Artificial Neural Network-Based Fault-Tolerant Control Strategy for Cascaded Multilevel Inverters with Auxiliary Bridge Integration
- lms editor
- 2 days ago
- 6 min read
Abstract
This research presents a robust fault-tolerant control (FTC) strategy for a 15-level Cascaded H-Bridge Multilevel Inverter (CHMI) by integrating an Artificial Neural Network (ANN) for high-speed fault detection together with an auxiliary H-bridge to provide hardware redundancy. Cascaded multilevel inverter topologies are inherently susceptible to component-level failures including semiconductor switch malfunctions, depletion of DC sources, or failure of an entire module. Such faults disturb output voltage symmetry and increase harmonic distortion.
The proposed method employs an ANN model trained using an eight-dimensional input vector consisting of individual module voltages and the load voltage to classify faults with high accuracy. Once a failure in any of the seven primary modules is detected, a multiport control logic enables seamless substitution of the faulty module by activating the auxiliary bridge through a coordinated bypass mechanism.
Simulation studies performed in the MATLAB/Simulink environment demonstrate that the proposed strategy preserves the 15-level staircase output waveform even under various fault conditions. Consequently, rated output voltage and acceptable power quality are maintained, thereby ensuring load reliability and improving the operational robustness of the inverter system.
Keywords
Cascaded Multilevel Inverter (CHMI), Fault-Tolerant Control, Artificial Neural Network (ANN), MATLAB/Simulink, Total Harmonic Distortion (THD), Power Reliability.
2. Introduction
The demand for highly reliable power electronic systems has increased significantly in modern energy infrastructures, where unexpected downtime can lead to severe operational and economic consequences. Multilevel inverter technologies, particularly the Cascaded H-Bridge (CHB) inverter, are widely utilized in medium- and high-power applications due to their modular architecture and capability to produce high-quality voltage waveforms.
Despite these advantages, the increased number of semiconductor devices and isolated DC sources in CHB topologies raises the probability of component-level failures. These failures may occur in the form of switch open-circuit faults, short-circuit faults, or depletion of DC sources. When such faults occur, conventional CHB inverters are unable to maintain the required voltage levels, leading to a reduction in output voltage resolution and an increase in Total Harmonic Distortion (THD).
A reduction in voltage levels often violates power quality standards such as IEEE 519 and can adversely affect sensitive loads connected to the inverter. Therefore, implementing fault-tolerant control strategies is essential for maintaining system reliability.
The present research addresses these limitations by introducing an intelligent fault detection mechanism based on Artificial Neural Networks (ANN) combined with an auxiliary hardware redundancy approach. By integrating the ANN-based detection layer with a redundant H-bridge module, the system maintains the full 15-level output voltage structure even in the presence of hardware faults. This capability significantly enhances the reliability and operational continuity of multilevel inverter systems used in critical power conversion applications.
3. Proposed System Configuration and Topology
The proposed inverter architecture is a 15-level Cascaded H-Bridge Multilevel Inverter (CHMI) consisting of seven primary H-bridge modules connected in series. An additional auxiliary H-bridge is incorporated to provide hardware redundancy during fault conditions.
Each H-bridge module contains an isolated DC source and four switching devices that generate three possible voltage levels:
where
· – output voltage of the H-bridge
· – DC source voltage
The total output voltage of the cascaded inverter is obtained by summing the contributions of all H-bridge modules:
Under normal operating conditions, the seven H-bridge modules collectively synthesize a 15-level staircase output waveform.
Auxiliary Integration and Bypass Logic
The integration of the auxiliary module is governed by bypass switching variables and . These switches determine whether the auxiliary bridge remains isolated or inserted into the power path.
Normal Operating State
In this condition, the auxiliary bridge is bypassed and electrically isolated from the system. The seven healthy modules generate the required output voltage.
Fault-Tolerant Operating State
When a fault is detected in any primary H-bridge module, the switching states are reversed. The auxiliary module is inserted into the inverter string to replace the faulty module.
The switching transition must be carefully synchronized to prevent voltage spikes and current discontinuities during the module substitution process.
4. Neural Network-Based Fault Detection and Control Strategy
Traditional threshold-based detection methods often exhibit limited accuracy under varying load and operating conditions. Artificial Neural Networks (ANN) provide a powerful alternative because they can learn nonlinear relationships and identify complex fault patterns in multilevel inverter systems.
Data Acquisition and Mathematical Representation
The ANN receives an eight-dimensional input vector representing the voltage states of the inverter modules and load voltage.
where
· represent the voltages of the seven H-bridge modules
· represents the output load voltage
The ANN performs a classification operation defined by
where
Interpretation of output:
· → Normal operating condition
· → Failure detected in the H-bridge module
Training Specification and Generalization
The neural network is trained using the MATLAB Neural Network Fitting App, employing the backpropagation learning algorithm to adjust network weights.
Initially, the dataset contained 16 samples representing various operational conditions. To enhance generalization capability and avoid overfitting, the dataset was expanded to 160 samples by introducing additional operating scenarios.
The regression performance of the trained network is expressed as
which indicates a perfect correlation between predicted and target outputs during training.
Multiport Switching Logic Transformation
The ANN output is utilized as the selection input for a 7-to-1 multiport selector (multiplexer).
If
then the gating signals originally intended for the faulty module are redirected to the auxiliary bridge.
This dynamic switching ensures that the auxiliary bridge reproduces the switching pattern of the failed module without recalculating PWM signals or modulation indices.
5. Simulation Framework and Parameters
The proposed fault-tolerant control system is evaluated through detailed simulation using the MATLAB/Simulink environment.
Fault conditions are artificially introduced by interrupting the DC source connection of a specific module using Ideal Switch blocks controlled by step signals.
A binary logic variable is used to represent module health:
This configuration allows controlled injection of faults and observation of the system response.
Table 1
Simulation Parameters
Parameter | Specification |
Inverter Topology | 15-Level Cascaded H-Bridge (CHB) |
Primary Module Count | 7 H-Bridges (H₁–H₇) |
Redundancy Configuration | 1 Auxiliary H-Bridge with Bypass Logic |
Input Vector Dimension | (7 Bridge Voltages + Load Voltage) |
Detection Mechanism | ANN-Based Multiport Selector |
ANN Training Dataset | 160 Samples |
Training Performance | |
Fault Injection Method | Constant Blocks with Ideal Switch Interruption |
6. Results and Performance Evaluation
The effectiveness of the proposed fault-tolerant control strategy is evaluated by examining the inverter output waveform under normal operation and fault conditions.
Maintaining the 15-level staircase waveform is the primary performance criterion, as it ensures rated voltage delivery and acceptable harmonic distortion levels.
Scenario Analysis
Case 1: Normal Operation
Under normal conditions, the seven primary H-bridge modules generate a symmetric 15-level staircase output waveform. The auxiliary bridge remains bypassed.
Case 2: Fault Condition (Without FTC)
When a module failure occurs (for example in ), the output voltage levels reduce from 15 to 13. This reduction causes distortion in the output waveform and decreases the fundamental voltage magnitude.
Case 3: Fault-Tolerant Operation (ANN Activated)
When the ANN detects a fault, the corresponding module index is identified immediately. The control logic then performs the following actions:
1. Switch transition of and
2. Activation of the auxiliary bridge
3. Routing of PWM signals through the multiport selector
As a result, the auxiliary module reproduces the switching behavior of the failed bridge and restores the 15-level waveform.
Performance Impact
Simulation results confirm that the ANN-based detection and auxiliary bridge activation occur almost instantaneously. The inverter output waveform quickly regains its original staircase structure, preventing excessive harmonic distortion and maintaining the rated voltage at the load.
7. Conclusion and Future Scope
This research has presented an ANN-based fault-tolerant control strategy for a 15-level Cascaded H-Bridge Multilevel Inverter. By combining intelligent software-based detection with hardware redundancy, the system effectively addresses the reliability challenges associated with multilevel inverter structures.
Key Findings
1. Detection Accuracy
The ANN model achieved a regression value of
indicating perfect classification of faults among seven inverter modules.
2. Operational Continuity
The coordinated bypass switching logic
along with multiport signal routing ensures smooth transition to the redundant auxiliary bridge.
3. Waveform Integrity
The inverter consistently maintains a 15-level staircase waveform, preserving the rated output voltage and minimizing harmonic distortion.
Future Scope
Future investigations will extend the proposed strategy to handle multiple simultaneous module failures, which may occur in large-scale inverter systems. Additionally, the implementation of the control strategy in a Real-Time Hardware-in-the-Loop (HIL) platform will enable validation under practical switching frequencies and industrial operating conditions. Such studies will further enhance the robustness and practical applicability of fault-tolerant multilevel inverter systems.
VII. YouTube Video
VIII. Purchase link of the Model
SKU: 0232
Comments