Research · Fault Detection & Isolation

Observer-based Actuator Fault Detection

A comparative study of four model-based observers for detecting partial and complete loss of actuator effectiveness in a quadrotor under uncertainty.

PROJECT 01 / 06 OBSERVER RESIDUAL THRESHOLD FAULT DETECTED
Period2025–2026
ContextSemester thesis · TUM
My RoleResearch & Simulation
ToolchainMATLAB · Simulink
01Project overview

Detecting a failing actuator before control authority is lost.

A quadrotor has little room for hidden actuator degradation. The project asked whether analytical redundancy—comparing measured motion with model-based estimates—could expose a fault early without creating nuisance alarms during nominal uncertainty.

?

The Engineering Question

How do observer choice, model complexity and uncertainty change detection sensitivity for partial and total loss-of-effectiveness faults?

!

The Central Constraint

Residuals must react to real faults while remaining quiet enough under noise and model mismatch to avoid false positives.

02My Contribution

From nonlinear vehicle model to comparable detection evidence.

01

Observer Implementation

Implemented Luenberger, Thau, extended Kalman and unscented Kalman observer schemes as residual generators.

02

Fault Modelling

Defined partial and complete actuator loss-of-effectiveness scenarios so every method saw the same controlled test cases.

03

Robustness Evaluation

Tested nominal, faulty and parametrically uncertain conditions to separate genuine sensitivity from model-dependent behaviour.

04

Comparative Analysis

Assessed detection behaviour, implementation complexity and false-alarm tendencies rather than selecting a method from one idealised run.

03Technical approach

One test framework, four observer philosophies.

Model

Establish the quadrotor dynamics, actuator inputs and measurable states.

Estimate

Run deterministic and stochastic observers against a common plant model.

Inject

Introduce controlled effectiveness losses and parameter uncertainty.

Evaluate

Compare residual response, sensitivity, robustness and alarm behaviour.

Why four observers?The linear Luenberger observer provides a transparent baseline; the Thau observer addresses nonlinear dynamics; EKF and UKF add noise-aware state estimation with different treatments of nonlinearity.
04Team & contributors

Research developed within TUM’s flight-systems environment.

Researcher

Aqib Habib

Observer implementation, simulation campaign, comparative evaluation and thesis documentation.

Supervisor

Z. Mbikayi

Supervisor to the semester-thesis work.

Host institute

Institute of Flight System Dynamics

05Results

A validated comparison, not just four isolated implementations.

The completed study characterises how the four residual-generation methods respond to actuator degradation and uncertainty, creating a structured basis for selecting detection architecture in future fault-tolerant flight-control work.

4
Observer strategies implemented and compared
2
Fault classes: partial and complete effectiveness loss
1.3
Semester-thesis grade at TUM