Choose your reading route
eduKateSG · Why Mathematics? · Dynamic Principal Component Analysis for Reliability Monitoring: Lagged Variables, Autocorrelation, T-Squared, SPE and Fault Isolation
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 the relationship between present and past carries more fault information than the current measurement alone. Dynamic principal component analysis (DPCA) augments each observation with lagged variables, then applies PCA to the expanded vectors. It can model simple autocorrelation and detect departures from learned temporal relationships.
For the scalar sequence 1,2,3,4 with one lag, embedded rows are [2,1], [3,2] and [4,3]. Their mean is [3,2], giving centred rows [−1,−1], [0,0], [1,1] and covariance [[1,1],[1,1]], with eigenvalues 2 and 0. One component captures the normal lag relation. A new centred pair [1,−1] lies perpendicular to it, giving residual squared length 2 even though each coordinate alone is modest. This is an idealised example; real lag order and thresholds require validation.
Did You Know? Adding lags turns time dependence into geometry: a temporal rule becomes a direction or subspace in an expanded data table.
This article answers one intent: The mechanism embeds current and past measurements in one vector so PCA can model lag relationships and detect temporal departures with score and residual statistics.
Dynamic PCA turns temporal dependence into geometry by embedding current and lagged measurements in one vector. The worked lag-one example shows how an ordinary trend can lie in the retained dynamic direction while a reversed relation appears as residual energy. Later chapters add lag choice, alignment, missing data and online deployment. Students transfer sequences, covariance and eigenvectors; practitioners transfer auditable treatment of autocorrelation before interpreting alarms.
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. Static PCA models same-time correlation but can miss timing relationships. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A sensor response normally follows another one sample later. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Ignoring lag creates patterned residuals. Include justified history. 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: Static PCA models same-time correlation but can miss timing relationships. 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. Form rows containing current and selected previous samples. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
One-lag rows are [2,1],[3,2],[4,3]. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Embedding shortens usable data and creates missing starts. Record alignment. 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: One-lag rows are [2,1],[3,2],[4,3]. 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. Subtract a mean for each current and lagged variable. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Column mean [3,2] yields symmetric centred rows. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Using future data for online centering leaks information. Freeze baseline transforms. 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: Using future data for online centering leaks information. 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. Compute covariance in expanded space. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
[[1,1],[1,1]] captures perfect lag relation. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Overlapping rows are not independent. Calibrate empirically. 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: Calibrate empirically. 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. PCA finds dominant temporal directions. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Eigenvalues 2 and 0 show one exact pattern. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Real systems rarely have zero residual variance. Retain noise-aware components. 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: PCA finds dominant temporal directions. 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. Project centred [1,−1] onto the learned [1,1] direction. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Its retained score is zero and residual squared is 2. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A large residual may come from timing misalignment. Check clocks. 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: Its retained score is zero and residual squared is 2. 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.
The mathematical value appears when a pattern is connected to one reproducible calculation. Monitor retained lag-space scores weighted by eigenvalues. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A strong shift along the learned trajectory raises T². Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Score variance changes across regimes. Use conditional baselines. 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: Score variance changes across regimes. 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.
| Object | Value |
|---|---|
| Embedded rows | [2,1],[3,2],[4,3] |
| Column mean | [3,2] |
| Centred rows | [−1,−1],[0,0],[1,1] |
| Covariance | [[1,1],[1,1]] |
| Eigenvalues | 2,0 |
| New centred pair | [1,−1] |
| Residual squared | 2 |
Separate the mechanism from the interpretation so a number does not become unsupported certainty. Monitor distance from the retained lagged subspace. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
The new pair raises SPE despite a small retained score. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
SPE does not state which lag broke. Inspect contributions. 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: Inspect contributions. Explain the result aloud, including the assumption that would most change the answer.
Keep timing, units and assumptions visible before applying a threshold. Use process knowledge, autocorrelation and validation to select lags. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A response delay of two samples suggests at least two lags. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Too many lags inflate dimension. Compare out-of-sample detection. 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: Use process knowledge, autocorrelation and validation to select lags. 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. Lag meaning depends on time between samples. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
One lag at 1 second differs from one lag at 1 minute. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Irregular sampling breaks fixed-lag interpretation. Use timestamps or resampling. 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: One lag at 1 second differs from one lag at 1 minute. 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. DPCA can absorb linear serial relationships into retained components. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A ramp sequence follows a stable lag relation. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Not all dynamics are linear or stationary. Check residual whiteness. 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: Not all dynamics are linear or stationary. 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. Different faults alter current and lagged columns in different patterns. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A stuck sensor produces repeated equal values. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Contribution smearing across lags complicates isolation. Map back to time. 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: Map back to time. 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. Decompose T² and SPE across variables and lags. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Current sensor one and its previous value both contribute. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Overlapping windows repeat one physical event. Avoid double counting. 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: Decompose T² and SPE across variables and lags. 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. The first lag-order samples cannot form complete rows. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A one-lag model starts at sample two. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Padding invents history. Mark warm-up explicitly. 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 one-lag model starts at sample two. 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. A gap contaminates several subsequent embedded rows. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
One missing point affects current and lag copies. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Mean filling creates artificial temporal consistency. Use a stated missing-data method. 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: Mean filling creates artificial temporal consistency. 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. p variables with L lags create p(L+1) columns. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Twenty sensors and five lags create 120 features. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Covariance becomes unstable. Use parsimonious lags and regularisation. 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: Use parsimonious lags and regularisation. 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. Different operating modes have different temporal subspaces. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Start-up differs from steady operation. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
One DPCA model can confuse normal transitions with faults. Use mode-aware models. 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: Different operating modes have different temporal subspaces. 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. Maintain lag buffers, apply frozen scaling and project each complete row. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A ring buffer supplies the newest vector. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Software indexing errors shift the model. Test alignment fixtures. 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 ring buffer supplies the newest 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.
Keep timing, units and assumptions visible before applying a threshold. A dynamic SPE signal can trigger timing, sensor and process checks. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A broken lag relation prompts communication-latency review. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
It does not prove physical failure. Confirm with independent data. 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: It does not prove physical failure. 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 connects sequences, tables, covariance and geometry. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A learner can build the three embedded rows. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
More columns do not mean more independent information. Explain overlap. 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 overlap. 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 sampling interval, lags, scaling, components, limits, warm-up, 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 delay, stuck sensor, step and regime transition. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Unlogged alignment prevents audit. Archive buffers and transforms. 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 sampling interval, lags, scaling, components, limits, warm-up, 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. Static PCA uses contemporaneous vectors; DPCA adds lags to represent simple dynamics. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
DPCA can whiten residuals better. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Expanded models cost data and interpretation. Use the simplest adequate model. 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: DPCA can whiten residuals better. 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. DPCA is not a full state-space model merely because it includes lags. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Add a second lag and count columns. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Lag embedding approximates dynamics in a chosen window. State that boundary. 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: Lag embedding approximates dynamics in a chosen window. 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. Ku, Storer and Georgakis developed DPCA disturbance monitoring; PCA residual research supplies the score and Q foundations. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Recent eduKate articles cover PCA and innovation whiteness. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Lag selection remains application-specific. Validate temporal faults. 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 temporal faults. Explain the result aloud, including the assumption that would most change the answer.
The nearest published companions are multivariate EWMA monitoring, Hotelling T-squared monitoring, parity-space sensor fault detection and unknown-input observers. 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/10.1016/0169-7439(95)00076-3, https://doi.org/10.1080/00401706.1979.10489779, https://doi.org/10.1016/0169-7439(95)00036-9.
Frequently asked questions
What makes DPCA dynamic?
It applies PCA to vectors containing current and lagged measurements, turning time relationships into geometry.
How should lag order be chosen?
Use process timing, autocorrelation and out-of-sample monitoring performance.
Why do missing values affect several rows?
Each observation appears as the current value and as lagged copies in later embedded rows.
Is DPCA a state-space model?
No. Lag embedding captures selected temporal patterns but does not automatically provide a full physical state model.
Can DPCA isolate a fault?
It can support isolation through variable-and-lag contributions or model banks, but independent confirmation remains necessary.
Extended practice laboratory
Practice case 1: Embed sequence
Setup: Use 1,2,3,4 with one lag.
Checking target: Build three rows.
Practice case 2: Center columns
Setup: Subtract [3,2].
Checking target: Get [−1,−1],[0,0],[1,1].
Practice case 3: Covariance
Setup: Use sample covariance.
Checking target: Confirm [[1,1],[1,1]].
Practice case 4: Eigenvalues
Setup: Diagonalise covariance.
Checking target: Confirm 2 and 0.
Practice case 5: New pair
Setup: Use [1,−1].
Checking target: Confirm residual squared 2.
Practice case 6: Second lag
Setup: Add x(t−2).
Checking target: Count usable rows.
Practice case 7: Stuck sensor
Setup: Repeat one value.
Checking target: Inspect lag residual.
Practice case 8: Timing shift
Setup: Delay a channel one sample.
Checking target: Inspect contributions.
Practice case 9: Missing point
Setup: Remove x3.
Checking target: Track affected rows.
Practice case 10: Decision memo
Setup: Report temporal inconsistency.
Checking target: Recommend clock and sensor checks.