QUALITY IS NOT SIMPLY THE ABSENCE OF FAILURE
In automotive engineering, quality does not end when a vehicle successfully leaves the factory gate.
It begins there.
A vehicle is expected to perform consistently through thousands of kilometres, changing environmental conditions, varying customer usage patterns and years of ownership. Therefore, the real measure of engineering excellence is not merely “Did the vehicle pass the test?” but rather:
“How consistently does the vehicle perform throughout its intended life?”
This is where Genichi Taguchi’s Loss Function, reliability engineering and life-cycle analysis come together to create a powerful approach to automotive quality.
🔧 Design the quality.
📊 Measure the variation.
🔍 Understand the degradation.
🚗 Predict the reliability.
♻️ Improve the entire life-cycle.
GENICHI TAGUCHI AND THE LOSS FUNCTION 🎯
The conventional understanding of quality often treats quality as a specification problem.
If a characteristic lies between the Upper Specification Limit (USL) and Lower Specification Limit (LSL), the product is considered acceptable.
Taguchi challenged this thinking.
His philosophy was that quality loss generally begins as a product characteristic moves away from its target, even when the product remains technically within specification. The loss may be economic, functional, environmental or customer-related, and its actual form depends on the characteristic and the consequences of deviation.
For a nominal-the-best characteristic, the classical quadratic Taguchi Loss Function is expressed as:
L(y) = k(y − m)²
Where:
- L(y) = estimated quality loss associated with the measured value
- y = actual measured value
- m = target value
- k = proportionality constant
The important message is:
QUALITY LOSS IS NOT NECESSARILY ZERO JUST BECAUSE THE PRODUCT IS WITHIN SPECIFICATION.
The further the characteristic moves from the target, the greater the estimated loss under the quadratic model.
The constant k is often estimated from the loss assigned to a specified deviation, such as the specification limit:
k = A / Δ²
Where:
- A = estimated loss at the reference deviation
- Δ = distance from the target to that reference point
The quadratic form is an engineering approximation, not a universal physical law. It is most appropriate when loss is reasonably symmetric and increases smoothly around the target.
Taguchi also described other characteristic types:
- Nominal-the-best: performance is best at a target value.
- Smaller-the-better: loss tends to increase as the response increases, with the ideal value often being zero.
- Larger-the-better: loss tends to decrease as the response increases, with the ideal value tending toward infinity.
For smaller-the-better and larger-the-better characteristics, the loss model is not simply the same symmetric parabola used for nominal-the-best characteristics.
📈 This transforms quality thinking from:
“Is it within specification?”
to:
“How close are we consistently operating to the intended target, and what is the consequence of deviation?”
THE TAGUCHI LOSS FUNCTION CURVE 📈
For a nominal-the-best characteristic, the classical loss function is a U-shaped parabola, not a bell-shaped curve. The optimum target is at its centre.
At the target:
Deviation = 0 → Modelled loss is minimum
As deviation increases in either direction:
Deviation ↑ → Modelled loss ↑
Under the quadratic model:
Loss increases with the square of the deviation.
This principle has important significance in automotive engineering.
Consider a component dimension, steering characteristic, brake response, NVH parameter, compressor performance, battery characteristic or suspension property.
Two components may both satisfy the specification.
However:
- Component A is very close to target.
- Component B is close to the specification boundary.
Both may be classified as OK.
But their expected performance, margin to functional limits and potential customer impact may not be equivalent.
The Taguchi philosophy therefore encourages engineers to reduce variation around the target, while also ensuring that the target itself is correctly defined and that the loss model reflects the actual functional and customer consequences.
🎯 Target orientation + variation reduction = a basis for robust quality.
FROM QUALITY TO RELIABILITY 🔧📊
Reliability engineering takes this philosophy one step further.
Reliability is the probability that an item performs its intended function for a specified duration under specified conditions.
Reliability is therefore a conditional and time-dependent concept. It depends on the definition of the required function, the operating environment, the usage profile, maintenance assumptions and the population being studied.
In an automobile, reliability is not restricted to one component.
It exists at multiple levels:
Material → Part → Sub-system → System → Vehicle → Customer
A vehicle may contain thousands of components, but the customer experiences them as one integrated product.
A failure of a relatively small component can therefore become a major customer experience, particularly when it affects safety, mobility, comfort or vehicle availability.
This is why reliability engineering must connect:
Design → Manufacturing → Testing → Validation → Field Performance → Customer Usage → Life-cycle
RELIABILITY IS A LIFE-CYCLE PHILOSOPHY 🔄
Automotive reliability should be viewed across the complete product life-cycle.
1️⃣ DESIGN
Reliability begins with design intent and clearly defined requirements.
Tools such as:
- Design FMEA
- Reliability prediction
- Robust design
- Design of Experiments
- Stress-strength analysis
- Tolerance analysis
- Reliability Block Diagrams
- Fault Tree Analysis
- Physics-of-failure analysis
help engineers anticipate potential failure mechanisms and quantify risk.
These tools support reliability engineering, but none of them alone guarantees field reliability. Their effectiveness depends on sound assumptions, representative loads, accurate failure mechanisms and effective verification.
2️⃣ DEVELOPMENT & VALIDATION
The design must then be challenged under representative and, where justified, accelerated conditions.
Typical activities include:
- Durability testing
- Environmental testing
- Thermal cycling
- Vibration testing
- Water and dust ingress testing
- Corrosion testing
- Electrical endurance
- Powertrain durability
- Battery endurance
- System-level validation
Validation should demonstrate that the product meets defined requirements for the intended usage envelope. A test that is more severe than customer usage is not automatically a valid reliability test; the applied stresses must remain relevant to the intended failure mechanisms.
3️⃣ PRODUCTION
A robust design can still suffer from manufacturing variation.
Therefore:
Process capability + statistical control + mistake-proofing + traceability
become critical.
Process capability indices are meaningful only when the measurement system is adequate, the process is reasonably stable and the specification and target are appropriate.
4️⃣ FIELD
This is where engineering assumptions meet reality.
Customer usage generates valuable reliability information.
Warranty claims, field failures, roadside assistance data, service records, customer complaints and technical campaigns become important sources of reliability intelligence. However, field data must be interpreted carefully because reporting behaviour, exposure, mileage, repair practices, vehicle mix and warranty policy can introduce bias.
5️⃣ END-OF-LIFE
A mature reliability system also considers degradation, repairability, maintainability, recyclability and the eventual end-of-life behaviour of the vehicle.
♻️ True reliability thinking therefore spans the entire product life-cycle.
THE IMPORTANCE OF MIS ANALYSIS IN AUTOMOTIVE RELIABILITY 🚗📅
One commonly used approach in automotive reliability monitoring is MIS — Months In Service analysis.
The vehicle population is grouped according to the duration for which vehicles have been operating in the field.
For example:
0 MIS → Newly commissioned vehicles or vehicles with negligible recorded service exposure
6 MIS → Approximately six months in service
12 MIS → Approximately one year in service
24 MIS → Approximately two years in service
36 MIS → Approximately three years in service
and so on.
The exact MIS buckets and definitions may vary by OEM, product and reliability programme. MIS is an exposure or age measure; it is not automatically equivalent to mileage, operating hours, number of duty cycles or component life.
A vehicle at 12 MIS may have accumulated very different mileage and environmental exposure from another vehicle at 12 MIS. Therefore, MIS analysis should often be supplemented with mileage, engine hours, starts, cycles, climate, road conditions, usage severity and vehicle population at risk.
The fundamental objective remains:
Understand how failures and claims evolve as vehicles accumulate real-world operating time and usage exposure.
Rates should generally be calculated using an appropriate denominator, such as vehicles at risk, vehicle-months, kilometres or operating hours. Raw claim counts by MIS can be misleading when the number of vehicles exposed differs between age groups.
6 MIS: THE EARLY FIELD SIGNAL 🔎
The 6 MIS window can provide an early indication of field behaviour, but it is not a universal reliability milestone.
Engineers may examine:
- Early component failures
- Supplier-related issues
- Assembly-related problems
- Software or calibration issues
- Initial customer usage effects
- Early degradation mechanisms
- Warranty trends
A high failure rate during the early months may indicate an infant-mortality pattern, but it may also reflect reporting delays, campaign activity, exposure differences, installation effects or a failure mechanism unrelated to manufacturing quality.
A sudden increase in failures during the early months should therefore be treated as a signal for investigation, not as proof of infant mortality.
This can trigger deeper investigation into:
Design → Process → Supplier → Assembly → Usage → Reporting and service factors
12 MIS: THE STABILITY CHECK 📊
At 12 MIS, the organisation may obtain a more mature picture of field behaviour for some failure modes.
The analysis can reveal:
- Recurring failures
- Emerging failure modes
- Component durability trends
- Warranty incidence
- Repair frequency
- Supplier performance
- Design robustness
- Usage-related patterns
The important question becomes:
Are failures stabilising, reducing or beginning to increase after accounting for exposure, vehicle mix and reporting effects?
This distinction is extremely important.
A low number of failures at one point in time does not automatically indicate strong reliability.
The trend, the exposure-adjusted rate, the confidence interval and the failure-mode distribution are often more informative than an isolated number.
24 MIS AND BEYOND ⏳
The 24 MIS analysis may become valuable for understanding longer-term reliability behaviour, but the appropriate observation period depends on the product, component, warranty coverage, usage profile and expected life.
Certain failure mechanisms may not appear during initial validation or early ownership.
Examples may include:
- Wear-out mechanisms
- Corrosion
- Seal degradation
- Fatigue
- Thermal ageing
- Electrical degradation
- Battery capacity deterioration
- Bearing deterioration
- Suspension wear
- HVAC performance degradation
As the vehicle population ages, reliability engineers may identify patterns consistent with:
Early-life failures → Approximately constant or random failure behaviour → Wear-out failures
These are analytical categories, not guaranteed sequential phases for every component or vehicle. A product may show no clear bathtub pattern, and different failure modes within the same vehicle may have different life distributions.
This provides valuable information for product improvement and future design generations when exposure, censoring, competing risks and failure-mode definitions are properly considered.
THE BATHTUB CURVE 🛁📈
A classical reliability concept is the Bathtub Curve.
It represents a possible pattern of hazard rate, or instantaneous failure risk conditional on survival to a given age, rather than simply the number of failures observed.
It broadly describes three phases:
🔹 INFANT MORTALITY
A decreasing hazard rate, often associated with:
- Manufacturing defects
- Assembly errors
- Supplier issues
- Installation problems
- Weak units being removed from the population
Design weaknesses can contribute to early failures, but they do not automatically produce an infant-mortality curve.
🔹 USEFUL LIFE
An approximately constant hazard-rate region in which failures may be treated as broadly random with respect to age, subject to the assumptions of the selected model.
🔹 WEAR-OUT
An increasing hazard rate associated with mechanisms such as:
- Wear
- Fatigue
- Corrosion
- Ageing
- Repeated thermal or mechanical stress
The bathtub curve is a conceptual model, not a universal law. Many products and individual automotive components do not exhibit all three phases. Mixtures of failure modes, maintenance, warranty replacement, design changes and changing usage can also distort the observed pattern.
This model can provide a useful conceptual bridge between MIS analysis and life-cycle reliability engineering, but MIS data alone cannot establish a bathtub curve without suitable exposure data and statistical analysis.
TAGUCHI + RELIABILITY ENGINEERING = ROBUST DESIGN 🎯
The real power comes when Taguchi’s philosophy is integrated with reliability engineering.
Taguchi asks:
How can we reduce sensitivity to controllable and uncontrollable variation?
Reliability engineering asks:
How will the product behave over time and under defined operating conditions?
Together, they can support:
ROBUST RELIABILITY
A robust automotive product is not merely one that performs well under ideal laboratory conditions.
It should remain dependable despite variations in:
- Temperature 🌡️
- Humidity 💧
- Road conditions 🛣️
- Driving behaviour 🚗
- Manufacturing variation 🏭
- Component variation ⚙️
- Voltage fluctuations ⚡
- Environmental exposure 🌧️
- Customer usage 👨👩👧
- Ageing and degradation ⏳
Robust design reduces sensitivity to relevant noise factors, but it does not eliminate the need for appropriate requirements, validation, maintenance assumptions and field monitoring.
RELIABILITY DATA SHOULD TELL A STORY 📖
A reliability dashboard should not become merely a collection of numbers.
It should answer engineering questions.
What failed?
Where did it fail?
When did it fail?
Why did it fail?
How frequently did it fail relative to the population at risk?
What was the operating condition and exposure?
Is the failure random, systematic or associated with a specific failure mechanism?
Is the failure rate increasing with MIS, mileage, hours or cycles?
Is the failure linked to a particular supplier, plant, variant or application?
What is the recurrence after corrective action?
This converts data into engineering knowledge.
FROM FAILURE DATA TO FAILURE MECHANISM 🔬
One of the most important principles of reliability engineering is:
Do not stop at the failure mode. Understand the failure mechanism.
For example:
Customer complaint: Compressor not cooling.
That is only the symptom.
Further investigation may reveal:
Insufficient cooling → refrigerant loss → leakage → seal degradation → material incompatibility or installation issue → environmental and operating exposure
This is an illustrative chain, not a predetermined diagnosis. Other causes may include compressor damage, control faults, blockage, incorrect charge, sensor errors or software issues.
The engineering investigation must move progressively towards the physical mechanism using evidence from testing, inspection, teardown, data analysis and reproduction of the failure.
Tools such as:
- 5 Why
- Fishbone Analysis
- Fault Tree Analysis
- FMEA
- Weibull Analysis
- Pareto Analysis
- Regression
- DOE
- Accelerated Life Testing
can then be deployed according to the problem. Statistical tools identify patterns; physical analysis is usually required to confirm the mechanism.
WEIBULL ANALYSIS — A POWERFUL RELIABILITY TOOL 📈
Weibull analysis is one of the widely used statistical approaches in reliability engineering.
It can help engineers model time-to-failure or life data and assess whether the observed data are consistent with decreasing, approximately constant or increasing hazard behaviour.
The Weibull shape parameter, commonly represented by β, provides important insight into the fitted Weibull model:
β < 1 → decreasing modelled hazard rate
β = 1 → constant modelled hazard rate
β > 1 → increasing modelled hazard rate
These interpretations apply to the fitted Weibull distribution and should not be treated as automatic proof of a physical failure mechanism.
The Weibull distribution is a model, not a universal description of all reliability data. Its suitability should be checked using engineering knowledge, graphical diagnostics, goodness-of-fit assessment, confidence intervals and comparison with alternative distributions where appropriate.
Field-return and durability data also require careful treatment of:
- Right-censored units that have not failed
- Warranty truncation
- Different exposure measures
- Multiple failure modes
- Competing risks
- Design or process changes
- Repeated repairs
- Non-random sampling
- Differences in vehicle usage
A Weibull plot can be useful, but it should not be interpreted without considering these factors.
ACCELERATED LIFE TESTING ⚡
Automotive products cannot always be tested for their entire intended life in real time.
Therefore, engineers may use Accelerated Life Testing (ALT).
The objective is to expose the product to controlled stresses that accelerate degradation while maintaining a scientifically defensible relationship with normal operating conditions.
Potential stresses include:
- Temperature
- Voltage
- Vibration
- Pressure
- Humidity
- Mechanical load
- Thermal cycling
The appropriate acceleration model depends on the failure mechanism. Examples include Arrhenius-type relationships for some temperature-driven chemical processes, inverse-power relationships for certain stress-life behaviours and other models for humidity, voltage or mechanical loading.
The challenge is critical:
Acceleration must reproduce the relevant failure mechanism and preserve the relationship between test conditions and use conditions.
If the stress level is too severe, the test may activate a different failure mechanism, create unrealistic damage or alter the failure-mode ranking.
Therefore, an accelerated test should normally include:
- A clearly defined failure mechanism
- Justification for the selected stress
- Appropriate use-condition reference data
- Verification that the failure mode remains relevant
- A defined acceleration model
- Adequate sample size and statistical planning
- Consideration of interactions between stresses
Otherwise, the test may become fast — but not representative.
OEM RELIABILITY ENGINEERING 🏭🚗
For an automotive OEM, reliability is inherently cross-functional.
It involves:
R&D + Quality + Manufacturing + Supplier Quality + Purchasing + Service + Warranty + Product Planning + Customer Experience
Reliability engineering therefore cannot remain confined to a laboratory.
It needs a closed-loop system:
CUSTOMER → FIELD DATA → FAILURE ANALYSIS → ROOT CAUSE → COUNTERMEASURE → DESIGN / PROCESS CHANGE → VALIDATION → FIELD CONFIRMATION
This is the essence of a closed-loop quality system.
🔄 Learn → Improve → Validate → Standardise → Learn Again
The loop should also include configuration control and change management so that improvements can be traced to the affected vehicles, parts and production periods.
LIFE-CYCLE QUALITY: THE BIGGER PICTURE 🌍
A truly mature organisation does not optimise only one stage of the product life-cycle.
It considers:
Concept
⬇️
Design
⬇️
Development
⬇️
Validation
⬇️
Industrialisation
⬇️
Production
⬇️
Customer Usage
⬇️
Service & Warranty
⬇️
Ageing
⬇️
End-of-Life
Every stage generates information for the next generation.
Therefore, one of the most powerful principles in automotive quality is:
FIELD EXPERIENCE MUST BECOME DESIGN KNOWLEDGE.
THE ULTIMATE OBJECTIVE: CUSTOMER TRUST 🤝
Customers rarely talk about statistical distributions, Weibull parameters or Taguchi coefficients.
They simply expect the vehicle to:
Start.
Move.
Perform.
Remain safe.
Remain dependable.
Provide consistent performance.
And continue doing so year after year.
That is the ultimate purpose of reliability engineering.
FROM “PASS/FAIL” TO “PREDICT/PREVENT” 🚀
The evolution of automotive quality can be viewed as a journey:
Inspection
⬇️
Detection
⬇️
Prevention
⬇️
Robust Design
⬇️
Reliability Prediction
⬇️
Field Monitoring
⬇️
Predictive Reliability
The future belongs to organisations that can move from:
“We detected a failure.”
to:
“We predicted the degradation before the customer experienced the failure.”
That is where AI, connected vehicles, telematics, digital twins, predictive analytics and advanced reliability modelling are creating new opportunities.
Prediction remains probabilistic and model-dependent. It should support engineering judgement and verification rather than replace them.
THE FINAL CONNECTION 🎯
Genichi Taguchi’s Loss Function teaches us that deviation from target can create increasing loss, with the classical quadratic form applying to a particular nominal-the-best modelling assumption.
Reliability engineering teaches us that performance must be sustained over time and under defined conditions.
MIS analysis helps engineers examine how failures and claims evolve with vehicle age and service exposure, provided that appropriate denominators and confounding factors are considered.
Weibull analysis helps engineers model life distributions and assess hazard-rate behaviour, but the fitted model must be validated against the data and the physical failure mechanism.
Life-cycle thinking ensures that learning continues from design to field and back into the next generation of products.
Together, these principles create a powerful philosophy:
QUALITY IS DESIGNED.
RELIABILITY IS ENGINEERED.
ROBUSTNESS IS BUILT.
FAILURE IS LEARNED FROM.
AND CUSTOMER TRUST IS EARNED OVER THE ENTIRE LIFE-CYCLE.
🚗⚙️📊🔬🎯📈🔄🏭🌍🤝♻️🚀
RECOMMENDED REFERENCE BOOKS 📚
- QUALITY ENGINEERING USING ROBUST DESIGN — GENICHI TAGUCHI
- INTRODUCTION TO QUALITY ENGINEERING — GENICHI TAGUCHI
- RELIABILITY ENGINEERING — E. E. LEWIS
- RELIABILITY ENGINEERING AND RISK ANALYSIS — T. LEWIS / RELATED RELIABILITY ENGINEERING REFERENCES
- STATISTICAL METHODS FOR RELIABILITY DATA — W. Q. MEAKER / RELIABILITY STATISTICS REFERENCES
For deeper automotive application, engineers should also refer to established OEM reliability methodologies, AIAG/VDA FMEA, automotive durability standards and recognised reliability-analysis practices.
CONCLUSION 🚘
The most mature definition of automotive quality is not simply conformance to specification.
It is the ability to deliver consistent, robust and reliable performance throughout the intended life of the product, despite the inevitable variation in manufacturing, environment and customer usage.
And perhaps that is the most important lesson from Taguchi:
A specification tells us the boundary.
A target tells us the aspiration.
Reliability tells us whether we can sustain it.
🎯 DESIGN FOR QUALITY.
🔧 ENGINEER FOR RELIABILITY.
📊 LEARN FROM DATA.
🚗 IMPROVE THROUGH THE LIFE-CYCLE.
🤝 CREATE LASTING CUSTOMER TRUST.
