Internet das Coisas é um conceito novo que define a forma que os dispositivos se comunicam entre si sem a necessidade de interferência humana, possibilitando tomadas de decisão mais inteligentes, baseadas em dados coletados de sensores, tags e outros dispositivos. O objetivo do trabalho é a criação e validação de um sistema usando elementos de Internet das Coisas com mineração de dados para detectar padrões de comportamentos de animais, possibilitando o proprietário tomar decisões para segurança e saúde dos animais. Para tanto, foi feito um estudo da arte das soluções atuais de monitoramento animal e das tecnologias disponíveis para aplicação no sistema. Dois hardwares foram desenvolvidos para coleta dos dados, e foram integrados a uma ferramenta de mineração chamada Damicore. Apesar de algumas dificuldades envolvendo cobertura com a internet, foram realizados dois experimentos com êxito e seus dados foram processados pela ferramenta de mineração, produzindo clusters hierárquicos que possibilitam destacar determinados comportamentos animais através dos clusters encontrados. Finalmente, foram sugeridos alguns pontos de melhoria do sistema, para aumentar ainda mais a abrangência da solução. / Internet of Things is a new concept that defines the way devices communicate with each other without the need for human interference, enabling smarter decision-making based on data collected from sensors, tags and other devices. The goal of this project is to create and validate a system using Internet of Things features with data mining to detect patterns of animal behavior, enabling the owner to make decisions for animal safety and health. For this purpose, current animal monitoring solutions and technologies were reviewed. Based on that, two hardware devices were developed for data collection and integrated with a data mining tool, called Damicore. Despite some difficulties involving internet coverage, two experiments were successfully carried out and their data were processed by the data mining tool, producing hierarchical clusterings that enable to highlight certain animal behaviors by means of the found clusters. Finally, some factors that can benefit the evolution of the system were suggested to further increase the scope of the solution.

In dieser Arbeit wurden Untersuchungen zur Ontogenese des Fressverhaltens von Milchkühen durchgeführt und zwei mitentwickelte Sensorsysteme zur Überwachung des tierindividuellen Fress- und Wiederkauverhaltens wissenschaftlich bewertet. Das Fressverhalten erstlaktierender Kühen unterschied sich deutlich von dem älterer Kühe. Besonders große Unterschiede fanden sich hier in der Fressgeschwindigkeit, die bei älteren Kühen deutlich größer war. Auch die Fresszeit unterschied sich deutlich zwischen Jung- und Altkühen, allerdings nicht im gleichen Maß wie die Fressgeschwindigkeit. Zur Überwachung des Fress-und Wiederkauverhaltens wurden zwei verschiedene sensorbasierte Systeme eingesetzt. Das RFID-basiertes System erwies sich bei der Erfassung des Fressverhaltens als ungeeignet aufgrund des großen infrastrukturellen Aufwands und der unzureichenden Erfassung der Tiere am Fressgitter. Dennoch konnten vergleichbare Ergebnisse mit der Literatur bei der Charakterisierung des Fressverhaltens erzielt werden. Auch konnten Unterschiede im Fressverhalten zwischen gesunden und auffälligen Kühen festgestellt werden. Das elektromyographiebasierte Sensorsystem erwies sich als gut geeignet, um das tierindividuelle Wiederkauverhalten zu erfassen und zu analysieren. Anhand der Detailanalyse der Signale konnte ein Wiederkaukriterium, ähnlich dem Mahlzeitenkriterium, ermittelt werden. Des Weiteren zeigten sich die Vorteile im Abgleich mit den RFID-basierten Wiegetrögen. Letztere weisen in den Fresszeiten Ungenauigkeiten auf, da sie die tatsächliche Maulaktivität nicht erfassen können. Trotzdem sind sie wichtig, da die quantitative Erfassung der Futteraufnahme mit anderen Systemen derzeit nicht möglich ist und die gezeigten Zusammenhänge zwischen Fress- und Wiederkauverhalten zu stark variieren, um allgemeingültige Aussagen zur Menge des aufgenommenen Futters zu treffen. An dieser Stelle müssen weitere Forschungsvorhaben ansetzen. / In this work the ontogenesis of feeding behavior of dairy cows was investigated and two co-developed sensor systems for monitoring feeding and ruminating scientifically evaluated. It appears that the feeding behavior of first-lactation cows significantly differs from that of older cows. Particularly noticeable differences were found for the feeding speed, which was much higher for older cows. The feeding times of young and old cows differed as well, but to a lesser extent. In order to monitor feeding and ruminating two different sensor-based systems wer used. It turned out that the RFID-based system is a rather unfit choice due to its extensive infrastructure requirements and the insufficient registration of the animals at the feed fence. However, measured results are comparable to those found in previous studies of typical feeding behavior. In addition, differences in the behavioral patterns of healthy cows and of ailing cows were noted. The electromyography-based sensor system appears to be quite suitable for documenting and evaluating animal-specific rumination patterns. Based on the detailed analysis of the signals, it was possible to determine a rumination variable, similar to the meals variable. This approach is also advantageous compared to the RFID-based weighing troughs in that the latter suffer from inaccuracies in the feeding times due to their inability to measure actual mouth activity. Nevertheless, these RFID-based systems play an important role as they currently provide the only option for quantitative measurements of feed intake. Furthermore, the noted correlations between feeding and rumination patterns vary too much to allow absolute statements on the amount of feed intake. Additional studies will have to explore these issues further.