Natural Gas Boiler Optimization to Improve Efficiency and Reduce Emissions

Improved boiler designs are critical in the quest to produce energy more efficiently and with reduced emissions of pollutants. Boilers are central to numerous industrial processes, particularly in producing steam and hot water for heating and power generation. Boiler systems seek to maximize heat transfer efficiency while reducing fuel consumption and emissions of CO, NO_x, and other pollutants.

This blog post will explain an advanced optimization technique developed by Bosch in collaboration with Cadence and the SU2 team for enhancing the performance of boiler heat exchangers while minimizing harmful emissions using a gradient-based adjoint approach with SU2 code and Fidelity Pointwise.

Design Objectives for Boilers

Boilers are heat exchangers that work on the principles of heat transfer, i.e., heating water by extracting heat from flue gas. Exhaust gases, primarily composed of CO₂ and water vapor, exit the appliance, carrying pollutants like CO and NO_x. The primary goals while designing boilers are

• Maximizing Heat Transfer: Achieving optimal efficiency to meet customer demands.
• Reducing CO and NO_x Emissions: Ensuring regulatory compliance when selling the boiler.

Impact of Heat Exchanger Design on Emissions

Experimental results show that the heat exchanger's shape significantly affects emissions. For instance, CO emissions were significantly reduced by designing the heat exchanger to use straight-cut fins instead of tapered fins. Experts employed both intuitive design and simulations for these adjustments.

Impact of heat exchanger design on CO₂ emissions

Simultaneous Reduction of NO_x and CO Emissions

The simultaneous reduction of NO_x and CO emissions is a complex task because they have inherently counteracting behaviors. NO_x emissions typically increase at higher combustion temperatures, while CO emissions tend to rise with incomplete combustion, often at lower temperatures. Therefore, finding a way to reduce both emissions demand a carefully optimized strategy. Alongside reducing emissions, it is crucial to enhance thermal efficiency and decrease the appliance's pressure drop to improve operational efficiency. The goal is to achieve these outcomes using a smart approach rather than relying on traditional trial-and-error methods. The adjoint-based optimization approach is a suitable method for achieving these objectives.

Deformation Approach: To Simplify Optimization Process

Test case geometry of a heat exchanger model

Before diving deep into the intricacies of adjoint optimization, it is essential to elucidate the method for managing the deformation process. The deformation approach leverages Free-Form Deformation Boxes (FFDB), illustrated in green in the image below. The core concept involves encapsulating the geometry to be modified within an FFDB. This technique facilitates a highly controllable, smooth, and robust deformation, even for highly complex geometries. FFDBs streamline the optimization problem by focusing on the translation of the FFDB vertices.

Free-form deformation box encapsulating the original geometry

Each translation can be uniquely characterized by the magnitude of movement in each spatial dimension. This technique simplifies the optimization problem, reducing it from potentially thousands of points on the geometry to just the translation of the FFDB vertices. In a particular scenario, an optimization problem may have around 108 design variables. This is because each vertex translation occurs in two dimensions within an FFDB composed of 18 vertices, resulting in 108 design variables.

Gradient-based Optimization: The Adjoint Method

A naive gradient-based approach requires conducting a large number of computational fluid dynamics (CFD) simulations (nDV + 1 baseline CFD simulation for a single design iteration), making it computationally expensive. Conversely, the adjoint method offers a promising alternative. This method enables the differentiation of the entire calculation from the optimization target to the displacement of the FFD vertices.

Utilizing the discrete adjoint method involves an iterative solution process for determining gradients rather than relying on precise analytical gradients. During baseline CFD simulation, derivatives for every mathematical operation are stored. Once the simulation converges, these gradients are propagated back to efficiently compute the derivatives for the design variables. This whole process only requires computational resources equivalent to two CFD simulations and is independent of the number of design variables.

Workflow of adjoint optimization with SU2 and Fidelity Pointwise

For Robust Deformation

There are two major challenges while using deformation, which are explained below, along with their possible solutions.

Challenge 1: Geometry Folding

One significant issue that can occur with large deformations is the generation of invalid geometries, which we call "geometry folding." This happens when two FFD vertices move in opposite directions with significant magnitudes, causing the geometry to overlap. This is a problem within our current framework that needs to be resolved.

Geometrical constraints should be introduced for the FFD box vertices to solve this issue. This approach will restrict the overall deformation process, increasing its robustness.

Invalid geometry due to geometry folding (left); FFD box around vertices to restrict deformation (right)

Challenge 2: Mesh Folding

Mesh folding is another issue that can arise during deformations. The mesh may fold towards the boundary, creating erroneous boundaries that complicate the meshing tool's ability to identify deformations correctly.

This issue can be addressed by performing deformations on a very coarse mesh. The coarse mesh can handle substantial deformation, allowing the generation of cleaner mesh at a later stage using Fidelity Pointwise.

Results from Single and Multi-Objective Optimization

Single objective optimization for different cases

While both singular pollutant-focused optimizations effectively reduced the respective pollutants, they led to trade-offs that negatively impacted overall system efficiency and environmental compliance.

In multi-objective optimizations, cases 9 and 10, as shown in the graph below (right), were able to optimize all targets. In case 10, the optimizer expanded the trailing edge of the heat exchanger, leading to cooling when most chemical conversion is complete, lowering the temperature without significantly increasing pollutants. It also created more space for CO to oxidize further to CO₂, diminishing CO levels. To reduce NO_x, the optimizer pushed the leading edge of the heat exchanger closer to the flame front, reducing NO_x formation.

Multi-objective optimization for different cases

By leveraging advanced optimization techniques like the gradient-based adjoint method, boiler efficiency and performance can be significantly improved. FFDB simplifies the complex task of geometric deformation, offering a robust and controlled approach to shape modifications. This intelligent, simulation-based method enables users to move beyond traditional trial-and-error practices, fostering innovations that meet both customer demands and regulatory compliance. Ultimately, these advancements contribute to the creation of boilers that are not only more efficient but can also reduce harmful emissions, underlining the substantial benefits of employing adjoint-based optimization in engineering design.

For more details, check out the recent publication in Fuel:

Daniel Mayer, Nijso Beishuizen, Heinz Pitsch, Thomas D. Economon, Travis Carrigan, Automatic adjoint-based design optimization for laminar combustion applications, Fuel, Volume 370, 2024, 131751, ISSN 0016-2361, https://doi.org/10.1016/j.fuel.2024.131751