Unlike many other techniques used in process control, which are widely applied in practice and play significant roles, abnormal situation management (ASM) still relies heavily on human experience, not least because the problem of fault detection and diagnosis (FDD) has not been well addressed. In this paper, a process fault diagnosis method using multi-time scale dynamic feature extraction based on convolutional neural network (CNN) consisting of similarity measurement, variable ranking, and multi-time scale dynamic feature extraction is proposed. The CNN-based model containing the fixed multiple sampling (FMS) layer can extract dynamic characteristics of process data at different time scales. The benchmark Tennessee Eastman (TE) process is used to verify the performance of the proposed method. K E Y W O R D S convolutional neural network, dynamic feature extraction, fault diagnosis, TE process

Wavelength selection is widely accepted as an important step in near-infrared (NIR) spectroscopic model development. In quantitative online applications, the robustness of the established NIR model is often jeopardized by instrument response changes, process condition variations or new sources of chemical variation. However, to the best of our knowledge, online updating of wavelength selection has not been considered in NIR modeling and property prediction. In this article, a new model-updating approach is proposed that can adjust to process changes by recursively selecting the NIR model structure in terms of wavelength. The advantage of the presented approach is that it can recursively adjust both wavelength selection and model coefficients according to real process variations. The performance of the method was tested on a spectroscopic data set from a refinery process. Compared with traditional PLS, locally weighted PLS, and several other updating strategies, the proposed method was found to achieve good accuracy in the prediction of diesel properties.

No abstract

Oil sands development is both a costly and technically complex business with potential concerns in land use, water consumption and greenhouse gas emissions. Therefore, it is of practical interest to further investigate novel techniques to improve profitability while diligently maintaining environmental compliance. Our approach for finding solutions to achieve this objective is to develop innovative strategies for advanced monitoring, optimisation and control of plant operations. Development of reliable process models is a key requirement for investigating the behaviour of complex systems. Such descriptive models can help to improve analysis, simulation, optimisation, design, control and operation of process systems at both micro and macro levels. This paper presents a summary of some of the successful applications focussed on development and implementation of inferential process models, also known as soft sensors for oil sands processes.Keywords: soft sensor, oil sands, bitumen extraction INTRODUCTIONT he very nature of oil sands extraction processes is to separate highly viscous bitumen from other components, which are mainly water and solids (e.g. clays and sand). This task is performed in multiple steps that involve: breaking the feed received from the mine into smaller chunks; screening out oversized materials; preparing slurry through addition of hot process water and separating bitumen, solids and water through a series of gravity and/or centrifugal separation units and usage of process aids. Despite the apparent simplicity of the overall process, different factors render the operation of the process quite complex. Some of these factors are: variation of the feed quality and rate; mechanical and electrical constraints related to equipment size; existence of multiple operating conditions; restrictions of downstream processes; and availability of process utilities such as hot process water or steam.Oil sands development is both a costly and technically complex business with potential concerns raising due to land use, water consumption and greenhouse gas emissions. Therefore, real-time process monitoring and control techniques are needed in many areas in order to ensure a smooth operation, achieve high throughput, stay close to the optimal boundaries of the process operating envelope, and minimise environmental footprints. An essential prerequisite for a successful process monitoring and control scheme is continuous collection of reliable and accurate data reflecting the equipment and plant operating status. However, extraction operations are naturally subject to harsh operating conditions with climate (ambient temperature and rain) playing a secondary role as an enhancement agent. Such conditions would affect reliability, accuracy and efficiency (i.e. service factor) of the instrumentation system. Moreover, measurements of quality variables (e.g. stream composition) are normally obtained through on-line quality analysers or off-line laboratory analysis with time delays of a few hours. Discontinuity and signif...

No abstract