Search

To solve the problems of limited expression of feature information in a discrete evaluation sequence, the high cost of model optimization, and insufficient discrimination information to detect missing classes, a zero-small sample capability evaluation (ZES) method based on model structure self-optimization is proposed. Spatial transformation of the gram angle field was used to optimize the information richness and data correlation of the sample sequences. A multi-mechanism fusion strategy (MFS) was designed for self-optimization of the visual feature extraction model. Through self-adaptive global optimization of ten key parameters in the model, a computational model suitable for small samples and with self-learning ability was constructed. Sample semantic attributes and a visual-semantic feature mapping were constructed, and the unseen class labels were predicted according to cosine similarity. The experimental results verify the feasibility and effectiveness of the proposed method in the aspects of model adaptive optimization and evaluation of unseen samples and provide a new idea for constructing a comprehensive evaluation, prediction, and diagnosis system and method design in related fields.

The accurate prediction of remaining useful life is a significant issue for ensuring the reliable operation of the system. Considering the dynamic transfer of degenerate state can improve the prediction accuracy and reduce the number of late prediction. Firstly, a time series density peak clustering algorithm suitable for real-time manifold data clustering is proposed. By choosing larger truncation distance at points with high sample density, the cluster centers can be found more accurately. Then, different degradation state patterns according to clustering results can be divided. Moreover, the smoothing parameters can be adaptively updated according to the sample density under different degradation modes and an adaptive kernel density remaining useful life estimation model is established. The test of the gearbox verifies the necessity and accuracy of the proposed model by comparison with the remaining useful life predictions of kernel density estimation without considering degraded state transitions.

Remaining useful life (RUL) prediction plays an important role in improving the availability and productivity of systems. To improve the accuracy of real-time RUL prediction during system operation, we propose a modeling method for real-time RUL prediction based on adaptive kernel window width density. First, a non-parametric kernel density estimation (KDE) real-time RUL prediction model is proposed, and a window width model with adaptive kernel window width density is established by introducing a local density factor in the window width selection. The local density of sample points is calculated by the k-nearest neighbor distance, and the KDE is performed by adaptively selecting the window width value according to the local density of sample points in the region of non-uniform distribution of sample points. As the monitoring data changes in real time, the kernel density estimates of known samples are used to recursively update the kernel density estimates of new samples. Moreover, the logarithmic transformation of random variables and space mapping are used in the establishment of the RUL prediction model. A model of logarithmic kernel diffeomorphism transformation is established to solve the boundary shift problem of kernel estimation in the prediction to improve the prediction accuracy. Finally, the validity of the method is verified through case studies, and the accuracy of the model is judged using evaluation quasi-measures.

Mechanical fault vibration signals are non-stationary, which causes system instability. The traditional methods are difficult to accurately extract fault information and this paper proposes a local mean decomposition and least squares support vector machine fault identification method. The article introduces waveform matching to solve the original features of signals at the endpoints, using linear interpolation to get local mean and envelope function, then obtain production function PF vector through making use of the local mean decomposition. The energy entropy of PF vector take as identification input vectors. These vectors are respectively inputted BP neural networks, support vector machines, least squares support vector machines to identify faults. Experimental result show that the accuracy of least squares support vector machine with higher classification accuracy has been improved.

Three-parameter and four-parameter sine-wave fitting algorithms are powerful tools for estimating the parameters of the excited single-tone sine-wave for ADC. In the dynamic performance testing processes of instruments, the angle frequency, amplitude, phase and dc component of the input sine-wave are all unknown, so the fitting procedure is nonlinear. This paper proposes and analyzes a test method based on iteration Interpolated Discrete Fourier Transform (IpDFT) and sine-wave fitting method for evaluating the effective number of bit (ENOB) of the acquisition channel. Mathematical expressions of the Least-square fitting residual error and the proposed ENOB evaluation based on iteration IpDFT method are derived. These expressions are then particularized for acquisition circuit output noise composed of single-tone and additive white noise. Simulation results show that the DFT-based golden section searching algorithm (DGSSA) is an effective algorithm under coherent and non-coherent sampling conditions. The accuracy of the derived expressions and estimated parameters are verified through both the computer simulations and experimental results.

Dempster-Shafer (DS) theory is a kind of classical reasoning method to deal with uncertainty data, but in case of data are made up with serious interference, the result is often contrary to common sense. Based on the analysis of the completeness of DS theory, a modified evidence combination rule is proposed. The rule introduce a concept of credibility of evidence, divide the evidence into reliable and unreliable parts for pretreatment and finally fusion information by DS theory. Compared with other methods, this method, under the premise of without changing DS fusion rule, has solved the problem that DS theory cannot be used in the situation when evidence data are highly conflict.

Mechanical fault diagnosis is a feature extraction and pattern recognition process. The feature extraction is related to the accuracy of fault diagnosis and the reliability of the prediction. How to extract fault features effectively and build an accurate model to recognize fault is the key of diagnosis technology. Accordingly, a feature extraction method based on LMD-approximate entropy was proposed, and combined it with LSSVM to diagnose mechanical fault. Firstly, the decomposition of fault feature by LMD, and then the approximate entropy of product function were taken to extract fault features accurately. Finally, the eigenvectors were input to LS-SVM for fault recognition. Compared with EMD and wavelet decomposition, the results show that it can extract fault features effectively and can improve the accuracy and speed of fault diagnosis. DOI: http://dx.doi.org/10.11591/telkomnika.v11i2.2028

In view of the application scenarios and demand of mechanical equipment pump truck, design a pump truck MRO system based on the Internet of Things and Cloud Computing, and the overall structure, General structure are given, and the system function is designed. Using the system, the functions can be realized, such as pump truck remote data acquisition, diagnosis processing and production management.

Injection molding materials selection based on neural network has broad application prospects in material selection problem. This paper puts forward a PSO-H-BP algorithm and applies to injection material selection on the basis of traditional BP algorithm. This method uses Hopfield network which optimized by particle swarm algorithm for original data pretreatment then selects material by BP network. The MATLAB simulation and result shows that: this method has higher classification accuracy and good expansibility compared with traditional BP neural network and H- BP neural network.

By the analysis of NOx formation principle in the coal-fired process of power station boiler, the boiler operation parameter related with NOx emission are determinate. Then the furnace temperature field is gained by using combustion flame image processing technique and dual-color high temperature measurement method. A data matrix is defined with the boiler operation parameter and the furnace temperature field, as the input of partial least squares. It is concluded that the recognition results agree basically with that by partial least squares, which shows that the method can be regarded as a detective method of NOx emission in high temperature air combustion.

A modified multi-sensor information fusion method for hydraulic fault diagnosing system is proposed in this paper. Combined with the improved JDL data fusion model and the hierarchical processing idea, it can solve some difficult fault diagnosis problems of hydraulic system. The adaptive weighted least squares estimation method is used to clean the data and extract the feature in data layer. The multi-parallel particle swarm optimisation (PSO)-Hopfield neural network is applied in feature level for local diagnosis. When the time-airspace integrates, there is a direct data communication and feedback between each level based on modified Dempster-Shafer (D-S) evidence theory in decision-making level. The final diagnosis has a direct data communication and feedback between each level, and it can make the information of each level based on data mining as soon as possible. Experimental results show that the method in conflicted evidence has high correct rate and can avoid index explosion and fix the fault exactly.

D-S evidence theory has a strong advantage in dealing with uncertain information. For the problem of the classic combination rule which is a failure in conflict evidence and the decision-making rule which is difficult to choose, a modified D-S combination rule and a new method of decision-making rule are proposed. At the same time, basic probability assignment BPA of the proposition is obtained by PSO-RBF ANN, which avoided the imperfection that the BPA is obtained by experience. In addition, we apply this algorithm to hydraulic fault diagnosis system and the simulating experiment shows that the method is beneficial to final decision-making under the premise that the risk of decision-making is not increasing, and it is more accurate to locate the fault.