Online monitoring supports plant reliability and performance management by providing real time information about the condition of equipment. However, the intricate geometries and harsh operating environment of coal fired power plant boilers inhibit the ability to do online measurements of all process related variables. A low-cost alternative lies in the possibility of using knowledge about boiler operation to extract information about its condition from standard online process measurements. This approach is evaluated with the aim of enhancing online condition monitoring of a boiler’s convective pass heat exchanger network by respectively using a soft sensor and a data-driven method. The soft sensor approach is based on a one-dimensional thermofluid process model which takes measurements as inputs and calculates unmeasured variables as outputs. The model is calibrated based on design information. The data-driven method is one developed specifically in this study to identify unique fault signatures in measurement data to detect and quantify changes in unmeasured variables. The fault signatures are initially constructed using the calibrated one-dimensional thermofluid process model. The benefits and limitations of these methods are compared at the hand of a case study boiler. The case study boiler has five convective heat exchanger stages, each composed of four separate legs. The data-driven method estimates the average conduction thermal resistance of individual heat exchanger legs and the flue gas temperature at the inlet to the convective pass. In addition to this, the soft sensor estimates the average fluid variables for individual legs throughout the convective pass and therefore provides information better suited for condition prognosis. The methods are tested using real plant measurements recorded during a period which contained load changes and on-load heat exchanger cleaning events. The cleaning event provides some basis for validating the results because the qualitative changes of some unmeasured monitored variables expected during this event are known. The relative changes detected by both methods are closely correlated. The data-driven method is computationally less expensive and easily implementable across different software platforms once the fault signatures have been obtained. Fault signatures are easily trainable once the model has been developed. The soft sensors require the continuous use of the modelling software and will therefore be subject to licencing constraints. Both methods offer the possibility to enhance the monitoring resolution of modern boilers without the need to install any additional measurements. Implementation of these monitoring frameworks can provide a simple and low-cost contribution to optimized boiler performance and reliability management.

京都大学 / 0048 / 新制・課程博士 / 博士(工学) / 甲第18311号 / 工博第3903号 / 新制||工||1599(附属図書館) / 31169 / 京都大学大学院工学研究科化学工学専攻 / (主査)教授 長谷部 伸治, 教授 大嶋 正裕, 教授 宮原 稔 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

No / Polymer extrusion is regarded as an energy-intensive production process, and the real-time monitoring of both energy consumption and melt quality has become necessary to meet new carbon regulations and survive in the highly competitive plastics market. The use of a power meter is a simple and easy way to monitor energy, but the cost can sometimes be high. On the other hand, viscosity is regarded as one of the key indicators of melt quality in the polymer extrusion process. Unfortunately, viscosity cannot be measured directly using current sensory technology. The employment of on-line, in-line or off-line rheometers is sometimes useful, but these instruments either involve signal delay or cause flow restrictions to the extrusion process, which is obviously not suitable for real-time monitoring and control in practice. In this paper, simple and accurate real-time energy monitoring methods are developed. This is achieved by looking inside the controller, and using control variables to calculate the power consumption. For viscosity monitoring, a 'soft-sensor' approach based on an RBF neural network model is developed. The model is obtained through a two-stage selection and differential evolution, enabling compact and accurate solutions for viscosity monitoring. The proposed monitoring methods were tested and validated on a Killion KTS-100 extruder, and the experimental results show high accuracy compared with traditional monitoring approaches.

As inferências têm diversas aplicações na indústria de processos químicos, sendo essenciais no sucesso de projetos de controle avançado. O desempenho do controle será sempre ligado ao desempenho da inferência, sendo importante a manutenção da sua qualidade ao longo do tempo. Neste trabalho, uma nova metodologia é sugerida para o desenvolvimento de inferências baseadas em dados seguindo uma abordagem segmentada com o objetivo de facilitar a sua manutenção. A nova proposta consiste em modelar a parte estacionária separada da parte dinâmica, diferentemente do que é feito na metodologia tradicional, onde o modelo dinâmico é gerado diretamente dos dados de processo. O modelo estacionário é obtido através de uma regressão PLS (Partial Least Squares), enquanto as dinâmicas são inseridas posteriormente utilizando-se um algoritmo de otimização. A técnica é aplicada a uma coluna de destilação e o resultado obtido é semelhante ao de inferências dinâmicas e estáticas desenvolvidas com métodos tradicionais. Outras etapas do desenvolvimento de inferências também são investigadas. Na seleção de variáveis, métodos estatísticos são comparados com a busca exaustiva e se conclui este último deve ser usado como padrão, visto que custo computacional não é mais um problema. Também são apresentadas boas práticas no pré-tratamento de dados, remoção do tempo morto do cromatógrafo modelado e detecção de estados estacionários. / Soft-sensors have several applications in the chemical processes industry and are essential for the success of advanced control projects. Its performance will always be linked to the performance of the soft-sensor, so it is important to maintain its quality over time. In this paper, a new methodology is suggested for the development of data-based soft-sensors following a segmented approach in order to facilitate its maintenance. The new proposal is to model the stationary part separated from the dynamic, unlike the traditional methodology where the dynamic model is generated directly from process data. The stationary model is obtained by a PLS (Partial Least Squares) regression, while the dynamics are inserted using an optimization algorithm. The technique is applied to a distillation column and its performance is similar to dynamic and static soft-sensors developed using traditional methods. Other steps in the development of soft-sensors are also investigated. In variable selection issue, statistical methods are compared with the testing of all possibilities; the latter should be used as default, since computational cost is no longer a problem. We also present best practices in data pre-processing, gas chromatograph dead-time removal and steady state detection.

Increasing demand on more effective cell culture reactors has driven optimization works to increase output of products. This has led to development of soft sensors that uses mathematical formulas to increase the available information for the parameters during runs. In the project two parameters was evaluated for use in such a soft sensor, viability by measuring on-line capacitance with Aber probe and L-lactate production using BioSenz apparatus. To determine how well these could be used both were used on batch reactors measuring on a mouse-mouse B cell hybridoma culture which produced IgG1. On-line measurements were performed by probes which measured directly on the cell suspension or withdrew sterile sample from the reactor. Measuring viability gave results with low error, which can be concluded to the variation in reference cell count, but it could not be determined if measuring L-lactate production with BioSenz works in reactors of this size. More work needs to be done on other types of reactors, like fed-batch or perfusion, or lower working volumes.

As inferências têm diversas aplicações na indústria de processos químicos, sendo essenciais no sucesso de projetos de controle avançado. O desempenho do controle será sempre ligado ao desempenho da inferência, sendo importante a manutenção da sua qualidade ao longo do tempo. Neste trabalho, uma nova metodologia é sugerida para o desenvolvimento de inferências baseadas em dados seguindo uma abordagem segmentada com o objetivo de facilitar a sua manutenção. A nova proposta consiste em modelar a parte estacionária separada da parte dinâmica, diferentemente do que é feito na metodologia tradicional, onde o modelo dinâmico é gerado diretamente dos dados de processo. O modelo estacionário é obtido através de uma regressão PLS (Partial Least Squares), enquanto as dinâmicas são inseridas posteriormente utilizando-se um algoritmo de otimização. A técnica é aplicada a uma coluna de destilação e o resultado obtido é semelhante ao de inferências dinâmicas e estáticas desenvolvidas com métodos tradicionais. Outras etapas do desenvolvimento de inferências também são investigadas. Na seleção de variáveis, métodos estatísticos são comparados com a busca exaustiva e se conclui este último deve ser usado como padrão, visto que custo computacional não é mais um problema. Também são apresentadas boas práticas no pré-tratamento de dados, remoção do tempo morto do cromatógrafo modelado e detecção de estados estacionários. / Soft-sensors have several applications in the chemical processes industry and are essential for the success of advanced control projects. Its performance will always be linked to the performance of the soft-sensor, so it is important to maintain its quality over time. In this paper, a new methodology is suggested for the development of data-based soft-sensors following a segmented approach in order to facilitate its maintenance. The new proposal is to model the stationary part separated from the dynamic, unlike the traditional methodology where the dynamic model is generated directly from process data. The stationary model is obtained by a PLS (Partial Least Squares) regression, while the dynamics are inserted using an optimization algorithm. The technique is applied to a distillation column and its performance is similar to dynamic and static soft-sensors developed using traditional methods. Other steps in the development of soft-sensors are also investigated. In variable selection issue, statistical methods are compared with the testing of all possibilities; the latter should be used as default, since computational cost is no longer a problem. We also present best practices in data pre-processing, gas chromatograph dead-time removal and steady state detection.

As inferências têm diversas aplicações na indústria de processos químicos, sendo essenciais no sucesso de projetos de controle avançado. O desempenho do controle será sempre ligado ao desempenho da inferência, sendo importante a manutenção da sua qualidade ao longo do tempo. Neste trabalho, uma nova metodologia é sugerida para o desenvolvimento de inferências baseadas em dados seguindo uma abordagem segmentada com o objetivo de facilitar a sua manutenção. A nova proposta consiste em modelar a parte estacionária separada da parte dinâmica, diferentemente do que é feito na metodologia tradicional, onde o modelo dinâmico é gerado diretamente dos dados de processo. O modelo estacionário é obtido através de uma regressão PLS (Partial Least Squares), enquanto as dinâmicas são inseridas posteriormente utilizando-se um algoritmo de otimização. A técnica é aplicada a uma coluna de destilação e o resultado obtido é semelhante ao de inferências dinâmicas e estáticas desenvolvidas com métodos tradicionais. Outras etapas do desenvolvimento de inferências também são investigadas. Na seleção de variáveis, métodos estatísticos são comparados com a busca exaustiva e se conclui este último deve ser usado como padrão, visto que custo computacional não é mais um problema. Também são apresentadas boas práticas no pré-tratamento de dados, remoção do tempo morto do cromatógrafo modelado e detecção de estados estacionários. / Soft-sensors have several applications in the chemical processes industry and are essential for the success of advanced control projects. Its performance will always be linked to the performance of the soft-sensor, so it is important to maintain its quality over time. In this paper, a new methodology is suggested for the development of data-based soft-sensors following a segmented approach in order to facilitate its maintenance. The new proposal is to model the stationary part separated from the dynamic, unlike the traditional methodology where the dynamic model is generated directly from process data. The stationary model is obtained by a PLS (Partial Least Squares) regression, while the dynamics are inserted using an optimization algorithm. The technique is applied to a distillation column and its performance is similar to dynamic and static soft-sensors developed using traditional methods. Other steps in the development of soft-sensors are also investigated. In variable selection issue, statistical methods are compared with the testing of all possibilities; the latter should be used as default, since computational cost is no longer a problem. We also present best practices in data pre-processing, gas chromatograph dead-time removal and steady state detection.

Unter dem Titel „Neue Wege in der Umformtechnik“ werden am 27. und 28.11.2023 anlässlich der 29. Sächsischen Fachtagen für Umformtechnik Vertreter aus Industrie und Forschung Ihre Entwicklungen und Forschungsarbeiten im Bereich der Umformtechnik in Dresden vorstellen. Die Professur Formgebende Fertigungsverfahren der Technischen Universität Dresden lädt dazu alle interessierten Fachbesucher in das Maritim Hotel & Internationales Congress Center Dresden nahe der Altstadt ein.

Manufacturing equipments in semiconductor factories (fabs) provide abundant data and opportunities for data-driven process monitoring and modeling. In particular, virtual metrology (VM) is an active area of research. Traditional monitoring techniques using univariate statistical process control charts do not provide immediate feedback to quality excursions, hindering the implementation of fab-wide advanced process control initiatives. VM models or inferential sensors aim to bridge this gap by predicting of quality measurements instantaneously using tool fault detection and classification (FDC) sensor measurements. The existing research in the field of inferential sensor and VM has focused on comparing regressions algorithms to demonstrate their feasibility in various applications. However, two important areas, data pretreatment and post-deployment model maintenance, are usually neglected in these discussions. Since it is well known that the industrial data collected is of poor quality, and that the semiconductor processes undergo drifts and periodic disturbances, these two issues are the roadblocks in furthering the adoption of inferential sensors and VM models. In data pretreatment, batch data collected from FDC systems usually contain inconsistent trajectories of various durations. Most analysis techniques requires the data from all batches to be of same duration with similar trajectory patterns. These inconsistencies, if unresolved, will propagate into the developed model and cause challenges in interpreting the modeling results and degrade model performance. To address this issue, a Constrained selective Derivative Dynamic Time Warping (CsDTW) method was developed to perform automatic alignment of trajectories. CsDTW is designed to preserve the key features that characterizes each batch and can be solved efficiently in polynomial time. Variable selection after trajectory alignment is another topic that requires improvement. To this end, the proposed Moving Window Variable Importance in Projection (MW-VIP) method yields a more robust set of variables with demonstrably more long-term correlation with the predicted output. In model maintenance, model adaptation has been the standard solution for dealing with drifting processes. However, most case studies have already preprocessed the model update data offline. This is an implicit assumption that the adaptation data is free of faults and outliers, which is often not true for practical implementations. To this end, a moving window scheme using Total Projection to Latent Structure (T-PLS) decomposition screens incoming updates to separate the harmless process noise from the outliers that negatively affects the model. The integrated approach was demonstrated to be more robust. In addition, model adaptation is very inefficient when there are multiplicities in the process, multiplicities could occur due to process nonlinearity, switches in product grade, or different operating conditions. A growing structure multiple model system using local PLS and PCA models have been proposed to improve model performance around process conditions with multiplicity. The use of local PLS and PCA models allows the method to handle a much larger set of inputs and overcome several challenges in mixture model systems. In addition, fault detection sensitivities are also improved by using the multivariate monitoring statistics of these local PLS/PCA models. These proposed methods are tested on two plasma etch data sets provided by Texas Instruments. In addition, a proof of concept using virtual metrology in a controller performance assessment application was also tested.

No description available.