By Tom Korb, Ph.D., P.E.
The petrochemical and refining industries are facing unprecedented challenges to the current role of fired equipment in industrial processes. Political, social, legislative, and environmental factors are driving both economic and compliance pressures to reduce the reliance on and impact of carbon-intensive production methods.
The petroleum refining industry is facing the greatest economic pressures driven in part by the trend towards the “electrification of everything,” which has been most visible in the transportation sector with hybrid or fully electric cars becoming much more common. With Tesla paving the way, Ford is investing more than $50B in electric vehicles by 2026. GM has made a similar investment pledge and has announced aspirations to eliminate gas and diesel light duty vehicles from their portfolio by 2035. While not yet commercially viable, the transition toward electric and autonomous systems extends to the aviation sector as well. Joby Aviation, which has a current valuation of $4.3B, Lilium Air Mobility, and Volocopter are three examples of companies that are making significant progress in developing electric, autonomous, vertical takeoff and landing vehicles. However, due to the energy density challenge, aviation is likely to lag 10 – 20 years behind ground-based transportation. The acceleration of these trends will continue to drive a decline in demand for hydrocarbon-based transportation fuels and increase economic pressures on the refining industry. The COVID-19 pandemic amplified the trend and has resulted in the permanent closure of several refineries and the development of plans for conversion of some sites to other value streams.
While the refining industry is possibly the most heavily and directly impacted economically, all operators of fired equipment are facing increasing pressures resulting from a global focus on reducing carbon and other greenhouse gas emissions. Greenhouse gas emissions from industrial processes account for 23% of all global greenhouse gas emissions. Together, industry (23%), transportation (29%), and electricity (25%) produce 77% of global greenhouse gas emissions. Carbon emissions from industry and electricity are produced primarily through the direct combustion of fossil fuels in fired equipment. Carbon emissions from transportation are produced through the combustion of fossil fuels in internal combustion engines and from the fired equipment used in the production processes for hydrocarbon-based transportation fuels. Consequently, efforts to reduce carbon emissions will be focused on these three sectors.
The preceding discusses how five of the six factors of a PESTLE (political, economic, social, technical, legislative, and environments) analysis are threats or head winds for industries that rely heavily on the combustion of fossil fuels in fired equipment. The sixth component of the PESTLE analysis, “technology” is unique because it is almost always both a threat and an opportunity. Disruptive products and processes that either eliminate the need for combustion systems (e.g., alternate reforming or cracking, fuel cells, solar, wind) or reduce the societal impact of combustion systems (carbon sequestration, etc.) will likely have a significant impact in the long term. However, optimization of the operational base-case provides many opportunities to make significant economically viable improvements in the immediate and near term. Digital technologies, and the many solutions enabled by them, play a critical role in both long-term disruptive technologies and in solutions that can produce significant immediate impact.
The current state of constant connectivity, combined with low-cost processing and data storage, lies at the root of the current digital or “data” revolution. Constant high-speed connectivity has enabled access to real-time data on a scale not previously possible. Low-cost compute and data storage capability have made it possible to apply computationally intensive analytical methods such as “machine learning” to massive datasets to produce practical solutions to real-world problems that have too many degrees of freedom and are too complex to solve via first principles or deterministic algorithms. The image processing algorithms used in automotive anti-collision and driver-assist systems, and natural language processing used in Alexa, Siri and Bixby are examples of such solutions that are in widespread commercial and consumer use today.
Relative to consumer and commercial products, the integration of advanced digital technologies into industrial processes has been much slower, likely due to the greater consequences of failure and accompanying risk profile. However, in recent years this has started to change rapidly and the promise of “Industry 4.0” or the Industrial Internet of Things (IIoT) is beginning to materialize. Due to the “size of the prize”, there is currently an enormous amount of investment being made in this space. GE reportedly spent approximately $7B on Predix, Honeywell spent more than $2B on Honeywell Forge to date, investment in C3.ai exceeded $1B in just six months following the IPO, and all the major digital companies (Google, Amazon, Microsoft, etc.) have major industrial programs and partnerships. There are also hundreds of smaller startup companies attempting to address various portions of this market. Solutions available to industrial operators range from instruments like Siemens, Yokogowa or ZoloSCAN tunable diode lasers, to first generation remote monitoring and diagnostics options like Atonix Asset 360, to data analytics and controls engines like H2O.ai, Seeq, Kelvin, to large-scale machine learning based platforms like Imubit, C3.ai, and Algorithmica Technologies. OEMs are also providing “Smart” or “Connected” solutions to augment their legacy products and services.
Some major industrial operators are already executing plans to implement live monitoring, anomaly detection, diagnostics, event prediction, etc. on every major process unit and significant piece of equipment in their fleet. The integration of Artificial Intelligence (AI), in the form of deep learning process control (DLPC), into closed loop control systems has also resulted in millions of dollars of cost reduction and value generation in major processing units that have been in operation for decades.
While digital transformation is an opportunity for industrial plant operators, it also presents some significant challenges. One challenge, which is faced by all industrial solution providers, is simply keeping up with the accelerating rate of change caused by this revolution. The changes that have occurred over the last two decades in the automotive world are now occurring in the industrial world. Just as a new car without sophisticated electronic features and controls would not be considered complete, soon any industrial process or equipment solution that doesn’t leverage digital technology to optimize the base case and maximize the return on physical upgrades, won’t likely be a competitive solution.
Many of the companies now developing digital solutions for industry are strong in software development, analytics, electronics, etc. but often have limited knowledge of industrial equipment or processes and of the struggles industrial owner/operators face while operating their plant. As a result, some of these digital companies are struggling to successfully solve problems for their customers and commercially scale their solutions. The companies that are successful in creating value for industrial owner/operators are leveraging deep subject matter expertise and knowledge of their customer’s world in combination with an understanding of how to develop and apply digital solutions to solve specific high-value problems for their customers. This is especially true when considering digital solutions related to specialized equipment like fired heaters and boilers vs. plant-wide historians, SCADA systems, etc.
Industrial operators, who are experts in running their plants, and not necessarily subject matter experts in specific types of equipment or digital solutions, may be overwhelmed by the number of options available to them and the rate of evolution of the landscape. While some of the majors have made significant investments in internal capabilities to drive digital transformation of their operation, many industrial owner/operators need help navigating this complex change.
Based on experiences in our personal lives, it is tempting to conclude that a digital solution that utilizes sophisticated AI would not require the assistance of human subject matter experts, just let the AI solve the problem. However, depending on the type of equipment, the governing physics, and the environment in which it is installed, it may not be possible for AI alone to understand, control, and optimize equipment performance. Like humans, in the absence of fundamental principles, AI learns empirically and cannot know something to which it has not been exposed. Consequently, AI is only as good as the data from which it learned, and this is where the difficulty lies in using AI to optimize unique and specialized industrial equipment such as a fired heater.
An almost incomprehensible amount of labeled data is publicly available for training image processing, natural language processing, or large language models like ChatGPT. Teaching a machine learning model to optimize the performance of a complex piece of process equipment like a fired heater is much more difficult than teaching it to recognize a cat, a bus, or human face for several reasons. Every fired heater and the circumstances in which it operates are unique and the operating data are not publicly available. A plant may have many years of operating data to train a model for their heater, but it often doesn’t include critical input parameters that have not been historically recorded like manual air damper settings or an indication of how many burners are operating. Plants work very hard to operate consistently and, as a result, there may not be enough variability in the controllable parameters for the model to understand the response resulting from a given change. Finally, the operating data range used to train the model may not include data at the optimal operation for a given set of conditions. As a result, if fundamental scientific principles are not included in the model, it will only be capable of local optimizing within the operating range that was used to empirically train the model and cannot extrapolate outside the learned range to a global optimum.
The good news is that all the learning data challenges mentioned above can be overcome by integrating first principles and subject matter expertise into the digital solution and model development. Thus, the key to successfully selecting and implementing digital and machine learning solutions aimed at improving processing equipment performance is to combine deep subject matter expertise with the digital and data analytics capabilities.
In addition to deep subject matter expertise in fired equipment and heat transfer, XRG Technologies has experience successfully deploying digital solutions to significantly improve fired heater operation. XRG can help customers evaluate their current equipment operation, evaluate what types of digital solutions can improve performance, specify performance targets, evaluate potential solutions, assist with selection, and support the installation and commissioning.