Multiphase Flow Facility Research Articles

Comparative Study of Air–Water and Air–Oil Frictional Pressure Drops in Horizontal Pipe Flow

Experimental data for frictional pressure drop using both air–water and air–oil mixtures are reported, compared and used to evaluate predictive methods. The data were gathered using the 2-inch (54.8 mm) flow loop of the multiphase flow facility at the National University of Singapore. Experiments were carried out over a wide range of flow conditions of superficial liquid and gas velocities that were varied from 0.05 to 1.5 m/s and 2 to 23 m/s, respectively. Pressure drops were measured using pressure transducers and a differential pressure (DP) cell. A hitherto unreported finding was achieved, as the pressure drop in air–oil flow can be lower than that in air–water flow for the higher range of flow conditions. Using flow visualization to explain this phenomenon, it was found that it is related to the higher liquid holdup that occurs in the case of air–oil around the annular flow transition and the resulting interfacial friction. This additional key finding can have applications in flow assurance to improve the efficiency of oil and gas transportation in pipelines. Models and correlations from the open literature were tested against the present data.

OWFD-UCPM: An open-world fault diagnosis scheme based on uncertainty calibration and prototype management

In real-world fault diagnosis tasks, it is never possible to enumerate all fault types beforehand since there are always unseen situations that may arise unexpectedly. Therefore, fault diagnosis is essentially an open-world recognition task. A desirable open-world fault diagnosis model must be able to perform open-set recognition (OSR) and incremental learning (IL). Classifier calibration and prototype replay are popular in two subtasks. However, existing calibration and replay strategies suffer from tampering with raw data and ignoring resource allocation, respectively. In this work, we formulate an open-world fault diagnosis scheme to address both issues. Firstly, we calibrate the uncertainty of the classifier’s prediction by assigning soft labels to the samples based on their distance from the class center. Then, Shannon entropy is employed to quantify the uncertainty as an estimate of the probability that the sample belongs to the unknown. Secondly, we adaptively manage the memory of old and new known classes according to the training accuracy. Then, based on this memory budget, a dissimilarity-based sparse subset selection algorithm is utilized to pick more diverse exemplars as prototypes for replay. Furthermore, we extend the open-set F-Measure (OSFM) to the weighted OSFM to make it applicable even when the classes are imbalanced. Experimental results on multiphase flow facility and wastewater treatment plant demonstrate the feasibility and superiority of the proposed method.

Deep Probabilistic Principal Component Analysis for Process Monitoring.

Probabilistic latent variable models (PLVMs), such as probabilistic principal component analysis (PPCA), are widely employed in process monitoring and fault detection of industrial processes. This article proposes a novel deep PPCA (DePPCA) model, which has the advantages of both probabilistic modeling and deep learning. The construction of DePPCA includes a greedy layer-wise pretraining phase and a unified end-to-end fine-tuning phase. The former establishes a hierarchical deep structure based on cascading multiple layers of the PPCA module to extract high-level features. The latter builds an end-to-end connection between the raw inputs and the final outputs to further improve the representation of the model to high-level features. After constructing the model structure of DePPCA, we first present the detailed training processes of the pretraining and fine-tuning stages, then clarify the theoretical merits of the proposed model from the perspective of variational inference. For process monitoring purposes, we develop two statistics based on the established DePPCA. The monitoring performance of these two statistics can remain superior even if the features extracted by DePPCA are significantly compressed to univariate. This makes the feature extraction process and online monitoring procedure of DePPCA quite fast. In other words, the proposed DePPCA can achieve accurate and efficient process monitoring by only extracting one feature for each sample. Finally, the effectiveness of DePPCA is evaluated on the Tennessee Eastman (TE) process and the multiphase flow (MPF) facility.

Variational Discriminative Stacked Auto-Encoder: Feature Representation Using a Prelearned Discriminator, and Its Application to Industrial Process Monitoring.

In deep-learning-based process monitoring, obtaining an effective feature representation is a critical step in constructing a reliable deep-learning monitoring model. Conventional deep-learning methods like stacked auto-encoders (SAEs) capture feature representation by minimizing the data reconstruction errors, which lack the expression of essential information and ultimately lead to degradation of the monitoring performance. To solve this problem, variational discriminative SAE (VDSAE) is proposed in this article. First, a variational generative discriminative structure is designed to obtain a reliable prelearned discriminator. Based on this new variational discriminator, the authenticity of the reconstructed data is evaluated as an important criterion for feature learning. Then, an SAE incorporating the prelearned discriminator is trained by both minimizing the reconstruction error and maximizing the data authenticity. In this way, the prelearned discriminator makes the network effectively capture the essential expression of the reconstructed data. The proposed approach enables SAE to learn a better feature representation owing to the excellent reconstruction performance. Finally, the feature representation and fault detection performance of VDSAE are verified in two cases. The results show that the average fault detection rates (FDRs) of the multiphase flow facility and the waste-water treatment process (WWTP) can be improved to 72% and 97%, respectively, compared with the other fault detection methods.

Enhanced robust multimode process monitoring under dirty data via difference-based decomposition of matrix

Traditional data-driven methods generally suppose the training dataset is not corrupted by outliers. However, outliers are inevitable in the real industrial processes even with a relatively high ratio, which degrades the accuracy of data-based models. For multimode process monitoring, outliers may deteriorate the accuracy of both mode identification and fault detection. However, the existing robust methods can hardly deal with a large percentage of outliers, i.e., dirty data in neither single nor multimode processes. In this paper, a robust multimode process monitoring scheme is developed by alleviating the negative effect of dirty data. A difference-based decomposition of matrix (DDM) algorithm is first proposed to divide the data into a basic subpart and a non-basic subpart. The optimization function of the proposed DDM algorithm is solved by the alternating direction method of multipliers (ADMM). In the off-line procedure, an iterative decomposition strategy is designed based on the proposed DDM approach to identify dirty data and obtain the clean dataset of each mode. In the on-line procedure, an on-line sample identification strategy is developed by the type indicator derived from the DDM algorithm to determine whether the current sample belongs to a mode, the dirty data, or the fault. A numerical example and an industrial-scale multiphase flow facility indicate the proposed method can improve the accuracy of both mode identification and fault detection for multimode processes even under dirty data.

A Novel Data-Driven Fault Detection Method Based on Stable Kernel Representation for Dynamic Systems.

With the steady improvement of advanced manufacturing processes and big data technologies, modern industrial systems have become large-scale. To enhance the sensitivity of fault detection (FD) and overcome the drawbacks of the centralized FD framework in dynamic systems, a new data-driven FD method based on Hellinger distance and subspace techniques is proposed for dynamic systems. Specifically, the proposed approach uses only system input/output data collected via sensor networks, and the distributed residual signals can be generated directly through the stable kernel representation of the process. Based on this, each sensor node can obtain the identical residual signal and test statistic through the average consensus algorithms. In addition, this paper integrates the Hellinger distance into the residual signal analysis for improving the FD performance. Finally, the effectiveness and accuracy of the proposed method have been verified in a real multiphase flow facility.

Influence of design parameters of upstream Venturi pipeline on multiphase flow measurement

A multiphase flowmeter (MPFM) is used in the upstream oil and gas industry for continuous, in-line, real-time, oil-gas-water flow measurement without fluid separation. An MPFM typically consists of phase-fraction and velocity measurements. It is desirable to have homogeneous flow at the measurement location so that the phase-fraction measurement is representative. A horizontal blind-tee pipe-section is often installed to homogenize flow in the downstream vertical Venturi-based flowmeters; however, little information is available on the effect of horizontal blind-tee depth (HBD) on flow homogeneity. In addition, the Venturi vertical entrance length (VEL) leading to the Venturi inlet from the horizontal blind-tee outlet is another design parameter that may potentially affect the downstream phase distribution. The phase-fraction measurement principle requires liquid properties. The local liquid richness makes the horizontal blind-tee an ideal location for measuring liquid properties; however, an excessive HBD may affect the reliability of the measurements of liquid properties, because local vortices may degrade liquid measurement representativeness if the local liquid velocity is too low. This study uses a computational fluid dynamics approach to evaluate the effect of HBD and VEL on multiphase flow measurement. The computational results are validated with experimental data collected in a multiphase flow facility.

Development of Gas-Liquid Flow Regimes Identification Using a Noninvasive Ultrasonic Sensor, Belt-Shape Features, and Convolutional Neural Network in an S-Shaped Riser.

The problem of classifying gas-liquid two-phase flow regimes from ultrasonic signals is considered. A new method, belt-shaped features (BSFs), is proposed for performing feature extraction on the preprocessed data. A convolutional neural network (CNN/ConvNet)-based classifier is then applied to categorize into one of the four flow regimes: 1) annular; 2) churn; 3) slug; or 4) bubbly. The proposed ConvNet classifier includes multiple stages of convolution and pooling layers, which both decrease the dimension and learn the classification features. Using experimental data collected from an industrial-scale multiphase flow facility, the proposed ConvNet classifier achieved 97.40%, 94.57%, and 94.94% accuracy, respectively, for the training set, testing set, and validation set. These results demonstrate the applicability of the BSF features and the ConvNet classifier for flow regime classification in industrial applications.

Change Point Detection for Process Data Analytics Applied to a Multiphase Flow Facility

Change point detection becomes increasingly important because it can support data analysis by providing labels to the data in an unsupervised manner. In the context of process data analytics, change points in the time series of process variables may have an important indication about the process operation. For example, in a batch process, the change points can correspond to the operations and phases defined by the batch recipe. Hence identifying change points can assist labelling the time series data. Various unsupervised algorithms have been developed for change point detection, including the optimisation approach which minimises a cost function with certain penalties to search for the change points. The Bayesian approach is another, which uses Bayesian statistics to calculate the posterior probability of a specific sample being a change point. The paper investigates how the two approaches for change point detection can be applied to process data analytics. In addition, a new type of cost function using Tikhonov regularisation is proposed for the optimisation approach to reduce irrelevant change points caused by randomness in the data. The novelty lies in using regularisation-based cost functions to handle ill-posed problems of noisy data. The results demonstrate that change point detection is useful for process data analytics because change points can produce data segments corresponding to different operating modes or varying conditions, which will be useful for other machine learning tasks.

Hierarchical Latent Variable Extraction and Multisegment Probability Density Analysis Method for Incipient Fault Detection

The incipient fault is difficult to detect because of its small amplitude and insignificant impact, however, ignoring such fault may cause irreversible damage to the system. In this article, a hierarchical latent variable extraction and multisegment probability density analysis method is proposed to detect the incipient fault. First, three data subspaces are constructed, which are named dominant, intermediate, and residual spaces, and key latent variables which contain more offline variance or online variation information will be retained. Afterward, the expanded data distribution interval and multiple data segmentsare constructed for the probability density estimation. Based on the improved symmetric divergence index, the distribution distance between the online data and offline modeling data can be evaluated, which has achieved 95.3% and 86.8% average detection rates for the faults in numerical case and Tennessee Eastman process. Finally, a real multiphase flow facility is used to demonstrate the effectiveness of the proposed method.

Classification of flow regimes using a neural network and a non-invasive ultrasonic sensor in an S-shaped pipeline-riser system

A method for classifying flow regimes is proposed that employs a neural network with inputs of extracted features from Doppler ultrasonic signals of flows using either the Discrete Wavelet Transform (DWT) or the Power Spectral Density (PSD). The flow regimes are classified into four types: annular, churn, slug, and bubbly flow regimes. The neural network used in this work is a feedforward network with 20 hidden neurons. The network comprises four output neurons, each of which corresponds to the target vector's element number. 13 and 40 inputs are used for features extracted from PSD and DWT respectively. Experimental data were collected from an industrial-scale multiphase flow facility. Using the PSD features, the neural network classifier misclassified 3 out of 31 test datasets in the classification and gave 90.3% accuracy, while only one dataset was misclassified with the DWT features, yielding an accuracy of 95.8%, thus showing the superiority of the DWT in feature extraction of flow regime classification. The approach demonstrates the applicability of a neural network and DWT for flow regime classification in industrial applications using a clamp-on Doppler ultrasonic sensor. The scheme has significant advantages over other techniques as only a non-radioactive and non-intrusive sensor is used. To the best of our knowledge, this is the first known successful attempt for the classification of liquid-gas flow regimes in an S-shape riser system using an ultrasonic sensor, PSD-DWTs features, and a neural network.

Multiphase flowrate measurement with time series sensing data and sequential model

Accurate multiphase flowrate measurement is challenging but vital in the energy industry to monitor the production process. Machine learning has recently emerged as a promising method for estimating multiphase flowrates based on different conventional flow meters. In this paper, we propose a Convolutional Neural Network (CNN)-Long-Short Term Memory (LSTM) model and a Temporal Convolutional Network (TCN) model to estimate the volumetric liquid flowrate of oil/gas/water three-phase flow based on the Venturi tube. The volumetric flowrates of the liquid and gas phase vary from 0.1–10 m3/h and 7.6137–86.7506 m3/h, respectively. We collected time series sensing data from a Venturi tube installed in a pilot-scale multiphase flow facility and utilized single-phase flowmeters to acquire reference data before mixing. Experimental results suggest that the proposed CNN-LSTM and TCN models can effectively deal with the time series sensing data from the Venturi tube and achieve a good accuracy of multiphase flowrate estimation under different flow conditions. TCN achieves a better accuracy for both liquid and phase flowrate estimation than CNN-LSTM. The results indicate the possibility of leveraging conventional flow meters for multiphase flowrate estimation under various flow conditions.

Anomaly Detection and Mode Identification in Multimode Processes Using the Field Kalman Filter

A process plant can have multiple modes of operation due to varying demand, availability of resources, or the fundamental design of a process. Each of these modes is considered as normal operation. Anomalies in the process are characterized as deviations away from normal operation. Such anomalies can be indicative of developing faults which, if left unresolved, can lead to failures and unplanned downtime. The field Kalman filter (FKF) is a model-based approach, which is adopted in this article for monitoring a multimode process. Previously, the FKF has been applied in process monitoring to differentiate normal operation from known faulty modes of operation. This article extends the FKF so that it may detect occurrences of anomalies and differentiate them from the various normal modes of operation. A method is proposed for off-line training an FKF monitoring model and online monitoring. The off-line part comprises training an FKF model based on multivariate autoregressive state-space (MARSS) models fitted to historical process data. A monitoring indicator is also introduced. Online monitoring, based on the FKF for anomaly detection and mode identification, is demonstrated using a simulated multimode process. The performance of the proposed method is also demonstrated using data obtained from a pilot-scale multiphase flow facility. The results show that the method can be applied successfully for anomaly detection and mode identification.

Nonstationary Discrete Convolution Kernel for Multimodal Process Monitoring.

Data-driven process monitoring has benefited from the development and application of kernel transformations, especially when various types of nonlinearity exist in the data. However, when dealing with the multimodality behavior that is frequently observed in the process operations, the most widely used radial basis function (RBF) kernel has limitations in describing process data collected from multiple normal operating modes. In this article, we highlight this limitation via a synthesized example. In order to account for the multimodality behavior and improve the fault detection performance accordingly, we propose a novel nonstationary discrete convolution kernel, which derives from the convolution kernel structure, as an alternative to the RBF kernel. By assuming the training samples to be the support of the discrete convolution, this new kernel can properly address these training samples from different operating modes with diverse properties and, therefore, can improve the data description and fault detection performance. Its performance is compared with RBF kernels under a standard kernel principal component analysis framework and with other methods proposed for multimode process monitoring via numerical examples. Moreover, a benchmark data set collected from a pilot-scale multiphase flow facility is used to demonstrate the advantages of the new kernel when applied to an experimental data set.

Fault detection and identification combining process measurements and statistical alarms

Classification-based methods for fault detection and identification can be difficult to implement in industrial systems where process measurements are subject to noise and to variability from one fault occurrence to another. This paper uses statistical alarms generated from process measurements to improve the robustness of the fault detection and identification on an industrial process. Two levels of alarms are defined according to the position of the alarm threshold: level-1 alarms (low severity threshold) and level-2 alarms (high severity threshold). Relevant variables are selected using the minimal-Redundancy-Maximal-Relevance criterion of level-2 alarms to only retain variables with large variations relative to the level of noise. The classification-based fault detection and identification fuses the results of a discrete Bayesian classifier on level-1 alarms and of a continuous Bayesian classifier on process measurements. The discrete classifier offers a practical way to deal with noise during the development of the fault, and the continuous classifier ensures a correct classification during later stages of the fault. The method is demonstrated on a multiphase flow facility.

A slow independent component analysis algorithm for time series feature extraction with the concurrent consideration of high-order statistic and slowness

For the process data analytics, numerous statistical methods are designed to extract informative features to reveal latent characteristics and correlation patterns among the variables. As a classic statistical method, independent component analysis extracts mutually independent components by searching for the maximum direction of non-Gaussianity from the mixed signals. However, the algorithm does not work for the latent Gaussian components, and the disorder of extracted features also brings trouble to the practical applications. In the present work, a slow independent component analysis algorithm is proposed to tackle the problems by concurrently considering the high-order statistic and slowness. It constructs a dual-objective optimization criterion function, simultaneously incorporating independence and slow varying characteristics into the feature extraction procedure, which ties together the ideas of independent component analysis and slow feature analysis under the same mathematical umbrella. Through adjusting the controllable sub-optimization objective weight, the proposed slow independent component analysis adds insight into the different roles of independence and slow varying characteristics in calibration modeling, and, thus, provides possibilities to combine the advantages of independent component analysis and slow feature analysis. Compared with the traditional method, the new algorithm no longer relies solely on higher-order statistic but also utilizes the slow varying characteristic to extract latent Gaussian or non-Gaussian components from mixed signals. Besides the unsupervised paradigm, the proposed algorithm is extended to the supervised form for classification task by utilizing the temporal structure as well as the label information. The effectiveness of the proposed algorithm is illustrated by blind source separation task and industrial fault diagnosis task using real-world datasets including speech signals and multiphase flow facility.

A heterogeneous benchmark dataset for data analytics: Multiphase flow facility case study

Improvements in sensing, connectivity and computing technologies mean that industrial processes now generate data from a variety of disparate sources. Data may take a number of forms, from time-domain signals, sampled at various rates using a variety of sensors, to alarm and event logs. Novel techniques need to be developed to tackle the challenges of heterogeneous data. Testing such algorithms requires benchmark datasets that allow direct comparison of the performance of the methods. This work presents the PRONTO heterogeneous benchmark dataset. Experiments were conducted on a multiphase flow facility under various operational conditions with and without induced faults. Data were collected from heterogeneous sources, including process measurements, alarm records, high frequency ultrasonic flow and pressure measurements. The presented dataset is suitable for developing and validating algorithms for fault detection and diagnosis and data fusion concepts. Three algorithms are tested using the dataset, illustrating the applicability of the dataset.

Variable Selection for Fault Detection and Identification based on Mutual Information of Alarm Series

Abstract Reducing the dimensionality of a fault detection and identification problem is often a necessity, and variable selection is a practical way to do it. Methods based on mutual information have been successful in that regard, but their applicability to industrial processes is limited by characteristics of the process variables such as their variability across fault occurrences. The paper introduces a new estimation strategy of mutual information criteria using alarm series to improve the robustness of the variable selection. The minimal-redundancy-maximal-relevance criterion on alarm series is suggested as new reference criterion, and the results are validated on a multiphase flow facility.

Investigation of the diagnostic properties of sensors and features in a multiphase flow facility case study

Abstract Feature design and selection is one of the first steps towards successful fault detection and diagnosis. Data from different sources can contain complimentary information about a monitored system. Hence methods which fuse features from multiple sources can often detect and diagnose a greater number of fault modes with higher confidence. However, solutions that require data from multiple sensors as inputs can be susceptible to failure if one or more of those sensors cease to function. Optimally a solution will fuse data from a sufficient number of sensors so that the advantages of sensor fusion are realized, while the robustness of the system is retained. In this paper the authors investigate how the best subset of features might differ for fault detection and fault severity diagnosis in a multiphase flow facility case study. ReliefF, which is a K-nearest neighbors-based feature selection filter, is used to rank the features for different problems. The dataset used for the analysis contains data from various operating conditions and induced faults with various severities. It was found that the optimal subset of features varied for different monitoring problems. It was also shown that including features that are ranked as being uninformative into a fault classifier can also impact the robustness of the classifier to sensor failures.

Multi-layer Contribution Propagation Analysis for Fault Diagnosis

The recent development of feature extraction algorithms with multiple layers in machine learning and pattern recognition has inspired many applications in multivariate statistical process monitoring. In this work, two existing multilayer linear approaches in fault detection are reviewed and a new one with extra layer is proposed in analogy. To provide a general framework for fault diagnosis in succession, this work also proposes the contribution propagation analysis which extends the original definition of contribution of variables in multivariate statistical process monitoring. In fault diagnosis stage, the proposed contribution propagation analysis for multilayer linear feature extraction algorithms is compared with the fault diagnosis results of original contribution plots associated with single layer feature extraction approach. Plots of variable contributions obtained by the aforementioned approaches on the data sets collected from a simulated benchmark case study (Tennessee Eastman process) as well as an industrial scale multiphase flow facility are presented as a demonstration of the usage and performance of the contribution propagation analysis on multilayer linear algorithms.