Input-output Mapping Research Articles

Efficient representation and feature extraction for neural network-based 3D object pose estimation

This paper introduces an efficient representation and feature extraction technique for 3D pose estimation of objects, incorporating a novel mechanism for the exploitation of the extracted visual cues. A combination of a fuzzy clustering technique for the input space, with supervised learning, results in a problem of reduced dimensionality and an efficient mapping of the input–output space. While other neural network-based approaches for 3D pose estimation focus on reducing dimensionality based on input space characteristics, such as with PCA-based approaches, the proposed scheme directly targets the input–output mapping, based on the available visual data. Evaluation results provide evidence of low generalization error when estimating the 3D pose of objects, with the best performance achieved when employing Radial Basis Functions. The proposed system can be adopted in several computer vision applications requiring object localization, pose estimation and target tracking.

Dynamic neural network-based fault diagnosis of gas turbine engines

In this paper, a neural network-based fault detection and isolation (FDI) scheme is presented to detect and isolate faults in a highly nonlinear dynamics of an aircraft jet engine. Towards this end, dynamic neural networks (DNN) are first developed to learn the input–output map of the jet engine. The DNN is constructed based on a multi-layer perceptron network which uses an IIR (infinite impulse response) filter to generate dynamics between the input and output of a neuron, and consequently of the entire neural network. The trained dynamic neural network is then utilized to detect and isolate component faults that may occur in a dual spool turbo fan engine. The fault detection and isolation schemes consist of multiple DNNs or parallel bank of filters, corresponding to various operating modes of the healthy and faulty engine conditions. Using the residuals that are generated by measuring the difference of each network output and the measured engine output various criteria are established for accomplishing the fault diagnosis task, that is addressing the problem of fault detection and isolation of the system components. A number of simulation studies are carried out to demonstrate and illustrate the advantages, capabilities, and performance of our proposed fault diagnosis scheme.

Semigroup approach for identification of the unknown diffusion coefficient in a linear parabolic equation with mixed output data

This article presents a semigroup approach for the mathematical analysis of the inverse coefficient problems of identifying the unknown coefficient in the linear parabolic equation with mixed boundary conditions , . The aim of this paper is to investigate the distinguishability of the input-output mappings , via semigroup theory. In this paper, we show that if the null space of the semigroup consists of only zero function, then the input-output mappings and have the distinguishability property. It is shown that the types of the boundary conditions and the region on which the problem is defined have a significant impact on the distinguishability property of these mappings. Moreover, in the light of measured output data (boundary observations) or/and , the values and of the unknown diffusion coefficient at and , respectively, can be determined explicitly. In addition to these, the values and of the unknown coefficient at and , respectively, are also determined via the input data. Furthermore, it is shown that measured output data and can be determined analytically by an integral representation. Hence the input-output mappings , are given explicitly in terms of the semigroup.

An online self-adaptive modular neural network for time-varying systems

We propose an online self-adaptive modular neural network (OSAMNN) for time-varying systems. Starting with zero subnetworks, OSAMNN uses a single-pass subtractive cluster algorithm to update the centers of radial-basis function (RBF) neurons for learning. Then the input space can be partitioned. The OSAMNN structure is capable of growing or merging subnetworks to maintain suitable model complexity, and the centers of RBF neurons can also be dynamically adjusted according to changes in the data environment. A fuzzy strategy is applied to select suitable subnetworks to learn the current sample. This method yields improved learning efficiency and accuracy. OSAMNN can adapt its architecture to realize online modeling of time-varying nonlinear input–output maps. Results for experiments on benchmark and real-world time-varying systems support the proposed techniques.

Enhancing Knowledge Hyper Surface Method for Casting Diagnosing

The diagnosis of defective castings has always been a centre of attention in the manufacturing industry. This is mainly because the cause and effect relationship in a casting process is complex and non-linear. Furthermore, a large number of parameters are needed to be coordinated with each other in an optimal way to minimise the occurrence of defective castings. An intelligent diagnosis system is needed to diagnose effectively the causal representation and also justify its diagnosis. A previous method, known as the Knowledge Hyper-surface method which used Lagrange Interpolation polynomials has gained more popularity in learning cause and effect analysis in casting processes. The current method show that the belief value of the occurrence of cause with respect to the change in the belief value in the occurrence of effect can be modeled by linear, quadratic or cubic relationships and the method retained the advantages of neural networks and overcomes their limitations in learning the input-output mapping function in the presence of noisy, limited and sparse data. However, the methodology was unable to model exponential increase/decrease in belief values in cause and effect relationships . This paper proposed an enhancement to the current Knowledge Hyper-surface method by introducing midpoints in the existing shape formulation which further constrains the shape of the Knowledge hyper-surfaces to model an exponential rise in belief values but without exposing the dataset to the limitations of ‘over fitting’. The ability of the proposed method to capture the exponential change in the belief variation of the cause when the belief in the effect is at its minimum is compared to the current method on real casting data. DOI: http://dx.doi.org/10.11591/ij-ai.v2i1.584

Genetic interval neural networks for granular data regression

Granular data and granular models offer an interesting tool for representing data in problems involving uncertainty, inaccuracy, variability and subjectivity have to be taken into account. In this paper, we deal with a particular type of information granules, namely interval-valued data. We propose a multilayer perceptron (MLP) to model interval-valued input–output mappings. The proposed MLP comes with interval-valued weights and biases, and is trained using a genetic algorithm designed to fit data with different levels of granularity. In the evolutionary optimization, two implementations of the objective function, based on a numeric-valued and an interval-valued network error, respectively, are discussed and compared. The modeling capabilities of the proposed MLP are illustrated by means of its application to both synthetic and real world datasets.

An artificial neural network approach to the estimation of stem water potential from frequency domain reflectometry soil moisture measurements and meteorological data

Stem water potential seems to be a sensitive measure of plant water status. Nonetheless, it is a labour-intensive measurement and is not suited for automatic irrigation scheduling or control. This study describes the application of artificial neural networks to estimate stem water potential from soil moisture at different depths and standard meteorological variables, considering a limited data set. The experiment was carried out with ‘Navelina’ citrus trees grafted on ‘Cleopatra’ mandarin. Principal components analysis and multiple linear regression were used preliminarily to assess the relationships among observations and to propose other models to allow a comparative analysis, respectively. Two principal components account for the systematic data variation. The optimum regression equation of stem water potential considered temperature, relative humidity, solar radiation and soil moisture at 50cm as input variables, with a determination coefficient of 0.852. When compared with their corresponding regression models, ANNs presented considerably higher performance accuracy (with an optimum determination coefficient of 0.926) due to a higher input-output mapping ability.

Revisiting age-of-acquisition effects in Spanish visual word recognition: The role of item imageability.

Word age-of-acquisition (AoA) affects reading. The mapping hypothesis predicts AoA effects when input-output mappings are arbitrary. In Spanish, the orthography-to-phonology mappings required for word naming are consistent; therefore, no AoA effects are expected. Nevertheless, AoA effects have been found, motivating the present investigation of how AoA can affect reading in Spanish. Four experiments were run to examine reading with a factorial design manipulating AoA and frequency. In Experiments 1 and 2 (immediate and speeded naming), only word frequency affected word naming. In Experiment 3 (lexical decision), both AoA and frequency affected word recognition. In Experiment 4 (immediate naming with highly imageable items), both frequency and AoA affected naming. The results suggest that highly imageable items induce a larger reliance on semantics in reading aloud. Such reliance causes faster naming of earlier acquired words because the corresponding concepts have richer visual and sensory features acquired mainly through direct sensory experience.

Structural Steel Plate Damage Detection using Non Destructive Testing, Frame Energy based Statistical Features and Artificial Neural Networks

Abstract This paper discusses about the detection of damages present in the steel plates using nondestructive vibration testing. A simple experimental model has been developed to hold the steel plate complying with the simply supported boundary condition. Vibration patterns from the steel structure are captured based on the impact testing using a simple protocol. The vibration signals in normal condition of the steel plate are recorded. The damages of size 512 |x m to 1852 |j, m are simulated manually on the steel plate using drill bits. The vibration signals in the fault condition of the steel plate are collected. The captured vibration signals are preprocessed and time domain based feature extraction algorithms are developed to extract features from the vibration signals. The conditions of the steel plate namely healthy and faulty are associated with the extracted features to establish input output mapping. A feed-forward neural network is modeled to classify the condition. The neural network parameters are adjusted to train the network. The performance of the network is calculated using Falhman criterion.

Neural Network for WGDOP Approximation and Mobile Location

This paper considers location methods that are applicable in global positioning systems (GPS), wireless sensor networks (WSN), and cellular communication systems. The approach is to employ the resilient backpropagation (Rprop), an artificial neural network learning algorithm, to compute weighted geometric dilution of precision (WGDOP), which represents the geometric effect on the relationship between measurement error and positioning error. The original four kinds of input-output mapping based on BPNN for GDOP calculation are extended to WGDOP based on Rprop. In addition, we propose two novel Rprop–based architectures to approximate WGDOP. To further reduce the complexity of our approach, the first is to select the serving BS and then combines it with three other measurements to estimate MS location. As such, the number of subsets is reduced greatly without compromising the location estimation accuracy. We further employed another Rprop that takes the higher precision MS locations of the first several minimum WGDOPs as the inputs into consideration to determine the final MS location estimation. This method can not only eliminate the poor geometry effects but also significantly improve the location accuracy.

Prediction of Backbreak in Open Pit Blasting by Adaptive Neuro-Fuzzy Inference System / Prognozowanie Spękań Skał Przy Pracach Strzałowych W Kopalniach Odkrywkowych Przy Użyciu Metod Neuronowych I Wnioskowania Rozmytego (Anfis) Zastosowanych W Modelu Adaptywnym

Abstract Adaptive neuro-fuzzy inference system (ANFIS) is powerful model in solving complex problems. Since ANFIS has the potential of solving nonlinear problem and can easily achieve the input-output mapping, it is perfect to be used for solving the predicting problem. Backbreak is one of the undesirable effects of blasting operations causing instability in mine walls, falling down the machinery, improper fragmentation and reduction in efficiency of drilling. In this paper, ANFIS was applied to predict backbreak in Sangan iron mine of Iran. The performance of the model was assessed through the root mean squared error (RMSE), the variance account for (VAF) and the correlation coefficient (R2) computed from the measured of backbreak and model-predicted values of the dependent variables. The RMSE, VAF, R2 indices were calculated 0.6, 0.94 and 0.95 for ANFIS model. As results, these indices revealed that the ANFIS model has very good prediction performance.

Risk Score Inference for Bridge Maintenance Project Using Evolutionary Fuzzy Least Squares Support Vector Machine

Bridges are essential infrastructure of the transportation network. Therefore, maintenance tasks are mandatory to prevent these structures from degradation over time. In practice, funding availability for maintenance projects is often confined and this necessitates prioritization of different bridges that are in need of remedial activities. The writers aim to construct an artificial intelligence (AI) approach, evolutionary fuzzy least-squares support-vector machine (LSSVM) inference model (EFLSIM), for prioritizing bridges based on risk scores (RSs). In EFLSIM, fuzzy logic (FL) is utilized to enhance the capability of approximate reasoning and to deal with subjective information, which is obtained from human judgment. The inference model employs LSSVM as a supervised learning technique to infer the fuzzy input-output mapping relationship. Differential evolution (DE) is integrated into the model to optimize its tuning parameters. Experimental results and comparison illustrates that EFLSIM can successfully absorb and simulate human knowledge in the bridge-assessment process. Additionally, the newly built model has outperformed other benchmark approaches in terms of both reliability and accuracy. A 10-fold cross-validation process has demonstrated that the EFLSIM has achieved more than 38% reduction in RMS error compared to other benchmark methods. Thus, the proposed AI approach is a promising tool to support decision-makers in bridge-maintenance planning.

Defective Casting Diagnosing via an Enhanced Knowledge Hyper-Surface Method

The research on the analysis of cause and effect relationships in castings has always been a centre of attention in the manufacturing industry. An intelligent diagnosis system should be able to diagnose effectively the causal representation and also justify its diagnosis. Recently, a method, known as the Knowledge Hyper-surface method which used Lagrange Interpolation polynomials has gained more popularity in learning cause and effect analysis in casting processes. The current method show that the belief value of the occurrence of cause with respect to the change in the belief value in the occurrence of effect can be modelled by linear, quadratic or cubic relationships and the method retained the advantages of neural networks and overcomes their limitations in learning the input-output mapping function in the presence of noisy, limited and sparse data. However, the methodology was unable to model exponential increase/decrease in belief values in cause and effect relationships. This paper proposed an enhancement to the current Knowledge Hyper-surface method by introducing midpoints in the existing shape formulation which further constrains the shape of the Knowledge hyper-surfaces to model an exponential rise in belief values but without exposing the dataset to the limitations of ‘over fitting’. The ability of the proposed method to capture the exponential change in the belief variation of the cause when the belief in the effect is at its minimum is compared to the current method on real casting data.

Nonsingular decoupled terminal sliding-mode control for a class of fourth-order nonlinear systems

This paper presents a nonsingular decoupled terminal sliding mode control (NDTSMC) method for a class of fourth-order nonlinear systems. First, the nonlinear fourth-order system is decoupled into two second-order subsystems which are referred to as the primary and secondary subsystems. The sliding surface of each subsystem was designed by utilizing time-varying coefficients which are computed by linear functions derived from the input–output mapping of the one-dimensional fuzzy rule base. Then, the control target of the secondary subsystem was embedded to the primary subsystem by the help of an intermediate signal. Thereafter, a nonsingular terminal sliding mode control (NTSMC) method was utilized to make both subsystems converge to their equilibrium points in finite time. The simulation results on the inverted pendulum system are given to show the effectiveness of the proposed method. It is seen that the proposed method exhibits a considerable improvement in terms of a faster dynamic response and lower IAE and ITAE values as compared with the existing decoupled control methods.

Hierarchical multivariate regression-based sensitivity analysis reveals complex parameter interaction patterns in dynamic models

Dynamic models of biological systems often possess complex and multivariate mappings between input parameters and output state variables, posing challenges for comprehensive sensitivity analysis across the biologically relevant parameter space. In particular, more efficient and robust ways to obtain a solid understanding of how the sensitivity to each parameter depends on the values of the other parameters are sorely needed.We report a new methodology for global sensitivity analysis based on Hierarchical Cluster-based Partial Least Squares Regression (HC-PLSR)-based approximations (metamodelling) of the input–output mappings of dynamic models, which we expect to be generic, efficient and robust, even for systems with highly nonlinear input–output relationships. The two-step HC-PLSR metamodelling automatically separates the observations (here corresponding to different combinations of input parameter values) into groups based on the dynamic model behaviour, then analyses each group separately with Partial Least Squares Regression (PLSR). This produces one global regression model comprising all observations, as well as regional regression models within each group, where the regression coefficients can be used as sensitivity measures. Thereby a more accurate description of complex interactions between inputs to the dynamic model can be revealed through analysis of how a certain level of one input parameter affects the model sensitivity to other inputs. We illustrate the usefulness of the HC-PLSR approach on a dynamic model of a mouse heart muscle cell, and demonstrate how it reveals interaction patterns of probable biological significance not easily identifiable by a global regression-based sensitivity analysis alone.Applied for sensitivity analysis of a complex, high-dimensional dynamic model of the mouse heart muscle cell, several interactions between input parameters were identified by the two-step HC-PLSR analysis that could not be detected in the single-step global analysis. Hence, our approach has the potential to reveal new biological insight through the identification of complex parameter interaction patterns. The HC-PLSR metamodel complexity can be adjusted according to the nonlinear complexity of the input–output mapping of the analysed dynamic model through adjustment of the number of regional regression models included. This facilitates sensitivity analysis of dynamic models of varying complexities.

Respiratory Motion Estimation With Hybrid Implementation of Extended Kalman Filter

The extended Kalman filter (EKF) can be used for the purpose of training nonlinear neural networks to perform desired input-output mappings. To improve the computational requirements of the EKF, Puskorius proposed the decoupled EKF (DEKF) as a practical remedy for the proper management of computational resources. This approach, however, sacrifices computational accuracy of estimates because it ignores the interactions between the estimates of mutually exclusive weights. To overcome such a limitation, therefore, we proposed hybrid implementation based on EKF (HEKF) for respiratory motion estimation, which uses the channel number for the mutually exclusive groups and the coupling technique to compensate the computational accuracy. Moreover, the authors restricted to a DEKF algorithm in which the weights connecting the inputs to a node are grouped together. If there are multiple input training sequences with respect to the time stamp, the complexity can increase by the power of the input channel number. To improve the computational complexity, we split the complicated neural network into a couple of simple neural networks to adjust separate input channels. The experimental results validated that the prediction overshoot of the proposed HEKF was improved by 62.95% in the average prediction overshoot values. The proposed HEKF showed a better performance of 52.40% improvement in the average of the prediction time horizon. We have evaluated that the proposed HEKF can outperform DEKF by comparing the prediction overshoot values, the performance of the tracking estimation value, and the normalized root-mean-squared error.

Control Communication Complexity of Distributed Actions

Recent papers have treated control communication complexity in the context of information-based, multiple agent control systems including nonlinear systems of the type that have been studied in connection with quantum information processing. The present paper continues this line of investigation into a class of two-agent distributed control systems in which the agents cooperate in order to realize common goals that are determined via independent actions undertaken individually by the agents. A basic assumption is that the actions taken are unknown in advance to the other agent. These goals can be conveniently summarized in the form of a target matrix, whose entries are computed by the control system responding to the choices of inputs made by the two agents. We show how to realize such target matrices for a broad class of systems that possess an input-output mapping that is bilinear. One can classify control-communication strategies, known as control protocols, according to the amount of information sharing occurring between the two agents. Protocols that assume no information sharing on the inputs that each agent selects and protocols that allow sufficient information sharing for identifying the common goals are the two extreme cases. Control protocols will also be evaluated and compared in terms of cost functionals given by integrated quadratic functions of the control inputs. The minimal control cost of the two classes of control protocols are analyzed and compared. The difference in the control costs between the two classes reflects an inherent trade-off between communication complexity and control cost.

System identification of smart structures using a wavelet neuro-fuzzy model

This paper proposes a complex model of smart structures equipped with magnetorheological (MR) dampers. Nonlinear behavior of the structure–MR damper systems is represented by the use of a wavelet-based adaptive neuro-fuzzy inference system (WANFIS). The WANFIS is developed through the integration of wavelet transforms, artificial neural networks, and fuzzy logic theory. To evaluate the effectiveness of the WANFIS model, a three-story building employing an MR damper under a variety of natural hazards is investigated. An artificial earthquake is used for training the input–output mapping of the WANFIS model. The artificial earthquake is generated such that the characteristics of a variety of real recorded earthquakes are included. It is demonstrated that this new WANFIS approach is effective in modeling nonlinear behavior of the structure–MR damper system subjected to a variety of disturbances while resulting in shorter training times in comparison with an adaptive neuro-fuzzy inference system (ANFIS) model. Comparison with high fidelity data proves the viability of the proposed approach in a structural health monitoring setting, and it is validated using known earthquake signals such as El-Centro, Kobe, Northridge, and Hachinohe.

Computational Evidence That Frequency Trajectory Theory Does Not Oppose But Emerges From Age‐of‐Acquisition Theory

According to the age-of-acquisition hypothesis, words acquired early in life are processed faster and more accurately than words acquired later. Connectionist models have begun to explore the influence of the age/order of acquisition of items (and also their frequency of encounter). This study attempts to reconcile two different methodological and theoretical approaches (proposed by Lambon Ralph & Ehsan, 2006 and Zevin & Seidenberg, 2002) to age-limited learning effects. The current simulations extend the findings reported by Zevin and Seidenberg (2002) that have shown that frequency trajectories (FTs) have limited and specific effects on word-reading tasks. Using the methodological framework proposed by Lambon Ralph and Ehsan (2006), which makes it possible to compare word-reading and picture-naming tasks in connectionist networks, we were able to show that FT has a considerable influence on age-limited learning effects in a picture naming task. The findings show that when the input-output mappings are arbitrary (simulating picture naming tasks), the links formed by the network become entrenched as a result of early experience and that subsequent variations in frequency of exposure of the items have only a minor impact. In contrast, when the mappings between input-output are quasi-systematic or systematic (simulating word-reading tasks), the training of new items was generalized and resulted in the suppression of age-limited learning effects. At a theoretical level, we suggest that FT, which simultaneously takes account of time and the level of exposure across time, represents a more precise and modulated measure compared with the order of introduction of the items and may lead to innovative hypotheses in the field of age-limited learning effects.

Center of mass acceleration feedback control of standing balance by functional neuromuscular stimulation against external postural perturbations.

This study investigated the use of center of mass (COM) acceleration feedback for improving performance of a functional neuromuscular stimulation control system to restore standing function to a subject with complete, thoracic-level spinal cord injury. The approach for linearly relating changes in muscle stimulation to changes in COM acceleration was verified experimentally and subsequently produced data to create an input-output map driven by sensor feedback. The feedback gains were systematically tuned to reduce upper extremity (UE) loads applied to an instrumented support device while resisting external postural disturbances. Total body COM acceleration was accurately estimated (>89% variance explained) using 3-D outputs of two accelerometers mounted on the pelvis and torso. Compared to constant muscle stimulation employed clinically, feedback control of stimulation reduced UE loading by 33%. COM acceleration feedback is advantageous in constructing a standing neuroprosthesis since it provides the basis for a comprehensive control synergy about a global, dynamic variable and requires minimal instrumentation. Future work should include tuning and testing the feedback control system during functional reaching activity that is more indicative of activities of daily living.