Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Most existing HVAC (Heating, Ventilation and Air Conditioning) systems control a spatial and temporal average level of environmental variable(s) (e.g., temperature, humidity and gas concentration, …) based upon feedback of inside environmental variable(s) measured at a single position in the whole room.Such control systems make use of set-points of these inside-environmental-variables that are assumed to be; (1), homogeneous inside the whole space of the indoor environment; (2), optimal for a defined theoretical population ‘average’ for the considered species of occupants. However, in reality, such environmental variables are never perfectly mixed but rather revealing a three-dimensional gradient due to the spatially and temporally heterogeneous heat and mass transfer phenomena inside the ventilated structures. Furthermore, living organisms (human, animal, plant or biological product) they are Complex, Individually different, Time-Varying and Dynamic ‘CITD’ systems. Consequently they are responding in individually different ways to changes in surrounding environmental variables rather than the ‘average organism’ concept due to several exogenous and endogenous variables, which also influence the bioresponses of these organisms. Every living organism reacts to its own individual ‘micro-environment’, which is defined as the collection of all environmental variables that reach the individual living organism. The micro-environment around the living organism is considered as a well-mixed-zone, which can be defined as the maximal 3D zone of improved mixing with acceptable low spatial gradients in heat and mass variables (micro-environmental variables such as temperature, humidity, …).A first part of this work focused on active control of the well-mixed-zone of temperature in a ventilated structure.The Data-Based Mechanistic approach was used to model the dynamic spatial temperature distribution of well-mixed-zone in a ventilated airspace to changes in the heat supply (control input). Dynamic response of indoor temperature to varying heat supply could be explained by a second-order transfer function model with a high coefficient of determination (R2T > 0.99), a low Young Identification Criterion (YIC < &#8722;2.3) and a low model standard error (SE < 0.028 &#9702;C). The physically meaningful model parameters as local heat load fraction (J/s) and the coefficient of thermal impedance (&#9702;C/J) were revealed. These two parameters were defining the relation between the heat and temperature spatiotemporal distribution inside the ventilated airspace (Chapter 2). The identified aforementioned data-based mechanistic models were the bases for a model-based control system to control the micro-environmental temperature in a ventilated test installation (Chapter 3). The designed control system (Proportional Integral Plus, PIP) was robust enough to control the temperature in different well-mixed-zones individually using ventilation rate and inlet temperature as control variables.By modifying the test installation, to increase the controllability of the system, it was possible to form and control two thermal well-mixed-zone of the same temperature (e.g., 37 oC &plusmn;1 oC) sandwiching another with a different temperature (e.g., 30.5 oC &plusmn;1 oC), simultaneously.A second part of this thesis focused on using the measured bioresponses of living organism to control the micro-environment around them.The bioresponses of individual living organisms to changes in their micro-environmental variables (mainly temperature) were modelled using the DBM approach (Chapters 4 and 5).The dynamic thermal bioresponses (eggshell temperature) of incubated chicken embryos were modelled from embryonic day 8 till embryonic day 19 to step changes in their ambient air te mperature. The incubated-embryo system was proposed in the present work to represent the incubated embryo and its surrounding micro-environment. It was shown that a first-order transfer function model was suitable (coefficient of determination, R2T > 0.98 and Young Identification Criterion, YIC < -10.2) to model the dynamic response (eggshell temperature) of the incubated-embryo system between the embryonic days 8 and 13. However, between embryonic days 14 and 19 a second-order transfer function model was more suitable to represent the dynamic response (eggshell temperature) of the incubated-embryo system, which was explained by the thermoregulatory effect of the developing embryo. Analogies between the incubated-embryo biosystem and physical systems (electric circuit and thermal systems) were used to quantify the physiological responses, thermoregulation and thermal capacity, during the biological processes. It was shown that the incubated egg acts as a reactive electric circuit to changes in its ambient temperature.The dynamic behavioural bioresponse (activity and position) of 7 days old broiler chickens in a ventilated chamber were modelled to stepwise changes in inlet temperature and ventilation rate(Chapter 5). A first-order transfer function model structure was suitable (coefficient of determination, R2T > 0.89 and Young Identification Criterion, YIC < -11) to describe the dynamic activity of the birds. It was noticed that the birds were tending to occupy the zones of low air velocity (< 0.2 m/s) when the inside temperature was lower than their thermal comfort temperature (< 28 oC) or when they were under cold stress. On the other hand, the birds were tending more to occupy the zones of higher air velocity when they were undergoing heat stress (>36 oC), which indicated a behavioural response to reduce their heat stress by moving to the zones of high air velocity to increase the convictive heat losses. It was possible to control the activity level and position of a group of chickens by changing their micro-environmental conditions (local temperature and air velocity). However, and surprisingly, by applying such control approach on the individual level it was found that the birds were tending to respond as one group instead of responding individually. Thus, we can say that the social effect of the group, flock behaviour, was more dominating than the micro-environmental effect on choosing their position in the room.A novel non-contact and motion tolerant technique was proposed and implemented in this thesis (Chapter 6) for real-time monitoring of embryo heart rate and vasodilatation and vasoconstriction in incubated eggs. Using a microscope camera (with 640 × 480 resolution) videos of the egg’s vascular structure were recorded with 10 frame per second from embryonic day 13 till embryonic day 19. Using image processing technique different embryogenic motions were captured in a form of time series of dynamic activity index. The measured heart rate of the chicken embryo was within the range (252-295 bpm) during the measurement period (between embryonic days 13 and 19).The cardiogenic signals were identified using signal processing techniques to calculate the heart rate of the embryo. The measured heart rate of the chicken embryo was within the range (252-295 bpm) during the measurement period.

Academic collection --- Theses

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

The poultry production sector is being challenged and a lot of pressure is placed on its producers. Together with the increasing demand for animal welfare, there are many and necessary restrictions and environmental standards. Since most of the poultry production takes place in warm climate areas such as China, Brazil, Latin America and the Asia-Pacific regions, both outdoor as indoor bred animals are threatened by rising temperatures. Meat-type chickens, called broilers, are more sensitive to heat stress than egg-type chickens, called layers, due to their higher body weight and faster growth. Thermal discomfort from the heat poses a serious challenge for sustainable broiler production. Even in more moderate climate regions such as North America and Europe, heat stress occurs more often due to the increase in the number and intensity of heat waves. Several heat stress alleviating techniques are available ranging from management to genetic strategies. One of the more promising strategies is a process called thermal manipulation in which broiler eggs are exposed to high temperatures during their incubation. Epigenetic adaptations will prepare these animals for later life when challenged by heat and this by lowering their body temperature and basal metabolism. Despite the plethora of research conducted, it appears that thermal manipulation affects not only the thermoregulatory system but also performance parameters such as chick quality, hatchability and growth. Contradicting effects are found for similar or even identical treatments indicating the lack of knowledge about the functioning of thermal manipulations and the thermoregulatory system. Additionally, it has been shown that thermal manipulation influences fear and social behaviour in reptiles and the first indications in layers are found. Therefore, it is important to assess these effects as well. In this work, the potential of thermal manipulation is further explored by looking into the mechanisms behind it, investigated from a molecular level as well as evaluating its effects on fear and social behaviour. All of this while monitoring performance parameters. In the first part of this work, the mRNA expression levels of 22 potentially thermosensitive ion channels are measured in the brains of developing broiler embryos. Several of these channels were TRP (transient receptor potential) ion channels as these are known to be thermosensitive and responsible for peripheral and brain temperature sensing in mammals. However, very limited research is conducted in poultry. To study these channels in broilers, eggs were incubated in different temperature profiles (T1 and T2) during which brain samples were collected from incubation day 15 until hatch. We found 8 of the ion channels to be affected by treatment or by the interaction of treatment and day. In general, the expression of potentially heat sensitive ion channels was increased in both T1 and T2. However, the more intense T2 treatment managed to change the expression of more channels than T1. It seems that both treatments had an acclimatisation effect, probably as a result of epigenetic mechanisms, to adapt the animal to a life in which heat stress can occur. As validation, a heat challenge test in later life showed a numerical lowering for the number of mortalities for both T1 and T2, although the difference was not statistically significant. To evaluate the effects of T1 and T2 on fear and social behaviour, seven behaviour tests were conducted throughout the rearing period of 42 days. Both T1 and T2 increased social stress during isolation with T1's effects visible from week 4 onwards, whereas T2 increased the stress levels only in week 6. Interestingly, the social motivation to be with conspecifics of T2 males was increased from week 1 onwards. Prosocial behaviours are crucial for good animal welfare, especially as a high sociality can aid in coping with other stressors. The behaviour effects of two cold incubation profiles were also researched with one referred to as IT treatment, designed to increase performance variables, and the other one referred to as NT, designed to represent a nature like incubation. IT was found to increase the fear of a novel environment whereas NT males showed increased levels of social stress when in isolation. For all treatments, a variety of effects were found on performance parameters. The same as in literature studies. It is likely that the background of the broiler breeder farm and age of the breeders, the exact storage conditions and uncontrolled elements in the incubators such as humidity, O2 and CO2 interfered with the results. Even though some treatments had a negative effect on hatch performance such as lowering hatchability or chick quality, none of the treatments affected their growth during the full rearing period. All in all, treatment T2 seems like a potentially commercially applicable treatment due to its promising heat resilience improving effects, improvements in social motivation and no decrease in performance. Still, many questions are remaining especially concerning the more fundamental research on thermal manipulation, thermosensitivity, integration of signals etc. Epigenetic research is just starting to elucidate further mechanisms behind thermoregulation in general and the way thermal manipulation influences it. Many opportunities remain open in this field of research. The combination of epigenetic studies with computer supported data analysis such as Precision livestock farming techniques analysing biological data in real-time could be valuable for future monitoring and control of the incubation process concerning its temperature profile.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Critically ill patients are admitted to the intensive care unit (ICU) to reduce morbidity and prevent mortality related to acute illness, surgical procedures, or trauma. They are characterized by severe illnesses of large diversity, resulting in the need for continuous monitoring. A consequence of this need for intensive treatment and continuous monitoring is a significantly higher cost per patient compared to patients in general hospital wards. The global number of intensive care patients is expected to grow during the coming decades as being noticed during the pandemic nowadays. This growth would require more equipment, medical staff and an efficient decision making process. Increasing the operational efficiency of the ICU is crucial to provide the best patient care and sustain costs. Possible solutions to this problem are the implementation of new technologies and the use of patients’ data for predictive machine learning models or data mining. In this thesis, several machine learning algorithms are used to gain a clinical insight into ICU patients’ mortality or survival. The aim of this thesis is to investigate the error performance and the interpretability of these machine learning algorithms with manually engineered features that are designed to obtain a clinical insight. The total number of the engineered features was 294. These features comprise statistical, dynamic and physiological features based on the patients’ vital signs, as well as features containing information about comorbidities and reasons for hospitalization. Sequential forward floating feature selection is performed to obtain the best performing and most relevant features for each model. The used techniques are logistic regression, decision trees, and Gaussian naïve Bayes. The used dataset are of 797 ICU patients from the hospital Ziekenhuis Oost-Limburg, Genk, Belgium. The data contained time-series data of the patients’ vital signs, as well as other patients’ information. Training the algorithms on the total patient population resulted in poor mortality prediction performance. Therefore, the patients’ population is split up into subpopulations based on the reason of hospitalization. A special focus was on the patients admitted due to an out-of-hospital cardiac arrest. This subpopulation only contained seventeen patients. Decision trees were trained with windowed features on the different subpopulations, and showed an average prediction accuracy, precision, recall, and F1-score of 90.0, 87.1, 72.6, and 78.7% respectively across all subpopulations. For the decision trees using non-windowed features, the average accuracy, precision, recall, and F1-score were 87.4, 80.7, 64.4, and 70.9% respectively across all subpopulations. These results indicate that it is possible to gain an insight into the patients’ pathophysiology by using a data-based approach using interpretable machine learning models. For the Gaussian naïve Bayes algorithm, the average accuracy, precision, recall, and F1-score were of 88.1, 77.3, 74.0, and 76.0% respectively across all subpopulations. Systolic blood pressure’s features were the most frequently selected features, meaning that this is an informative vital sign for ICU mortality prediction. A large fraction of the selected features considered the patient's evolution, indicating that it is important to consider the evolution of the patients’ vital signs throughout their stay and not only rely on static vital signs.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

When Flemish students are graduated from high schools with a diploma, they have much freedom of choosing subject fields to be enrolled at a public university. In spite of all the benefits of such freedom, students can also get confused and wonder whether they are going to perform well in the field they plan to choose. On the other hand, minimal entrance requirement often leads to a higher dropout rate reaching up to about 30% after the first academic year. In this thesis, only the students enrolled in engineering technology programmes are taken into account as an example. The objective of this thesis is to give advice to high school graduates who plan to follow engineering technology programmes before being enrolled and to help decrease dropout rate in corresponding university programmes by predicting students’ performances with machine learning. Student's study efficiency after first academic year was chosen as the standard for evaluating student's performance. After setting up proper thresholds for the study efficiency, the case became a typical classification problem. Artificial neural networks (ANNs) and support vector machines (SVMs), two of the most popular supervised learning techniques models for classification problems nowadays were applied to build prediction models on MATLAB. A dataset containing student samples of the academic year 2015-2016 (N=459) and 2016-2017 (N=414) was used to train and evaluate our models. Each sample contained 22 parameters, which were the features chosen to describe the student. Three thresholds of study efficiency were chosen in this case: 30%, 50%, and 70%. Two metrics were used to examine the performances of the three classifiers: (1) accuracy of students being correctly classified, which means the probability that the prediction made before a student entering university matches the actual performance after first year’s study. (2) Posterior probability, which represents the probability of belonging to a specific class. Three important error performance metrics were used to evaluate the model: (1) sensitivity, (2) precision and (3) F1 score. Based on those metrics, following conclusions were drawn: Firstly, for three-classification, the model was not able to identify the average student. Secondly, for binary classification, the model with 70% threshold had the best performance because it had more balanced data input. The model with 30% threshold had the worst performance on sensitivity, precision, and F1 Score due to data imbalance. Thirdly, the model performances after the feature selection hardly improved.