Monitoring, fault diagnosis, fault-tolerant control and optimization: Data driven methods

Fault Diagnostics Based on the Analysis of Probability Distributions Estimated Using a Particle Filter

This paper proposes a monitoring procedure based on characterizing state probability distributions estimated using particle filters. The work highlights what types of information can be obtained during state estimation and how the revealed information helps to solve fault diagnosis tasks. If a failure is present in the system, the output predicted by the model is inconsistent with the actual output, which affects the operation of the estimator. The heterogeneity of the probability distribution of states increases, and a large proportion of the particles lose their information content. The correlation structure of the posterior probability density can also be altered by failures. The proposed method uses various indicators that characterize the heterogeneity and correlation structure of the state distribution, as well as the consistency between model predictions and observed behavior, to identify the effects of failures.The applicability of the utilized measures is demonstrated through a dynamic vehicle model, where actuator and sensor failure scenarios are investigated.

Data-Driven Formation Control for Unknown MIMO Nonlinear Discrete-Time Multi-Agent Systems With Sensor Fault

A data-driven distributed formation control algorithm is proposed for an unknown heterogeneous non-affine nonlinear discrete-time MIMO multi-agent system (MAS) with sensor fault. For the considered unknown MAS, the dynamic linearization technique in model-free adaptive control (MFAC) theory is used to transform the unknown MAS into an equivalent virtual dynamic linearization data model. Then using the virtual data model, the structure of the distributed model-free adaptive controller is constructed. For the incorrect signal measurements due to the sensor fault, the radial basis function neural network (RBFNN) is first trained for the MAS under the fault-free case, then using the outputs of the well-trained RBFNN and the actual outputs of MAS under sensor fault case, the estimation laws of the unknown fault and system parameters in the virtual data model are designed with only the measured input-output (I/O) data information. Finally, the boundedness of the formation error is analyzed by the contraction mapping method and mathematical induction method. The effectiveness of the proposed algorithm is illustrated by simulation examples.

Fault Detection of Non-Gaussian and Nonlinear Processes Based on Independent Slow Feature Analysis

Independent component analysis (ICA) is an excellent latent variables (LVs) extraction method that can maximize the non-Gaussianity between LVs to extract statistically independent latent variables and which has been widely used in multivariate statistical process monitoring (MSPM). The underlying assumption of ICA is that the observation data are composed of linear combinations of LVs that are statistically independent. However, the assumption is invalid because the observation data are always derived from the nonlinear mixture of LVs due to the nonlinear characteristic in industrial processes. Under this circumstance, the ICA-based fault detection is unable to provide accurate detection for specific faults of industrial processes. Since the observation data come from the nonlinear mixing of LVs, this makes the observation data change faster than the intrinsic LVs on the time scale. The temporal slowness can be regarded as an additional criterion in the extraction of LVs. The slow feature analysis (SFA) derived from the temporal slowness has received extensive attention and application in MSPM in recent years. Simultaneously, the temporal slowness is expected to make up for the problem that the LVs extracted by ICA have difficulty accurately describing the characteristics of the process. To solve the above problems, this work proposes to monitor non-Gaussian and nonlinear processes using the independent slow feature analysis (ISFA) that combines statistical independence and temporal slowness in extracting the LVs. When the observation data are composed of a nonlinear mixture of LVs, the extracted LVs of ISFA can describe the characteristics of the processes better than ICA, thereby improving the accuracy of fault detection for the non-Gaussian and nonlinear processes. The superiority of the proposed method is verified by a numerical example design and the Tennessee-Eastman process.

A Survey on Learning-Based Model Predictive Control: Toward Path Tracking Control of Mobile Platforms

The learning-based model predictive control (LB-MPC) is an effective and critical method to solve the path tracking problem in mobile platforms under uncertain disturbances. It is well known that the machine learning (ML) methods use the historical and real-time measurement data to build data-driven prediction models. The model predictive control (MPC) provides an integrated solution for control systems with interactive variables, complex dynamics, and various constraints. The LB-MPC combines the advantages of ML and MPC. In this work, the LB-MPC technique is summarized, and the application of path tracking control in mobile platforms is discussed by considering three aspects, namely, learning and optimizing the prediction model, the controller design, and the controller output under uncertain disturbances. Furthermore, some research challenges faced by LB-MPC for path tracking control in mobile platforms are discussed.

Big data analytics opportunities for applications in process engineering

Big data is an expression for massive data sets consisting of both structured and unstructured data that are particularly difficult to store, analyze and visualize. Big data analytics has the potential to help companies or organizations improve operations as well as disclose hidden patterns and secret correlations to make faster and intelligent decisions. This article provides useful information on this emerging and promising field for companies, industries, and researchers to gain a richer and deeper insight into advancements. Initially, an overview of big data content, key characteristics, and related topics are presented. The paper also highlights a systematic review of available big data techniques and analytics. The available big data analytics tools and platforms are categorized. Besides, this article discusses recent applications of big data in chemical industries to increase understanding and encourage its implementation in their engineering processes as much as possible. Finally, by emphasizing the adoption of big data analytics in various areas of process engineering, the aim is to provide a practical vision of big data.

Single Fault Diagnosis Method of Sensors in Cascade System Based on Data-Driven

The reliability and safety of the cascade system, which is widely applied, have attached attention increasingly. Fault detection and diagnosis can play a significant role in enhancing its reliability and safety. On account of the complexity of the double closed-loop system in operation, the problem of fault diagnosis is relatively complex. For the single fault of the second-order valued system sensors, a real-time fault diagnosis method based on data-driven is proposed in this study. Off-line data is employed to establish static fault detection, location, estimation, and separation models. The static models are calibrated with on-line data to obtain the real-time fault diagnosis models. The real-time calibration, working flow and anti-interference measures of the real-time diagnosis system are given. Experiments results demonstrate the validity and accuracy of the fault diagnosis method, which is suitable for the general cascade system.

Diagnosis and Monitoring of Volatile Fatty Acids Production from Raw Cheese Whey by Multiscale Time-Series Analysis

Anaerobic treatment is a viable alternative for the treatment of agro-industrial waste. Anaerobic digestion reduces organic load and produces volatile fatty acids (VFA), which are precursors of value-added products such as methane-rich biogas, biohydrogen, and biopolymers. Nowadays, there are no low-cost diagnosis and monitoring systems that analyze the dynamic behavior of key variables in real time, representing a significant limitation for its practical implementation. In this work, the feasibility of using the multiscale analysis to diagnose and monitor the key variables in VFA production by anaerobic treatment of raw cheese whey is presented. First, experiments were carried out to evaluate the performance of the proposed methodology under different operating conditions. Then, experimental pH time series were analyzed using rescaled range (R/S) techniques. Time-series analysis shows that the anaerobic VFA production exhibits a multiscale behavior, identifying three characteristic regions (i.e., three values of Hurst exponent). In addition, the dynamic Hurst exponents show satisfactory correlations with the chemical oxygen demand (COD) consumption and VFA production. The multiscale analysis of pH time series is easy to implement and inexpensive. Hence, it could be used as a diagnosis and indirect monitoring system of key variables in the anaerobic treatment of raw cheese whey.

A review of data-driven fault detection and diagnosis methods: applications in chemical process systems

Fault detection and diagnosis (FDD) systems are developed to characterize normal variations and detect abnormal changes in a process plant. It is always important for early detection and diagnosis, especially in chemical process systems to prevent process disruptions, shutdowns, or even process failures. However, there have been only limited reviews of data-driven FDD methods published in the literature. Therefore, the aim of this review is to provide the state-of-the-art reference for chemical engineers and to promote the application of data-driven FDD methods in chemical process systems. In general, there are two different groups of data-driven FDD methods: the multivariate statistical analysis and the machine learning approaches, which are widely accepted and applied in various industrial processes, including chemicals, pharmaceuticals, and polymers. Many different multivariate statistical analysis methods have been proposed in the literature, such as principal component analysis, partial least squares, independent component analysis, and Fisher discriminant analysis, while the machine learning approaches include artificial neural networks, neuro-fuzzy methods, support vector machine, Gaussian mixture model, K-nearest neighbor, and Bayesian network. In the first part, this review intends to provide a comprehensive literature review on applications of data-driven methods in FDD systems for chemical process systems. In addition, the hybrid FDD frameworks have also been reviewed by discussing the distinct advantages and various constraints, with some applications as examples. However, the choice for the data-driven FDD methods is not a straightforward issue. Thus, in the second part, this paper provides a guideline for selecting the best possible data-driven method for FDD systems based on their faults. Finally, future directions of data-driven FDD methods are summarized with the intent to expand the use for the process monitoring community.

A Review of Kernel Methods for Feature Extraction in Nonlinear Process Monitoring

Kernel methods are a class of learning machines for the fast recognition of nonlinear patterns in any data set. In this paper, the applications of kernel methods for feature extraction in industrial process monitoring are systematically reviewed. First, we describe the reasons for using kernel methods and contextualize them among other machine learning tools. Second, by reviewing a total of 230 papers, this work has identified 12 major issues surrounding the use of kernel methods for nonlinear feature extraction. Each issue was discussed as to why they are important and how they were addressed through the years by many researchers. We also present a breakdown of the commonly used kernel functions, parameter selection routes, and case studies. Lastly, this review provides an outlook into the future of kernel-based process monitoring, which can hopefully instigate more advanced yet practical solutions in the process industries.

Steady-State Data Reconciliation Framework for a Direct Continuous Tableting Line

Purpose

Reliable process monitoring in real-time remains a challenge for the pharmaceutical industry. Dealing with random and gross errors in the process measurements in a systematic way is a potential solution. In this paper, we present a process model-based framework, which for given sensor network and measurement uncertainties will predict the most likely state of the process. Thus, real-time process decisions, whether for process control or exceptional events management, can be based on the most reliable estimate of the process state.

Methods

Reliable process monitoring is achieved by using data reconciliation (DR) and gross error detection (GED) to mitigate the effects of random measurement errors and non-random sensor malfunctions. Steady-state data reconciliation (SSDR) is the simplest forms of DR but offers the benefits of short computational times. We also compare and contrast the model-based DR approach (SSDR-M) to the purely data-driven approach (SSDR-D) based on the use of principal component constructions.

Results

We report the results of studies on a pilot plant-scale continuous direct compression-based tableting line at steady-state in two subsystems. If the process is linear or mildly nonlinear, SSDR-M and SSDR-D give comparable results for the variables estimation and GED. SSDR-M also complies with mass balances and estimate unmeasured variables.

Conclusions

SSDR successfully estimates the true state of the process in presence of gross errors, as long as steady state is maintained and the redundancy requirement is met. Gross errors are also detected while using SSDR-M or SSDR-D. Process monitoring is more reliable while using the SSDR framework.

Neighborhood preserving neural network for fault detection

A novel statistical feature extraction method, called the neighborhood preserving neural network (NPNN), is proposed in this paper. NPNN can be viewed as a nonlinear data-driven fault detection technique through preserving the local geometrical structure of normal process data. The "local geometrical structure " means that each sample can be constructed as a linear combination of its neighbors. NPNN is characterized by adaptively training a nonlinear neural network which takes the local geometrical structure of the data into consideration. Moreover, in order to extract uncorrelated and faithful features, NPNN adopts orthogonal constraints in the objective function. Through backpropagation and eigen decomposition (ED) technique, NPNN is optimized to extract low-dimensional features from original high-dimensional process data. After nonlinear feature extraction, Hotelling T² statistic and the squared prediction error (SPE) statistic are utilized for the fault detection tasks. The advantages of the proposed NPNN method are demonstrated by both theoretical analysis and case studies on the Tennessee Eastman (TE) benchmark process. Extensive experimental results show the superiority of NPNN in terms of missed detection rate (MDR) and false alarm rate (FAR). The source code of NPNN can be found in https://github.com/htzhaoecust/npnn.