10.3 Robust Design, Parameter Design, and Verification

Creating a high-performance design architecture requires configuring systems so that functional performance is virtually immune to real-world operational variation. Robust design, tolerance allocation, stochastic simulation, and rigorous verification testing form the quantitative core of Design for Six Sigma (DFSS).

---

Robust Design & The Taguchi Methodology

Developed by Dr. Genichi Taguchi, Robust Design focuses on making product performance insensitive (robust) to variation caused by uncontrollable environmental and operational noise factors, without eliminating the noise sources themselves.

The Taguchi Loss Function

Traditional manufacturing operates under a binary "goal-post" mentality: any product measuring within Lower and Upper Specification Limits (LSL and USL) is deemed 100% acceptable with zero financial loss, whereas products outside limits incur total loss. Taguchi refuted this concept, proving that quality loss occurs continuously as performance departs from the ideal target value ($T$).

Taguchi modeled societal loss using a symmetric Quadratic Loss Function:

$L(y) = k(y - T)^2$

Where:

• $L(y)$ is the economic loss per unit in monetary units ($).
• $y$ is the actual value of the quality characteristic.
• $T$ is the nominal target value.
• $k$ is the financial loss coefficient, calculated as:

$k = \frac{A_0}{\Delta^2}$

Where $A_0$ is the average financial loss incurred if the product exceeds the specification limit (e.g., warranty repair, replacement cost), and $\Delta$ is the distance from the target to the specification limit ($\Delta = |USL - T|$).

                  TAGUCHI QUADRATIC LOSS FUNCTION
    Loss ($)
      ^                                           * (LSL/USL Boundary)
  A0 -|---- *                                   * 
      |      *                                 *  
      |       *                               *   L(y) = k(y - T)^2
      |        *                             *    
   0 -+---------+-------------+-------------+------------>
             LSL             Target (T)    USL       Characteristic (y)

Three-Stage Design Process

Taguchi categorized system design into three sequential stages:

1. System Design: Primary engineering design phase where technology concepts, basic circuit topographies, structural layouts, and material classes are selected.
2. Parameter Design: The critical optimization phase where nominal settings of control factors are systematically adjusted to minimize sensitivity to noise factors. This phase maximizes product quality at zero added cost.
3. Tolerance Design: The final phase, applied only if parameter design fails to reduce functional variation sufficiently. Tight component tolerances are specified, requiring higher-precision materials and manufacturing equipment, thereby increasing unit manufacturing cost.

Control Factors vs. Noise Factors

• Control Factors: Design parameters ($X_i$) easily set and maintained by the design engineer (e.g., resistor values, material thickness, gear ratio, target surface finish).
• Noise Factors: Variables impossible or cost-prohibitive to control during customer use. Categorized into three types:
  • External Noise: Operational environment variations (e.g., ambient temperature, humidity, road vibration, input voltage surges).
  • Internal Noise: Physical deterioration over time (e.g., component wear, material oxidation, mechanical fatigue, thermal aging).
  • Unit-to-Unit Noise: Manufacturing process variation between individual production units.

Signal-to-Noise ($S/N$) Ratios

Taguchi introduced Signal-to-Noise ($S/N$) Ratios expressed in decibels ($dB$) to evaluate performance robustness. A higher $S/N$ ratio indicates superior quality, as functional signal dominates unwanted noise variation. Formulas depend on the CTQ objective:

• Nominal-is-the-Best (Targeted value, e.g., output voltage, gap dimension):

$S/N = 10 \log_{10} \left( \frac{\bar{y}^2}{s^2} \right)$

• Smaller-is-Better (Minimizing undesirable attribute, e.g., wear rate, shrinkage, parasitic resistance):

$S/N = -10 \log_{10} \left( \frac{1}{n} \sum_{i=1}^n y_i^2 \right)$

• Larger-is-Better (Maximizing desirable attribute, e.g., tensile strength, battery runtime, thermal efficiency):

$S/N = -10 \log_{10} \left( \frac{1}{n} \sum_{i=1}^n \frac{1}{y_i^2} \right)$

---

Transfer Functions ($Y = f(X)$) & Sensitivity Analysis

A Transfer Function is a mathematical model explicitly linking critical output characteristics ($Y$) to input variables ($X_1, X_2, \dots, X_n$):

$Y = f(X_1, X_2, \dots, X_n)$

Transfer functions are derived either analytically from fundamental physical laws (e.g., Ohm's Law $V = IR$, stress equations $\sigma = F/A$) or empirically using Design of Experiments (DOE) and Response Surface Methodology (RSM).

Sensitivity Analysis via Partial Derivatives

Using the Delta Method (Error Propagation), output variance $\sigma_Y^2$ is estimated by expanding the transfer function via Taylor series partial derivatives:

$\sigma_Y^2 \approx \sum_{i=1}^n \left( \frac{\partial f}{\partial X_i} \right)^2 \sigma_{X_i}^2 + 2 \sum_{i<j} \left( \frac{\partial f}{\partial X_i} \right) \left( \frac{\partial f}{\partial X_j} \right) \text{Cov}(X_i, X_j)$

Where $\frac{\partial f}{\partial X_i}$ is the sensitivity partial derivative of output $Y$ with respect to input parameter $X_i$. If the derivative magnitude $|\frac{\partial f}{\partial X_i}|$ is large, small variations in input $X_i$ cause massive instability in output $Y$. Robust parameter design operates by selecting operating points where the partial derivatives approach zero.

---

Monte Carlo Simulation in Design Verification

When transfer functions $Y = f(X)$ are highly non-linear or input variables follow non-normal probability distributions (e.g., Weibull, Lognormal, Uniform), analytical error propagation formulas break down. In these cases, Monte Carlo Simulation is utilized to stochastically model system variation.

Monte Carlo Simulation Algorithm

1. Define Mathematical Model: Input the transfer function equation $Y = f(X_1, X_2, \dots, X_n)$.
2. Assign Input Distributions: Specify probability density functions (PDFs), means ($\mu$), standard deviations ($\sigma$), and shape parameters for each input variable $X_i$.
3. Random Pseudo-Sampling: The simulation engine draws tens of thousands ($N \ge 10,000 \text{ to } 100,000$) of random input value combinations from their designated distributions.
4. Compute Output Vector: Evaluate output $Y_k$ for every trial run $k$.
5. Analyze Output Histogram: Generate simulated output distribution histograms, calculate predicted capability indices ($P_p, P_{pk}$), estimate DPMO defect rates, and perform sensitivity ranking (Tornado Charts).

---

Prototype Testing, Design Verification & Validation

Before releasing a design to production, physical and virtual prototypes must undergo formal testing defined in the Design Verification Plan and Report (DVP&R).

Verification vs. Validation

• Design Verification: Confirms that design outputs satisfy specified design inputs ("Did we design the product right?"). Evaluated through FEA simulation, dimensional inspection, bench testing, and schematic analysis.
• Design Validation: Confirms that the final product satisfies real-world user requirements, environmental conditions, and operational intended use ("Did we design the right product?"). Evaluated through customer field testing and mission profile simulation.

Accelerated Testing: HALT vs. HASS

• HALT (Highly Accelerated Life Testing): Applied during early product development to discover design weaknesses and stress boundaries. Test samples are subjected to extreme multi-axis vibration, rapid thermal cycling (e.g., $-60^\circ\text{C to } +150^\circ\text{C}$), and electrical overstress well beyond specification limits until destruction occurs. Uncovered failure mechanisms are mitigated through redesign.
• HASS (Highly Accelerated Stress Screening): Implemented in manufacturing production to screen 100% of manufactured units or lot samples using stress levels lower than HALT to catch latent assembly workmanship defects before shipment.