The Role of Synthetic Data in Aerial Object Detection

The purpose of this study is to explore the characteristics of developing a machine learning application using synthetic data. The study is structured to develop the application for the purpose of deploying the computer vision model. The findings discuss the realities of attempting to develop a computer vision model for practical purpose, and detail the processes, tools and techniques that were used to meet accuracy requirements. The research reveals that synthetic data represent another variable that can be adjusted to improve the performance of a computer vision model. Further, a suite of tools and tuning recommendations are provided.

Method for Tuning Level Control Loops Based on Internal Model Control and Closed Loop Step Test Data

This paper describes a two-stage methodology derived from IMC (Internal Model Control) for tuning a PID (Proportional-Integral-Derivative) controller for levels or other integrating processes in an industrial environment. Focus is ease of use and implementation speed which are critical for an industrial application. Tuning can be done with minimum effort and without the need of time-consuming open-loop step tests on the plant. The first stage of the method applies to levels only: the vessel residence time is calculated from equipment dimensions and used to derive a set of preliminary PI (Proportional-Integral) settings with IMC. The second stage, re-tuning in closed-loop, applies to levels as well as other integrating processes: a tuning correction mechanism has been developed based on a series of closed-loop simulations with model errors. The tuning correction is done from a simple closed-loop step test and application of a generic correlation between observed overshoot and integral time correction. A spin-off of the method is that an estimate of the vessel residence time (levels) or open-loop process gain (other integrating process) is obtained from the closed-loop data.

Gaits Stability Analysis for a Pneumatic Quadruped Robot Using Reinforcement Learning

Deep reinforcement learning (deep RL) algorithms leverage the symbolic power of complex controllers by automating it by mapping sensory inputs to low-level actions. Deep RL eliminates the complex robot dynamics with minimal engineering. Deep RL provides high-risk involvement by directly implementing it in real-world scenarios and also high sensitivity towards hyperparameters. Tuning of hyperparameters on a pneumatic quadruped robot becomes very expensive through trial-and-error learning. This paper presents an automated learning control for a pneumatic quadruped robot using sample efficient deep Q learning, enabling minimal tuning and very few trials to learn the neural network. Long training hours may degrade the pneumatic cylinder due to jerk actions originated through stochastic weights. We applied this method to the pneumatic quadruped robot, which resulted in a hopping gait. In our process, we eliminated the use of a simulator and acquired a stable gait. This approach evolves so that the resultant gait matures more sturdy towards any stochastic changes in the environment. We further show that our algorithm performed very well as compared to programmed gait using robot dynamics.

Numerical and Experimental Analyses of a Semi-Active Pendulum Tuned Mass Damper

Modern structures such as floor systems, pedestrian bridges and high-rise buildings have become lighter in mass and more flexible with negligible damping and thus prone to vibration. In this paper, a semi-actively controlled pendulum tuned mass dampers (PTMD) is presented that uses air springs as both the restoring (resilient) and energy dissipating (damping) elements; the tuned mass damper (TMD) uses no passive dampers. The proposed PTMD can readily be fine-tuned and re-tuned, via software, without changing any hardware. Almost all existing semi-active systems have the three elements that passive TMDs have, i.e., inertia, resilient, and dissipative elements with some adjustability built into one or two of these elements. The proposed semi-active air suspended TMD, on the other hand, is made up of only inertia and resilience elements. A notable feature of this TMD is the absence of a physical damping element in its make-up. The required viscous damping is introduced into the TMD using a semi-active control scheme residing in a micro-controller which actuates a high-speed proportional valve regulating the flow of air in and out of the air springs. In addition to introducing damping into the TMD, the semi-active control scheme adjusts the stiffness of the TMD. The focus of this work has been the synthesis and analysis of the control algorithms and strategies to vary the tuning accuracy, introduce damping into air suspended PTMD, and enable the PTMD to self-tune itself. The accelerations of the main structure and PTMD as well as the pressure in the air springs are used as the feedback signals in control strategies. Numerical simulation and experimental evaluation of the proposed tuned damping system are presented in this paper.

A Study of the Trade-off Energy Consumption-Performance-Schedulability for DVFS Multicore Systems

Dynamic Voltage and Frequency Scaling (DVFS) multicore platforms are promising execution platforms that enable high computational performance, less energy consumption and flexibility in scheduling the system processes. However, the resulting interleaving and memory interference together with per-core frequency tuning make real-time guarantees hard to be delivered. Besides, energy consumption represents a strong constraint for the deployment of such systems on energy-limited settings. Identifying the system configurations that would achieve a high performance and consume less energy while guaranteeing the system schedulability is a complex task in the design of modern embedded systems. This work studies the trade-off between energy consumption, cores utilization and memory bottleneck and their impact on the schedulability of DVFS multicore time-critical systems with a hierarchy of shared memories. We build a model-based framework using Parametrized Timed Automata of UPPAAL to analyze the mutual impact of performance, energy consumption and schedulability of DVFS multicore systems, and demonstrate the trade-off on an actual case study.

Practical Techniques of Improving State Estimator Solution

State Estimator became an intrinsic part of Energy Management Systems (EMS). The SCADA measurements received from the field are processed by the State Estimator in order to accurately determine the actual operating state of the power systems and provide that information to other real-time network applications. All EMS vendors offer a State Estimator functionality in their baseline products. However, setting up and ensuring that State Estimator consistently produces a reliable solution often consumes a substantial engineering effort. This paper provides generic recommendations and describes a simple practical approach to efficient tuning of State Estimator, based on the working experience with major EMS software platforms and consulting projects in many electrical utilities of the USA.

Model Reference Adaptive Approach for Power System Stabilizer for Damping of Power Oscillations

In recent years, electricity trade between neighboring countries has become increasingly intense. Increasing power transmission over long distances has resulted in an increase in the oscillations of the transmitted power. The damping of the oscillations can be carried out with the reconfiguration of the network or the replacement of generators, but such solution is not economically reasonable. The only cost-effective solution to improve the damping of power oscillations is to use power system stabilizers. Power system stabilizer represents a part of synchronous generator control system. It utilizes semiconductor’s excitation system connected to the rotor field excitation winding to increase the damping of the power system. The majority of the synchronous generators are equipped with the conventional power system stabilizers with fixed parameters. The control structure of the conventional power system stabilizers and the tuning procedure are based on the linear control theory. Conventional power system stabilizers are simple to realize, but they show non-sufficient damping improvement in the entire operating conditions. This is the reason that advanced control theories are used for development of better power system stabilizers. In this paper, the adaptive control theory for power system stabilizers design and synthesis is studied. The presented work is focused on the use of model reference adaptive control approach. Control signal, which assures that the controlled plant output will follow the reference model output, is generated by the adaptive algorithm. Adaptive gains are obtained as a combination of the "proportional" term and with the σ-term extended "integral" term. The σ-term is introduced to avoid divergence of the integral gains. The necessary condition for asymptotic tracking is derived by means of hyperstability theory. The benefits of the proposed model reference adaptive power system stabilizer were evaluated as objectively as possible by means of a theoretical analysis, numerical simulations and laboratory realizations. Damping of the synchronous generator oscillations in the entire operating range was investigated. Obtained results show the improved damping in the entire operating area and the increase of the power system stability. The results of the presented work will help by the development of the model reference power system stabilizer which should be able to replace the conventional stabilizers in power systems.

Improving the Frequency Response of a Circular Dual-Mode Resonator with a Reconfigurable Bandwidth

In this paper, a method for reconfiguring bandwidth in a circular dual-mode resonator is presented. The method concerns the optimized geometry of a structure that may be used to host the tuning elements, which are typically RF (Radio Frequency) switches. The tuning elements themselves, and their performance during tuning, are not the focus of this paper. The designed resonator is able to reconfigure its fractional bandwidth by adjusting the inter-coupling level between the degenerate modes, while at the same time improving its response by adjusting the external-coupling level and keeping the center frequency fixed. The inter-coupling level has been adjusted by changing the dimensions of the perturbation element, while the external-coupling level has been adjusted by changing one of the feeder dimensions. The design was arrived at via optimization. Agreeing simulation and measurement results of the designed and implemented filters showed good improvements in return loss values and the stability of the center frequency.

Doubly Fed Induction Generator Based Variable Speed Wind Conversion System Control Enhancement by Applying Fractional Order Controller

In an electric power grid connected wind generation system, dynamic control strategy is essential to use the wind energy efficiently as well as for an energy optimization. The present study has focused on decoupled power regulation of doubly fed induction generator, operating in wind turbine, in accordance with the vector control approach by applying fractional order proportional integral (FOPI) controller. The FOPI controller is designed based on a simple method; up such that the response of closed loop process is similar to the response of a specified fractional model whose transfer function is Bode’s ideal function. In this tuning operation, the parameters of the proposed fractional controller are established analytically using the impulse closed-loop response of the controlled process. To show the superior action of the developed FOPI controller in comparison with standard PI controller in different function conditions, the study is validated through simulation using the software MATLAB/Simulink.

Fast Adjustable Threshold for Uniform Neural Network Quantization

The neural network quantization is highly desired procedure to perform before running neural networks on mobile devices. Quantization without fine-tuning leads to accuracy drop of the model, whereas commonly used training with quantization is done on the full set of the labeled data and therefore is both time- and resource-consuming. Real life applications require simplification and acceleration of quantization procedure that will maintain accuracy of full-precision neural network, especially for modern mobile neural network architectures like Mobilenet-v1, MobileNet-v2 and MNAS. Here we present a method to significantly optimize training with quantization procedure by introducing the trained scale factors for discretization thresholds that are separate for each filter. Using the proposed technique, we quantize the modern mobile architectures of neural networks with the set of train data of only ∼ 10% of the total ImageNet 2012 sample. Such reduction of train dataset size and small number of trainable parameters allow to fine-tune the network for several hours while maintaining the high accuracy of quantized model (accuracy drop was less than 0.5%). Ready-for-use models and code are available in the GitHub repository.

Performance Evaluation of Distributed Deep Learning Frameworks in Cloud Environment

2016 has become the year of the Artificial Intelligence explosion. AI technologies are getting more and more matured that most world well-known tech giants are making large investment to increase the capabilities in AI. Machine learning is the science of getting computers to act without being explicitly programmed, and deep learning is a subset of machine learning that uses deep neural network to train a machine to learn  features directly from data. Deep learning realizes many machine learning applications which expand the field of AI. At the present time, deep learning frameworks have been widely deployed on servers for deep learning applications in both academia and industry. In training deep neural networks, there are many standard processes or algorithms, but the performance of different frameworks might be different. In this paper we evaluate the running performance of two state-of-the-art distributed deep learning frameworks that are running training calculation in parallel over multi GPU and multi nodes in our cloud environment. We evaluate the training performance of the frameworks with ResNet-50 convolutional neural network, and we analyze what factors that result in the performance among both distributed frameworks as well. Through the experimental analysis, we identify the overheads which could be further optimized. The main contribution is that the evaluation results provide further optimization directions in both performance tuning and algorithmic design.

Design of Identification Based Adaptive Control for Fermentation Process in Bioreactor

The biochemical technology has been developing extremely fast since the middle of the last century. The main reason for such development represents a requirement for large production of high-quality biologically manufactured products such as pharmaceuticals, foods, and beverages. The impact of the biochemical industry on the world economy is enormous. The great importance of this industry also results in intensive development in scientific disciplines relevant to the development of biochemical technology. In addition to developments in the fields of biology and chemistry, which enable to understand complex biochemical processes, development in the field of control theory and applications is also very important. In the paper, the control for the biochemical reactor for the milk fermentation was studied. During the fermentation process, the biophysical quantities must be precisely controlled to obtain the high-quality product. To control these quantities, the bioreactor’s stirring drive and/or heating system can be used. Available commercial biochemical reactors are equipped with open loop or conventional linear closed loop control system. Due to the outstanding parameters variations and the partial nonlinearity of the biochemical process, the results obtained with these control systems are not satisfactory. To improve the fermentation process, the self-tuning adaptive control system was proposed. The use of the self-tuning adaptive control is suggested because the parameters’ variations of the studied biochemical process are very slow in most cases. To determine the linearized mathematical model of the fermentation process, the recursive least square identification method was used. Based on the obtained mathematical model the linear quadratic regulator was tuned. The parameters’ identification and the controller’s synthesis are executed on-line and adapt the controller’s parameters to the fermentation process’ dynamics during the operation. The use of the proposed combination represents the original solution for the control of the milk fermentation process. The purpose of the paper is to contribute to the progress of the control systems for the biochemical reactors. The proposed adaptive control system was tested thoroughly. From the obtained results it is obvious that the proposed adaptive control system assures much better following of the reference signal as a conventional linear control system with fixed control parameters.

Bidirectional Pendulum Vibration Absorbers with Homogeneous Variable Tangential Friction: Modelling and Design

Passive resonant vibration absorbers are among the most widely used dynamic control systems in civil engineering. They typically consist in a single-degree-of-freedom mechanical appendage of the main structure, tuned to one structural target mode through frequency and damping optimization. One classical scheme is the pendulum absorber, whose mass is constrained to move along a curved trajectory and is damped by viscous dashpots. Even though the principle is well known, the search for improved arrangements is still under way. In recent years this investigation inspired a type of bidirectional pendulum absorber (BPA), consisting of a mass constrained to move along an optimal three-dimensional (3D) concave surface. For such a BPA, the surface principal curvatures are designed to ensure a bidirectional tuning of the absorber to both principal modes of the main structure, while damping is produced either by horizontal viscous dashpots or by vertical friction dashpots, connecting the BPA to the main structure. In this paper, a variant of BPA is proposed, where damping originates from the variable tangential friction force which develops between the pendulum mass and the 3D surface as a result of a spatially-varying friction coefficient pattern. Namely, a friction coefficient is proposed that varies along the pendulum surface in proportion to the modulus of the 3D surface gradient. With such an assumption, the dissipative model of the absorber can be proven to be nonlinear homogeneous in the small displacement domain. The resulting homogeneous BPA (HBPA) has a fundamental advantage over conventional friction-type absorbers, because its equivalent damping ratio results independent on the amplitude of oscillations, and therefore its optimal performance does not depend on the excitation level. On the other hand, the HBPA is more compact than viscously damped BPAs because it does not need the installation of dampers. This paper presents the analytical model of the HBPA and an optimal methodology for its design. Numerical simulations of single- and multi-story building structures under wind and earthquake loads are presented to compare the HBPA with classical viscously damped BPAs. It is shown that the HBPA is a promising alternative to existing BPA types and that homogeneous tangential friction is an effective means to realize systems provided with amplitude-independent damping.

Noise Performance of Magnetic Field Tunable Avalanche Transit Time Source

The effect of magnetic field on the noise performance of the magnetic field tunable avalanche transit time (MAGTATT) device based on Si, designed to operate at W-band (75 – 110 GHz), has been studied in this paper. A comprehensive two-dimensional (2D) model has been developed. The simulation results show that due to the presence of applied external transverse magnetic field, both the noise spectral density and noise measure of the MAGTATT device increase significantly. The noise performance of the device has been found to be further deteriorated if the magnetic field strength is further increased. Hence, in order to achieve the magnetic field tuning of the radio frequency (RF) properties of impact avalanche transit time (IMPATT) source, the noise performance of it has to be sacrificed in fair extent. Moreover, it clearly indicates that an IMPATT source must be covered with appropriate magnetic shielding material to avoid undesirable shift in operating frequency and output power and objectionable amount of deterioration in noise performance due to the presence of external magnetic field.

A Hybrid Algorithm for Collaborative Transportation Planning among Carriers

In this paper, there is concentration on collaborative transportation planning (CTP) among multiple carriers with pickup and delivery requests and time windows. This problem is a vehicle routing problem with constraints from standard vehicle routing problems and new constraints from a real-world application. In the problem, each carrier has a finite number of vehicles, and each request is a pickup and delivery request with time window. Moreover, each carrier has reserved requests, which must be served by itself, whereas its exchangeable requests can be outsourced to and served by other carriers. This collaboration among carriers can help them to reduce total transportation costs. A mixed integer programming model is proposed to the problem. To solve the model, a hybrid algorithm that combines Genetic Algorithm and Simulated Annealing (GASA) is proposed. This algorithm takes advantages of GASA at the same time. After tuning the parameters of the algorithm with the Taguchi method, the experiments are conducted and experimental results are provided for the hybrid algorithm. The results are compared with those obtained by a commercial solver. The comparison indicates that the GASA significantly outperforms the commercial solver.

Crude Oil Price Prediction Using LSTM Networks

Crude oil market is an immensely complex and dynamic environment and thus the task of predicting changes in such an environment becomes challenging with regards to its accuracy. A number of approaches have been adopted to take on that challenge and machine learning has been at the core in many of them. There are plenty of examples of algorithms based on machine learning yielding satisfactory results for such type of prediction. In this paper, we have tried to predict crude oil prices using Long Short-Term Memory (LSTM) based recurrent neural networks. We have tried to experiment with different types of models using different epochs, lookbacks and other tuning methods. The results obtained are promising and presented a reasonably accurate prediction for the price of crude oil in near future.

A Tuning Method for Microwave Filter via Complex Neural Network and Improved Space Mapping

This paper presents an intelligent tuning method of microwave filter based on complex neural network and improved space mapping. The tuning process consists of two stages: the initial tuning and the fine tuning. At the beginning of the tuning, the return loss of the filter is transferred to the passband via the error of phase. During the fine tuning, the phase shift caused by the transmission line and the higher order mode is removed by the curve fitting. Then, an Cauchy method based on the admittance parameter (Y-parameter) is used to extract the coupling matrix. The influence of the resonant cavity loss is eliminated during the parameter extraction process. By using processed data pairs (the amount of screw variation and the variation of the coupling matrix), a tuning model is established by the complex neural network. In view of the improved space mapping algorithm, the mapping relationship between the actual model and the ideal model is established, and the amplitude and direction of the tuning is constantly updated. Finally, the tuning experiment of the eight order coaxial cavity filter shows that the proposed method has a good effect in tuning time and tuning precision.

The Design Optimization for Sound Absorption Material of Multi-Layer Structure

Sound absorbing material is used as automotive interior material. Sound absorption coefficient should be predicted to design it. But it is difficult to predict sound absorbing coefficient because it is comprised of several material layers. So, its targets are achieved through many experimental tunings. It causes a lot of cost and time. In this paper, we propose the process to estimate the sound absorption coefficient with multi-layer structure. In order to estimate the coefficient, physical properties of each material are used. These properties also use predicted values by Foam-X software using the sound absorption coefficient data measured by impedance tube. Since there are many physical properties and the measurement equipment is expensive, the values predicted by software are used. Through the measurement of the sound absorption coefficient of each material, its physical properties are calculated inversely. The properties of each material are used to calculate the sound absorption coefficient of the multi-layer material. Since the absorption coefficient of multi-layer can be calculated, optimization design is possible through simulation. Then, we will compare and analyze the calculated sound absorption coefficient with the data measured by scaled reverberation chamber and impedance tubes for a prototype. If this method is used when developing automotive interior materials with multi-layer structure, the development effort can be reduced because it can be optimized by simulation. So, cost and time can be saved.

12x12 MIMO Terminal Antennas Covering the Whole LTE and WiFi Spectrum

A broadband resonant terminal antenna has been developed. It can be used in different MIMO arrangements such as 2x2, 4x4, 8x8, or even 12x12 MIMO configurations. The antenna covers the whole LTE and WiFi bands besides the existing 2G/3G bands (700-5800 MHz), without using any matching/tuning circuits. Matching circuits significantly reduce the efficiency of any antenna and reduce the battery life. They also reduce the bandwidth because they are frequency dependent. The antenna can be implemented in smartphone handsets, tablets, laptops, notebooks or any other terminal. It is also suitable for different IoT and vehicle applications. The antenna is manufactured from a flexible material and can be bent or folded and shaped in any form to fit any available space in any terminal. It is self-contained and does not need to use the ground plane, the chassis or any other component of the terminal. Hence, it can be mounted on any terminal at different positions and configurations. Its performance does not get affected by the terminal, regardless of its type, shape or size. Moreover, its performance does not get affected by the human body of the terminal’s users. Because of all these unique features of the antenna, multiples of them can be simultaneously used for MIMO diversity coverage in any terminal device with a high isolation and a low correlation factor between them.

Batteryless DCM Boost Converter for Kinetic Energy Harvesting Applications

In this paper, a bidirectional boost converter operated in Discontinuous Conduction Mode (DCM) is presented as a suitable power conditioning circuit for tuning of kinetic energy harvesters without the need of a battery. A nonlinear control scheme, composed by two linear controllers, is used to control the average value of the input current, enabling the synthesization of complex loads. The converter, along with the control system, is validated through SPICE simulations using the LTspice tool. The converter model and the controller transfer functions are derived. From the simulation results, it was found that the input current distortion increases with the introduced phase shift and that, such distortion, is almost entirely present at the zero-crossing point of the input voltage.