Complex systems are playing increasingly important roles in the modernization process, so it's necessary to improve their operational reliability. The improvement of system operational reliability can be achieved through faster and more accurate fault diagnosis and prognosis of key components in the system. Traditional machine learning methods and model-based methods have become difficult to effectively diagnose and predict faults, while deep learning, with its excellent data learning ability, has enormous potential in fault diagnosis and prognosis. This paper elaborates on the application research and development of four common deep learning models applied in the fields of fault diagnosis and prognosis: autoencoder and its variants, deep belief network, convolutional neural network and recurrent neural network. The advantages and limitations of various models of deep learning are discussed, and their possible future research directions are prospected.
In response to the conventional approach of designing reliability for redundant systems, which often assumes independence among component failures and may not fully capture the true system reliability, this study introduces a novel reliability simulation design methodology that accounts for the correlation of component failures. Initially, a reliability design model for redundant systems, incorporating component failure correlation, is formulated using the copula function. Subsequently, an innovative simulation solution algorithm, grounded in the generalized reduced gradient method model, is developed. Finally, the efficacy of the proposed model and algorithm is demonstrated through a numerical illustration. The outcomes underscore the pronounced disparities in system reliability design strategies when considering the interdependency of component failures. This research furnishes valuable insights into optimizing the reliability design of equipment redundancy systems.
This paper addresses the challenge of inadequate observation data for the reliability assessment of newly deployed wind turbines. It introduces a methodology for evaluating the reliability of key wind turbine components, grounded in multi-source data. By constructing a fusion algorithm of fault data and performance degradation data, using Monte Carlo sampling technology to realize unit fault data expansion and fusion. Based on the fusion data, distribution function primary, goodness of fit test, parameter estimation, and reliability estimation are carried out. Finally, the reliability of key components of the unit, MTBF, and reliable life of the point estimation and interval estimation. The effectiveness of the method is verified by taking a gearbox as an example.
To achieve the optimal reliability allocation scheme that meets the system reliability requirements while minimizing the cost, an improved particle swarm optimization (PSO) algorithm is proposed in this paper. The optimal allocation scheme is iteratively solved by taking the lowest total allocation cost as the objective function and taking into account the upper and lower limit constraints of component reliability and system reliability index constraints. This paper adopts the four parameter cost reliability model commonly used in engineering to establish an objective function. For the upper and lower limit constraints of component reliability, a correction link is added to the classical PSO algorithm to ensure that the particles generated in each iteration meet the component reliability constraints. For system reliability index constraints, a discard link is added to the classical PSO algorithm to ensure that each particle is a feasible solution. Finally, the above improved PSO algorithm is realized through a numerical example. And compared with other algorithms in the literature, it shows that the improved PSO algorithm can simultaneously obtain the optimization results of the lowest total allocation cost and the highest system reliability.
Electrical connectors are essential components in electrical systems. During their usage, frictional wear can occur between the pin and socket of the contact components. Excessive contact resistance resulting from this phenomenon can lead to system failure, emphasizing the critical importance of ensuring the stability of the contact component's performance. In this study, the commonly used slotted contact component was selected, and its mechanical characteristics were initially analyzed to determine the required parameters for experimentation. Subsequently, vibration tests under different vibration conditions were conducted on the contact component at room temperature, revealing the wear patterns of the contact component under various vibration conditions. Finally, a physical model for the failure of frictional wear was established for the selected contact component model based on the parameters obtained from the experiments.
Industry risk is one of the non-systematic risks affecting the credit risk of listed companies. At present, most of the listed companies implement diversification strategies, but the existing studies often use a single industry to measure the risk of enterprises, which is not scientific. This study thus endeavors to fill this research gap by using quantitative industry prosperity to study the prosperity of individual enterprises based on a comprehensive consideration of the multiple industries that companies are involved in. This article firstly calculates the industry development prosperity index based on financial indicators, and then quantifies the business of listed companies by industry. After that, we calculate the quantified ratio of each company's business in different industries. Then we calculate the development prosperity index and the excess development index of enterprises. Our research data are obtained from the Wind database, and the indicators related to the company's industry information are calculated from the composition of its main business disclosed. Finally, the empirical analysis of the development index is conducted, and the study shows that the index can provide certain incremental information to predict the future development level of the enterprises more effectively than single industry classification.
To address the challenge of limited training data for entity relation extraction in the military domain, training instances within this domain are automatically generated employing the distant supervision method. Subsequently, to mitigate the problem of noise in corpus construction, a two-level attention model for entity relation extraction (RE) is introduced. Following the bidirectional gated recurrent unit (BI-GRU) network, instance-level and word-level attention mechanisms are incorporated. This allows the model to initially capture bidirectional semantic information of the training instances through BI-GRU. Subsequently, it employs word-level attention to identify crucial words within each training instance, and introduces instance-level attention across multiple training instances, emphasizing the key examples. Experiments conducted on the automatically constructed training corpus in the military domain illustrate the model's effective entity relation extraction capabilities. Furthermore, it mitigates the influence of noise data introduced through distant supervision, thereby enhancing the accuracy of relation extraction.
Many computational models used in structural design produce multivariate outputs. This paper aims at quantifying how the subranges of model input variables affect the uncertainty of compositive multivariate outputs. Generalized variance in multivariate statistical theory is firstly introduced to describe the uncertainties of multivariate outputs. From geometric perspective, it is clear that the generalized variance represents both uncertainties of a single output and the correlations between two different outputs. The generalized variance ratio function is then defined to investigate the effect on the multivariate outputs when the uncertainty of one or a set of input variables are reduced to their subranges. Two efficient computational procedures called single-loop Monte Carlo simulation and sparse grid integration are developed to calculate the proposed regional sensitivity index. At last, by applying the proposed method to two typical cases, it is illustrated in detail about how to use the information provided by the developed regional sensitivity theory.
The structures of industrial robots are complex, and they consist of numerous parts. The fault analysis of current industrial robots usually targets a single component or a part, ignoring the fault propagation relationship of the parts. Due to the large number of parts and complex coupling, it is difficult to find the key unit that affects the whole machine, thus this paper proposes a fault propagation analysis method based on meta-action. Firstly, the whole machine is decomposed using the Function-Motion-Action (FMA) decomposition method to obtain each meta-action unit. Secondly, the fault data are combined with the expert experience after fuzzification to jointly construct the initial relationship matrix of the meta-action units and draw the corresponding fault propagation directed graph. Thirdly, the Decision-Making Trial and Evaluation Laboratory (DEMATEL) and the Interpretive Structural Model (ISM) are used to stratify the fault propagation directed graph, to rank the importance of meta-action units to get the key meta-action units, and to carry out the impact analysis between meta-action unit faults. The analysis of a certain model of robot R500 as an example verifies the practicality of the method in this paper.
GCW system is one of the hot research objects of in-situ remediation of contaminated groundwater. However, the system will be difficult to maintain if there is a failure during underground operation. The uncertainty of the source of the failure will make it more difficult to perform maintenance on the system, which will delay the remediation process. Therefore, a method that can diagnose or predict system failures in a timely manner is needed. Based on RPL, this study designs an intelligent PHM system to predict potential failures of GCW systems based on the relationship between performance degradation and failure time of GCW systems. And the maturity of the intelligent PHM system is evaluated into 5 levels, and 15 maturity evaluation factors in 4 dimensions are proposed. The maturity model and its application process are obtained. This modeling method provides ideas for decision-makers and managers to carry out early maintenance, and estimate its development level in related fields.
The use of a protective coating to enhance the weatherability of blades has become a reliable, economical, and important means to ensure aerodynamic performance and prolong the life of blades. In this paper, the model of the 3MW horizontal axis wind turbine blade was established by the Glauert method, and then the 3D model of a 2mm PU coating was established. According to the measured data, the parameters of the Weibull distribution model of wind speed are obtained, and the particle size distribution and final velocity of raindrops are calculated using Bester's empirical formula. The coupling field of blade coating was established in Workbench, and the damage and life of blade coating were analyzed with ANSYS software. Simulation analysis shows that the maximum instantaneous stress response of the coating is 2.79 MPa, which is located far from the blade rotation center and close to the blade tip. This area is close to the maximum cumulative damage area of the coating, and the fatigue life of the coating is about 6.3 years. The research results have guiding value for the design and maintenance of wind turbine blade coatings.
Within the software industry, the pace of software upgrades remains swift due to fierce market competition and rapidly evolving requirements. The fluctuating composition of development teams poses challenges in testing upgraded software. Despite this, limited attention has been devoted to testing strategies and reliability estimation for upgraded software systems, leveraging past testing data. Therefore, it is important and necessary to develop an effective testing strategy in order to ensure the product quality. In this paper, an upgradation software testing strategy adapted to requirements change will be proposed combined with the software development process of Parking Lot Mobile Payment System (PLMPS). In this approach, the testing information of the pre-versions and software functional structure will be fully used in upgradation software testing. Moreover, an upgradation software reliability model can be constructed based on proposed testing strategy. The proposed reliability model also can be applied to multi upgradation software systems. Furthermore, the proposed approach can help managers and developers generate testing cases and predict software reliability. A numerical example based on simulation data is given to illustrate its application in the modeling process.
Instead of the real finite element model, the the approximate model can achieve a balance between calculation accuracy and computational cost, and computation efficiency can also be improved. There are two main ways to implement approximate Modelling: (1) Single step Modelling method; (2) Adaptive Modelling method. When constructing the approximate model, the single-step method employs one-time sampling to obtain sample points. If there are too many sample points, significant computational resources and costs are required for solving; conversely, too few points can lead to inaccuracy in the approximation. Consequently, the single-step method is not typically utilized. Acknowledging these limitations, this paper introduces a Support Vector Regression (SVR) adaptive modeling method, grounded in the infill strategy of maximum entropy. This approach augments the training sample with points of the highest entropy value and updates the approximation model, utilizing the entropy value of the sample to represent the model's error. To prevent over-sampling in any specific area, newly acquired sample points must also adhere to the principle of minimizing distance while maximizing coverage. The computational efficacy of this proposed method is substantiated through three numerical examples.
Based on the improved RPN method, this paper analyzes the reliability of each system of CNC machine tools, studies the fault types, failure modes, and fault causes of each subsystem, and determines its influence on the overall function of the machine tool. Through FMEA analysis and improved subjective and objective data RPN analysis method, the weak position of each system of CNC machine tools is found, the overall impact of each fault on CNC machine tools is understood, and finally, the reliability information of CNC machine tools is obtained. The entropy weight-TOPSIS distance formula is sorted for RPN values, to sort the faults with closer values more accurately, which overcomes the mathematical defects of traditional RPN compared with the traditional RPN method. The research results show that this method can accurately find out the main fault parts of the machine tool, and put forward corresponding preventive measures and solutions for these fault parts, so as to improve the operating life of the machine tool and optimize the performance and reliability of the system. The research results of this paper have important theoretical and practical significance for improving the safety and reliability of CNC machine tool systems.
In order to deeply study the fatigue failure process of wind turbine blade adhesive structure, a corresponding theoretical model is constructed to express the process more intuitively and accurately. Based on the fatigue failure mechanism of the adhesive structure, the improved three-dimensional Hashin criterion, the improved Yeh-stratton failure criterion, and the quadratic nominal strain criterion were selected to determine the failure mode at the local failure of the adhesive structure, and the stiffness of the coefficient reduction method and the cohesive model was used to characterize the sudden drop of local stiffness after the failure of the adhesive structure. The fatigue failure process of the structure was compiled into a program using the FORTRAN language and inserted into the ABAQUS software to establish the progressive damage model of the adhesive structure. The fatigue failure process of the adhesive structure of the wind turbine blade I-beam was analyzed by applying the progressive damage model of the adhesive structure, and the results showed that the progressive damage model of the adhesive structure established in this paper could effectively simulate the fatigue failure process of the adhesive structure of the I-beam.
The State of Health (SOH) for lithium batteries is a crucial safety index, which is usually subjected to the many influencing factors including current, voltage, temperature, and so on. The existing direct indices to predict SOH have limitations with agreeable reliability. A new approach using multi-sensor fusion technology is proposed to construct a comprehensive health indicator for predicting SOH. Firstly, we calculate the probability density of each sensor for the charge/discharge cycle using the kernel probability density method. Secondly, we use the peak value of each probability density function as a new health index. Thirdly, we fuse the strongly correlated health indices using an improved grey correlation analysis method and use the comprehensive health indicator to estimate SOH. Thereafter, we conduct an experimental comparison study using publicly available degradation data of lithium batteries, and the results demonstrate that our method can effectively address the issues of incomplete consideration for influencing factors and low reliability of the SOH characterization indicators.
YOLOv5s algorithm has made significant progress in object detection. However, it contains a large number of parameters, and its detection speed is low. To overcome the problem, an improved YOLOv5s algorithm is proposed. Firstly, ghost convolution and ghost bottleneck layers of the GhostNet are added to the convolution and bottleneck layers of the YOLOv5s algorithm. Secondly, a feature pyramid network is employed to continuously update the weights of the branches during training, which can fuse features of different resolutions. Finally, an Attention Mechanism is added to the backbone network to extract the deep features of mechanical parts. The experimental results indicate that the proposed algorithm can more accurately and quickly identify mechanical parts.
Wind turbines are devices that convert wind energy into mechanical energy, a process largely governed by the interaction of the wind with the turbine's blades. Accurate recognition of wind characteristics, therefore, forms a vital premise for the effective operation of the wind turbine impeller. The probability distribution of wind speed and direction at specific points within a wind field has substantial implications for turbine location, layout, and fan design, and plays a critical role in the reliability analysis of the system. Since wind speed is subject to a high degree of uncertainty, it constitutes a major factor in assessing the risk of system failure. This uncertainty is influenced by various factors, including temperature, atmospheric pressure, terrain, and geography, all of which contribute to random fluctuations in wind speed. However, despite these uncertainties, the convergence time and spatial characteristics of wind speed do follow certain patterns. Therefore, the establishment of a reliable wind speed model holds significant value for studying the characteristics of wind turbine impellers. Such a model would enable more precise predictions and optimizations, enhancing the overall performance and reliability of wind energy systems.
Autocorrelation of the immediate coda of P waves provides an alternative to traditional receiver functions. Working at higher frequencies enhanced information can be extracted on the Moho surface. We show that the autocorrelation process is equivalent to a virtual point source at the surface, and this enables the extraction of local reflectivity. ACC (Auto-Correlogram Calculation in Seismology) software is used to process and analyze teleseismic P-wave coda recorded along the Bilby array aligned north to south and the Marla array aligned east to west in central Australia. Clear signals can be extracted for the Moho, mid lithospheric discontinuity (MLD), and the transition from the lithosphere to the asthenosphere (LAT), exploiting reflectivity estimates in different frequency bands (up to 5 Hz upper bound). The Moho of Bilby lies at 40 km or deeper, the MLD around 80-90 km depth, and the LAT between 130 and 180 km depth. A cross-check on the nature of the seismic structure is provided by the intersection with the dense Marla array (MAL) in an east-west direction near stations BL19 and BL20. Consistent reflectivity results are extracted in orthogonal directions.
The dynamic model of rotary vector (RV) reducer is capable of simulating the actual running state of RV reducer and providing the simulation data of RV reducer running in various states. However, the dynamic model parameters of the RV reducer need to carefully selected to improve the precision of the corresponding dynamic responses, which is the focus of this work. Twentyfour dynamic model parameters, speed, torque, stiffnesses and dampings, are optimized using the whale optimization algorithm (WOA). Initial parameters and experimental data for an RV reducer are used as the starting point for the optimization. The optimized dynamics model parameters are determined, resulting in simulation data that exhibits a higher similarity with the experimental data than that using initial parameters.
