Loading field map
Process Systems Engineering (PSE) is the branch of chemical engineering concerned with the holistic design, operation, control, and optimization of chemical processes and the larger systems in which they are embedded. Where traditional chemical engineering focuses on the behavior of individual units—a reactor, a distillation column, a heat exchanger—PSE focuses on how these units interact as an integrated system, and how that system performs over time under varying conditions. Its central questions concern how to allocate limited resources, how to manage uncertainty and variability, how to coordinate multiple objectives, and how to make decisions that are robust across a range of possible futures.
Chemical engineering emerged in the late nineteenth and early twentieth centuries around the concept of unit operations—the idea that diverse industrial processes share common physical steps, such as distillation, filtration, or heat exchange. This framing allowed engineers to study each operation in isolation and then scale it up. But by the mid-twentieth century, it became clear that optimizing each unit independently did not optimize the whole plant. A change in one unit's operating conditions propagates through the network of interconnected units, affecting product quality, energy consumption, and safety elsewhere. Moreover, plants do not run at a single steady state; they face feed variations, equipment degradation, market price shifts, and regulatory changes. The need to address these system-level issues gave rise to PSE.
The field is defined less by a specific physical phenomenon than by a methodological stance: it treats the chemical plant, and often the broader supply chain or energy network, as a dynamic, interacting system that can be represented mathematically and optimized computationally. This stance unites several distinct activities that are often grouped under the PSE umbrella: process design, process control, process operations, and process synthesis.
The intellectual roots of PSE lie in the post-World War II era, when the increasing availability of digital computers made it possible to solve the large systems of equations that describe interconnected process units. Early work in the 1950s and 1960s focused on steady-state process simulation—computing the material and energy balances of an entire flowsheet given the design of each unit. This was a natural extension of unit operations thinking, but it required new numerical methods for solving large nonlinear equation systems and new ways of representing flowsheet topology.
A second major strand emerged from process control theory, which had developed during the war for military applications and was adapted to chemical processes in the 1960s and 1970s. Chemical processes are notoriously difficult to control because they involve large time delays, nonlinear dynamics, and strong interactions between variables. The development of model predictive control (MPC) in the 1970s and 1980s was a landmark: instead of tuning a controller to respond to a single measured error, MPC uses a dynamic model of the process to predict future behavior and compute a sequence of control actions that optimizes performance over a future horizon. This approach, originally developed in industry, became a cornerstone of modern process control.
A third strand concerned process synthesis—the problem of generating the structure of a process, not just analyzing a given structure. Given a set of raw materials and desired products, what sequence of reactors, separators, and recycle streams should be used? This is a combinatorial problem of enormous complexity. The field of synthesis developed systematic methods, including hierarchical decomposition (breaking the design problem into levels from reactor selection to heat integration), evolutionary modification (starting from a base design and improving it), and mathematical programming approaches that formulate the synthesis problem as an optimization problem over discrete and continuous variables.
By the 1980s, these strands were increasingly unified under the banner of PSE. The field's identity was reinforced by the founding of dedicated journals, conferences, and academic groups, and by the recognition that the same mathematical tools—optimization, dynamic simulation, numerical analysis—underlie all of its activities.
PSE is not organized around a single paradigm but around a set of complementary approaches that address different aspects of the system-level problem. These approaches are best understood in terms of the problem each addresses and the assumptions each makes.
The most pervasive approach in modern PSE is the formulation of engineering decisions as optimization problems. The engineer defines decision variables (e.g., reactor temperatures, flow rates, equipment sizes), an objective function (e.g., profit, energy consumption, environmental impact), and constraints (e.g., safety limits, product purity requirements, physical laws). The task is then to find the values of the decision variables that optimize the objective while satisfying the constraints.
This approach has been enormously successful because it provides a rigorous, systematic way to compare alternatives. Its power depends on the structure of the resulting mathematical problem. Linear programs can be solved efficiently even at large scale; nonlinear programs require more care; mixed-integer problems, which arise when decisions are discrete (e.g., whether to install a unit, which of several technologies to select), are harder still. A significant portion of PSE research has been devoted to developing algorithms that exploit the special structure of process engineering problems—for example, the sparsity of the equations, the presence of recycle streams, or the hierarchical nature of design decisions.
The optimization approach has a characteristic limitation: it requires a model, and the model is always a simplification. If the model omits important phenomena, or if the parameters are uncertain, the optimal solution may be poor in practice. This has led to the development of robust optimization and stochastic programming, which explicitly account for uncertainty in parameters or future conditions, and to a continuing tension between model fidelity and computational tractability.
Underlying all PSE activities is the construction of mathematical models of process units and their interconnections. Simulation—the computation of a process's behavior given its design and inputs—is the most fundamental PSE tool. Steady-state simulators, which solve the algebraic equations describing a process at equilibrium, are standard in industry for design and debottlenecking. Dynamic simulators, which solve differential-algebraic equations to capture time-varying behavior, are used for control design, safety analysis, and operator training.
Modeling in PSE is not a single method but a spectrum of practices. At one end are first-principles models based on conservation laws and physical chemistry; at the other are data-driven models built from plant measurements using statistical or machine-learning techniques. The choice depends on the purpose: first-principles models generalize better to new conditions but are expensive to build and may be inaccurate if the underlying physics is poorly understood; data-driven models are easier to construct but extrapolate poorly. Modern practice increasingly uses hybrid models that combine both, and the field has a long tradition of model reduction—simplifying complex models while preserving the behavior relevant to the decision at hand.
Process control addresses the problem of keeping a process operating at desired conditions despite disturbances. The classical approach, developed in the mid-twentieth century, uses feedback controllers that measure a variable (e.g., temperature) and adjust a manipulated variable (e.g., valve position) to correct deviations from a setpoint. The standard proportional-integral-derivative (PID) controller remains ubiquitous in industry because it is simple, robust, and well understood.
The system-level perspective of PSE becomes essential when processes have multiple interacting variables. A distillation column, for example, has multiple temperatures, flows, and compositions that affect one another; controlling one variable may disturb others. Multivariable control methods, and especially model predictive control, address this by coordinating all manipulated variables simultaneously based on a dynamic model. MPC's key advantage is its ability to handle constraints—for example, keeping a temperature below a safety limit while still optimizing throughput—which classical controllers cannot do directly.
Control is not a separate activity from design; the two are deeply intertwined. A process that is poorly designed may be impossible to control well, and a control system cannot compensate for fundamental design flaws. This insight has motivated simultaneous design and control approaches, which consider controllability during the design phase, though these remain computationally challenging and are not yet standard industrial practice.
Process synthesis is the creative part of PSE: generating the structure of a process. The synthesis problem is fundamentally different from analysis because the set of possible structures is enormous and not enumerable in practice. Systematic synthesis methods fall into several families.
Hierarchical methods break the design into levels, starting with the reactor network (since the reactor determines the species present), then the separation system, then the heat exchanger network, then the utility system. At each level, the designer makes decisions that constrain the next level. This approach, associated with the work of James Douglas and others, is intuitive and widely taught, but it can miss solutions that require coordination across levels.
Mathematical programming approaches to synthesis formulate the entire design problem as a mixed-integer optimization, where binary variables represent the presence or absence of units and continuous variables represent their sizes and operating conditions. This is conceptually appealing because it searches the entire space of alternatives, but the resulting problems are extremely hard to solve, and the approach is limited by the quality of the models and the computational resources available.
Evolutionary and heuristic methods start from a base design and modify it—adding or removing units, changing interconnections—guided by rules of thumb or by optimization algorithms such as simulated annealing or genetic algorithms. These methods are flexible and can handle complex constraints, but they do not guarantee optimality and their performance depends heavily on the initial design and the modification rules.
A particularly successful subfield is heat exchanger network synthesis, which seeks to minimize the total cost of heating and cooling utilities by maximizing heat recovery between process streams. The development of pinch analysis in the 1970s and 1980s provided a systematic, graphical method for identifying the minimum energy requirement of a process and designing networks to approach it. Pinch analysis is notable because it is simple enough to be applied by hand for small problems, yet rigorous in its identification of thermodynamic limits. It remains widely used in industry, and it illustrates a general principle of PSE: understanding the fundamental limits of a system is often more valuable than finding an optimal solution to an approximate model.
Beyond design and control lies the problem of operations: given an existing plant, how should it be run over time to maximize performance? This includes production scheduling (which products to make, in what order, for how long), maintenance planning, and the management of inventories and utilities. These problems are typically formulated as optimization problems over time, often with discrete decisions (e.g., whether to switch a batch reactor from one product to another).
Operations problems are distinguished from design problems by their time scale and by the presence of uncertainty. A design is made once and fixed; operations decisions are made repeatedly under changing conditions. This has led to the development of rolling-horizon approaches, where decisions are made for a near-term window and revised as new information arrives, and to the integration of scheduling with control, so that the control system's targets are consistent with the scheduling decisions.
A more recent development extends the PSE perspective beyond the plant boundary to the supply chain and the energy system. The same mathematical tools used for process design can be applied to the design and operation of a network of plants, warehouses, and markets, or to the planning of energy production and distribution. This enterprise-wide view treats the chemical plant as one node in a larger system, and it raises new questions about coordination across organizational and temporal boundaries.
This extension is a natural consequence of the PSE mindset: the field has always been defined by the question "how should this system be designed and operated?" rather than by a specific physical scale. As computational power has grown, the systems that can be addressed have grown from single units to entire plants to global networks.
Contemporary PSE is characterized by several durable features. First, it is computationally intensive: the field's progress has been driven by advances in algorithms and hardware, and its practitioners are typically as comfortable with numerical analysis and software development as with chemical engineering fundamentals. Second, it is model-centric: nearly every PSE activity begins with the construction of a mathematical model, and debates within the field often concern the appropriate level of model fidelity and the treatment of uncertainty. Third, it is application-driven: PSE methods are developed in response to industrial problems, and the field maintains close ties to the chemical, pharmaceutical, energy, and materials industries.
Several contemporary trends are reshaping the field. The increasing availability of data from sensors and historical records has renewed interest in data-driven modeling and in the integration of machine learning with first-principles models. The push toward sustainability has expanded the scope of PSE from economic optimization to multi-objective optimization that includes environmental impact, resource depletion, and social factors. The growth of renewable energy has created new problems in the design and operation of processes that must handle intermittent inputs, such as the production of fuels from variable wind or solar power. And the development of digital twins—dynamic, data-connected models of operating plants—promises to bring the PSE toolkit into real-time operations more directly than ever before.
These developments do not replace the core approaches described above; they extend and combine them. The field's identity remains rooted in the conviction that chemical processes are best understood and improved as integrated systems, and that rigorous mathematical methods, applied to well-constructed models, can guide decisions that would otherwise be left to intuition or trial and error. That conviction, and the methods built on it, constitute the durable contribution of process systems engineering to the broader discipline of chemical engineering.