Application of AI in Electrical/Electronics Engg. (EX-802 (D)) - Unit 4 Short Notes

UNIT 4: Application of AI in Electrical/Electronics Engineering

1.0 AI Fundamentals and Learning Paradigms

Introduction to AI in Electrical Engineering

AI techniques are applied for optimization, control, prediction, and fault diagnosis in power systems, motor drives, and power quality.

Learning Types

[!TIP]

Exam Focus: Distinguish learning types by their training data and goal. SOM is both unsupervised and competitive.

2.0 Artificial Neural Networks

Radial Basis Function (RBF) Networks

• Architecture: 3 layers – Input, Hidden (RBF units), Output.

• Hidden Layer: Uses Gaussian basis functions:

$$\phi_j(\mathbf{x}) = \exp\left(-\frac{\|\mathbf{x} - \mathbf{c}_j\|^2}{2\sigma_j^2}\right)$$

where $$\displaystyle \mathbf{c}_j $$ = center, $$\displaystyle \sigma_j $$ = spread.

• Training:

  1. Centers via k-means clustering.

  2. Widths ($$\displaystyle \sigma_j $$) based on nearest neighbor distance.

  3. Output weights via linear least squares.

• Application: Function approximation, time-series prediction (e.g., load forecasting).

Functional Link Networks (FLN)

• Architecture: Single-layer perceptron with functional expansion of input.

• Operation: Input $\mathbf{x}$ expanded to nonlinear basis functions $$\displaystyle \mathbf{F}(\mathbf{x}) = [1, f_1(x_1), f_2(x_2), \dots] $$.

• Advantage: Handles nonlinearity without hidden layer; faster training.

• Example: $$\displaystyle F(x) = [1, x_1, x_2, x_1x_2, x_1^2, x_2^2] $$.

Self-Organizing Maps (SOM) – Kohonen Network

• Principle: Unsupervised competitive learning; preserves topological relationships.

• Training Steps:

  1. Initialize weight vectors $$\displaystyle \mathbf{w}_j $$.

  2. For input $\mathbf{x}$, find winner neuron $c$: $$\displaystyle \|\mathbf{x} - \mathbf{w}_c\| = \min_j \|\mathbf{x} - \mathbf{w}_j\| $$.

  3. Update winner and neighbors: $$\displaystyle \mathbf{w}_j(t+1) = \mathbf{w}_j(t) + \alpha(t) h_{cj}(t) [\mathbf{x} - \mathbf{w}_j(t)] $$.

  4. Decay learning rate $\alpha(t)$ and neighborhood $$\displaystyle h_{cj}(t) $$.

• Application in EE: Clustering of PQ disturbances, fault pattern recognition.

[!TIP]

Common Pitfall: RBF uses fixed nonlinear basis (Gaussian); FLN uses predefined polynomial expansion; SOM uses competitive learning for topology preservation.

3.0 Fuzzy Logic Systems

Fuzzy Sets & Relations

• Fuzzy Set: $$\displaystyle \tilde{A} = \{(x, \mu_{\tilde{A}}(x)) \mid x \in X\} $$, $$\displaystyle \mu_{\tilde{A}}(x) \in [0,1] $$.

• Operations on Fuzzy Relations (for $$\displaystyle R_1, R_2 $$ on $X \times Y$):

  • Union: $$\displaystyle \mu_{R_1 \cup R_2}(x,y) = \max[\mu_{R_1}(x,y), \mu_{R_2}(x,y)] $$

  • Intersection: $$\displaystyle \min[\mu_{R_1}(x,y), \mu_{R_2}(x,y)] $$

  • Complement: $$\displaystyle 1 - \mu_R(x,y) $$

Fuzzy Reasoning

• Mamdani Inference:

  • Rule: IF $x$ is $A$ AND $y$ is $B$ THEN $z$ is $C$.

  • Aggregation: $$\displaystyle \mu_{C'}(z) = \max(\min(\mu_A(x), \mu_B(y))) $$ over all rules.

  • Defuzzification: Centroid method: $$\displaystyle z^* = \frac{\int z \cdot \mu_{C'}(z) dz}{\int \mu_{C'}(z) dz} $$.

• Sugeno Inference:

  • Rule consequent is linear: $$\displaystyle z = p x + q y + r $$.

  • Defuzzification: Weighted average: $$\displaystyle z^* = \frac{\sum_i w_i z_i}{\sum_i w_i} $$, $$\displaystyle w_i = \mu_{A_i}(x) \cdot \mu_{B_i}(y) $$.

Example: Motor Speed Control

• Inputs: Error ($e$), Change in error ($\Delta e$).

• Output: Control voltage ($\Delta V$).

• Rules: "IF $e$ is Positive Big AND $\Delta e$ is Positive Small THEN $\Delta V$ is Negative Medium."

[!TIP]

Exam Key: Mamdani uses fuzzy output sets (max-min); Sugeno gives crisp output directly (weighted average). Sugeno is computationally efficient for adaptive control.

4.0 Evolutionary Algorithms

Genetic Algorithms (GA)

• Overview: Population-based search inspired by natural selection.

• Steps:

  1. Initialization: Random population of chromosomes (binary/real).

  2. Fitness Evaluation: Objective function (e.g., fuel cost in ELD).

  3. Selection: Choose parents (e.g., roulette wheel, tournament).

  4. Crossover: Recombine parents (e.g., single-point, multi-point, uniform).

  5. Mutation: Random perturbation (e.g., bit-flip for binary, Gaussian for real).

  6. Repeat until convergence.

Genetic Operators

Mutation Operator Application

• In ELD: Gaussian mutation on real-valued generation schedules to escape local optima.

• In Motor Parameter Tuning: Mutate controller gains (e.g., PID) to optimize response.

[!TIP]

Key Point: GA is stochastic; balance exploration (mutation) and exploitation (crossover). Tournament selection is robust to fitness scaling.

5.0 Special Electrical Machines (as AI Application Domains)

5.1 Stepper Motors

Types

Construction & Working

• Rotor-Stator Interaction: Sequential energization of stator phases creates magnetic attraction, moving rotor in discrete steps.

• Step Angle Calculation:

$$\theta_s = \frac{360^\circ}{N_s \times N_r}$$

$$\displaystyle N_s $$ = stator phases/poles, $$\displaystyle N_r $$ = rotor teeth/poles.

• Example (May 2023): 8-pole stator castellated to 5 teeth, rotor 50 teeth:

  $$\displaystyle \theta_s = \frac{360}{8 \times 50} = 0.9^\circ $$ (if 5 teeth per pole, effective $$\displaystyle N_s = 8 \times 5 = 40 $$? Clarify: Castellated increases effective poles. Usually $$\displaystyle \theta_s = \frac{360}{N_s N_r} $$ with $$\displaystyle N_s $$ = number of stator teeth.)

Driver Circuits

• Unipolar: Center-tapped windings; current flows in one direction per phase. Simple but half torque.

• Bipolar: Full winding used; higher torque, requires H-bridge.

• Dual Voltage: High voltage for fast current rise, low voltage for holding. Current build-up:

$$i(t) = \frac{V}{L} \left(1 - e^{-t/\tau}\right), \quad \tau = L/R$$

• Chopper Drive: PWM to regulate current; constant torque.

Microstepping

• Concept: Divide full step into smaller increments by proportioning phase currents sinusoidally.

• Waveform Synthesis:

  $$\displaystyle I_A = I_m \sin \theta $$, $$\displaystyle I_B = I_m \cos \theta $$ for 2-phase motor.

• Advantages: Smooth motion, reduced resonance, higher resolution.

Static & Dynamic Characteristics

• Pull-in Torque: Max torque at given speed without missing steps (static load).

• Pull-out Torque: Max torque at speed without losing synchronism (dynamic load).

• Resonance: Occurs at step rates matching natural frequency; avoid via microstepping or damping.

Torque Equation

$$T = K_t \cdot I \cdot \sin(\theta)$$

$$\displaystyle K_t $$ = torque constant, $\theta$ = load angle (rotor offset from equilibrium).

Speed Control Methods

1. Pulse Rate Control: Vary step rate (most common).

2. Series Resistance: Insert resistor in winding; reduces torque, inefficient.

3. Voltage Control: Adjust supply voltage (affects current rise time).

Applications

• Robotics, printers, CNC machines, electric vehicles (adjustable seats, windows), PV water pumping (solar-powered, precise flow control).

[!TIP]

Common Error: Confusing step angle formula. For hybrid: $$\displaystyle \theta_s = \frac{360}{N_s N_r} $$ where $$\displaystyle N_s $$ = stator teeth, $$\displaystyle N_r $$ = rotor teeth. Pull-in < Pull-out torque.

5.2 Switched Reluctance Motors (SRM)

Construction

• Stator: Salient poles with windings.

• Rotor: Salient, no windings or PMs; made of laminated steel (or solid for low cost).

• Cross-section: Doubly salient structure.

Principle of Operation

• Variable Reluctance: Torque produced by tendency to minimize magnetic reluctance.

• Sequence: Phase A excited → rotor pole aligns → Phase A off, Phase B on → continues rotation.

Torque Production

• Expression:

$$T = \frac{1}{2} i^2 \frac{dL(\theta)}{d\theta}$$

where $L(\theta)$ = phase inductance profile.

• Torque positive when $$\displaystyle \frac{dL}{d\theta} > 0 $$ (inductance increasing with $\theta$).

• Torque-Angle Characteristics:

  • At low current: approximately sinusoidal.

  • At high current: saturation flattens peak, shifts to $$\displaystyle \theta < 90^\circ $$.

Shaft Position Sensing

• Hall Effect Sensors: Common, low cost.

• Optical Encoders: High precision.

• Resolvers: Robust, for harsh environments.

• Sensorless: Use phase inductance measurement (e.g., detect zero-crossing of $$\displaystyle \frac{dL}{d\theta} $$).

Advantages

• Simple, rugged construction (no PM, no brushes).

• High torque/inertia ratio.

• Fault-tolerant (phase independent).

• Wide speed range.

Disadvantages

• Torque ripple (due to pulsed excitation).

• Acoustic noise (from torque pulsations).

• Requires position sensor (adds cost/complexity).

Solid vs Laminated Rotors

Applications

• Traction ( EVs, locomotives), industrial drives (pumps, fans), appliances (washing machines).

[!TIP]

SRM Torque Calculation (May 2024): Given $L(\theta)$ profile, compute $$\displaystyle T = \frac{1}{2} i^2 \frac{\Delta L}{\Delta \theta} $$. For aligned ($$\displaystyle L_{max} $$) and unaligned ($$\displaystyle L_{min} $$), maximum energy per stroke: $$\displaystyle W_{max} = \frac{1}{2} I_{max}^2 (L_{max} - L_{min}) $$. Average torque: $$\displaystyle T_{avg} = \frac{W_{max}}{\text{stroke angle}} $$.

5.3 Brushless DC Motors (BLDC)

Construction

• Stator: Three-phase windings (similar to AC induction motor).

• Rotor: Surface-mounted permanent magnets (NdFeB, SmCo).

• Position Sensors: Hall effect sensors (3 sensors, 60° electrical apart).

Working Principle

• Electronic Commutation: Controllers switch phases based on rotor position (from Hall sensors) to maintain continuous torque.

• Six-Step Commutation: Each phase conducts for 120° electrical; sequence: A→B→C (for 3-phase).

Torque Production

• Lorentz Force: $$\displaystyle F = I \times B $$.

• Torque: $$\displaystyle T = K_t \cdot I $$, where $$\displaystyle K_t $$ = torque constant (Nm/A).

• Back-EMF: Trapezoidal waveform; $$\displaystyle E_b = K_e \cdot \omega $$, $$\displaystyle K_e $$ = back-EMF constant (V/(rad/s)).

Speed Control

• Voltage Control: Adjust DC bus voltage (inefficient).

• PWM Control: Vary duty cycle while keeping voltage constant (standard).

• Current Control: Torque directly proportional to phase current.

Winding Patterns

• Series (Delta): Windings in Δ; higher line current, lower phase current.

• Parallel (Wye): Windings in Y; neutral point available, common in BLDC.

Commutation

• Sequence: Based on Hall signals (e.g., Hall: 101 → 100 → 110 → 010 → 011 → 001).

• Sensorless: Detect back-EMF zero-crossing in unenergized phase.

Armature Reaction

• Effect: Stator MMF distorts/weakens rotor PM field.

• Consequence: Demagnetization risk (especially at high currents), torque ripple.

• Mitigation: Overdesign PMs, use magnetic shunts.

Permanent Magnet Materials

Applications

• Fans, pumps, electric vehicles (traction motor), aerospace, consumer electronics.

[!TIP]

BLDC vs PMSM: BLDC has trapezoidal back-EMF, 6-step commutation; PMSM has sinusoidal back-EMF, sinusoidal drive (FOC). BLDC simpler control, PMSM smoother torque.

5.4 Permanent Magnet Synchronous Motors (PMSM)

Construction

• Stator: Three-phase AC winding.

• Rotor:

  • SPM (Surface-mounted PM): Magnets on surface.

  • IPM (Interior PM): Magnets embedded; provides reluctance torque.

Operation

• Synchronous Speed: $$\displaystyle n_s = \frac{120 f}{P} $$ (P = poles).

• Sinusoidal Back-EMF: $$\displaystyle e(t) = E_{peak} \sin(\omega t) $$.

EMF Equation Derivation

$$E_{rms} = 4.44 \cdot f \cdot N \cdot \phi \cdot K_w$$

where:

• $f$ = frequency (Hz),

• $N$ = turns per phase,

• $\phi$ = flux per pole (Wb),

• $$\displaystyle K_w $$ = winding factor (accounts for pitch/distribution).

\boxed{E_{rms} = 4.44 \cdot f \cdot N \cdot \phi \cdot K_w}

Torque-Speed Characteristics

• Phasor Diagram: $$\displaystyle V = E + I \cdot Z $$.

• Circle Diagram: Constant current circles in dq-plane; maximum torque at $$\displaystyle \delta = 90^\circ $$ for SPM.

• Regions:

  • Constant Torque: Below base speed, $$\displaystyle I_d = 0 $$ (max torque per amp).

  • Torque Weakening: Above base speed, $$\displaystyle I_d < 0 $$ to reduce back-EMF.

Torque Ripple Reduction

• Design: Skewing, fractional-slot winding.

• Control: Current profiling (inject $$\displaystyle I_d $$ to compensate cogging), FOC with high resolution sensors.

Speed Control Methods

1. Scalar Control (V/f): Simple, open-loop; poor dynamic response.

2. Vector Control (FOC): Decouple $$\displaystyle I_d $$, $$\displaystyle I_q $$; independent torque/flux control.

3. Direct Torque Control (DTC): Direct torque/flux control; fast response, torque ripple.

Comparison: BLDC vs PMSM

Applications

• High-performance drives, robotics, electric vehicles, compressors, CNC machines.

[!TIP]

PMSM EMF: Derived from $$\displaystyle E = 4.44 f N \phi K_w $$. $\phi$ from PM flux density $$\displaystyle B_r $$ and air-gap area. FOC requires accurate machine model; DTC is sensorless-friendly.

5.5 Other Motors & Magnetic Concepts

Permanent Magnet DC Motors

• Applications: Traction (e.g., forklifts), power tools, automotive (window lifts).

AC Servomotors

• Construction: Two-phase stator (main + control), salient-pole rotor.

• Torque-Speed: Linear relationship; high torque at low speed.

• Control: Voltage control of control phase; used in precise positioning.

Magnetic Fundamentals

• B-H Relationship:

  • Hysteresis: Energy loss per cycle; area of loop.

  • Saturation: $B$ saturates at high $H$; limits flux density.

• Leakage Flux: Flux that does not link both windings; causes leakage inductance.

• Fringing: Flux bulging at air-gap edges; increases effective air-gap area.

• Stacking Factor: $$\displaystyle k_{stack} = \frac{\text{actual iron volume}}{\text{stack volume}} $$; accounts for insulation/varnish (typically 0.9–0.95).

6.0 Power Quality (as AI Application Domains)

6.1 Power Quality Issues

Voltage Phenomena

Transients

• Causes: Switching operations, faults, lightning.

• Transient Recovery Voltage (TRV): Factors: system voltage, fault current, grounding, breaker speed.

Harmonics

• Sources: Non-linear loads (rectifiers, arc furnaces, LED drivers).

• Definitions:

  • Fundamental frequency $$\displaystyle f_1 $$; harmonic order $$\displaystyle h = n f_1 $$.

  • THD (Total Harmonic Distortion):

$$\text{THD}_V = \frac{\sqrt{\sum_{h=2}^{\infty} V_h^2}}{V_1} \times 100\%$$

• TDD (Total Demand Distortion): THD relative to demand current $$\displaystyle I_{1,demand} $$.

• Crest Factor: $$\displaystyle \frac{I_{peak}}{I_{rms}} $$; indicates harmonic stress.

• Interharmonics: Frequencies not integer multiples of $$\displaystyle f_1 $$; from cycloconverters, arcing.

Classification of Disturbances

1. Short-duration: Sag, swell, transient.

2. Long-duration: Interruption, voltage variation.

3. Waveform distortion: Harmonics, notching, interharmonics.

6.2 Mitigation Techniques

Power Factor Correction

• Benefits: Reduce losses, increase capacity, avoid penalties.

• Capacitor Bank Locations:

  • At Load: Most effective, but high inrush.

  • At Bus: Balances system PF.

  • At Substation: Bulk correction.

FACTS Devices

Custom Power Devices

• DSTATCOM: Shunt VSC; acts as active power filter (APF) for current harmonics/reactive power.

• UPQC: Series VSC (voltage sag/swell correction) + shunt VSC (current compensation).

• Active Power Filter (APF):

  • Shunt: Injects harmonic currents.

  • Series: Blocks harmonic voltages.

  • Hybrid: Combines passive + active stages.

Passive Compensation

• Series/Shunt/Hybrid Filters: Tuned to specific harmonics; risk of resonance with system inductance.

Surge Protection

• SPDs (Surge Protective Devices): Clamp overvoltages; installed at service entrance, subpanels.

• Zero-Voltage Crossing Switching: For capacitor banks; closes at voltage zero to minimize inrush/TRV.

Voltage Sag Protection

• Sag Ride-Through: Equipment tolerates sags (ITIC curve).

• DVR (Dynamic Voltage Restorer): Series-connected VSC; injects voltage to compensate sag.

6.3 Harmonic Management

Harmonic Sources

• Industrial: Arc furnaces, rectifiers (HVDC, drives), welding.

• Commercial: Computers, LED drivers, UPS.

Active Harmonic Filters

• Principle: Detect harmonics (via FFT or instantaneous theory), inject compensating currents.

• Advantages: Dynamic, selective, no resonance.

• Disadvantages: Costly, complex control, limited bandwidth.

Passive Filters

• Design: Tuned LC branches for 5th, 7th, 11th, 13th harmonics.

• Risks: Detuning due to system impedance change; parallel resonance.

Harmonic Analyzer

• Operation: Samples voltage/current, computes THD, individual harmonics via FFT.

• Standards: IEEE 519-2014 limits harmonic currents.

6.4 UPQC

Principle

• Series Converter: Injects voltage to correct sag/swell, harmonics.

• Shunt Converter: Injects current to compensate load harmonics/reactive power.

• Back-to-Back Converters: Share common DC bus (capacitor).

Control Strategies

• Series: Voltage sag detection, reference generation (e.g., instantaneous symmetrical components).

• Shunt: Instantaneous p-q theory or synchronous reference frame for harmonic extraction.

Classification

• UPQC-Q: Focus on power quality (voltage/current correction).

• UPQC-P: Also controls power flow (acts as power conditioner).

Advantages over DSTATCOM

• Mitigates both voltage disturbances (sag/swell) and current harmonics.

• Provides sag ride-through for critical loads.

Applications

• Hospitals, data centers, semiconductor plants, sensitive industrial processes.

[!TIP]

Key Distinction: DSTATCOM = shunt APF only; UPQC = series + shunt. STATCOM is shunt VSC for voltage support (reactive power), not harmonics. SVC uses thyristors, slower; STATCOM uses IGBTs, faster.

7.0 AI Applications in Power Systems

7.1 Economic Load Dispatch (ELD)

• Problem: Minimize total fuel cost $$\displaystyle \sum F_i(P_i) $$ subject to:

$$\sum P_i = P_D + P_{loss}, \quad P_{i,\min} \leq P_i \leq P_{i,\max}$$

• AI Methods:

  • GA/PSO: Handle non-convex cost (valve-point loading).

  • Fuzzy Logic: Model uncertainty in cost coefficients.

Example (Valve-Point Loading):
$$\displaystyle F_i(P_i) = a_i P_i^2 + b_i P_i + c_i + |e_i \sin(f_i (P_{i,\min} - P_i))| $$

AI optimizes $$\displaystyle P_i $$ to minimize sum.

7.2 Load Frequency Control (LFC)

Single Area

• ACE (Area Control Error): $$\displaystyle ACE = \Delta P_{tie} + B \Delta f $$

  $B$ = frequency bias constant.

• Integral Control: $$\displaystyle \frac{d}{dt} \Delta P_c = K_I \cdot ACE $$.

Two-Area System

• Tie-line Power: $$\displaystyle \Delta P_{tie} = T_{12} (\Delta \theta_1 - \Delta \theta_2) $$.

• ACE per area: $$\displaystyle ACE_i = \Delta P_{tie,i} + B_i \Delta f_i $$.

• Parameters: Governor time constant $$\displaystyle T_g $$, droop $R$, turbine time constant $$\displaystyle T_t $$.

AI-based LFC

• Fuzzy Logic Controller: Handles nonlinearities, adapts to load changes.

• Neural Network: Learns system dynamics; adaptive control.

7.3 Stability Analysis

• Small Signal Stability: Damping of oscillations after disturbance; assessed via eigenvalues of state matrix.

• AI Applications:

  • Neural Networks: Classify stable/unstable from input features (e.g., generation pattern).

  • Fuzzy Logic: Damping controllers (PSS) with adaptive rules.

7.4 Load Forecasting

• Types:

  • Short-term: Hourly/daily (unit commitment).

  • Medium-term: Weekly/monthly (maintenance).

  • Long-term: Yearly (expansion planning).

• AI Techniques:

  • RBF Networks: Approximate nonlinear load-weather relationships.

  • SOM: Cluster similar days (weekday/weekend, holiday).

  • Fuzzy Logic: Handle uncertainty in temperature/humidity impact.

  • GA: Feature selection (optimal input variables).

7.5 AI for Power Quality Improvement

• Active Filters:

  • Neural Networks: Detect harmonics (trained on distorted waveforms).

  • Fuzzy Logic: Generate reference currents robustly under noisy conditions.

• STATCOM/DVR Control:

  • ANNs: Tune PI controllers for sag mitigation.

  • Fuzzy Logic: Adaptive voltage support during faults.

• FACTS Optimal Placement: GA/PSO to minimize losses, improve stability.

8.0 Integration and Case Studies

AI in Electric Vehicles

• BLDC/PMSM Control: FOC for traction motors; AI optimizes efficiency, regenerative braking.

• Battery Management: Neural networks for SOC/SOH estimation; fuzzy logic for thermal management.

• Energy Optimization: GA for route planning, energy consumption minimization.

AI in Renewable Energy

• PV Water Pumping: BLDC motor directly coupled; MPPT with fuzzy logic (perturb & observe adaptive).

• Grid Integration: AI forecasts solar/wind output; stabilizes grid with virtual inertia.

Case Studies

• Implementation: AI-based LFC in interconnected grids (e.g., Indian power system).

• PQ Monitoring: SOM-based classification of disturbances in industrial plants.

• Motor Drives: GA-optimized PID for SRM torque ripple reduction.

Future Trends

• Deep Learning: CNN/LSTM for PQ event classification from waveform images/sequences.

• Digital Twins: AI-driven real-time simulation of machines for predictive maintenance.

• Edge AI: On-device neural networks for motor control in EVs.

[!TIP]

Exam Focus: Link AI techniques to specific EE problems: RBF for load forecasting, GA for ELD, fuzzy for LFC, SOM for PQ clustering. In motors, AI for SRM torque ripple, BLDC sensorless control.

Final Note: This summary aligns strictly with RGPV past papers (2022–2025). Prioritize Special Machines (Stepper, SRM, BLDC, PMSM) and Power Quality (sag/swell, harmonics, UPQC) for high-mark questions. AI sections are shorter but cover asked topics (RBF, SOM, GA, fuzzy). Always include diagrams where possible (e.g., SRM cross-section, UPQC block diagram, SOM training steps).