Introduction

We are living through a digital revolution, often referred to as ‘Industry 4.0’, where production operations are being connected to smart digital technologies such as machine learning, artificial intelligence and big data. The drive is to connect these elements and derive real-time insights across the full spectrum of operations. The timing of this revolution coincides with advances in supportive technologies such as data science, where more data of higher quality is being collected from a wider range of sources and computation, where more processing power is available at significantly lower cost. Data analytics also helps, with rapidly evolving technologies in machine leaning, artificial intelligence and data mining.

In applying Industry 4.0 principles, there is a push to capture knowledge through experience and data at each step of the process lifecycle. For this paper, we introduce how such thinking affects our approach to the ultimate safe and optimal operation of industrial reactors and in particular, the steam-methane reformer. The focus is model-based – using a ‘high-fidelity’ model which can be developed throughout the cycle and which ultimately embodies the deep process knowledge which it accumulates along the way, including elements from reactor development, scale-up, design and operation. As new information becomes available, the model will be updated accordingly, preserving the knowledge within the model itself. The ultimate goal is to ensure the final reactor as installed is safe and exhibits maximum profitability for the operator.

Reactor model development

To satisfy the objective of a digital approach to reactor development, a systematic approach is followed to develop the reactor model. The preferred route is to utilize a high-fidelity model which completely describes the behaviour of the reaction. The useful modelling capabilities that are required include:

• ability to execute steady-state and dynamic simulation and optimization
• parameter estimation and data analysis capabilities to help determine accurate reaction kinetic parameters and heat-transfer coefficients from experimental and pilot data
• results management and visualization capabilities.

With a validated reaction kinetics model established from laboratory and/or pilot trials, further layers of detail can be added to cover the effects of catalyst loading, design arrangement, heat-transfer properties and pressure drops.

At the end of the reactor model development, the Digital Reactor model now contains enough knowledge to predict reaction and thermal performance for all anticipated reactor operating conditions. The model can now be used to optimize the operation of the commercial-scale reactor.

Steam-methane reforming (SMR) process

A detailed equation-based model has been built for a small steam-methane reforming process generating hydrogen for use in transportation. Such small-scale reactors are predicted to grow significantly in number as the hydrogen economy grows. The performance of the whole SMR process is described in detail, from catalyst behaviour to heat-transfer. The model has been tuned and validated against actual plant operating data so that it can be used in a digital twin role.

Reactor Operations

The SMR model has sufficient fidelity and predictive accuracy to be used on-line. It contains all the deep knowledge, developed through tuning against actual plant performance. The model can be deployed online, interfaced to plant data systems, to be used for the following applications:

• Data reconciliation. Plant data measurements are subject to drift and error. A confidence level can be applied to each plant measurement and the Digital Reactor model used to establish the ‘best-fit’ heat and material balance to the plant data.
• Soft-sensing. Many difficult or impossible-to-measure key performance indicators such as temperature and composition profiles within the reactor beds can instead be calculated using the model. Exposing such soft-sensed information leads to better decision-making such as improved control and enhanced safety.
• Long-term health monitoring. By combining a predictive model with live plant data, the model can be used to track and quantify drift in key process parameters such as catalyst activity.
• Real-time optimization. The Digital Reactor model can be used to maximize profit in real time, taking into account equipment and product quality constraints as well as current product demands, product prices, utility costs and feedstock costs.
• Run-length prediction. The Digital Reactor model can sometimes include forecasting to predict, for example, end-of-run dates when a catalyst bed may need regenerating or replacing.

To be able to achieve these modelling objectives, a standard IT solution framework is required which allows the model to run robustly on-line. A Digital Applications Platform (DAP) is selected with the following elements:

• Execution Schedule Manager controls how the platform operates, manages the timing and execution of the components listed below and controls the way in which the system recovers and corrects for errors
• External Data Manager handles communications and collects information from external data historians, distributed control systems and other plant databases to feed to the data validation module
• Data Validation Module applies extensive checking of all data read from the external servers and applies any remedial action to correct missing or invalid data. This module also applies validation rules to any results calculated by the models in the platform.
• Computational Modules are wrappers for the Digital Reactor and other models which perform simulations with a discrete focus such as those outlined above (from data reconciliation onwards).

Typically, the DAP is mounted onto a physical or virtual computer located somewhere on the plant network. The plant data, either from the Distributed Control System or the data historian, are read by the DAP. Once validated, the data are passed to the Digital Reactor model where the various modelling objectives outlined above can be carried out. The results of the Digital Reactor model can be used in 2 ways:

• On-line closed-loop control. Some of the calculated results will be sent back to the DCS or data historian and the plant control system will take over, adjusting the plant set-points to reflect the calculated optimum settings
• On-line decision support. The results will be presented to the operator in an easily readable format and the operator will then have the choice to implement the results or not. Generally, this is the preferred choice, especially early on when the model needs to be tested to ensure it is giving the right information.

In the system used by the authors, the validated plant data and Digital Reactor calculation results are stored in a SQL database. This allows the results to be more easily made available to operators and it is compliant with Industry 4.0 as its contents are available for use by any other App residing on the plant network. The information for engineers and operators is most efficiently provided via dashboards which employ graphics and easy-to-view operator actions, advising the operator where the plant is now and where it should be in the future.

Conclusions

The key take-aways from this paper are:

• Progress in digital technologies such as computation, data science and analytics are driving the digitalization of our approaches. The key emphasis is to use data to embed deep process knowledge into our models and to allow models to ‘learn’ as they are used to predict process performance
• Digital models should be developed for the whole process lifecycle, from R&D through to operations, to capture knowledge and always be using the full extent of this knowledge. The underlying software platform in which the Digital Reactor model is to be built should be carefully selected so as to have as much functionality as possible across the lifecycle
• A Digital Applications Platform (DAP) has been described which enables a reactor model to be run on-line as part of a standard approach. The DAP allows the model to access plant data, to allow it to be tuned and therefore maximizes its predictive accuracy. As a result, the model can be used as part of an advanced process control system or as an operations advisor in a decision support tool. In the latter, visualization is very important, to ensure the users’ needs are satisfied.

A real-life small-scale industrial SMR process has been used to demonstrate the techniques.