Hi Pushkar,

Say you want to control the indoor air temperature using an air-conditioning system.

The surface temperature of external walls is affected by the ambient air temperature and solar radiation intensity on the walls. The heat transfer from the external walls also affects the dynamic response of a heat exchanger in the room. In this case, the ambient air temperature and solar radiation intensity can be considered as input disturbances, di.

If the indoor air temperature measured by a thermometer that is not shielded from radiation (a/c unit installed too close to window or light source) and moisture (without desiccant air dryer), then the true air temperature (dry-bulb temperature) must be determined. In this situation, the radiation and moisture are considered as output disturbances, do.

So in Yew-Chung Chak's diagram, the tranfer functions that link the various inputs (r, d_i & d_o) to the output (y) are all slightly different.

To understand the origins of these disturbances, think about the nature of your plant model and the units of the plant input and output ...

e.g. for a dc motor plant with volts input and rpm output:

an output disturbance might be caused by friction from worn bearings;

an input disturbance might be due to a power supply glitch.

Hi Pushkar

Say, you want to control voltage of an islanded microgrid using PID controller, some types of dusturbance have been arised. The difference between input and output disturbance have entirely discussed in the following address

https://link.springer.com/article/10.1007/s40435-017-0364-0

Many interesting answers from others with which I agree. First, your question asks for differences between effects of input noise and output noise: the main one is that input noise gets filtered by the plant in the figure from Chak, while the output noise is added unaltered to plant response. Hence any analysis on plant responses needs to correct for contamination that could be white, mixed with filtered input noise that is non-white.

From another viewpoint, and that is what are you planning to do with the controller-plant model? Produce a more robust performance despite input and output disturbances, or identify a model for your controller, actuator and plant cascade, despite the presence of disturbances (noise) in the data. If the second goal is your focus, I would try and characterize the input noise by measuring control signal to the plant when the actuator is fixed (but active) or driven by a known simple signal (like steps or ramps). Given calibration, you can extract the input noise by removing the controller input and examining the residual, and then find its statistical properties and bandwidth profile. Similarly, you can disconnect the actuator and replace it with a waveform generator for the plant input, measure the plant response and find a satisfactory transfer function or non-linear model for your plant. At this point you can predict the expected plant response to any input...

So reconnect system, measure input (including noise), measure output; predict expected response from total input and plant model, subract from plant response and isolate output 'noise'. Find output noise statistics and bandwidth. You may be able to use appropriate filters to exclude the noise signals from these tests, depending on the bandwidth required from your controller. In fact, the last test can be used to tune a pant model until the residual (output noise) is zero-mean and nearly white with numerous tests available in matlab or other libraries

It depends on the real system, for motion control, i/p disturbance means disturbance on input torque. O/p disturbance appears in load frequency control in power systems, which means fluctuations in the frequency and called speed disturbance.

modeling these disturbance are very easy and can be cancelled very easily by active disturbance rejection control methodology.

Going back to design problem, one always needs to check the transfer function from the input/output disturbance to the process variable (controlled variable) and also to controller output (manipulated variable). These relations are given by different transfer functions (sometimes called Gang of Four). For robust feedback loop, all of them must satisfy certain shaping conditions.

The input disturbance is generated in actuator and output disturbance is generated in sensor, this is the basic difference between input and output disturbances. The main causes of these disturbances are abnormal behavior of actuator (in case of motor, its magnetic field becomes weak or fault in system for which this motor is used) and environmental effect in sensors' performances.

Dear Pushkar Prakash Arya

Exogenous disturbance (input disturbance) affects model (physical or mathematical) behavior, most probably assimilated to account for model-uncertainty (inaccuracy), time-varying parameters, perturbation (as wind-gust in aircrafts), actuator unmodeled dynamics, and the same ilk. Input disturbance could be modeled as a constant signal D(s)=1/s, in Laplace domain, or d(t)=1 in time-domain, or a sinusoidal signal, or a stochastic process, or something like that.

Input disturbance signal is usually added before plant dynamics in block diagram.

But output disturbance, is mainly due to sensor noise which makes the output signal measurement perturbed. It is simply a noisy signal added in the measurement location.