Fault Immunization Concept for Self-Organizing Mapping Neural Networks

A fault detection and diagnosis method for the hydraulic servo system based on adaptive threshold and self-organizing map (SOM) neural network is proposed in this study. The nonlinear, time-varying, fluid-solid coupling properties of the hydraulic servo system are considered. Fault detection is realized based on a two-stage radial basis function (RBF) neural network model. The first-stage RBF neural network is adopted as a fault observer for the hydraulic servo system; the residual error signal is generated by comparing the estimated observer output with the actual measurements. To overcome the drawback of false alarms when the traditional fixed fault threshold is used, an adaptive threshold producer is established by the second-stage RBF neural network. Fault occurrence is detected by comparing the residual error signal with the adaptive threshold. When a system fault is detected, the SOM neural network is employed to implement fault classification and isolation by analyzing the features of the residual error signal. Three types of common faults are simulated to verify the performance and effectiveness of the proposed method. Experimental results demonstrate that the proposed method based on adaptive threshold and SOM neural network is effective in detecting and isolating the failure of the hydraulic servo system.

In this work, we propose a solution for the recognition of protein folds using Self-Organizing Map (SOM) neural network and present a comparison between few approaches. We use SOM, Fisher’s Linear Discriminant Analysis (FLD), K-Nearest Neighbor (KNN), Support Vector Machine (SVM) and Multi-Layer Perceptron (MLP) methods for the recognition of three SCOP folds with six attributes (amino acid composition, predicted secondary structure, hydrophobicity, normalized van der Waals volume, polarity and polarizability). Then we classify the most common 27 SCOP folds using 125 dimensional data formed by the six attributes. This paper has a novelty in the way of applying SOM to these six attributes, and also portrays the capabilities of SOM among the other methods in protein fold classification. Firstly for the threeclass problem, the methods are tested on 120 proteins by applying 10-fold cross-validation technique and 93.33% classification performance is obtained with SOM. Secondly for the 27-class problem SOM is tested on 694 proteins by applying one-versus-others technique and 73.37% classification performance is obtained. Keywords: Protein fold recognition, protein fold classification, neural networks, self-organizing map.

The main goal of this paper is to show that the fractal analysis based on the continuous wavelet transform is not able to improve lithofacies classification using the self-organizing map (SOM) neural network model from well-logs data. The proposed idea consists to inject many inputs in SOM neural network machines and to choose the best map. These inputs are: data set 1: the five raw well-logs data which are: the gamma ray, density, neutron porosity, photoelectric absorption coefficient and sonic well-log; data set 2: the estimated Holder exponents using the continuous wavelet transform of the data set 1; data set 3: data set 1 and the three radioactive elements concentrations; data set 4: the estimated Holder exponents of the data set 1 and the Holder exponents of the radioactive elements concentrations; data set 5: the estimated Holder exponents of the data set 1 and the three radioactive elements concentrations logs. Application of the proposed idea at two boreholes located in the Algerian Sahara shows that the Holder exponents estimated with the continuous wavelet transform as an input of the SOM neural network are not able to give geological details. However, the raw well-logs as an input give more details and precision especially when they are enhanced with the natural gamma ray spectrometry data.

To improve the robustness of performance assessment for aileron actuators, a novel performance assessment for aileron actuators using multi-fractal detrended fluctuation analysis (MF-DFA) and self-organizing mapping (SOM) neural network is first proposed. First, a generalized regression neural network (GRNN) is employed to establish a fault observer, then the residue is generated by subtracting the actual actuator system output from the observer output. Second, the residue's feature is extracted by MF-DFA. Third, SOM neural network, trained only by the normal residue's feature, is used to obtain the minimum quantization error (MQE) between the current feature and the normal feature, and the health confidence value (CV) is obtained by normalizing MQE. Lastly, four types of faults, namely electronic amplifier gain abrupt change, electronic amplifier gain gradual change, sensor gain change and cylinder leakage, are applied to validate the effectiveness of the proposed method.

The use of neural networks in quality control has been a popular research topic over the last decade. An adaptive self-organizing mapping (SOM) neural network algorithm is proposed to overcome the shortages of traditional neural networks in this paper. In order to improve the classification effectiveness of SOM neural network, this paper designs an improved SOM neural network, which improved the algorithm formula based on input vector, the number setting of competitive layer neurons and the initializing weight vector. And the method is used to classify the product of cement slide shoe bearing in manufacturing process quality control, and experiment results show that the algorithm adapts well the unsupervised learning problems. Keywords-Self-organizing Mapping (SOM), artificial neural network, manufacturing process quality control

The concept of software-defined satellite (SDS) enables in-orbit software reconfiguration and payload virtualization, significantly enhancing the configurability and cross-mission reusability of imaging payloads. However, this flexibility also exposes traditional star identification methods-which depend on fixed intrinsic camera parameters-to severe robustness limitations when dealing with multiple cameras and varying operational conditions. Therefore, a universal and adaptive star identification algorithm is essential to ensure stable and precise attitude determination within the SDS architecture. To this end, this paper proposes a parameter-independent star identification algorithm based on topological invariance features. The method integrates a topological invariant feature coordinate system with a self-organizing map (SOM) neural network, enabling consistent recognition across diverse payload configurations and observation scenarios. Extensive experiments under conditions such as stellar position noise, false-star noise, missing-star noise, focal length variation, and dynamic fields of view demonstrate that the proposed algorithm consistently outperforms three benchmark methods and maintains stable accuracy. These results confirm that the proposed method not only aligns with the SDS philosophy of hardware-software decoupling but also provides a necessary algorithmic foundation for its flexible, reconfigurable, and multi-payload attitude determination framework.

In this paper, we focus on hardware Trojans inserted into IC (Integrated Circuit) chips during IC design. We propose a hardware Trojan detection technique which is based on the chip temperature characterization by using SOM neural network. This method can achieve a high detection rate for Trojans without gold First, we use a tool called HotSpot to get the steady-state heatmap from running IC. Then, we make use of 2DPCA (Dimensional Principal Component Analysis) to extract the features of the heatmap profile and feed them into a SOM (Self Organizing Map) neural network. Finally, the SOM neural network automatically distinguishes Trojan-free chips from Trojaninfected chips. The experimental results show that our method can achieve high detection rate, with 20% chip PV (Process Variability), for all centralized Trojans which are located at different positions.

Generalized part family formation through fuzzy self-organizing feature map neural network

A method for automatic detection of weld defects of short-circuit gas metal arc welding is presented. It is based on the extraction of arc signal features as well as classification of the obtained features using self-organizing feature map (SOM) neural networks in order to get the weld quality information, for example, to determine if there is a defect in the product. This is important for the online monitoring of weld quality especially in robotic welding and lays the foundation for further real-time control of weld quality.

Decomposition of interacting features using a Kohonen self-organizing feature map neural network

PET detectors based on continuous crystal could be an ideal alternative for current high resolution pixelated PET detectors if the issues of high performance position estimation and its real-time implementation are solved. Unfortunately, current existing position estimators are not much feasible for implementation on FPGA. In this paper, we propose a new self-organizing map (SOM) neural network based positioning scheme aiming at not only providing a high performance, but also being realistic for FPGA implementation. Benefitting from SOM feature mapping mechanism, the large set of input reference event data is approximated by a smaller set of prototype, which makes it easier to find the best pattern match for unknown events. Using our experimental data, the scheme is evaluated and compared with others. The test results show that SOM scheme has an equivalent positioning performance with previous k-NN method, and it is realistic to be implemented on FPGA achieving a high event processing throughput.

This paper describes the architecture design of novel massively parallel self-organizing map (SOM) neural networks. The proposed architecture, referred to as the planar SOM (PSOM), is described as a soft IP core synthesized in VHDL. The SOM neural network’s size and the input data vectors’ dimension are adjustable parameters. In this work, several SOM architectures are synthesized and their performance is evaluated for Xilinx Virtex-7 FPGAs. The presented hardware architecture allows online learning and can be easily adapted to a large variety of SOM topologies without a considerable design effort. A [Formula: see text] SOM hardware is validated through the FPGA implementation and its performances with an estimated working frequency of 297[Formula: see text]MHz for a 23-element input vector will reach 21,970 MCUPS in the learning phase and 35,902 MCPS in the recall one.

The Self-Organizing Neural Network Map (SOM) may be considered as good tools for classification and clustering process. Therefore, their performance is investigated in this work to make out the main flaw and propose the suitable improvements to overpass their limitations. These latter were extracted by the use of the SOM to classify a set of values with two different dimensions, the computational time cost and the input space normalization were the main deficiencies faced in this process. It's to note that the SOM was used without any kind of supervising or pre-training. The proposed improvement to overpass these deficiencies relies on enhancing the normalization mechanism with more intelligent techniques that can reduce the computational time and regularize the input space to avoid these limitations.

Music is an indispensable part of our life and study and is one of the most important forms of multimedia applications. With the development of deep learning and neural network in recent years, how to use cutting-edge technology to study and apply music has become a research hotspot. Music waveform is not only the main form of music frequency but also the basis of music feature extraction. This paper first designs a method of note extraction based on the fast Fourier transform principle of the audio signal packet route under the self-organizing map (SOM neural network) which can accurately extract the musical features of the note, such as amplitude, loudness, period, and so on. Secondly, the audio segments are divided into summary by adding window moving matching method, and the music features such as amplitude, loudness, and period of each bar are obtained according to the performance of audio signal in each bar. Finally, according to the similarity of the audio music theory of the adjacent summary of each bar, the audio segments are divided, and the music features of each segment are obtained. The traditional recurrent neural network (RNN) is improved, and the SOM neural network is used to recognize the audio emotion features. The final experimental results show that the proposed method based on SOM neural network and big data can effectively extract and analyze music waveform features. Compared with previous studies, this paper creatively proposed a new algorithm, which can more accurately and quickly extract and analyze the data sound waveform, and used SOM neural network to analyze the emotion model contained in music for the first time.

The paper presents a readily implementable approach for sensor fault detection, identification (SFD/I) and faulted sensor data reconstruction in complex systems based on self-organizing map neural networks (SOMNNs). Two operational regimes are considered, i.e. the steady operation and operation with transients. For steady operation, SOMNN based estimation error (EE) are used for SFD. EE contribution plots are employed for SFI. For operation with transients, SOMNN classification maps are used for SFD/I comparing with the `fingerprint' maps. In addition, extension algorithm of SOMNNs is developed for faulted sensor data reconstruction. The validation of the proposed approach is demonstrated through experimental data during the commissioning of industrial gas turbines.