Time Delay Neural Network Research Articles

Human Identification by Gait Using Time Delay Neural Networks

<p>This paper proposed human identification method by gait. Human gait is a type of biometric features and related to the physiological and behavioral features of a human. In this paper, a feature vector of gait motion parameters is extracted from each frame using image segmentation methods, and categorized into different categories. Two of these categories were used to form the gait motion trajectories; Category one: Gait angle velocity: angle velocity hip, angle velocity knee, angle velocity thigh and angle velocity shank. Category two: Gait angle acceleration: angle acceleration hip, angle acceleration knee, angle acceleration thigh and angle acceleration shank for each image sequence. Finally, the TDNN method with different training algorithms is used for recognition purpose. This experiment is done on our own database. This research developed a method which achieves a higher recognition rate in the training set 100% and in the testing set 83%. Also, category one establishes gait motion features to be used in human gait identification applications using different training algorithms, While category two achieved a higher recognition rate by trainrb algorithm.</p>

Fault detection and isolation of gas turbine engines using a bank of neural networks

The main goal of this paper is to design and develop a fault detection and isolation (FDI) scheme for aircraft gas turbine engines by using neural networks. Towards this end, first for the fault detection task two types of dynamic neural networks are used and compared to learn the engine dynamics. Specially, the dynamic neural model (DNM) and the time delay neural network (TDNN) are utilized. For both architectures a bank of neural networks is trained separately to capture the dynamic relationships among the engine measurable variables. The results show that certain engine parameters have better detection capabilities as compared to the others. Finally, the fault isolation task is accomplished by using a multilayer perception (MLP) network functioning as a pattern classifier applied to the residual signals that are generated by the two dynamic neural networks used for the purpose of the fault detection task. The simulation results do indeed substantiate and verify that our proposed FDI scheme represents a promising tool for aircraft engine diagnostics and health monitoring.

Modeling nonlinear behavior of Buckling-Restrained Braces via different artificial intelligence methods

Five AI models are presented to model the dynamic nonlinear behavior of Buckling-Restrained Braces (BRBs). The AI techniques utilized in the models are: Time-Delayed Neural Networks (TDNN), Nonlinear Auto-Regressive eXogenous (NARX) neural networks, Gaussian-Mixture Models Regression (GMMR), Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Polynomial Classifier Regression (PCR). The models are developed using time-delayed brace displacements inputs and brace force outputs to predict updated brace forces during load reversals. The training and testing of the AI models are performed using experimental data from BRB specimens tested at the Pacific Earthquake Engineering Research (PEER) Center. The training stage for every method makes use of the experimental data from one specimen. In order to assess the models’ learning and generalization capabilities, three sets of experimental data for different specimens are used. To arrive at an optimized architecture that best models the phenomenon, the model performance with different parameters is evaluated. The brace force predicted by the proposed model shows excellent resemblance to the experimental results for the training sample, for all techniques. The predicted behavior of the testing samples shows noticeable accuracy and further demonstrates the generalization and prediction capability of the proposed modeling techniques. The various techniques are compared on the basis of selected performance criteria. It is found that the performance of two AI techniques standout among the others: the NARX and the PCR. Although the NARX demonstrates a slight advantage in the prediction accuracy over the PCR, the latter is far more superior in terms of computational efficiency. Thus, the PCR would be recommended for scenarios where online training is needed. The BRB design and performance investigation processes can be facilitated by the developed modeling techniques thus minimizing the need for, and extent of, experimental testing.

Neuro-computational frameworks for non-linear stochastic wireless channels

Inter symbol interference and related distortions observed in high data rate wireless systems are mitigated using equalizers and channel estimation methods which are traditionally based on data driven tapped delay line (TDL) models. With increased fading due to mobility of transmit receive terminals, data driven schemes prove to be inefficient. Further, the TDL model assumes the dispersive channel to be linear in nature unlike the practical case which is non-linear, resulting in degraded performance. In such a situation, extrapolation of the channel state information is required which has been done by using artificial neural networks (ANN) optimized for non-linear channels. We explore a few neuro-computational approaches namely focused time delay neural network, feed forward ANN (FANN)–decision feedback equalizer (DFE) (FANN–DFE) and a functional link neural network while modeling non-linear wireless channels. Further, implementation of FANN–DFE based algorithm has been carried out using a specialized hardware framework to show the reduction of computational latency during training to establish its suitability for real time applications.

Mean square synchronization of neural networks with Lévy noise via sampled-data and actuator saturating controller

The problem of synchronization via sampled-data and saturating controller is considered for stochastic time-delay neural networks with Lévy noise and Markovian switching parameters in this paper. By using of the generalized Itô׳s formula and the Lyapunov functional method, an LMI-based sufficient condition is established to ensure the mean square synchronization of the master system and the slave system. Meanwhile, the gain of the sample data and saturating controller is determined. The sufficient condition depends on not only the switching mode and time-delay, but also the upper and the lower bound of sampling intervals. A numerical example is provided to verify the usefulness of the criterion proposed in this paper.

A dynamic EMG-torque model of elbow based on neural networks.

In this paper, a dynamic EMG-torque model of the elbow joint is developed based on ANN, and two novel test methods are proposed to validate its generalization performance. A time-delay neural network (TDNN) model is built and proved to have less risk of overfitting than the most-used multilayer feedfoward neural network (MFNN) model for dynamic EMG-torque modeling. Both EMG and kinematic features are included in the input of ANN, but the zero-EMG test shows that the trained ANN is part of the inverse joint dynamics rather than the EMG-torque model, and some random samples for ANN training are added to overcome this problem. The single-muscle test shows that an inappropriate choice of the motion type may cause the model to estimate wrong torque directions. After tuning and testing, the root mean square error (RMSE) across all subjects is 0.60±0.20 N.m.

Speech frame recognition based on less shift sensitive wavelet filter banks

The wavelet transform possesses multi-resolution property and high localization performance; hence, it can be optimized for speech recognition. In our previous work, we show that redundant wavelet filter bank parameters work better in speech recognition task, because they are much less shift sensitive than those of critically sampled discrete wavelet transform (DWT). In this paper, three types of wavelet representations are introduced, including features based on dual-tree complex wavelet transform (DT-CWT), perceptual dual-tree complex wavelet transform, and four-channel double-density discrete wavelet transform (FCDDDWT). Then, appropriate filter values for DT-CWT and FCDDDWT are proposed. The performances of the proposed wavelet representations are compared in a phoneme recognition task using special form of the time-delay neural networks. Performance evaluations confirm that dual-tree complex wavelet filter banks outperform conventional DWT in speech recognition systems. The proposed perceptual dual-tree complex wavelet filter bank results in up to approximately 9.82 % recognition rate increase, compared to the critically sampled two-channel wavelet filter bank.

Mixed H-Infinity and Passive Filtering for Discrete Fuzzy Neural Networks With Stochastic Jumps and Time Delays.

In this brief, the problems of the mixed H-infinity and passivity performance analysis and design are investigated for discrete time-delay neural networks with Markovian jump parameters represented by Takagi-Sugeno fuzzy model. The main purpose of this brief is to design a filter to guarantee that the augmented Markovian jump fuzzy neural networks are stable in mean-square sense and satisfy a prescribed passivity performance index by employing the Lyapunov method and the stochastic analysis technique. Applying the matrix decomposition techniques, sufficient conditions are provided for the solvability of the problems, which can be formulated in terms of linear matrix inequalities. A numerical example is also presented to illustrate the effectiveness of the proposed techniques.

Neural network fault diagnosis of a trolling motor based on feature reduction techniques for an unmanned surface vehicle

This article presents a novel approach to the diagnosis of unbalanced faults in a trolling motor under stationary operating conditions. The trolling motor being typically of that used as the propulsion system for an unmanned surface vehicle, the diagnosis approach is based on the use of discrete wavelet transforms as a feature extraction tool and a time-delayed neural network for fault classification. The time-delayed neural network classifies between healthy and faulty conditions of the trolling motor by analysing the stator current and vibration. To overcome feature redundancy, which affects diagnosis accuracy, several feature reduction methods have been tested, and the orthogonal fuzzy neighbourhood discriminant analysis approach is found to be the most effective method. Four faulty conditions were analysed under laboratory conditions, where one of the blades causing damage to the trolling motor is cut into 10%, 25%, half and then into full to simulate the effects of propeller blades being damaged partly or fully. The results obtained from the real-time simulation demonstrate the effectiveness and reliability of the proposed methodology in classifying the different faults faster and accurately.

Time-Delay Neural Network for Continuous Emotional Dimension Prediction From Facial Expression Sequences.

Automatic continuous affective state prediction from naturalistic facial expression is a very challenging research topic but very important in human-computer interaction. One of the main challenges is modeling the dynamics that characterize naturalistic expressions. In this paper, a novel two-stage automatic system is proposed to continuously predict affective dimension values from facial expression videos. In the first stage, traditional regression methods are used to classify each individual video frame, while in the second stage, a time-delay neural network (TDNN) is proposed to model the temporal relationships between consecutive predictions. The two-stage approach separates the emotional state dynamics modeling from an individual emotional state prediction step based on input features. In doing so, the temporal information used by the TDNN is not biased by the high variability between features of consecutive frames and allows the network to more easily exploit the slow changing dynamics between emotional states. The system was fully tested and evaluated on three different facial expression video datasets. Our experimental results demonstrate that the use of a two-stage approach combined with the TDNN to take into account previously classified frames significantly improves the overall performance of continuous emotional state estimation in naturalistic facial expressions. The proposed approach has won the affect recognition sub-challenge of the Third International Audio/Visual Emotion Recognition Challenge.

Forecasting the Term Structure of Crude Oil Futures Prices with Neural Networks

The paper contributes to the rare literature modeling term structure of crude oil markets. We explain term structure of crude oil prices using dynamic Nelson-Siegel model, and propose to forecast them with the generalized regression framework based on neural networks. The newly proposed framework is empirically tested on 24 years of crude oil futures prices covering several important recessions and crisis periods. We find 1-month, 3-month, 6-month and 12-month-ahead forecasts obtained from focused time-delay neural network to be significantly more accurate than forecasts from other benchmark models. The proposed forecasting strategy produces the lowest errors across all times to maturity.

Stabilization of Coupled Time-delay Neural Networks with Nodes of Different Dimensions

In reality, the neural networks (NNs) with different dimensions may be coupled as a complex network. This paper deals with the stabilization problem for an array of coupled NNs with time delays, in which nodes of the isolated NNs have different state dimensions from each other. Without assuming the boundedness or monotonicity of the activation functions and the symmetry or irreducibility of the coupling configuration matrices, several sufficient criteria are derived for ensuring the existence and uniqueness of the equilibrium point of such coupled NNs. The decentralized linear controllers are employed to stabilize the system. By constructing suitable Lyapunov functional, the criteria are formulated in terms of linear matrix inequalities and can be easily verified by using MATLAB LMI Toolbox. Two numerical examples are given to demonstrate the effectiveness of the proposed theoretical results.

A hierarchical neuro-fuzzy architecture for human behavior analysis

Analysis and detection of human behaviors from video sequences has became recently a very hot research topic in computer vision and artificial intelligence. Indeed, human behavior understanding plays a fundamental role in several innovative application domains such as smart video surveillance, ambient intelligence and content-based video information retrieval. However, the uncertainty and vagueness that typically characterize human daily activities make frameworks for human behavior analysis (HBA) hard to design and develop. In order to bridge this gap, this paper proposes a hierarchical architecture, based on a tracking algorithm, time-delay neural networks and fuzzy inference systems, aimed at improving the performance of current HBA systems in terms of scalability, robustness and effectiveness in behavior detection. Precisely, the joint use of the aforementioned methodologies enables both a quantitative and qualitative behavioral analysis that efficiently face the intrinsic people/objects tracking imprecision and provide context aware and semantic capabilities for better identifying a given activity. The validity and effectiveness of the proposed framework have been verified by using the well-known CAVIAR dataset and comparing our system’s performance with other similar approaches working on the same dataset.

Recent Developments on Time-Delay Neural Networks

Over the past several decades, an amount of attention has been devoted to neural networks, which have wide applications in many areas, such as signal processing, pattern recognition, and combinatorial optimization. On the other hand, time delays occur in electronic implementation of analog neural networks because of the transmission of signal and the finite switching speed of amplifiers and can lead to the instability and poor performance of systems; thus neural networks with time delay have also been studied extensively because numerous interesting results have been obtained over the last decade, including stability analysis, state estimation, and passivity. In this special issue, we have accepted seven papers. The paper titled “New Pinning Synchronization of Complex Networks with Time-Varying Coupling Strength, NonDelayed and Delayed Coupling” by G. Wang et al. studies the pinning synchronization problem for a class of complex networks by a stochastic viewpoint, in which both time-varying coupling strength and nondelayed and delayed coupling are included. The paper titled “A Novel Model of Conforming Delaunay Triangulation for Sensor Network Configuration” by Y. Ma et al. proposes a new method for solvingDelaunayTriangulation problem,which is called endpoint triangle’s circumcircle model (ETCM).The paper titled “Multimodel Modeling and Predictive Control for DirectDrive Wind Turbine with Permanent Magnet Synchronous Generator” by L. Wang et al. establishes a method based on multimodel modeling and predictive control is proposed for the optimal operation of direct-drive wind turbine with permanent magnet synchronous generator. The paper titled “Independent Component Analysis Based on Information Bottleneck” by Q. Ke et al. provides the equivalence of two algorithms of independent component analysis (ICA) based on the information bottleneck (IB). In “Nonlinear Time-Delay Suspension Adaptive Neural Network Active Control” by Y. Zhu and S. Zhu, a quarter-vehicle magnetorheological active suspension nonlinear model with time delay is established, and an adaptive neural network structure for magnetorheological active suspension is presented. The paper titled “An Adaptive Regulator for Space Teleoperation System in Task Space” by C. Ge et al. studies the problem of the gravity information for bilateral teleoperation. The paper titled “Robust Stabilization of Linear Switching Systems with both Input and Communication Delays” by L. Jia and Z. Sheng considers the stabilization control of a class of linear switching systems with time delays in both the input and the communication channels.

Distance Prediction for Commercial Serial Crime Cases Using Time Delay Neural Networks

The prediction of the next serial criminal time is important in the field of criminology for preventing the recurring actions of serial criminals. In the associated dynamic systems, one of the main sources of instability and poor performances is the time delay, which is commonly predicted based on nonlinear methods. The aim of this study is to introduce a dynamic neural network model by using nonlinear autoregressive time series with exogenous (external) input (NARX) and Back Propagation Through Time (BPTT), which is verified intensively with MATLAB to predict and model the crime times for the next distance of serial cases. Recurrent neural networks have been extensively used for modeling of nonlinear dynamic systems. There are different types of recurrent neural networks such as Time Delay Neural Networks (TDNN), layer recurrent networks, NARX, and BPTT. The NARX model for the two cases of input- output modeling of dynamic systems and time series prediction draw more attention. In this study, a comparison of two models of NARX and BPTT used for the prediction of the next serial criminal time illustrates that the NARX model exhibits better performance for the prediction of serial cases than the BPTT model. Our future work aims to improve the NARX model by combining objective functions.

An empirical analysis of a neural network model for the time series forecasting of different industrial segments

This paper aims to propose a neural network model for forecasting the production time series of 11 different industries in Brazil. The data was collected from Brazilian Institute of Geography and Statistics (IBGE). Firstly, we study different networks topologies that have been implemented in the literature in recent years, such as perceptron, linear networks, multi-layer perceptron (MLP), probabilistic network, Hopfield model, Kohonen model, time delay neural network (TDNN), Elman and Jordan network, in addition to the backpropagation and Levenberg-Marquadt algorithms. Studying the behaviour of these time series and the main characteristics of the each network topology, we conclude that the TDNN with multi-layer perceptron is the best to estimate the production time series of 11 industrial segments. The neural network was then applied considering two different strategies of structural model. We conclude that the neural network model proposed was effective for forecasting production time series in these industries.

Discrimination algorithm using voiced detection method and time-delay neural network system by 3 FFT sub-bands

In the field of speech processing, the main application of a neural network is the category discrimination of phoneme recognition and speech/speaker recognition. In a speech signal, information on the time variation is significant when training the neural network for the speech signal input. Therefore, this paper proposes a speech discrimination algorithm in noisy speech signals using a voiced detection method and a time-delay neural network with a time structure. Thus, this algorithm first detects voiced sections using the proposed neural network at each frame in the condition of background noises, then discriminates the speech signals using the time-delay neural network based on three fast Fourier transform sub-bands, in the noisy environments. The effectiveness of the proposed algorithm is experimentally confirmed based on measuring the correct discrimination rates for speech degraded by various noises.

LMI-Based Stability Criterion for Impulsive Delays Markovian Jumping Time-Delays Reaction-Diffusion BAM Neural Networks via Gronwall-Bellman-Type Impulsive Integral Inequality

Lyapunov stability theory, variational methods, Gronwall-Bellman-type inequalities theorem, and linear matrices inequality (LMI) technique are synthetically employed to obtain the LMI-based global stochastic exponential stability criterion for a class of time-delays Laplace diffusion stochastic equations with large impulsive range under Dirichlet boundary value, whose backgrounds of physics and engineering are the impulsive Markovian jumping time-delays reaction-diffusion BAM neural networks. As far as the authors know, it is the first time to derive the LMI-based criterion by way of Gronwall-Bellman-type inequalities, which can be easily and efficiently computed by computer Matlab LMI toolbox. And the obtained criterion improves the allowable upper bounds of impulse against those of some previous related literature. Moreover, a numerical example is presented to illustrate the effectiveness of the proposed methods.

Stochastic Synchronization of Neutral-Type Neural Networks with Multidelays Based onM-Matrix

The problem of stochastic synchronization of neutral-type neural networks with multidelays based onM-matrix is researched. Firstly, we designed a control law of stochastic synchronization of the neural-type and multiple time-delays neural network. Secondly, by making use of Lyapunov functional andM-matrix method, we obtained a criterion under which the drive and response neutral-type multiple time-delays neural networks with stochastic disturbance and Markovian switching are stochastic synchronization. The synchronization condition is expressed as linear matrix inequality which can be easily solved by MATLAB. Finally, we introduced a numerical example to illustrate the effectiveness of the method and result obtained in this paper.

A real-time topography of maximum contact pressure distribution at medial tibiofemoral knee implant during gait: Application to knee rehabilitation

Knee contact pressure is a crucial factor in the knee rehabilitation programs. Although contact pressure can be estimated using finite element analysis, this approach is generally time-consuming and does not satisfy the real-time requirements of a clinical set-up. Therefore, a real-time surrogate method to estimate the contact pressure would be advantageous.This study implemented a novel computational framework using wavelet time delay neural network (WTDNN) to provide a real-time estimation of contact pressure at the medial tibiofemoral interface of a knee implant. For a number of experimental gait trials, joint kinematics/kinetics and the resultant contact pressure were computed through multi-body dynamic and explicit finite element analyses to establish a training database for the proposed WTDNN. The trained network was then tested by predicting the maximum contact pressure at the medial tibiofemoral knee implant for two different knee rehabilitation patterns; “medial thrust” and “trunk sway”. WTDNN predictions were compared against the calculations from an explicit finite element analysis (gold standard).Results showed that the proposed WTDNN could accurately calculate the maximum contact pressure at the medial tibiofemoral knee implant for medial thrust (RMSE¯=1.7MPa, NRMSE¯=6.2% and ρ¯=0.98) and trunk sway (RMSE¯=2.6MPa, NRMSE¯=9.3%, ρ¯=0.96) much faster than the finite element method. The proposed methodology could therefore serve as a cost-effective surrogate model to provide real-time evaluation of the gait retraining programs in terms of the resultant maximum contact pressures.