2.5: System Linearization

Mechatronics: Fundamentals, Design, Integration, and Validation · LibreTexts

0%

Contents

    Text

    Size
    100%
    Theme
    Typeface
    Line length
    Spacing

    Read aloud

    How

    Speed: 1×
    Light up
    Highlight

    Symbols: Mulberry Symbols, © 2018–2026 Steve Lee, CC BY-SA 4.0 (licence, the set), the site's own copies.

    From LibreTexts, under CC BY-NC 4.0, as the source states it; its equations are written in TeX, which the PDF shows typeset. This page at LibreTexts.

    2.5: System Linearization
    Learning Objectives

    After completing this section, you will be able to:

    1. Describe the motivation of linearizing a nonlinear system model.
    2. Linearize a nonlinear system model with the given operational condition by following the six-step process described in this section.

    This section discusses how to linearize a nonlinear system model in the time domain. Most practical physical systems are nonlinear in nature. However, most analysis and design tools for mechatronic systems are limited to linear systems. Thus, we often need to convert a nonlinear system model into a linear one near its operational condition of interest. Before diving into linearization, we need to define a linear system with \(u(t)\) as input and \(y\left(t\right)\) as output.

    Definition 2.1

    If a dynamic system is linear, the principle of superposition is satisfied. Assume the input is the linear combination of \(u_1(t)\) and \(u_2\left(t\right)\), that is, \(α_1u_1\left(t\right)+α_2u_2\left(t\right)\). In this case, \(α_1\) and \(α_2\) are real numbers, and the associated system output is \(α_1y_1\left(t\right)+α_2y_2\left(t\right)\), assuming that \(y_1(t)\) and \(y_2\left(t\right)\) are time responses associated with inputs \(u_1(t)\) and \(u_2\left(t\right)\), respectively.

    Note that a nonlinear system does not satisfy the principle of superposition. Furthermore, a linear system with affine input does not satisfy the principle of superposition. Consider a linear system with affine excitation described by the differential equation \ref{2.135}.

    \[\dot{y}\left(t\right)+y(t)=au(t)+b \label{2.135}\]

    Let \(y\left(t\right)=y_1(t)\) and \(y\left(t\right)=y_2\left(t\right)\) be output time responses associated with inputs \(u\left(t\right)=u_1(t)\) and \(u\left(t\right)=u_2\left(t\right)\), respectively. In this case, the time response of system \ref{2.135}, with input \(u\left(t\right)=α_1u_1\left(t\right)+α_2u_2\left(t\right)\), is not equal to \(α_1y_1\left(t\right)+α_2y_2\left(t\right)\). They are equal when \(b=0.\) That is, a bias input \(b\) makes the system nonlinear.

    Now we would like to introduce the concept of a time-invariant system. The system response of a time-invariant system is not affected by time. Suppose \(y(t)\) is the system response with control input \(u(t)\). The left plot of Figure 2.39 shows the input is shifted by time \(T\mathrm{-}\mathrm{t}\mathrm{hat\ is,\ }u(t+T)\). In turn, the associated system response is \(y(t+T)\), as seen in the right plot of Figure 2.39.

    A system block diagram showing system time responses of the original and time-delayed inputs.
    Figure 2.39: Time response of a time-shifted (delayed) input.
    Definition 2.2

    For a time-invariant system, suppose \(y(t)\) is the system response with control input \(u(t)\). When the control input is shifted by time \(T\)—that is, \(u(t+T)\)— the associated system response is \(y(t+T)\).

    A linear system that is time-invariant is called a linear time-invariant (LTI) system; it can be represented by a transfer function in the form of a Laplace transform. An LTI system is easy to understand. It utilizes existing analysis and control design tools. Note that a linear system can be described by ordinary differential equations (ODEs). We can divided solutions of an ordinary different equation into two parts: homogeneous and particular responses.

    The homogeneous solution is related to the system transient response introduced by the system initial condition; the particular solution is the response caused by the forced (control) input. Due to the principle of superposition, solutions of linear combinations of multiple forced inputs can be solved individually and combined afterwards. In addition, system stability (see Chapter 3) can be easily analyzed based on the system transfer function by calculating its pole locations (roots of the denominator polynomial).

    A graphic representation of linearizing a nonlinear function at a given operational location marked with both global and local coordinates.
    Figure 2.40: Linearization of a nonlinear function.

    The actual physical system is always nonlinear, but it could be assumed to be linear over a small operational range. Often, a mechatronic system is designed to operate near the linear range. For example, a DC motor can be considered to be linear when it is operated within a certain power output range. When its power output is outside that range (e.g., beyond its output power rating), the DC-motor system is no longer linear. The main motivation of linearizing a nonlinear system is for easy control system analysis and design; many control design techniques are available only for linear system models. Note that nonlinear system models are difficult to deal with mathematically. In practice, we often linearize the nonlinear system model before conducting analysis and control design. Now the question is how to linearize a nonlinear system.

    Before discussing how to linearize a nonlinear system model, we will address the question of linearizing a nonlinear function \(f(x)\); see Figure 2.40. Let \(x_0\) be the operating location to be linearized. Deploying the Taylor expansion around the operating condition \(x_0\) leads to the following

    \[f\left(x\right)=f\left(x_0\right)+{\left.\dfrac{\partial f(x)}{\partial x}\right|}_{x=x_0}δx+HOT(δx) \label{2.136}\]

    where the term \(HOT(δx)\) in equation \ref{2.136} represents high order terms of \(δx=x-x_0\). That is,

    \[{\mathop{\mathrm{l}\mathrm{im}}_{δx\to 0} \dfrac{HOT(δx)}{δx}\ }=0 \label{2.137}\]

    When \(δx\) is small, we can approximate \(f(x)\) by setting \(HOT(δx)\) to zero.

    \[δf\left(δx\right)=f\left(x\right)-f\left(x_0\right)\approx {\left.\dfrac{\partial f(x)}{\partial x}\right|}_{x=x_0}δx \label{2.138}\]

    Because the above function is linear in \(δx\), we call it a linearization of \(f(x)\) at \(x_0\). Note that the linearized function \(δf\left(δx\right)\) is a function of \(δx\) (not a function of \(x\)).

    For a nonlinear dynamic system model with input \(u(t)\) and output \(y(t)\), we can use the following six-step process for linearization.

    Six-Step Linearization Process
    1. Identify the system model input \(u(t)\) and output \(y\left(t\right)\) with \({\dot{u}}_0=0,\) \({\dot{y}}_0=0,{\ddot{u}}_0=0,\ {\ddot{y}}_0=0,\dots\).
    2. Express the model differential equation in the form of \(f\left(u,\ \dot{u},\dots ,y,\dot{y},\dots \right)=0\) by shifting the terms on the right side to left.
    3. Find equilibrium operational points \(u_0\) and \(y_0\) and set all their corresponding derivatives to zero at the equilibrium point \(({\dot{u}}_0=0,{\ddot{u}}_0=0,\dots ,{\dot{y}}_0=0,\ {\ddot{y}}_0=0,\dots )\).
    4. Perform Taylor expansion about \((u_0,y_0)\) and retain only the first-order derivatives.
    5. Change the system variables to their incremental ones as below.

    \[δu=u-u_0,\ δ\dot{u}=\dot{u}-{\dot{u}}_0,\ \dots ; \tilde{y}=y-y_0,\ \dot{\tilde{y}}=\dot{y}-{\dot{y}}_0,\ \dots \label{2.139}\]

    We will use an example below to illustrate how to use the six-step process to linearize a nonlinear system model.

    Example 2.16

    For the pendulum system shown in Figure 2.41, derive the nonlinear system equation of motion, linearize it at \(θ_o=π\) (an inversed pendulum), and find the transfer function \(G\left(s\right)\) from \(ΔT(s)\) to \(ΔΘ(s)\).

    A mechanical diagram of a pendulum system for Example 2.16 with control torque as input and angular displacement as output.
    Figure 2.41: Pendulum system of Example 2.16.

    Using Newton’s law, we have the following equation of motion.

    \[ml^2\ddot{θ}\left(t\right)+mgl{\mathrm{s}\mathrm{in} θ(t)\ }=τ(t) \label{2.140}\]

    Now, we will linearize equation \ref{2.140} around \(θ_o=π\) using the six-step process.

    1. Identify the system model input \(τ(t)\) and output \(θ\left(t\right)\).
    2. Express the model in the form of \(f\left(τ,\ \dot{τ},\dots ,θ,\dot{θ},\dots \right)=0\).

    \[f\left(θ,\ddot{θ},τ\right)=ml^2\ddot{θ}\left(t\right)+mgl{\mathrm{s}\mathrm{in} θ(t)\ }-τ\left(t\right)=0 \label{2.141}\]

    1. Find the equilibrium operation points \(θ_0\) and \(τ_0\) with \(θ_0=π,\) \({\ddot{θ}}_0=0\).

    \[ml^2{\ddot{θ}}_0+mgl{\mathrm{s}\mathrm{in} θ_0\ }-τ_0=0\ \to τ_0=0 \label{2.142}\]

    1. Perform Taylor expansion about \((τ_0,θ_0,{\ddot{θ}}_0)\) with \(τ\left(t\right)=τ_0+δτ\left(t\right)\), \(θ\left(t\right)=θ_0+δθ\left(t\right)\), and \(\ddot{θ}\left(t\right)={\ddot{θ}}_0+δ\ddot{θ}\left(t\right)\), and retain only the first-order derivatives.

    \[\begin{array}{ccc} f\left(θ,\ddot{θ},τ\right) & \cong & \overbrace{f\left(θ_0,{\ddot{θ}}_0,τ_0\right)}^{0}+{\left.\dfrac{\partial f}{∂θ}\right|}_{x_0}δθ+{\left.\dfrac{\partial f}{\partial \ddot{θ}}\right|}_{x_0}δ\ddot{θ}+{\left.\dfrac{\partial f}{∂τ}\right|}_{x_0}δτ \\ & = & {\left.mgl{\mathrm{c}\mathrm{os} θ\ }\right|}_{θ=π}\left(θ-π\right)+{\left.ml^2\right|}_{\ddot{θ}=0}\left(\ddot{θ}-0\right)-{\left.1\right|}_{τ=0}(τ-0) \\ & = & -mgl\left(θ-π\right)+ml^2\left(\ddot{θ}-0\right)-1\left(τ-0\right) \ \end{array} \label{2.143}\]

    1. Change the system variables to their incremental ones for equation \ref{2.143}.

    \[f\left(θ,\ddot{θ},τ\right)=ml^2δ\ddot{θ}-mglδθ-δτ=0\ \label{2.144}\]

    1. Rewrite the function defined above into the ODE below and conduct the Laplace transform.

    \[ml^2δ\ddot{θ}-mglδθ=δτ \to \ ml^2s^2ΔΘ(s)-mglΔΘ(s)=ΔT(s) \label{2.145}\]

        We have the following linearized transfer function.

    \[G\left(s\right)=\dfrac{ΔΘ(s)}{ΔT(s)}=\dfrac{1}{ml^2s^2-mgl} \label{2.146}\]

        With the linearized system represented by the transfer function \(G(s)\), the relationship between the original nonlinear and linearized systems can be shown in Figure 2.42.

    Two block diagrams showing the relationship between a nonlinear system and its linearized one.
    Figure 2.42: Relationship between nonlinear and linearized systems.

    Note that the linearized model represented by the transfer function \(G(s)\) describes only the system relationship from \(δτ(t)\) to \(δθ(t)\).

    Example 2.17

    Find the nonlinear differential equation described by the following circuit in Figure 2.43, linearize it at \(v^0_i=10v\), and find the transfer function \(G\left(s\right)\) from \(ΔV_i(s)\) to \(ΔV_o(s)\), where \(R_1=1Ω\), \(R_{21}=1Ω\), \(R_{22}=200Ω\mathrm{/}A^2,\) and \(C=0.1F\).

    A schematic circuit diagram for Example 2.17 with two series-connected resistors connected to a capacitor in series where one of resistors is a nonlinear function of current.
    Figure 2.43: A nonlinear circuit for Example 2.17.

    Using Kirchhoff’s voltage law, we can calculate the output voltage \(v_o(t)\) based on the capacitance formula by the following equation, assuming that \(v_o\left(0\right)=0\).

    \[v_o\left(t\right)=\dfrac{1}{C}\int^t_0{i\left(τ\right)dτ} \to i\left(t\right)=C{\dot{v}}_o(t) \label{2.147}\]

    and \(v_i(t)\) is

    \[v_i\left(t\right)=\left(R_1+R_{21}\right)i\left(t\right)+R_{22}i^3\left(t\right)+v_o(t) \label{2.148}\]

    Substituting equation \ref{2.147} into \ref{2.148} yields the following.

    \[v_i\left(t\right)=\left(R_1+R_{21}\right)C{\dot{v}}_o(t)+R_{22}C^3{\dot{v}}^3_o\left(t\right)+v_o(t) \label{2.149}\]

    In turn, substituting \(R_1=1Ω\), \(R_{21}=1Ω\), \(R_{22}=200Ω\mathrm{/}A^2,\) and \(C=0.1F\) into the above equation leads to the following nonlinear differential equation.

    \[v_i\left(t\right)=0.2{\dot{v}}_o(t)+0.2{\dot{v}}^3_o\left(t\right)+v_o(t) \label{2.150}\]

    Now, we are ready to linearize the above nonlinear equation at \(v^0_i=10V\) using the six-step process.

    1. Identify the system model input \(v_i(t)\) and output \(v_o\left(t\right)\).
    2. Express the model in the form of \(f\left(v_i,~{\dot{v}}_i,\dots ,v_o,{\dot{v}}_o,\dots \right)=0\).

    \[f\left(v_i,v_o,{\dot{v}}_o\right)=0.2{\dot{v}}_o(t)+0.2{\dot{v}}^3_o\left(t\right)+v_o(t)-v_i\left(t\right)=0 \label{2.151}\]

    1. Find equilibrium operation point \(v^0_o\) with \(v^0_i=10\mathrm{V},\) \({\dot{v}}^0_o=0\).

    \[0.2\times 0+0.2\times 0^3+v_o(t)-10=0\ \to v^0_o=10V \label{2.152}\]

    1. Perform Taylor expansion about \(x_0=(v^0_i,v^0_o,\ {\dot{v}}^0_o)\) with \(v_i\left(t\right)=v^0_i+δv_i\left(t\right)\), \(v_o\left(t\right)=v^0_o+δv_o\left(t\right)\), and \({\dot{v}}_o\left(t\right)={\dot{v}}^0_o+δ{\dot{v}}_o\left(t\right)\), and retain only the first-order derivatives.

    \[\begin{aligned} f\left(v_i,v_o,{\dot{v}}_o\right) & \cong \overbrace{f\left(v^0_i,v^0_o,0\right)}^{0}+{\left.\dfrac{\partial f}{\partial v_i}\right|}_{x_0}δv_i+{\left.\dfrac{\partial f}{\partial v_o}\right|}_{x_0}δv_o + {\left.\dfrac{\partial f}{\partial {\dot{v}}_o}\right|}_{x_0}δ{\dot{v}}_o. \\[4pt] & = \left(-1\right)\left(v_i-v^0_i\right)+\left(1\right)\left(v_o-v^0_o\right)+{\left.(0.2+0.6{\dot{v}}^2_o)\right|}_{\dot{v}^0_o=0}({\dot{v}}_o-{\dot{v}}^0_o) \\[4pt] & = -\left(v_i-v^0_i\right)+\left(v_o-v^0_o\right)+0.2\left({\dot{v}}_o-{\dot{v}}^0_o\right) \end{aligned} \label{2.153}\]

    1. Change the system variables to their incremental ones for equation \ref{2.153}

    \[f\left(v_i,v_o,{\dot{v}}_o\right)=-δv_i+δv_o+0.2δ{\dot{v}}_o=0\ \label{2.154}\]

    1. Rewrite the function defined above into the ODE below and calculate the Laplace transform.

    \[δv_o(t)+0.2δ{\dot{v}}_o(t)=δv_i(t)\ \to \ (1+0.2s)ΔV_o(s)=ΔV_i(s) \label{2.155}\]

        We have the following transfer function.

    \[G\left(s\right)=\dfrac{ΔV_o(s)}{ΔV_i(s)}=\dfrac{1}{0.2s+1} \label{2.156}\]

    Before ending this section, we will discuss how to linearize a time delay transfer function in the frequency domain. Consider the following time delay system.

    \[y(t)=x\left(t-T_d\right)u(t-T_d) \label{2.157}\]

    where \(T_d\) is the time delay and \(u(\cdot )\) is a unit step function. Applying the Laplace transform to the above equation, based upon Table 2.3, yields the following.

    \[Y\left(s\right)=e^{-T_ds}X\left(s\right) \to \ \dfrac{Y(s)}{X(s)}=e^{-T_ds} \label{2.158}\]

    With the help of Taylor expansion in the frequency domain around \(s=0\), we have

    \[e^{0.5T_ds}\ \approx {\left.e^{0.5T_ds}\right|}_{s=0}+{\left.\dfrac{\partial e^{0.5T_ds}}{\partial s}\right|}_{s=0}(s-0)=1+0.5T_ds \label{2.159}\]

    and

    \[e^{-0.5T_ds}\ \approx {\left.e^{-0.5T_ds}\right|}_{s=0}+{\left.\dfrac{\partial e^{-0.5T_ds}}{\partial s}\right|}_{s=0}\left(s-0\right)=1-0.5T_ds \label{2.160}\]

    As a result,

    \[e^{-T_ds}=\dfrac{e^{-0.5T_ds}}{e^{0.5T_ds}}\approx \dfrac{1-0.5T_ds}{1+0.5T_ds} \label{2.161}\]

    Applying the Taylor expansion to \(e^{T_ds}\) directly at \(s=0\), we have the following linearized approximation.

    \[e^{-T_ds}\approx \dfrac{1}{1+T_ds} \label{2.162}\]

    Note that the approximation in equation \ref{2.161} is preferred because the magnitude of the linearized model equals one at any frequency; this matches the original time delay transfer function. However, both approximated linear transfer functions \ref{2.161} and \ref{2.162} have the same phase characteristics, which are quite different from that of the original time delay transfer function, especially at high frequency.

    Section Summary

    This section introduces the concept of system model linearization. Note that all physical systems are nonlinear but are approximated or modeled as linear systems. When the system is modeled using a nonlinear model, it often needs to be linearized so that linear control theory can be applied. This section presents a six-step linearization process in time-domain, using Taylor expansion, to obtain the linearized model. Once the model has been linearized in time domain, the time-domain model can be converted into the frequency domain in the form of a transfer function by applying the Laplace transform.

    Section Keywords

    Practice Problems

    1. Find the equation of motion for the vehicle system below, where \(f_m\) is the vehicle input traction force, \(f_v\left(t\right)=K_vv(t)\) is the vehicle viscous (or rolling resistance) force, \(v(t)\) is vehicle speed, and \(f_d\left(t\right)=K_dv^2(t)\) is the vehicle aerodynamic drag force. Then linearize the nonlinear equation of motion at \(v_0=30\ m/sec\), and find the linearized transfer \(G(s)\) function from \(ΔF_m\left(s\right)\) to \(ΔV\left(s\right)\).

    image54.png

    1. For the equation of motion shown below (obtained by solving P2.5),
      1. Solve for the operating point \(x_0\) when \(f_{a0}=10N\) (assuming that \(x>0,\ {\dot{x}}_0={\ddot{x}}_0=0)\).
      2. Derive the linearized model as a function of \(\delta x=x-x_0\) and \(\delta f_a=f_a-f_{a 0}\), and associated transfer function \(G(s)=\Delta X(s) / \Delta F_a(s)\). \[2 \ddot{x}+6 \dot{x}+3 x+x^2=f_a(t) \nonumber\]

    This page titled 2.5: System Linearization was last modified on Sat, 22 Aug 2026 21:42:39 GMT and is shared under a CC BY-NC 4.0 license and was authored, remixed, and/or curated by Guoming Zhu (Co-DREAM OER) .