Choose your reading route
eduKateSG · Why Mathematics? · Multivariate EWMA for Sensor-Network Reliability: Vector Memory, Covariance Weighting, Mean Shifts and Run-Length Design
Start with the question, audit the worked example, or jump to the assumptions and practice. Every chapter stays open for uninterrupted reading.
Understand the question
Define the quantity and follow the mechanism.Route 2 · Chapters 6–10
Reproduce the example
Audit the numbers one step at a time.Route 3 · Chapters 11–15
Check assumptions
See where design and calibration matter.Route 4 · Chapters 16–20
Explore applications
Connect the method to a concrete investigation.Route 5 · Chapters 21–24
Study and verify
Practise, inspect failure cases and read further.
Full chapter index · Practice and checking · Mathematics Learning Hub
Why mathematics is important becomes clear when several sensors change together and their correlations matter. A multivariate exponentially weighted moving average (MEWMA) extends EWMA from one scalar to a vector. It preserves decaying memory in every channel, then measures the smoothed vector with a covariance-aware quadratic form.
Let λ=0.2, z0=[0,0], y1=[1,2] and y2=[0.5,1]. The update zt=λyt+(1−λ)z(t−1) gives z1=[0.2,0.4] and z2=[0.26,0.52]. If baseline sensor covariance is the identity, the long-run MEWMA covariance is λ/(2−λ)I=1/9 I. The quadratic statistic at step two is z2' (1/9 I)^−1 z2=9(0.26²+0.52²)=3.042. This calculation is illustrative: start-up covariance, estimated correlations and chosen limits must be handled explicitly in a real chart.
Did You Know? MEWMA can detect a coordinated shift spread across channels even when no single sensor looks extreme on its own.
This article answers one intent: The mechanism combines vector exponential memory with inverse-covariance distance so correlated sensor evidence is accumulated without double counting.
MEWMA joins exponential memory with multivariate geometry. The worked two-channel update makes the vector recurrence and covariance-weighted statistic visible. Later chapters add start-up limits, run length, missing channels, high dimension and diagnosis. Students transfer percentages, vectors and matrices; practitioners transfer joint calibration and transparent sensor preprocessing.
Full chapter index
Chapters 1–5 · Understand the question
Chapters 6–10 · Reproduce the example
Chapters 11–15 · Check assumptions
Chapters 16–20 · Explore applications
Chapters 21–24 · Study and verify
Begin with the object being measured, because a detector name is less useful than the statistic it changes. Sensor networks produce related channels whose small joint shifts can matter. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Temperature and vibration drift together. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Separate charts ignore cross-channel structure. Model the vector. Transfer the idea by changing one input at a time and explaining the direction before computing.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: Sensor networks produce related channels whose small joint shifts can matter. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
The mathematical value appears when a pattern is connected to one reproducible calculation. Update zt=Λyt+(I−Λ)z(t−1), often with Λ=λI. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
λ=.2 gives one-fifth weight to the new vector. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Different channel dynamics may need unequal weights. State the matrix. Keep raw evidence beside the transformed statistic so diagnosis remains possible.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: λ=.2 gives one-fifth weight to the new vector. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. Each previous vector receives a geometrically decaying matrix weight. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
y1 still contributes to z2. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Changing sensor availability breaks a simple fixed recursion. Define missing-channel handling. Compare with a simple baseline and an alternative cause before deciding.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: Changing sensor availability breaks a simple fixed recursion. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Keep timing, units and assumptions visible before applying a threshold. Set z0 to the trusted baseline mean, often zero after standardisation. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
The example starts at [0,0]. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A contaminated baseline biases every later update. Complete Phase I review. Explain the key equation aloud and point to the term carrying each meaning.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Complete Phase I review. Explain the result aloud, including the assumption that would most change the answer.
Make the example auditable by predicting direction, calculating and checking a nearby case. Apply the recurrence component by component. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
z1=[.2,.4] and z2=[.26,.52]. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Rounding can distort small quadratic statistics. Store full precision. Report observation, assumption, calculation and authorised action as separate layers.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: Apply the recurrence component by component. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
Begin with the object being measured, because a detector name is less useful than the statistic it changes. Estimate covariance from stable multichannel data. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Identity covariance means uncorrelated unit-variance channels. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Small samples make the inverse unstable. Use enough data or regularisation. Transfer the idea by changing one input at a time and explaining the direction before computing.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: Identity covariance means uncorrelated unit-variance channels. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
| Step | Observation vector | MEWMA vector |
|---|---|---|
| 0 | — | [0,0] |
| 1 | [1,2] | [0.20,0.40] |
| 2 | [0.5,1] | [0.26,0.52] |
| Long-run statistic | Σ=I | 3.042 |
The mathematical value appears when a pattern is connected to one reproducible calculation. For scalar λ and independent vectors, steady-state covariance is λ/(2−λ)Σ. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
With λ=.2 and Σ=I it is I/9. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Autocorrelation changes the formula. Model time dependence. Keep raw evidence beside the transformed statistic so diagnosis remains possible.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: Autocorrelation changes the formula. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. Compute zt prime Σz(t)^−1 zt to measure joint distance. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
The example gives 3.042. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A singular covariance cannot be inverted directly. Use stable factorisation or dimension reduction. Compare with a simple baseline and an alternative cause before deciding.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Use stable factorisation or dimension reduction. Explain the result aloud, including the assumption that would most change the answer.
Keep timing, units and assumptions visible before applying a threshold. Inverse covariance discounts directions that naturally vary together. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Two positively correlated sensor rises count less than if independent. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Wrong covariance can hide or exaggerate evidence. Check calibration. Explain the key equation aloud and point to the term carrying each meaning.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: Inverse covariance discounts directions that naturally vary together. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
Make the example auditable by predicting direction, calculating and checking a nearby case. Finite-time covariance grows toward the steady state. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Early chart limits differ from long-run limits. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Using only asymptotic values can misstate early alarms. Implement the exact start-up formula. Report observation, assumption, calculation and authorised action as separate layers.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: Early chart limits differ from long-run limits. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Begin with the object being measured, because a detector name is less useful than the statistic it changes. Compare the quadratic statistic with a design limit chosen for run-length behaviour. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A calibrated boundary defines the signal. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A chi-square shortcut may not match estimated parameters or start-up. Use theory and simulation. Transfer the idea by changing one input at a time and explaining the direction before computing.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: A chi-square shortcut may not match estimated parameters or start-up. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
The mathematical value appears when a pattern is connected to one reproducible calculation. Evaluate expected samples to signal under baseline and shifted mean vectors. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Small coordinated shifts may take longer than large ones. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Average run length hides tails. Report quantiles. Keep raw evidence beside the transformed statistic so diagnosis remains possible.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Report quantiles. Explain the result aloud, including the assumption that would most change the answer.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. The scalar chart statistic shows distance but not directly which direction changed. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Contribution plots inspect channel patterns. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Naive contributions can be unstable under correlation. Confirm with raw vectors. Compare with a simple baseline and an alternative cause before deciding.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: The scalar chart statistic shows distance but not directly which direction changed. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
Keep timing, units and assumptions visible before applying a threshold. A large earlier excursion can make the smoothed vector slow to respond in the opposite direction. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A process crosses then reverses. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
MEWMA alone may delay a new signal. Pair with pointwise checks. Explain the key equation aloud and point to the term carrying each meaning.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: A process crosses then reverses. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Make the example auditable by predicting direction, calculating and checking a nearby case. Update only observed subspaces or use a principled model-based treatment. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
One channel drops out for three samples. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Filling with zero creates false evidence. Track availability masks. Report observation, assumption, calculation and authorised action as separate layers.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: Filling with zero creates false evidence. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Begin with the object being measured, because a detector name is less useful than the statistic it changes. Standardise or use covariance so volts, degrees and acceleration remain comparable. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Covariance scaling makes units coherent in the quadratic form. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Poor scaling causes numerical instability. Audit units. Transfer the idea by changing one input at a time and explaining the direction before computing.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Audit units. Explain the result aloud, including the assumption that would most change the answer.
The mathematical value appears when a pattern is connected to one reproducible calculation. When sensors outnumber baseline samples, covariance estimation needs structure. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Shrinkage or sparse models stabilise the inverse. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Regularisation changes the effective chart. Validate it end to end. Keep raw evidence beside the transformed statistic so diagnosis remains possible.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: When sensors outnumber baseline samples, covariance estimation needs structure. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. Dynamic residuals may retain temporal structure. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A shared cycle raises MEWMA repeatedly. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Independent-vector limits become invalid. Whiten or simulate the dynamics. Compare with a simple baseline and an alternative cause before deciding.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: A shared cycle raises MEWMA repeatedly. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Keep timing, units and assumptions visible before applying a threshold. A signal can trigger localisation, inspection or safe-mode review. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Joint temperature-vibration drift prioritises a bearing check. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
The chart does not prove physical cause. Seek confirming evidence. Explain the key equation aloud and point to the term carrying each meaning.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: The chart does not prove physical cause. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Make the example auditable by predicting direction, calculating and checking a nearby case. The topic combines percentages, vectors, matrices and covariance. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A learner can reproduce [0.26,0.52] and 3.042. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A bigger vector is not always more surprising. Explain correlation. Report observation, assumption, calculation and authorised action as separate layers.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Explain correlation. Explain the result aloud, including the assumption that would most change the answer.
Begin with the object being measured, because a detector name is less useful than the statistic it changes. Specify vector, baseline, λ, covariance estimator, start-up limits, missing data and actions. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Tests include common shifts, opposite shifts and single-channel faults. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Unlogged preprocessing prevents audit. Version every setting. Transfer the idea by changing one input at a time and explaining the direction before computing.
A useful audit begins by writing the claim in plain language, then identifying the observation, reference model and comparison rule. In this chapter, the central claim is: Specify vector, baseline, λ, covariance estimator, start-up limits, missing data and actions. Test it first on a short sequence where every intermediate value can be inspected. Then change one condition and explain why the conclusion does or does not survive.
The mathematical value appears when a pattern is connected to one reproducible calculation. Scalar EWMA tracks one series; MEWMA tracks joint direction and covariance. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Several weak channels can combine into evidence. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A multivariate chart is harder to interpret. Keep channel-level diagnostics. Keep raw evidence beside the transformed statistic so diagnosis remains possible.
For transfer, sketch a before-and-after case and label what is measured, what is predicted and what is inferred. The concrete case here is: Several weak channels can combine into evidence. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. MEWMA is not the average of separate EWMA alarm decisions. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Change Σ to include correlation and recompute. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Correlation changes distance, not just the picture. Show the matrix multiplication. Compare with a simple baseline and an alternative cause before deciding.
Reliability work needs a counterexample as well as a successful example. Here the important boundary is: Correlation changes distance, not just the picture. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Keep timing, units and assumptions visible before applying a threshold. Lowry and colleagues introduced the MEWMA chart and studied run length; NIST documents covariance-aware multivariate distance. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Recent eduKate articles cover EWMA and Hotelling monitoring. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Published limits do not replace local calibration. Validate with representative streams. Explain the key equation aloud and point to the term carrying each meaning.
A student can practise this idea without specialised equipment. Start with a tiny table, reproduce the stated operation, and predict how the output changes when one entry is doubled, delayed or removed. The practical next step is: Validate with representative streams. Explain the result aloud, including the assumption that would most change the answer.
The nearest published companions are sequential probability ratio tests, EWMA monitoring, Bayesian model averaging and observability and identifiability. The eduKateSG Mathematics Learning Hub, eduKateSingapore Mathematics World, eduKate Sengkang Mathematics Hub, Bukit Timah Tutor Mathematics Hub and eduKate Punggol Mathematics Pathway provide wider learning routes. Primary sources: https://doi.org/10.1080/00401706.1992.10485232, https://stat.cmu.edu/technometrics/90-00/vol-34-01/v3401046.pdf, https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc341.htm.
Frequently asked questions
What does MEWMA monitor?
It monitors a smoothed vector of measurements or residuals and evaluates it with a covariance-aware distance.
Why not use separate EWMA charts?
Separate charts ignore correlation and can miss weak coordinated shifts or double count shared variation.
What does lambda control?
Lambda controls how strongly the newest vector updates the remembered vector.
Why is covariance estimation important?
The inverse covariance determines which joint directions are common and which are surprising.
Does a signal identify the faulty sensor?
No. It indicates a joint anomaly; channel-level diagnostics and independent evidence are still needed.
Extended practice laboratory
Practice case 1: Vector update
Setup: Use λ=.2 and two vectors.
Checking target: Confirm [.2,.4] then [.26,.52].
Practice case 2: Long-run covariance
Setup: Use Σ=I.
Checking target: Confirm I/9.
Practice case 3: Quadratic statistic
Setup: Use z2 and I/9.
Checking target: Confirm 3.042.
Practice case 4: Add correlation
Setup: Set off-diagonal covariance .7.
Checking target: Recompute distance.
Practice case 5: Change lambda
Setup: Use λ=.5.
Checking target: Compare memory.
Practice case 6: Common shift
Setup: Raise both channels slightly.
Checking target: Measure detection delay.
Practice case 7: Opposite shift
Setup: Raise one and lower one.
Checking target: Inspect direction.
Practice case 8: Missing channel
Setup: Remove channel two at step two.
Checking target: Compare valid treatments.
Practice case 9: Regularise covariance
Setup: Use a small ridge.
Checking target: Check stability.
Practice case 10: Decision memo
Setup: Report joint evidence without naming a fault.
Checking target: Recommend confirmation.