Multiway discrete hidden Markov model‐based approach for dynamic batch process monitoring and fault classification

A new multiway discrete hidden Markov model (MDHMM)‐based approach is proposed in this article for fault detection and classification in complex batch or semibatch process with inherent dynamics and system uncertainty. The probabilistic inference along the state transitions in MDHMM can effectively extract the dynamic and stochastic patterns in the process operation. Furthermore, the used multiway analysis is able to transform the three‐dimensional (3‐D) data matrices into 2‐D measurement‐state data sets for hidden Markov model estimation and state path optimization. The proposed MDHMM approach is applied to fed‐batch penicillin fermentation process and compared to the conventional multiway principal component analysis (MPCA) and multiway dynamic principal component analysis (MDPCA) methods in three faulty scenarios. The monitoring results demonstrate that the MDHMM approach is superior to both the MPCA and MDPCA methods in terms of fault detection and false alarm rates. In addition, the supervised MDHMM approach is able to classify different types of process faults with high fidelity. © 2011 American Institute of Chemical Engineers AIChE J, 2012     

Operating condition diagnosis based on HMM with adaptive transition probabilities in presence of missing observations

A new approach for modeling and monitoring of the multivariate processes in presence of faulty and missing observations is introduced. It is assumed that operating modes of the process can transit to each other following a Markov chain model. Transition probabilities of the Markov chain are time varying as a function of the scheduling variable. Therefore, the transition probabilities will be able to vary adaptively according to different operating modes. In order to handle the problem of missing observations and unknown operating regimes, the expectation maximization algorithm is used to estimate the parameters. The proposed method is tested on two simulations and one industrial case studies. The industrial case study is the abnormal operating condition diagnosis in the primary separation vessel of oil‐sand processes. In comparison to the conventional methods, the proposed method shows superior performance in detection of different operating conditions of the process. © 2014 American Institute of Chemical Engineers AIChE J, 61: 477–493, 2015     

Multiway independent component analysis mixture model and mutual information based fault detection and diagnosis approach of multiphase batch processes

Batch process monitoring is a challenging task, because conventional methods are not well suited to handle the inherent multiphase operation. In this study, a novel multiway independent component analysis (MICA) mixture model and mutual information based fault detection and diagnosis approach is proposed. The multiple operating phases in batch processes are characterized by non‐Gaussian independent component mixture models. Then, the posterior probability of the monitored sample is maximized to identify the operating phase that the sample belongs to, and, thus, the localized MICA model is developed for process fault detection. Moreover, the detected faulty samples are projected onto the residual subspace, and the mutual information based non‐Gaussian contribution index is established to evaluate the statistical dependency between the projection and the measurement along each process variable. Such contribution index is used to diagnose the major faulty variables responsible for process abnormalities. The effectiveness of the proposed approach is demonstrated using the fed‐batch penicillin fermentation process, and the results are compared to those of the multiway principal component analysis mixture model and regular MICA method. The case study demonstrates that the proposed approach is able to detect the abnormal events over different phases as well as diagnose the faulty variables with high accuracy. © 2013 American Institute of Chemical Engineers AIChE J, 59: 2761–2779, 2013     

Orthogonal nonnegative matrix factorization based local hidden Markov model for multimode process monitoring

Traditional data driven fault detection methods assume that the process operates in a single mode so that they cannot perform well in processes with multiple operating modes. To monitor multimode processes effectively, this paper proposes a novel process monitoring scheme based on orthogonal nonnegative matrix factorization (ONMF) and hidden Markov model (HMM). The new clustering technique ONMF is employed to separate data fromdifferent processmodes. ThemultipleHMMs for various operating modes lead to highermodeling accuracy. The proposed approach does not presume the distribution of data in each mode because the process uncertainty and dynamics can bewell interpreted through the hidden Markov estimation. The HMM-based monitoring indication named negative log likelihood probability is utilized for fault detection. In order to assess the proposed monitoring strategy, a numerical example and the Tennessee Eastman process are used. The results demonstrate that this method provides efficient fault detection performance.     

Dynamic stationary subspace analysis based on Gaussian mixture models for ironmaking process monitoring

Blast furnace ironmaking process monitoring is an important and challenging task. Due to the influence of hot blast stove switching and large fluctuations in the quality of raw materials, the measurements of ironmaking processes show obvious non-stationary characteristics, and in addition, the observed data are also characterized by time-series dynamic and non-Gaussian characteristics. In this paper, a dynamic stationary subspace analysis method based on the Gaussian mixture model (DSSA–GMM) is proposed to address the difficulties in blast furnace ironmaking process monitoring. The time-series dynamic relationship of the data is conducted by introducing a sliding time window. The Gaussian mixture model (GMM) is used to deal with the non-Gaussian characteristics of the data, and the parameters of the GMMs are estimated using the expectation–maximization algorithm. The stationary projection matrix is obtained by optimizing the Kullback–Leibler (K–L) divergence between GMMs of different periods to realize the stationary subspace separation. Finally, the convex hull of the stationary subspace is established for fault detection, thus realizing the monitoring for non-stationary and non-Gaussian dynamic processes. The effectiveness of the DSSA–GMM method is verified by a numerical simulation and a dataset collected from an actual blast furnace ironmaking process.     

Bayesian analysis of multivariate Gaussian hidden Markov models with an unknown number of regimes

Multivariate Gaussian hidden Markov models with an unknown number of regimes are introduced here in the Bayesian setting and new efficient reversible jump Markov chain Monte Carlo algorithms for estimating both the dimension and the unknown parameters of the model are presented. Hidden Markov models are an extension of mixture models that can be applied to time series so as to classify the observations in a small number of groups, to understand when change points occur in the dynamics of the series and to model data heterogeneity through the switching among subseries with different means and covariance matrices. These aims can be achieved by assuming that the observed phenomenon is driven by a latent, or hidden, Markov chain. The methodology is illustrated through two different examples of multivariate time series.     

Hidden Markov model based approach for diagnosing cause of alarm signals

When a fault occurs in a process, it slowly propagates within the system and affects the measurements triggering a sequence of alarms in the control room. The operators are required to diagnose the cause of alarms and take necessary corrective measures. The idea of representing the alarm sequence as the fault propagation path and using the propagation path to diagnose the fault is explored. A diagnoser based on hidden Markov model is built to identify the cause of the alarm signals. The proposed approach is applied to an industrial case study: Tennessee Eastman process. The results show that the proposed approach is successful in determining the probable cause of alarms generated with high accuracy. The model was able to identify the cause accurately, even when tested with short alarm sub-sequences. This allows for early identification of faults, providing more time to the operator to restore the system to normal operation.     

Cycle temporal algorithm-based multivariate statistical methods for fault diagnosis in chemical processes

Multivariate statistical process monitoring methods are often used in chemical process fault diagnosis. In this article,(I) the cycle temporal algorithm(CTA) combined with the dynamic kernel principal component analysis(DKPCA) and the multiway dynamic kernel principal component analysis(MDKPCA) fault detection algorithms are proposed, which are used for continuous and batch process fault detections,respectively. In addition,(II) a fault variable identification model based on reconstructed-based ...     

Fault detection and isolation in hybrid process systems using a combined data‐driven and observer‐design methodology

A combined data‐driven and observer‐design methodology for fault detection and isolation (FDI) in hybrid process systems with switching operating modes is proposed. The main contribution is to construct a unified framework for FDI by integrating Gaussian mixture models (GMM), subspace model identification (SMI), and results from unknown input observer (UIO) theory. Initially, a GMM is built to identify and describe the multimodality of hybrid systems using the recorded input/output process data. A state‐space model is then obtained for each specific operating mode based on SMI if the system matrices are unknown. An UIO is designed to estimate the system states robustly, based on which the fault detection is laid out through a multivariate analysis of the residuals. Finally, by designing a set of unknown input matrices for specific fault scenarios, fault isolation is performed through the disturbance‐decoupling principle from the UIO theory. A significant benefit of the developed framework is to overcome some of the limitations associated with individual model‐based and data‐based approaches in dealing with the problem of FDI in hybrid systems. Finally, the validity and effectiveness of the proposed monitoring framework are demonstrated using a numerical example, a simulated continuous stirred tank heater process, and the Tennessee Eastman benchmark process. © 2014 American Institute of Chemical Engineers AIChE J, 60: 2805–2814, 2014     

D-vine copulas混合模型及其在故障检测中的应用

过程监控技术是保证现代流程工业安全平稳运行及产品质量的有效手段。传统的过程监控方法大多采用维度约简方法提取数据特征,且要求过程数据必须服从高斯分布、线性等限制条件,对复杂工况条件下发生的故障难以取得较好的检测效果。因此,提出了混合D-vine copulas故障诊断模型,在不降维的情况下直接刻画数据中存在的复杂相关关系,构建过程变量的统计模型实现对存在非线性与非高斯性过程的精确描述。通过EM算法和伪极大似然估计优化混合模型参数,然后结合高密度区域（HDR）与密度分位数法等理论,构建广义贝叶斯概率（GBIP）指标实现对过程的实时监测。数值例子及在TE过程上的仿真结果说明了该混合模型的有效性及在故障检测中的良好性能。     

Sequential Anomaly Detection in Wireless Sensor Networks and Effects of Long-Range Dependent Data

Abstract

Anomaly detection is important for the correct functioning of wireless sensor networks. Recent studies have shown that node mobility along with spatial correlation of the monitored phenomenon in sensor networks can lead to observation data that have long range dependency, which could significantly increase the difficulty of anomaly detection. In this article, we develop an anomaly detection scheme based on multiscale analysis of the long-range dependent traffic to address this challenge. In this proposed detection scheme, the discrete wavelet transform is used to approximately de-correlate the traffic data and capture data characteristics in different timescales. The remaining dependencies are then captured by a multilevel hidden Markov model in the wavelet domain. To estimate the model parameters, we develop an online discounting expectation maximization (EM) algorithm, which also tracks variations of the estimated models over time. Network anomalies are detected as abrupt changes in the tracked model variation scores. Statistical properties of our detection scheme are evaluated numerically using long-range dependent time series. We also evaluate our detection scheme in malicious scenarios simulated using the NS-2 network simulator.     

一种鲁棒半监督建模方法及其在化工过程故障检测中的应用

复杂化工过程的观测样本往往包含着测量噪声与少量的离群点数据,而这些受污染的数据会影响数据驱动的过程建模与故障检测方法的准确性。本文考虑了化工过程测量样本的这一实际情况,提出了一种鲁棒半监督PLVR模型（RSSPLVR）,并利用核方法将其扩展为非线性的形式（K-RSSPLVR）。此类算法利用基于样本相似度的加权系数作为概率模型的先验参数,能有效消除离群点对建模的影响。利用加权后的建模样本,本文通过EM算法训练了RSSPLVR和K-RSSPLVR的模型参数,并提出了相应的故障检测算法。最后,通过TE过程仿真实验验证了所提出方法的有效性。     

基于PCA混合模型的多工况过程监控

针对传统多元统计故障检测方法大多假设测量数据服从单一高斯分布的不足,提出了一种基于PCA(principal component analysis)混合模型的多工况过程监测方法。首先通过直接对混合模型的各高斯成分的协方差进行PCA降维变换,使得协方差阵对角化,既减少了运算量又避免了变量相关而导致的奇异性问题;同时采用BYY增量EM算法自动获取混合模型的最佳混合分量数目,避免了常规EM算法的不足。所得的混合模型,除包括均值、协方差和先验概率等参数外,还包括了PCA载荷阵,即对每个混合元建立了PCA模型。然后给出了统计量定义,实现对多工况过程的故障检测。数值例子和TE过程的应用表明,本文提出的方法无需过程先验知识,能自动获取工况数目、精确估计各个工况的统计特性,并更准确及时地检测出多工况过程的各种故障。     

一种不等长的多模态间歇过程故障检测方法

提出一种不等长的多模态间歇过程故障检测方法。首先,运用局部加权算法对不等长批次数据进行预处理。在训练样本中确定不等长数据的最大可保留长度,利用k近邻信息,通过加权重构出不等长批次缺失的数据点。其次,对等长的训练集构造局部近邻标准化矩阵,运用K-means算法进行模态聚类,使用局部离群因子方法确定第一控制限,并剔除离群样本。最后,对各个模态建立MPCA模型并确定第二控制限。根据各个模态控制限的匹配系数计算统一的统计量和控制限,在统一的控制限下进行多模态故障检测。将提出方法应用于半导体工业过程,仿真结果表明,与传统的故障检测算法相比,本文算法提高了故障检测率,验证了该方法的有效性。     

Development of hidden semi-Markov models for diagnosis of multiphase batch operation

This paper deals with automatic on-line detection and diagnosis of fault patterns in multiphase batch processes. A novel and flexible approach based on the combination of hidden segmental semi-Markov models (HSMM) and multiway principal component analysis (MPCA) is proposed. In all batch operations, process variables may have correlations with each other, and MPCA is used to handle cross-correlation among process variables. In multiphase batch processes, the effect of external factors on process variables is phase-specific and the duration of each phase varies from batch to batch. HSMM is used to model the multiphase batch operation by representing each phase with a macro-state whose duration is determined by a phase-specific probability distribution of a number of micro-states. The output of each micro-state corresponds to the values of the monitored variables at a specific point in time. Given this structure, MPCA-HSMM parameters are trained by the batch operation data and recursive Viterbi algorithm is used to find out the optimum state sequence from each batch. Probability values of the optimum state sequence are collected to construct the probabilistic model which is used to compute the corresponding control limit for the specified operating condition. One MPCA-HSMM model is to be built for each type of previously known operating condition—normal and fault events. The power and advantages of the proposed method are successfully demonstrated in a simulated fed-batch penicillin cultivation process. MPCA-HSMM correctly identifies the type of fault from the batch operation data.     

Parameter estimation in models with hidden variables : An application to a biotech process

Biological processes are often characterised by significant nonlinearities, noisy measurements and hidden process variables. The dynamic behaviour of such processes can be represented by stochastic differential equations obtained from physical laws. We propose a Bayesian algorithm for parameter estimation in stochastic nonlinear biological processes with unmeasured (or hidden) variables. The proposed algorithm, involves drawing random samples iteratively from a posterior density functions of the parameters and the hidden variables. A Bayesian sampling techniques is used to approximate these posterior density functions. Both Metropolis–Hastings algorithm and Gibbs sampling are used for sample generation. The algorithm is extended to handle multiple data sets and missing observations. The algorithm is applied to an experimental data set collected from an algal bioreactor system. © 2011 Canadian Society for Chemical Engineering     

Is Average Run Length to False Alarm Always an Informative Criterion?

Abstract

Apart from Bayesian approaches, the average run length (ARL) to false alarm has always been seen as the natural performance criterion for quantifying the propensity of a detection scheme to make false alarms, and no researchers seem to have questioned this on grounds that it does not always apply. In this article, we show that in the change-point problem with mixture prechange models, detection schemes with finite detection delays can have infinite ARLs to false alarm. We also discuss the implication of our results on the change-point problem with either exchangeable prechange models or hidden Markov models. Alternative minimax formulations with different false alarm criteria are proposed.     

Soft sensor modelling of acrolein conversion based on hidden Markov model of principle component analysis and fireworks algorithm

Based on the existing propylene oxidation process, it is important to measure acrolein conversion for the production of acrylic acid. The gas chromatographic analyzer is generally used to analyze the acrolein conversion as an off‐line method. In this paper, a soft sensor modelling method of acrolein conversion based on the hidden Markov model with principle component analysis (PCA) and the fireworks algorithm (FWA) is proposed. Firstly, PCA is used to decrease the input variables of hidden Markov model. Then, FWA is applied to optimize the initial parameters of the hidden Markov model. Finally, the hidden Markov model based on PCA and the FWA is employed to predict the acrolein conversion. The proposed method is compared with the support vector machine (SVM), the artificial neural network (ANN), and the hidden Markov method (HMM) to show its superior performance.     

A note on the mixture transition distribution and hidden Markov models

We discuss an interpretation of the mixture transition distribution (MTD) for discrete‐valued time series which is based on a sequence of independent latent variables which are occasion‐specific. We show that, by assuming that this latent process follows a first order Markov Chain, MTD can be generalized in a sensible way. A class of models results which also includes the hidden Markov model (HMM). For these models we outline an EM algorithm for the maximum likelihood estimation which exploits recursions developed within the HMM literature. As an illustration, we provide an example based on the analysis of stock market data referred to different American countries.     

Regression‐based analysis of multivariate non‐Gaussian datasets for diagnosing abnormal situations in chemical processes

This article presents a regression‐based monitoring approach for diagnosing abnormal conditions in complex chemical process systems. Such systems typically yield process variables that may be both Gaussian and non‐Gaussian distributed. The proposed approach utilizes the statistical local approach to monitor parametric changes of the latent variable model that is identified by a revised non‐Gaussian regression algorithm. Based on a numerical example and recorded data from a fluidized bed reactor, the article shows that the proposed approach is more sensitive when compared to existing work in this area. A detailed analysis of both application studies highlights that the introduced non‐Gaussian monitoring scheme extracts latent components that provide a better approximation of non‐Gaussian source signal and/or is more sensitive in detecting process abnormities. © 2013 American Institute of Chemical Engineers AIChE J, 60: 148–159, 2014