Search

The increasing use of unmanned aerial vehicles in areas such as rescue, mapping, and transportation have made it necessary to study more accurate techniques for calculating flight time estimates. Such calculations require knowing the battery discharge profile. Simplified flight time calculation methods provide data with uncertainties as they are based solely on manufacturer datasheet information. This study presents a setup to measure the battery discharge curve using a LabVIEW interface with a low-cost acquisition system. The acquired data passes through a nonlinear optimisation algorithm to find the battery coefficients, which enables the more precise estimation of its range and endurance. The great advantage of this model is that it makes it possible to predict how the battery will discharge at different rates using just one experimental curve. The methodology was applied to three different batteries and the model was validated with different discharge rates in a controlled environment, which resulted in endurance lower than 3.0% for most conditions and voltage estimation error lower than 3.0% in operational voltage. The work also presented a methodology for estimating cruise time based on the current used during each flight stage.

New technologies have pushed Unmanned Underwater Vehicles use in the industry where the focus is often oil and gas applications and, high end systems. Operation in broad underwater tunnels where water turbidity is high, make it difficult to orient Observation Class UUVs inside them. Poor sensory orientation inside the tunnel can put the vehicle integrity and operation effectiveness at risk. In this work, we present a compact UUV prototype which is designed for operation in flooded tunnels, in special, those of hydro power plant. The system architecture is described and each subsystem is tested. Experiments present the effectiveness of each subsystem, with focus on the tunnel, discussing each system in the context of future mapping and autonomous operations. We present a system with a set of sensors aiming manual and autonomous navigation: an imaging sonar for forward looking information and navigation; an array of echosounder sensors used to measure the distance between top, bottom and side walls, which complement the forward looking sonar information; a profiling sonar for tunnel cross-section profiling; and a low light camera for close range inspections. Tests with the system are performed in a swimming pool, a lake, and, finally, inside a hydro power plant tunnel, during a four day-trial at Ceran Castro Alves power plant.

The stock price movements result from many factors that are often difficult to be detected and modelled. The investigation of price trends and the use of the information available to evaluate investments and identify trading opportunities can be promising. However, financial data are non-stationary, i.e., their statistical characteristics constantly change. Therefore, the financial market is a challenging environment for the application of Machine Learning techniques, since they can only make reliable predictions to data consistent with what they have seen before. This paper test the use of a Machine Learning technique known as Support Vector Machines (SVM) aiming at being a tool to support the decision making process of trading at the stock market. SVM aggregates some input signals and, based on a set of technical indicators and historical price changes, create buy/sell recommendations of a given security as outputs. The dataset comprises several Brazilian stocks time-series traded on both the Brazilian (B3) and American (NYSE) stock exchange. These time-series belong to various economic sectors and present different market dynamics. The computational simulations are based on a fictitious strategy that does not consider the trading costs and only long positions are allowed. Using two risk-adjusted performance metrics, the results show that strategies based on the SVM model achieve better performance than the Buy & Hold benchmark.

Hydroelectric power plants often make use of tunnels to redirect the flow of water to the plant power house. Such tunnels are often flooded and can span considerable distances. Periodical inspections of such tunnels are highly desirable since a tunnel collapse will be catastrophic, disrupting the power plant operation. In many cases, the use of Unmanned Underwater Vehicles (UUVs) equipped with mechanical profiling sonars is a suitable and affordable way to gather data to generate 3D mapping of flooded tunnels. In this paper, we study the resolution of 3D tunnel maps generated by one or more mechanical profiling sonars working in tandem, considering synchronization and occlusion problems. The article derives the analytical equations to estimate the sampling of the underwater tunnels using mechanical profiling sonars (scanning sonars). Experiments in a simulated environment using up to four sensors simultaneously are presented. We also report experimental results obtained by a UUV inside a large power plant tunnel, together with a first map of this environment using a single sonar sensor.

Presently, time-based airline maintenance scheduling does not take fault predictions into account, but happens at fixed time-intervals. This may result in unnecessary maintenance interventions and also in situations where components are not taken out of service despite exceeding their designed risk of failure. To address this issue we propose a framework that can predict when a component/system will be at risk of failure in the future, and therefore, advise when maintenance actions should be taken. In order to facilitate such prediction, we employ an auto-regressive moving average (ARMA) model along with data-driven techniques, and compare the performance of multiple data-driven techniques. The ARMA model adds a new feature that is used within the data-driven model to give the final prediction. The novelty of our work is the integration of the ARMA methodology with data-driven techniques to predict fault events. This study reports on a real industrial case of unscheduled removals of a critical valve of the aircraft engine. Our results suggest that the support vector regression model can outperform the life usage model on the evaluation measures of sample standard deviation, median error, median absolute error, and percentage error. The generalized linear model provides an effective approach for predictive maintenance with comparable results to the baseline. The remaining data-driven models have a lower overall performance.

This paper describes an iterative decentralized planning and learning method, based on stochastic learning automata theory and heuristic search techniques, to generate construction and motion strategies to build different types of three-dimensional structures using multiple quadrotors. This architecture is proposed to simultaneously solve three main problems: 1) the iterative generation of feasible construction and motion plans for each quadrotor; 2) the optimization with constraints on power and assembly while taking into account the dynamic nature of the environment, and 3) the planning of the translational speeds and selection of breakpoints for each vehicle. The quadrotors learn the optimal action policy to construct the structures while avoiding collisions during the loading and unloading procedures. In order to demonstrate the generality of the solution, simulated trials of the proposed autonomous construction system are presented where different three-dimensional structures are built.

In this paper, we describe an architectural framework by which a mobile robot can learn how to autonomously assemble solid 3D brick structures according to user-specified designs. The policies of actions to perform the construction task are obtained from a simulation environment using Reinforcement Learning (RL) and Particle Swarm Optimization (PSO) approaches. The proposed planning architecture is used to simultaneously solve three problems: 1) to generate feasible construction policies, 2) to define the set of maneuvers for the vehicle to carry out the assembly task, and 3) to obtain the set of trajectories for handling and mounting parts while avoiding fixed obstacles. During the learning process the power limitation of the ground robot is taken into account. Simulation results show that the set of learned actions may efficiently perform the construction procedures without resulting in conditions which prevent the fulfillment of the assembly procedures. The synthesis of this system opens the way to the development of intelligent construction approaches using ground robots.

Prognostics are a key activity in repair and maintenance operations. A recent approach to condition-based maintenance is the data-driven approach. This approach has been mostly based on past failure time measures, and sensed measurements of component degradation to derive estimates of the remaining useful life of equipment. An alternative source of data, rarely used in these models, is the stream of automatic messages derived from diagnostics systems, which consist of fault codes indicating abnormal events or deviations from optimal operation. Despite the richness and concise nature of these messages, their difficult interpretation poses significant challenges to its use in prognostics. This paper aims to show that data-driven prognostics based on this type of messages can be better suited to maintenance than time-based approaches. We illustrate this comparison with an industrial case study involving the removal times of a bleed valve from the aircraft air management system. Our experimental results reveal a significant accuracy improvement over the contrasting time-based models. We also establish the contribution to this improvement of the data-driven methods and message-related predictors.

The objective of this work is to analyze the productivity of different Brazilian research promotion agencies by using the Method of Interactive Decision Making and Multicriteria (Acronym TODIM in Portuguese), which is based on the Prospect Theory. The proposed model classify of 22 Brazilian research agencies (two federal and twenty state agencies) under six different criteria. The result shows that some small state agencies are more efficient than bigger ones in terms of the funding/scholarship ratio. This suggests that the winner model could be further investigated to be promoted among the other agencies of this country.

Singular Spectrum Analysis (SSA) is a nonparametric approach that can be used to decompose a time series as trends, oscillations and noise. In online applications, the SSA algorithm must be recalculated for each new sample available. The so called Causal SSA have been used in this context. In this manuscript, an alternative version of online SSA, the Recursive SSA, is proposed as a technical indicator. Based on a “forgetting factor” parameter, it is possible to control the amount of previous samples that are used in the SSA algorithm. This functionality may confer adaptive and hybrid features, thereby providing a crucial characteristic to technical indicators. The Recursive SSA technical trading rules (SSA-TTR) are applied to the DJIA time-series and compared against popular technical indicators. The results show the advantages of SSA-TTR over the popular ones.

This manuscript introduces a Matlab computational framework for a multiagent system simulation of financial markets. The main motivation of using Matlab is that this platform is very familiar among researches in the area of engineering. This could motivate that researchers to investigate this area. The aim of our platform is to replicate the interactions among agents through a double auction environment, leading to an artificial market. Every agent is assigned with unlimited credit, and short selling is allowed. These two degrees of freedom might ensure that a particular agent is able to trade whenever it becomes active. Agents should be able to take their own decisions. The key innovative aspect of this work is the development of a computational platform which integrates such heterogeneous financial agents. This framework intends to mimic behavioral patterns followed by the market practitioners. Even though the proposed framework uses a simplified version of the financial market, the time series resulted from the interactions of those agents reproduces some of the financial stylized facts. Therefore, at least some market dynamics are successfully emulated by the proposed approach.

This work intends to replicate the interactions among intelligent agents in a double auction environment, leading to an artificial market. Aiming at making agents able to take their own decisions, different methodologies for agents intelligence are proposed, ranging from zero intelligence agents to reinforcement learning. Regarding the study of financial markets, agents can rely on a myriad of techniques in their making-decision process. In this manuscript, the Fundamental and/or Technical Analysis are the tools used by the agents to take their own decisions. A different set of parameters is assigned to each agent, thereby heterogeneity is included within each strategy. Based on these simple strategies some agents can also learn with their past experience, updating the probability to select a trading strategy according to switching rules as a function of the previous result. Several experiments are carried out. Results show that the time series produced by the interactions of those agents resembles to financial time series, reproducing some stylized facts encountered thereof. The performance of different methods for agent design are also compared.

Developing exact models of combustion instabilities is not an easy task to carry out and requires a great deal of time prior to obtaining success. The present study proposes a low-order model for pressure oscillations that does not require any knowledge of the systems, any new physical findings nor intricate details regarding its operating condition. This new approach is obtained using a Modified Van der Pol’s equation (MVDP) which is tuned by use of a Dual Extended Kalman Filter (DKEF) as a recursive estimator with perspectives in control by computer. This phenomenological model is used to predict the pressure signal from a variety of different combustors. Input data were taken from experimental cases such as a Rijke tube, a gas turbine and a liquid-fuel aero-engine combustor. Furthermore, a simulation considering high frequency oscillations to show the capability of the new approach is presented. In all cases, the results demonstrated the feasibility of applying the tractable model MVDP and DKEF running together to investigate pressure oscillations in practical cases.

This work consisted in developing a data driven solution to identify and isolate faults on aircraft bleed valves using supervised machine learning algorithms. A physical computer model of the valve with noisy observations and variable environmental conditions was used to simulate a set of operationing conditions submitted to three different failures modes: leakage between the valve chambers, degraded return spring and excessive piston friction. Features extracted from the model simulations were used to train and test a Support Vector Machine (SVM) as well as a Classification Tree. Different SVM Kernels and tree pruning limits were considered at the training step. Both solutions were evaluated at the testing step by evaluating their accuracy and precision. The results evidenced that both algorithms could identify and classify the faults with a slightly better performance of the SVM. Also, results evidenced that this application could automate the fault isolation process of these types of valves. This work evidences the importance of the proper definition of the classification algorithms as well as their metaparameters in order to obtain the most suitable solution for fault isolation developments.

A comparison between Moving Averages (MA) and two versions of Singular Spectrum Analysis (SSA) methodology — the Caterpillar and the Toeplitz — is presented. Caterpillar had already been studied in this manner but the same is not true for the Toeplitz SSA. Toeplitz SSA assumes the stationarity of the time-series, which means that the process needs to be mean-reverting. However, such assumption is not a necessary condition for the Caterpillar SSA. In this paper both approaches are applied to a trend estimation problem in order to be used as an indicator in trend-following technical rules design. Similarities and differences between these techniques are addressed. The obtained results suggest that, although SSA approaches provides more flexibility to achieve a desired trend resolution compared to the traditional MA, the Toeplitz SSA exhibit some issues that might put it off its use in this particular application.

The central theme of this paper is the prioritization of failing equipment using a Multicriteria Decision Aid method. Maintenance prioritization decision criteria is needed in productive processes with several similar machines. This decision is based on the inherent characteristics of the production process and the failed equipment. PROMETHEE II method is proposed in this paper, and has the characteristic of being non-compensatory and allows a pairwise comparison of alternatives, resulting in an order of alternatives. The criteria used were previously defined by specialists along of the production process. A model based on PROMETHEE II method is developed and applied in order to prioritize the failed equipment to be served. This model is proposed to be used on daily basis, and its validation was performed by the process specialists.

In this paper, we introduce a novel method to select the components of a portfolio of securities. This method is based on a reinforcement learning technique known as learning automata. Several heuristic solutions for the portfolio weights selection problem have been introduced in literature. The point is that these applications assumes that portfolio components are given. The difference of our work is that we propose some heuristic in order to select the portfolio components instead of the weights. In terms of heuristic, we propose learning automata because its ability to solve complex systems such as a the optimal portfolio components. We test the use of learning automata in terms of financial indicators optimization. Our findings show that our proposed method improves the portfolio optimization performance in terms of accuracy and computational effort.

In this paper, we develop an adaptive scheme based on reinforcement learning (RL) for planning the construction tasks using a quadrotor. Moreover, an autonomous construction system to assemble user-specified 3-D structures is proposed. Nowadays, complex construction tasks using mobile robots are characterized by three fundamental problems: assembly planning, motion planning, and path tracking control. The high-level plan to perform the construction task consists of assembly mode algorithms that are derived offline in a simulation environment through learning and heuristic search. A promising approach to design and optimize the path tracking controllers for a quadrotor as well as the attitude controllers using RL is presented. This paper describes a comprehensive validation framework that enables an aerial robot to build structures in a robust and safe manner. The experimental trials for building the 3-D structures using the designed high-level plans and path tracking controllers have provided encouraging results.

The implementation of condition-based maintenance continues to face several challenges especially in the aeronautics field. While it is true that time-based maintenance dominates the industry today, it is believed that condition monitoring could yield promising results with a better compromise on cost over effectiveness in the long run. The aim of condition-based monitoring in aeronautics is, based on the available system data (e.g., flight, event and maintenance data), to evaluate the current health state of an aircraft component and to estimate its remaining useful life. Several approaches have been studied in condition-based maintenance with the most promising being data-driven modeling. This paper proposes a comparison of a set of data-driven modeling techniques to perform prognostics on a critical component of the jet engine bleed system. The novelty of our work is twofold. First, we perform this comparative study on a real case study of a critical valve of the aircraft bleed system. Fielded data from different data sources are used in the models. To our knowledge, this is the first case study that merges data from the computer central maintenance system (fault messages), maintenance data, and flight data on a prognostics system. Second, a variety of data-driven techniques are compared from neural nets to regression support machines. The models are compared using the standard metrics of absolute, mean, and squared errors. A regressive accuracy curve is also used to compare the models along different prediction window sizes. The results show the best model comprised information from all data sources. The data that most contributed to the performance improvement was the maintenance, flight and fault data, in this order. This result comes to reinforce the notion that it is more difficult to extract quantitative information from fault events than flight data with data-driven regressive methods.

This article presents a concept for obstacle avoidance in dynamic environment suitable for mobile robot. The task of obstacle avoidance is divided in three principal groups: local, global and for emergencies. The local avoidance is here approached, in which the concept used is based on reinforcement learning, in such a way that the situations are divided into four states and two kinds of actions are possible. The states define in what situation the movement relationship between the robot and the dynamic obstacles is present, and the actions decide in which direction the robot must follow, in order to avoid a possible collision. And besides, here it is showed also how the state-action matrix was filled and its representation using neural network.

In this paper, we present a FastSLAM particle filter algorithm used to efficiently map large indoor environments features. The proposed filter uses an unknown data association to match the extracted environment characteristics, such as walls and doors. Data association (DA) is chosen due to two reasons: 1) permit to rearrange the filter particles in the prediction phase of the filter, and 2) enable to incorporate the extracted features in the map of each particle. Indoor SLAM experiments were conducted in a long corridor composed by several wooden walls. These provisional walls were used to create a more challenging environment. From the map obtained by the mapping process, the robot is capable of navigating through the environment using the set of 22 predefined poses. The SLAM filter measurements are compared with their actual measured values.

This paper describes the implementation of a low-cost Autonomous Surface Vehicle (ASV) using a behavior-based software, the MOOS-IvP. The platform used is a catamaran boat driven by two direct current motors as the propulsion system. It is embedded with a processing board with an Arduino microcontroller, a low-cost Inertial Measurement Unit (IMU) with accelerometers, gyroscopes and magnetometers, a GPS receiver and a wireless RF serial modem. The ASV communicates with a Ground Control Station (GCS) sending telemetry data and receiving navigation commands for the propulsion motors. The GCS uses the MOOS-IvP software to implement the autonomous navigation procedures and the GPS/Compass/IMU sensor fusion algorithm. Simulation and experimental results in real conditions are presented and discussed. The experimental and simulation results consist of path following missions in the presence of virtual obstacles.

This paper presents a scheme of multiple neural networks (MNNs) with a new strategy of combination. This combination can obtain an accumulative learning: the knowledge is increased by gradually adding more neural networks to the system. This scheme is applied to flexible link control via feedback-error-learning (FEL) strategy, here called multi-network-feedback-error-learning. Three different neural control approaches are used to control a flexible link, and it is shown that a better inverse dynamic model of the plant is obtained in this case.

This article is concerned with the gait synthesis problem of a hybrid robot (in this case, a four-legged robot with free wheels on its feet) considering multiple criteria. It is assumed that the position of each leg actuator over time is described by a periodic function with parameters that are determined using the learning automata reinforcement learning algorithm. Analysis of the robot morphology is used to group similar legs and decrease the number of actuator functions that must be determined. MATLAB/Simulink/SimMechanics Toolbox are used to simulate the robot gait. The simulated robot response is evaluated by the reinforcement learning algorithm considering: 1) the robot frontal speed, 2) the “smoothness” of the robot movements, 3) the largest torque required by all leg actuators, and 4) the robot energy consumption. When the reinforcement learning algorithm converges to a good solution, it is applied to the real robot which was built using the Bioloid Comprehensive Kit, an educational robot kit manufactured by ROBOTIS. The responses of the simulated and real robot are then compared and are shown to be similar.

This paper presents an adaptive approach based on the Reinforcement Learning (RL) method to manipulate and transport parts and also assemble 3-D structures in a moderately constrained and dynamic environment using a quad-rotor. Nowadays, complex construction tasks using mobile robots are characterized by two fundamental problems such as task planning and motion planning. However, to obtain the task and path planning that define a specific sequence of operations for construction of a given structure is generally very complex. In this context, we propose and investigate a system in which an aerial robot learns the assembly and construction tasks of multiple 3-D structures. This process involves the learning of the sequence of maneuvers of a vehicle, the assembly sequence of the parts and also the correct types of structural elements for each assembly point of the structure. A heuristic search algorithm is used in the learning process to find the optimal path for the quad-rotor so that its navigation through the dynamic environment is performed. The experimental results show that a 3-D structure can be built using the task planning approach derived from a learning algorithm combined with a heuristic search method.