Hello everyone, This is Pon Adithi, ME22B029.

I will be presenting the topics- State space

<next slide>

Starting with the State space model of feed drive control systems, a state space model is a
mathematical framework that describes the dynamics of a system in terms of the state variables,
where state variables include position, velocity, current and more.

The purpose of this is usually to verify the validity of another mathematical model of the system,
which is derived either through first principles or experimental data and approximations.

The servo is divided into continuous and discrete sections for state-space modelling.

The continuous part of the system includes the velocity control loop, which ensures the motor
maintains a desired speed despite disturbances or changes in operating conditions and the up-
down counter, whose output is used to determine the tool's current position relative to a reference
point.

Three states can be derived - the armature current denoted by Ia, the angular velocity denoted by
omega, and the actual position denoted by Xa.

Starting with the armature current, an expression for the armature voltage can be developed from
both the block diagram of the feed drive control system and the electrical diagram of a DC
motor.

<next slide>

Here, in the block diagram, Vc is the velocity command input voltage. Multiplied by the
differential preamplifier gain Sg, it gives the net amplified command signal. This signal is
compared with the actual velocity feedback obtained from the tacho generator. Here, Tg is the
tachometer gain; Hg is the tachometer constant, which relates the motor's rotational speed and
output voltage. The difference between the two gives the velocity error signal.
The demand current can be obtained when the velocity error signal is multiplied by the current
amplifier gain Ki. The armature voltage can hence be expressed as the difference between the
demand current, which is this whole term, and the amplified armature current, multiplied by the
PWM amplification gain, Kv.

Here, Kirchoff's voltage law is applied in the electrical diagram of the DC motor, which gives
this expression for the armature voltage. Ra is the armature resistance; La is the armature
inductance; Kb is the motor's back emf constant.

Equating the two equations, we obtain this expression for the rate of change of the armature
current in terms of the angular velocity, velocity command voltage and the armature current
itself.

<next silde>

An expression for the angular velocity can be obtained from the expressions for the motor's
torque.

According to this mathematical relation, the motor's magnetic field produces a useful torque Tm
proportional to the armature current.

The motor's useful torque is mainly utilised in accelerating the reflected inertia on the motor's
shaft, overcoming the friction in the motor's bearings and guideways and resisting feed cutting
forces and friction loads reflected as a disturbance torque on the motor's shaft, which is
formulated mathematically here. Here, Je is the inertia reflected on the motor shaft, B is the
equivalent friction coefficient, and Ts is the static disturbance reflected on the motor shaft.

Equating the two equations expresses the rate of change of angular velocity.

Finally, the actual position can be obtained by applying the inverse Laplace transform to the
block diagram of the up-down counter.

<next slide>
Summarising the three state equations here, they can be combined into this standard state-space
form, where xc(t) is the state vector, containing the state variables Ia, omega and Xa, and uc(t) is
the input vector containing the velocity command input and the static disturbance torque.

Ac and Bc are constant matrices presented here.

<next slide>

The state equation has this form of discrete equivalent solution for an observation interval of T.
Here, k is the sampling interval, and the state and input vectors at the sampling interval are
shown.

This matrix phi(T) can be computed either from the eigenvalues of the Ac matrix or its Taylor
series expansion for the discrete-time equivalent of the continuous-time system.

In the case of Taylor series expansion, since the sampling interval usually lies between 0.1-1ms
for CNC systems, the first three terms of the approximation will suffice for most applications.

<next slide>

In the discrete section of the modelling, the components consist of the digital filter, represented
by D(z) and a Digital-to-Analog converter with gain Kd. The digital filter has a transfer function
in this form, where Kp is the position control gain filter, a is the filter's zero, and b is the filter's
pole.

Furthermore, the velocity command signal can be expressed in the z-domain, where Xr(k) is the
discrete reference position. Rearranging the terms, Vc can be written as ---------------, where
Vd(k) is -------------------------. Vc and Vd can now be treated as a fourth state. Taking the inverse
z-transform of the equation, we obtain an iterative expression for Vc and Vd.

<next slide>
Now, combining the discrete-time state equation and the discrete-time equivalent of the
continuous part of the feed drive servo, we obtain the complete state equations for the feed drive
as follows, where x(k) is the state vector, u(k) is the input vector and y(k) is the output vector.

G(T) is the state matrix, Gamma(T) is the input matrix, Cs is the output matrix, and Ds is the
transmission matrix.

From the output vector y(k), three dynamic parameters- the armature current, the angular
velocity and the position of the table for a given position command and applied torque are
obtained.

Let us look at an example where all the matrices are calculated. A list of parameters of the feed
drive system is given. Using these parameters, the constant matrices Ac and Bc are computed.
Phi(T) and H(T) are computed for T = 1ms. Similarly, The state matrix, the input matrix, the
output matrix and the transmission matrix are found. Having computed all required matrices, the
dynamic parameters can be calculated.

<next slide>

The sliding mode controller is a robust control technique for nonlinear and uncertain systems. It
forces the system's dynamics to "slide" along a predefined surface in the state space, representing
the desired dynamics and ensuring the desired system behaviour despite disturbances or model
uncertainties. The control objective is to drive the system's states to this surface and keep them
on it.

The open loop block diagram shows a ball-screw drive system controlled in a current mode.

The open-loop transfer function between the table position x(s) and the amplifier command
generated in the CNC u(s) can be derived from the block diagram and is further simplified. Here,
K is the gain, p is the velocity loop pole, Tc is the torque disturbance caused by friction and the
cutting process, J is the inertia on the motor shaft, and B is the equivalent friction coefficient.
In linear motor drives, the inertia (J) is replaced by the table–workpiece mass, and the
disturbance torque (Tc) is replaced by the cutting force and friction in the linear motor-driven
systems.

Rearranging the terms of the equation and applying the inverse Laplace transform to arrive at
this equation. Here, d(t) is the reflected disturbance at the input. Hence, rearranging the terms
one last time expresses the drive's acceleration.

<next slide>

The basic functionality of the feed drive control system is that the controller must be capable of
following the commanded trajectory and velocity with minimum error at high speeds, regardless
of the slight variations in the inertia (Je), viscous friction (Be), and external disturbances (d).

The Conventional drive control systems, such as pole placement and feed-forward control
techniques, work exceptionally well when the system is modelled accurately, and external
disturbances are rejected by having a large bandwidth in linear drives.

In addition, the large transmission ratio in the ballscrew drives significantly reduces the reflected
torque on the rotating drive motor shaft.

But, when the friction is not modelled and compensated accurately, and when the external forces,
such as in linear drives, are not determined before, conventional control techniques exhibit low
tracking accuracy in high-speed machining, leading to inaccuracies in contour machining
operations. The sliding mode controller is robust to such uncertainties and time variations in the
drive dynamics and, hence, is effective for nonlinear and uncertain systems.

<next slide>

A sliding state surface and a Lyapunov function must be selected to design a sliding mode
controller. This form of sliding surface is chosen for accurately tracking position and velocity, an
essential requirement for a high-speed machine. Here, lambda is the desired and achievable
bandwidth of the drive, xr and x are the reference command and actual positions, and xr_dot and
x_dot are the drive's reference command and actual velocities, respectively.

The control input (u) must be manipulated so that, after a finite time, both the position and
velocity of the drive approach reference command values, forcing the value of position and
velocity errors to be zero, S → 0.

Inertia (Je) and viscous damping (Be) on the machine drive are assumed to change slowly and
insignificantly. The variation in the external disturbance caused by the cutting process and
friction is considered strong, but with upper (d+) and lower (d−) limits measured on the machine.
This simple observer can track external disturbances.

Here, d_cap is the estimated disturbance, T[s] is the control period, k is the control interval
counter in the discrete-time domain, ρ is the parameter adaptation gain (ρ≈0.005), and κ is used
to limit the integral control of the disturbance by imposing the following criterion.

The second design step is to select a Lyapunov function, which is used to obtain a stable control
law for a nonlinear system. This form of the function resembles the summation of kinetic energy
and the square of the disturbance prediction error.

The derivative of the above function must be negative or zero for asymptotic stability of
nonlinear systems. That is, mathematically, this inequality must hold good. Substituting the
values of S_dot and d_cap_dot, the inequality is simplified.

Re-writing these terms, it is observed that this term is always either negative or zero. Given that
kappa is either 0 or 1, the sliding surface 𝑆 and estimation error is designed to interact such that
their product is negative. A negative multiple of a square represents this term to ensure a
negative derivative of the function. Here, Ks is a control gain.

<next slide>
The control law can be derived by isolating u(t) from the rest of the terms. The reference
position, xr(k), velocity · xr(k), and acceleration ·· xr(k) are obtained from the command
generation algorithm running in the CNC system.

The actual position x(k) is measured from the encoder, and the actual velocity · x(k) is estimated
by taking the derivative of the measured position from the linear drive or measuring directly
from a tacho generator.

However, calculating velocity and acceleration from discrete position commands and encoder
readings may be noisy. A simple low-pass filter can be used to smooth them.

The modified signals have this form, where xm_dot is the velocity the tacho generator measures.
The filter gain alpha is usually set to 0.6.