Model based operation of industrial steam methane reformers using large scale sensor data
MetadataShow full item record
Large quantities of hydrogen are consumed in refineries and for production of important chemicals such as ammonia and methanol. Declining crude-oil quality and increased fertilizer demands, among others, have led to further increase in hydrogen demand. A significant portion (~80%) of industrial hydrogen consumption is met via natural-gas steam methane reforming. This process takes place in a large scale, high-temperature, and highly energy-intensive unit called a steam methane reformer (SMR), where endothermic reforming reactions are carried out in hundreds of catalyst-filled tubes placed in a gas-fired furnace. A typical modern hydrogen production plant consumes a substantial amount (~10^5 GJ) of natural gas per day. The overall productivity (energy consumed per unit H2 produced) of the plant is strongly dependent on how efficiently the SMR is operated, which further depends on the spatial temperature distribution inside the furnace, where a more uniform distribution paves the way for reduced plant-wide energy use. Controlling the temperature distribution is, however, a challenging task due to the distributed nature of the system and the difficulty of obtaining distributed temperature measurements (the latter associated with the extreme operating conditions and the complex geometry of the furnace). In this thesis, results concerning the monitoring of temperature distribution in an industrial SMR furnace using a large array of infrared camera sensors, which produce a significant stream of data regarding the furnace temperature distribution, are presented. Specifically, strategies for homogenization of reformer tube-wall temperature distribution, also called furnace balancing, using reduced-order and physics-based models are developed. First, for a proof-of-concept study, a computational fluid dynamics (CFD) model of a small scale SMR system is developed as a substitute for a real plant. A proper orthogonal decomposition-based reduced-order linear model is used to modulate the fuel distribution among the burners. It is shown that a reduced-order empirical model with much lower computational requirements, when trained with sufficiently rich data, can be a viable substitute to the detailed modeling of the complex thermal and flow interactions in the furnace. Next, the data-driven modeling approach is extended to a real full-scale industrial SMR furnace. Shortcomings in popular empirical modeling approaches such as partial least squares (PLS) and ordinary least squares (OLS) are highlighted and a novel egg-crate SMR (EC-SMR) model is proposed. The model is calibrated using temperature measurements from the infrared cameras. Experimental results confirm that the proposed framework has excellent performance providing a $44\%$ improvement in temperature distribution non-uniformity. While computationally intensive CFD models are not suitable for use in furnace efficiency optimization calculations, empirical models (data-driven reduced-order models) have limited accuracy when wide changes in operating conditions are required. To overcome these limitations, a physics-based furnace model is presented that relaxes the assumption of identical temperatures for all the tubes which is commonly employed in the currently available literature. Supported by distributed temperature sensing from infrared cameras, an empirical scheme is used to explain the spatially non-uniform tube-wall temperature distribution. Reasonably low computational time makes the model suitable for online optimization purposes. Further, a multi-resolution model of a complete hydrogen plant is developed that includes the high-resolution physics-based model of the furnace, and low-resolution models, adequate for the purpose of plantwide optimization, of other unit operations of the plant. The model is then used for maximization of the plant thermal efficiency post furnace balancing. Reasonable computational time makes the developed optimization scheme suitable to be included as part of a hydrogen plant start-up procedure. In the above furnace balancing approaches, while the fuel distribution among the burners is manipulated to control the furnace temperature distribution, adequate temperature measurements are a prerequisite. Typical furnaces have hundreds of tubes and burners, and economic considerations dictate that the number of temperature sensors and flow actuators required for automatic temperature optimization be minimized. In this thesis, several formulations for the design of the optimal sensor and actuation configurations for an industrial furnace are investigated. The optimal sensor placement problem is initially formulated as a bi-level optimization problem, and the problem structure is exploited to obtain an equivalent MILP formulation. An extension to the combined sensor and actuator placement is then provided. The efficacy of the proposed approach is demonstrated through simulation case studies based on industrial data. This work comprises of multiple firsts, including the only open-literature discussion of using continuous massive scale image information for the monitoring of an industrial SMR, as well as the first application of distributed-parameter control of steam methane reforming. Furthermore, the implementation of the monitoring and control algorithms in a readily-deployable smart-manufacturing (SM) computational infrastructure is reported.