Subband coding for sigmoidal nonlinear operations

In previous work, hybrid wavelet packets were introduced as a generalization of wavelet packets in which the choice of quadrature mirror filter (QMF) is selected adaptively within the wavelet packet analysis. This was motivated by the observation that for certain classes of signals, the choice of appropriate QMF is not only signal dependent, but may be scale dependent as well. Best Basis selection was generalized to provide a means of optimizing the representation of a given signal. This method can be viewed as an adaptive partitioning of signal subspaces. In such a scheme it is important to determine a small number of QMFs which will provide some diversity in the partitioning of the subspaces. In the current work, canonical correlations and canonical angles are used to quantify the difference between subspaces spanned by two filter pairs. Preliminary results indicate the utility of this method as a data independent metric for filter comparison.

Reconfigurable logic gates in biological crossbar neural networks using STDP learning.

In this article, an efficient method is proposed to diagnose urban rail vehicle auxiliary inverter faults based on wavelet packet neural network and genetic algorithm. Firstly, the original signals are decomposed into different frequency subbands by wavelet packet. Secondly, the wavelet packet energy eigenvector is constructed. Finally, those wavelet packet energy eigenvectors are taken as fault samples to train neural network, In order to improve the function approximation accuracy and general capability of the neural network system, an efficient genetic algorithm approach is used to adjust the parameters of translation and weights functions. The experiment shows that the GA-ANN model gives superior result. This approach can be used as a useful tool for the auxiliary inverter fault diagnosis. DOI : http://dx.doi.org/10.11591/telkomnika.v12i1.3847

The paper presents a modification of the structure of a biological neural network (BNN) based on spiking neuron models. The proposed modification allows to influence the level of the stimulus response of particular neurons in the BNN. We consider an extended, three-dimensional Hodgkin-Huxley model of the neural cell. A typical BNN composed of such neural cells have been expanded by addition of resistors in each branch point. The resistors can be treated as the weights in such BNN. We demonstrate that adding these elements to the BNN significantly affects the waveform of the potential on the membrane of the neuron, causing an uncontrolled excitation. This provides a better description of processes that take place in nervous cell. Such BNN enables an easy adaptation of the learning rules used in artificial or spiking neural networks. The modified BNN has been implemented on Graphics Processing Unit (GPU) in the CUDA C language. This platform enables a parallel data processing, which is an important feature in such applications.

This study presents a fault diagnosis system for vehicle heating, ventilation and air conditioning (HVAC) acoustic signal with various feature extractions in deep learning neural network. Traditionally, sound used for fault diagnosis or signal classification is observed the difference of energy in time or frequency domains. Unfortunately, the frequency smearing effect often arises in some critical conditions. In the present study, discrete wavelet transform (DWT) and wavelet packet transform (WPT) are proposed in fault diagnosis. Meanwhile, when using mechanical learning methods, the data are relatively large, in order to reduce the amount of data, DWT and WPT low-frequency decomposition could be used to improve the performance. Furthermore, the signal characteristics more comprehensive, this study attempts to use the feature extraction method of wavelet packet conversion to improve the signal characteristics. In the experiment process, the operation state of the blade blower in the vehicle air conditioner, four different faults were designed, test database was established through sound to classify, and identify the data using deep neural networks to achieve the purpose of blower fault diagnosis. In data analysis, the original signal is presented through wavelet packet decomposition and discrete packet conversion technology, compared with traditional time and frequency domain signals to explore the identification rate, identification speed and related issues. Experimental results show that using WPT combined with deep neural networks have good fault diagnosis and discrimination capabilities, training, and identification time is shorter than time-frequency domain signals.

In the actual project, the structural response often shows the effect of multiple excitation sources because of the random characteristics of the operating environment. So it is difficult for the traditional single source excitation method to meet its actual response characteristics. Therefore, the research on the characteristics of multi-source excitation signal is one of the development trends of mechanical structure detection technology. In this paper, combined with the principle of multi-source excitation and wavelet transform, wavelet packet is used to analyze the multi-source signal. The accuracy of the detection signal can be greatly improved through wavelet packet to reduce the noise of the signal. At the same time, through the comparison between the original signal and the fault signal analysis, we find the superiority of wavelet packet analysis in the structural fault detection.

One of the central goals in systems neuroscience is to understand how information is encoded in the brain, and the standard approach is to identify the relation between a stimulus and a neural response. However, the feature of a stimulus is typically defined by the researcher's hypothesis, which may cause biases in the research conclusion. To demonstrate potential biases, we simulate four likely scenarios using deep neural networks trained on the image classification dataset CIFAR-10 and demonstrate the possibility of selecting suboptimal/irrelevant features or overestimating the network feature representation/noise correlation. Additionally, we present studies investigating neural coding principles in biological neural networks to which our points can be applied. This study aims to not only highlight the importance of careful assumptions and interpretations regarding the neural response to stimulus features but also suggest that the comparative study between deep and biological neural networks from the perspective of machine learning can be an effective strategy for understanding the coding principles of the brain.

A roller bearing fault diagnosis method based on EMD energy entropy and ANN

Presented in this paper is a distortion-free filtering approach for cancellation of sinusoidal harmonic interference using wavelet packet decomposition (WPD). Under a special sampling condition, the transform coefficients of the harmonic interference in each subband become a constant, exhibiting as a shift of the baseline of the WPD coefficients in that subband. The harmonic interference is thus removed by shifting the baselines back to zero. Unlike the use of a notch or a comb filter, the new approach does not cause any distortion to the original signal in principle. This advantage makes this approach especially suitable for the removal of harmonic interference that is located within the spectrum of the original signal. Evaluation of the new approach has been conducted using a simulated piecewise smooth signal and a real, irregularly pulsating signal. Results obtained from the studies have demonstrated the effectiveness of the new approach.

The purpose of this study is to familiarise the reader with the foundations of neural networks. Artificial Neural Networks (ANNs) are algorithm-based systems that are modelled after Biological Neural Networks (BNNs). Neural networks are an effort to use the human brain's information processing skills to address challenging real-world AI issues. The evolution of neural networks and their significance are briefly explored. ANNs and BNNs are contrasted, and their qualities, benefits, and disadvantages are discussed. The drawbacks of the perceptron model and their improvement by the sigmoid neuron and ReLU neuron are briefly discussed. In addition, we give a bird's-eye view of the different Neural Network models. We study neural networks (NNs) and highlight the different learning approaches and algorithms used in Machine Learning and Deep Learning. We also discuss different types of NNs and their applications. A brief introduction to Neuro-Fuzzy and its applications with a comprehensive review of NN technological advances is provided.

On-line detection of chatter in the cutting process can identify the chatter in time or before the chatter happens, so the initiative to change the cutting parameters can be taken to avoid chatter and improve the surface quality. At present, time-frequency analysis technology is performed to extract the time-frequency features of chatter by scholars. With respect to the modal aliasing problem in the process of empirical mode decomposition (EMD), the chatter identification method of combining EMD and wavelet packets decomposition (WPD) is proposed to eliminate the influence of modal aliasing. To fully extract the main features of signal, the intrinsic mode functions (IMFs) changing consistent with power spectrum or amplitude-frequency are selected for signal reconstruction. Then, WPD is used in the reconstructed signal. The two times reconstruction of signal is based on wavelet packet node with the maximum energy. The distribution of frequency and energy in the time domain is presented by Hilbert Huang Transform (HHT) spectrum, and the mean value and standard deviation of the HHT spectrum are extracted as the feature vectors. The chatter features can be extracted from the original simulation signal by this method. Three groups of experiments with different cutting depth which are on behalf of the three cutting conditions (stable cutting, slight chatter, and severe chatter) were carried out. More cutting tests were carried out under the same cutting condition. Experimental results show that this method can be used to effectively identify the chatter features in milling process.

The purpose of this paper is to familiarize readers with the fundamental ideas of neural networks. The algorithmic models known as Artificial Neural Networks (ANNs) are modeled after Biological Neural Networks (BNNs). Neural networks aim to address challenging AI problems by utilizing the brain's information processing power. The paper provides a concise synopsis of the development and significance of neural networks, contrasting ANNs with BNNs and analyzing their traits, benefits, and drawbacks. Discussion is held regarding the shortcomings of the perceptron model as well as the enhancements brought forth by the sigmoid and ReLU neurons. The paper also looks at different learning strategies and algorithms in machine learning and deep learning, and it gives a summary of several neural network models. In addition to an overview of Neuro-Fuzzy systems, many NN kinds and their uses are examined.

Independent component analysis (ICA) is a computational method for separating independent-source signals from a mixed-signal series. The common ICA method cannot be directly used for separating when there is only one sensor collecting the signals. To solve the problem of separating single-channel mixed signals, a novel method based on ICA and the convolutional wavelet packet is proposed to divide a blind source. The method provides a new approach for blind source separation when the number of original signals is limited. First, the sub-band data are obtained by a non-downsampling convolutional wavelet packet. Because the convolutional wavelet packet is not down-sampled, the lengths of different sub-band sequences are the same as the original signal and correspond to a certain frequency band. Second, to decrease the influence of edge frequency aliasing, each sub-band of the wavelet packet is processing using frequency division. Using preprocessing, the authors are able to create a multiple-channel mixed-signal series that can be used to obtain the inputs of the ICA. The method proposed in this paper was applied to blind source separation (BBS) using a simulated sine signal series; real signal series collected from a motor vibration testing platform are also used in testing. The results show that the method produces results in blind source separation that are equivalent to using multiple mixed signals directly.

Finite-state machines are the most pervasive models of com- putation, not only in theoretical computer science, but also in all of its applications to real-life problems, and constitute the best characterized computational model. On the other hand, neural networks -proposed almost sixty years ago by McCulloch and Pitts as a simplified model of nervous activity in living beings- have evolved into a great variety of so-called artificial neural networks. Artificial neural networks have be- come a very successful tool for modelling and problem solving because of their built-in learning capability, but most of the progress in this field has occurred with models that are very removed from the behaviour of real, i.e., biological neural networks. This paper surveys the work that has established a connection between finite-state machines and (mainly discrete-time recurrent) neural networks, and suggests possible ways to construct finite-state models in biologically plausible neural networks.

Neural networks are non-algorithmic, analog, distributive and massively parallel information processing systems that have a number of performance characteristics in common with biological neural networks or the human brain. Neural networks can simulate the nervous systems of living animals which work differently from conventional computing, to analyse, compute and solve some complex practical problems making use of computers. Neural networks are able to discover highly complex relationships between several variables that are presented to the network. Studies show that neural networks can be used to solve a great number of practical problems which occur in modeling, predictions, assessments, recognition and image processing. In particular, neural networks are suitable for application to problems where some results are known but the manner in which these results can be achieved are not known (or are difficult to implement) or the results themselves are not known. An important challenge for geologists, geophysicists and reservoir engineers is to accurately determine petrophysical parameters and to improve reservoir evaluation and description. It is important to be able to obtain realistic values of petrophysical parameters from well logs because core data are often not available either because of bore hole conditions or due to the high cost of coring. In lithologically complex formations conventional petrophysical evaluation methods cannot be used because of the lithological heterogeneity. This paper presents an application of neural networking to estimate petrophysical parameters from well logs and to evaluate reservoir quality in the Mardie Greensand in the Carnarvon Basin in Western Australia.