The Central Idea
Reliability and quality control both ask whether a system is performing acceptably.
Reliability focuses on whether components and systems continue to function.
Quality control focuses on whether outputs remain within expected and acceptable limits.
The shared mathematical idea is:
\[ \boxed{\text{measure variation, model failure or defect, and decide when intervention is justified}} \]Reliability as a Probability
If a component works successfully over a specified time interval with probability \(R\), then
\[ \boxed{R=P(\text{component works})}. \]If the only alternatives are working and failed, then failure probability is
\[ \boxed{1-R}. \]A First Reliability Example
Suppose a component works for the required mission with probability
\[ R=0.97. \]Then its failure probability is
\[ 1-0.97=\boxed{0.03}. \]The reliability statement is meaningful only relative to a specified time, environment, and definition of successful operation.
Reliability Depends on the Mission
A component might be highly reliable over one hour but much less reliable over ten years.
Reliability should therefore be tied to:
- a time interval;
- operating conditions;
- a definition of failure;
- the required function.
Series Systems
A series system succeeds only if every required component succeeds.
If independent components have reliabilities
\[ R_1,R_2,\ldots,R_n, \]then system reliability is
\[ \boxed{R_{\text{series}}=R_1R_2\cdots R_n}. \]Series Reliability Example
Suppose three independent required components have reliabilities
\[ 0.98,\qquad0.96,\qquad0.99. \]Then
\[ R_{\text{series}} =(0.98)(0.96)(0.99) \approx \boxed{0.9314}. \]The system is less reliable than any one component because every component must work.
Parallel Systems
In a basic parallel redundancy model, the system works if at least one component works.
For independent components,
\[ P(\text{all fail}) = (1-R_1)(1-R_2)\cdots(1-R_n). \]Therefore,
\[ \boxed{ R_{\text{parallel}} = 1-\prod_{i=1}^{n}(1-R_i) }. \]Parallel Reliability Example
Suppose two independent redundant components each have reliability
\[ 0.90. \]The probability both fail is
\[ (0.10)(0.10)=0.01. \]Therefore,
\[ R_{\text{parallel}} = 1-0.01 = \boxed{0.99}. \]Redundancy
Redundancy adds extra components or paths so that failure of one element does not necessarily cause system failure.
Redundancy can improve reliability, but it may also increase:
- cost;
- weight;
- maintenance;
- complexity;
- common-cause exposure.
Independence Matters
The simple series and parallel formulas above assume component failures are independent.
That assumption may fail when components share:
- the same power source;
- the same environment;
- the same software defect;
- the same supplier defect;
- the same physical enclosure.
Redundancy is less valuable when failures are strongly correlated.
Common-Cause Failure
A common-cause failure affects several components through one shared cause.
For example, two redundant servers may both fail if they depend on the same unavailable power source.
This is why physical, electrical, software, and organizational independence can matter in resilient design.
Failure Rate
A failure rate describes how frequently failures occur relative to operating exposure.
A simplified empirical rate might be
\[ \boxed{ \text{failure rate} = \frac{\text{number of failures}}{\text{total operating time}} }. \]More advanced reliability models distinguish between average rates and time-dependent hazard rates.
Exponential Reliability Model
Under a constant hazard-rate model with rate \(\lambda\), reliability over time \(t\) is
\[ \boxed{R(t)=e^{-\lambda t}}. \]The corresponding failure probability by time \(t\) is
\[ \boxed{F(t)=1-e^{-\lambda t}}. \]Exponential Reliability Example
Suppose
\[ \lambda=0.002 \]failures per hour.
Over \(100\) hours,
\[ R(100)=e^{-0.002(100)} =e^{-0.2} \approx \boxed{0.8187}. \]Mean Time to Failure
Under the exponential model, the mean time to failure is
\[ \boxed{\operatorname{MTTF}=\frac{1}{\lambda}}. \]If
\[ \lambda=0.002\text{ failures/hour}, \]then
\[ \operatorname{MTTF} = \frac{1}{0.002} = \boxed{500\text{ hours}}. \]Repairable Systems
Some systems can be restored after failure.
Useful measures then include:
- mean time between failures;
- mean time to repair;
- availability;
- downtime.
Reliability and availability are related but not identical.
Availability
In a simple repairable-system model,
\[ \boxed{ A = \frac{\operatorname{MTBF}} {\operatorname{MTBF}+\operatorname{MTTR}} }. \]Here:
- MTBF is mean time between failures;
- MTTR is mean time to repair.
High reliability and fast repair can both improve availability.
Availability Example
Suppose
\[ \operatorname{MTBF}=1000\text{ hours} \]and
\[ \operatorname{MTTR}=10\text{ hours}. \]Then
\[ A = \frac{1000}{1000+10} = \frac{1000}{1010} \approx \boxed{0.9901}. \]Quality Control
Quality control uses measurements and decision rules to evaluate whether products, services, or processes meet defined standards.
A quality-control system may monitor:
- dimensions;
- weights;
- defect counts;
- error rates;
- processing times;
- failure frequencies;
- customer-service outcomes.
Conformance and Performance
Quality can be evaluated in different ways.
Conformance quality asks whether output meets a specified requirement.
Performance quality concerns how well the output actually performs its intended function.
A product can conform to a poorly chosen specification and still perform badly in practice.
Defects and Defectives
A defect is a nonconformity or quality problem.
A defective unit is a unit classified as unacceptable according to the chosen criteria.
One unit can sometimes contain several defects.
Therefore, defect count and defective-unit count are different quantities.
Defect Proportion
If \(x\) defective units are found in a sample of size \(n\), the sample defect proportion is
\[ \boxed{\hat p=\frac{x}{n}}. \]If \(7\) defective units are found among \(200\),
\[ \hat p=\frac{7}{200}=\boxed{0.035}. \]The observed defective proportion is \(3.5\%\).
Sampling Inspection
Inspecting every unit can be expensive, slow, or destructive.
Sampling inspection uses a subset of units to infer something about a larger lot or process.
The sample should be chosen in a way that supports the intended statistical inference.
Acceptance Sampling
Acceptance sampling uses a sample to decide whether to accept or reject a lot.
A simple plan might specify:
- sample size \(n\);
- acceptance number \(c\).
The lot is accepted if the sample contains at most \(c\) defectives.
Acceptance-Sampling Example
Suppose
\[ n=50 \]and
\[ c=2. \]A sample contains \(1\) defective unit.
Because
\[ 1\le2, \]the lot is accepted under this plan.
Producer and Consumer Risk
Sampling decisions can make errors.
A good lot can be rejected.
A poor lot can be accepted.
These risks are often discussed as:
- producer's risk: rejecting acceptable quality;
- consumer's risk: accepting unacceptable quality.
Process Variation
No real process produces perfectly identical outputs.
Some variation is routine and built into the process.
Other variation is associated with identifiable changes or disturbances.
Statistical process control attempts to distinguish these types of variation.
Common-Cause and Special-Cause Variation
Common-cause variation is the routine variation inherent in a stable process.
Special-cause variation arises from a particular disturbance or change.
Examples of special causes may include:
- tool damage;
- incorrect settings;
- material changes;
- sensor failure;
- unusual environmental conditions.
Control Charts
A control chart plots a process statistic over time and compares it with reference limits.
A typical chart contains:
- a center line;
- an upper control limit;
- a lower control limit.
These limits are designed to help identify unusual process behavior.
Control Limits Are Not Specification Limits
Control limits describe expected process variation under a statistical model.
Specification limits describe external requirements for acceptable output.
A process can be statistically stable but centered outside the desired specification.
Conversely, a process can currently meet specifications while showing signs of instability.
Three-Sigma Control Limits
A common control-chart structure places limits approximately three standard errors from the center line:
\[ \boxed{ \text{UCL} = \text{center}+3(\text{standard error}) } \]and
\[ \boxed{ \text{LCL} = \text{center}-3(\text{standard error}) }. \]The exact formulas depend on the statistic being monitored.
p-Charts
A p-chart monitors the proportion of defective units.
If the long-run process proportion is estimated by \(\bar p\) and samples all have size \(n\), then the standard error is
\[ \boxed{ \sqrt{\frac{\bar p(1-\bar p)}{n}} }. \]p-Chart Control Limits
The center line is
\[ \boxed{\bar p}. \]Approximate three-sigma limits are
\[ \boxed{ \operatorname{UCL} = \bar p + 3\sqrt{\frac{\bar p(1-\bar p)}{n}} } \]and
\[ \boxed{ \operatorname{LCL} = \bar p - 3\sqrt{\frac{\bar p(1-\bar p)}{n}} }. \]If the lower expression is negative, a practical p-chart lower limit is often set to \(0\).
p-Chart Example
Suppose
\[ \bar p=0.04 \]and
\[ n=100. \]The standard error is
\[ \sqrt{\frac{0.04(0.96)}{100}} = \sqrt{0.000384} \approx0.0196. \]The upper limit is
\[ 0.04+3(0.0196) \approx \boxed{0.0988}. \]The lower expression is negative, so the practical lower limit is
\[ \boxed{0}. \]Mean Charts
For quantitative measurements, sample means can be monitored over time.
If the process mean is \(\mu\), process standard deviation is \(\sigma\), and sample size is \(n\), then
\[ \operatorname{SE}(\bar X)=\frac{\sigma}{\sqrt n}. \]A theoretical three-sigma structure is
\[ \mu\pm3\frac{\sigma}{\sqrt n}. \]Operational control charts often estimate these quantities from process data.
Patterns Matter, Not Just Single Points
A point outside a control limit is an obvious signal, but other unusual patterns can also matter.
Examples include:
- long runs on one side of the center line;
- persistent upward or downward trends;
- unusual clustering;
- sudden shifts in level;
- unexpected changes in variability.
Formal run rules depend on the quality-control standard or procedure being used.
False Alarms
Even a stable process can occasionally produce a statistic outside a control limit by chance.
Therefore, a control signal does not prove that a special cause exists.
It indicates that investigation may be warranted.
Missed Signals
A process can change without immediately producing an obvious control-chart signal.
Quality monitoring therefore involves a tradeoff between:
- false alarms;
- missed or delayed detection.
This is closely related to Type I and Type II error reasoning.
Reliability and Quality Interact
Manufacturing quality can affect future reliability.
For example, poor assembly quality may increase the probability of early component failure.
Likewise, reliability failures can reveal quality problems in design, production, maintenance, or use.
Defect Probability and Binomial Models
If each inspected unit is independently defective with probability \(p\), then the number of defectives \(X\) in a sample of size \(n\) can be modeled as
\[ \boxed{X\sim\operatorname{Binomial}(n,p)}. \]Then
\[ P(X=k) = \binom{n}{k}p^k(1-p)^{n-k}. \]Binomial Quality Example
Suppose the defect probability is
\[ p=0.02 \]and \(10\) independent units are inspected.
The probability of finding no defectives is
\[ P(X=0) = (0.98)^{10} \approx \boxed{0.8171}. \]Probability of At Least One Defective
Using the complement rule,
\[ P(X\ge1) = 1-P(X=0). \]In the previous example,
\[ P(X\ge1) = 1-(0.98)^{10} \approx \boxed{0.1829}. \]Poisson Defect Counts
When defects occur as counts over an area, length, time interval, or other exposure, a Poisson model may sometimes be appropriate.
If
\[ X\sim\operatorname{Poisson}(\lambda), \]then
\[ P(X=k) = \boxed{ \frac{e^{-\lambda}\lambda^k}{k!} }. \]Whether the Poisson assumptions are reasonable depends on the process.
Process Capability
A stable process may still produce too much output outside specification limits.
Process capability compares the spread and centering of a stable process with the allowed specification range.
Capability analysis should normally be interpreted only after checking whether the process is sufficiently stable for the analysis to be meaningful.
Basic Capability Ratio
One common capability index is
\[ \boxed{ C_p = \frac{\operatorname{USL}-\operatorname{LSL}}{6\sigma} }. \]Here:
- USL is the upper specification limit;
- LSL is the lower specification limit;
- \(\sigma\) is the process standard deviation.
\(C_p\) compares allowed width with process spread but does not measure centering.
Centered Capability
A related index accounts for process centering:
\[ \boxed{ C_{pk} = \min\left( \frac{\operatorname{USL}-\mu}{3\sigma}, \frac{\mu-\operatorname{LSL}}{3\sigma} \right) }. \]If the process mean moves near one specification boundary, \(C_{pk}\) will reflect the reduced margin.
Inspection Does Not Create Quality
Inspection can detect problems, but it does not automatically improve the underlying process.
Long-term quality improvement usually requires identifying and reducing causes of variation, defect, or failure.
Measurement is a control tool, not a substitute for process design.
Common Error: Adding Series Reliabilities
If all required components must work, their reliabilities are multiplied under independence.
For
\[ R_1=0.9,\qquad R_2=0.8, \]series reliability is
\[ (0.9)(0.8)=\boxed{0.72}, \]not \(1.7\).
Common Error: Multiplying Parallel Reliabilities Directly
If either parallel component can keep the system working, multiply their failure probabilities and use the complement.
For independent reliabilities \(0.9\) and \(0.8\),
\[ R = 1-(0.1)(0.2) = \boxed{0.98}. \]Common Error: Ignoring Time in Reliability Statements
Saying a component is \(99\%\) reliable is incomplete unless the operating period and conditions are understood.
Reliability is generally a function of mission duration.
Common Error: Confusing Reliability and Availability
Reliability concerns failure-free operation over an interval.
Availability concerns whether the system is operational when needed, including the effect of repair.
A frequently failing system can still have high availability if repairs are extremely fast, although that may still be undesirable for other reasons.
Common Error: Treating Control Limits as Customer Specifications
Control limits are calculated from process behavior.
Specification limits come from design, customer, regulatory, or engineering requirements.
One should not be substituted for the other.
Common Error: Reacting to Every Random Fluctuation
If a stable process is adjusted after every ordinary fluctuation, the adjustment itself can increase variation.
Statistical control methods are intended to help distinguish routine noise from meaningful signals.
Common Error: Ignoring Sample Size
Sampling variability decreases as sample size increases.
For a mean,
\[ \operatorname{SE}(\bar X)=\frac{\sigma}{\sqrt n}. \]Therefore, control limits and uncertainty calculations must use the correct sample size.
Common Error: Confusing Defects with Defective Units
One unit may contain several defects.
A defect-count chart and a defective-proportion chart measure different things.
The selected model must match the actual quality measure.
A Reliable Procedure for Reliability Problems
- Define successful system operation.
- Define the mission time and operating conditions.
- Identify the required components.
- Determine whether the structure is series, parallel, or mixed.
- Identify component reliabilities or failure models.
- Check whether independence is reasonable.
- Calculate system reliability.
- Identify critical components or common-cause risks.
- Evaluate whether redundancy improves the design sufficiently.
A Reliable Procedure for Quality-Control Problems
- Define the quality characteristic.
- Decide whether it is quantitative, defective/not defective, or a defect count.
- Choose the sampling plan.
- Estimate the relevant process center and variability.
- Select an appropriate control chart or acceptance rule.
- Calculate the center line and limits correctly.
- Plot or evaluate observations over time.
- Investigate unusual points or patterns.
- Separate control limits from specification limits.
- Address root causes rather than merely screening output.
A Reliable Procedure for Acceptance Sampling
- Define lot size and quality criteria.
- Choose the sample size \(n\).
- Choose the acceptance number \(c\).
- Select the sample according to the sampling plan.
- Count defective units or defects as specified.
- Compare the result with the acceptance rule.
- Recognize producer and consumer risk.
- Document the decision and assumptions.
Example: Full Reliability Workflow
Suppose a system contains two required stages.
Stage 1 has reliability
\[ 0.97. \]Stage 2 contains two independent redundant components, each with reliability
\[ 0.90. \]Reliability of Stage 2 is
\[ 1-(0.10)(0.10)=0.99. \]Because both stages are required, total system reliability is
\[ (0.97)(0.99)=\boxed{0.9603}. \]Example: Full p-Chart Workflow
Suppose a stable process has estimated defective proportion
\[ \bar p=0.03 \]with sample size
\[ n=200. \]Standard error:
\[ \sqrt{\frac{0.03(0.97)}{200}} \approx0.01206. \]Upper control limit:
\[ 0.03+3(0.01206) \approx \boxed{0.0662}. \]Lower expression:
\[ 0.03-3(0.01206) \approx-0.0062. \]The practical lower limit is therefore
\[ \boxed{0}. \]Practice 1: Failure Probability
A component has reliability
\[ 0.94. \]Find its failure probability.
Practice 2: Series Reliability
Two independent required components have reliabilities
\[ 0.95 \]and
\[ 0.98. \]Find system reliability.
Practice 3: Parallel Reliability
Two independent redundant components each have reliability
\[ 0.85. \]Find system reliability if either component is sufficient.
Practice 4: MTTF
Under an exponential model,
\[ \lambda=0.004 \]failures per hour.
Find the mean time to failure.
Practice 5: Availability
A repairable system has
\[ \operatorname{MTBF}=500 \]hours and
\[ \operatorname{MTTR}=5 \]hours.
Find availability.
Practice 6: Defect Proportion
A sample of \(250\) units contains \(10\) defectives.
Find the sample defective proportion.
Practice 7: Acceptance Sampling
A plan uses
\[ n=40,\qquad c=1. \]The sample contains \(2\) defectives.
Is the lot accepted?
Practice 8: Binomial No-Defective Probability
If
\[ p=0.05 \]and \(n=5\), find the probability of no defective units.
Practice 9: Capability Width
A process has
\[ \operatorname{USL}=110, \qquad \operatorname{LSL}=90, \qquad \sigma=2. \]Find \(C_p\).
Practice 10: Control versus Specification
Which type of limit represents external acceptable-product requirements: control limits or specification limits?
Practice Answers
Practice 1
\[ 1-0.94=\boxed{0.06}. \]Practice 2
\[ (0.95)(0.98)=\boxed{0.931}. \]Practice 3
Both fail with probability
\[ (0.15)(0.15)=0.0225. \]Therefore,
\[ 1-0.0225=\boxed{0.9775}. \]Practice 4
\[ \operatorname{MTTF} = \frac{1}{0.004} = \boxed{250\text{ hours}}. \]Practice 5
\[ A = \frac{500}{500+5} = \frac{500}{505} \approx \boxed{0.9901}. \]Practice 6
\[ \hat p = \frac{10}{250} = \boxed{0.04}. \]Practice 7
No.
The plan accepts only when the number of defectives is at most \(1\).
Since
\[ 2>1, \]the lot is rejected.
Practice 8
\[ P(X=0) = (0.95)^5 \approx \boxed{0.7738}. \]Practice 9
\[ C_p = \frac{110-90}{6(2)} = \frac{20}{12} \approx \boxed{1.67}. \]Practice 10
External acceptable-product requirements are represented by
\[ \boxed{\text{specification limits}}. \]What to Remember
Reliability and quality control use probability and statistical monitoring to evaluate whether systems and processes perform acceptably.
- reliability is the probability of successful operation over a defined mission;
- series reliabilities multiply under independence;
- parallel redundancy uses the complement of joint failure;
- common-cause failure can reduce the benefit of redundancy;
- exponential reliability uses \(R(t)=e^{-\lambda t}\);
- availability incorporates repair as well as failure;
- defect proportion is \(\hat p=x/n\);
- acceptance sampling makes lot decisions from samples;
- control charts distinguish routine variation from unusual signals;
- control limits and specification limits are different;
- quality monitoring should lead to process understanding, not merely inspection.
The central habit is: define failure or unacceptable quality precisely, match the probability or statistical model to the process, and distinguish routine variation from evidence that the system has changed.
Where We Go Next
Reliability and quality control examine how systems perform under failure, variation, and defect. The next applied problem family shifts from system performance to strategic interaction, where outcomes depend not only on one decision maker's choice but also on the choices of others.
The next lesson, Games and Strategic Behavior, develops models of players, strategies, payoffs, dominance, best responses, equilibrium, and repeated interaction.