Batch process monitoring with tensor factorization

A batch process monitoring method using tensor factorization, tensor locality preserving projections (TLPP), is proposed. In many existing vector-based methods on batch process monitoring such as MPCA and MLPP, a batch data is represented as a vector in high-dimensional space. But vectorizing batch data will lead to information loss. Essentially, a batch data is presented as a second order tensor, or a matrix. In this case, tensor factorization may be used to deal with the two-way batch data matrix directly instead of performing vectorizing procedure. Furthermore, tensor representation has some advantages such as low memory and storage requirements and less estimated parameters for normal operating condition (NOC) model. On the other hand, different from principal component analysis (PCA) which aims at preserving the global Euclidean structure of the data, the TLPP aims to preserve the local neighborhood information and to detect the intrinsic manifold structure of the data. Consequently, TLPP may be used to find more meaningful intrinsic information hidden in the observations. The effectiveness and advantages of the TLPP monitoring approach are tested with the data from a benchmark fed-batch penicillin fermentation and two industrial fermentation processes, penicillin and cephalosporin, respectively.     

基于MFA-ICA的间歇过程在线监控

针对间歇过程独特的数据特点,提出1种将因子分析(FA)作为独立成分分析(ICA)白化预处理手段的多向因子分析白化独立成分分析(multiway factor analysis-independent component analysis,MFA-ICA)间歇过程监控方法.因子分析充分考虑了模型误差的普遍意义,拥有优秀的噪声建模能力.将其代替上成分分析用于白化,可以更好的提取数据集的本质信息.首先将间歇过程三维数据依次按批次和变量展开得到二维数据矩阵,接着把上述方法用于展开后的数据,利用ICA的,I2统计图实现在线故障检测.该方法用于标准仿真平台Pensim,结果表明上述方法对于提高间歇过程故障检测的快速性,降低漏报率有明显效果.     

基于改进MPLS的间歇过程监测与故障诊断

针对间歇过程三维数据预处理中不同展开方式的多向偏最小二乘(MPLS)方法在线应用时存在的缺陷,提出改进的MPLS方法。该方法结合传统沿变量展开与批次展开的优势,不仅包含了批次间的信息,在一定程度上去除了过程的非线性及动态性,而且解决了在线应用时数据填充的问题;其次,该方法采用随时间更新的协方差代替固定的主元协方差充分考虑了得分向量的动态特性:最后,引进时变贡献图的故障诊断方法,实现了对故障源的实时跟踪。将该方法应用到工业青霉素发酵过程中,并与传统的MPLS方法进行比较。结果表明:该方法具有更好的监控性能,并能够及时检测故障及跟踪故障源。     

基于自适应多向独立成分分析的间歇过程监控的研究

针对间歇过程批次与批次之间,操作条件缓慢变化的特性,提出一种基于自适应多向独立成分分析(MICA)的监控算法。该方法首先用MICA法建模,然后在历史数据集中加入新的正常批次并剔除最早批次,逐渐更新模型,同时引入遗忘因子,提高对新过程特性的适应性。青霉素发酵过程的仿真结果表明,自适应MICA比MICA更准确地描述过程行为,并有效减少检测故障时的误报。     

相空间重构-最小二乘支持向量机用于间歇过程变量在线预报

时间序列预测技术可实现过程参数未来变化趋势的早期预报,从而为分析判断工况是否正常、确定转入下一工序的时机提供依据.针对间歇过程数据长度短、非线性、动态、不同批次数据不等长等特点,提出了一种基于相空间重构-最小二乘支持向量机的非线性时间序列预测方法.首先将多批次数据随机的拼接组成长数据向量,差分处理后采用相空间重构关联积分C-C方法计算该序列的延迟时间τ和嵌入维数m,从而构建训练集和检验集,然后采用最小二乘支持向量机算法建立预测模型.对某间歇蒸馏过程上升气温度建立的5步预测模型可用于生产现场的在线预报.     

基于FDA-KDE间歇过程在线监控

针对基于传统的多向主元分析(Multiway Principal Component Analysis,MPCA)方法用于间歇过程在线监控时需要对新批次未反应完的数据进行预估,从而易导致误诊断,且统计量控制限的确定是以主元得分呈正态分布为假设前提的缺陷,结合Fisher判别分析(Fisher Discriminant Analysis,FDA)在数据分类及非参数统计方法核密度估计(Kernel Density Estimation,KDE)在计算概率密度函数方面的优势,提出了一种FDA-KDE的间歇过程监控方法。该方法首先利用FDA求取正常工况数据和故障数据的Fisher特征向量和判别向量,获得Fisher特征向量的相似度:然后在提出偏平均集成平方误差(Biased Mean Integrated Squared Error,BMISE)交叉验证法确定KDE的带宽从而获得相似度统计量控制限的基础上,利用已获得的数据测量值对过程进行监控,避免了基于MPCA方法对未来测量值的预估;最后采用基于Fisher判别向量权重的贡献图方法来进行故障诊断。通过对青霉素发酵间歇过程应用表明,所提出的方法比传统的MPCA方法能更及时地监测出过程异常情况,更准确地判断异常发生的原因。     

基于改进的MPCA与DTW方法及其在批过程故障诊断中的应用

针对传统的多向主元分析(Multiway Principal Component Analysis,MPCA)常会导致误诊断,且对批生产过程难以保证在线状态监测和故障诊断的实时性,提出了一种改进的MPCA与动态时间错位(Dynamitc Time Warping,DTW)方法,该方法采用多模型非线性结构代替传统的MPCA单模型线性化结构,并利用对称式DTW算法解决了多元轨迹同步化的问题。将该方法应用到青霉素发酵批过程的在线故障监测中,结果表明它克服了MPCA不能处理非线性过程和实时性问题,并避免了MPCA在线应用时预报未来测量值带来的误差,提高了批过程性能监测和故障诊断的准确性。     

Batch process modelling with mixtures of Gaussian processes

In recent years Gaussian processes have attracted a significant amount of interest with the particular focus being that of process modelling. This has primarily been a consequence of their good predictive performance and inherent analytical properties. Gaussian processes are a member of the family of non-parametric Bayesian regression models and can be derived from the perspective of neural networks. Their behaviour is controlled through the structure of the covariance function. However, when applied to batch processes, whose data exhibits different variance structures throughout the duration of the batch, a single Gaussian process may not be appropriate for the accurate modelling of its behaviour. Furthermore there are issues with respect to the computational costs of Gaussian processes. The implementation of a Gaussian process model requires the repeated computation of a matrix inverse whose order is the cubic of the number of training data points. This renders the algorithm impractical when dealing with large data sets. To address these two issues, a mixture model of Gaussian processes is proposed. The resulting prediction is attained as a weighted sum of the outputs from each Gaussian process component, with the weights determined by a Gaussian kernel gating network. The model is implemented through a Bayesian approach utilising Markov chain Monte Carlo algorithms. The proposed methodology is applied to data from a bench-mark batch simulation polymerization process, methyl methacrylate (MMA), and the results are compared with those from a single Gaussian process to illustrate the advantages of the proposed mixture model approach.     

Event-based process monitoring

This article presents an efficient Intelligent Supervision System (ISS) architecture for the monitoring of a plant. The ISS detects relevant events which are later used to identify the state of the plant. The ISS layered structure implements the sliding window paradigm to detect significative events from measured signals. This methodology allows for the design of flexible ISS interfaces, that can be easily configured to detect the desired events. The behavioral model of the plant is described by an automaton, which matches event sequences to the state of the plant. Expert knowledge is used in the design of the whole ISS architecture. The ISS has been implemented in Simulink, and applied to a biotechnological process.     

Real space process algebra

The real time process algebra of Baeten and Bergstra [Formal Aspects of Computing,3, 142–188 (1991)] is extended to real space by requiring the presence of spatial coordinates for each atomic action, in addition to the required temporal attribute. It is found that asynchronous communication cannot easily be avoided. Based on the state operators of Baeten and Bergstra [Information and Computation,78, 205–245 (1988)] and following Bergstra et al. [Proc. Seminar on Concurrency, LNCS 197, Springer, 1985, pp. 76–95], asychronous communication mechanisms are introduced as an additional feature of real space process algebra. The overall emphasis is on the introductory explanation of the features of real space process algebra, and characteristic examples are given for each of these.     

Temporal stability in predictive process monitoring

Predictive process monitoring is concerned with the analysis of events produced during the execution of a business process in order to predict as early as possible the final outcome of an ongoing case. Traditionally, predictive process monitoring methods are optimized with respect to accuracy. However, in environments where users make decisions and take actions in response to the predictions they receive, it is equally important to optimize the stability of the successive predictions made for each case. To this end, this paper defines a notion of temporal stability for binary classification tasks in predictive process monitoring and evaluates existing methods with respect to both temporal stability and accuracy. We find that methods based on XGBoost and LSTM neural networks exhibit the highest temporal stability. We then show that temporal stability can be enhanced by hyperparameter-optimizing random forests and XGBoost classifiers with respect to inter-run stability. Finally, we show that time series smoothing techniques can further enhance temporal stability at the expense of slightly lower accuracy.     

Statistical process control: analyzing space shuttle onboardsoftware process

Demand for increased software process efficiency and effectiveness places measurement demands on the software engineering community beyond those traditionally practiced. Statistical- and process-thinking principles lead to the use of statistical process control (SPC) methods to determine the consistency and capability of the processes used to develop software. The authors use data and analysis from a collaborative effort between the Software Engineering Institute (a federally funded research and development center sponsored by the US Department of Defense) and the Space Shuttle Onboard Software Project as a vehicle to illustrate the analytic processes analysts frequently encounter when using SPC     

Petri net-based process monitoring: a workflow management system for process modelling and monitoring

Nowadays business process management is becoming a fundamental piece of many industrial processes. To manage the evolution and interactions between the business actions it is important to accurately model the steps to follow and the resources needed by a process. Workflows provide a way of describing the order of execution and the dependencies between the constituting activities of business processes. Workflow monitoring can help to improve and avoid delays in industrial environments where concurrent processes are carried out. In this article a new Petri net extension for modelling workflow activities together with their required resources is presented: resource-aware Petri nets (RAPN). An intelligent workflow management system for process monitoring and delay prediction is also introduced. Resource aware-Petri nets include time and resources within the classical Petri net workflow representation, facilitating the task of modelling and monitoring workflows. The workflow management system monitors the execution of workflows and detects possible delays using RAPN. In order to test this new approach, different services from a medical maintenance environment have been modelled and simulated.     

Prototyping a process monitoring experiment

Features are often the basic unit of development for a very large software system and represent long-term efforts, spanning up to several years from inception to actual use. Developing an experiment to monitor (by means of sampling) such lengthy processes requires a great deal of care in order to minimize casts and to maximize benefits. Just as prototyping is often a necessary auxiliary step in a large-scale, long-term development effort, so, too, is prototyping a necessary step in the development of a large-scale, long-term process monitoring experiment. Therefore, we have prototyped our experiment using a representative process and reconstructed data from a large and rich feature development. This approach has yielded three interesting sets of results. First, we reconstructed a 30-month time diary for the lead engineer of a feature composed of both hardware and software. These data represent the daily state (where the lead engineer spent the majority of his time) for a complete cycle of the development process. Second, we found that we needed to modify our experimental design. Our initial set of states did not represent the data as well as we had hoped. This is exemplified by the fact that the “Other” category is too large. Finally, the data provide evidence for both a waterfall view and an interactive, cyclic view of software development. We conclude that the prototyping effort is a necessary part of developing and installing any large-scale process monitoring experiment     

Process sequencing and process clustering in process planning using state space search

Process sequencing, as a very important part of process planning, has been the subject of many research reports in the area of process planning, but is usually treated as a feature-sequencing problem. This paper presents a novel algorithm for process sequencing, which considers the feature precedence network, different process candidates, and machine and tool constraints. The algorithm consists of two parts: process clustering and process sequencing. For clustering we used a notion of the same resource usage for different features, while for sequencing we applied the best-first search method algorithm to generate an optimal process sequence. The algorithm has been applied on several examples with realistic complexities, and it showed satisfactory results.     

Cyber physical process monitoring systems

Manufacturing process monitoring systems is evolving from centralised bespoke applications to decentralised reconfigurable collectives. The resulting cyber-physical systems are made possible through the integration of high power computation, collaborative communication, and advanced analytics. This digital age of manufacturing is aimed at yielding the next generation of innovative intelligent machines. The focus of this research is to present the design and development of a cyber-physical process monitoring system; the components of which consist of an advanced signal processing chain for the semi-autonomous process characterisation of a CNC turning machine tool. The novelty of this decentralised system is its modularity, reconfigurability, openness, scalability, and unique functionality. The function of the decentralised system is to produce performance criteria via spindle vibration monitoring, which is correlated to the occurrence of sequential process events via motor current monitoring. Performance criteria enables the establishment of normal operating response of machining operations, and more importantly the identification of abnormalities or trends in the sensor data that can provide insight into the quality of the process ongoing. The function of each component in the signal processing chain is reviewed and investigated in an industrial case study.     

Reconstruction-based contribution for process monitoring

This paper presents a new method to perform fault diagnosis for data-correlation based process monitoring. As an alternative to the traditional contribution plot method, a reconstruction-based contribution for fault diagnosis is proposed based on monitored indices, SPE, T² and a combined index φ. Analysis of the diagnosability of the traditional contributions and the reconstruction-based contributions is performed. The lack of diagnosability of traditional contributions is analyzed for the case of single sensor faults with large fault magnitudes, whereas for the same case the proposed reconstruction-based contributions guarantee correct diagnosis. Monte Carlo simulation results are provided for the case of modest fault magnitudes by randomly assigning fault sensors and fault magnitudes.     

OWave control chart for monitoring the process mean

In this paper a control chart for monitoring the process mean, called OWave (Orthogonal Wavelets), is proposed. The statistic that is plotted in the proposed control chart is based on weighted wavelets coefficients, which are provided through the Discrete Wavelets Transform using Daubechies $\mathit{db}2$ wavelets family. The statistical behavior of the wavelets coefficients when the mean shifts are occurring is presented, and the distribution of wavelets coefficients in the case of normality and independence assumptions is provided. The on-line algorithm of implementing the proposed method is also provided. The detection performance is based on simulation studies, and the comparison result shows that OWave control chart performs slightly better than Fixed Sample Size and Sampling Intervals control charts ($\overline{X}$, EWMA, CUSUM) in terms of Average Run Length. In addition, illustrative examples of the new control chart are presented, and an application to Tennessee Eastman Process is also proposed.