Applications of Numerical Methods in Engineering: Real-World Systems, ODEs, PDEs, and Optimization

TAKEAWAYS:

• Numerical methods turn real-world engineering problems (e.g., Ncell’s network load balancing, Daraz’s delivery route optimization) into solvable math models using ODEs, PDEs, and linear algebra.
• Ordinary Differential Equations (ODEs) model dynamic systems (e.g., heat dissipation in a CPU, Pathao’s rider demand spikes), solved via Euler, RK4, or shooting methods.
• Partial Differential Equations (PDEs) govern spatial phenomena (e.g., NTC’s power grid voltage drops, NEPSE’s price fluctuations), discretized via finite difference or finite element methods.
• Eigenvalue problems optimize systems (e.g., Google’s PageRank uses eigenvectors to rank websites; Ncell’s antenna placement maximizes coverage).
• Error analysis ensures reliability: truncation errors in RK4 vs. round-off errors in floating-point arithmetic (critical for eSewa’s transaction limits).
• Case studies tie theory to exams: solve a bank loan amortization schedule (ODE), traffic flow at Thapathali (PDE), or NEPSE’s moving average (curve fitting).

1. Why Numerical Methods in Engineering?

Engineering problems are rarely solvable by pencil-and-paper math. Numerical methods provide approximate but practical solutions using computers. For example:

• eSewa’s transaction limits: Uses root-finding methods (Newton-Raphson) to model fraud detection thresholds.
• Pathao’s rider allocation: Solves linear systems (Gauss-Seidel) to balance supply-demand in real time.
• NTC’s power grid: Relies on PDE solvers to predict voltage drops across Kathmandu’s network.

2. Modeling Real Systems with ODEs

Ordinary Differential Equations (ODEs) describe how systems change over time. Key applications:

A. Population Growth (e.g., Kathmandu’s Traffic)

Problem: Predict vehicle count at Thapathali intersection over 1 hour. ODE Model: $$\frac{dP}{dt}=rP-k{P}^2\text{(Logistic growth)}$$ where:

• $P(t)$ = number of vehicles at time $t$,
• $r$ = growth rate (vehicles/min),
• $k$ = congestion factor.

Solution via RK4 Method:

# Pseudocode for RK4 (h = 0.1 min)
def rk4(P, t, h):
    k1 = h * (r*P - k*P**2)
    k2 = h * (r*(P + k1/2) - k*(P + k1/2)**2)
    k3 = h * (r*(P + k2/2) - k*(P + k2/2)**2)
    k4 = h * (r*(P + k3) - k*(P + k3)**2)
    return P + (k1 + 2*k2 + 2*k3 + k4)/6

Worked Example: Given $P(0)=500$ vehicles, $r=0.05$, $k=0.0001$, compute $P(0.3)$:

B. Financial Modeling (e.g., NEPSE Stock Prices)

Problem: Model the price $S(t)$ of a stock using the Black-Scholes ODE: $$\frac{\partial S}{\partial t}+\frac{1}{2}{\sigma }^2{S}^2\frac{{\partial }^{2}S}{\partial S^2}+rS\frac{\partial S}{\partial S}-rS=0$$ Numerical Solution: Finite difference method (exam favorite!). Worked Example: Discretize for $S(0)=1000$, $\sigma =0.2$, $r=0.05$, $\Delta t=0.1$:

3. Spatial Problems: PDEs in Engineering

Partial Differential Equations (PDEs) model phenomena across space and time. Examples:

A. Heat Equation (e.g., CPU Cooling)

PDE: $$\frac{\partial u}{\partial t}=\alpha \frac{{\partial }^{2}u}{\partial x^2}$$ where $u(x,t)$ = temperature at position $x$ and time $t$.

Finite Difference Solution: Worked Example: For a CPU with $u(0,t)=100°C$, $u(L,t)=20°C$, $\alpha =0.01$, $\Delta x=0.1$, $\Delta t=0.001$:

B. Traffic Flow (e.g., Kathmandu Gridlock)

PDE Model (LWR Model): $$\frac{\partial \rho }{\partial t}+\frac{\partial (\rho v)}{\partial x}=0$$ where $\rho (x,t)$ = vehicle density, $v(\rho )$ = speed.

Numerical Solution: Godunov’s method (exam tip: compare with Euler’s method). Real-World Tie:

• Problem: Simulate traffic at Thapathali-Chabahil during peak hours.
• Data: ${\rho }_{\text{max}}=200$ vehicles/km, $v_{\text{max}}=50$ km/h.
• Result: Predict bottlenecks to optimize signal timings.

4. Linear Algebra in Engineering

A. Circuit Analysis (e.g., NTC Power Grid)

Problem: Solve for currents $I_1,I_2$ in a 3-node circuit using Gauss-Seidel. Equations: $$\{\begin{array}{l}5I_1-2I_2=10\\ -2I_1+4I_2-I_3=0\\ -I_2+3I_3=5\end{array}$$ Gauss-Seidel Iteration: $$I_1^{(k+1)}=\frac{10+2I_2^{(k)}}{5},I_2^{(k+1)}=\frac{2I_1^{(k+1)}+I_3^{(k)}}{4},I_3^{(k+1)}=\frac{5+I_2^{(k+1)}}{3}$$ Convergence:

B. Eigenvalues in Structural Engineering

Problem: Find natural frequencies of a bridge modeled as a matrix: $$A=\left[\begin{array}{ccc}2& -1& 0\\ -1& 2& -1\\ 0& -1& 2\end{array}\right]$$ Power Method:

1. Start with ${\mathbf{x}}_0=[1,1,1{]}^T$.
2. Iterate: ${\mathbf{x}}_{k+1}=A{\mathbf{x}}_k/\Vert A{\mathbf{x}}_k\Vert$.
3. Largest eigenvalue $\lambda \approx \frac{{\mathbf{x}}_k^{T}A{\mathbf{x}}_k}{{\mathbf{x}}_k^T{\mathbf{x}}_k}$.

Result:

5. Curve Fitting and Optimization

A. NEPSE Stock Prediction (Polynomial Fit)

Data: Closing prices over 5 days:

Lagrange Interpolation: $$P(x)=2000\frac{(x-2)(x-3)(x-4)(x-5)}{(1-2)(1-3)(1-4)(1-5)}+\cdots +2300\frac{(x-1)(x-2)(x-3)(x-4)}{(5-1)(5-2)(5-3)(5-4)}$$ Prediction for Day 6: $$P(6)\approx 2400\text{ Rs.}$$

B. Optimization: Daraz Delivery Routes

Problem: Minimize delivery time for 3 orders using linear programming. Constraints: $$\{\begin{array}{l}2x_1+x_2\le 10\text{ (time)}\\ x_1+2x_2\le 8\\ x_1,x_2\ge 0\end{array}$$ Solution via Simplex Method:

flowchart LR
    A["Initial Tableau"] --> B["Pivot on (10,2)"] --> C["Optimal: x₁=4, x₂=2"]
    C --> D["Total time = 2*4 + 2 = 10 units"]

6. Error Analysis in Engineering

Sources of Error:

Worked Example: Compute ${\int }_0^1{e}^{-x^2}dx$ using Simpson’s Rule with $h=0.2$: $$\text{Error}=-\frac{h^4}{180}{f}^{(4)}(\xi )\approx -0.0003$$

In the Real World

1. eSewa’s Fraud Detection:

   • Idea: Root-finding (Newton-Raphson) identifies anomalous transaction patterns.
   • How: Solves $f(x)=\text{transaction\_amount}-\text{user\_limit}=0$ to flag $x>\text{threshold}$.

2. Pathao’s Rider Allocation:

   • Idea: Linear systems (Gauss-Seidel) balances riders across zones.
   • How: Solves $A\mathbf{x}=\mathbf{b}$ where $A$ = demand matrix, $\mathbf{x}$ = rider counts.

3. NTC’s Power Grid:

   • Idea: PDEs (finite difference) predicts voltage drops.
   • How: Solves ${\nabla }^{2}V=0$ over Kathmandu’s grid to optimize transformer placement.

4. NEPSE’s Moving Averages:

   • Idea: Curve fitting (least squares) smooths stock data.
   • How: Fits $S(t)=a+bt+c{t}^2$ to 30-day prices to predict trends.

Exam Tip

1. Always show iterations: For Gauss-Seidel or RK4, write out at least 2 iterations (examiners deduct marks for skipping steps).
2. Link to real systems: In part (a), solve a circuit or traffic flow; in part (b), optimize a delivery route or stock prediction.
3. Error analysis is key: For numerical integration, state the error formula (e.g., Simpson’s $O(h^4)$).
4. PDEs: Use finite difference stencils (e.g., 5-point Laplacian) and label boundary conditions clearly.
5. Eigenvalues: For the power method, show the Rayleigh quotient $\lambda \approx \frac{{\mathbf{x}}^{T}A\mathbf{x}}{{\mathbf{x}}^T\mathbf{x}}$.

Pro Tip: Memorize these exam-friendly mappings:

In the real world

• eSewa’s fraud detection uses Newton-Raphson root-finding to flag transactions where the error between predicted and actual spending exceeds a threshold (e.g., Rs. 5000 deviation triggers a manual review).
• Pathao’s rider allocation solves linear systems (Gauss-Seidel) in real-time to match supply (riders) with demand (orders) across Kathmandu’s 15 zones, reducing wait times by 30%.
• NTC’s power grid employs finite difference PDE solvers to simulate voltage drops across 220kV transmission lines, preventing blackouts during peak demand (e.g., monsoon season).