1,721,234 research outputs found
Range measurements of a High Frequency Radio Frequency Identification (HF RFID) system for registering feeding patterns of growing-finishing pigs
No full textMonitoring the feeding pattern of a pig enables early detection of diseases and other problems. To monitor the individual feeding pattern of group-housed pigs, it has been suggested to equip the pigs with High Frequency Radio Frequency Identification (HF RFID) tags and the feeding trough with an antenna. The detection range of the HF RFID system is crucial to guarantee that all feeding pigs are detected without detecting the pigs located further from the feeder. The current study examines the factors that influence whether an antenna attached to a round feeding trough (such as those used in group housing of growing–finishing pigs) detects stationary HF RFID tags placed in various orientations and distances from the antenna. Four experiments were performed using a custom-built test set-up that allowed determining the RFID registrations for 70 tag positions, at seven distances from the antenna and for seven orientations of the tags in relation to the antenna. In the first experiment there was determined that which tag side is closest to the antenna had very little influence on the range of registration. The results of the second experiment revealed that all eight HF RFID antennas in the pig house performed similarly, with symmetry observed in their range of registration. In the third experiment the range of the HF RFID system was measured while accounting for tag, tag position and tag orientation, whilst the last experiment was designed to test the effect of interference between tags. Reproducibility between (the order of) the tags and the average agreement between five repetitions of all tests was very high. In total, the sensitivity was 51.0%, with a standard deviation of 43.1percentage point (pp). The specificity was 87.1% with a standard deviation of 19.4pp. It was concluded that the performance of the HF RFID system in terms of sensitivity and specificity of the range depends greatly on the height and orientation of the tags. This causes irregular gaps to appear between subsequent RFID registrations of a feeding pig. To improve the performance of the system in practice, it is suggested to adjust the height of the antenna to better match the size of the pigs and to develop algorithms and criteria to merge raw RFID registrations into relevant feeding variables for individual pigs.sponsorship: Jarissa Maselyne is funded by a PhD grant from the Agency for Innovation by Science of Technology (IWT Flanders - project SB 111447). These results are part of the ICT-AGRI ERA-net project PIGWISE "Optimizing performance and welfare of fattening pigs using High Frequent Radio Frequency Identification (HF RFID) and synergistic control on individual level" (Call for transnational research projects 2010). Special thanks go to the technical staff of ILVO for the work and technical support provided during the course of the project and to Miriam Levenson for English-language editing. (Agency for Innovation by Science of Technology (IWT Flanders)|SB 111447)status: Publishe
Laser verstrooiingsbeeldvorming voor de niet-destructieve inspectie van agro-voedingsproducten
No full textIn the agro-food industry, there is an increasing demand for non-destructive, fast and cost-efficient methods for the objective determination of product quality. Several parameters influence the quality of a product, while the importance of each of these parameters may vary amongst people. Optical measurement techniques are often used because of their non-destructive nature. However, the robust determination of both chemical and physical quality attributes remains difficult. In this work, a non-destructive determination of quality was performed using laser scatter imaging. This technique allows to obtain more information on both the absorption and scattering of light, by retrieving spatial information. The bulk optical properties (bulk absorption coefficient, bulk scattering coefficient and scattering anisotropy factor) are used to characterize absorption and scattering properties. The absorption of light is related to the chemical composition of a product, like present pigments, water or sugars. Scattering of light is more related to the physical structure, possibly allowing to retrieve more information on physical quality attributes, like firmness, tenderness or porosity. Optical properties of light were derived from the obtained scatter images using a data-based modelling technique, possibly allowing a better prediction of product quality. This approach was tested in two case studies, apples and bovine meat, selected because of their economic importance for Belgium.
First, a hyperspectral laser scatter imaging (HLSI) system was developed. A combination of a supercontinuum laser with monochromator was used to illuminate samples with monochromatic light in the 550 nm to 1000 nm range, while a CCD camera was used to take images of the diffuse reflectance glow spots. From this glow spots, a diffuse reflectance profile with the light intensity relative to the distance from the point illumination was obtained. Models were constructed to optimize the detector size and the source-detector distance, estimating different quality parameters of Braeburn apples. A detector size of 0.82 mm was found to be adequate for the estimation of all parameters, including the starch value, firmness, SSC and Streif index. Different source-detector distances were found to be of importance for predicting different quality traits. Photons exiting the sample closer to the point illumination, which have interacted less with the sample, were more important for SSC prediction, while the prediction of physical parameters like firmness relied on photons which had more interaction with the sample. Moreover, using variable selection, the wavelength regions of pigment and water absorption were found to be most informative. A classification of apples into ripeness classes based on these models was possible, with a misclassification rate of 12.5%. Nevertheless, these models still used mixed information, including both the effects of absorption and scattering.
Using the double integrating spheres (DIS), the golden standard method for determining bulk optical properties (BOP), the interaction of light with both apple and bovine meat samples was studied. Both the apple skin and cortex were studied separately during maturation of bi-colored (Braeburn and Kanzi) and green cultivars (Greenstar). The bulk absorption coefficient µa of the skin showed features of anthocyanins at 550 nm, chlorophyll at 678 nm and water at 970 nm, 1200 nm and 1450 nm, while the µa of the cortex showed an overall lower absorption attributed to carotenoids, chlorophyll and water. During maturation of apples, an increase in the absorption by anthocyanins was observed in the red cultivar’s skin, while a decrease in absorption by chlorophyll was observed in the cortex. Both the bulk scattering coefficient µs and anisotropy factor g of the skin were significantly higher in comparison to the cortex. Both skin and cortex were found to be highly forward scattering with anisotropy factors above 0.9 in the entire wavelength range between 500 nm and 1850 nm. During maturation, no clear evolutions in the anisotropy factor were observed, while µs decreased in the fruit cortex. It was hypothesized that the shape and size of the scattering particles hardly changed during maturation.
The DIS analysis showed changes in the optical properties of apple during maturation. This indicates that the non-destructive estimation of BOP could be beneficial in determining apple quality. To go from scatter images towards an estimation of the optical properties, a data-based model was used. To train this model, a set of optical phantoms with known optical properties was measured using the HLSI system. These diffuse reflectance measurements, in combination with the BOP from the DIS, were used as an input for a metamodel, linking the BOP to diffuse reflectance profiles. The metamodel showed a good performance for a set of validation phantoms, with an R2V of 0.9977 and 0.957 in combination with an RMSEV of 0.20 cm-1 and 3.21 cm-1, for respectively µa and µs’. Nevertheless, at wavelengths with extreme BOP values, the predictions were less accurate. The prediction of apple BOP showed an expected course for the µa spectra, with absorption features of anthocyanin, chlorophyll and water. However, an incomplete separation between µa and µs’ was obtained, as µs’ still showed some distinct absorption features. Nevertheless, at wavelengths with low absorption, the estimation of µs’ was according to expectations. In addition, the same evolutions in BOP as found with the DIS setup were also found with the non-destructive HLSI technique. Though, no clear relation was obtained between apple quality and the estimated BOP. Possibly, the low variability of both SSC and firmness during maturation, together with a high variability amongst apples from the same harvest day, could have complicated the prediction models. Moreover, still mostly information on the changes of apple pigments was used in the models.
Next, two bovine muscles were measured using the DIS as well. Both the longissimus lumborum (LL) and the biceps femoris (BF) were considered, while the BF was further divided into an outer and inner part, due to two-toning. Clear absorption features of myoglobin, mainly oxymyoglobin at 544 nm and 582 nm, and water were found in the µa spectra. A higher absorption of myoglobin was found in the BF samples, while also a higher µs and a lower g was found in this muscle compared to the LL. The inner and outer BF showed significant scattering differences, possibly related to an increased degree of protein denaturation in the inner BF. During wet aging, when meat tenderness increases, a decrease in the measured g was noticed in both muscle samples.
When measuring muscle samples using spatially resolved spectroscopy, anisotropic light scattering by the present muscle fibers should be accounted for. By measuring muscle samples with different initial fiber orientations, it was shown that muscle fibers can change the shape of the diffuse reflectance glow spots from circular to a rhombus shape. This effect was mainly present in the samples with the muscle fibers running parallel to the measurement surface. Moreover, the rhombus’ major axis orientation was related to the distance from the point of illumination. Close to the point illumination the major axis orientation was found to be perpendicular to the muscle fibers, while at larger distance a 90° shift occurred, aligning the major axis with the muscle fibers. In these samples with muscle fibers parallel to the measurement surface, the fiber orientation could be predicted based on the fitted rhombuses, with an R2P of 0.993 and RMSEP of 3.95°. These results show the importance of the 3D fiber orientation when measuring diffuse reflectance signals. Moreover, this 3D fiber orientation could possibly be determined using the obtained diffuse reflectance signals.
The prediction of muscle BOP values using the metamodel also showed similarities with the DIS measurement. Clear absorption features of oxymyoglobin and water were present, while the absorption by metmyoglobin was observed as well, related to the ticker samples measured using HLSI. A thicker sample allows a gradient of myoglobin forms to exist, relative to different depths inside the sample. Again, larger µa and µs’ values were observed for the inner BF samples, while both the outer BF and LL samples showed lower values for µs’. During wet aging, a significant increase was obtained for the µs’ of the LL muscle. Moreover, due to the non-destructive nature of HLSI, measurements were possible through the plastic vacuum pack. Due to the lack of oxygen inside the vacuum pack, mainly deoxymyoglobin with an absorption peak around 557 nm was present. Measurements on the exact same sample during wet aging, through the vacuum pack, also showed an increase in µs’ of the LL muscle. As the changes in meat tenderness were most prominent in the LL muscle, it was suggested that the increase in meat tenderness could explain the observed increase in µs’ for this muscle.
Finally, a limited number of wavelengths were selected to design a multispectral hand-held measurement device. Four laser diode modules emitting at wavelengths of 533.3 nm, 674 nm, 800.7 nm and 981.1 nm, were selected and mounted around a CCD camera. Using shutters, the laser light of the different modules was guided onto the sample sequentially. Again, a metamodel was built by measuring a set of liquid optical phantoms with a wide range of both µa and µs’. In validation, the metamodel showed an R2V of 0.9724 and 0.9377 in combination with an RMSEV of 0.56 cm-1 and 5.13 cm-1 for µa and µs’ respectively. However, for the prediction of apple and pear samples, with and without the skin, an incomplete separation between absorption and scattering properties was obtained, mainly at wavelengths with high absorption values. Nevertheless, the estimation with the HLSI and multispectral device agreed for the apple samples. Moreover, fruit samples of different density showed different µs’ values. Like this, Golden Delicious apples had the lowest density, while the highest µs’ values were obtained.
Based on the obtained results, it was concluded that laser scatter imaging shows potential for the non-destructive monitoring of product quality. The technique can be of added value for applications in which both the evolution of chemical composition and microstructure are of importance. Moreover, the spatial measurements could be useful in characterizing the 3D structure of anisotropic products. Nevertheless, still improvements in the modelling and setup configuration could be introduced before evolving towards measurements in the field or on-site. A first step was made by the introduction of a multispectral, portable and low-cost measurement device. This type of optical measurements could be beneficial in the entire agro-food industry, as the estimation of BOP and the spatial nature of the measurements offers perspectives for monitoring product quality.status: Publishe
Gebruik van expertkennis in de multivariate calibratie van een implanteerbare glucosesensor
No full textThe World Health Organization estimates that 347 million people worldwide are diabetic. Although diabetes is presently not curable, continuous glucose monitoring and strict insulin therapy to control the blood glucose levels in diabetic patients can dramatically delay the onset of serious complications. Near infrared (NIR) spectroscopy offers a promising technological platform for continuous glucose monitoring in the human body. The investigations carried out in this research work focus on utilizing NIR spectroscopic data for building reliable glucose prediction models. Principal Component or Partial Least Squares based regression (PLSR) methods are by far the most often used chemometric approaches for the calibration of spectroscopic glucose monitoring sensors. To ensure good prediction performance of the PLSR model, the regression coefficients have to be estimated using a representative calibration set, i.e., a data set containing all relevant variation in the measured NIR spectra which can be expected in future test samples. One may never be able to generate such a representative calibration set for in vivo glucose monitoring primarily because of two reasons: First, the chemical composition of the biological fluid may vary over time. Second, a thin tissue layer may grow in the optical path of an implantable in vivo glucose sensor. This will have an adverse effect on the signal to noise ratio of the net collected signal, and could dramatically worsen the prediction ability of multivariate calibration models. This underscores the importance of robust multivariate calibration, and characterization of optical properties of biological tissues.Accordingly, in the first part of this work, in vitro multivariate calibration models have been built for aqueous glucose and human serum solutions. As protein molecules are the most important interferents in human serum, the effect of overall serum protein concentration and glycated serum protein concentration on glucose predictions was studied in detail. In the next part of the work, the possibilities to robustify the multivariate calibration model by inclusion of expert information on chemical interferents and scatterers were investigated. To quantify the effect of presence of a thin tissue layer in the optical path, the bulk optical properties of tissue samples grown on sensor dummies which had been implanted for several months in goats were characterized using Double Integrating Spheres and unscattered transmittance measurements. Overall, it was found that the multivariate calibration models can successfully predict the glucose concentration in aqueous and biological media. Robustification of the multivariate models by utilizing expert knowledge was successful especially when Spectral Interference Subtraction was used to preprocess the NIR data, or when Augmented Classical Least Squares method was used to build multivariate calibration model. Based on the optical characterization of tissues, the diffuse transmittance measurement in the combination band of NIR region was recommended as the optimal configuration for an implantable glucose sensor.status: Publishe
Karakterisatie en optimalisatie van het optisch pad verbetert hyperspectrale beeldvorming van vruchten in het SWIR
No full textQuality assessment and process monitoring are essential for today's fruit industry sector and the world's economy. From picking fruit in orchards, to transport and handling practices, to storage and packaging, each step will influence the quality of the end-product when presented to the consumers. The appearance, consisting of the shape, size, colour or absence of any damage are essential criteria relevant to consumers, and influence their will on buying. It is therefore essential to provide a fast, consistent quality assessment of each fruit to match the expectations of the market. To remain competitive to this demand at low cost, fast and efficiently, non-destructive automated quality sorting lines are needed. Among the different defects affecting fruit quality, bruises are one of the most problematic industrial losses. The detection of bruises in fruit such as apples during handling is therefore required. The browning process of bruises results in progressive apple tissue softening and colour changes. As this natural process takes time before it becomes visible, there is a gap of a few days between the mechanical damage causing the bruise and the consequent visible brown spots, which lower the price of apples. It is therefore important to detect bruises on each apple as early as possible after damaging to limit consequent economical losses.
Light is the fastest known information carrier. In ambient conditions, it is also harmless for fruit or the surrounding workers, and a cheap technology. Among the different non-destructive and non-invasive techniques, the usage of light and the analysis of the information it can carry is therefore the most promising path. By shining light onto apples, and observing the absorbed, reflected and scattered light, bruises may be detected non-destructively at high speed. Among the most recent technologies reported, hyperspectral imaging (HSI), being the combination of the machine vision and spectroscopy fields, is showing promising paths. More particularly, the short-wave infra-red (SWIR) range has been demonstrated to promote successful detection of bruises in apples at early stages. However, there are still limiting factors when using SWIR HSI prior its success in industry. Among them, the most predominant are high noise levels arising from the detectors, non-uniform illumination, specular reflections and real-time HSI data handling. This research aims to tackle those problems be first describing and modelling the different components consisting of a SWIR HSI sorting system being the illumination and the imager, and further optimizes their configuration for better image quality. Those building blocks are further put together combined with improved data handling more robust and efficient usage of SWIR HSI in industry. This research is split into 10 chapters.
Chapter 1 covers the current practices in image based fruit sorting, with a stronger emphasis on hyperspectral imaging. It further compares visible and near infra-red (Vis-NIR) to SWIR HSI and where are the additional challenges when using SWIR over Vis-NIR. The chapter then describes the relevance of apples in industry and why early bruise detection. The chapter ends with the outline of this thesis.
Chapter 2 describes the state-of-the-art in light, its interaction with matter, with a focus on polarisation and vibrational spectroscopy. The browning bio-chemical process of apples is further described. The algorithms used to process light spectral information, also referred to multivariate data analysis which are applied within this research are then described.
Chapter 3 is focused on characterising the noise and sensitivity of a SWIR hyperspectral imager, to quantify the signal to noise ratio (SNR). To quantify the pixel-to-pixel variation or the detector’s response, a radiometric calibration method is proposed which dynamically removes the detector noise. This approach removed 6% further noise compared to conventional sequential noise sampling methods. The average detector noise or dark current evolution through time is then shown, which was noted to vary non-linearly, with a sub-linear trend one hour after start-up to stabilize after 3 hours, with up to 12% of the imager’s dynamic range. Contrast of each spectral image is also described using a novel custom-made checkerboard calibration rig. The following showed a ratio per wavelength up to 15000 versus 1 raw values with 100-1700 nm. The checkerboard also enabled accurate spatial calibration using a thin lens model.
Chapter 4 describes how to measure, model analytically and using non-sequential ray-tracing software the spectral and angular distribution of halogen tungsten (HT) spots, considered as the standard in SWIR HSI illumination. The far field angular distribution was modelled with a Gaussian distribution with an R² of 0.99, while the spectra using a Plank based 5th order polynomial with and R² of 0.98. The modelled spectra enabled to convert photometric measurements into radiometric units, and estimate the energy contribution of the spots in the SWIR spectral range, with up to 63% of the total spectral power. Further, near-field spot distribution is measured and modelled within the ray-tracing software, comparing irradiance distributions when using or not diffusers. It was shown that the irradiance patterns could be reproduced with a peak relative error of 12% when using diffusers, while up to 30% without.
Conventional illumination distribution and light beam shaping are non-linear problems, which often are solved using iterative methods such as simplex or simulated annealing (SA) optimization algorithms, which can result in sub-optimal solutions or time consuming searches. Chapter 5 introduces novel constrained non-linear global optimization algorithms which can handle more efficiently such problems while simultaneously offering information on the sensitivity of a configuration near its optimum. A design is proposed using 4 HT spots placed around a flat target, using the source models from chapter 4. The two proposed optimization methods are referred to as Design of Computer Experiments with Design Augmentation (DACEDA) and Design of Computer Experiments with Simplex post-optimization (DACES). A 2 variables analytical version of the problem using isotropic source models enabled to compare DACES and DACEDA’s modelled design space with an overall average relative error of 2%, with a peak up to 10% at the corners of the design space. The SNR of ray-traced near-field sources modelled in chapter 4 is quantified using the Rose model, setting the stop criterion of the proposed optimization algorithms. The simulated irradiance distribution uniformity is then optimized for a 2 and 5 variables case studies with DACES, DACEDA, simplex and SA. In the 2D case, it was shown that DACES performed best after 30 simulations while in the 5D case, DACEDA performed best after 65 simulations. Both algorithms were further used in a case study with DACEDA for tolerance analysis, and DACES for optimization of a configuration for apples, which was used within the remaining of the thesis.
Among the main challenges when using SWIR HSI for fruit quality inspection, are the glossy regions observed from their arbitrary deformed toroidal shape and waxy surface. Therefore, this research further aimed at reducing the influence of those specular reflections, both numerically and optically.
In chapter 6, a first proposed approach is to use chemometrics tools combined with image analysis to reduce or remove those artefacts, using a multiclass classifier or a stepwise approach. The proposed method using iterative steps to remove progressively automatically unwanted regions resulted in 6% higher prediction accuracy than a multi-class partial least-squares discriminant analysis (PLS-DA) classifier. Appropriate wavelength selection using interval PLS-DA enabled to improve further by 4%. Furthermore, the stepwise algorithm enabled to detect for multiple cultivars up to 80% six hours after bruising.
Chapter 7 uses the multi-class PLS-DA classifier from chapter 6 on a real-time case study of one cultivar, and compare pixel based calibration models to conventionally used region based ones. Pixel based models, encountering for variations described in chapter3, improved prediction accuracy at the pixel level up to 2%. With a cultivar based model built for 2 hours after damage, using area normalization as spectral pre-processing and image post-processing, a pixel-based prediction of accuracy of 95.6% was obtained, while up to 98% at the sample level. The following was demonstrated on a real-time SWIR HIS sorting system at a rate of 200 ms per apple at a scanning speed of 0.3 m/s.
To further improve those results, chapter 8 aims at quantifying the degree of glossiness for apples as a function of the light geometrical path, also referred to as surface scatter properties or bi-directional scatter distribution function (BSDF). It was shown that apples have a Gaussian gloss trend around the specular angle, and are Lambertian outside the glossy region.
Moreover, polarization properties of apples are then investigated in chapter 9, in the aim to remove optically gloss arising from apples using a cross-polarized imaging system. It was shown that gloss could be removed for multiple cultivars using cross-polarization, and that the resulting scattered reflected light was Lambertian, thus improving the image uniformity and bruise-sound contrast region.
Finally, the combination of the conclusions drawn from the different chapters and future research perspectives are given in chapter 10. It can be concluded that near-field ray-traced sources with diffusers are the best choice for SWIR illumination, which can be optimized using DACES or DACEDA for improved uniformity. It was shown in this research that real-time SWIR HSI is possible, and using broad spectral has significant added value. It was shown that using the knowledge of polarization and surface scatter properties of apples, linear cross-polarized imaging configuration is the most efficient solution to remove gloss of fruit in images. Alternatives are also possible, by numerically removing gloss using area-normalization and PLS-DA, with image post-processing, which are cheaper, but more time consuming.status: Publishe
Fast ingredient quantification in multigrain flour mixes using hyperspectral imaging
No full textsponsorship: This work was carried out in the context of the iFAST project with the support from Flanders' FOOD and VLAIO (Agentschap Innoveren & Ondernemen), research and innovation program under grant agreement No 140992. (Flanders' FOOD, VLAIO (Agentschap Innoveren & Ondernemen), research and innovation program|140992, Academy of Finland (AKA)|140992)status: Publishe
De ontwikkeling van een automatisch detectiesysteem voor kreupelheid bij melkvee: het GAITWISE systeem
No full textAs a result of the growth in dairy production, dairy farms have intensified with more cattle on fewer farms and higher productivity per animal and per caretaker. Consequently, farmers have less time to observe and monitor cows and technology is being used to support the farmers in their management, especially by monitoring the cows health.Besides reduced fertility and mastitis, lameness is one of the top most costly health problems in dairy cows. Unfortunately, not only its effect on farm profitability (caused by drug treatment, veterinary costs, reduced milk production, reduced reproductive performance and shorter life expectancy) is underestimated, but also its detrimental effect on cow health and welfare. With a lameness prevalence of up to 72 %, the levels in European dairy herds are unacceptably high and hence, minimizing the occurrence is one of the greatest challenges the dairy industry is currently facing. To apply proper treatment farmers must be able to detect their lame cows in the herd in an early stage. Therefore, the general aim of this PhD is to develop an automatic system for lameness detection to help farmers in better detecting lame cows in their herd. To achieve this, a walk-over device - called Gaitwise - with an integrated pressure sensitive mat and specific software was developed. Gaitwise measures spatial (e.g. step length), temporal (e.g. stance time) and force related gait variables of claw-floor interactions of cows walking over the measurement zone. Assuming that these gait variables change when a cow develops lameness, Gaitwise could serve as a lameness detection system that alerts the farmer of cows that show abnormal changes in these variables that are related to lameness. In order for this system to be used in daily practice, measurements were fully automated and gait variables are available in real time. In a next step, two groups of gait variables are calculated: the basic gait variables describing the basic gait of cows and ten more specific gait variables that are often used for locomotion scoring in lameness research (Stride length, stride time, stance time, step overlap, abduction and variables covering asymmetry between left and right limbs in step width, step length, step time, stance time and force). Relevant associations between these measured specific gait variables and corresponding lameness attributes scored by an observer were found. This suggests that the measured variables are functionally relevant for lameness research and detection. Also, a test-retest study revealed that the measurements of the cows walking on the measurement zone are highly reliable. This makes the Gaitwise system a good instrument for gait analysis in the context of lameness detection, because the changes in gait variables due to lameness are likely to be attributed to lameness and not to cow variability. In order to select the specific variables that were most suited for a detection model, the specific gait variables were compared to a reference method for lameness detection usingobserver locomotion scoring of the cows. All the tested variables were different between groups of non-lame, mildly lame and severely lame cows. Variables of asymmetry in step length, asymmetry in stance time, asymmetry in step time and stance time, step overlap and abduction seemed to have the highest potential for the detection of lameness in cattle. The lameness detection model that was built based on these results, showed promising overall sensitivity and specificity, especially in detecting severely lame cows. However, the results of this validation of the detection model revealed that detecting mildly lame cows is the most challenging as the differences with non-lame cows are much smaller. The added value of lameness detection systems would increase considerably when - besides severely lame cows which are more easily spotted by the farmers - also the mildly lame cows could be detected. To improve the detection of mildly lame cases, two approaches were investigated: (1) the potential of other gait variables for detecting lameness in such early stages was investigated and (2) the normal variation of the specific gait variables caused by cow (age, lactation stage, etc.) or environmental (dark environment, wet flooring)) factors was evaluated, because this might cause misclassification of cows and hence hamper the success rate of the detection system. The first approach is based on the fact that in human gait research, increased stride-to-stride fluctuations (i.e. gait inconsistency) were found to be more closely related to early health problems compared to average gait variables. Therefore, the potential of gait inconsistency variables was investigated for early lameness detection in cows. In other words, will a cow that develops lameness first alter its gait by occasionally taking a shorter stride before altering its gait to shorter strides in general? Using two case-control studies, both the basic gait variables and the new gait inconsistency variables were compared between non-lame and severely lame cows and non-lame and mildly lame cows). The inconsistency gait variables were able to show significant differences between non-lame and mildly lame cows, where the more basic gait variables could not. In addition, this data set was used to build a lameness detection model using solely the basic variables and a second model using both the basic and the inconsistency variabels. The second model using the inconsistency variables outperformed by far the model based on only basic gait variables. It was therefore concluded that these new variables of inconsistent gait are promising for assessing lameness at an early stage. In addition, they might even aid in detecting the location of the lameness problem, i.e. which leg is lame. Whether these inconsistency variables are more promising in detecting lameness at an early stage compared to basic or specific gait variables should be investigated in future research using individual detection models. Finally, as cow (e.g. age, gestation stage) or environmental (e.g. wet flooring) factors can also influence cow gait, such changes in gait that are not related to lameness can cause misclassification of cows and hence hamper the success rate of the detection system. Therefore, in the second approach, this normal variation of specific gait variables was investigated in a pilot study. The results show significant influences of wet surfaces, age, lactation and gestation stage on the gait variables. Hence, this normal variation within cows together with the specific way of walking of individual cows should be taken into account when developing a lameness detection algorithm. An individual cow model might therefore outperform the models using group thresholds resulting in better detection of mildly lame cows. This special emphasis on detection of the mildly lame cows will provide more added value for the farmers compared to detecting the severely lame cows.Based on the results obtained in this PhD research, some short-term future work to further improve the Gaitwise system as a lameness detection tool has been suggested. First, future work should focus on pointing out the benefits (less financial losses) to the farmers and decrease the systems production costs. Therefore, the cost-benefit of a lameness detection system should be analysed, the farmers expectations of such lameness detection system should be gathered and more awareness for the effect of lameness on their farm profit and on the health and welfare of their herd should be created. Also possible ways to downscale Gaitwise in order to make it less expensive should be investigated. In addition, several approaches should be tested to further improve the detection level of Gaitwise. The inconsistency variables, adjusted specific variables or other new variables could be tested for their accuracy to detect mildly lame cows. The detection algorithm could be improved by setting individual thresholds instead of group thresholds, by taking the normal variation of gait variables during gestation or lactation into account, and by combining Gaitwise data with data already available on farm and data of other sensortechnologies (accelerometers, StepMetrixTM, etc.).status: Publishe
Automatisch monitoren van eet- en drinkpatronen van vleesvarkens: naar een waarschuwingssysteem voor productiviteits-, gezondheids- en welzijnsproblemen bij individuele varkens
No full textA pig farmer aims to maximize profit whilst maintaining the health and welfare status of the animals at optimal level. This requires adequate follow-up of each individual animal to spot and treat upcoming or present health, welfare and productivity problems. Currently, this follow-up is done through visual observation of the animals. However, with the economically-driven trends towards larger farms and larger groups of pigs in one pen, this visual observation is becoming too time-consuming and difficult. Therefore, in this PhD thesis, an automated warning system for performance and welfare monitoring of individual fattening pigs is developed to support the pig farmer’s activities.
The aim is to signal upcoming performance, health and welfare problems to the pig farmer. Changes in the daily feeding and drinking patterns of an individual pig are used to detect these problems. A High Frequent (HF) Radio Frequency Identification (RFID) system attached to a standard feeding and drinking system used in group-housing allowed to measure this behaviour automatically. Each pig was equipped with a passive RFID tag with a unique number in its ear. These measurement systems were validated thoroughly through comparison with observations. Irregular intervals occurred between the registrations of a feeding or drinking pig. Range measurements of the RFID system revealed that changes in tag position and orientation relative to the antenna during feeding and drinking most likely caused this phenomenon.
From registrations of the RFID tag at the feeder or drinker; number, duration and timing of feeding and drinking visits can be derived. However, feeding and drinking visits first had to be constructed from the raw RFID registrations. Visit criteria were found to be the optimal method to do this. Then, several variables of the feeding and drinking pattern were extracted from the RFID data and compared to the observations. Correlations between observed and RFID based feeding or drinking variables were found. This was especially the case for the duration of feeding and drinking, which was highly correlated to both the duration of the RFID based visits and the raw number of RFID registrations. In addition, also water usage could be estimated by the RFID drinking system.
Four warning systems were designed to monitor changes in the feeding patterns of the individual pigs. The number of registrations per pig and the average gap between RFID based feeding visits were chosen as variables to be used in the detection algorithms. For each variable, fixed limits that were constant for all the pigs during the entire fattening period were compared to Synergistic Control (SGC) limits that were individual and time-varying. The concept of Synergistic Control allows differentiating normal variation in the pigs’ feeding and drinking behaviour, such as age-effects, from abnormal variation pointing towards problems. Because every fattening pig acts as its own reference and the limits are pig-specific when using SGC, possible problems can be signalled on an individual level.
Abnormal points detected by each of the warning systems were signalled daily in the form of an ‘alert’ for each pig that crossed its individual threshold value. An extensive validation of the warning systems was performed on a group of pigs that was closely monitored on a daily basis to determine the number of true and false alerts and the number of missed problems. The best performing warning system was with Synergistic Control limits on the number of RFID registrations and the use of historical data to initialize the warning system. This led to a sensitivity of 66 %, specificity of 98 %, accuracy of 97 % and precision of 67 % for all health, welfare and productivity problems spotted by the observers. The average time till the first false alert was 82 days for individual pigs and severe problems were detected after 1.1 day on average.
Further research is required to quantify the added value of an automated warning system for the farmer at different levels (e.g. health, performance, efficiency, labour, costs, welfare, sound use of antibiotics). The alerts of the warning systems were now compared to problems detected by observers. However, there is no information available yet on which problems a pig farmer would detect (and when) and more importantly, which problems are most important to detect (e.g. because they require treatment, are difficult to detect visually). Future research should also focus on increasing the performance of the warning system in terms of sensitivity and precision. The optimal variables for problem detection should still be determined, as well as the best combination of variables and warning systems. Such combinations could also include automatic monitoring of the drinking behaviour.status: Publishe
Lerende controle voor autonome voertuig navigatie
No full textIn agricultural field work, machines must be accurately navigated to achieve an optimal result. The entire field should be covered with minimal overlap during tillage, fertilizing and spraying. The rows must be nicely parallel nicely for sowing and planting and evenly distributed so that the weed hoes can be easily driven between them. The fact that this is a difficult task, can be clearly seen from the considerable overlap and variation in plant distances observed in the field. Moreover, this limits the speed at which the tasks can be performed. Apart from navigating the machine, the operator must also supervise the work performed by the machine. Switching between paying attention to the steering and paying attention to the machine control results in an increase in the deviation from the optimal path in practice. To alleviate the task of the operator and allow him to concentrate on the quality of work performed, systems were developed for driver assistance and semi-autonomous control. These systems initially worked primarily based on local positioning. Thanks to the developments in the field of RTK GNSS systems with cm accuracy, this is nowadays mainly done through global positioning. The current commercial systems use simple controllers to minimize the deviation from the target path by adjusting the steering angle. These systems work well for the following straight lines under uniform soil conditions with a constant speed. However, when the soil conditions or speed change, the controllers must be tuned again. Furthermore, they use independent controllers for the absolute steering of the tractor and the relative steering of the trailer. Since both controllers will exhibit selfish behavior, this often leads to a sub-optimal result, especially for curved target paths in which the steering action of the tractor works against that of the trailer. Advanced control methods are needed to achieve higher control accuracy for the trailer for both straight and curved target paths under varying soil conditions and to make tractor control subordinate to the trailer control. Therefore different advanced model predictive control structures were elaborated in this PhD research and implemented on an autonomous tractor-trailer system to investigate its potential in practice. Firstly, an adaptive kinematic model and a yaw dynamics model of the tractor-trailer system have been derived from first principles. Then, the longitudinal dynamics and the dynamics of the steering mechanisms have been identified. The overall resulting model can serve as a benchmark system for evaluating model based controller and estimator designs. When model-based control structures have to deal with uncertain and varying process conditions, it is inevitable to use adaptive models. Real-time estimators allow to make these model adaptations through online parameter estimation. In this study, nonlinear moving horizon estimation method has been chosen as a state and parameter estimation algorithm because it considers the state and parameter estimation within the same problem and allows to incorporate constraints both on states and parameters. Secondly, a nonlinear model predictive controller has been designed in which a full model of tractor-trailer system was used including all interactions. This centralized controller managed to let the system follow the target trajectory with a high accuracy. However, it requires a relatively high computational cost due to the complexity of the optimization problem up to 7.2 ms. To reduce the computational cost a fully decentralized nonlinear model predictive controller was also designed and implemented. However, as these decentralized controllers ignore the interactions between the subsystems, they affect each other's performance negatively and may destabilize each other. In order to make the system robust against the differences between the real system and the sub-models, a tube-based approach was applied for the interactions between the two subsystems. Thanks to dividing the optimization problem into two smaller problems, the computation time could be reduced by a factor of 5. However, the trajectory error increased by 20 to 50% by ignoring the interactions. As both the centralized and decentralized approaches had their merits and limitations, a compromise was sought in the form of a distributed nonlinear model predictive controller in which the interactions are partially taken into account. These distributed controllers were between the centralized and decentralized controllers in terms of both computation time and tracking error. Finally, as an alternative approach to reduce the required computation time, two linear model predictive controllers have been designed: a linear time-invariant and a linear time-varying. Only the linear time-invariant model predictive controller was capable of driving the tractor-trailer system on any desired trajectory with an acceptable accuracy. While the computation time decreased to ms which is a factor of 10 shorter than the centralized nonlinear model predictive control, the trajectory error increased by 50 to 100%. Nonlinear model predictive controllers and nonlinear moving horizon estimators are capable of tracking straight and curved lines accurately. Thanks to the input-state linearization, successful results have been reported for the linear time-invariant model predictive controller. However, the linear time-varying model predictive controller obtained by linearizing around the reference trajectory is not able to follow curvilinear trajectories due to the fact that the trajectory tracking is a nonlinear control problem from its nature. As future studies, nonlinear control and estimation methods are more than welcome. In addition to efficient numerical methods, decentralized and distributed nonlinear moving horizon estimators can be studied to decrease computation time in practice. Robot sensing is also welcome to make agricultural robots being aware of environment and making decisions without human intervention.status: Publishe
Een discrete elementen aanpak voor het simuleren van de compressie van vezelrijke biomassa
No full textDue to the growing world population, the agricultural sector needs more productive and more energy-efficient agricultural machinery in order to adequately address the growing demand for food and biomass. Therefore, the sector significantly invests in the optimization of agricultural machinery.
Historically, the optimization of agricultural machinery was done by trial and error. Design improvements of agricultural machinery are still often based on the experience and the insights of engineers and farmers. To test whether an adjustment has a positive effect, a prototype is developed and validated during field tests. This optimization method, however, has some disadvantages. Developing and constructing a prototype is costly. Moreover, prototypes can only be validated in field conditions during the growing season. These two factors oblige agricultural machinery manufacturers to opt for small adaptations with a high success rate. The current generation of agricultural machinery is, therefore, the result of decades of evolution.
Now that computing power increases, another opportunity to improve the design of agricultural machinery presents itself. Models and simulations facilitate the optimization of machines. However, an accurate virtual crop model is missing. A simulation model that accurately describes the interactions between individual crop stems and the interactions between crop stems and machine components could be used to improve the design of stem processing agricultural machinery. In this thesis, such a simulation model was developed for the processing of crop stems in a baler.
It has previously been shown that the Discrete Element Method (DEM) can be used to simulate and optimize particulate processes. For this, two requirements need to be met. Realistic particle geometries and realistic deformation models are required to obtain accurate simulation results. The virtual crop stems, therefore, need to be compressible and bendable. Also the frictional and tensional properties need to be realistic.
A first step in the development of the DEM simulation model included an analysis of the bending behaviour of crop stems. It was observed that there are two phases during bending: ovalisation and buckling. The forces that occur during ovalisation result in a flattening of the cross-section of the stem and this reduces the bending resistance. This process continues until the maximum force has been reached and the stem buckles. Buckling is associated with a strong reduction of the resistance to bending. The influence of the stem diameter, the thickness of the stem wall and the presence of a core-rind structure were examined for wheat and barley stems. All were found to affect the bending behaviour significantly.
The acquired knowledge was used to develop a data based bending model for flexible particles (i.e. crop stems) in DEM. The influences of the stem length, the support distance and the number of segments which make up the virtual stem, were examined. The same data based method was also used for developing a compression model for virtual stems in DEM. For this purpose, the interactions between individual stems and the interaction between a stem and a plate were studied and modelled. The models were successfully validated by comparing bulk compression simulations and measurements. For this purpose 250 stems were compressed in a box by the movement of a plunger.
To study the influence of friction on the processing of stems, measurements were performed on the stem level. The measured coefficients of friction were significantly lower than those found in the literature, which have been measured on a bulk level. The influence of friction on bulk compression was evaluated and it was found that a small change in the coefficient of friction at stem level has a significant effect on the bulk behaviour.
The last stem parameter that was studied was the tensional stiffness. Stem measurements were again performed for this purpose. The force was found to increase linearly with increasing deformation up to the point where the stem broke. A linear tensional model was therefore implemented in DEM. Afterwards, the influence of the tensional resistance on the bulk deformation behaviour was examined. The effects of the tensional model parameters were found to be very limited. Therefore, the model parameters of the tensional model were selected in such a way that the computation time was minimized.
The effect of strain rate on the force-deformation behaviour at the stem level was studied using a pendulum device. However, no significant effects could be observed for the tests at low and high speed with the used set-up.
When the stem properties were measured and after they were modelled in DEM, the influence of the stem variability (e.g. the variability in physical and mechanical properties) on the bulk deformation behaviour was determined. To this end, simulations were performed with different degrees of variability. As a validation, bulk compression tests were performed. It was observed that a limited number of stem measurements can be sufficient to obtain accurate DEM simulations. As more stems are measured and as the stem database becomes larger, the accuracy of the simulations increases. However, the accuracy gained by measuring additional stems decreases with an increasing crop database. A statistical method was therefore presented to determine the minimum number of stem measurements needed to obtain accurate DEM simulations.
When the behaviour of crop stems was fully characterized and modelled and after insights were obtained on the influence of stem variability, DEM simulations were performed regarding the processing of crop stems by the rotor of a large square baler. First, a method was developed to create virtual swaths. Scalability was demonstrated with these swaths. This reduced the computation time of thesimulations. Again, friction was found to have a significant impact on the crop processing as a higher coefficient of friction led to a higher energy consumption. When stems are damaged, less energy is required for their processing. Also, the feed rate was found to have an influence. The energy demand increased as more stems were processed simultaneously. Finally, the shape of the swath was also found to have a major impact on the required torque. An evenly filled swath was found to require less processing energy than an unevenly filled swath.
In a final step, the filling of the pre-compression chamber was also simulated and successfully validated with stationary measurements. An increasing swath mass, the presence of the trip sensors (determining when the the pre-compression chamber is full) and a reduced rotor speed were found to have a negative impact on the required energy. The crop flow in the simulations was visually compared to the crop flow in the measurements. A high-speed camera was used for this purpose. The crop flows were found to be similar. However, a quantitative analysis should be performed to confirm this.
The simulation model is now ready for the optimization of the design of the pick-up and feeding sections of the agricultural baler. The knowledge that was gained in this dissertation is more broadly applicable and could, for example, also be used for optimizing sections of the combine and the forage harvester. However, more research is required to accurately simulate nodes, leafs and ears. The influence of the strain rate and the number of stem measurements required to obtain accurate DEM simulations of crops should also be studied in more detail. Modelling the cutting and breaking of stems would have a positive effect on the accuracy and the applicability of the simulations. Since in many processes air currents have a major impact, a coupling with a CFD software is also necessary.status: Publishe
Optimalisatie van het samenpersen van biomassa
No full textThe compression of biomass is essential for facilitating transport and storage. Biomass, like wood, landscape refuse, straw and hay, is mostly compressed with a press moulding machine because of continuous operation which results in high capacities.
The working principle of press moulding machines is based on friction of the material with the compression chamber. Higher friction means more compression but also results in high energy consumption for pushing the material out of the chamber. Studies show that the actual compression energy is less than 50% of the total required energy of the compression process. This low energy efficiency is inherent to the current design of press moulding machines in agriculture and allows for optimizing the process.
High forces are needed for the compression and the transportation of the biomaterial and thus the metal work has to be very solid. A tractor driving mobile press moulding machines then requires a lot of energy for off-road transportation.
This research will found the development of new balers with higher energy efficiency and a lighter construction and will consist of two parts. The first and biggest part will model the interaction between the baler and the biomaterial in the compression chamber. This will result in a three dimensional dynamic finite element model that correlates the applied forces with the density of the biomass. The (time dependent) inputs of the compression process consist of machine and crop parameters together with the settings of the machine. Literature, lab research and fieldwork will explore the dependency of the model to the crops. The model validation will be done on the baler that is currently present at the lab.
The second part of this research will show the advantages of a model-based optimization and will show the power of in-silico design optimization. This part will combine insights and expertise from experienced drivers with the modelling to eventually determine new machine parameters. The optimization will aim to minimize the work for transportation of the compressed biomaterial in the compression chamber while compressing till 200 kg/m³. Mechanical adaptations and suggestions for alternative control algorithms will be validated in the field with a large square baler to evaluate the optimal, theoretical efficiency in practice.status: Publishe
- …
