Choose your reading route
eduKateSG · Why Mathematics? · Moving Horizon Estimation for Constrained Condition Monitoring: Sliding Windows, Arrival Costs, Optimisation and Physical Bounds
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 5–8
Reproduce the example
Audit the numbers one step at a time.Route 3 · Chapters 9–12
Check assumptions
See where design and calibration matter.Route 4 · Chapters 13–16
Connect to condition monitoring
Connect the method to a condition-monitoring investigation.Route 5 · Chapters 17–20
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 a state estimate must obey physical limits. Moving horizon estimation, or MHE, solves an optimisation problem over a recent window of states, disturbances and measurements. It can impose bounds such as nonnegative degradation, rate limits or conservation constraints that a basic recursive Gaussian filter does not enforce directly. The window moves forward as new data arrive, while an arrival cost summarises information from earlier data.
For a two-state teaching candidate x₀=2 and x₁=3, take an arrival estimate 1.8 with variance weight 0.25, measurements 2.2 and 2.8 with variance weight 0.04, and model x₁=x₀+1 with variance weight 0.09. The weighted objective is (2−1.8)²/0.25 + [(2.2−2)²+(2.8−3)²]/0.04 + (3−2−1)²/0.09 = 0.16+2+0=2.16. The candidate satisfies x≥0. An optimiser compares feasible candidates; evaluating one objective is an audit step, not proof it is optimal.
Full chapter index
Chapters 1–4 · Understand the question
Chapters 5–8 · Reproduce the example
Chapters 9–12 · Check assumptions
Chapters 13–16 · Connect to condition monitoring
Chapters 17–20 · Study and verify
Place states, residuals and bounds inside one optimisation problem. Physical knowledge can rule out states that fit noisy data numerically. Wear depth cannot be negative. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Incorrect constraints can force biased estimates. Justify every bound.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make why constraints matter the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Did You Know? The arrival cost is the mathematical memory of data that have fallen outside the moving window; a poor arrival approximation can bias the entire current optimisation. A small observation like this often reveals the conceptual boundary that a formula alone hides.
The sliding window keeps recent evidence while the arrival cost carries older information. Estimate only the most recent N states and shift the window as each observation arrives. A ten-minute window supports online monitoring. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. A short window forgets slow information. Test horizon length.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make the moving window the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Every weight changes the trade-off between measurements and dynamics. States, disturbances, parameters or measurement errors may be optimised together. A degradation rate and latent condition are estimated. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Too many variables increase ambiguity and cost. Use identifiability checks.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make decision variables the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
A feasible numerical solution still needs physical validation. Summarise information before the window with a penalty on the first state. The teaching arrival centre is 1.8. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. A weak or biased arrival cost distorts the solution. Update it consistently.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make arrival cost the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Primary research source: https://doi.org/10.1109/TAC.2002.808470. Read its problem definition, mathematical criterion and assumptions before transferring the method to a reliability setting.
Place states, residuals and bounds inside one optimisation problem. Penalise differences between predicted and observed measurements using declared scaling. Two residuals of magnitude 0.2 contribute two units together. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Weights are model assumptions, not decorations. Calibrate measurement error.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make measurement residual cost the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Did You Know? The arrival cost is the mathematical memory of data that have fallen outside the moving window; a poor arrival approximation can bias the entire current optimisation. A small observation like this often reveals the conceptual boundary that a formula alone hides.
The sliding window keeps recent evidence while the arrival cost carries older information. Penalise departures from the state-transition model or constrain them exactly. Candidate states follow x one equals x zero plus one, so penalty is zero. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. An exact model constraint can be too rigid. Allow justified process uncertainty.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make dynamic residual cost the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Every weight changes the trade-off between measurements and dynamics. Square 0.2 and divide by 0.25. The result is 0.16. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Using standard deviation instead of variance changes scaling. Label denominators.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make worked arrival penalty the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
A feasible numerical solution still needs physical validation. Square both residuals and divide their sum by 0.04. The result is two. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Correlated errors require a matrix quadratic form. Use the full covariance.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make worked measurement penalty the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Place states, residuals and bounds inside one optimisation problem. Add arrival, measurement and dynamic penalties. The candidate score is 2.16. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. A low score is meaningful only relative to other feasible candidates and the model. Compare alternatives.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make worked total objective the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Did You Know? The arrival cost is the mathematical memory of data that have fallen outside the moving window; a poor arrival approximation can bias the entire current optimisation. A small observation like this often reveals the conceptual boundary that a formula alone hides.
| Term | Calculation | Value |
|---|---|---|
| Arrival | (2-1.8)^2 / 0.25 | 0.16 |
| Measurements | ((2.2-2)^2 + (2.8-3)^2) / 0.04 | 2.00 |
| Dynamics | (3-2-1)^2 / 0.09 | 0.00 |
| Total | 2.16 | |
| Constraint | x0 and x1 are nonnegative | Satisfied |
The sliding window keeps recent evidence while the arrival cost carries older information. Impose lower, upper or rate bounds directly in the optimisation. Both candidate states satisfy x≥0. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Active bounds can create non-Gaussian uncertainty. Report active constraints.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make inequality constraints the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Every weight changes the trade-off between measurements and dynamics. Enforce balances or exact relationships when physically warranted. A conservation equation links component states. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. False exactness can cause infeasibility. Allow tolerances where needed.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make equality constraints the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
A feasible numerical solution still needs physical validation. Replace quadratic penalties with robust losses to reduce sensitivity to outliers. A single sensor spike receives limited influence. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Robust loss choices change statistical interpretation. State thresholds and rationale.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make robust losses the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Supporting authoritative source: https://doi.org/10.1016/S0005-1098(01)00115-7. Compare the published scope with this teaching example, especially the information set, numerical representation and conditions under which the result is claimed.
Place states, residuals and bounds inside one optimisation problem. MHE naturally accommodates nonlinear dynamics and observations through nonlinear programming. Temperature-dependent wear is estimated inside the window. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Local minima and initialization matter. Use multi-start or diagnostics.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make nonlinear models the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Did You Know? The arrival cost is the mathematical memory of data that have fallen outside the moving window; a poor arrival approximation can bias the entire current optimisation. A small observation like this often reveals the conceptual boundary that a formula alone hides.
The sliding window keeps recent evidence while the arrival cost carries older information. The optimiser must return a feasible solution within the monitoring deadline. Warm starts use the shifted previous solution. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Timeouts can create stale estimates. Define fallback behaviour.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make solver feasibility the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Every weight changes the trade-off between measurements and dynamics. Longer windows use more evidence but cost more computation and rely on older model validity. A longer horizon resolves slow drift. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Bigger is not automatically better. Plot accuracy and runtime.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make horizon trade-off the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
A feasible numerical solution still needs physical validation. Curvature, ensembles, profiles or repeated simulations can approximate uncertainty around the constrained solution. A Hessian-based interval is compared with bootstrap results. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Bounds and nonlinearity can invalidate simple Gaussian intervals. Use suitable methods.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make uncertainty quantification the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Place states, residuals and bounds inside one optimisation problem. Track objective components, residuals, constraint activity, solver status and coverage. A rising dynamic penalty signals model mismatch. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. A converged solver does not guarantee a correct model. Inspect physical residuals.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make diagnostics the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Did You Know? The arrival cost is the mathematical memory of data that have fallen outside the moving window; a poor arrival approximation can bias the entire current optimisation. A small observation like this often reveals the conceptual boundary that a formula alone hides.
The sliding window keeps recent evidence while the arrival cost carries older information. Constrained estimates can prevent impossible health states and support threshold timing. A nonnegative degradation path feeds an alert rule. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. Constraints do not prove the failure mechanism. Combine with inspection evidence.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make reliability decisions the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Every weight changes the trade-off between measurements and dynamics. MHE connects algebra, graphs, inequalities, weighted squares and optimisation. A learner can verify the 2.16 objective by hand. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. The lowest score is conditional on the chosen model and weights. Explain assumptions.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make student and parent guidance the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Useful next reading: H-Infinity filtering, Kalman filtering foundations, Mathematics Learning Hub, eduKate Mathematics Learning System, approach to learning mathematics, Bukit Timah Mathematics curriculum overview, cross-subject learning approach. Follow the link that answers the next question in your own reasoning, rather than treating the list as a reading quota.
A feasible numerical solution still needs physical validation. Moving horizon estimation balances recent data, dynamics, old information and physical constraints in one auditable optimisation. Change one measurement and compare objective terms. The mathematics is useful only when the state, units, timing and information boundary stay attached to every number. The window is not the whole history, and the arrival cost is not optional memory. End with a feasibility check.
A careful audit has four moves. First write the claim in ordinary language. Next preserve the intermediate quantities instead of jumping to the final estimate. Then check dimensions, signs, matrix validity or probability totals. Finally compare the result with the physical story and an alternative model. Save window length, arrival cost, objective terms, constraints, scaling, solver version, tolerances, warm start, status, runtime and fallback action. This record lets another reader locate whether disagreement comes from data, assumptions, computation or interpretation.
For practice, make practice, misconceptions and next reading the focus and evaluate arrival penalty zero point one six, measurement penalty two and dynamic penalty zero for candidate states two and three, giving total objective two point one six while checking nonnegative bounds. Predict the direction before using a calculator. Change one defensible input, recompute the result and explain why it moved. Finish with one supported conclusion, one uncertainty and one next check. That routine develops mathematical problem-solving skills that transfer beyond this one algorithm.
Quick answers
Why is mathematics important here? It turns a verbal concern about uncertainty, later evidence, shared information or robustness into quantities that can be checked and challenged.
Does a successful calculation prove the physical model? No. It shows what follows from the declared model, inputs and criterion. Validation against independent evidence remains necessary.
Which school mathematics transfers? Ratios, inequalities, functions, weighted averages, variance, graphs and matrices all reappear, joined by careful language about time and assumptions.
How should a student practise? Reproduce the small example, keep intermediate values, predict the direction of change and explain one limitation without jargon.
What should a practitioner report? Equations, data boundaries, initialization, uncertainty settings, numerical method, diagnostics, sensitivity tests and the rule that connects estimates to decisions.
How can a parent help? Ask what each number represents, when the information became available and what assumption would most change the conclusion. Explanation is part of mathematical mastery.
Extended practice laboratory
Practice case 1: Why constraints matter
Setup: Use this chapter’s illustrative case: Wear depth cannot be negative.
Task: Justify every bound.
Checking target: Incorrect constraints can force biased estimates.
Practice case 2: The moving window
Setup: Use this chapter’s illustrative case: A ten-minute window supports online monitoring.
Task: Test horizon length.
Checking target: A short window forgets slow information.
Practice case 3: Decision variables
Setup: Use this chapter’s illustrative case: A degradation rate and latent condition are estimated.
Task: Use identifiability checks.
Checking target: Too many variables increase ambiguity and cost.
Practice case 4: Arrival cost
Setup: Use this chapter’s illustrative case: The teaching arrival centre is 1.8.
Task: Update it consistently.
Checking target: A weak or biased arrival cost distorts the solution.
Practice case 5: Measurement residual cost
Setup: Use this chapter’s illustrative case: Two residuals of magnitude 0.2 contribute two units together.
Task: Calibrate measurement error.
Checking target: Weights are model assumptions, not decorations.
Practice case 6: Dynamic residual cost
Setup: Use this chapter’s illustrative case: Candidate states follow x one equals x zero plus one, so penalty is zero.
Task: Allow justified process uncertainty.
Checking target: An exact model constraint can be too rigid.
Practice case 7: Worked arrival penalty
Setup: Use this chapter’s illustrative case: The result is 0.16.
Task: Label denominators.
Checking target: Using standard deviation instead of variance changes scaling.
Practice case 8: Worked measurement penalty
Setup: Use this chapter’s illustrative case: The result is two.
Task: Use the full covariance.
Checking target: Correlated errors require a matrix quadratic form.
Practice case 9: Worked total objective
Setup: Use this chapter’s illustrative case: The candidate score is 2.16.
Task: Compare alternatives.
Checking target: A low score is meaningful only relative to other feasible candidates and the model.
Practice case 10: Inequality constraints
Setup: Use this chapter’s illustrative case: Both candidate states satisfy x≥0.
Task: Report active constraints.
Checking target: Active bounds can create non-Gaussian uncertainty.