A benchmark data set for two-phase Coriolis metering

For more than a decade there has been growing interest in the use of Coriolis mass flow metering applied to two-phase (gas/liquid) and multiphase (oil/water/gas) conditions. It is well-established that the mass flow and density measurements generated from multiphase flows are subject to large errors, and a variety of physical models and correction techniques have been proposed to explain and/or to compensate for these errors. One difficulty is the absence of a common basis for comparing correction techniques, because different flowtube designs and configurations, as well as liquid and gas properties, may result in quite different error curves. Furthermore, some researchers with interests in the modelling aspects of the field may not have suitable multiphase laboratory facilities to generate their own data sets. This paper offers a small data set that may be used by researchers as a benchmark i.e. a common data set for comparing correction techniques. The data set was collected at the UK National Flow Laboratory TUV-NEL, using air and a viscous oil, and provides experimental points over a wide flow range (8:1 turndown) and with Gas Volume Fraction (GVF) values up to 60%. As a first investigation using the benchmark data set, we consider how data sparsity (i.e. the flow rate and GVF spacing in the experimental grid) affects the accuracy of a correction model. A range of neural network models are evaluated, based on different subsets of the benchmark data set. The data set and some exemplary code are provided with the paper. Additional data sets are available on a web site created to support this initiative.     

基于SVM多元非线性回归的微波谐振腔谷物含水率测量法

使用微波谐振腔对物料含水率测量过程中，减少谐振参量与含水率多元非线性回归过程的误差是影响测量精度的主要因素。针对这一问题，建立了一种基于支持向量机多元非线性回归模型，并确定了其中谐振频率、品质因数和环境温度的特征值、贡献率。应用SVM—KM对该模型进行实验研究，利用50组数据对模型进行训练并验证其学习性能，利用另外15组数据验证其泛化能力。实验表明，该方法能够实现微波谐振腔物料含水率的软测量，且小样本条件下比神经元网络具有优势。对SVM多元非线性回归泛化性能进行测试，其均方根相对误差为1．06％，平均绝对相对误差为0．96％，最大绝对相对误差为1．16％。     

Predicting oil flow rate through orifice plate with robust machine learning algorithms

Measuring fluid flow rate passing through pipelines is a basic strategy for developing the infrastructure of fluid-dependent industries. It is a challenging issue for trade, transportation, and reservoir management purposes. Predicting the flow rate of fluid is also regarded as one of the crucial steps for the development of oil fields. In this study, a novel deep machine learning model, convolutional neural network (CNN), was developed to predict oil flow rate through orifice plate (Qo) from seven input variables, including fluid temperature (T_f), upstream pressure (Pu), root differential pressure (√ΔP), percentage of base sediment and water (BS&W%), oil specific gravity (SG), kinematic viscosity (ν), and beta ratio (β, the ratio of pipe diameter to orifice diameter). Due to the absence of accurate and credible methods for determining Qo, deep learning can be a useful alternative to traditional machine learning methods. Justifying the promising performance of the developed CNN model over conventional machine learning models, three different machine learning algorithms, including radial basis function (RBF), least absolute shrinkage and selection operator (LASSO), and support vector machine (SVM), were also developed and their prediction performance was compared with that of the CNN model. A sensitivity analysis was also performed on the influence degree of each input variable on the output variable (Qo). The study outcomes indicate that the CNN model provided the highest Qo prediction accuracy among all the four models developed by presenting a root mean squared error (RMSE) of 0.0341 m3/s and a coefficient of determination (R²) of 0.9999, when applied to the dataset of 3303 data records compiled from oil fields around Iran. The Spearman correlation coefficient analysis results display that √ΔP, Pu, and Tf were the most influential variables on the oil flow rate in respect of the large dataset evaluated.     

Time series modeling of endocardial border motion in ultrasonic images comparing support vector machines, multilayer perceptrons and linear estimation techniques

This paper presents a novel study of endocardial boundary motion tracking from sequences of echocardiogram images using neural network and linear estimation techniques. Contrary to the majority of previous studies theoretically analyzing endocardial motion physical parameters, a time series modeling approach is herein adopted. Such a modeling approach is demonstrated, by extensive experimentation, to be very efficient in terms of endocardial border motion tracking performance. The tracking performance of the different modeling techniques involved is evaluated quantitatively by defining suitable error measures as well as qualitatively. A thorough experimental investigation shows the importance of the highly correlated nature of endocardial contour motion within a cardiac cycle. Moreover, it shows that its short term dynamics can be almost equally well captured by support vector machines (SVM) for non-linear regression, multilayer perceptrons (MLP) and two matrix-parameters vector autoregressive (VAR) process models. Longer term dynamics, however, can be described more effectively using SVM for non-linear regression rather than MLPs. Additionally, the latter is shown to describe more effectively longer term dynamics than the VAR modeling approach. Such results are important for modeling the endocardial motion process, aiming at introducing improved adaptive focusing of ultrasonic scanners in order to enhance the quality of heart ultrasonic images.     

Research on calculation method of free flow discharge based on artificial neural network and regression analysis

Taking the Huaidian Sluice on the Shaying River in China as an example, this paper establishes the calculation model of the free flow based on artificial neural network and regression analysis. Four forms of discharge coefficient calculation equations were obtained by regression analysis, and three neural network models were established. The model is fully verified by using the measured data. The experimental results show that the third-order polynomial and multilayer perceptron neural network have better adaptability. The advantages and disadvantages of the different methods are analyzed and the cause of the error is identified. It provides a theoretical basis for dealing with the discharge calculation of small and medium dam.     

Modeling the flow rate of dry part in the wet gas mixture using decision tree/kernel/non-parametric regression-based soft-computing techniques

Owing to its importance in extraction of natural gas from underground gas storage as well as its crucial role in determination of final gas mixture in the production facilities of gas/oil industry, the dry content of wet gas mixture needs to be calculated precisely. The present study explores the potential of different soft-computing techniques in estimation of the dry gas flow rate (kg/h) (output variable) of wet gas mixture based on two input variables of wet gas flow rate (kg/h) and absolute gas humidity (g/m³). Decision tree-based methods (M5P tree, random forest (RF), random tree (RT), and reduced error pruning tree (REPT) models), kernel function-based approaches (Gaussian process regression (GPR) and support vector machines (SVM)), and non-parametric regression-based technique (multivariate adaptive regression splines (MARS)) were implemented for the first time to estimate the dry gas flow rate, and their respective prediction performances were analyzed statistically. Coefficient of correlation (CC), Nash–Sutcliffe efficiency (NSE), root mean square error (RMSE), mean absolute error (MAE), Legates and McCabe's index (LMI), and Willmott's Index (WI) were used as the statistical indicators for validating the performance of each soft-computing model. While M5P model (MAE = 122.2382 kg/h, RMSE = 580.5626 kg/h, CC = 0.9875 for the testing data set) was better than other tree-based models (MAE = 363.2802–542.6119 kg/h, RMSE = 871.9363–1025.3444 kg/h, CC = 0.9587–0.9706 for the testing data set) and MARS model (MAE = 128.0083 kg/h, RMSE = 622.9515 kg/h, CC = 0.9852 for the testing data set), the statistical indicators approved the superiority of the radial basis kernel function-based GPR model (GPR-RBKF) model (MAE = 163.3266 kg/h, RMSE = 483.1359 kg/h, CC = 0.9915 for the testing data set) over other implemented models in predicting the dry gas flow rate. The findings highlighted the potential of soft-computing methodologies in precise estimation of dry gas flow rate in wet gas mixture, particularly, in situations where the measurement of such parameters with traditional deterministic models is practically not possible.     

Intelligent fault diagnosis of rolling element bearing based on SVMs and fractal dimension

The development of non-linear dynamic theory brought a new method for recognising and predicting the complex non-linear dynamic behaviour. Fractal dimension can quantitatively describe the non-linear behaviour of vibration signal. In the present paper, the capacity dimension, information dimension and correlation dimension are applied to classify various fault types and evaluate various fault conditions of rolling element bearing, and the classification performance of each fractal dimension and their combinations are evaluated by using SVMs. Experiments on 10 fault data sets showed that the classification performance of the single fractal dimension is quite poor on most data sets, and for a given data set, each fractal dimension exhibited different classification ability, this indicates that various fractal dimensions contain various fault information. Experiments on different combinations of the fractal dimensions demonstrated that the combination of all these three fractal dimensions gets the highest score, but the classification performance is still poor on some data sets. In order to improve the classification performance of the SVM further, 11 time-domain statistical features are introduced to train the SVM together with three fractal dimensions, and the classification performance of the SVM is improved significantly. At the same time, experimental results showed that the classification performance of the SVM trained with 11 time-domain statistical features in tandem with three fractal dimensions outperforms that of the SVM trained only with 11 time-domain statistical features or with three fractal dimensions.     

Discharge prediction for rectangular sharp-crested weirs by machine learning techniques

The stage-discharge relationship of a weir is essential for posteriori calculations of flow discharges. Conventionally, it is determined by regression methods, which is time-consuming and may subject to limited prediction accuracy. To provide a better estimate, the machine learning models, artificial neural network (ANN), support vector machine (SVM) and extreme learning machine (ELM), are assessed for the prediction of discharges of rectangular sharp-crested weirs. A large number of experimental data sets are adopted to develop and calibrate these models. Different input scenarios and data management strategies are employed to optimize the models, for which performance is evaluated in the light of statistical criteria. The results show that all three models are capable of predicting the discharge coefficient with high accuracy, but the SVM exhibits somewhat better performance. Its maximum and mean relative error are respectively 5.44 and 0.99%, and 99% of the predicted data show an error below 5%. The coefficient of determination and root mean square error are 0.95 and 0.01, respectively. The model sensitivity is examined, indicative of the dominant roles of weir Reynolds number and contraction ratio in discharge estimation. The existing empirical formulas are assessed and compared against the machine learning models. It is found that the relationship proposed by Vatankhah exhibits the highest accuracy. However, it is still less accurate than the machine learning approaches. The study is intended to provide reference for discharge determination of overflow structures including spillways.     

Support-vector-regression-based prediction of water holdup in horizontal oil-water flow by using a bicircular conductance probe array

This paper presents a water holdup prediction method based on support vector regression (SVR) for horizontal oil-water two-phase flow when using a bicircular conductance probe array that consists of 24 conductance probes. The support vector machine (SVM) was employed to establish a nonlinear SVR model mapping the probe array responses into water holdup directly. Experiments were carried out in the 16 m long and 125 mm inner diameter horizontal pipe of an industrial scale experimental setup. The experimental data obtained under 220 flow conditions were first divided into modeling data set and comparing data set. The modeling data set is used for establishing a nonlinear SVR and a linear least squares regression (LSR) models, while the comparing data set is used for comparing both models with the equi-weight and optimal weight estimate methods. Comparison results obtained by using the comparing data set show that when the binary data of the probes’ responses are used only, the measurement accuracy of the optimal weight estimate method is the best. If the analog data can be obtained, the measurement accuracy of both regression methods are better than those of both weighting estimate methods, especially, the nonlinear SVR method provide the best measurement accuracy.     

Performance prediction of high-pressure coolant assisted turning of Ti-6Al-4V

The present study focuses on the development of predictive models of average surface roughness, chip-tool interface temperature, chip reduction coefficient, and average tool flank wear in turning of Ti-6Al-4V alloy. The cutting speed, feed rate, cutting conditions (dry and high-pressure coolant), and turning forces (cutting force and feed force) were the input variables in modeling the first three quality parameters, while in modeling tool wear, the machining time was the only variable. Notably, the machining environment influences the machining performance; yet, very few models exist wherein this variable was considered as input. Herein, soft computing-based modeling techniques such as artificial neural network (ANN) and support vector machines (SVM) were explored for roughness, temperature, and chip coefficient. The prediction capability of the formulated models was compared based on the lowest mean absolute percentage error. For surface roughness and cutting temperature, the ANN and, for chip reduction coefficient, the SVM revealed the lowest error, hence recommended. In addition, empirical models were constructed by using the experimental data of tool wear. The adequacy and good fit of tool wear models were justified by a coefficient of determination value greater than 0.99.     

Prediction of oil rates using Machine Learning for high gas oil ratio and water cut reservoirs

Allocated well oil rates are essential well performance evaluation. Flow meters are not reliable at high gas-oil ratio (GOR) and high water-cut (WC). Most of the available formulas are based on Gilbert-type formulas with neglecting the differential pressure across the choke. Adaptive network-based fuzzy inference system (ANFIS), and functional networks (FN) were used to generate different models to predict the production rates for high GOR and WC wells. A set of data (550 wells) was obtained from oil fields in the Middle East. GOR varied from 1,000 to 9,351 scf/stb, WC ranged from 1 to 60%. Around 300 wells were flowing under critical flow conditions, while the rest were subcritical. The developed AI models were compared against the previous published formulas. For each AI method, two models were developed for subcritical flow and critical flow conditions. The average absolute percent error (AAPE) in the subcritical flow for ANFIS and FN were 1.25, and 0.95%, respectively. While in the critical flow, the AAPE values were 1.1, and 1.35% for ANFIS and FN models, respectively. All developed AI models outperform the published formulas by 34%. The findings from this study will significantly assist production engineers to predict the oil rate in real-time without adding any cost or field intervention.     

Practical applications of artificial neural networks in chemical process development

Artificial neural networks (ANN) have the ability to map non-linear relationships without a-priori information about process or system models. This significant feature allows the network to “learn” the behavior of a system by example when it may be difficult or impractical to complete a rigorous mathematical solution. Recently ANN technology has been leaving the academic arena and placed in user-friendly software packages. This paper will offer an introduction to artificial neural networks and present a case history of two problems in chemical process development that were approached with ANN. Both optimal PID control tuning parameters and product particle size predictions were constructed from process information using neural networks. The ANN provides a rapid solution to many applications with little physical insight into the underlying system function. The amount of data preparation and performance limitations using a neural network will be discussed. However, the properly applied ANN will generally provide insight to which variables are most influential to the model and evolve dynamically to the minimum performance surface squared error. Neural networks have been used successfully with non-linear dynamic systems and can be applied to chemical process development for system identification and multivariate optimization problems.     

基于RS-SVM的TN软测量建模研究

采用粗糙集理论(RS)约简属性,在保留重要信息的前提下消除冗余信息,简化了模型结构。而支持向量机(SVM)是一种基于统计学习理论的新型学习机,本文根据TN(总氮)难于在线测量的情况,采用RS-SVM方法,用某城市污水处理厂的实际水质参数数据,建立了出水TN基于粗糙集-支持向量机的软测量模型。和未经粗糙集预处理的支持向量机模型及粗糙集-BP神经网络(RS-BPNN)模型进行了比较,选择RS-SVM模型作为最终的软测量模型。结果表明,有粗糙集预处理后,不仅测量值的误差值更小,而且大大降低了输人数据的维数,减小了模型的规模,更有利于软测量模型的实用化。同时也表明支持向量机作为建立软测量模型的工具,具有良好的性能,比神经网络更加具有优势。     

Fault diagnosis method of rolling bearing using principal component analysis and support vector machine

To effectively extract the fault feature information of rolling bearings and improve the performance of fault diagnosis, a fault diagnosis method based on principal component analysis and support vector machine was presented, and the rolling bearings signals with different fault states were collected. To address the limitation on effectively dealing with the raw vibration signals by the traditional signal processing technology based on Fourier transform, wavelet packet decomposition was employed to extract the features of bearing faults such as outer ring flaking, inner ring flaking, roller flaking and normal condition. Compared with the previous literature on fault diagnosis using principal component analysis (PCA) and support vector machine (SVM), one-to-one and one-to-many algorithms were taken into account. Additionally, the effect of four kernel functions, such as liner kernel function, polynomial kernel function, radial basis function and hyperbolic tangent kernel function, on the performance of SVM classifier was investigated, and the optimal hype-parameters of SVM classifier model were determined by genetic algorithm optimization. PCA was employed for dimension reduction, so as to reduce the computational complexity. The principal components that reached more than 95 % cumulative contribution rate were extracted by PCA and were input into SVM and BP neural network classifiers for identification. Results show that the fault feature dimensionality of the rolling bearing is reduced from 8-dimensions to 5-dimensions, which can still characterize the bearing status effectively, and the computational complexity is reduced as well. Compared with the raw feature set, PCA has a higher fault diagnosis accuracy (more than 97 %), and a shorter diagnosis time relatively. To better verify the superiority of the proposed method, SVM classification results were compared with the results of BP neural network. It is concluded that SVM classifier achieved a better performance than BP neural network classifier in terms of the classification accuracy and time-cost.     

A comparative study of the ANN and RSM modeling approaches for predicting burr size in drilling

This paper describes the comparison of the burr size predictive models based on artificial neural networks (ANN) and response surface methodology (RSM). The models were developed based on three-level full factorial design of experiments conducted on AISI 316L stainless steel work material with cutting speed, feed, and point angle as the process parameters. The ANN predictive models of burr height and burr thickness were developed using a multilayer feed forward neural network, trained using an error back propagation learning algorithm (EBPA), which is based on the generalized delta rule. The performance of the developed ANN models were compared with the second-order RSM mathematical models of burr height and thickness. The comparison clearly indicates that the ANN models provide more accurate prediction compared to the RSM models. The details of experimentation, model development, testing, and performance comparison are presented in the paper.     

势函数聚类的优化下采样SVM分类方法

为了改善大样本集下支持向量机(SVM)的训练效率和泛化性能,提出一种新算法。该算法运用采样优化和学习器优化相结合的策略,通过构建势函数对原始样本空间进行密度度量,建立了不同参数的高斯核,以实现对样本空间不同区域的逐次覆盖,并以增量学习的方式生成下采样集。然后,在所获取的下采样集上进行SVM初始训练,通过寻找原始训练集中的边界样本,进行SVM二次优化。最后,将新算法应用于人工数据集及基准数据集,结果表明,该算法在有效改善训练效率的同时,保证了分类器的泛化性能。     

Intelligent recognition of gas-oil-water three-phase flow regime and determination of volume fraction using radial basis function

The problem of how to accurately measure the flow rate of oil–gas–water mixtures in a pipeline remains one of the key challenges in the petroleum industry. This paper proposes a new methodology for identifying flow regimes and predicting volume fractions in gas-oil-water multiphase systems using dual energy fan-beam gamma-ray attenuation technique and artificial neural networks. The novelty of this study in comparison with previous works, is using just 4 extracted features (photo peaks of ²⁴¹Am and ¹³⁷Cs in 2 detectors) from the gamma ray spectrums instead of using the whole gamma ray spectrum, which reduces the undesired noises and also improves the speed of recognition in real situations. Radial basis function was used for developing the neural network model in MATLAB software in order to classify the flow patterns (annular, stratified and homogenous) and predict the value of volume fractions. The ideal and static theoretical models for flow regimes have been developed using MCNP-X code. The proposed networks could correctly recognize all the three different flow regimes and also determine volume fractions with mean absolute error of less than 5.68% according to the recognized regime.     

Modeling subcritical multi-phase flow through surface chokes with new production parameters

The produced hydrocarbons from underground reservoirs must eventually pass through surface chokes installed to control the surface flow rate at an optimum value, which should regularly be checked against the recommendations of the production engineers to prevent problems such as water coning. Accurate prediction of the surface flow rate is, therefore, crucial as it will lead to fulfilling the development plan goals of the reservoir and production optimization. In this regard, many correlations have been developed to predict the flow rate through surface choke and most of them being developed from only one dataset gathered from a single reservoir, hence with limited prediction capability and high error. Furthermore, these correlations predict the oil flow rate only as a function of wellhead pressure, gas-oil ratio, and choke size. In this study, two machine learning techniques are used to develop models for better prediction of the multi-phase flow rate for the oil wells using two new parameters of basic sediment and water (BS&W) and fluid temperature which were overlooked previously. A total of 182 production tests were utilized in developing these models which are covering a wide range of data. Graphical and statistical approaches are utilized to compare the forecasted values against the field data. Furthermore, absolute error is used as a statistical approach to assess the developed models based on machine learning in comparison to conventional correlations available in the published literature. The findings illustrate that an acceptable relation exists between the field data and predicted values with coefficients of determination equal to 0.9840 and 0.9706 for artificial neural network (ANN) and least squares support vector machine coupled simulated annealing (LSSVM-CSA), respectively, based on total datapoints. The results from this study will greatly assist petroleum engineers to have particular estimations of liquid flow rates from wellhead chokes.     

Prediction of contraction scour using ANN and GA

This paper emphasizes on the application of soft computing tools such as artificial neural network (ANN) and genetic algorithm (GA) in the prediction of scour depth within channel contractions. The experimental data of earlier investigators are used in developing the models and ANN and GA Toolboxes of MATLAB software are utilized for the purpose. The multilayered perceptron (MLP) neural networks with feed-forward back-propagation training algorithms were designed to predict the scour depth. The mean squared error and correlation coefficient are used to check the performance of networks. It is found that the ANN architecture 4-16-1 having trained with Levenberg-Marquardt ‘trainlm’ function had best performance having mean squared error of 0.001 and correlation coefficient of 0.998. In addition, the suitability of ‘trainlm’ method over other training methods is also discussed. The scour depths predicted by ANN model were compared with those computed by the two analytical models (with and without sidewall correction for contracted zone) and an empirical model proposed by Dey and Raikar [1]. In addition, heuristic search technique called genetic algorithm is used to develop the predictor for maximum scour depth within channel contraction. The population size for GA was 500 members with total generations of 1000, crossover fraction of 0.8 and Gaussian operator for mutation. It is promising to observe that the GA model predicts the maximum scour depth equally well as that of empirical model of Dey and Raikar [1]. Hence, both ANN and GA models can be satisfactorily used to predict the scour depth within channel contractions.     

一种新的机电设备状态趋势智能混合预测模型

针对机电设备运行状态受多因素影响，变化趋势复杂，难以用单一预测方法进行有效预测的问题，提出一种新的基于改进灰色系统一支持向量机一神经模糊系统的智能混合预测模型。该模型首先利用改进灰色系统弱化数据序列波动性、支持向量机处理小样本和模糊神经系统处理非线性模糊信息的优点，分别进行趋势预测，然后通过改进遗传算法对这三者的预测结果进行自适应加权组合。将该模型应用于信号随机波动性较强、趋势变化复杂的标准算例和某机组振动趋势的预测中，研究结果表明，该模型的预测性能均优于上述三种单一预测方法。