Search

A multivariate statistical process monitoring scheme should be able to describe multimodal data. Multimodality typically arises in process data due to varying production regimes. Moreover, multimodality may influence how easy it is for process operators to interpret the monitoring results. To address these challenges, this paper proposes an on-line monitoring framework for anomaly detection where an anomaly may either indicate a fault occurring and developing in the process or the process moving to a new operating mode. The framework incorporates the Dirichlet process, which is an unsupervised clustering method, and kernel principal component analysis with a new kernel specialized for multimode data. A monitoring model is trained using the data obtained from several healthy operating modes. When on-line, if a new healthy operating mode is confirmed by an operator, the monitoring model is updated using data collected in the new mode. Implementation issues of this framework, including the parameter tuning for the kernel and the selection of anomaly indicators, are also discussed. A bivariate numerical simulation is used to demonstrate the performance of anomaly detection of the monitoring model. The ability of this framework in model updating and anomaly detection in new operating modes is shown on data from an industrial-scale process using the PRONTO benchmark dataset. The examples will also demonstrate the industrial applicability of the proposed framework.

This paper reviews existing data-driven methods according to three criteria: the properties of the method, the impact of the systems through which the disturbance propagates on the results of the method, and the types of disturbances covered by the method. The comparative review is used to derive guidelines for Root Cause Analysis (RCA) of oscillatory and non-oscillatory disturbances in industrial processes. In particular, it suggests an approach for RCA of transient disturbances based on methods for detection of transient disturbances. The comparative review is targeting primarily engineers working on Fault Detection and Diagnosis (FDD). RCA methods provide tools to speed up the work of the process engineer troubleshooting an abnormal situation, and the review should help in selecting appropriate methods. It is intended as a tool to increase the awareness and the understanding of data-driven RCA methods and to promote their use. The focus on transient disturbances answers a need for methods to automate or facilitate RCA of transient disturbances in industrial processes or industrial machinery that was observed by the authors during discussions with service engineers.

The paper discusses operation and control of a process for the liquefaction of natural gas in which the refrigeration compressors are driven by electric motors. The aim is to enable the plant to accommodate contingencies in the availability of electrical power and to continue running when there is a shortage of electrical power, avoiding the significant economic impact of a shutdown. The article provides a detailed first principles analysis of the relationships between the electrical power consumption of the process, the production rate of the liquefied natural gas, its exit temperature, and its purity. By doing this, it is possible to ascertain settings for operating the process at various levels of power consumption. The results show that the process can operate with reductions of electrical power of 30 percent or more. Hence, power shortages could be managed by operating the process flexibly to make best use of the available remaining power, rather than by shutting down. The paper also discusses how such a system could be implemented industrially and identifies aspects that require further study.

Kernel Principal Component Analysis (KPCA) using Radial Basis Function (RBF) kernels can capture data nonlinearity by projecting the original variable space to a high-dimensional kernel feature space and obtaining the kernel principal components. This article examines the tuning of the kernel width when using RBF kernels in KPCA, showing that inappropriate kernel widths result in RBF-KPCA being unable to capture nonlinearity present in data. The paper also considers the choice of monitoring statistics when RBF-KPCA is applied to anomaly detection. Linear PCA requires two monitoring statistics. The Hotelling’s $T^{2}$ monitoring statistic detects when a sample exceeds the healthy operating range, while the Squared Prediction Error (SPE) monitoring statistic detects the case when the sample does not follow the model of the training data. The analysis in this article shows that SPE for RBF-KPCA can detect both cases. Moreover, unlike the case of linear PCA, the $T^{2}$ monitoring statistic for RBF-KPCA is non-monotonic with respect to the magnitude of the anomaly, making it not optimal as a monitoring statistic. The paper presents examples to illustrate these points. The paper also provides a detailed mathematical analysis which explains the observations from a theoretical perspective. Tuning strategies are proposed for setting the kernel width and the detection threshold of the monitoring statistic. The performance of optimally tuned RBF-KPCA for anomaly detection is demonstrated via numerical simulation and a benchmark dataset from an industrial-scale facility.

Sensor installation on water infrastructure is challenging due to requirements for service interruption, specialised personnel, regulations and reliability as well as the resultant high costs. Here, a minimally invasive installation method is introduced based on hot-tapping and immersion of a sensor probe. A modular architecture is developed that enables the use of interchangeable multi-sensor probes, non-specialist installation and servicing, low-power operation and configurable sensing and connectivity. A prototype implementation with a temperature, pressure, conductivity and flow multi-sensor probe is presented and tested on an evaluation rig. This paper demonstrates simple installation, reliable and accurate sensing capability as well as remote data acquisition. The demonstrated minimally invasive multi-sensor probes provide an opportunity for the deployment of water quality sensors that typically require immersion such as pH and spectroscopic composition analysis. This design allows dynamic deployment on existing water infrastructure with expandable sensing capability and minimal interruption, which can be key to addressing important sensing parameters such as optimal sensor network density and topology.

Normal practice in a compressor station with compressors in parallel is to allocate the mass flows equally. However, this strategy is not optimal if the compressors are not identical. A common reason why compressors become non-identical is because their performance degrades over time. Degradation increases the power necessary to run the compressor station and changes the optimal allocation of mass flows. This paper presents a framework for optimal operation in a compressor station with degrading compressors. The optimisation framework proposed in this work explicitly includes a model of degradation in the optimisation problem and analyses how the optimal load-sharing changes when the compressors are degrading. The optimisation framework was applied in an industrial case study of a compressor station in which three parallel compressors are subject to degradation. The case study confirms that it is possible to minimise the extra power consumption due to degradation by adjusting the operating conditions of the compressor station. The analysis also gives insights into the impact of degradation on the optimal solution when compressors work at their limits.

The analysis of equipment degradation has traditionally developed in two main directions. One approach, of great interest for control system design, has been to consider that degradation causes fundamental changes to the behaviour of a system. Another approach, used in optimal maintenance planning and production scheduling, considers degradation as a separate process that affects performance but does not necessarily change the behaviour. This article provides a review of mathematical models of degradation that will facilitate the integration of degradation modelling into control and optimisation schemes. To this end, a new unified classification is proposed. It takes into account the influence of degradation on the behaviour of the system, as well as the factors influencing degradation. Understanding these mutual influences will enable improved optimization, design and operation of control systems. The flexibility of the proposed classification is demonstrated in an industrial application to a multi-product batch scheduling process.

Varying production regimes and loading conditions on equipment often result in multiple operating modes in process operations. The data recorded from such processes will typically be multimodal in nature leading to challenges in applying standard data-driven process monitoring approaches. Moreover, even if a monitoring approach is able to account for the variability present in a training set comprised of historical process data, in order to be robust and reliable the method will need to account for any new operating modes which might emerge during production. Therefore, it is desirable to have a monitoring algorithm that can both handle data multimodality in off-line training and, when implemented on-line, can actively update in order to incorporate new operating modes. This paper proposes a monitoring framework which combines an unsupervised clustering approach with a kernel-based Multivariate Statistical Process Monitoring (MSPM) algorithm. A monitoring model is trained off-line and is subsequently used to detect anomalies on-line. An anomaly might be indicative of either a developing fault or a change in the process to a new operating mode. In the latter case, the monitoring model can be updated to account for the new mode whilst still being able to detect faults under this framework. The advantages of the off-line training procedure relative to a standard kernel-based method are demonstrated via a numerical simulation. Additionally, the monitoring performance in the presence of faults and the capability of updating the model in the presence of new operating modes is demonstrated using a benchmark data set from an experimental pilot plant.

Alarm systems based on engineering and safety considerations are the prime source of information for operators when it comes to abnormal situations. Conversely, the presence of fault detection and diagnosis algorithms in process plants is still limited, in comparison with other process control technologies. This work presents a simple way to integrate the information contained in the alarm systems into the fault detection and diagnosis algorithm. A normalisation of the process measurements based on the alarm thresholds is proposed, improving the robustness of the algorithm with regard to the variability of the measurements across fault occurrences in industrial systems.

Oscillations occurring in industrial process plants often reflect the presence of severe disturbances affecting process operations. Accurate detection and root-cause analysis of oscillations is of great interest for the economic viability of the process operation. Standard oscillation detection and root cause analysis methods require a large enough number of data samples. Unrelated transient changes superimposed on the oscillation pattern reduce the number of useful data samples. The present paper proposes simple heuristic methods to effectively detect and remove two types of transient changes from oscillatory signals, namely step changes and spikes. The proposed methods are used to pre-process oscillatory time series. The accuracy gained when using auto-correlation function method for oscillation detection and transfer entropy method for oscillation propagation is experimentally evaluated. The methods are carried out on a 1.3-Butadiene production process where several measurements showed an established oscillation occurring after a production level change.

The paper provides a detailed first principles analysis of a process of liquefaction of natural gas in which the refrigeration compressors are driven by electric motors. The aim is to determine and understand the impact of the operation of the refrigeration cycles on the power consumption of the process. This study gives insights into the relationship between the discharge pressure of the compressors and the thermodynamic performance of the refrigeration cycles. By doing this, the findings of this research highlight the importance of the vapour liquid equilibrium. The results show that, for a given production rate, there is a minimum in the power consumption when the discharge pressures are close to the refrigerant saturation points. The paper provides a discussion and analysis of the advantages and disadvantages of operating at such points.

Performance-based maintenance of machinery relies on detection and prediction of performance degradation. Degradation indicators calculated from process measurements need to be approximated with degradation models that smooth the variations in the measurements and give predictions of future values of the indicator. Existing models for performance degradation assume that the performance monotonically decreases with time. In consequence, the models yield suboptimal performance in performance-based maintenance as they do not take into account that performance degradation can reverse itself. For instance, deposits on the blades of a turbomachine can be self-cleaning in some conditions. In this study, a data-driven algorithm is proposed that detects if the performance degradation indicator is increasing or decreasing and adapts the model accordingly. A moving window approach is combined with adaptive regression analysis of operating data to predict the expected value of the performance degradation indicator and to quantify the uncertainty of predictions. The algorithm is tested on industrial performance degradation data from two independent offshore applications, and compared with four other approaches. The parameters of the algorithm are discussed and recommendations on the optimal choices are made. The algorithm proved to be portable and the results are promising for improving performance-based maintenance.

Alarm systems designed according to engineering and safety considerations provide the primary source of information for operators when it comes to abnormal situations. Still, alarm systems have rarely been exploited for fault detection and diagnosis. Recent work has demonstrated the benefits of alarm logs for fault detection and diagnosis. However, alarm settings conceived during the alarm design stage can also be integrated into fault detection and diagnosis methods. This paper suggests the use of those alarm settings in the preprocessing of the process measurements, proposing a normalization based on the alarm thresholds of each process variable. Normalization is needed to render process measurements dimensionless for multivariate analysis. While common normalization approaches such as standardization depend on the historical process measurements available, the proposed alarm-range normalization is based on acceptable variations of the process measurements. An industrial case study of an offshore oil gas separation plant is used to demonstrate that the alarm-range normalization improves the robustness of popular methods for fault detection, fault isolation, and fault identification.

This paper presents the control loop data of industrial controllers that are recently made available online. All data is confirmed and some of it has been published previously to develop fault detection and diagnosis methods. Methods to detect faults that occur during the operation of an industrial process are important and have attracted attention previously but are not always widely used in industry. One of the reasons is that any method needs to be robust and fully automated. The purpose of the data repository is to present data to test methods so that false positives and negatives are reduced to an insignificant number. Three previously published methods – oscillation detection based on the autocorrelation function, the idle index and a method for quantization detection – together with a simple, novel saturation detection method and one new detection methods are applied to all industrial data. The results are discussed and ways to improve the robustness and automation potential of these methods.

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.

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.

Process industries are becoming increasingly reliant on electrical power for reasons of efficiency and sustainability. A large industrial site typically has its own power management system to distribute electricity to the process and to manage electrical contingencies such as partial loss of supply. Recent work has illustrated more flexible alternatives to load shedding whereby an industrial process plant can continue to operate at a lower level making use of available electrical power. This paper presents a way for achieving such flexibility in a Liquefied Natural Gas (LNG) plant. It analyzes the consequences for production of varying the consumed power, and assesses the maximum flexibility within the feasible operating envelope of the process. The study has been conducted by modeling and simulation of an LNG plant using the Linde process with three refrigeration cycles. The results also show the relationships between electrical power consumption and production in terms of production rate and product characteristics. They also show that the vapour-liquid equilibrium plays a crucial role in establishing the operating points and setting the boundaries in which the process has to work. Thus, through the assessment and simulation of an LNG plant, this work demonstrates that flexible operation has benefits over alternatives. It achieves more operating points and therefore adds more flexibility.

A global multiscale method based on a dynamic mode decomposition (DMD) algorithm to characterize the global behavior of transient processes recorded using wide-area sensors is proposed. The method interprets global dynamic behavior in terms of both, spatial patterns or shapes and temporal patterns associated with dynamic modes containing essentially single-frequency components, from which the mode shapes, frequencies and growth and decay rates of the modes can be extracted simultaneously. These modes are then used to detect the coherent and dominant structures within the data. The technique is well suited for fast wide-area monitoring and assessment of global instability in the context of modern data fusion-based estimation techniques. Results of the application of the proposed method to large, high-dimensional data sets are encouraging.

A fault-tolerant observer design methodology is proposed. The aim is to guarantee a minimum level of closed-loop performance under all possible sensor fault combinations while optimizing performance under the nominal, fault-free condition. A novel approach is proposed to tackle the combinatorial nature of the problem, which is computationally intractable even for a moderate number of sensors, by recasting the problem as a robust performance problem, where the uncertainty set is composed of all combinations of a set of binary variables. A procedure based on an elimination lemma and an extension of a semidefinite relaxation procedure for binary variables is then used to derive sufficient conditions (necessary and sufficient in the case of one binary variable) for the solution of the problem which significantly reduces the number of matrix inequalities needed to solve the problem. The procedure is illustrated by considering a fault-tolerant observer switching scheme in which the observer outputs track the actual sensor fault condition. A numerical example from an electric power application is presented to illustrate the effectiveness of the design.

The adoption of wireless technology for industrial wireless instrumentation requires high-quality communication performance. The use of transmission power control (TPC) can help address industrial issues concerning energy consumption, interference, and fading. This paper presents a TPC algorithm designed for industrial applications based on theoretical and empirical studies. It is shown that the proposed algorithm adapts to variations in link quality, and is hardware-independent and practical.

Performance deterioration in turbomachinery is an unwanted phenomenon that changes the behaviour of the system. It can be described by a degradation indicator based on deviations from expected values of process variables. Existing models assume that the degradation is strictly increasing with fixed convexity and that there are no additional changes during the considered operating period. This work proposes the use of an exponential trend approximation with shape adaptation and apply it in a moving window framework. The suggested method of adjustment makes it possible for the model to follow the evolution of the indicator over time. The approximation method is then applied for monitoring purposes, to predict future degradation. The influence of the tuning parameters on the accuracy of the algorithm is investigated and recommendations for the values are derived. Finally directions for further work are proposed.

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.