With continuously changing external conditions, controlling power electronic devices such that they always operate at best efficiency point cannot be achieved by conventional control methods. This is where artificial-intelligence techniques come into the picture.
A wide range of applications based on power electronic devices are possible because of the ability to convert electrical energy into useful forms like heat, light, motion, and sound with utmost efficiency. Motor drives are a prime example, with application in almost every industry. More than 70% of industrial loads are motor loads, with induction motors forming a major part of it. Hence, precise control of these motors is important for industries, as it can save them a ton of money and resources. All of these high-frequency switching devices require precise control to operate properly. With continuously changing external conditions, controlling power electronic devices such that they always operate at best efficiency point (BEP) cannot be achieved by conventional control methods. This is where artificial-intelligence techniques come into the picture.
AI techniques, such as expert systems, fuzzy logic, artificial neural networks (ANNs), and genetic algorithms, have recently been applied widely in power electronics and motor drives. The goal of these AI techniques is to build a control system having human-like intelligence in machines. These systems are designed such that they have self-learning, self-adapting, and self-organizing capabilities. While expert systems and fuzzy logic are more rule-based systems, neural networks are more generic in nature and tend to emulate the biological neural network directly. Although ANNs have been there from the 1940s, they have seen major advancements only in the early 2000s. ANNs have had one of the greatest impacts on power electronics and motor drives among all the branches of AI.
The ANN can be considered a machine copy of a biological neuron, with different parts of the ANN performing different tasks just like an actual neuron (see Figure 1). ANNs are generally of two types: feedforward and feedback type. Most applications in power electronics use a feedforward ANN, but to precisely control and monitor motors using motor drives, the feedback ANN is used.
Role of neural networks in core power electronics
The ANN is the next step in the evolution of power electronics design, control, and application. It can be used for different renewable energy applications like grid-tied inverters, solar PV inverters, and electric charging stations. One such application, which has seen a lot of progress recently, is the grid-tied PV inverter. ANN networks are being used to improve design, operation, and maintenance of PV cells. Traditional PV controllers use PI controllers or PR algorithms, which are sometimes sluggish in their response to sudden disturbances. In grid-tied operations, disturbances are quite frequent, and hence, these controllers cause loss in efficiency and precision of operation. When AI algorithms are added to the controller, the response time to disturbances and the accuracy of the converter are improved.
One of the major reasons why people do not use EVs is the long charging time taken to fully charge the vehicle. AI-powered smart EV charging systems optimize charging by efficiently monitoring the charging current, battery type, and other charging parameters to charge the battery faster (see Figure 2). Continuous monitoring while charging will also help predict battery life and prevent faults.
Role of neural networks in induction motor drives
Losses in an induction motor can be minimized to find the BEP during lower-load conditions by maintaining just the right amount of flux at a given value of speed and torque. But determining the right amount of flux in a complete dynamic environment is a major challenge faced by engineers.
This is where neural networks are of great help, as they can be trained to solve complex nonlinear functions with variable parameters, which cannot be attained by conventional mathematical tools. In the case of induction motors, the optimal rotor flux for maximum efficiency is a nonlinear function of both rotor shaft speed and load torque, and the machine parameters change with increase in machine temperature. Hence, a neural network model for optimal rotor magnetizing current prediction and field control can be used.
Future scope
The neural network is a vast discipline in AI, and there has been a lot of progress in the field in recent times. The use of neural networks in power electronics and motor drives at present is limited because of the relatively young technology with low reliability, although the technique has an immense amount of promise in the coming days. The major challenge faced by these technologies today is the lack of availability of good training sets and data. Generating data specifically for this purpose is not economical, and hence, low-data high-accuracy models, which require less processing power, can be researched for the industry to adapt and implement. With more and more industries looking for sustainability and with Industry 4.0 around the corner, AI and machine learning will surely take over the conventional control methodologies in place today.
