project purpose and development stage.
The main implication is that deliberate choices need to be made within projects about the relevance of stakeholders at different stages, how they can realistically participate, and what opportunities they have to contribute. The study recommends more flexible approaches to involvement, explicit attention to representation and power, and participation methods that are adapted to stakeholder capacities and project needs. The main limitations are the retrospective design, the focus on three cases within two experimental environments, and the emphasis on actors who were already connected to the projects. Future research should therefore also investigate the perspectives of stakeholders who were not involved but could have been relevant and examine how meaningful stakeholder involvement can be organised under the practical and institutional constraints of real-world experimentation.
...
Accelerating sustainability transitions requires moving innovations from experimental testbeds into mainstream practice. Across the Netherlands, Living Labs and Field Labs have emerged as environments in which technological, organisational and social innovations can be tested in real-world settings. However, successful experimentation within these testing environments does not necessarily result in development or adoption beyond the pilot. This is reflected in the ‘pilot paradox’, where projects can achieve their experimental objectives while remaining limited in their ability to develop beyond the pilot.
Stakeholder involvement is important in this process because stakeholders can provide knowledge, resources and access to the contexts in which innovations need to be implemented. At the same time, previous research shows that identifying, attracting and maintaining stakeholder involvement can be difficult. This thesis therefore investigates the challenges that projects themselves encounter when organising stakeholder involvement to support developments beyond the pilot.
The aim of this thesis is to understand how stakeholder involvement is organised in Dutch Living Labs and Field Labs and to identify the challenges that constrain stakeholder involvement in supporting scaling beyond the pilot. The central research question asks: What challenges do projects within Dutch Living Labs and Field Labs encounter when involving stakeholders to support scaling beyond the pilot?
A qualitative multiple-case study was conducted on three innovation projects in a Dutch Living Lab and Field Lab. Two cases took place at The Green Village in Delft and one case took place at Energie-Lab Zuid Oost in Amsterdam. Data were collected through project documentation, participant observations and 15 semi-structured interviews with actors from governmental institutions, Living and Field Labs, industry, and residents. The data were analysed using abductive thematic coding.
The findings show that stakeholder involvement was dynamic rather than continuous. Several challenges in involving stakeholders to support scaling beyond the pilot were identified within the case studies. These challenges occur on both sides of the involvement process, as both project teams and stakeholders face challenges that can limit involvement. 6 categories were identified: (1) limited resources and capacity; (2) identifying and reaching relevant stakeholders; (3) perceived relevance and willingness to participate; (4) meaningful participation and influence; (5) institutional and commercial constraints; and (6) project purpose and development stage.
The main implication is that deliberate choices need to be made within projects about the relevance of stakeholders at different stages, how they can realistically participate, and what opportunities they have to contribute. The study recommends more flexible approaches to involvement, explicit attention to representation and power, and participation methods that are adapted to stakeholder capacities and project needs. The main limitations are the retrospective design, the focus on three cases within two experimental environments, and the emphasis on actors who were already connected to the projects. Future research should therefore also investigate the perspectives of stakeholders who were not involved but could have been relevant and examine how meaningful stakeholder involvement can be organised under the practical and institutional constraints of real-world experimentation.
Stakeholder involvement is important in this process because stakeholders can provide knowledge, resources and access to the contexts in which innovations need to be implemented. At the same time, previous research shows that identifying, attracting and maintaining stakeholder involvement can be difficult. This thesis therefore investigates the challenges that projects themselves encounter when organising stakeholder involvement to support developments beyond the pilot.
The aim of this thesis is to understand how stakeholder involvement is organised in Dutch Living Labs and Field Labs and to identify the challenges that constrain stakeholder involvement in supporting scaling beyond the pilot. The central research question asks: What challenges do projects within Dutch Living Labs and Field Labs encounter when involving stakeholders to support scaling beyond the pilot?
A qualitative multiple-case study was conducted on three innovation projects in a Dutch Living Lab and Field Lab. Two cases took place at The Green Village in Delft and one case took place at Energie-Lab Zuid Oost in Amsterdam. Data were collected through project documentation, participant observations and 15 semi-structured interviews with actors from governmental institutions, Living and Field Labs, industry, and residents. The data were analysed using abductive thematic coding.
The findings show that stakeholder involvement was dynamic rather than continuous. Several challenges in involving stakeholders to support scaling beyond the pilot were identified within the case studies. These challenges occur on both sides of the involvement process, as both project teams and stakeholders face challenges that can limit involvement. 6 categories were identified: (1) limited resources and capacity; (2) identifying and reaching relevant stakeholders; (3) perceived relevance and willingness to participate; (4) meaningful participation and influence; (5) institutional and commercial constraints; and (6) project purpose and development stage.
The main implication is that deliberate choices need to be made within projects about the relevance of stakeholders at different stages, how they can realistically participate, and what opportunities they have to contribute. The study recommends more flexible approaches to involvement, explicit attention to representation and power, and participation methods that are adapted to stakeholder capacities and project needs. The main limitations are the retrospective design, the focus on three cases within two experimental environments, and the emphasis on actors who were already connected to the projects. Future research should therefore also investigate the perspectives of stakeholders who were not involved but could have been relevant and examine how meaningful stakeholder involvement can be organised under the practical and institutional constraints of real-world experimentation.
Lift Estimation from Wind Tunnel Wall Pressure
An experimental approach carried out at the Research & Development department of Econowind
At Econowind, wind-tunnel testing is used to evaluate the aerodynamic performance of boundary-layer suction wings, with lift primarily determined through wall pressure measurements (WPM).
A baseline experiment was conducted using a NACA0015 airfoil without suction. Lift coefficients obtained from WPM were compared with independently derived surface pressure measurements (SPM) at Reynolds numbers of Re=200,000, Re=400,000, and Re=600,000.
The results show good agreement between WPM and SPM in the pre-stall regime, while discrepancies increase around stall and become substantial in the post-stall region, where WPM consistently overpredicts lift relative to SPM.
Further analysis revealed that the measured wall-pressure footprint alone yields lift estimates close to those obtained from SPM in the post-stall regime, whereas the existing corrections and empirical extrapolations generally lead to overprediction.
These findings demonstrate the potential of WPM for lift estimation under predominantly attached-flow conditions while identifying strongly separated flow as a major limitation of the current approach. ...
A baseline experiment was conducted using a NACA0015 airfoil without suction. Lift coefficients obtained from WPM were compared with independently derived surface pressure measurements (SPM) at Reynolds numbers of Re=200,000, Re=400,000, and Re=600,000.
The results show good agreement between WPM and SPM in the pre-stall regime, while discrepancies increase around stall and become substantial in the post-stall region, where WPM consistently overpredicts lift relative to SPM.
Further analysis revealed that the measured wall-pressure footprint alone yields lift estimates close to those obtained from SPM in the post-stall regime, whereas the existing corrections and empirical extrapolations generally lead to overprediction.
These findings demonstrate the potential of WPM for lift estimation under predominantly attached-flow conditions while identifying strongly separated flow as a major limitation of the current approach. ...
At Econowind, wind-tunnel testing is used to evaluate the aerodynamic performance of boundary-layer suction wings, with lift primarily determined through wall pressure measurements (WPM).
A baseline experiment was conducted using a NACA0015 airfoil without suction. Lift coefficients obtained from WPM were compared with independently derived surface pressure measurements (SPM) at Reynolds numbers of Re=200,000, Re=400,000, and Re=600,000.
The results show good agreement between WPM and SPM in the pre-stall regime, while discrepancies increase around stall and become substantial in the post-stall region, where WPM consistently overpredicts lift relative to SPM.
Further analysis revealed that the measured wall-pressure footprint alone yields lift estimates close to those obtained from SPM in the post-stall regime, whereas the existing corrections and empirical extrapolations generally lead to overprediction.
These findings demonstrate the potential of WPM for lift estimation under predominantly attached-flow conditions while identifying strongly separated flow as a major limitation of the current approach.
A baseline experiment was conducted using a NACA0015 airfoil without suction. Lift coefficients obtained from WPM were compared with independently derived surface pressure measurements (SPM) at Reynolds numbers of Re=200,000, Re=400,000, and Re=600,000.
The results show good agreement between WPM and SPM in the pre-stall regime, while discrepancies increase around stall and become substantial in the post-stall region, where WPM consistently overpredicts lift relative to SPM.
Further analysis revealed that the measured wall-pressure footprint alone yields lift estimates close to those obtained from SPM in the post-stall regime, whereas the existing corrections and empirical extrapolations generally lead to overprediction.
These findings demonstrate the potential of WPM for lift estimation under predominantly attached-flow conditions while identifying strongly separated flow as a major limitation of the current approach.
Master thesis
(2026)
-
S.E.G. Du Marchie Sarvaas, Lorant Tavasszy, A.J. van Binsbergen, M.A. de Bok, Alex Vermeulen, Bastiaan Possel
Dutch regional traffic models derive freight trip ends from employment per zone and treat every zone as independent. This thesis asks whether freight trip generation improves when zones are related to each other, through geographic proximity and through tour co-occurrence: the sharing of truck tours. Six models of increasing complexity, from current practice through a log-log hurdle model, a spatial error model and geographically weighted regression to models with tour-lagged predictors, were estimated on the 6,465 zones of the national freight model BasGoed, whose synthetic truck tours supply both the dependent variables and the tour network. Each model was judged on in-sample fit and on spatially blocked ten-fold cross-validation, then transferred without re-estimation to Goudappel’s regional traffic model of Drechtsteden and the Alblasserwaard and compared with 814 traffic counts.
Every extension improves the fit to the national data, yet none predicts the trip ends of unseen blocks of zones better than current practice when errors are counted in trips. The models with local coefficients are 5 to 7 percent worse in trips but more accurate in relative terms for small zones. A random split would have overstated accuracy by a factor of two. Robust tests reject a spatial lag; the remaining clustering is consistent with a regional level that BasGoed’s allocation builds into the data. Tour partners rarely coincide with geographic neighbours (8.7 percent), and industrial employment among them has a significant elasticity of about 0.09. In the regional model, all variants order busy and quiet roads correctly and differ far less from each other than from the counts.
The extensions add reach rather than accuracy. For base-year volumes the simplest model is as good as any; the tour model can show how growth in one place affects freight in places linked to it by tours. These findings hold within BasGoed’s simulated tours. ...
Every extension improves the fit to the national data, yet none predicts the trip ends of unseen blocks of zones better than current practice when errors are counted in trips. The models with local coefficients are 5 to 7 percent worse in trips but more accurate in relative terms for small zones. A random split would have overstated accuracy by a factor of two. Robust tests reject a spatial lag; the remaining clustering is consistent with a regional level that BasGoed’s allocation builds into the data. Tour partners rarely coincide with geographic neighbours (8.7 percent), and industrial employment among them has a significant elasticity of about 0.09. In the regional model, all variants order busy and quiet roads correctly and differ far less from each other than from the counts.
The extensions add reach rather than accuracy. For base-year volumes the simplest model is as good as any; the tour model can show how growth in one place affects freight in places linked to it by tours. These findings hold within BasGoed’s simulated tours. ...
Dutch regional traffic models derive freight trip ends from employment per zone and treat every zone as independent. This thesis asks whether freight trip generation improves when zones are related to each other, through geographic proximity and through tour co-occurrence: the sharing of truck tours. Six models of increasing complexity, from current practice through a log-log hurdle model, a spatial error model and geographically weighted regression to models with tour-lagged predictors, were estimated on the 6,465 zones of the national freight model BasGoed, whose synthetic truck tours supply both the dependent variables and the tour network. Each model was judged on in-sample fit and on spatially blocked ten-fold cross-validation, then transferred without re-estimation to Goudappel’s regional traffic model of Drechtsteden and the Alblasserwaard and compared with 814 traffic counts.
Every extension improves the fit to the national data, yet none predicts the trip ends of unseen blocks of zones better than current practice when errors are counted in trips. The models with local coefficients are 5 to 7 percent worse in trips but more accurate in relative terms for small zones. A random split would have overstated accuracy by a factor of two. Robust tests reject a spatial lag; the remaining clustering is consistent with a regional level that BasGoed’s allocation builds into the data. Tour partners rarely coincide with geographic neighbours (8.7 percent), and industrial employment among them has a significant elasticity of about 0.09. In the regional model, all variants order busy and quiet roads correctly and differ far less from each other than from the counts.
The extensions add reach rather than accuracy. For base-year volumes the simplest model is as good as any; the tour model can show how growth in one place affects freight in places linked to it by tours. These findings hold within BasGoed’s simulated tours.
Every extension improves the fit to the national data, yet none predicts the trip ends of unseen blocks of zones better than current practice when errors are counted in trips. The models with local coefficients are 5 to 7 percent worse in trips but more accurate in relative terms for small zones. A random split would have overstated accuracy by a factor of two. Robust tests reject a spatial lag; the remaining clustering is consistent with a regional level that BasGoed’s allocation builds into the data. Tour partners rarely coincide with geographic neighbours (8.7 percent), and industrial employment among them has a significant elasticity of about 0.09. In the regional model, all variants order busy and quiet roads correctly and differ far less from each other than from the counts.
The extensions add reach rather than accuracy. For base-year volumes the simplest model is as good as any; the tour model can show how growth in one place affects freight in places linked to it by tours. These findings hold within BasGoed’s simulated tours.