Why Mathematics? | Extended Kalman Filtering for Nonlinear Condition Monitoring: Jacobians, Linearisation, Prediction, Update and Covariance

Start with the nonlinear function and its local slope. A nonlinear function can turn a symmetric state distribution into a skewed observation distribution. A squared sensor response treats positive and negative state values identically. The calculation matters only when its probability statement, units and observation boundary remain attached to it. One mean and covariance do not preserve every shape. Identify the nonlinear step before choosing an approximation.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use why nonlinear filtering needs care as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Did You Know? The EKF applies the nonlinear functions to the mean but uses their Jacobians to move covariance, so two different mathematical operations happen in one filter cycle. The algorithm becomes easier to trust when this hidden structure is made visible.

Contents · Next chapter

At this step, calculus turns model curvature into a matrix used by the filter. The EKF separates nonlinear state transition f from nonlinear measurement h and attaches process and measurement noise. Wear evolves with load and a voltage sensor observes a curved response. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Writing only code hides units and assumptions. Write both stochastic equations explicitly.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use state and measurement equations as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

The update is easy to compute only after the linearisation point is explicit. The filter linearises around its current predicted or updated state. The local slope at condition two guides the next update. The calculation matters only when its probability statement, units and observation boundary remain attached to it. A slope far from the true state can point in the wrong direction. Inspect initialization and large innovations.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use nominal trajectory as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

A derivative carries local sensitivity, not global truth. Partial derivatives of f form the local transition matrix used to propagate covariance. A degradation rate changes with temperature. The calculation matters only when its probability statement, units and observation boundary remain attached to it. A Jacobian is local and depends on the evaluation point. Save derivative formulas and evaluation states.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use jacobian of the transition as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Primary authoritative source: https://ntrs.nasa.gov/citations/19860003843. Check the source's state definition, approximation and computational conditions before transferring the method to reliability work.

Contents · Previous chapter · Next chapter

Start with the nonlinear function and its local slope. Partial derivatives of h map state uncertainty into predicted sensor uncertainty. For h(x)=x squared, H equals 2x and is four at x two. The calculation matters only when its probability statement, units and observation boundary remain attached to it. At a flat derivative, the filter may see little local information. Check sensitivity across the operating range.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use jacobian of the measurement as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Did You Know? The EKF applies the nonlinear functions to the mean but uses their Jacobians to move covariance, so two different mathematical operations happen in one filter cycle. The algorithm becomes easier to trust when this hidden structure is made visible.

Contents · Previous chapter · Next chapter

At this step, calculus turns model curvature into a matrix used by the filter. The mean is propagated through the full nonlinear transition function. The predicted wear follows f of the previous mean. The calculation matters only when its probability statement, units and observation boundary remain attached to it. In general E of f(X) is not f of E(X). Treat the mean propagation as an approximation.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use nonlinear mean prediction as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

The update is easy to compute only after the linearisation point is explicit. The Jacobian sandwiches prior covariance and process noise is added. F P F transpose plus Q expands uncertainty. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Omitting Q creates overconfidence. Audit units and positive semidefiniteness.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use covariance prediction as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

A derivative carries local sensitivity, not global truth. The nonlinear sensor function predicts a reading and the observed-minus-predicted difference is the innovation. Five minus four gives one. The calculation matters only when its probability statement, units and observation boundary remain attached to it. A large innovation may mean noise, state error or model error. Do not diagnose from one residual.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use predicted measurement and innovation as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

Start with the nonlinear function and its local slope. H P H transpose plus R describes expected innovation spread under the linearisation. Four squared times one plus one gives seventeen. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Correlated sensor noise needs a full covariance matrix. Use the actual measurement model.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use innovation covariance as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Did You Know? The EKF applies the nonlinear functions to the mean but uses their Jacobians to move covariance, so two different mathematical operations happen in one filter cycle. The algorithm becomes easier to trust when this hidden structure is made visible.

Contents · Previous chapter · Next chapter

At this step, calculus turns model curvature into a matrix used by the filter. The cross covariance P H transpose divided by innovation covariance balances state and sensor uncertainty. Four divided by seventeen is about 0.2353. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Gain is not an arbitrary learning rate. Derive it from the declared matrices.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use worked kalman gain as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

The update is easy to compute only after the linearisation point is explicit. Gain times innovation shifts the predicted mean. Two plus four seventeenths gives about 2.2353. The calculation matters only when its probability statement, units and observation boundary remain attached to it. A sign error reverses the correction. Predict the direction before calculating.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use worked state update as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

A derivative carries local sensitivity, not global truth. The measurement reduces covariance according to the local information supplied. One minus sixteen seventeenths gives one seventeenth. The calculation matters only when its probability statement, units and observation boundary remain attached to it. The simplified formula can lose symmetry numerically. Use the Joseph or square-root form when needed.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use worked covariance update as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Second authoritative source: https://doi.org/10.1016/B978-1-4831-6716-9.50011-4. Compare its notation and scope with the teaching example, especially where finite precision or finite ensembles matter.

Contents · Previous chapter · Next chapter

Start with the nonlinear function and its local slope. Taylor terms beyond the Jacobian are discarded. A sharply curved sensor makes the local line inaccurate. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Small covariance does not guarantee small bias after a bad linearisation. Compare with simulation, UKF or particle filtering.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use first-order approximation error as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Did You Know? The EKF applies the nonlinear functions to the mean but uses their Jacobians to move covariance, so two different mathematical operations happen in one filter cycle. The algorithm becomes easier to trust when this hidden structure is made visible.

Contents · Previous chapter · Next chapter

At this step, calculus turns model curvature into a matrix used by the filter. Re-linearising the measurement around successive updated states can improve a difficult measurement update. Several local steps fit the same observation. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Iteration is not new independent evidence. Do not count the reading multiple times.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use iterated ekf as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

The update is easy to compute only after the linearisation point is explicit. Positive or bounded states may require transformed coordinates or constrained updates. Log wear remains positive after back transformation. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Back-transforming a mean changes interpretation. Propagate uncertainty through the transform.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use constraints and transformations as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

A derivative carries local sensitivity, not global truth. Prediction continues through gaps using elapsed-time-specific dynamics and Q. A three-hour outage grows more uncertainty than one interval. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Missingness may depend on condition. Record timestamps and dropout causes.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use missing and irregular readings as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

Start with the nonlinear function and its local slope. Standardised innovations test local consistency, bias and temporal dependence. Residuals become positive during high load. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Good averages can hide regime-specific failure. Check by time, sensor and operating regime.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use innovation diagnostics as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Did You Know? The EKF applies the nonlinear functions to the mean but uses their Jacobians to move covariance, so two different mathematical operations happen in one filter cycle. The algorithm becomes easier to trust when this hidden structure is made visible.

Contents · Previous chapter · Next chapter

At this step, calculus turns model curvature into a matrix used by the filter. EKF uses derivatives, UKF uses deterministic sigma points and particle filters use weighted samples. The same nonlinear sensor can be processed three ways. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Complexity alone does not decide accuracy. Benchmark under realistic simulations and later data.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use comparison with ukf and particles as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Contents · Previous chapter · Next chapter

The update is easy to compute only after the linearisation point is explicit. Posterior state and covariance support threshold probabilities and expected-loss decisions. A high probability of excessive wear prompts inspection. The calculation matters only when its probability statement, units and observation boundary remain attached to it. An EKF estimate is not a root-cause diagnosis. Combine estimates with costs and engineering review.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use maintenance interpretation as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Useful next reading: Kalman filtering foundations, unscented Kalman filtering, Mathematics Learning Hub, eduKate Mathematics Learning System, approach to learning mathematics, Bukit Timah Mathematics curriculum overview, cross-subject learning approach. Choose the route that answers the next question rather than treating the links as a checklist.

Contents · Previous chapter · Next chapter

A derivative carries local sensitivity, not global truth. EKF study connects calculus, matrices, variance, functions and feedback. Students can reproduce the h(x)=x squared update by hand. The calculation matters only when its probability statement, units and observation boundary remain attached to it. Extended does not mean exact for nonlinear or non-Gaussian systems. Explain the linearisation point, evidence and limitations.

A dependable audit follows four moves. Write the conditioning information in words, preserve the intermediate quantities, check dimensions and probability totals, then compare the direction of the result with the physical story. Save nonlinear equations, Jacobians, evaluation points, Q, R, initial conditions and innovation diagnostics. Reproduction is useful because it reveals whether a surprising conclusion came from data, approximation, code or interpretation.

For practice, use practice, misconceptions and next reading as the focus and differentiate h of x equals x squared at x equals 2, calculate innovation variance 17, gain four seventeenths, updated mean 2.2353 and variance one seventeenth. Before calculating, predict which term should rise or fall. Change one defensible input and explain the response without jargon. End with one supported statement and one limit. This turns mathematics education into transferable reasoning rather than formula recall.

Quick answers

Why is mathematics important here? It exposes how uncertainty moves through a model, which evidence changes the estimate, and where approximation enters.

Does a successful update prove the physical model? No. It shows what follows from the declared model and data; external validation remains necessary.

Which mathematics transfers from school? Functions, derivatives, weighted averages, probability, variance, graphs and matrices all reappear.

How should a student practise? Reproduce the small example, keep intermediate values, predict directions and explain one limitation.

What should a practitioner report? State equations, data boundaries, initialization, uncertainty settings, numerical method, diagnostics, sensitivity checks and decision rules.

How can a parent help? Ask the learner what each symbol represents and what assumption would change the conclusion. Clear explanation is part of mathematical mastery.

Contents · Previous chapter · Mathematics Learning Hub