Linearization and deviation variables are required when a nonlinear pilot-plant model must be converted into a tractable local control model. Linearization approximates nonlinear equations around a chosen steady-state operating point, while deviation variables measure changes relative to that point. Together, they produce linear dynamic equations that support transfer functions, Laplace analysis, controller tuning, stability analysis, and process identification.

The essential purpose is to turn a nonlinear process description into a local model of how small disturbances affect the process around normal operation. This simplification is accurate only within a sufficiently small region around the selected operating point.

Why Nonlinear Unit Operations Need Simplification

Chemical processes are inherently nonlinear

Pilot-scale equipment such as CSTRs, heat exchangers, distillation columns, and liquid-level systems is governed by nonlinear relationships.

Examples include the reaction-rate expression:

[ k = k_0 e^{-E/(RT)} ]

and the liquid outflow relationship:

[ F_o = \beta \sqrt{h} ]

In these cases, a change in temperature or liquid level does not produce a proportional change in reaction rate or flow.

Nonlinear equations complicate control design

A nonlinear dynamic model generally cannot be represented by one fixed transfer function over all operating conditions.

Its response characteristics can change with temperature, concentration, level, flow rate, catalyst activity, or fouling. This makes standard linear control tools difficult to apply directly.

Pilot equipment still needs predictable operating behavior

Control experiments usually focus on keeping variables such as temperature, level, pressure, or concentration near a desired operating condition.

The practical question is therefore often local: How will the process respond to a small change in an input near its current steady state? Linearization provides an answer to that question without requiring the full nonlinear model during every analysis or design step.

What Linearization Provides

It approximates the nonlinear equations locally

Suppose a nonlinear process model is written as:

[ \frac{dx}{dt} = f(x,u) ]

where (x) represents process states and (u) represents manipulated inputs.

Around a steady-state condition ((x_0,u_0)), a first-order Taylor expansion gives:

[ \frac{dx'}{dt} \approx \left.\frac{\partial f}{\partial x}\right|_0 x' + \left.\frac{\partial f}{\partial u}\right|_0 u' ]

The nonlinear function is replaced by its local slope with respect to each process variable.

It creates a linear dynamic model

The resulting equation has the general form:

[ \frac{dx'}{dt} = A x' + B u' ]

This is a linear state-space model. For a single-input, single-output process, it can often be converted into a transfer function such as:

[ G(s) = \frac{Y'(s)}{U'(s)} ]

The model then becomes compatible with established frequency-domain and feedback-control methods.

It captures small-disturbance behavior

Near the nominal steady state, the first-order approximation often represents the actual pilot-plant response closely.

This is particularly useful for controller design because many feedback loops are intended to reject relatively small disturbances around a normal operating condition.

What Deviation Variables Provide

They remove the steady-state baseline

A deviation variable is defined as the difference between a measured value and its nominal steady-state value:

[ C_A' = C_A - C_{A,0} ]

[ T' = T - T_0 ]

[ u' = u - u_0 ]

At the nominal steady state, every deviation variable is zero.

They isolate the process response

Deviation variables separate the constant operating condition from the dynamic change caused by a disturbance or control action.

Instead of modeling the absolute temperature and its steady-state offset, the model focuses on the temperature change (T') caused by a change in heating input, flow, or feed conditions.

They simplify the dynamic equations

Consider a nonlinear level balance:

[ A\frac{dh}{dt} = F_i - \beta\sqrt{h} ]

At steady state:

[ F_{i,0} = \beta\sqrt{h_0} ]

Using (h = h_0 + h') and (F_i = F_{i,0} + F_i'), the square-root term can be approximated as:

[ \beta\sqrt{h} \approx \beta\sqrt{h_0} + \frac{\beta}{2\sqrt{h_0}}h' ]

After applying the steady-state relationship, the constant terms cancel. The resulting model describes only the deviations:

[ A\frac{dh'}{dt} + \frac{\beta}{2\sqrt{h_0}}h'

F_i' ]

This cancellation is one of the main reasons deviation variables are used.

How the Two Concepts Work Together

Linearization supplies the approximation

Linearization determines how the nonlinear process changes near the operating point.

For example, the local coefficient

[ \frac{\beta}{2\sqrt{h_0}} ]

is the slope of the nonlinear outlet-flow relationship at (h_0). It becomes a constant parameter in the local linear model.

Deviation variables supply the coordinate system

Deviation variables express the process relative to the operating point.

They ensure that the nominal steady state corresponds to the origin, which makes the linearized equations homogeneous in the process changes and simplifies initial conditions for dynamic analysis.

Together they enable transfer functions

Once the model is written in deviation form, Laplace transformation can be applied under zero initial deviation conditions.

For the level example:

[ A s H'(s) + \frac{\beta}{2\sqrt{h_0}}H'(s)

F_i'(s) ]

Therefore:

[ \frac{H'(s)}{F_i'(s)}

\frac{1}{ A s + \frac{\beta}{2\sqrt{h_0}} } ]

This transfer function relates a deviation in inlet flow to a deviation in liquid level.

Why This Matters for Control Design

Controller tuning depends on a usable process model

Methods for tuning proportional-integral-derivative controllers commonly require process gain, time constant, and delay estimates.

A local linear model provides these quantities in a form that can be analyzed and used for controller design.

Stability analysis becomes practical

Linear models allow engineers to study poles, zeros, frequency response, gain, phase margin, and closed-loop stability.

These tools provide direct insight into whether a proposed controller will produce stable or oscillatory behavior near the operating point.

Experimental identification becomes easier

Pilot plants are often used to estimate process parameters from measured input-output data.

Using deviation variables allows the experiment to focus on changes from steady state, making process gains and dynamic time constants easier to identify.

It supports practical control architectures

The same linearized representations can be used when developing cascade, ratio, split-range, or multivariable control strategies.

For interacting processes, local linear models also support decoupling and Relative Gain Array analysis.

Understanding the Trade-offs

The model is valid only near its operating point

A first-order Taylor approximation neglects higher-order terms.

If the temperature, level, concentration, or flow changes substantially, the linearized model may no longer predict the actual process accurately.

One model may not describe the entire operating range

A CSTR may have significantly different dynamics at different temperatures or feed concentrations.

Similarly, the effective gain of a square-root level system changes with (h_0). A controller based on one operating point can therefore perform poorly at another.

Nonlinear effects can affect stability and performance

Large disturbances may produce asymmetric responses, saturation, multiple steady states, or instability that a local linear model does not reveal.

The linearized model should therefore be treated as a design approximation, not as a replacement for the underlying nonlinear process model.

Fixed-parameter controllers may need adaptation

When operating conditions vary widely, gain scheduling, adaptive control, or repeated linearization may be necessary.

For pilot-plant training, this limitation is also valuable: it demonstrates why a controller that works well at one condition may require retuning elsewhere.

Linearizing data is not the same as linearizing dynamics

Transforming an empirical relationship, such as plotting mass flow against (\sqrt{T}), can help estimate parameters.

That procedure is different from deriving a local linear dynamic model for control. Both use simplification, but they address different purposes: parameter fitting versus dynamic response and feedback design.

Making the Right Choice for Your Goal

The appropriate modeling approach depends on the size of expected disturbances and the operating range of the pilot equipment.

If your primary focus is controller design near one steady state: Linearize the governing equations around that operating point and express all inputs, outputs, and states as deviation variables.
If your primary focus is large operating changes: Retain the nonlinear model or use multiple local models with gain scheduling or adaptive control.
If your primary focus is transfer-function analysis: Use deviation variables so the steady-state offsets cancel and the dynamic equations can be transformed using Laplace methods.
If your primary focus is parameter estimation from experiments: Remove the nominal baseline before analyzing the measured input-output changes, while verifying that the chosen mathematical transformation matches the physical relationship.
If your primary focus is understanding pilot-plant limitations: Compare the linearized prediction with the nonlinear model or measured response as the process moves farther from its nominal operating point.

Linearization and deviation variables make nonlinear pilot-plant control problems manageable while clearly defining the operating region in which the resulting model can be trusted.

Summary Table:

Concept	Purpose	Main Benefit	Limitation
Linearization	Approximates nonlinear equations around a steady-state operating point	Enables state-space models, transfer functions, and controller tuning	Accurate only near the selected operating point
Deviation variables	Express states, inputs, and outputs relative to nominal values	Removes steady-state offsets and simplifies dynamic equations	Must be updated when the operating point changes
Combined use	Converts nonlinear dynamics into a local linear model	Supports stability analysis, process identification, and feedback design	May require gain scheduling or adaptive control for wide operating ranges
Nonlinear model	Represents full process behavior	Captures large disturbances, saturation, and multiple operating conditions	More difficult to analyze and control directly

Need practical equipment for teaching and validating nonlinear process control? Contact LABPARK to discuss your requirements. LABPARK provides Educational and Vocational Unit Operations Pilot Plants for chemical engineering, bioprocess and biotechnology, and environmental and water treatment applications. Our pilot systems help universities, research institutes, and enterprises demonstrate steady-state behavior, linearization, deviation-variable modeling, controller tuning, process identification, and real-world control limitations. Tell us your target process, operating range, and training or research goals, and our team can help you select a suitable pilot plant configuration.