Richardson Ribeiro

In Vivo Pattern Classification of Ingestive Behavior in Ruminants Using FBG Sensors and Machine Learning.

Pattern classification of ingestive behavior in grazing animals has extreme importance in studies related to animal nutrition, growth and health. In this paper, a system to classify chewing patterns of ruminants in in vivo experiments is developed. The proposal is based on data collected by optical fiber Bragg grating sensors (FBG) that are processed by machine learning techniques. The FBG sensors measure the biomechanical strain during jaw movements, and a decision tree is responsible for the classification of the associated chewing pattern. In this study, patterns associated with food intake of dietary supplement, hay and ryegrass were considered. Additionally, two other important events for ingestive behavior were monitored: rumination and idleness. Experimental results show that the proposed approach for pattern classification is capable of differentiating the five patterns involved in the chewing process with an overall accuracy of 94%.

A sociologically inspired heuristic for optimization algorithms: A case study on ant systems

This paper discusses how social network theory can provide optimization algorithms with social heuristics. The foundations of this approach were used in the SAnt-Q (Social Ant-Q) algorithm, which combines theory from different fields to build social structures for state-space search, in terms of the ways that interactions between states occur and reinforcements are generated. Social measures are therefore used as a heuristic to guide exploration and approximation processes. Trial and error optimization techniques are based on reinforcements and are often used to improve behavior and coordination between individuals in a multi-agent system, although without guarantees of convergence in the short term. Experiments show that identifying different social behavior within the social structure that incorporates patterns of occurrence between states explored helps to improve ant coordination and optimization process within Ant-Q and SAnt-Q, giving better results that are statistically significant.

In-vivo determination of chewing patterns using FBG and artificial neural networks

This paper reports the process of pattern classification of the chewing process of ruminants. We propose a simplified signal processing scheme for optical fiber Bragg grating (FBG) sensors based on machine learning techniques. The FBG sensors measure the biomechanical forces during jaw movements and an artificial neural network is responsible for the classification of the associated chewing pattern. In this study, three patterns associated to dietary supplement, hay and ryegrass were considered. Aditionally, two other important events for ingestive behavior studies were monitored, rumination and idle period. Experimental results show that the proposed approach for pattern classification has been capable of differentiating the materials involved in the chewing process with a small classification error.

Modeling and control of flexible context-dependent manufacturing systems

Abstract In emerging Manufacturing Systems (MSs), flexibility is a key issue. It is related to the ability for a MS to recognize the context and switch its workflow accordingly. Although the literature has provided automated options to model and control MSs, programming context-dependent controllers remains challenging. This is an event-based construction that integrates a large and intricate combination of events and states in order to make the controller flexible, i.e., include context-sensitiveness strategies subject to switching at runtime. Without self-adaptation, each system configuration may require an entire control solution to be recalculated, which implies redesigning the whole modeling and implementation structures. This paper shows that a system model can nevertheless be enriched with elements collected from the context, which optimizes the design of formula-based constraints that can then be integrated to control frameworks for synthesis and code generation. The result is a controller that recognizes the context and takes control decisions accordingly. Examples are provided to illustrate the approach.

An Economic Approach for Generation of Train Driving Plans using Continuous Case-based Planning

We present an approach for reusing and sharing train driving plans P using continuous (or without human intervention) Case-Based Planning (CBP). P is formed by a set of actions, which when applied, can move a train in a stretch of railroad. This is a complex task due to the variations in the (i) composition of the train, (ii) environmental conditions, and (iii) stretches travelled. To overcome these difficulties we provide to the driver a support system to help the driver in this complex task. CBP was chosen because it allows directly reuse the human drivers experience as well as from other sources. The main steps of the CBP are distributed among specialized agents with different roles: Planner and Executor. Our approach was evaluated by different metrics: (i) accuracy of the case recovery task, (ii) efficiency of task adaptation and application of such cases in realistic scenarios and (iii) fuel consumption. We show that the inclusion of new experiences reduces the efforts of both the Planner and the Executor, reduces significantly the fuel consumption and allow the reuse of the obtained experiences in similar scenarios with low effort.

A Learning Model for Intelligent Agents Applied to Poultry Farming

This paper proposes a learning model for taking-decision problems using intelligent agents technologies combined with instance-based machine learning techniques. Our learning model is applied to a real case to support the daily decisions of a poultry farmer. The agent of the system is used to generate action policies, in order to control a set of factors in the daily activities, such as food-meat conversion, amount of food to be consumed, time to rest, weight gain, comfort temperature, water and energy to be consumed, etc. The perception of the agent is ensured by a set of sensors scattered by the physical structure of the poultry. The principal role of the agent is to perform a set of actions in a way to consider aspects such as productivity and profitability without compromising bird welfare. Experimental results have shown that, for the decision-taking process in poultry farming, our model is sound, advantageous and can substantially improve the agent actions in comparison with equivalent decision when taken by a human specialist.

A Hybrid Interaction Model for Multi-Agent Reinforcement Learning

The main contribution of this paper is to implement a hybrid method of coordination from the combination of interaction models developed previously. The interaction models are based on the sharing of rewards for learning with multiple agents in order to discover interactively good quality policies. Exchange of rewards among agents, when not occur properly, can cause delays in learning or even cause unexpected behavior, making the cooperation inefficient and converging to a non-satisfactory policy. From these concepts, the hybrid method uses the characteristics of each model, reducing possible conflicts between different policy actions with rewards, improving the coordination of agents in reinforcement learning problems. Experimental results show that the hybrid method can accelerate the convergence, rapidly gaining optimal policies even in large spaces of states, exceeding the results of classical approaches to reinforcement learning.

Resources Planning in Database Infrastructures

Anticipating resources consumption is essential to project robust database infrastructures able to support transactions to be processed with certain quality levels. In Database-as-a-Service (DBaaS), for example, it could help to construct Service Level Agreements (SLA) to intermediate service customers and providers. A proper database resources assessment can avoid mistakes when choosing technology, hardware, network, client profiles, etc. However, to be properly evaluated, a database transaction usually requires the physical system to be measured, which can be expensive an time consuming. As most information about resource consumption are useful at design time, before developing the whole system, is essential to have mechanisms that partially open the black box hiding the in-operation system. This motivates the adoption of predictive evaluation models. In this paper, we propose a simulation model that can be used to estimate performance and availability of database transactions at design time, when the system is still being conceived. By not requiring real time inputs to be simulated, the model can provide useful information for resources planning. The accuracy of the model is checked in the context of a SLA composition process, in which database operations are simulated and model estimations are compared to measurements collected from a real database system.

Supervisory control of multiple robots subject to context switching

Coordinating concurrent multiple robots, in an efficient manner, is a task that challenges control engineers by its complexity. To be properly addressed, this task usually requires the support of formal approaches, such as the Supervisory Control Theory (SCT), which provides an automated way to calculate control strategies for event-based robots coordination. Little has been reported, however, on how to program controllers that automatically adapt themselves to the system context. In fact, self-adaptation support implies providing the robot brain with context-sensitive control strategies subject to changes at runtime. Without self-adaptation, on the other hand, each system configuration may require an entire control solution to be recalculated, which implies reestablishing the whole modeling structure. This paper shows that a system model can nevertheless be enriched with elements collected from the system context, which optimizes the design of formula-based constraints that can then be integrated to the SCT framework for control synthesis and posterior code generation. The result is a controller that recognizes the context and take control decisions accordingly. An example of multiple robots coordination illustrates the approach.

In vivo analysis of bone strain using fiber Bragg grating sensor and decision tree algorithm in bovine during masticatory movements

This study focused on the development of a biosensor able to follow the deformations of the bone tissue and to allow for future studies in the areas of health sciences. The biosensor is designed with a titanium mesh, which is fixed in the jaws of the animal to be monitored. The animal received different kinds of foods to allow analysis of the signal acquired during feeding. The acquired signal was then subjected to a processing, which has been classified and capable of identifying the animal chewing process for each type of food. This technology has application in grazing, and can be useful in studies related to nutrition and animal health. The classification of the masticatory patterns is based on decision trees algorithm. The results demonstrate that the sensor is effective and is able to capture the differences in the deformation generated during the chewing process, generating a signal that can be identified by the algorithm presented.