This paper presents a novel data-driven Reference Governor with Model Predictive Control integrating local motion replanning and path following for collision avoidance. Employing a model-free Reference Governor, the proposed solution utilises system knowledge through Bayesian Optimisation to augment predetermined evasive trajectories, minimising path-following errors and simultaneously ensuring obstacle safety margins. A single-track vehicle model in combination with non-linear tyre models is used to capture the vehicle’s dynamics. The optimised control action is the vehicle steering angle, whilst the Reference Governor optimises parameters of a sigmoid reference signal to minimise the tracking error and guarantee safety with respect to obstacles in emergency manoeuvres. The proposed approach is evaluated on a single lane change using a high-fidelity simulation environment and its performance is compared to a baseline controller integrating path following and obstacle avoidance. The results show a 14% reduction of safety critical overshoot, maximising obstacle safety distance and a four times lower controller cycle time compared to the baseline. Furthermore, through a robustness analysis, it is demonstrated that the proposed approach is more robust towards model mismatches and perception-based errors, as seen by average 30% and 40% reductions in near-miss and collision rates.

Automation of machines is becoming increasingly widespread and advanced, of which an example is the use robots for Prime Vision, which sorts parcels for postal services. The coordination of scheduling a fleet of robots picking up and dropping off many parcels while avoiding collisions, within a limited space, following predefined routes in a floorplan, is a complex problem.

This logistical challenge can be effectively modelled using max-plus linear algebra to allow an optimization for the route scheduling as was previously done by L. Smeets. The goal of this research is to improve the existing scheduling model and use this to develop a reinforcement learning-based algorithm that determines the optimal floorplan for the parcel delivery robots. Two methods are applied to improve the existing scheduling model. Firstly, nodes where no decisions are made are identified and removed. Secondly, certain constraints are also removed to simplify the model.

The results of the scheduler are used to determine a key performance indicator to allow a reinforcement learning based algorithm to identify the optimal floorplan for the robots. The reinforcement learning algorithm employed a deep Q-learning approach, with the neural network trained using various action space approaches, tuned rewards and hyper-parameters. The greedy-epsilon method was applied to address the exploration vs. exploitation problem. While the scheduler improvements significantly enhanced its computational costs, the neural network did not converge, and the potential causes are thoroughly discussed.

Traversability estimation is a key component in autonomous driving tasks. In many applications, semantic segmentation is used to pixel-wise classify a visual scene. The pixel-wise segmented map is used to estimate the traversability of different environments. The semantic segmentation accuracy can drop if environmental conditions change. The introduction of shadow in images can cause the segmentation network to misclassify pixels, which leads to an
inaccurate traversability estimation. This inaccurate estimation could lead an autonomous vehicle to deviate from traversable paths, leaving it unable to continue or even cause accidents.

To increase the segmentation accuracy in shadow conditions, shadow removal before semantic segmentation is proposed. In this research, the supervised Dual Hierarchical Aggregation Network (DHAN) and unsupervised cycle-based Shadow Generative Adversarial Network (SGAN) are used for shadow removal before semantic segmentation with Segmentation Network (SegNet). The shadow removal networks are evaluated on two datasets, containing image triplets, consisting of shadow, shadow-free and shadow-mask images. The structural
similarity is calculated for complete images and the non-shadow regions by inverting
the shadow-masks. The networks are applied to the Cambridge-driving Labeled Video Database (CamVid) dataset to evaluate the change in segmentation accuracy. In a second set of experiments, the DHAN shadow removal network is retrained on multiple datasets containing synthetic shadows. The obtained DHAN networks are tested on shadows with increasing intensity. The retrained DHAN networks are compared to evaluate the training dataset for shadow removal to increase segmentation accuracy.

The experiments show that the DHAN increases the structural similarity of 99.8% and the SGAN for 74.5% of the datasets. After retraining on the synthetic shadow dataset, segmentation accuracy after DHAN shadow removal increases the segmentation Pixel Accuracy (PA) for a maximum of 91% and the segmentation mean Intersection over Union (mIoU) for a maximum of 88% of the images of the CamVid dataset. We conclude that segmentation accuracy increases after DHAN shadow removal if the DHAN is trained on the synthetic shadow dataset.

Railway networks are subjected to disruptions on a daily basis, which may make the timetable unimplementable, which in its turn may significantly influence passenger satisfaction. In practice, train dispatchers are responsible for mitigating the influence of disruptions, such as delays of trains, train cancellations, and overcrowdedness at stations. The solutions they propose are highly dependent on their experience, often resulting in low quality solutions. In addition, the disruptions must be solved in a matter of minutes, which is challenging because of the problem scale and computational complexity. Effective models and solution approaches are required to mitigate the influence of disruptions.

In recent decades, railway rescheduling models have been developed to support train dispatchers and to improve rail services. A recent and promising model is the event-activity network model, which is a graph-based formulation that supports a wide variety of rescheduling measures. This thesis extends the event-activity network model by including rolling stock circulation with depot entry and exit operations to increase the practicability of the operator-centric model. In addition, a passenger-centric model is proposed by embedding detailed passenger-related factors into the operator-centric model, where the train capacity is included, and the detailed number of passengers in the railway network is calculated. Therefore, the effect of delays on passengers can be handled properly. The passenger-centric model can help minimize the number of waiting passengers on platforms to avoid overcrowding and to improve passenger satisfaction. In practice, the resulting passenger-centric mixed-integer linear programming (MILP) problem is hard to solve due to the introduction of binary variables for train orders, which are important for calculating the detailed number of passengers. An adaptive large neighborhood search (ALNS) algorithm is introduced to address the complexity due to train orders and to improve the computational efficiency of the passenger-centric method. With properly designed destroy and repair operators, the ALNS algorithm can explore the solution space efficiently. Therefore, a balanced trade-off between solution time and quality can be made.

Case studies are conducted based on the train lines operating between the stations of Utrecht and 's-Hertogenbosch in the Netherlands. The simulation results show that the developed model can explicitly include the number of passengers while considering the rolling stock circulation plan. Compared to directly solving an MILP problem using a commercial solver, ALNS can calculate solutions more efficiently while maintaining a high level of solution quality.

In this thesis, an economic-engineering model of the ripple effect in the housing market is put forward. The ripple effect in the housing market is modeled by integrating into one model the local price dynamics of the rental market, the spatial price dynamics between local rental markets, and the valuation of real estate. Such a model is lacking in current housing-market literature.

The economic-engineering model design has two main aspects. First, economic-engineering model design uses analogs between mechanical systems and the housing market, ensuring that the model contains only economically and mechanically interpretable parameters. In this way input optimization and parameter optimization have a direct implementation in the real world. Second, economic-engineering model design relies on classical-mechanical modeling techniques that make the model suitable for optimally and robustly determining governmental policy using control formalism.

This thesis takes the perspective of the housing market as a heterogeneous economic space, where the dimensions are geographic proximity and economic influence. The model put forward in this thesis spatially discretizes this heterogeneous economic space into homogeneous local markets, such that local price dynamics govern the local markets and inter-local differences add spatial price dynamics.

The model put forward in this thesis is designed to predict where shortages will arise due to the introduction of rent control. Such a model guides policymakers in where to build for effectively relieving shortages in the housing market. Finally, such a model informs investors about the value change to expect due to policy changes and migration trends.

Mapping an environment with a Light Detection and Ranging (LiDAR) sensor through the use of a LiDAR Simultaneous Localization And Mapping (SLAM) algorithm is a powerful technology that allows for the creation of detailed 3D models. Recently various LiDAR sensors have been developed based on Micro-Electro-Mechanical System (MEMS) technology. These LiDARs are very low cost and considerably smaller than conventional LiDARs. They also often incorporate other sensors such as Inertial Measurement Unit (IMU)s and cameras into the same device.
Performing LiDAR SLAM with MEMS based LiDAR is challenging due to the short range, the smaller Field of View (FOV) and the sensitivity to ambient light of MEMS based LiDAR. In this thesis the objective is to reduce the effect of these factors when doing LiDAR SLAM by incorporating IMU measurements into the position estimation of the sensor.
A graph based positioning approach is proposed to achieve tight coupling of the IMU sensor and LiDAR position estimates. The method is made more robust by incorporating an outlier detection mechanism that reduces the influence of wrong LiDAR position estimates caused by insufficient points in the LiDAR FOV or by ambient light disturbance.
The method was built in ROS and implemented on the Intel ® L515 sensor. The performance is evaluated in indoor situations with varying presence of ambient sunlight and where room size approaches the maximum limit of the sensor range. The algorithm achieves lower drift than the current state of the art for the Intel ® L515. The algorithm especially achieves altitude drift reduction and increases robustness to outliers in the LiDAR positioning.

Maintenance is a necessary measure to keep the asset, in this case a road network, in good condition. Spending too much on maintenance is clearly not efficient, while not spending enough may cause the condition to drop below a desired value; moreover, it will almost always cost more to correct the emerging damages afterwards.
In this report, a moving horizon optimization approach is presented as a conceptual model to improve the efficiency of maintenance of a road network, compared to the currently used maintenance approach.

The platooning of trucks provides significant benefits in the existing transportation systems regarding social, economic and environmental aspects. Due to the platoon's high dependence on sensors, it is critically important to have fault-tolerant countermeasures that deal efficiently with failures by achieving minimal risk condition. This thesis tackles designing an emergency escape manoeuvre for a faulty truck that has lost its ability to monitor the driving environment while moving in a platoon formation on a highway. A fallback strategy is proposed that leads the faulty truck to park on the shoulder of the road and allows the rest of the platoon participants to continue their journey. To solve this problem, first, a functional platooning system should be designed. Therefore, a nonlinear bicycle dynamic model is employed, while a new Bidirectional (BD) Cooperative Adaptive Cruise Control (CACC) system is utilized for control purposes. The emergency escape manoeuvre has to be pre-computed and ready to be triggered when the failure occurs, as concerning defined specifications, limitations and constraints. It is generated using a quintic splines methodology, while a new optimization approach regarding the minimization of jerk peaks is adopted, enhancing the efficiency of the design procedure. Furthermore, a gap-closing controller is employed and tuned accordingly to reform the platoon after the faulty truck abandons the formation. Moreover, a suitable decision logic is defined to achieve smooth transition among regular performance and fallback state at both platoon and inter-vehicle level. The functionality of each component is firstly evaluated individually, based on specific performance metrics. Finally, all the mentioned components are put together and the fallback strategy that executes emergency escape manoeuvre in platoon context is derived. Several simulation tests illustrate proof of principle for the proposed framework. It is shown that the proposed strategy indeed achieves minimal risk condition for the faulty truck achieving minimal tracking errors. A global applied strategy is adopted for the gap-closing controller by making healthy follower(s) travel with the maximum possible acceleration. At the same time, the leading truck of the formation maintains a constant speed until the time gap between them becomes $0.8s$, before the switch back to the CACC, to complete this action as fast as possible and safety issues in the driving performance. Additionally, the CACC system is evaluated to ensure that the string stability criterion is satisfied in the mode of operation. Moreover, a virtual leader mechanism is introduced to fill in faulty truck's blind spot in the road that arise from the sensor failure. It is shown that it can provide a sufficient window of movement for the platoon before switching to the fallback strategy.

The DelFly Nimble is a type of tailless flapping-wing micro air vehicles (FWMAVs) that has received an increasing amount of attention. FWMAVs show efficient and agile flight possibilities at small scale. The aerodynamics and dynamics of these flapping vehicles are challenging and not fully understood. In this work, a strategy to implement a 3D dynamic model and a trajectory control algorithm is proposed for the DelFly Nimble by the use of a quasi-steady state framework. The design of a suitable tracking control algorithm is the essence of this task. A globally defined smooth nonlinear geometric framework of the flapping-wing vehicle’s rigid body dynamics is introduced as a basis for the analysis. This grants an unambiguous coordinate-free dynamic model in which problem of singularities are avoided. The Nimble has four inputs used to control the six translational and rotational degrees of freedom. A nonlinear tracking controller is chosen on the special Euclidean group 푆퐸(3) for the underactuated aerial vehicle, where position and yaw trajectory tracking are achieved. The full system is classified into the coupled attitude and position subsystems. Using the Lyapunov Stability theorem, the nonlinear controller is shown to achieve almost global asymptotic tracking of the attitude error dynamics of the Nimble and almost global asymptotic tracking of the position error dynamics of the center of mass of the Nimble, enabling sufficient tracking of aggressive maneuvers. Finally, the dynamic model and the controller are examined with numeric simulations. From the results can be concluded that the nonlinear control design allows for aggressive aerobatic maneuvers while maintaining stability of the closed-loop system, provided that the control inputs and damping forces remain moderate.

During the preparation for the Olympic Sailing Competition, held in 2021 in Tokyo, Japan, the Dutch National Sailing Team encountered days with unpredicted wind behaviour. To gain more understanding in the wind patterns occurring, a deep learning based approach is taken. The goal of this research is to find out if unsupervised learning methods can contribute to wind pattern classification. It can then be investigated if the classification can increase understanding in specific wind patterns. The input data for the unsupervised learning model consists of 40 years of reanalysis wind speed data of an area including Japan. To classify the wind patterns, the dimensionality of the input data is reduced using different autoencoders. This reduced dimensional form is then clustered using K-means clustering. The results of the K-means algorithm are compared and the best autoencoder is chosen. The resulting clusters are analyzed for extreme wind patterns, such as typhoons. It is expected that these wind patterns will be clustered together. To check this, the cluster containing typhoon Jebi, the typhoon which caused the highest insurance cost ever in Japan, is analyzed. If this cluster contains typhoons, unsupervised learning is able to provide useful information regarding wind patterns. The best working autoencoder used in this research is the 3D CNN autoencoder. Using the 3D CNN autoencoder, some clusters with specific wind patterns are found. The cluster containing typhoon Jebi consists of 95.8% of typhoons, from which it can be concluded that unsupervised learning is a valid method for wind pattern classification.

This research aims at determining a suitable order release strategy for a warehouse featuring a shuttle-based storage and retrieval system with multiple goods-to-person picking workstations. These picking workstations facilitate efficient picking; however, current release strategies do not take similarity between orders into account, while this could potentially lead to better capacity utilisation. A meta-heuristic approach has been developed in literature to exploit this similarity, thereby using less stock retrievals to fulfill customer orders. However, its capabilities are limited. In this thesis, the meta-heuristic is extended to deal with multiple pick stations and a stock-multiplicity constraint. The algorithm consists of multiple steps. First, a random-start greedy algorithm generates an initial solution for a large set of orders. However, this initial solution will be of poor quality in general. As the computation cost scales exponentially with the order set size, heuristically optimising the entire order set is infeasible in practice. Secondly, the initial order sequence is split into subsets of equal length, each of which will be picked by a separate picking workstation. The picking workstations are then sequentially optimised using the meta-heuristic approach assuming an infinite stock. Afterwards, a check on the stock-multiplicity constraint is conducted, i.e. it is checked if the warehouse is capable of supplying all picking stations simultaneously. If an order line violates this constraint, the order is selected and put in a random position of the order completion sequence of that picking station. It was found that this method is capable of avoiding violations of the stock-multiplicity constraints, although it was also found that this constraint will hardly be violated in a realistic goods-to-person warehouse setting, i.e. with a large number of orders, picking stations, and order lines per order. Furthermore, our simulations show that the algorithm is capable of significantly reducing the number of stock retrievals per picking workstation.

The integration of variable renewable energy sources (RESs) in the electrical power grid leads to larger and faster variations in the power demanded from controllable power sources. This is a problem, because flexibility of (base load) power plants is limited. Solid oxide reversible cells(SORCs) can be used as load-shifting devices to reduce these power variations by converting electricity to hydrogen (solid oxide electrolysis cell (SOEC) mode) when power demand is low and converting hydrogen to electricity (solid oxide fuel cell (SOFC) mode) when power demand is high. However, the introduction of SORCs is challenging. It is a promising long-term energy storage technology, but it is in its development stage. Apart from prohibitive costs, challenges also lie within durability and efficiency under dynamic operation. Development of control strategies is essential for maintaining optimal operating conditions. Therefore, this study researches the ability of SORCs to operate in a mixed power grid by developing an SORC model and power disturbance rejection controller which ensures safe operating conditions.
A dynamic 0D SORC model was developed. It describes a single cell at the center of a large stack of identical cells, which makes it representative for large-scale SORCs. The model is based on SOFC models and uses the current density to indicate the operating mode of the SORC. The benefit of this approach is that one continuous model describes both operating modes. Validation of the model is based on comparison of static cell voltage-current density curves from literature and from a small stack experiment. Open-loop analysis of the model showed that the system is stable and can be decoupled. It also showed that development of gain-scheduling controllers was necessary to handle the exothermic, hydrogen consuming SOFC mode and endothermic, hydrogen producing SOEC mode. This motivated the design of gain-scheduling H-infinity tuned proportional-integral (PI) controller, which were used to control the positive electrode, electrolyte, negative electrode (PEN) structure temperature and fuel channel composition by manipulating the air and fuel flow rate, respectively. Two methods were compared for specifying the performance of the controller. The first method was based on the desired closed-loop bandwidths and the second method was based on the bandwidth of the disturbance. The first method was superior to the second method, because the obtainable closed-loop bandwidths are faster than the bandwidth of the disturbance.
This study shows that gain-scheduling PI controllers allow SORCs to be used for load shifting applications in a mixed power grid. Further research is needed to validate the dynamics of the model and to identify the influence of balance of plant (BOP) dynamics on controller performance.

Time-discounting in behavioural economics is modelled using mechanical system dynamics through the economic engineering framework. The economic engineering framework is being developed at the Delft Center for Systems and Control, and uses mechanical system dynamics to model economic processes and systems. Time-discounting is the calculation of the present value of the received utility from future consumption. Presently behavioural economists have
not been able to reach a consensus on how to model time-discounting behaviour. Two theories dominate economic literature: exponential discounting theory and hyperbolic discounting theory. These theories are treated separately by economists and have separate fields of application. Exponential discounting theory and hyperbolic discounting theory are shown to be related through the dynamics of the damped harmonic oscillator. Exponential and hyperbolic discounting theory are linked to the dynamics of the critically damped and overdamped mechanical system respectively. The dynamics of the underdamped mechanical system are linked to the time-discounting behaviour of a trader. Moreover, the parameters of the damped harmonic oscillator are interpreted economically, resulting in the following analogues: the natural frequency is analogous to the risk-free discount rate, the damping ratio is analogous to time-preference, and the real part of the eigenvalues are analogous to the exponential discount rate. Modelling time-discounting using mechanical system dynamics therefore results in a time-discounting model based on economic first principles.

Combined chemo-radiation treatments are often used to enhance tumor cell kill, with respect to their separate treatments. Currently, chemo-radiation therapy treatments are prescribed as a separately optimized radiotherapy treatment with the addition of a generic chemotherapy regimen. The dosing of the combined therapies is mostly based on clinical experience. Hence there is a clear need for the optimization of dosing for combined chemo-radiation therapy treatments. The goal of this thesis is to propose novel approaches to optimize cancer treatment efficiency when combining chemotherapy and radiotherapy by using control-theoretical methods as well as radiation therapy models. A positive switched system is a dynamic system consisting of a set of Metzler state-space matrices and a specific switching law, which determines when and how to switch between the subsystems. Positive switched systems seem particularly appropriate to model cancer evolution under different treatments, and therefore to determine optimal treatment scheduling. They are used to model the effect of combined chemo-radiation treatments, thus enabling the systematic design of optimal therapy planning. The proposed model is especially appropriate for heterogeneous tumors, since it describes the impact of therapies on different tumor cell lines, and includes the possibility of mutations. Metastases are also included since the model considers separate tumor compartments, and the possibility of migrations between compartments are modeled as well. Undesired consequences due to toxicities are incorporated, and an upper bound can be set to limit the possible damage. A clinical dataset was obtained from the Erasmus MC, which was used to validate chemotherapy-alone model outcomes. Combined chemo-radiation treatments were also developed for a set of initial tumors. The model shows promising applications for cancer treatment design. The use of positive switched systems for combined chemo-radiation treatments is a new approach to obtain optimal drug usage and radiation fractionation schemes for each body compartment. The achieved findings could have a significant impact on effective treatment planning by choosing optimal drug durations or by using radiotherapy in combination with chemotherapy, for instance, to target certain chemotherapy-resistant areas, to achieve tumor control with minimal side effects.

Successful implementation of fully autonomous vehicles is a much desired objective. Within the area of unmanned marine craft or unmanned surface vehicles (USV), the implementation of truly autonomous navigation is still a challenge. During times in which humans are still actively involved in the processof maritime navigation, regulations are needed to prevent accidents from happening. It is therefore essential for the transition towards more autonomy that USVs are able to obey the regulations as well and clearly exhibit the right behavior to enable other ships to anticipate the behavior of USVs and reactappropriately and timely. In maritime navigation, ships have to obey the International Regulations for preventing collisions at sea, also known as COLREGS (COLlision REGulationS). This thesis is focused on the guidance of USVs, and is specifically aimed at COLREGS compliant path re-planning to avoid potential collisions. A first method has been conceived based upon the SaturatedFast Marching Square method where trailing points are used to correctly guide the USV past dynamic vessels according to COLREGS. A second novel method builds upon the previous one, by combining the Saturated Fast Marching Square method with a path generating Genetic Algorithm. The first methodperforms as supposed in simple environments such as when encountering a single dynamic obstacle. However, the first method starts to under-perform as complexity increases through the presence of both static and dynamic obstacles. The additional second method performs accurately in complex scenariosas well. Both methods have been tested in single case simulations and in extensive random simulations to assess their COLREGS compliant path re-planning accuracy.

Because of the increasing e-commerce volume, resulting in increasing demands on the speed of delivery, logistical processes become more and more automated. Order picking is one of the last tasks that is done by humans in warehouses, because humans are flexible with respect to the large variability and changeability in items. However, it is a labour intensive and monotonous task, resulting in fatigue and a shortage of order picking personnel. This motivates the development of automated pick-and-place systems.
One of the challenges for such systems is the heterogeneity of items. In warehouses there is a big diversity in items so the system has to be able to deal with all of them. Another challenge is dealing with items that are deformable. Current systems often make use of suction cups but integrating sensors that can be used to handle deformable items is hard. Fingered robotic grippers have more potential in grasping these kind of items, but grasping deformable items is one of the least addressed topics in robotics. Therefore, the objective of this thesis is to design a control strategy for a fingered robotic gripper to grasp and hold deformable items in a pick-and-place task.
Inspired by the underlying principles that humans use to execute a pick-and-place task, a multi-level controller is proposed for a three-fingered gripper with capacitive pressure pads. The multi-level controller consists of a low-level computed torque controller and a high-level numerical optimisation based extremum seeking controller. The computed torque controller uses an internal model of the kinematics and dynamics, which is derived with screw theory, to compute the torques required to comply with the fundamental grasping constraint and the setpoint on the gripping force. The controller is tuned in such a way that the grasp quality is maximised, given a constant reference gripping force. Because of the fact that the properties of the items are unknown, an intelligent control system has to be able to determine the gripping force setpoint autonomously. This is the task of the high-level controller, that uses tactile sensors to derive the slip. This slip is used to determine the setpoint on the gripping force that the low-level controller has to follow, while maximising the grasp quality and not damaging the products as a result of applying excessive gripping force.
The proposed control strategy is tested and tuned in a simulation environment. The pick-and-place task is executed for the products from a virtual product inventory. The controller is optimised with respect to the control goal on a wide variety of deformable items. Designing controllers according to the proposed principle will increase the diversity of items that can be handled in a pick-and-place environment, while increasing the quality of the grasp and minimising the risk of damaged products.

The demand from the automotive industry for thinner, stronger and more ductile steels led to development of the so called Advanced High Strength Steels (AHSS). These new steel grades are produced at one of the hot dip galvanizing lines, “Dompel Verzinklijn 3” (DVL3), at Tata Steel IJmuiden. The annealing furnace in DVL3 consists out of a number of consecutive sections. One of these sections is the Slow-Cooling Section.

The Slow-Cooling Section, a so-called hybrid furnace, is installed with both heating and cooling capabilities to effectively adjust the temperature required by the metallurgical prescription of a specific steel strip product. The required temperature and associated cooling or heating demand can vary from strip to strip, depending on the desired mechanical properties and strip dimensions, which complicates the control of the Slow-Cooling Section. This is further complicated by process variations in production speed and system delays in control instrumentation. It becomes apparent that in order to produce these high quality steel grades, an improved solution to achieve precise temperature control in the slow cooling process has to be found.

In this research a Model Predictive Controller (MPC) is developed for the control of the Slow-Cooling Section of DVL3. This advanced control system is based on a mathematical model, developed in the first part of this research, incorporating the heat transfer from and to the steel strips inside the Slow-Cooling Section. The nonlinear furnace model is validated with historical furnace data, showing good agreement for various process conditions. MPC is applied using two different methods. The first method is based on one linear internal model. The second method approximates the nonlinear mathematical model by the use of piecewise affine models. Case studies are performed to test the performance of the new controllers. The performance of the new controllers are compared with the current temperature control of DVL3.

A critical discussion is presented based on the results of both Model Predictive Controllers. It is shown that the predictive ability, especially in transient conditions is improved with the newly developed MPC and more accurate temperature control is achieved. Recommendations are given to further improve the control performance.