Autoassociative Neural Network Research Articles

Machine Learning for Automatic Processing of Modal Analysis in Damage Detection of Bridges

Autonomous structural health monitoring (SHM) of a large number of bridges became a topic of paramount importance for maintenance purposes and safety reasons. This article proposes a set of machine learning (ML) tools to perform automatic detection of anomalies in a bridge structure from vibrational data. As a case study, we considered the Z-24 bridge for which an extensive database of accelerometric data is available. The proposed framework starts from the stabilization diagram obtained through operational modal analysis (OMA) to perform the clustering of modal frequencies and their tracking by density-based time-domain filtering. The features extracted are then fed to a one-class classification (OCC) algorithm to perform anomaly detection. In particular, we propose two new anomaly detectors, namely, one-class classifier neural network (OCCNN) and OCCNN2, that find the normal class (the boundary of the features space in normal operating conditions) through a two-step approach: coarse and fine boundary estimate. The detection algorithms are then compared with known methods based on the principal component analysis (PCA), the kernel PCA (KPCA), the Gaussian mixture model (GMM), and the autoassociative neural network (ANN). The proposed OCCNN solution presents increased accuracy and F1 score over conventional algorithms, without the need to set critical parameters, while OCCNN2 provides the best performance in terms of F1 score, accuracy, and responsiveness.

A Fuzzy Cognitive Map method for integrated and participatory water governance and indicators affecting drinking water supplies

Drinking water governance is challenging with different perceptions and priorities among stakeholders in different countries. To make provision for drinking water protection in agricultural areas, governance systems need to be mapped for bottlenecks to be identified and solutions highlighted. To address this a system thinking approach was used in an explanatory network analysis of Fuzzy Cognitive Maps (FCM) that were created during face to face interviews with stakeholder representative groups (individuals, policy developers, researchers, and regulators). Two exercises were designed and facilitated to obtain stakeholder maps on A) the water governance framework from stakeholders' own perspective with a ranking of actors in terms of their perceived importance and B) a list of importance factors and how these were connected for the provision of good drinking water quality supplies in agricultural areas. Causal relationships were subsequently drawn around each subject allowing mapping. A graph theory Hierarchy Index (h) approach examined if stakeholder groups preferred top down hierarchical governance or a more inclusive democratic governance approach. Finally, an auto-associative neural network method was deployed on group maps for examination during steady-state conditions for three scenarios to be explored i.e. changing “Farmers knowledge”, “best management practice (BMP) uptake” and “Farmers behaviour and belief” to the highest level of influence and seeing how the system reacted. Results of Exercise A showed that all stakeholder representative groups had a different perception of the water governance framework. Most stakeholder groups had a democratic point of view regarding water governance structures and the ranking and importance of the actors within the framework. Results of Exercise B demonstrated that most of the groups have similar opinions regarding the highest ranked factors affecting drinking water quality and the possible environmental ecological policy options. In this second exercise, only one representative group showed a democratic outlook whereas all others had a hierarchal outlook. Scenario testing of policy options enabled bottlenecks and possible solutions to be identified. By boosting “Farmers behaviour and belief” to the highest possible level, resulted in a large increase in other factors – a scenario where farmers could benefit from the outcome. This would be achieved by enhancing farmers' willingness and intention to participate and implement BMPs. Better results would be achieved if farmers believed in the method and could benefit from the outcome. Also keeping “Farmers knowledge” at the highest point had a positive influence on the other factors. This can be achieved by enhancing farmers training and knowledge transfer by local and national actors. This method is widely applicable and should be considered for more integrated and participatory approaches to drinking water governance.

Early damage detection under massive data via innovative hybrid methods: application to a large-scale cable-stayed bridge

Application of massive data to structural health monitoring (SHM) may lead to serious problems such as difficulty, computational inefficiency, and low damage detectability. This paper proposes innovative hybrid methods in order to detect damage under massive data, ambient vibration, and environmental and/or operational variability. Each of the proposed methods consists of a three-stage algorithm including response modelling via a time series representation, Gaussian mixture model (GMM) for dimensionality reduction, and an outlier detector. Due to the importance of response modelling under ambient vibration, this article employs the combination of AR with ARX called the ARARX model. In the second stage, a GMM is individually fitted to residual samples extracted from ARARX regarding the undamaged and damaged conditions to obtain low-dimensional features from massive data. Eventually, outlier analyses are carried out by the Mahalanobis distance and an auto-associative neural network to make a decision about the occurrence of damage. Two different kinds of threshold limits are also considered to examine the performances of the proposed methods. A large-scale cable-stayed bridge is applied to validate the reliability and effectiveness of these methods with several comparative studies. Results demonstrate that both proposed methods are effective tools for damage detection.

The experimental study of the effectiveness of Kohonen maps and autoassociative neural networks in the qualitative analysis of multidimensional data by the example of real data describing coal susceptibility to fluidal gasification

The qualitative analysis of multidimensional data using their visualization allows to observe some characteristics of data in a way which is the most natural for a human, through the sense of sight. Thanks to such an approach, some characteristics of the analyzed data are simply visible. This allows to avoid using often complex algorithms allowing to examine specific data properties. Visualization of multidimensional data consists in using the representation transforming a multidimensional space into a two-dimensional space representing a computer screen. The important information which can be obtained in this way is the possibility to separate points belonging to different classes in the multidimensional space. Such information can be directly obtained if images of points belonging to different classes occupy other areas of the picture presenting these data. The paper presents the effectiveness of the qualitative analysis of multidimensional data conducted in this way through their visualization with the application of Kohonen maps and autoassociative neural networks. The obtained results were compared with results obtained using the perspective-based observational tunnels method, PCA, multidimensional scaling and relevance maps. Effectiveness tests of the above methods were performed using real seven-dimensional data describing coal samples in terms of their susceptibility to fluidal gasification. The methods’ effectiveness was compared using the criterion for the readability of the multidimensional visualization results, introduced in earlier papers.

The examination of the effect of the criterion for neural network\u2019s learning on the effectiveness of the qualitative analysis of multidimensional data

A variety of multidimensional visualization methods are applied for the qualitative analysis of multidimensional data. One of the multidimensional data visualization methods is a method using autoassociative neural networks. In order to perform visualizations of n-dimensional data, such a network has n inputs, n outputs and one of the interlayers consisting of two outputs whose values represent coordinates of the analyzed sample’s image on the screen. Such a criterion for the network’s learning consists in that the same value as the one at the ith input appears at each ith output. If the network is trained in this way, the whole information from n inputs was compressed to two outputs of the interlayer and then decompressed to n network outputs. The paper shows the application of different learning criteria can be more beneficial from the point of view of the results’ readability. Overall analysis was conducted on seven-dimensional real data representing three coal classes, five-dimensional data representing printed characters, 216-dimensional data representing hand-written digits and, additionally, in order to illustrate additional explanations using artificially generated seven-dimensional data. Readability of results of the qualitative analysis of these data was compared using the multidimensional visualization utilizing neural networks for different learning criteria. Also, the obtained results of applying all analyzed criteria on 20 randomly selected sets of multidimensional data obtained from one of the publicly available repositories are presented.

Відмовостійкий алгоритм ідентифікації бортової математичної моделі авіаційного двигуна ТВ3- 117 в складі його системи автоматичного управління

The subject matter of the article is TV3-117 aircraft engine and methods for monitoring and diagnosing its technical condition. The goal of the work is development of fault-tolerant algorithms for identification of the onboard mathematical model of the TV3-117 aircraft engine as part of its automatic control system in flight modes. The following tasks were solved in the article: recovering of lost information by an auto-associative neural network in case of a single sensor failure, recovering of lost information by an «optimal» auto-associative neural network in case of single sensor failures of the on-board control and diagnostic system, recovering of lost information by an auto-associative neural network and an on-board control and diagnostic system from the gas temperature registration sensor in front of the turbine compressor in case of failure. The following methods used are – substantiation of expediency of Kalman-filtration use in TV3-117 aircraft engine automatic control system, determination of Kalman filter transfer function, determination of algorithm of detection and localization of channel failure of two-channel sensor, determination of frequency characteristics of TV3-117 aircraft engine automatic control system, proof of equality of unit coefficient fault-tolerant filter unit . Conclusions: The frequency properties of the automatic control system of the aircraft engine TV3-117 are investigated, the equality of the unit of the gain of the developed fault-tolerant filtering unit and the absence of additional phase shifts introduced by the possible net delay due to the peculiarities of the implemented algorithms mathematical model. The absence of their influence on the stability of aircraft engine TV3-117 automatic control system is proved. Approbation of the developed algorithms showed that the average relative error in dynamics does not exceed 0.15 %, and in statics at the maximum expense – decreases to 0.01 % that corresponds to modern requirements of accuracy of algorithms of identification on a contour of dosing needle . Keywords: aircraft engine, neural network, Kalman filtering, failure detection and localization

An intelligent data filtering and fault detection method for gas turbine engines

In a gas turbine fault diagnostics, the removal of measurement noise and data outliers prior to the fault analysis is very essential. The conventional filtering methods, particularly the linear ones, are not sufficiently accurate, which might possibly lead to the loss of critically important features in the fault analysis process. Conversely, the recorded accuracies obtained from the non-linear filters are promising. Recently, the focus has been shifted to the artificial neural network (ANN) based nonlinear filters due to their capability of providing a robust identity map between the input and output data, which can be efficiently exploited in the process of fault diagnosis. This paper aims to present combined auto-associative neural network (AANN) and K-nearest neighbor (KNN) based noise reduction and fault detection method for a gas turbine engine application. The performance of the developed method has been evaluated using data obtained from a model simulation. The test results revealed that the developed hybrid method is more effective and reliable than the conventional methods for the fault detection of the gas turbine engine with negligible false alarms and missed detections.

SAR Data Fusion Using Nonlinear Principal Component Analysis

Synthetic Aperture Radar (SAR) images taken over a certain area at different bands and also with a short time interval are now more widely available. This is due to the increase of SAR acquisitions following the last space missions, such as Sentinel 1 and COSMO-SkyMed (CSK). New paradigms capable of performing effective analysis and synthesis stemming from such a type of information are then required in order to exploit better and disseminate the information contained in the data. In this letter, a data fusion technique between CSK and Sentinel-1 data is described. To this purpose, an ad hoc Nonlinear Principal Component Analysis (NLPCA) with Auto-Associative Neural Networks (AANNs) algorithm is designed and developed. The network extracts the most relevant features from the combination of the different scattering mechanisms. The extracted features are then used as inputs for a land cover classification exercise. A comparison between the results obtained with the original images and those yielded by the new synthesized data, with lower dimensionality, demonstrates the ability of the algorithm to generate useful final products.

Keyword Detection using Auto Associative Neural Network with Reference to Assamese Language

Speech and speaker recognition has yet not achieved the state-of-the-art position. Keyword detection in audio clips is gaining importance as it contributes to the audio recognition and detection systems. In this area, very few works have been carried out. In this paper, we present our experiment on keyword detection within recorded news clips. It is based on Assamese language spoken by Assamese native speakers. For this experiment, the audio clips are collected from local TV news debates, whereas the keywords are recorded by random speakers. The keywords are selected for recording considering the fact that they appear somewhere within the audio clips for a finite number of times. Mel Frequency Cepstral Coefficient (MFCC) is considered as feature and Auto Associative Neural Network (AANN) is considered as the classifier tool. With this detection model an average accuracy of 87% is achieved.

Sensor data validation and fault diagnosis using Auto-Associative Neural Network for HVAC systems

The Heating, Ventilation, and Air conditioning (HVAC) system is a major system in buildings for conditioning the indoor environment. Sensor data validation and fault diagnosis for HVAC systems are essentially important to secure a reliable and efficient operation since sensor measurements are vital for the HVAC closed-loop control system. The aim of this work is to address this matter by developing a data-driven approach using the system's normal operation data and without the need for the knowledge of the mathematical model of the system. It is based on an Auto-Associative Neural Network (AANN) that is structured and trained to construct an input-output mapping model based on data dimensionality reduction that is capable of validating sensor measurements in terms of sensor error correction, missing data replacement, noise filtering, and inaccuracy correction. It can be used for both single and multiple sensor faults diagnosis by monitoring the consistency between the actual and the AANN-estimated sensor reading. The validation of the proposed method is demonstrated on data obtained from a 3-zone HVAC system simulated in TRNSYS. The evaluation results show the effectiveness of the proposed approach and an improvement in terms of data validation and diagnostic accuracy when compared with a PCA-based method.

Statistical Process Monitoring of the Tennessee Eastman Process Using Parallel Autoassociative Neural Networks and a Large Dataset

In this article, the statistical process monitoring problem of the Tennessee Eastman process is considered using deep learning techniques. This work is motivated by three limitations of the existing works for such problem. First, although deep learning has been used for process monitoring extensively, in the majority of the existing works, the neural networks were trained in a supervised manner assuming that the normal/fault labels were available. However, this is not always the case in real applications. Thus, in this work, autoassociative neural networks are used, which are trained in an unsupervised fashion. Another limitation is that the typical dataset used for the monitoring of the Tennessee Eastman process is comprised of just a small number of data samples, which can be highly limiting for deep learning. The dataset used in this work is 500-times larger than the typically-used dataset and is large enough for deep learning. Lastly, an alternative neural network architecture, which is called parallel autoassociative neural networks, is proposed to decouple the training of different principal components. The proposed architecture is expected to address the co-adaptation issue of the fully-connected autoassociative neural networks. An extensive case study is designed and performed to evaluate the effects of the following neural network settings: neural network size, type of regularization, training objective function, and training epoch. The results are compared with those obtained using linear principal component analysis, and the advantages and limitations of the parallel autoassociative neural networks are illustrated.

Neural Networks Principal Component Analysis for estimating the generative multifactor model of returns under a statistical approach to the Arbitrage Pricing Theory. Evidence from the Mexican Stock Exchange

A nonlinear principal component analysis (NLPCA) represents an extension of the standard principal component analysis (PCA) that overcomes the limitation of the PCA’s assumption about the linearity of the model. The NLPCA belongs to the family of nonlinear versions of dimension reduction or the extraction techniques of underlying features, including nonlinear factor analysis and nonlinear independent component analysis, where the principal components are generalized from straight lines to curves. The NLPCA can be achieved via an artificial neural network specification where the PCA classic model is generalized to a nonlinear mode, namely, Neural Networks Principal Component Analysis (NNPCA). In order to extract a set of nonlinear underlying systematic risk factors, we estimate the generative multifactor model of returns in a statistical version of the Arbitrage Pricing Theory (APT), in the context of the Mexican Stock Exchange. We used an auto-associative multilayer perceptron neural network or autoencoder, where the ‘bottleneck’ layer represented the nonlinear principal components, or in our context, the scores of the underlying factors of systematic risk. This neural network represents a powerful technique capable of performing a nonlinear transformation of the observed variables into the nonlinear principal components, and to execute a nonlinear mapping that reproduces the original variables. We propose a network architecture capable of generating a loading matrix that enables us to make a first approach to the interpretation of the extracted latent risk factors. In addition, we used a two stage methodology for the econometric contrast of the APT involving first, a simultaneous estimation of the system of equations via Seemingly Unrelated Regression (SUR), and secondly, a cross-section estimation via Ordinary Least Squared corrected by heteroskedasticity and autocorrelation by means of the Newey-West heteroskedasticity and autocorrelation consistent covariances estimates (HEC). The evidence found shows that the reproductions of the observed returns using the estimated components via NNPCA are suitable in almost all cases; nevertheless, the results in an econometric contrast lead us to a partial acceptance of the APT in the samples and periods studied.

Parallel neural networks for improved nonlinear principal component analysis

In this paper, a parallel neural network architecture is proposed to improve the performance of neural-network-based nonlinear principal component analysis. There exist two typical approaches for such analysis: simultaneous extraction of principal components using a single autoassociative neural network (also known as autoencoder), and sequential extraction using multiple neural networks in series. The proposed architecture can be obtained by systematically pruning the network connections of a fully connected autoassociative neural network, resulting in decoupled neural networks. As a result, more independent (i.e., less correlated) principal components can be obtained than the simultaneous extraction approach. The proposed architecture can be also viewed as a rearrangement of multiple neural networks for the sequential extraction in a parallel setting, and thus, the network training becomes more efficient. Simulation case studies are performed to illustrate the advantages of the proposed architecture, and it was shown that it is particularly beneficial for deep neural networks.

Stress wave evaluation for predicting the properties of thermally modified wood using neuro-fuzzy and neural network modeling

AbstractThis study investigated using the stress wave method to predict the properties of thermally modified wood by means of an adaptive neuro-fuzzy inference system (ANFIS) and neural network (NN) modeling. The stress wave was detected using a pair of accelerometers and an acoustic emission (AE) sensor, and the effect of heat treatment (HT) on the physical and mechanical properties of wood as well as wave velocity and AE signal is discussed. The AE signal was processed in the time and time-frequency domains using wavelet analysis and different features were extracted for network training. The auto-associative NN is used as a dimensional reduction method to decrease the dimension of the extracted AE features and enhance the ANFIS performance. It was shown that while the stress wave velocity using the accelerometer did not result in an accurate model, the network performance significantly increased when trained with the AE features. The AE signal exhibited a significant correlation with wood treatment and porosity. The best ANFIS performance corresponded to predicting the wood swelling coefficient, equilibrium moisture content (EMC) and water absorption (WA), respectively. However, the AE signal did not seem suitable for predicting the wood density and hardness. The performance of ANFIS was compared with the “group method of data handling” (GMDH) NN. Both the ANFIS and GMDH networks showed higher accuracy than the multivariate linear regression (MVLR) model.

Competitive Autoassociative Neural Networks for Electrical Appliance Identification for Non-Intrusive Load Monitoring

This paper presents a new approach to electrical appliance identification for non-intrusive load monitoring (NILM). In the proposed method a set of autoassociative neural networks is trained so that each one is tuned with the characteristics of a particular electrical appliance. Then, the autoassociative neural networks are set up in a competitive parallel arrangement in which they compete with one another when a new input vector is entered and the closest recognition is accepted to identify the given electrical appliance. The system is trained to recognize specific types of electrical appliances and use the transient power signal obtained from the on/off events for each electrical appliance. To test the proposed method, three public datasets were used, they are, the reference energy disaggregation dataset (REDD), the United Kingdom recording domestic appliance-level electricity (UK-DALE) and the Tracebase dataset containing real residential measurements are used. The accuracy and F-score obtained for the three datasets show the applicability of the proposed method for NILM systems.

Enhanced auto associative neural network using feed forward neural network: an approach to improve performance of fault detection and analysis

Biosensors have played a significant role in many of present day's applications ranging from military applications to healthcare sectors. However, its practicality and robustness in its usage in real time scenario is still a matter of concern. Primarily issues such as prediction of sensor data, noise estimation and channel estimation and most importantly in fault detection and analysis. In this paper an enhancement is applied to the auto associative neural network (AANN) by considering the cascade feed forward propagation. The residual noise is also computed along with fault detection and analysis of the sensor data. An experimental result shows a significant reduction in the MSE as compared to conventional AANN. The regression based correlation coefficient has improved in the proposed method as compared to conventional AANN.

Effective analysis of malware detection in cloud computing

Cloud services are relied upon to be always on and have a significant nature; acordingly, security and versatility are progressively imperative perspectives. Accessing the cloud environment is performed via internet services in which information stored on cloud environment are easier to both internal and external Malwares. To detect intruders, a variety of malware detection systems were focused on machine learning techniques, which are not delivering major results based on detection rate. To overcome this drawback, we propose novel consolidated Weighted Fuzzy K-means clustering algorithm with Auto Associative Neural Network(WFCM-AANN). Our proposed classifier successfully identifies the malwares and it is obvious that the proposed framework can identify the anomalies with high detection precision thereby outperforming existing classifiers.

Gas path fault diagnostics using a hybrid intelligent method for industrial gas turbine engines

There are many challenges against an accurate gas turbine fault diagnostics, such as the nonlinearity of the engine health, the measurement uncertainty, and the occurrence of simultaneous faults. The conventional methods have limitations in effectively handling these challenges. In this paper, a hybrid intelligent technique is devised by integrating an autoassociative neural network (AANN), nested machine learning (ML) classifiers, and a multilayer perceptron (MLP). The AANN module is used as a data preprocessor to reduce measurement noise and extract the important features for visualisation and fault diagnostics. The features are extracted from the bottleneck layer output values based on the concept of the nonlinear principal component analysis (NLPCA). The nested classifier modules are then used in such a manner that fault and no-fault conditions, component and sensor faults, and different component faults are distinguished hierarchically. As part of the classification, evaluation of the fault classification performance of five widely used ML techniques aiming to identify alternative approaches is undertaken. In the end, the MLP approximator is utilised to estimate the magnitude of the isolated component faults in terms of flow capacity and isentropic efficiency indices. The developed system was implemented to diagnose up to three simultaneous faults in a two-shaft industrial gas turbine engine. Its robustness towards the measurement uncertainty was also evaluated based on Gaussian noise corrupted data. The test results show the derivable benefits of integrating two or more methods in engine diagnostics on the basis of offsetting the weakness of the one with the strength of another.

Performance-based fault diagnosis of a gas turbine engine using an integrated support vector machine and artificial neural network method

An effective and reliable gas path diagnostic method that could be used to detect, isolate, and identify gas turbine degradations is crucial in a gas turbine condition-based maintenance. In this paper, we proposed a new combined technique of artificial neural network and support vector machine for a two-shaft industrial gas turbine engine gas path diagnostics. To this end, an autoassociative neural network is used as a preprocessor to minimize noise and generate necessary features, a nested support vector machine to classify gas path faults, and a multilayer perceptron to assess the magnitude of the faults. The necessary data to train and test the method are obtained from a performance model of the case engine under steady-state operating conditions. The test results indicate that the proposed method can diagnose both single- and multiple-component faults successfully and shows a clear advantage over some other methods in terms of multiple fault diagnosis. Moreover, 5-8 sets of measurements have been used to assess the prediction accuracy, and only a 2.3% difference was observed. This result indicates that the proposed method can be used for multiple fault diagnosis of gas turbines with limited measurements.

Performance of Speaker Verification Using CSM and TM

In this paper, we presented the performance of a speaker verification system based on features computed from the speech recorded using a Close Speaking Microphone(CSM) and Throat Microphone(TM) in clean and noisy environment. Noise is the one of the most complicated problem in speaker verification system. The background noises affect the performance of speaker verification using CSM. To overcome this issue, TM is used which has a transducer held at the throat resulting in a clean signal and unaffected by background noises. Acoustic features are computed by means of Relative Spectral Transform-Perceptual Linear Prediction (RASTA-PLP). Autoassociative neural network (AANN) technique is used to extract the features and in order to confirm the speakers from clean and noisy environment. A new method is presented in this paper, for verification of speakers in clean using combined CSM and TM. The verification performance of the proposed combined system is significantly better than the system using the CSM alone due to the complementary nature of CSM and TM. It is evident that an EER of about 1.0% for the combined devices (CSM+TM) by evaluating the FAR and FRR values and the overall verification of 99% is obtained in clean speech.