查看更多>>摘要：In this paper, a new model updating-based damage detection method based on a modified sensitivity equation is proposed that can tackle the problem of damage detection in structures with closely-spaced eigenvalues. It is known that modal information, such as natural frequencies, in these structures can be of close proximity, making the procedure of damage detection hard if not impossible. The obtained sensitivity equation uses incomplete measurements from frequency response functions (FRFs) to conduct the challenging damage detection of structures with closely-spaced eigenvalues. Although using FRF for damage detection of this kind of structures can offer some advantages over modal data, there are still some challenges that need to be addressed. For instance, it is not possible to have the response of the structure measured in all of its degrees of freedom. As such, the proposed FRF-based model updating method is capable of overcoming these limitations and has the advantage of avoiding modal analysis errors. In order to evaluate the efficiency of the proposed method, two numerical examples of a 144-element three-layered laminated composite plate and a 120-element three-dimensional truss structure, as examples of structures with closely-spaced eigenvalues, are studied. The results demonstrate the capability of the proposed method in damage detection of structures with closely-spaced eigenvalues with incomplete measurement data. Moreover, the results of comparison between the proposed method with some other methods demonstrate the superiority of the proposed method in damage detection of structures with closely-spaced eigenvalues using noisy incomplete FRF data.
查看更多>>摘要：On-machine measurement (OMM) systems with touch-triggered probes are efficacious in productivity improvement. Machine motion errors impose major limitations on their accuracy for any OMM systems, especially when rotary axes motions are involved. This research proposes a precise measurement method to address this problem by simultaneously identifying the probe parameters and the location errors of rotary axes, using a touch-triggered probe to sample a sphere. This method is suitable for both periodic machine checks and OMM system recalibrations. The method formulates a mathematical model based on machine kinematics to represent errors between the theoretical probed data and the actual recorded data, minimizing which can solve for necessary parameters. Two proposed probing patterns simplify the solution to the model. Experiment results demonstrate that the method is effective and accurate. Monte Carlo simulations were carried out to prove the robustness of the method.
查看更多>>摘要：The power quality disturbances caused by large-scale grid connection of nonlinear loads and distributed generations seriously affect the safe and stable operation of precision computers and microprocessors in the power grid, and may cause serious security accidents and economic losses in some cases. Therefore, the accurate classification of power quality disturbances is of great significance for the power supply quality improvement, the power equipment condition monitoring, and the troubleshooting of power grid. For this reason, a novel method based on visual attention mechanism and feed-forward neural network is proposed to classify single and combined power quality disturbances caused by non-balanced, nonlinear loads and distributed generations in the power grid. In the first step of the proposed method, visual attention mechanism is utilized to extract the disturbance features of power quality disturbances, through performing disturbance region selection, multi-scale spatial rarity analysis, and disturbance feature fusion on the binary image converted from the original voltage signal successively. Then, four disturbance feature indexes are selected for the characterization of power quality disturbances. Finally, a classifier using feed-forward neural network is constructed to distinguish various single and combined power quality disturbances. The classification accuracy of the proposed method is compared with that of several existing methods for the classification of power quality disturbances from two types of datasources. The power quality disturbances from the simulation operating conditions include eight kinds of single and thirty-eight kinds of combined power quality disturbances. The power quality disturbances from the IEEE Work Group P1159.3 and P1159.2 Datasets include seven kinds of single and eleven kinds of combined power quality disturbances. Comparison results demonstrate that the proposed method can classify single and combined power quality disturbances more accurate than the compared classification methods, which verifies the effectiveness of the proposed method.
查看更多>>摘要：Differential scanning calorimetry (DSC) is a powerful technique to study protein stability, since the DSC test data provides valuable insights to characterize protein folding thermodynamics. Researchers in the drug discovery field need to manually analyze the DSC curves in multiple steps, such as baseline subtraction, data fitting, integration, and domain deconvolution. To improve the efficiency and consistency of data processing, machine learning methods for automatic DSC peak identification and baseline estimation were seen in prior research. However, the DSC's automatic peak separation remained unexplored, despite its significant role in explaining the multi-domain protein unfolding. In this research, we propose a method based on reinforcement learning to separate the overlapping peaks of the DSC signal. We use two types of protein data to verify the effectiveness of this method. It automatically deconvolutes the peak signals into multiple sub-peaks. Our automated analysis method could lead to improved efficiency in DSC signal analysis when high volume data is involved. The code and data for this work can be found at: https://github.com/shuyu-wang/DSC_analysis_peak_separation.
查看更多>>摘要：As an important part of integrated logging, gas logging can directly measure the content and composition of hydrocarbon gas in the formation, and plays an irreplaceable role in the discovery and evaluation of oil and gas while drilling and real-time drilling parameter monitoring. Traditional gas logging uses chromatographs to analyze downhole gas. However, due to its long measurement period, many additional instruments, long lines, strict environmental requirements, complex operations, and easy distortion, it has been difficult to meet the requirements of current exploration work in complex oil and gas reservoirs. With the advantages of fast analysis speed, rich detection parameters, short delay time, good stability, and strong anti-pollution ability, the quantitative infrared spectroscopy technique can replace chromatography for hydrocarbon gas measurement. However, when it is used for the quantitative analysis of downhole hydrocarbon gases, a large span of absorption spectra and overlapping characteristic peaks can occur due to the complexity of gas components, making quantitative identification more difficult. To address the above problems, this paper proposes a method of net signal analysis (NAS) combined with density peak clustering (DPCA) for local PLS modeling. The method first obtains the net signals of the unknown and correction samples by the net signal analysis and then clusters the net signals by DPCA to select the corresponding local correction for PLS modeling prediction. The experimental results show that the NAS-DPCA-PLS method proposed in this paper is significantly better than the traditional method in the quantitative analysis of hydrocarbon mixed gas components, and the average accuracy of mixed component prediction reaches more than 98%, which effectively improves the detection accuracy of quantitative analysis of mixed component infrared spectroscopy and provides important theoretical support for the use of infrared spectroscopy in gas logging.
查看更多>>摘要：Predicting the remaining useful life (RUL) of rolling bearings under different working conditions improved significantly by transfer learning. However, existing methods have not studied the following problems thoroughly: (1) The influence of the discrepancy between features of different dimensions on the feature transfer process; (2) The feature transfer process in the degradation stage with apparent discrepancy has a significant influence on the transfer prediction of remaining useful life. In this study, a degradation occurrence time identification method based on the distribution differences in reconstructing degradation indicators has been proposed to obtain samples of degradation stages. A stack convolutional autoencoder model based on a multidomain adversarial network is also proposed to reduce the impact of discrepancies among extracted degradation features on the feature transfer process. As per the experimental results, it was found that the proposed method can effectively improve the RUL prediction accuracy.
查看更多>>摘要：Initial in-flight alignment is the basis of accurate navigation for projectiles strap-down inertial navigation system (SINS). Due to complex and highly dynamic flight environment of projectiles, inertial sensors and GNSS are susceptible to interference, which causes measurement noise appear as non-Gaussian noise, resulting in low alignment accuracy and long alignment time. Thus, this paper proposes statistical confidence domain data driven based fast in-flight alignment method. Firstly, the K matrix is used as state variables to construct projectiles initial in-flight alignment filter model. The noise evaluation indexes are defined according to measurement information and estimation results to judge abnormal degree of measurement noise. Based on this, we propose an adaptive robust matrix Kalman filter (ARMKF) method. The measurement variance matrix formulas are derived based on additive noise, which provides theoretical support for parameter selection in practical applications. Simulation and test results show that alignment accuracy and alignment time of the proposed method are better than traditional methods.
查看更多>>摘要：Dimensional inspection, which consists in verifying the geometric conformity of parts in terms of ISO specifications, is an essential step in the product life cycle. In recent years, new optical measurement technologies, such as laser sensors or structured light sensors, have become increasingly important. However, the quality of the data limits their expansive use for metrology applications and touch probes with CMM (Coordinate Measurement Machine) remain the reference in terms of measurement quality. On the other hand, the cooperative use of an optical system and a contact sensor presents a good solution which takes advantage of the characteristics of both systems while maintaining a quality level satisfying the requirements of metrology. However, measurement with a laser sensor and a touch probe mounted on a CMM results in two heterogeneous point clouds that must be processed before the final evaluation of specifications. This paper presents a methodology that addresses the processing of measurement data in this scenario. The proposed approach consists of four phases. In the first phase, the measurement data resulting from the laser sensor is processed to remove outliers, reduce noise and partition the point cloud. In the second phase, the point cloud resulting from both sensors are unified in a unique coordinate system. In the third phase, a fusion method is proposed to fuse the point clouds. The specifications are finally evaluated based on the fused point clouds in the fourth phase. A case study is conducted to illustrate the proposed approach with the measurement data of a test part defined by its CAD model and specifications. The results are compared with a reference inspection report obtained by a touch probe measurement system.
查看更多>>摘要：Currently, it has been considered a nonlinear optimization problem to accurately evaluate axis straightness error of shaft and hole parts. Using intelligent optimization algorithm to solve this problem can avoid complex mathematical modeling process, while showing the advantages of high solution accuracy, fast search speed and easy convergence. By using the grey wolf optimization (GWO) algorithm with strong convergence performance, the global search performance was improved by regulating the linear convergence factor to nonlinear, and the wolf in the optimal position was endowed with the capability of receiving information and moving autonomously. Thus, an improved grey wolf optimization (IGWO) algorithm with better optimization accuracy was yielded. Moreover, the fitness function of optimization was rebuilt, thereby avoiding the unscientific setting of the parameter optimization range based on the subjective experience. Lastly, IGWO was successfully applied to the evaluation of axis straightness error of shaft and hole parts with good accuracy.
查看更多>>摘要：In industrial systems, the vibration signals of rolling bearings are influenced by changing operating conditions and strong environmental noise, therefore they are often characterized by high complexity. The multi-scale deep learning method can achieve bearing fault diagnosis under complex operating conditions, however, the importance of dynamic feature selection is neglected. To solve this problem, we propose a multi-scale dynamic adaptive residual network (MSDARN) fault diagnosis method. In the proposed method, we combine multi-scale learning and attention mechanism to construct a multi-scale dynamic adaptive convolutional layer (MSDAC). To learn vibration signal features, MSDAC can dynamically adjust the weights of different scale convolutional layers. In addition, we introduce a nonlinear function to adaptively determine the scaling rate parameter in MSDAC. Finally, in order to improve the feature learning ability of the proposed method, we use MSDAC and residual connections to construct residual blocks, and use multiple such residual blocks to construct MSDARN. The effectiveness of the proposed method is verified by noise, variable load and mixed fault experiments, and the proposed method has higher fault classification accuracy than other three deep learning methods.
NSTL
Elsevier