Now there are many control types; PID, fuzzy, robust, sliding mode, neural, etc. techniques.
Which method do you prefer and why? If someone decided to learn a new control method (other than PID), which one is the best in terms of simplicity, accuracy and quickness?
Dear Ahmad
As you mention there are many ways in which to skin the cat. However, I go with Dave. The best strategy is to start with the old classical PI controller. From my perspective it is best to try and understand how and what any controller is trying to control from a systems viewpoint whilst trying to achieve its control objective. Only then can one makes sensible rules for Fuzzy Logic, sliding mode control and neural networks.
Regardless of controller any and all power converters simply control the flow of electrical energy from source to load and vice versa. Whether one uses classical control, fuzzy, or neural all require sufficient information to be able achieve the control objective and this may not be "violated". In all power converters the only things we can normally control is the voltage and its phase and frequency. Whether one has current loops, speed loops, position loops or any other, one may only control the magnitude, phase and frequency of the output voltage.
If we take a variable speed AC drive for example, the fundamental control objective is to transfer electrical energy, via the converter and store this within, the mechanical inertia of the motor or drive system as fast and as stable as is possible. In reality we only have physical machines with physical properties (R, L, J etc) and physical converters with practical static limits (Vmax, Imax) and in reality we only have open loop parameters through which to transfer energy.
Through fast acting control we can make it appear we have faster time constants but all we are doing is controlling the voltage. The higher the voltage the more the ability or capability to transfer energy more rapidly between load and source and vice versa within the physical limits of the power electronics converter. The achievable closed loop time constant depends on the voltage limit we may apply and very little else. Hence we have to understand each design aspect of the system being controlled.
Why am I saying this is that many believe one may simply implement fuzzy and neural controllers without much knowledge of the system or system parameters but this is not true. When in fact you actually do this you realise that the Fuzzy rules depend on your (in depth) knowledge and experience of what is actually going in the old PID controller and the system it is trying to control. Same as for neuro controller what input, outputs, number of neurons / nodes etc. This is why I stated that all controllers must have sufficient information and as Dave has said the starting point is the old classical PID controller.
The answer to your question probably lies in what is most widely used in the industry. At very low levels like washing machines, disc drives etc fuzzy is being used, but at high power levels where risk, deterministic, reliable and repeatable operation and performance is required the old classical controller still rules with some non-linear add on's here and there gained from operational experience
As an example sliding mode control was quite popular when robotics emerged as it offered robust 3D point to point trajectory based control. However, this required inner stable speed and current loops which normally used classical old PI controllers and of course the power converter had to be able to provide the energy to allow the robot arm be able to robustly follow the said trajectory.
I think adaptive fuzzy controller because it can reduce THD(total harmonic distortions) to lowest level
I'm old fashioned when it comes to converters. I use digital PID, sometimes analog filters before my ADCs (depending on the noise in the feedback signals) and add digital filters to my coding if required. For most applications I've encountered, PI is adequate.
mathematical based controller (PID, H infinitive, etc) can be the effective one by evaluate the performance criteria. But we need to develop input-output transfer function of system to tune it well. Otherwise, if we want to avoid modelling of system, we can apply fuzzy/AI based controller. But we can not specify the performance criteria.
Normally will be controlled the output-voltage. Especially you feed a three-phase motor drive then you need only a simple voltage-controller for the best basic oscillation content. You can also control the output-current with a simple hysteresis controller! What is your goal?
I think there is no specified optimum control methodology . Each one has advantages and disadvantages. The simplest one is the best one. Therefore, I think you can start with a very simple control to achieve your goals and use the upper one if you do not satisfied with the output of the first one and so on.
Dear Ahmad
As you mention there are many ways in which to skin the cat. However, I go with Dave. The best strategy is to start with the old classical PI controller. From my perspective it is best to try and understand how and what any controller is trying to control from a systems viewpoint whilst trying to achieve its control objective. Only then can one makes sensible rules for Fuzzy Logic, sliding mode control and neural networks.
Regardless of controller any and all power converters simply control the flow of electrical energy from source to load and vice versa. Whether one uses classical control, fuzzy, or neural all require sufficient information to be able achieve the control objective and this may not be "violated". In all power converters the only things we can normally control is the voltage and its phase and frequency. Whether one has current loops, speed loops, position loops or any other, one may only control the magnitude, phase and frequency of the output voltage.
If we take a variable speed AC drive for example, the fundamental control objective is to transfer electrical energy, via the converter and store this within, the mechanical inertia of the motor or drive system as fast and as stable as is possible. In reality we only have physical machines with physical properties (R, L, J etc) and physical converters with practical static limits (Vmax, Imax) and in reality we only have open loop parameters through which to transfer energy.
Through fast acting control we can make it appear we have faster time constants but all we are doing is controlling the voltage. The higher the voltage the more the ability or capability to transfer energy more rapidly between load and source and vice versa within the physical limits of the power electronics converter. The achievable closed loop time constant depends on the voltage limit we may apply and very little else. Hence we have to understand each design aspect of the system being controlled.
Why am I saying this is that many believe one may simply implement fuzzy and neural controllers without much knowledge of the system or system parameters but this is not true. When in fact you actually do this you realise that the Fuzzy rules depend on your (in depth) knowledge and experience of what is actually going in the old PID controller and the system it is trying to control. Same as for neuro controller what input, outputs, number of neurons / nodes etc. This is why I stated that all controllers must have sufficient information and as Dave has said the starting point is the old classical PID controller.
The answer to your question probably lies in what is most widely used in the industry. At very low levels like washing machines, disc drives etc fuzzy is being used, but at high power levels where risk, deterministic, reliable and repeatable operation and performance is required the old classical controller still rules with some non-linear add on's here and there gained from operational experience
As an example sliding mode control was quite popular when robotics emerged as it offered robust 3D point to point trajectory based control. However, this required inner stable speed and current loops which normally used classical old PI controllers and of course the power converter had to be able to provide the energy to allow the robot arm be able to robustly follow the said trajectory.
Dear Prof. Diana
Thank you very much for your fantastic and invaluable answers and demonstrations. I find nowadays many researchers publish the old ideas of control for power converters with just changing the method of control. from PI to fuzzy or sliding mode, for example. This is why I posted this question to know more about this valuable subject and share ideas about it.
Thank you gain everyone replies
Dear Ahmad,
Thank you much appreciated. This is true but to be expected in what is a very mature industry and the need to publish or perish. In many instances we spend time reading papers only to find it has already been done or a modification or addition to previous work. This in itself is not bad as young engineers, educators and researchers all have to make their way up the ladder and as you will possibly agree with no one is born with experience which must be learned the good old fashion way. The difference is that today we have fantastic real-time hardware, simulation and visualisation tools that may be used to implement, test, demonstrate and educate and gain experience that was not previously possible or at least expensive and complex to do.
It depends of power level used. For VSC and industrial range you can use high frequency commutation range and technics like "Direct Power Controller" (DPC) and predictive control which show an excellent performance with very simple implementation
https://cds.cern.ch/record/987498/files/p13.pdf
Article Predictive Control in Power Electronics and Drives
https://books.google.com.tr/books?id=I2JgX5ghSq4C&pg=SA1-PA6&lpg=SA1-PA6&dq=the+most+useful+method+to+control+power+converters&source=bl&ots=XPXxm-8Gio&sig=YJqf28WCd7UhDZBF_XqcMYJFNtI&hl=tr&sa=X&ved=0ahUKEwjjsuP99anZAhWMK1AKHZMYAUEQ6AEITDAE#v=onepage&q=the%20most%20useful%20method%20to%20control%20power%20converters&f=false
Dear Dr. Ahmed Eid
The advanced model predictive control techniques could provide excellent performance for the power converters rather that the mentioned methods.
Regards
Dear Ahmad,
Into my opinion, it depends on several factors e.g. application, accuracy, cost ....etc.
Mohamed
I think, current control with PI is good enough for most of the applications. Although it may depend on the application, PI with constant frequency is most suitable since the passive filters of the converter can be designed based on the operating frequency.
The nonlinear controller (Hysteresis, SMC) is robust enough but require special attention when the order of the passive filter is high.
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Energy