Choose your reading route
eduKateSG · Why Mathematics? · Multiple-Model Adaptive Estimation for Sensor-Fault Isolation: Parallel Filters, Mode Probabilities, Likelihoods and Recovery
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 practical when a monitoring system has several credible explanations for its data: healthy operation, a biased sensor, a failed actuator or another regime. Multiple-model adaptive estimation (MMAE) runs a bank of filters matched to different hypotheses. Each innovation likelihood updates the probability of its model, and outputs can be combined by probability-weighted averaging or used for fault isolation.
Suppose healthy and biased-sensor models have prior probabilities 0.8 and 0.2. Their current innovation likelihoods are 0.1 and 0.6. Unnormalised weights are 0.08 and 0.12, so posterior mode probabilities are 0.4 and 0.6. If their state estimates are 10 and 13, the fused estimate is 0.4(10)+0.6(13)=11.8. The calculation shows Bayesian model weighting; it does not prove the second model contains the true fault.
Did You Know? A model bank can isolate only hypotheses that are distinguishable in the available measurements. More filters do not create information that the sensor layout does not contain.
This article answers one intent: The mechanism updates competing model probabilities from innovation likelihoods and uses them to isolate faults or fuse state estimates.
MMAE treats diagnosis as competition among explicit models. Each filter predicts a likelihood, Bayes’ rule updates probabilities, and the system may isolate a mode or fuse estimates. The simple two-model calculation makes normalisation and weighted averaging transparent. Later chapters address collapse, recovery, identifiability, common causes and unknown faults. Students can transfer conditional probability and weighted averages; practitioners can transfer auditable model-bank design. The leading model remains one hypothesis in a finite set.
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. One filter assumes one structure, while real systems can occupy several fault or operating hypotheses. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A bank includes healthy and biased-sensor cases. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Unlisted faults can force the wrong winner. Design the hypothesis set carefully. 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: One filter assumes one structure, while real systems can occupy several fault or operating hypotheses. 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. Run one state estimator per model using the same observations. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Each filter predicts a different innovation distribution. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Large banks consume computation and maintenance effort. Prioritise identifiable faults. 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: Each filter predicts a different innovation distribution. 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. Priors express belief before the current measurement. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
The example begins at 0.8 healthy and 0.2 biased. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Strong priors can delay recognition. Document and test them. 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: Strong priors can delay recognition. 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. Each filter evaluates how probable the new innovation is under its model. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Likelihoods 0.1 and 0.6 favour the bias model. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Poor covariance can dominate likelihood unfairly. Check calibration. 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: Check calibration. 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. Multiply prior probability by likelihood and normalise across models. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
0.08 and 0.12 normalise to 0.4 and 0.6. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Underflow occurs with long likelihood products. Use log probabilities. 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: Multiply prior probability by likelihood and normalise across models. 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. Average conditional estimates with posterior model probabilities when fusion is appropriate. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
0.4(10)+0.6(13)=11.8. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A fused state can sit between physically distinct modes. Keep mode-conditional outputs. 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: 0.4(10)+0.6(13)=11.8. 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.
| Quantity | Healthy | Bias model |
|---|---|---|
| Prior | 0.8 | 0.2 |
| Likelihood | 0.1 | 0.6 |
| Unnormalised | 0.08 | 0.12 |
| Posterior | 0.4 | 0.6 |
| Conditional state | 10 | 13 |
The mathematical value appears when a pattern is connected to one reproducible calculation. A high probability for a fault-matched model supports isolation. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
The biased-sensor model becomes most probable. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Highest probability is not certainty. Use thresholds and persistence. 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: Highest probability is not certainty. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. Tiny likelihoods can eliminate a model permanently. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A numerical floor preserves recovery possibility. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Too large a floor weakens evidence. Calibrate and report it. 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: Calibrate and report it. Explain the result aloud, including the assumption that would most change the answer.
Keep timing, units and assumptions visible before applying a threshold. Repeated evidence can drive one probability near one. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A long healthy period suppresses fault models. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Sudden faults may then recover slowly. Use transition or reinitialisation logic. 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: Repeated evidence can drive one probability near one. 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. The bank must recognise return to healthy operation. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A repaired sensor produces healthy innovations again. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Sticky probabilities can retain the fault label. Define recovery tests. 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: A repaired sensor produces healthy innovations again. 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. Classical MMAE may treat an unknown fixed model; switching formulations include transition probabilities. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Faults can appear and disappear. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Transition choices encode persistence assumptions. State them explicitly. 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: Transition choices encode persistence assumptions. 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. Interacting multiple-model filtering mixes model-conditioned states before each update. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
IMM is useful for switching dynamics. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
MMAE and IMM are not interchangeable labels. Describe the mixing step. 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: Describe the mixing step. 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. Several filters can represent different bias magnitudes. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A grid includes 0,1 and 2-unit bias. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A coarse grid creates approximation error and a large grid increases cost. Use adaptive estimation when needed. 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: Several filters can represent different bias magnitudes. 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. Models can encode specific sensor or actuator patterns. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A bank separates one sensor bias from one actuator loss. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Some simultaneous faults are structurally unidentifiable. Perform identifiability analysis. 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 bank separates one sensor bias from one actuator loss. 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. Distinct faults must produce distinguishable output distributions. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
NASA work shows conditions and impossibilities for bias-type faults. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Probability cannot resolve identical signatures. Improve sensing or merge hypotheses. 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: Probability cannot resolve identical signatures. 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. Several sensors may share a reference or power supply. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A common bias model competes with independent faults. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Assuming independence can mis-isolate the event. Include shared-cause structure. 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: Include shared-cause structure. 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. A residual model or catch-all hypothesis can flag evidence outside the bank. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
All named models receive low likelihood. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A vague catch-all may absorb normal mismatch. Escalate for investigation. 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: A residual model or catch-all hypothesis can flag evidence outside the bank. 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. Mode probabilities can support targeted inspection and graceful reconfiguration. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
The likely failed sensor is removed only after confirmation. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Automatic compensation carries safety and authority constraints. Separate estimation from authorised control. 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: The likely failed sensor is removed only after confirmation. 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. Measure detection delay, false alarm, isolation accuracy, probability calibration and recovery time. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A bank detects quickly but confuses two sensors. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
State error alone misses isolation quality. Report a confusion matrix. 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: State error alone misses isolation quality. Ask what other mechanism could create the same observed pattern, what additional measurement would separate the explanations, and how uncertainty should be recorded. This turns a formula into a defensible decision process.
Make the example auditable by predicting direction, calculating and checking a nearby case. The topic combines conditional probability, weighted averages and competing explanations. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
A learner can verify 0.4,0.6 and 11.8. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
The most probable model can still be wrong. Keep options open. 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: Keep options open. 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 model equations, priors, likelihoods, floors, transitions, thresholds, fusion and recovery. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Unit tests cover each model and an unknown fault. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Silent default priors make results irreproducible. Archive the bank. 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 model equations, priors, likelihoods, floors, transitions, thresholds, fusion and recovery. 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. GLR maximises unknown fault parameters; MMAE assigns probability across a discrete model bank. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Both use innovation evidence. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A finite bank can miss off-grid faults. Choose representation deliberately. 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: Both use innovation evidence. That example is illustrative rather than evidence about a particular school, learner or machine. A sound explanation keeps calculation, interpretation and action in separate sentences.
Separate the mechanism from the interpretation so a number does not become unsupported certainty. More models do not automatically improve diagnosis. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Add a third equally plausible model and update probabilities. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
Complexity can reduce calibration and interpretability. Add only distinguishable hypotheses. 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: Complexity can reduce calibration and interpretability. 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. Magill provides an early multiple-model estimation foundation; NASA analyses fault identifiability; Maybeck documents failure detection and compensation. The importance of mathematics here is that the statement becomes a quantity or logical test that another reader can reproduce.
Recent eduKate articles cover IMM, GLR and robust filtering. Follow the order of operations, keep the time index and units, and check whether the direction makes sense before trusting the number.
A model bank remains conditional on its hypotheses. Validate identification and recovery. 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 identification and recovery. Explain the result aloud, including the assumption that would most change the answer.
The nearest published companions are Student’s t filtering, Huber robust Kalman filtering, maximum correntropy Kalman filtering and normalized innovation squared. 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/10.1109/TAC.1965.1098191, https://ntrs.nasa.gov/citations/20120011947, https://doi.org/10.1002/(SICI)1099-1239(19991215)9:14%3C1051::AID-RNC452%3E3.0.CO;2-0.
Frequently asked questions
What is MMAE?
Multiple-model adaptive estimation runs parallel filters for competing hypotheses and updates their probabilities from innovation likelihoods.
How are mode probabilities updated?
Multiply each prior probability by its likelihood and normalise across all models.
What does the fused estimate mean?
It is a probability-weighted average of model-conditioned estimates; the conditional estimates should also be retained.
Can MMAE detect an unlisted fault?
It may show poor likelihood across the bank, but it cannot identify a fault model that is absent or indistinguishable.
Why is recovery logic needed?
After repair, probabilities may remain concentrated on the old fault model unless transitions, floors or reinitialisation allow return.
Extended practice laboratory
Practice case 1: Bayes update
Setup: Use priors 0.8,0.2 and likelihoods 0.1,0.6.
Checking target: Confirm posteriors 0.4,0.6.
Practice case 2: Fusion
Setup: Use states 10 and 13.
Checking target: Confirm 11.8.
Practice case 3: Equal likelihood
Setup: Set both likelihoods equal.
Checking target: Observe priors persist.
Practice case 4: Strong prior
Setup: Set healthy prior to 0.99.
Checking target: Measure detection delay.
Practice case 5: Likelihood floor
Setup: Apply a small probability floor.
Checking target: Test recovery.
Practice case 6: Third model
Setup: Add an actuator-fault hypothesis.
Checking target: Normalise three weights.
Practice case 7: Off-grid bias
Setup: Simulate bias between grid points.
Checking target: Inspect probability splitting.
Practice case 8: Common mode
Setup: Bias two sensors together.
Checking target: Compare independent and shared models.
Practice case 9: Repair
Setup: Return data to healthy.
Checking target: Measure recovery time.
Practice case 10: Decision memo
Setup: Report a 0.6 fault probability.
Checking target: Avoid certainty and specify confirmation.