IAES International Journal of Robotics and Automation (IJRA)

Vol. 14, No. 2, June 2025, pp. 214~226

ISSN: 2722-2586, DOI: 10.11591/ijra.v14i2.pp214-226  214

Design of H-/H∞ based fault detection filter for linear uncertain

systems using linear matrix inequalities

Masood Ahmad1,2, Rosmiwati Mohd-Mokhtar1

1
School of Electrical and Electronic Engineering, Engineering Campus, Universiti Sains Malaysia, Penang, Malaysia
2
Electrical and Computer Engineering Department, COMSATS University Islamabad (Lahore Campus), Pakistan

Article Info ABSTRACT

Article history: One of the significant challenges in model-based fault detection is achieving
robustness against disturbances and model uncertainties while ensuring
Received Nov 27, 2024 sensitivity to faults. This study proposes an optimized approach for
Revised Mar 7, 2025 designing fault detection filters for discrete-time linear systems with norm-
Accepted May 27, 2025 bounded model uncertainties. The design leverages the H-/H∞ optimization
framework and is expressed through linear matrix inequality constraints. The
filter is designed to produce a residual signal that balances two opposing
Keywords: objectives: minimizing the impact of disturbances and model uncertainties
while maximizing fault sensitivity. The effectiveness of the proposed
Fault detection method is demonstrated through simulations involving sensor and actuator
Linear matrix inequality fault detection in the well-known three-tank system. Simulation results
Model uncertainty illustrate the method's ability to maintain robustness against disturbances and
Observer design uncertainties while effectively detecting faults in the system.
Robustness
This is an open access article under the CC BY-SA license.

Corresponding Author:
Rosmiwati Mohd-Mokhtar
School of Electrical and Electronic Engineering, Engineering Campus, Universiti Sains Malaysia
Penang, Malaysia
Email: [email protected]

1. INTRODUCTION
Fault detection is essential in robotics and automation systems to ensure reliability, safety,
environmental sustainability, and achieve desired performance levels [1]. It identifies issues early, minimizes
downtime, prevents accidents, and maintains product quality. In the industry, it increases productivity,
ensures operational continuity, and contributes to sustainability by preventing waste and inefficiencies.
Robotics and automation systems frequently interact with humans or hazardous materials. Detecting faults
early prevents accidents, protects workers, and minimizes risks. As industries adopt smart manufacturing,
fault detection becomes integral to real-time monitoring, self-diagnosis, and autonomous decision-making,
key principles of Industry 4.0. Meeting these requirements often increases both system complexity and cost.
Faults or abnormal behaviors in such complex systems can degrade performance and potentially lead to
hazardous situations, posing risks to human safety and financial loss. Thus, early detection and identification
of abnormal system behaviors are essential to prevent these adverse outcomes [2]–[5]. Over the past two
decades, numerous advancements in resilient fault detection (FD) system design have been made, broadly
categorized into model-based and model-free approaches [6]–[9]. Among model-based methods, observer-
based techniques have gained popularity due to their simpler structure and relatively lower design complexity
[10], [11]. These approaches utilize a fault detection filter (FDF) to generate a residual, defined as the
difference between the system's measured outputs and the estimated outputs derived from its model. By
comparing the residual against a predefined threshold, the occurrence of faults can be identified [12], [13].

Journal homepage: http://ijra.iaescore.com

IAES Int J Rob & Autom ISSN: 2722-2586  215

The presence of external disturbances and model uncertainties complicates the residual generation
process, often producing non-zero residuals even in fault-free scenarios. Ideally, a fault-free system should
yield a zero-value residual, while faulty conditions should result in a non-zero residual. However,
disturbances and uncertainties may lead to false alarms, undermining the FD process. Therefore, robust
residual generation is critical for effective FD [14]–[17]. Addressing disturbances and model uncertainty in
model-based FD systems presents a significant challenge. To this end, the H∞ norm optimization technique
has been employed to enhance residual robustness against disturbances [18]. Conversely, the H- index, which
reflects the minimum fault sensitivity of the residual, has been used to design fault-sensitive FDFs, enhancing
their sensitivity to faults [19]–[21]. While H∞ optimization ensures robustness against disturbances, it also
reduces fault sensitivity, and similarly, H- index-based FDFs, although fault-sensitive, may amplify the
influence of disturbances [22]. Balancing these trade-offs is key to achieving optimal FD performance.
An optimal FD system aims to minimize the impact of unknown disturbances (minimizing the H∞
norm) while maximizing fault sensitivity (maximizing the H- index), framing the design as a multi-objective
optimization problem. A literature review indicates that most FD methods address robustness and sensitivity
issues for continuous-time or discrete-time linear systems with external disturbances only [23]–[26].
However, model uncertainties in system matrices can introduce biases in the residual, necessitating careful
handling to ensure robust residual generation. For uncertain continuous-time linear systems, an observer-
based FD system was proposed in [27], utilizing iterative linear matrix inequalities (LMIs) to generate robust
residuals. This approach provided an optimal balance between robustness to disturbances and fault sensitivity
for the multi-objective optimization problem. However, the method's complexity increased due to the need to
first derive a theoretically optimal solution and subsequently design the observer. In contrast, an H∞ based
FD residual generator for linear systems was developed in [28], demonstrating robustness against
disturbances and model uncertainty. Nevertheless, this approach did not address fault sensitivity issues. For
successful FD, it is crucial to simultaneously consider fault sensitivity and robustness.
Motivated by the scarcity of solutions addressing the multi-objective optimization problem for
discrete-time linear systems with norm-bounded model uncertainties, this study seeks to develop an optimal
observer-based residual generator. The proposed method ensures observer stability while achieving
robustness to disturbances, resilience against model uncertainties, and enhanced fault sensitivity. The
existence of the proposed observer is established through sufficient conditions expressed as LMIs. The
results obtained for the observer-based fault detection filter were illustrated through a simulation analysis of a
three-tank system. The proposed approach can be applied to any discrete-time linear system with norm-
bounded model uncertainties and disturbances. The key contributions of this research are outlined as follows:
a. Development of an H∞ observer-based filter to minimize the H∞ norm of Grd, the transfer function matrix
representing the disturbance-to-residual relationship, within the linear matrix inequality (LMI)
framework.
b. Development of an H- observer-based filter aimed at maximizing the H- norm of Grf, the transfer function
matrix from fault to residual, also using the LMI framework.
c. Design of an observer-based filter utilizing the H-/H∞ optimization method, which concurrently
minimizes the H∞ norm while maximizing the H- norm. This approach seeks to create an optimal
observer-based residual generator that satisfies both H∞ and H- performance criteria.
d. After constructing the proposed observer, the l2 norm is applied to assess and compare the generated
residual against a defined threshold to detect faults.
The structure of the paper is as follows. Section 2 introduces the problem formulation. Section 3
details the core contribution of the research, including the derivation of the filter gain matrix. Section 4
provides simulation results showcasing the filter's effectiveness, particularly in detecting sensor and actuator
faults within a three-tank system. Finally, section 5 presents concluding remarks.

2. PROBLEM FORMULATION
The discrete-time linear system in (1) is adopted to formulate the problem being solved in this paper.

𝑥(𝑘 + 1) = 𝐴𝑥(𝑘) + 𝐸𝑓 𝑓(𝑘) + 𝐸𝑑 𝑑(𝑘) + (𝐵 + 𝛥𝐵)𝑢(𝑘) + 𝛥𝐴𝑥(𝑘)

𝑦(𝑘) = 𝐶𝑥(𝑘) + 𝐹𝑓 𝑓(𝑘) + 𝐹𝑑 𝑑(𝑘) + (𝐷 + 𝛥𝐷)𝑢(𝑘) + 𝛥𝐶𝑥(𝑘) (1)

Let x(k) ∈ Rn represent the state vector, u(k) ∈ Rp the control input vector, and y(k) ∈ Rm the measurement
vector. The disturbance vector d(k) is l2 norm bounded, such that ∥d(k)∥2 ≤ δd, while f(k) is the l2 norm
bounded fault vector to be detected. The matrices Ed, Fd, Ef, and Ff define the locations where the disturbance
and fault vectors influence the system dynamics, respectively. The matrices A, B, C, and D are the nominal
system matrices with compatible dimensions, and ΔA, ΔB, ΔC, and ΔD represent norm-bounded model
uncertainties, given as (2).
Design of H-/H∞ based fault detection filter for linear uncertain systems … (Masood Ahmad)
216  ISSN: 2722-2586

∆𝐴 ∆𝐵 𝐻 ∑𝐺 𝐻1 ∑𝐺2
[ ]=[ 1 1 ] (2)
∆𝐶 ∆𝐷 𝐻2 ∑𝐺1 𝐻2 ∑𝐺2

where ∑ is an unknown scalar constant holds the condition, i.e., ∑𝑇 ∑ ≤ 𝐼. The assumptions listed below are
used consistently throughout this work [18]:
𝐴 − 𝑒 𝑗𝜃 𝐼 𝐸𝑑
A1: System (1) is observable; A2: [ ] has full row rank, while 𝜃 ∈ [0,2𝜋]; A3: (𝐴 + 𝛥𝐴) is
𝐶 𝐹𝑑
stable
As initially introduced, the model-based FD system comprises two subsystems: a residual generator
and a residual evaluator with thresholding and decision logic. An observer-based FD filter is used for
generating the residual, which is expressed as (3):

𝑥̂(𝑘 + 1) = 𝐴𝑥̂(𝑘) + 𝐿(𝑦(𝑘) − 𝑦̂(𝑘)) + 𝐵𝑢(𝑘)

𝑟(𝑘) = 𝑦(𝑘) − 𝑦̂(𝑘) (3)

(𝑦̂(𝑘) = 𝐶𝑥̂(𝑘) + 𝐷𝑢(𝑘)) ∈ 𝑅𝑚 and 𝑥̂(𝑘) ∈ 𝑅𝑛 represent the estimated output and the state estimation
vectors, respectively. The residual signal is denoted by 𝑟(𝑘), and the filter gain 𝐿 serves as the design
parameter for the proposed FD filter. The dynamics of the filter are described by the state estimation error
vector, 𝑒(𝑘) = 𝑥(𝑘) − 𝑥̂(𝑘). The following equations represent the error dynamics and the residual:

𝑒(𝑘 + 1) = (𝐴 − 𝐿𝐶)𝑒(𝑘) + (𝐸𝑓 − 𝐿𝐹𝑓 ) 𝑓(𝑘) + (𝐸𝑑 − 𝐿𝐹𝑑 )𝑑(𝑘) + (𝛥𝐴 − 𝐿𝛥𝐶)𝑥(𝑘) + (𝛥𝐵 − 𝐿𝛥𝐷)𝑢(𝑘) (4)

𝑟(𝑘) = 𝐶𝑒(𝑘) + 𝐹𝑓 𝑓(𝑘) + 𝐹𝑑 𝑑(𝑘) + 𝛥𝐶𝑥(𝑘) + 𝛥𝐷𝑢(𝑘) (5)

Undesired behavior in FD theory is caused by model uncertainty and disturbance, which influences
the estimation process and makes the residual sensitive to faults, control input, and the system's state [29].
For the sake of simplicity, the dynamics of (4) are governed by two new vectors:

𝑒(𝑘) 𝑢(𝑘)
𝑥̅ (𝑘) = [ ] and 𝑢̅(𝑘) = [ ]
𝑥(𝑘) 𝑑(𝑘)

Then, an augmented system is represented as (6), (7):

𝑥̅ (𝑘 + 1) = 𝐴̅𝑥̅ (𝑘) + 𝐵̅ 𝑢̅(𝑘) + 𝐸̅𝑓 𝑓(𝑘) (6)

𝑟(𝑘) = 𝐶̅ 𝑥̅ (𝑘) + 𝐷
̅ 𝑢̅(𝑘) + 𝐹𝑓 𝑓(𝑘) (7)

where

𝐴 − 𝐿𝐶 𝛥𝐴 − 𝐿𝛥𝐶 ̅ 𝛥𝐵 − 𝐿𝛥𝐷 𝐸𝑑 − 𝐿𝐹𝑑 𝐸𝑓 − 𝐿𝐹𝑓

𝐴̅ = [ ]; 𝐵 = [ ]; 𝐶̅ = [𝐶 ̅ = [𝛥𝐷
𝛥𝐶 ]; 𝐷 𝐹𝑑 ]; 𝐸̅𝑓 = [ ]
0 𝐴 + 𝛥𝐴 𝐵 + 𝛥𝐵 𝐸𝑑 𝐸𝑓

The residual signal in (7) can be represented in the frequency domain.

𝑟(𝑧) = 𝐺𝑟𝑢̅ (𝑧)𝑢̅ (𝑧) + 𝐺𝑟𝑓 (𝑧)𝑓(𝑧) (8)

where 𝐺𝑟𝑢̅ (𝑧) = 𝐶̅ (𝑧𝐼 − 𝐴̅ + 𝐿𝐶̅ )−1 (𝐵̅ − 𝐿𝐷

̅) + 𝐷̅ and 𝐺𝑟𝑓 (𝑧) = 𝐶̅ (𝑧𝐼 − 𝐴̅ + 𝐿𝐶̅ )−1 (𝐸̅𝑓 − 𝐿𝐹𝑓 ) + 𝐹𝑓
𝐺𝑟𝑢̅ (𝑧) and 𝐺𝑟𝑓 (𝑧) are the transfer function matrices from 𝑢̅(𝑘) and 𝑓(𝑘) to 𝑟(𝑘), respectively. The influence
of disturbance and model uncertainty on the residual is measured by 𝐻∞ norm and is represented as (9):
sup sup ∑∞ 𝑇
𝑘=0 𝑟 (𝑘)𝑟(𝑘)
𝐻∞ = ‖𝐺𝑟𝑢̅ (𝑧)‖∞ = 𝜎̅(𝐺𝑟𝑢̅ (𝑧)) = 𝑢̅ (𝑘) ∈ 𝑙 , ‖𝑢̅‖ ≠ 0 {∑∞ } (9)
𝜃 ∈ [0,2𝜋] 2 2 𝑘=0 ̅
𝑢 𝑇 (𝑘)𝑢
̅(𝑘)

Robustness against disturbance and model uncertainty is expressed by (10) [30].

‖𝐺𝑟𝑢̅ (𝑧)‖∞ < 𝛾; 𝛾 > 0 (10)

𝛾 represents the maximum effect of model uncertainty and disturbance on the residual, and the value of 𝛾
should be smaller. Likewise, the effect of fault on the residual is characterized by 𝐻− index [31].

IAES Int J Rob & Autom, Vol. 14, No. 2, June 2025: 214-226
IAES Int J Rob & Autom ISSN: 2722-2586  217

inf
𝐻− = ‖𝐺𝑟𝑓 (𝑒 𝑗𝜃 )‖− = 𝜎[𝐺𝑟𝑓 (𝑒 𝑗𝜃 )] (11)
𝜃 ∈ [0,2𝜋]

The residual's sensitivity to the fault is illustrated by (12).

‖𝐺𝑟𝑓 (𝑒 𝑗𝜃 )‖− > 𝛽 ; 𝛽 > 0 (12)

𝛽 denotes the worst-case fault sensitivity measurement of the residual signal. A larger value of 𝛽 shows that
residual is more sensitive to a fault.
𝐻
The solution of an optimal FD filter design based on − optimization for the nominal system
𝐻∞
(system uncertainty, 𝛥𝑠 = 0) can be easily obtained by solving a single Riccati equation [18]. Unfortunately,
𝐻
the Riccati equation cannot solve − optimization problems for dynamic systems with model uncertainties
𝐻∞
(𝛥𝑠 ≠ 0). The multi-objective optimization problem for linear systems subject to disturbance and model
𝐻
uncertainty is addressed in this study using the LMI technique. Using − optimization, the objective is to
𝐻∞
design an optimal FD filter by determining the filter gain matrix 𝐿 in a way that (a) makes the augmented
system (6) asymptotically stable, (b) makes the residual (7) robust to disturbance and model uncertainties in
the 𝐻∞ sense, and (c) makes the residual (7) fault-sensitive.

3. SYNTHESIS OF OPTIMAL FD FILTER

In this section, an optimal FD filter is designed for system (1) in the LMI framework. First, separate
solutions of 𝐻∞ and 𝐻− index conditions in (10) and (12) are obtained, followed by an algorithm for solving
𝐻
mixed − optimization problem. For onward discussion, the following lemmas help to derive the main
𝐻∞
results.
Lemma 1 [18]: The observer error dynamics (4) is asymptotically stable and meets the condition (10) for the
linear system (1) with zero model uncertainty in the system matrices if the following LMI is true for (𝑓(𝑘) =
0) then there exists a scalar 𝛾 ≥ 𝛾min , matrix 𝐿 and 𝑃 = 𝑃𝑇 > 0.

−𝑃 𝑃(𝐴 − 𝐿𝐶) 𝑃(𝐸𝑑 − 𝐿𝐹𝑑 ) 0

(𝐴 − 𝐿𝐶)𝑇 𝑃 −𝑃 0 𝐶𝑇
<0
(𝐸𝑑 − 𝐿𝐹𝑑 )𝑇 𝑃 0 −𝛾𝐼 𝐹𝑑 𝑇
[ 0 𝐶 𝐹𝑑 −𝛾𝐼]

The above lemma provides the necessary and sufficient condition for (10) and ensures that ‖𝐺𝑟𝑢̅ (𝑧)‖∞ < 𝛾.
Lemma 2 [18]: The observer error dynamics (4) is asymptotically stable and meets the condition (12) for the
linear system (1) with zero model uncertainty in the system matrices if the following LMI is true for (𝑢̅(𝑘) =
0), then there exists a scalar 𝛽 ≤ 𝛽max , matrix 𝐿 and 𝑃 = 𝑃𝑇 > 0.

𝑃 − (𝐴 − 𝐿𝐶)𝑇 𝑃(𝐴 − 𝐿𝐶) − 𝐶 𝑇 𝐶 𝐶 𝑇 𝐹𝑓 + (𝐴 − 𝐿𝐶)𝑇 𝑃(𝐸𝑓 − 𝐿𝐹𝑓 )

[ 𝑇 ]>0
(𝐸𝑓 − 𝐿𝐹𝑓 )𝑇 𝑃(𝐴 − 𝐿𝐶) + 𝐹𝑓 𝑇 𝐶 𝛽 2 𝐼 − 𝐹𝑓 𝐹𝑓 − (𝐸𝑓 − 𝐿𝐹𝑓 )𝑇 𝑃(𝐸𝑓 − 𝐿𝐹𝑓 )

The above lemma guarantees that the minimum fault sensitivity of the residual is greater than a constant, i.e.,
‖𝐺𝑟𝑓 (𝑒 𝑗𝜃 )‖− > 𝛽.
Lemma 3 [30]: If there exists a symmetric positive definite matrix 𝑃, and an arbitrary positive scalar 𝜀 > 0
that satisfy (𝜀𝐼 − 𝐻𝑇 𝑃𝐻)−1 > 0 then

(𝑊 + 𝐻∑𝐺)𝑇 𝑃(𝑊 + 𝐻∑𝐺) ≤ 𝑊 𝑇 𝑃𝑊 + 𝑊 𝑇 𝑃𝐻(𝜀𝐼 − 𝐻𝑇 𝑃𝐻)−1 𝐻 𝑇 𝑃𝑊 + 𝜀𝐺 𝑇 𝐺

Lemma 4 [32]: The following conditions are equivalent when the Schur complement principle is applied to
several symmetric matrices 𝐴11 , 𝐴12 and 𝐴22 .

𝐴 𝐴12
If 𝐴11 < 0 then [ 11 ] < 0 if and only if 𝐴22 − 𝐴21 (𝐴11 )−1 𝐴12 < 0
𝐴21 𝐴22
𝐴 𝐴12
If 𝐴22 < 0 then [ 11 ] < 0 if and only if 𝐴11 − 𝐴12 (𝐴22 )−1 𝐴21 < 0
𝐴21 𝐴22

Design of H-/H∞ based fault detection filter for linear uncertain systems … (Masood Ahmad)
218  ISSN: 2722-2586

Theorem 1: Consider system (1) with model uncertainties (∆𝐴, ∆𝐵, ∆𝐶, ∆𝐷 ≠ 0), under the assumptions A1
and A2, if there exist scalars 𝛽 > 0, 𝛾 > 0, a filter gain matrix 𝐿, a symmetric matrix 𝑃 > 0 and a scalar 𝜀 >
0 such that the augmented system (6) is asymptotically stable and the following matrix inequalities hold,
then conditions (10) and (12) are satisfied.

𝐻2 𝑇 𝐻2 + 𝐻3 𝑇 𝑃𝐻3 − 𝜀𝐼 𝐻2 𝑇 𝐶0 + 𝐻3 𝑇 𝑃𝐴0 𝐻2 𝑇 𝐷0 + 𝐻3 𝑇 𝑃𝐵0

[ 𝐶0 𝑇 𝐻2 + 𝐴0 𝑇 𝑃𝐻3 𝐶0 𝑇 𝐶0 + 𝐴0 𝑇 𝑃𝐴0 + 𝜀𝐺1 𝑇 𝐺1 − 𝑃 𝐶0 𝑇 𝐷0 + 𝐴0 𝑇 𝑃𝐵0 + 𝜀𝐺1 𝑇 𝐺2 ]<0 (13)
𝐷0 𝑇 𝐻2 + 𝐵0 𝑇 𝑃𝐻3 𝐷0 𝑇 𝐶0 + 𝐵0 𝑇 𝑃𝐴0 + 𝜀𝐺2 𝑇 𝐺1 𝐷0 𝑇 𝐷0 + 𝐵0 𝑇 𝑃𝐵0 + 𝜀𝐺2 𝑇 𝐺2 − 𝛾 2 𝐼

−𝐻2 𝑇 𝐻2 − 𝐻3 𝑇 𝑃𝐻3 + 𝜀𝐼 𝐻2 𝑇 𝐶𝑜̅ + 𝐻3 𝑇 𝑃𝐴̅𝑜 𝐻2 𝑇 𝐹𝑓𝑜 + 𝐻3 𝑇 𝑃𝐸̅𝑓𝑜

𝑇 𝑇 𝑇 𝑇 𝑇 𝑇
[ 𝐶𝑜̅ 𝐻2 + 𝐴̅𝑜 𝑃𝐻3 −𝐶𝑜̅ 𝐶𝑜̅ − 𝐴̅𝑜 𝑃𝐴̅𝑜 − 𝜀𝐺1 𝐺1 + 𝑃 𝑇
−𝐶𝑜̅ 𝐹𝑓𝑜 − 𝐴̅𝑜 𝑃𝐸̅𝑓𝑜 ]>0 (14)
𝑇 𝑇 𝑇 𝑇 𝑇 𝑇
𝐹𝑓𝑜 𝐻2 + 𝐸̅𝑓𝑜 𝑃𝐻3 −𝐹𝑓𝑜 𝐶𝑜̅ − 𝐸̅𝑓𝑜 𝑃𝐴̅𝑜 −𝐹𝑓𝑜 𝐹𝑓𝑜 − 𝐸̅𝑓𝑜 𝑃𝐸̅𝑓𝑜 + 𝛽 2 𝐼

In addition to solving (13) and (14), optimal filter gain 𝐿, can be determined by solving the following
optimization problem:

𝛽
max 𝐽 = (15)
𝛾

Proof of the Theorem

For the augmented system (6) and (7), (10) can be expressed as (16):

‖𝐺𝑟𝑢̅ ‖∞ < 𝛾 ↔ ∑∞ 𝑇 2 𝑇
𝑘=0[𝑟 (𝑘)𝑟(𝑘) − 𝛾 𝑢
̅ (𝑘)𝑢̅(𝑘)] < 𝛾; 𝑓(𝑘) = 0 (16)

Defining a Lyapunov function, 𝑉(𝑥̅ (𝑘)) = 𝑥̅ 𝑇 (𝑘)𝑃𝑥̅ (𝑘) > 0 where 𝑃 = diag[𝑃1 , 𝑃2 ] > 0. Suppose 𝑃 > 0,
the necessary stability condition listed below is ensured.

∑∞
𝑘=0 (𝑉(𝑥̅ (𝑘 + 1)) − 𝑉(𝑥̅ (𝑘))) < 0 (17)

The control objective (10) and 𝐻∞ FD filter stability is ensured by combining (16) and (17), which will yield

∑∞ 𝑇 2 𝑇 (𝑘)𝑢 (𝑘)]
𝑘=0[𝑟 (𝑘)𝑟(𝑘) + 𝑉(𝑥̅ (𝑘 + 1)) − 𝑉(𝑥̅ (𝑘)) − 𝛾 𝑢
̅ ̅ <0 (18)

From equation (6) and (7), it is easy to write

̅
𝑇
𝐶̅ 𝑇 −𝑃 0 𝑥̅ (𝑘)
[𝑥̅ 𝑇 (𝑘) 𝑢̅𝑇 (𝑘)] ([ 𝐴𝑇 ] 𝑃[𝐴̅ 𝐵̅] + [ ̅ 𝑇 ] [𝐶̅ ̅] + [
𝐷 ]) [ ]<0 (19)
𝐵̅ 𝐷 0 −𝛾 2 𝐼 𝑢̅(𝑘)

The constant and uncertain system matrices are divided as follows to avoid any ambiguity:

̅ ̅ 𝐶 𝐷0 ̅ ̅
[𝐶 𝐷 ]=[ 0 ] + [∆𝐶 ∆𝐷 ] (20)
𝐴̅ 𝐵̅ 𝐴0 𝐵0 ∆𝐴̅ ∆𝐵̅

𝐶 0 0 𝐹𝑑
𝐶 𝐷0
where [ 0 ]=[ 𝐴 − 𝐿𝐶 0 0 𝐸𝑑 − 𝐿𝐹𝑑 ];
𝐴0 𝐵0
0 𝐴 𝐵 𝐸𝑑
𝐻2
̅ ̅ 𝐻2
[∆𝐶 ∆𝐷 ] = [𝐻1 − 𝐿𝐻2 ] ∑[0 𝐺1 𝐺2 0] = [ ] ∑[𝐺1 𝐺2 ] = 𝐻 ∑ 𝐺
∆𝐴̅ ∆𝐵̅ 𝐻1
𝐻3

Representing the above matrices as:

𝐶 𝐷0 𝐴 − 𝐿𝐶 0 0 𝐸𝑑 − 𝐿𝐹𝑑 𝐻
𝐴𝑜 = [ 0 ]; 𝐴𝑜 = [ ]; 𝐵 = [ ]; 𝐶𝑜 = [𝐶 0]; 𝐷𝑜 = [0 𝐹𝑑 ]; 𝐻 = [ 2 ];
𝐴0 𝐵0 0 𝐴 𝑜 𝐵 𝐸𝑑 𝐻3
𝐻 − 𝐿𝐻2 −𝑃 0 𝐼 0
𝐻3 = [ 1 ]; 𝐺 = [𝐺1 𝐺2 ]; 𝐺1 = [0 𝐺1 ]; 𝐺2 = [𝐺2 0]; 𝑃 = [ ]; 𝑃 = [ ]
𝐻1 0 −𝛾 2 𝐼 0 𝑃

Applying Lemma 3 on (19) using (20) turns to (21).

IAES Int J Rob & Autom, Vol. 14, No. 2, June 2025: 214-226
IAES Int J Rob & Autom ISSN: 2722-2586  219

𝑇 𝑥̅ (𝑘)
[𝑥̅ 𝑇 (𝑘) 𝑢̅𝑇 (𝑘)] ((𝐴𝑜 + 𝐻 ∑ 𝐺) 𝑃(𝐴𝑜 + 𝐻 ∑ 𝐺) + 𝑃) [ ] (21)
𝑢̅(𝑘)
𝑇 −1
(𝐴𝑜 + 𝐻 ∑ 𝐺) 𝑃(𝐴𝑜 + 𝐻 ∑ 𝐺) + 𝑃 ≤ 𝐴𝑜 𝑇 𝑃𝐴𝑜 + 𝐴𝑜 𝑇 𝑃𝐻(𝜀𝐼 − 𝐻 𝑇 𝑃𝐻) 𝐻 𝑇 𝑃𝐴𝑜 + 𝜀𝐺 𝑇 𝐺 + 𝑃 (22)

Applying Lemma 4 on (22) will yield (23).

𝐻 𝑇 𝑃𝐻 − 𝜀𝐼 𝐻 𝑇 𝑃𝐴𝑜
[ ]<0 (23)
𝐴𝑜 𝑇 𝑃𝐻 𝐴𝑜 𝑇 𝑃𝐴𝑜 + 𝜀𝐺 𝑇 𝐺 + 𝑃

Expanding (23), one can write as (24).

𝐻2 𝑇 𝐻2 + 𝐻3 𝑇 𝑃𝐻3 − 𝜀𝐼 𝐻2 𝑇 𝐶0 + 𝐻3 𝑇 𝑃𝐴0 𝐻2 𝑇 𝐷0 + 𝐻3 𝑇 𝑃𝐵0

[ 𝐶0 𝑇 𝐻2 + 𝐴0 𝑇 𝑃𝐻3 𝐶0 𝐶0 + 𝐴0 𝑇 𝑃𝐴0 + 𝜀𝐺1 𝑇 𝐺1 − 𝑃
𝑇
𝐶0 𝐷0 + 𝐴0 𝑇 𝑃𝐵0 + 𝜀𝐺1 𝑇 𝐺2
𝑇
] < 0 (24)
𝑇 𝑇 𝑇 𝑇 𝑇 𝑇 𝑇 𝑇
𝐷0 𝐻2 + 𝐵0 𝑃𝐻3 𝐷0 𝐶0 + 𝐵0 𝑃𝐴0 + 𝜀𝐺2 𝐺1 𝐷0 𝐷0 + 𝐵0 𝑃𝐵0 + 𝜀𝐺2 𝐺2 − 𝛾2𝐼

Rewriting the above matrix inequality as (25).

𝐻2 𝑇 𝐻3 𝑇 0 𝐼 0 0 𝐻2 𝐶0 𝐷0 𝜀𝐼 0 0
[ 𝐶0 𝑇 𝐴0 𝑇 𝐺1 𝑇 ] [0 𝑃 0 ] [𝐻3 𝐴0 𝐵0 ] − [ 0 𝑃 0 ]<0 (25)
𝐷0 𝑇
𝐵0 𝑇
𝐺2 𝑇 0 0 𝜀𝐼 0 𝐺1 𝐺2 0 0 𝛾 2𝐼

Applying the Schur complement lemma given below, (25) can be represented as (26).

−𝜀𝐼 0 0 𝐻2 𝑇 𝐻3 𝑇 0
0 −𝑃 0 𝐶0 𝑇 𝐴0 𝑇 𝐺1 𝑇
0 0 −𝛾 2 𝐼 𝐷0 𝑇 𝐵0 𝑇 𝐺2 𝑇 < 0 (26)
𝐻2 𝐶0 𝐷0 −𝐼 0 0
𝐻3 𝐴0 𝐵0 0 −𝑃 −1 0
[ 0 𝐺1 𝐺2 0 0 −𝜀 −1 𝐼]

The nonlinear inequality is transformed into linear inequality form by performing matrix equivalent
transformation as (27).

−𝜀𝐼 0 0 𝐻2 𝑇 𝐻3 𝑇 𝑃 0
0 −𝑃 0 𝐶0 𝑇 𝐴0 𝑇 𝑃 𝐺1 𝑇 𝜀
0 0 −𝛾 2 𝐼 𝐷0 𝐵0 𝑃 𝐺2 𝑇 𝜀 < 0
𝑇 𝑇
(27)
𝐻2 𝐶0 𝐷0 −𝐼 0 0
𝑃𝐻3 𝑃𝐴0 𝑃𝐵0 0 −𝑃 0
[ 0 𝜀𝐺1 𝜀𝐺2 0 0 −𝜀𝐼 ]

By inserting the 𝐻3 , 𝐴0 , 𝐵0 , 𝐶0 , 𝐷0 , 𝐺1 , 𝐺2 and 𝑃 matrices in the above matrix, which completes the proof of
the first part of Theorem 1. Similarly, 𝐻− index-based fault sensitivity condition (12) can be derived as (28).

‖𝐺𝑟𝑓 ‖− > 𝛽 ↔ ∑∞ 𝑇 2 𝑇
𝑘=0[𝑟 (𝑘)𝑟(𝑘) > 𝛽 𝑓 (𝑘)𝑓(𝑘)] ; 𝑢
̅ (𝑘) = 0 (28)

Considering the Lyapunov function defined earlier, 𝑉(𝑘) = 𝑥̅ 𝑇 (𝑘)𝑃𝑥̅ (𝑘) > 0, 𝑃 > 0. The control objective
(12) and 𝐻− index filter stability is ensured by (29).

∑∞ 𝑇 2 𝑇
𝑘=0[𝑟 (𝑘)𝑟(𝑘) − 𝑉(𝑥̅ (𝑘 + 1)) + 𝑉(𝑥̅ (𝑘)) − 𝛽 𝑓 (𝑘)𝑓(𝑘)] > 0 (29)

After mathematical simplification, one can write (29) as (30).

∑∞ 𝑇 2 𝑇
𝑘=0[𝑟 (𝑘)𝑟(𝑘) − 𝛽 𝑓 (𝑘)𝑓(𝑘) − 𝑉(𝑥̅ (𝑘 + 1)) + 𝑉(𝑥̅ (𝑘))] < 0 (30)

Substituting matrices from (6) and (7) into (30) by taking 𝑢̅(𝑘) = 0, can be written in matrix form as (31)
and (32).
Design of H-/H∞ based fault detection filter for linear uncertain systems … (Masood Ahmad)
220  ISSN: 2722-2586

𝐴𝑇̅ 𝐶̅ 𝑇 −𝑃 0 𝑥̅
[𝑥̅ 𝑇 𝑓 𝑇 ] ([ ̅ 𝑇 ] 𝑃[𝐴̅ 𝐸̅𝑓 ] + [ 𝑇 ] [𝐶̅ 𝐹𝑓 ] + [ ]) [ ] < 0 (31)
𝐸𝑓 𝐹𝑓 0 −𝛽 2 𝐼 𝑓

𝐶̅ 𝑇 𝐴𝑇̅ 𝐼 0 𝐶̅ 𝐹𝑓 −𝑃 0 𝑥̅
[𝑥̅ 𝑇 𝑓 𝑇 ] ([ 𝑇] [ ][ ]+[ ]) [ ] < 0 (32)
𝐹𝑓 𝑇 ̅
𝐸𝑓 0 𝑃 𝐴̅ 𝐸̅𝑓 0 −𝛽 2 𝐼 𝑓

The constant and uncertain matrices are separated as:

𝐶̅ 𝐹𝑓 𝐶𝑜̅ 𝐹𝑓𝑜 ∆𝐶̅ ∆𝐹𝑓

[ ]=[ ]+[ ] (33)
𝐴̅ ̅
𝐸𝑓 𝐴̅𝑜 ̅
𝐸𝑓𝑜 ∆𝐴̅ ∆𝐸̅𝑓

𝐶 0 𝐹𝑓 𝐻2
𝐶𝑜̅ 𝐹𝑓𝑜 ∆𝐶̅ ∆𝐹𝑓
where [ ] = [𝐴 − 𝐿𝐶 0 𝐸𝑓 − 𝐿𝐹𝑓 ] and [ ̅ ] = [𝐻1 − 𝐿𝐻2 ] ∑[0 𝐺1 0]
𝐴̅𝑜 𝐸̅𝑓𝑜 ∆𝐴 ∆𝐸̅𝑓 𝐻
0 𝐴 𝐸𝑓 1

Defining the new matrices:

𝐶𝑜̅ 𝐹𝑓𝑜 𝐴 − 𝐿𝐶 0 ̅ 𝐸𝑓 − 𝐿𝐹𝑓 𝐻

𝐴̿𝑜 = [ ]; 𝐴̅𝑜 = [ ]; 𝐸𝑓𝑜 = [ ]; 𝐹𝑓𝑜 = 𝐹𝑓 ; 𝐶𝑜̅ = [𝐶 0]; 𝐻 = [ 2 ];
̅
𝐴𝑜 𝐸𝑓𝑜̅ 0 𝐴 𝐸𝑓 𝐻3
𝐻1 − 𝐿𝐻2 𝐼 0 −𝑃 0
𝐻3 = [ ]; 𝐺 = [𝐺1 0]; 𝐺1 = [0 𝐺1 ]; 𝑃 = [ ]; 𝑃1 = [ 2 ]
𝐻1 0 𝑃 0 −𝛽 𝐼

It is simple to write (32) as follows using Lemma 3:

𝑇 𝑥̅
[𝑥̅ 𝑇 𝑓 𝑇 ] ((𝐴̿𝑜 + 𝐻 ∑ 𝐺) 𝑃(𝐴̿𝑜 + 𝐻 ∑ 𝐺) + 𝑃1) [𝑓] < 0 (34)

𝑇 𝑇−1
𝐴̿𝑜 𝑃𝐴̿𝑜 + 𝐴̿𝑜 𝑃𝐻(𝜀𝐼 − 𝐻 𝑇 𝑃𝐻) 𝐻 𝑇 𝑃𝐴̿𝑜 + 𝜀𝐺 𝑇 𝐺 + 𝑃1 < 0 (35)

According to 𝐻− index criteria (‖𝐺𝑟𝑓 (𝑒 𝑗𝜃 )‖− > 0), the above inequality is written as:

𝑇 −1 𝑇
−𝐴̿𝑜 𝑃1𝐴̿𝑜 − 𝐴̿𝑜 𝑃1𝐻(𝜀𝐼 − 𝐻 𝑇 𝑃1𝐻) 𝐻 𝑇 𝑃1𝐴̿𝑜 − 𝜀𝐺 𝑇 𝐺 − 𝑃1 > 0 (36)

By applying Lemma 4, (36) becomes (37).

𝜀𝐼 − 𝐻 𝑇 𝑃𝐻 𝐻 𝑇 𝑃𝐴̿𝑜
[ 𝑇 𝑇 ]>0 (37)
𝐴̿𝑜 𝑃𝐻 −𝐴̿𝑜 𝑃𝐴̿𝑜 − 𝜀𝐺 𝑇 𝐺 − 𝑃1

Expanding matrix inequality (37)

−𝐻2 𝑇 𝐻2 − 𝐻3 𝑇 𝑃𝐻3 + 𝜀𝐼 𝐻2 𝑇 𝐶𝑜̅ + 𝐻3 𝑇 𝑃𝐴̅𝑜 𝐻2 𝑇 𝐹𝑓𝑜 + 𝐻3 𝑇 𝑃𝐸̅𝑓𝑜

𝑇 𝑇 𝑇 𝑇 𝑇 𝑇
𝐶𝑜̅ 𝐻2 + 𝐴̅𝑜 𝑃𝐻3 −𝐶𝑜̅ 𝐶𝑜̅ − 𝐴̅𝑜 𝑃𝐴̅𝑜 − 𝜀𝐺1 𝑇 𝐺1 + 𝑃 −𝐶𝑜̅ 𝐹𝑓𝑜 − 𝐴̅𝑜 𝑃𝐸̅𝑓𝑜 > 0 (38)
𝑇 𝑇 𝑇 𝑇 𝑇 𝑇
[ 𝐹𝑓𝑜 𝐻2 + 𝐸̅𝑓𝑜 𝑃𝐻3 −𝐹𝑓𝑜 𝐶𝑜̅ − 𝐸̅𝑓𝑜 𝑃𝐴̅𝑜 −𝐹𝑓𝑜 𝐹𝑓𝑜 − 𝐸̅𝑓𝑜 𝑃𝐸̅𝑓𝑜 + 𝛽2 𝐼 ]

This concludes the proof of Theorem 1's second part. FD filter gain can be calculated by solving the LMIs
(27) and (38) for the optimization problem (15).

𝐿 = 𝑃1 −1 𝑋1 (39)

3.1. Residual evaluation and threshold

In the second step of the FD process, the generated residual is evaluated using 𝑙2 signal norm and
further compared with the threshold, 𝐽𝑡ℎ > 0. The residual (3) that was generated using the proposed FD
filter can be shown as (40):

𝑟(𝑘) = 𝑟𝑑 (𝑘) + 𝑟𝑢 (𝑘) + 𝑟𝑓 (𝑘) (40)

IAES Int J Rob & Autom, Vol. 14, No. 2, June 2025: 214-226
IAES Int J Rob & Autom ISSN: 2722-2586  221

In fault-free case, 𝑟𝑓 (𝑘) = 0, then the residual evaluation function becomes 𝐽 = ‖𝑟𝑑 (𝑘) + 𝑟𝑢 (𝑘)‖22 . Thus, the
threshold can be computed as 𝐽𝑡ℎ = sup‖𝑟𝑑 (𝑘) + 𝑟𝑢 (𝑘)‖22 . In the end, the evaluated residual (𝐽) is compared
with the threshold (𝐽𝑡ℎ ) and the fault alarm is released when the following condition is satisfied:

𝐽 > 𝐽𝑡ℎ ; fault alarm

𝐽 ≤ 𝐽𝑡ℎ ; fault-free

4. APPLICATION TO A THREE-TANK SYSTEM

A three-tank system application is used in this study. The system is often used to illustrate the
principles of process control, system dynamics, and fault detection. In such a system, the liquid levels in the
tanks and the flow rates between them are managed using sensors, actuators, and controllers. Automation
plays a critical role in this setup by ensuring the precise regulation of these variables to achieve desired
outcomes, such as maintaining specific liquid levels or flow rates. Using advanced automation technologies,
such as programmable logic controllers (PLCs) and distributed control systems (DCS), the three-tank system
can operate autonomously, adjusting valves and pumps based on real-time feedback from level and flow
sensors. This level of automation improves accuracy, reduces manual intervention, and ensures consistent
operation even in complex scenarios. Moreover, integrating fault detection algorithms into the system
enhances reliability by identifying anomalies like sensor malfunctions, leaks, or blockages, enabling
proactive maintenance. Thus, the automation of a three-tank system serves as a foundational model for
understanding and implementing control strategies in larger industrial processes such as chemical
manufacturing, water treatment, and oil refining.
This section presents simulation results that demonstrate the effectiveness of the proposed FD
method. For simulation purposes, abrupt and intermittent faults are introduced in the sensors and actuators of
the advanced three-tank system illustrated in Figure 1. Such faults significantly degrade system performance
and are included in the study to evaluate the capability of the proposed method in identifying these critical
issues. Modeling errors from the system linearization process are incorporated as norm-bounded model
uncertainties. The behavior of the three-tank system is described by the following set of nonlinear equations,
which capture its dynamics.

𝐴ℎ̇1 = 𝑄1 − 𝑄13
𝐴ℎ̇2 = 𝑄2 + 𝑄32 − 𝑄20
𝐴ℎ̇1 = 𝑄13 − 𝑄32 (41)

with
𝑄13 = 𝑎1 𝑠13 𝑠𝑔𝑛(ℎ1 − ℎ3 )√2𝑔|ℎ1 − ℎ3 |
𝑄32 = 𝑎3 𝑠23 𝑠𝑔𝑛(ℎ3 − ℎ2 )√2𝑔|ℎ3 − ℎ2 |
𝑄20 = 𝑎2 𝑠0 √2𝑔ℎ2

Figure 1. A three-tank system [33]

Design of H-/H∞ based fault detection filter for linear uncertain systems … (Masood Ahmad)
222  ISSN: 2722-2586

The process outputs y(k), represented by ℎ3 , ℎ2 , ℎ1 , corresponding to the water levels in the
respective tanks. The process inputs u(k) are denoted by Q1 and Q2, while Qij represents the flow rate of water
from tank i-th to tank j-th. Additionally, s13 and s23 refer to the cross-sectional areas of the pipes connecting
the respective tanks. The cross-sectional area of the pipe connected to Tank 2 is 𝑠0 . sgn denotes the signum
function, which is defined as

−1 𝑖𝑓𝑥 < 0
sgn(𝑥) = { 0 𝑖𝑓𝑥 = 0
1 𝑖𝑓𝑥 > 0
𝑠13 = 𝑠23 = 𝑠0 = 𝑠𝑛

The system's primary parameters and coefficients are shown in Table 1. In the three-tank system, an
unknown disturbance arises from water falling into the tanks from the pumps. Additionally, the sensors used
to measure water levels introduce noise measurement. For fault detection (FD), a linear model of the system
is derived by applying Taylor series expansion and linearizing the dynamics around the equilibrium or
operating point. This process results in a linear nominal model of the discrete-time system in state-space
form, as shown in (1). The linearization is performed at the operating points ℎ1 = 45cm, ℎ2 = 15cm and
ℎ3 = 30cm. Nominal matrices are obtained after linearizing the nonlinear model of the system.

0.9915 0 0.0084 0.0065 0.0008 0.25 0 0

𝐴=[ 0 0.9807 0.0082]; 𝐵 = [0.0008 0.0065]; 𝐸𝑑 = [ 0 0.25 0 ];
0.0084 0.0082 0.9833 0 0 0 0 0.25
𝐶 = diag[1,1,1]; 𝐷 = 0; 𝐸𝑓 = 𝐵; 𝐹𝑑 = 𝐹𝑓 = diag[1,1,1]

The linearization process incorporates modelling errors known as norm-bounded model uncertainty into the
system matrices, which are denoted as:

−0.01 0 0 0.01 0 0.015 0.01 0 0

𝐻1 = 𝐻2 = [ 0 −0.01 0 ] ; 𝐺1 = [ 0 0.01 0.015];𝐺2 = [ 0 0.02 0]
0 0 −0.01 0.01 0.01 0.05 0 0 0

Table 1. Three-tank system's parameters [33]

Parameters Value Unit
𝐴 154 cm2
𝑠𝑛 0.5 cm2
𝐻𝑚𝑎𝑥 62 cm
𝑄1𝑚𝑎𝑥 100 cm3/sec
𝑄2𝑚𝑎𝑥 100 cm3/sec
𝑎1 0.46
𝑎2 0.60
𝑎3 0.45

The uncertain parameter (∑ = diag[0.9597,0.9597,0.9597]) is chosen at random and unknown

disturbance, 𝑑(𝑘) ∈ [−0.01,0.01] is used for simulations. The pump inflows are assumed to be constant with
specified values of 𝑄1 = 25.6 cm3/sec and 𝑄2 = 39.5 cm3/sec. After solving the linear matrix inequalities in
(13) and (14), a disturbance attenuation level of γ = 1.026 and a fault sensitivity level of β = 1.9891 are
achieved. The corresponding filter gain matrix is computed using (39) and is given as

0.3265 −0.0001 −0.0017

𝐿=[ ∗ 6.7344 −0.0017]
∗ ∗ 6.7265

Furthermore, the residual evaluation function is computed using 𝑙2 norm of the residual (40) and the
sup
threshold is computed as 𝐽𝑡ℎ|𝑓=0 = 𝑢̅(𝑘)∈𝑙2 ‖𝑟(𝑘)‖2 = 0.03. The residual in the sensor/actuator fault-free case
is shown in Figure 2.
Figure 3 displays the impact of an abrupt sensor fault in Tank 1 on the residual. A fault with a 10 cm
offset is introduced to the sensor input of Tank 1 at t = 80 sec. The simulation results confirm that the fault is
successfully detected. Comparable responses and successful detections are also observed for faults in the

IAES Int J Rob & Autom, Vol. 14, No. 2, June 2025: 214-226
IAES Int J Rob & Autom ISSN: 2722-2586  223

other sensors. As shown in Figure 3, the evaluation function remains below the threshold prior to the fault
occurrence but exceeds the threshold when the sensor fault occurs at t = 80 sec. Similarly, Figure 4
demonstrates the response when an intermittent fault is applied to the actuator of Pump 1. The results
highlight that the fault detection filter effectively identified faults in the discrete-time system, even in the
presence of unknown disturbances and model uncertainty.

Figure 2. Residual in a fault-free case

Figure 3. An abrupt sensor FD in Tank 1

Remark: In section 3, two linear matrix inequalities (LMIs) are derived for fault detection (FD) in linear
uncertain systems. An H- index-based fault-sensitive filter is designed to improve the fault sensitivity of the
residual. However, this filter also exhibits sensitivity to disturbances and model uncertainties. In contrast, the
H∞ FD filter ensures disturbance attenuation but also provides robustness against faults. To address these
challenges, a multi-objective H-/H∞ based FD filter is proposed, which simultaneously offers robustness to
disturbances and model uncertainties, as well as sensitivity to faults. Rather than maximizing β and
minimizing γ separately, the performance index, β/γ, is maximized in this design. It is important to note that
the residual generated by the H-/H∞ based filter may be less sensitive than that produced by the H- index-
based fault-sensitive filter. Similarly, the residual from the H-/H∞ based filter might be less robust to
disturbances and model uncertainties compared to the residual from the H∞ based filter. Nevertheless, the
proposed H-/H∞ based FD filter is advantageous as it achieves both disturbance attenuation and fault
sensitivity simultaneously.

Design of H-/H∞ based fault detection filter for linear uncertain systems … (Masood Ahmad)
224  ISSN: 2722-2586

Figure 4. Simulated fault (top); Evaluated residual and threshold in actuator fault (bottom)

5. CONCLUSION
Robotics and automation systems rely on advanced fault detection mechanisms to ensure seamless
operation, prevent downtime, and maintain safety standards. This paper addresses the fault detection problem
for discrete-time linear systems subjected to deterministic disturbances and norm-bounded model uncertainty.
The proposed FD filter generates a residual that simultaneously attenuates the effect of model uncertainty and
disturbance and enhances the sensitivity to faults. The filter optimizes the H-/H∞ performance index,
ensuring the best trade-off between robustness and sensitivity in all directions of the residual space. A
solution to the problem is formulated using linear matrix inequality constraints. Successful actuator and
sensor fault detection results of the three-tank system obtained from simulation validate the effectiveness of
the proposed observer-based FD system. By identifying and addressing the faults, these systems can enhance
reliability and efficiency across various industrial applications.

FUNDING INFORMATION
This research is funded by the Malaysia Fundamental Research Grant Scheme
(FRGS/1/2019/TK04/USM/02/12) and the Higher Education Commission (HEC) of Pakistan through the
HRDI-UESTPS scholarship program.

AUTHOR CONTRIBUTIONS STATEMENT

This journal uses the Contributor Roles Taxonomy (CRediT) to recognize individual author
contributions, reduce authorship disputes, and facilitate collaboration.

Name of Author C M So Va Fo I R D O E Vi Su P Fu
Masood Ahmad ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Rosmiwati Mohd- ✓ ✓ ✓ ✓ ✓ ✓
Mokhtar

C : Conceptualization I : Investigation Vi : Visualization

M : Methodology R : Resources Su : Supervision
So : Software D : Data Curation P : Project administration
Va : Validation O : Writing - Original Draft Fu : Funding acquisition
Fo : Formal analysis E : Writing - Review & Editing

CONFLICT OF INTEREST STATEMENT

Authors state no conflict of interest

IAES Int J Rob & Autom, Vol. 14, No. 2, June 2025: 214-226
IAES Int J Rob & Autom ISSN: 2722-2586  225

DATA AVAILABILITY
Data availability is not applicable to this paper as no new data were created in this study.

REFERENCES
[1] J. Zhang et al., “A fault detection and isolation method for independent metering control system based on adaptive robust
observer,” Measurement, vol. 241, p. 115707, Feb. 2025, doi: 10.1016/j.measurement.2024.115707.
[2] W. Lu, Y. Li, Y. Cheng, D. Meng, B. Liang, and P. Zhou, “Early Fault detection approach with deep architectures,” IEEE
Transactions on Instrumentation and Measurement, vol. 67, no. 7, pp. 1679–1689, Jul. 2018, doi: 10.1109/TIM.2018.2800978.
[3] J. Wang, X. Sun, C. Zhang, and X. Ma, “An integrated methodology for system-level early fault detection and isolation,” Expert
Systems with Applications, vol. 201, p. 117080, Sep. 2022, doi: 10.1016/j.eswa.2022.117080.
[4] S. Yin, T. Peng, C. Yang, C. Yang, Z. Chen, and W. Gui, “Dynamic hybrid observer-based early slipping fault detection for high-
speed train wheelsets,” Control Engineering Practice, vol. 142, p. 105736, Jan. 2024, doi: 10.1016/j.conengprac.2023.105736.
[5] K. R. Palepogu and S. Mahapatra, “Design of sliding mode control with state varying gains for a Benchmark twin rotor MIMO
system in horizontal motion,” European Journal of Control, vol. 75, p. 100909, Jan. 2024, doi: 10.1016/j.ejcon.2023.100909.
[6] S. X. Ding, “Advanced methods for fault diagnosis and fault-tolerant control,” Advanced Methods for Fault Diagnosis and Fault-
Tolerant Control, pp. 1–658, 2020, doi: 10.1007/978-3-662-62004-5.
[7] Z. Li, Y. Yang, L. Li, H. Ma, and X. Chen, “Model-free fault detection based on performance residual for feedback control
systems,” IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 69, no. 11, pp. 4453–4457, Nov. 2022, doi:
10.1109/TCSII.2022.3182655.
[8] M. A. Ziane, C. Join, M. C. Pera, N. Y. Steiner, M. Benne, and C. Damour, “A new method for fault detection in a free model
context,” IFAC-PapersOnLine, vol. 55, no. 6, pp. 55–60, 2022, doi: 10.1016/j.ifacol.2022.07.105.
[9] B. Taghezouit, F. Harrou, Y. Sun, and W. Merrouche, “Model-based fault detection in photovoltaic systems: A comprehensive
review and avenues for enhancement,” Results in Engineering, vol. 21, p. 101835, Mar. 2024, doi: 10.1016/j.rineng.2024.101835.
[10] E. Bernardi and E. J. Adam, “Observer-based fault detection and diagnosis strategy for industrial processes,” Journal of the
Franklin Institute, vol. 357, no. 14, pp. 10054–10081, Sep. 2020, doi: 10.1016/j.jfranklin.2020.07.046.
[11] S. A. Herve, Y. Tamtsia Aurelien, and N. Nneme Leandre, “Performance evaluation of industrial ethernet protocols for real-time
fault detection based adaptive observer in networked control systems with network communication constraints,” IAES
International Journal of Robotics and Automation (IJRA), vol. 10, no. 3, p. 261, Sep. 2021, doi: 10.11591/ijra.v10i3.pp261-274.
[12] M. Ahmad and R. M. Mokhtar, “A survey on model based fault detection techniques for linear time invariant systems with
numerical analysi,” Pertanika journal of science and technology, vol. 30, no. 1, pp. 53–78, 2022.
[13] M. Dihya, M. Moufid, B. Chemseddine, and B. Moussaab, “Switched time delay control based on neural network for fault
detection and compensation in robot,” IAES International Journal of Robotics and Automation (IJRA), vol. 10, no. 2, pp. 91–103,
Jun. 2021, doi: 10.11591/ijra.v10i2.pp91-103.
[14] M. Mazare, M. Taghizadeh, P. Ghaf-Ghanbari, and E. Davoodi, “Robust fault detection and adaptive fixed-time fault-tolerant
control for quadrotor UAVs,” Robotics and Autonomous Systems, vol. 179, p. 104747, Sep. 2024, doi:
10.1016/j.robot.2024.104747.
[15] S. Ali, Y. Jiang, H. Luo, M. T. Raza, S. Faisal, and F. Shahid, “Robust fault detection scheme for asynchronous switched systems
via sliding mode observer,” International Journal of Control, Automation and Systems, vol. 22, no. 4, pp. 1186–1200, Apr. 2024,
doi: 10.1007/s12555-023-0121-4.
[16] C. Shang, S. X. Ding, and H. Ye, “Distributionally robust fault detection design and assessment for dynamical systems,”
Automatica, vol. 125, p. 109434, Mar. 2021, doi: 10.1016/j.automatica.2020.109434.
[17] M. Hashemi, M. Stolz, and D. Watzenig, “Super-twisting algorithm-based sliding mode observer for open-circuit fault diagnosis
in PWM voltage source inverter in an in-wheel motor drive system,” in 2023 IEEE International Conference on Mechatronics
(ICM), Mar. 2023, pp. 1–6, doi: 10.1109/ICM54990.2023.10102000.
[18] S. X. Ding, “Model-based fault diagnosis techniques: Design schemes, algorithms, and tools,” Model-based Fault Diagnosis
Techniques: Design Schemes, Algorithms, and Tools, pp. 1–473, 2008, doi: 10.1007/978-3-540-76304-8.
[19] R. Chintala, J. Winkler, S. Ramaraj, and X. Jin, “Sensitivity analysis of an automated fault detection algorithm for residential air-
conditioning systems,” Applied Thermal Engineering, vol. 238, 2024, doi: 10.1016/j.applthermaleng.2023.121895.
[20] G. Floreale, P. Baraldi, X. Lu, P. Rossetti, and E. Zio, “Sensitivity analysis by differential importance measure for unsupervised
fault diagnostics,” Reliability Engineering and System Safety, vol. 243, 2024, doi: 10.1016/j.ress.2023.109846.
[21] J. Liu, J. L. Wang, and G. H. Yang, “An LMI approach to minimum sensitivity analysis with application to fault detection,”
Automatica, vol. 41, no. 11, pp. 1995–2004, 2005, doi: 10.1016/j.automatica.2005.06.005.
[22] W.-H. Chen, J. Yang, L. Guo, and S. Li, “Disturbance-observer-based control and related methods—an overview,” IEEE
Transactions on Industrial Electronics, vol. 63, no. 2, pp. 1083–1095, Feb. 2016, doi: 10.1109/TIE.2015.2478397.
[23] Z. Gao, C. Cecati, and S. X. Ding, “A survey of fault diagnosis and fault-tolerant techniques-part I: Fault diagnosis with model-
based and signal-based approaches,” IEEE Transactions on Industrial Electronics, vol. 62, no. 6, pp. 3757–3767, Jun. 2015, doi:
10.1109/TIE.2015.2417501.
[24] P. Thomas, “Fault detection and diagnosis in engineering systems,” Control Engineering Practice, vol. 10, no. 9, pp. 1037–1038,
2002, doi: 10.1016/s0967-0661(02)00064-3.
[25] C. Ding, M. Zhao, J. Lin, and J. Jiao, “Multi-objective iterative optimization algorithm based optimal wavelet filter selection for
multi-fault diagnosis of rolling element bearings,” ISA Transactions, vol. 88, pp. 199–215, May 2019, doi:
10.1016/j.isatra.2018.12.010.
[26] M. Ahmad, F. Azeem, and H. A. Zidan, “Robust residual generator design for sensor fault detection in twin rotor aerodynamic
system,” e-Prime - Advances in Electrical Engineering, Electronics and Energy, vol. 8, p. 100620, Jun. 2024, doi:
10.1016/j.prime.2024.100620.
[27] W. Li, Z. Zhu, G. Zhou, and G. Chen, “Optimal H i / H ∞ fault‐detection filter design for uncertain linear time‐invariant systems:
an iterative linear matrix inequality approach,” IET Control Theory & Applications, vol. 7, no. 8, pp. 1160–1167, May 2013, doi:
10.1049/iet-cta.2012.0954.
[28] M. Ahmad and R. Mohd-Mokhtar, “Fault detection full order filter apply to discrete time-invariant linear system,” Journal of
Engineering Science and Technology, vol. 16, no. 5, pp. 4221–4234, 2021.
[29] F. Xu, J. Tan, X. Wang, V. Puig, B. Liang, and B. Yuan, “Mixed active/passive robust fault detection and isolation using set-
theoretic unknown input observers,” IEEE Transactions on Automation Science and Engineering, vol. 15, no. 2, pp. 863–871,

Design of H-/H∞ based fault detection filter for linear uncertain systems … (Masood Ahmad)
226  ISSN: 2722-2586

Apr. 2018, doi: 10.1109/TASE.2017.2776998.

[30] J. Chen and R. J. Patton, “Standard H ∞ filtering formulation of robust fault detection,” IFAC Proceedings Volumes, vol. 33, no.
11, pp. 261–266, Jun. 2000, doi: 10.1016/S1474-6670(17)37370-6.
[31] J. L. Wang, G. H. Yang, and J. Liu, “An LMI approach to H- index and mixed H- / H∞ fault detection observer design,”
Automatica, vol. 43, no. 9, pp. 1656–1665, 2007, doi: 10.1016/j.automatica.2007.02.019.
[32] S. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan, Linear matrix inequalities in system and control theory. Society for
Industrial and Applied Mathematics, 1994.
[33] J. Chen and R. J. Patton, “Robust model-based fault diagnosis for dynamic systems,” Asian studies in computer and information
science, vol. 11, no. 14, p. 356, 1999, [Online]. Available: https://link.springer.com/content/pdf/10.1007%2F978-1-4615-5149-
2.pdf%0Ahttp://portal.acm.org/citation.cfm?id=316586&amp;dl=.

BIOGRAPHIES OF AUTHORS

Masood Ahmad is an assistant professor at the Electrical and Computer

Engineering Department (ECE) at COMSATS University Islamabad (CUI), Lahore
campus. He has been associated with the ECE department since 2012. He completed his
PhD in Electrical Engineering with a specialization in Control and Automation from
Universiti Sains Malaysia (USM), Malaysia, in June 2022, under the HEC Faculty
Development Program for UETs/UESTPs. During his PhD studies, he worked on model-
based observer design for fault detection. He has massive experience in control and
electronics and is engaged in various research activities. His areas of interest include
observer and controller design for LTI systems, optimal control, robust control, and
implementing advanced control techniques in several practical applications. He can be
contacted at [email protected].

Rosmiwati Mohd-Mokhtar received her B.Eng. degree with honours and an

M.Sc. in Electrical and Electronic Engineering from Universiti Sains Malaysia in 1999 and
2002, respectively. She received her Ph.D. in Electrical Engineering from the Royal
Melbourne Institute of Technology University, Australia, in 2008. She joined the School of
Electrical and Electronic Engineering, Universiti Sains Malaysia, in October 2008. She is
an Associate Professor at the School of Electrical and Electronic Engineering, USM. She is
also a Chartered Engineer (C.Eng.) of the Institution of Engineering and Technology (IET),
a Professional Engineer (P.Eng.) of the Board of Engineers Malaysia (BEM), a Senior
Member of the Institution of Electrical and Electronics Engineers (IEEE), and a member to
Institution of Engineers Malaysia (IEM). Her research interests include system
identification, advanced control system design, process modeling, process optimization,
mechatronics applications, and underwater system applications. She can be contacted at
[email protected].

IAES Int J Rob & Autom, Vol. 14, No. 2, June 2025: 214-226