High Computation Time Cost Research Articles

Maximum negative entropy deconvolution and its application to bearing condition monitoring

Blind deconvolution (BD) has proven to be an effective approach to detecting repetitive transients caused by bearing faults. However, BD suffers from instability issues including excessive sensitivity of kurtosis-guided BD methods to the single impulse and high computational time cost of the eigenvector algorithm-aided BD methods. To address these critical issues, this paper proposed a novel BD method maximizing negative entropy (NE), shortened as maximum negative entropy deconvolution (MNED). MNED utilizes NE instead of kurtosis as the optimization metric and optimizes the filter coefficients through the objective function method. The effectiveness of MNED in enhancing repetitive transients is illustrated through a simulation case and two experimental cases. A quantitative comparison with three existing BD methods demonstrates the advantages of MNED in fault detection and computational efficiency. In addition, the performance of the four methods under different filter lengths and external shocks is compared. MNED exhibits lower sensitivity to random impulse noise than the kurtosis-guided BD methods and higher computational efficiency than the BD methods based on the eigenvalue algorithm. The simulation and experimental results demonstrate that MNED is a robust and cost-effective method for bearing fault diagnosis and condition monitoring.

A Comparative Study of Nonlinear MPC and Differential-Flatness-Based Control for Quadrotor Agile Flight

Accurate trajectory-tracking control for quadrotors is essential for safe navigation in cluttered environments. However, this is challenging in agile flights due to nonlinear dynamics, complex aerodynamic effects, and actuation constraints. In this article, we empirically compare two state-of-the-art control frameworks: the nonlinear-model-predictive controller (NMPC) and the differential-flatness-based controller (DFBC), by tracking a wide variety of agile trajectories at speeds up to 20 m/s (i.e., 72 km/h). The comparisons are performed in both simulation and real-world environments to systematically evaluate both methods from the aspect of tracking accuracy, robustness, and computational efficiency. We show the superiority of the NMPC in tracking dynamically infeasible trajectories, at the cost of higher computation time and risk of numerical convergence issues. For both methods, we also quantitatively study the effect of adding an inner loop controller using the incremental nonlinear dynamic inversion method, and the effect of adding an aerodynamic drag model. Our real-world experiments, performed in one of the world’s largest motion capture systems, demonstrate more than 78% tracking error reduction of both NMPC and DFBC, indicating the necessity of using an inner loop controller and aerodynamic drag model for agile trajectory tracking.

Digital image correlation with topology-based matching algorithm on dots pattern and its application in large deformation measurement of nitrile–butadiene rubber

Digital image correlation (DIC) has inherent limitations in large deformation measurement, such as speckle pattern decorrelation and high computation time cost. A new dots pattern with fixed size and randomly distributed dots is devised for DIC, rather than the traditional grayscale speckle pattern. Topology-based matching correlation (TMC) based on dots pattern is proposed, in which topology connectivity property is used in correlation calculation for images before and after deformation. Experiments on simulated tensile deformation with up to 300% elongation and nitrile-butadiene rubber tensile tests with 260% elongation were designed and results showed that, compared with the incremental-DIC method, the proposed TMC based DIC (TMC-DIC) method could correctly obtain large deformation with no need of updating reference image, less time consumption and smaller errors. The proposed method also has advantages in the applications like real time deformation observation or deformation measurement of large-scale structure.

LogNADS: Network anomaly detection scheme based on log semantics representation

Semantics-aware anomaly detection based on log has attracted much attention. However, the existing methods based on the weighted aggregation of all word vectors might lose the semantic relationship of word order and cannot maintain the unique representation, and the methods based on word order-preserving by concatenating all word vectors might lead to a high computation time cost. To solve these issues and further improve the sequential anomaly detection, this paper proposes a network anomaly detection scheme LogNADS by designing a novel log semantics representation method and an adaptive sequence data construction method. It first discards the useless words and then selects theme words to hold the log abstraction and maintain a low time cost as well. Subsequently, it concatenates theme words’ vectors based on the original word order to maintain the unique representation and avoid the word order loss. Furthermore, to better detect the sequential anomalies, we utilize the sliding window scheme and design a method to compute the optimal window size for constructing the log sequence self-adaptively, and then LSTM is built to extract timing characteristics of the log sequences. Experimental results conducted on the public benchmark HDFS dataset and BGL dataset demonstrate the effectiveness of LogNADS through comparing with other state-of-the-art methods in the detection accuracy and time cost. Moreover, the statistical significance tests prove the superior performance.

New Efficient Technique for Finite Element Modeling of Macro Fiber Composite Piezoelectric Materials

A lot of interest to simulate piezocomposite actuators with finite element method has been increased recently. However, there are still open questions regarding the modeling methodology, accuracy, and computational time cost. In this work, a new technique for modeling macro fiber composite piezoelectric actuator by finite element analysis is proposed. The presented technique models the piezocomposite actuator as a simple monolithic piezoceramic material with just two electrodes along its longitudinal extremes instead of using the actual large number of electrodes which results in very fine finite element mesh with high computational time cost. The proposed technique is validated successfully by comparing its results with those of the actual detailed model as well as with the published experimental results and manufacturer’s data.

Two heuristics for the rainbow spanning forest problem

A rainbow spanning forest (say RF) of a given connected, undirected and edge-colored graph (G) is a spanning forest of a set of rainbow components, where a rainbow component is a tree whose edges have different colors. The rainbow spanning forest problem (RSFP) of G aims to find a RF of G with the minimum number of rainbow components. The RSFP which is recently introduced is NP-hard on general graphs and trees, but finds its applications in many situations in which one requires to distinguish between different types of connections in network. To the best of our knowledge, there exists only one paper (Carrabs et al. 2018b) that presents a problem-specific heuristic called greedy algorithm (say GH) and a multi-start scheme embedding this GH (say MSGH) in order to further improve the results. This paper presents a novel and fast problem-specific heuristic Heu_RSF. With the intent of improving solution quality at the cost of higher computational time, we present an artificial bee colony algorithm (ABC_RSFP) for the RSFP. Main components of ABC_RSFP such as randomized version of Heu_RSF in solution initialization and grouping-based neighborhood strategies in conjunction with the use of Heu_RSF as a repair operator help in making ABC_RSFP more effective framework. On a set of benchmark instance scenarios, experimental results suggest that Heu_RSF is effective particularly on instances of larger scenarios with higher density and is computationally much faster than GH and MSGH. ABC_RSFP outperforms MSGH on almost all instance scenarios and shows its effectiveness in terms of computational time.

Feature selection based on hybridization of genetic algorithm and competitive swarm optimizer

Feature selection is one of the hottest machine learning topics in recent years. The main purposes of it are to simplify the original model, improve the readability of the model, and prevent over-fitting by searching for a suitable subset of features. There are many methods for this problem, including evolutionary algorithms and particle swarm optimization. Among them, the competitive swarm optimizer is a new optimization algorithm proposed in recent years, which is based on particle swarm optimization algorithm, and has achieved good results in high-dimensional feature selection problems, but it also has the problems of high computation time cost and easily being premature. Aiming at these problems, this paper proposes to add the crossover operator and mutation operator in the genetic algorithm to the competitive swarm optimization, so as to improve the generation speed of new individuals in the algorithm and prevent premature population. After testing on UC Irvine Machine Learning Repository, the new algorithm not only improves the computational efficiency, but also avoids the problem that the competitive swarm optimization algorithm is easy to fall into the local optimum, which greatly improves the calculation effect.

Casting light on forcing and breaching scenarios that lead to marine inundation: Combining numerical simulations with a random-forest classification approach

Identifying the offshore forcing and breaching conditions that lead to marine inundation is of high importance for risk management. This task cannot be conducted by using a numerical hydrodynamic model due to its high computation time cost (of several minutes or even hours). In the present study, we show how the random forest (RF) classification technique can approximate the numerical model to explore these critical conditions. We focus on the Bouchôleurs site, which is located on the French Atlantic coast and exposed to overflow processes. An iterative strategy is developed for selecting the numerical simulations (a total of 200) to train the RF model. The sensitivity to the input parameters is studied using permutation-based importance measures and extended versions of the partial dependence plots. The results highlight the key interplay among the high-tide level, the surge peak and the phase difference, and the complex role of the breaching location.

Cooperative Resource Allocation in Cognitive Radio Networks With Wireless Powered Primary Users

This letter focuses on cooperative resource allocation in multi-carrier cognitive radio (CR) networks where the primary user (PU) is wirelessly powered by the energy harvested from the signals transmitted by the secondary user (SU) for its own information transfer. We adopt the two-stage harvest-then-transmit protocol, where the wireless power transfer for the PU is executed first and then the information transfer for the PU takes place. Specifically, two schemes, namely subcarrier sharing (SS) and subcarrier exclusivity, are considered for coordinating the information transfer of the PU and the SU in the second stage. The resource allocation problems to maximize the SU rate subject to the PU minimum rate constraint are investigated. It is shown that the SS scheme achieves higher SU rate than the SE scheme at the cost of higher computation time when the PU minimum rate constraint is stringent and the number of subcarriers is large.

Source characterisation by mixing long-running tsunami wave numerical simulations and historical observations within a metamodel-aided ABC setting

Uncertainty related to the source parameters of earthquake can largely impact the tsunami-induced wave characteristics, especially in the case of near-field tsunami source. The combination of numerical simulations and historical eyewitness accounts can be used to better constrain those uncertainties. In the present study, we propose a Bayesian procedure to infer (i.e. learn) the probability distribution of the source parameters of the earthquake. The strategy is based on the combination of: (1) kriging-based metamodelling techniques to overcome the high computation time cost of the numerical simulator; and (2) Approximate Bayesian Computation (ABC) procedure to perform the Bayesian inference. The procedure is applied to the Ligurian (North West of Italy) 1887 tsunami case, for which tsunami-induced sea surface elevations at the coast have been reported at four locations, namely Marseille, Imperia, Diano-Marina and Genoa. The kriging metamodels are trained using only 300 long-running numerical simulations that were performed using the FUNWAVE-TVD code. Contrary to recent inversion exercises that can benefit from current modern observation networks (like tide gauges, sea bottom pressure gauges, GPS-mounted buoys), the case of historical tsunami like Liguria is complicated by the imprecision and scarcity of the observations: this is accounted for by conducting the combined ABC-kriging procedure a large number of times (i.e. 1000); each time a new set of observations being randomly generated to account for this observational error. The combined analysis of the inference results and of the observation uncertainty reveals that: (1) the coseismic slip is the most important source parameter with a very peaky density distribution around low values ranging from 0.3 to 0.6 m; (2) The fault width has a peaky density distribution around low values ranging from 10 to 12 km; (3) The rake and azimuth distribution only slightly deviate from the uniform prior, hence indicating a low influence of those parameters; (4) The bi-modal distribution of the dip is also evidenced.

Parallel topic model and its application on document clustering

This paper presents PLDACOL, our parallel implementation on LDACOL model, to effectively cluster large-scale documents. Since phrases contain more semantic information than the sum of its individual word, we use topic model LDACOL for phrase discovery, and use Gibbs sampling for parameter inference. PLDACOL overcomes the high computation time cost in parameter inference by the distributed computing framework based on Hadoop. We show that our PLDACOL can be applied to the clustering of large-scale documents in different size and produces significant improvements on both effectiveness and efficiency compared with other related traditional algorithms.

Parallel topic model and its application on document clustering

This paper presents PLDACOL, our parallel implementation on LDACOL model, to effectively cluster large-scale documents. Since phrases contain more semantic information than the sum of its individual word, we use topic model LDACOL for phrase discovery, and use Gibbs sampling for parameter inference. PLDACOL overcomes the high computation time cost in parameter inference by the distributed computing framework based on Hadoop. We show that our PLDACOL can be applied to the clustering of large-scale documents in different size and produces significant improvements on both effectiveness and efficiency compared with other related traditional algorithms.

Gender and texture classification: A comparative analysis using 13 variants of local binary patterns

Among very popular local image descriptors which has shown interesting results in extracting soft facial biometric traits is the local binary patterns (LBP). LBP is a gray-scale invariant texture operator which labels the pixels of an image by thresholding the neighborhood of each pixel with the value of the center pixel and considers the result as a binary number. LBP labels can be regarded as local primitives such as curved edges, spots, flat areas etc. These labels or their statistics, most commonly the histogram, can then be used for further image analysis. Due to its discriminative power and computational simplicity, the LBP methodology has already attained an established position in computer vision. LBP is also very flexible: it can be easily adapted to different types of problems and used together with other image descriptors. Since its introduction, LBP has inspired plenty of new methods, thus revealing that texture based region descriptors can be very efficient in representing different images. Nowadays, many LBP variants can be found in the literature. This article reviews 13 variants and provides a comparative analysis on two different problems (gender and texture classification) using benchmark databases. The experiments show that basic LBP provides good results and generalizes well to different problems and hence can be a good starting point when trying to find an optimal variant for a given application. The best results are obtained with BSIF (binarized statistical image features) but at the cost of higher computational time compared to basic LBP. Furthermore, experiments on combining three best performing descriptors are conducted, pointing out useful insight into their complementarity.

Boosting Kernel-Based Dimension Reduction for Jointly Propagating Spatial Variability and Parameter Uncertainty in Long-Running Flow Simulators

Assessing the impact of multiple sources of uncertainty in flow models requires simulating the response associated with a large set of stochastic realizations ( $$>\!\!1,\!000$$ ) of both the spatial variable (e.g. permeability field) and of the large number of model parameters (e.g., of the relative permeability curve). Yet, this Monte Carlo approach may be hindered by the high computation time cost of long-running flow simulators (CPU time $$> $$ several hours). A possible strategy can then rely on the combination of meta-modelling techniques and of basis set expansion like kernel principal component analysis to reduce the high-dimensionality ( $$>\!\!10,\!000$$ grid cells) of the spatial variable. Using a synthetic heterogeneous channelized reservoir, it is shown that $$\sim $$ 50 principal components PCs are still necessary to retain $$\sim $$ 80 % of the total variability: this might pose difficulties for the construction of the meta-model by imposing a large set of training samples. This computational limitation can be overcome by improving the meta-modelling phase through boosting techniques, which allow selecting during the fitting process the input variables (PCs or model parameters), which are the most informative regarding the prediction accuracy, and allow in this manner reducing the number of necessary model runs. The procedure is applied to a synthetic flow model of a CO $$_{2}$$ geological storage with low CPU time. By comparing the P10, P50 and P90 quantiles of the pressure build-up estimated using the true simulator (1,500 simulations) with the ones estimated using the meta-model. The joint impact of spatial variability and of 12 uncertain parameters is shown to be accurately captured using a number of training samples of the order of the number of PCs ( $$\sim $$ a few tens). The level of performance proved to be as high as the one of the distance-based alternative proposed in the literature to solve the problem of joint propagation.

Towards spatially constrained gust models

With the trend of moving towards 10-20 MW turbines, rotor diameters are growing beyond the size of the largest turbulent structures in the atmospheric boundary layer. As a consequence, the fully uniform transients that are commonly used to predict extreme gust loads are losing their connection to reality and may lead to gross overdimensioning. More suiting would be to represent gusts by advecting air parcels and posing certain physical constraints on size and position. However, this would introduce several new degrees of freedom that significantly increase the computational burden of extreme load prediction. In an attempt to elaborate on the costs and benefits of such an approach, load calculations were done on the DTU 10 MW reference turbine where a single uniform gust shape was given various spatial dimensions with the transverse wavelength ranging up to twice the rotor diameter (357 m). The resulting loads displayed a very high spread, but remained well under the level of a uniform gust. Moving towards spatially constrained gust models would therefore yield far less conservative, though more realistic predictions at the cost of higher computation time.

Evaluating strong, weak and hybrid formulations of the single stage capacitated warehouse location problem

Lower bound (LB) of weak formulation of single stage capacitated warehouse location problem (SSCWLP) has low strength when compared to that of the strong formulation. Hence, the overall computational time to solve the weak formulation of SSCWLP is very high. Strong formulation take lesser branch-and-bound (B&B) nodes to solve SSCWLP but at the cost of higher computational time for linear programming (LP) relaxation at each node. Thus a trade-off is required, leading to the hybrid formulation, containing better characteristics of strong and weak formulations. Most promising constraints are utilised to reduce computational effort. Extensive computational experiments show that significantly superior LB are obtained by hybrid formulation compared to those obtained from weak formulation while significantly less computational time is taken for solving the LP relaxation compared to that taken by strong formulation. The gain in reduction of computational time using reformulation outweighs the cost of expanding the earlier formulation.

Selective vehicle routing problems under uncertainty without recourse

We argue that the selective vehicle routing problem is more appropriate than the conventional VRP in handling uncertainty with limited resources. However, previous formulations of selective VRPs have all been deterministic. Three new formulations are proposed to account for different optimization strategies under uncertain demand (or utility) level: reliable, robust, and fuzzy selective vehicle routing problems. Three parallel genetic algorithms (PGAs) and a classic genetic algorithm are developed and compared to the deterministic solution. PGAs differ based on their communication strategies and diversity in sub-populations. Results show that a PGA, wherein communication between demes, or subpopulations, occurs in every generation and does not eliminate repeated chromosomes, outperforms other algorithms at the cost of higher computation time. A faster variation of PGA is used to solve the non-convex reliable selective VRP, robust selective VRP and the large-scale fuzzy selective VRP, consisting of 200 nodes. Large scale application demonstrates the value of fuzzy selective vehicle routing problem FSVRP in humanitarian logistics.

Dynamic sensitivity analysis of long-running landslide models through basis set expansion and meta-modelling

Predicting the temporal evolution of landslides is typically supported by numerical modelling. Dynamic sensitivity analysis aims at assessing the influence of the landslide properties on the time-dependent predictions (e.g. time series of landslide displacements). Yet, two major difficulties arise: (1) Global sensitivity analysis require running the landslide model a high number of times (>1,000), which may become impracticable when the landslide model has a high computation time cost (>several hours); (2) Landslide model outputs are not scalar, but function of time, that is, they are n-dimensional vectors with n usually ranging from 100 to 1,000. In this article, I explore the use of a basis set expansion, such as principal component analysis, to reduce the output dimensionality to a few components, each of them being interpreted as a dominant mode of variation in the overall structure of the temporal evolution. The computationally intensive calculation of the Sobol’ indices for each of these components are then achieved through meta-modelling, that is, by replacing the landslide model by a “costless-to-evaluate” approximation (e.g. a projection pursuit regression model). The methodology combining “basis set expansion—meta-model—Sobol’ indices” is then applied to the Swiss La Frasse landslide to investigate the dynamic sensitivity analysis of the surface horizontal displacements to the slip surface properties during the pore pressure changes. I show how to extract information on the sensitivity of each main modes of temporal behaviour using a limited number (a few tens) of long-running simulations.

Large-scale Impact of CO2 Storage Operations: Dealing with Computationally Intensive Simulations for Global Sensitivity Analysis

Assessing the potential impacts associated with CO2 storage operations implies using large-scale models characterized by a very large number of grid cells (>500,000) and high computation time cost (> several hours). Yet, investigating the influence of the input parameters on the model results requires multiple simulations (>1,000), which might become impracticable due to the computation burden. A meta-modelling strategy is then proposed, basically consisting in approximating the long running model by a costless-to-evaluate model, for instance a Gaussian Process, based on a very limited number of simulations (e.g., 50). This strategy is tested to investigate the sensitivity of the overpressure induced by an industrial-scale CO2 injection into a fluvial heterogeneous reservoir, to the properties of the shale formation using a 3-dimensional long running multiphase flow model (with CPU time > 5 days).

Meta-learning for evolutionary parameter optimization of classifiers

The performance of most of the classification algorithms on a particular dataset is highly dependent on the learning parameters used for training them. Different approaches like grid search or genetic algorithms are frequently employed to find suitable parameter values for a given dataset. Grid search has the advantage of finding more accurate solutions in general at the cost of higher computation time. Genetic algorithms, on the other hand, are able to find good solutions in less time, but the accuracy of these solutions is usually lower than those of grid search. This paper uses ideas from meta-learning and case-based reasoning to provide good starting points to the genetic algorithm. The presented approach reaches the accuracy of grid search at a significantly lower computational cost. We performed extensive experiments for optimizing learning parameters of the Support Vector Machine (SVM) and the Random Forest classifiers on over 100 datasets from UCI and StatLib repositories. For the SVM classifier, grid search achieved an average accuracy of 81 % and took six hours for training, whereas the standard genetic algorithm obtained 74 % accuracy in close to one hour of training. Our method was able to achieve an average accuracy of 81 % in only about 45 minutes. Similar results were achieved for the Random Forest classifier. Besides a standard genetic algorithm, we also compared the presented method with three state-of-the-art optimization algorithms: Generating Set Search, Dividing Rectangles, and the Covariance Matrix Adaptation Evolution Strategy. Experimental results show that our method achieved the highest average accuracy for both classifiers. Our approach can be particularly useful when training classifiers on large datasets where grid search is not feasible.