A Data-Driven Architecture for Sensor Validation Based on Neural Networks

Question

1. What have the authors contributed in "A data-driven architecture for sensor validation based on neural networks" ?

2. What are the future works in "A data-driven architecture for sensor validation based on neural networks" ?

3. What is the purpose of the proposed SFDIA system?

4. How many MLP virtual sensors are used to estimate the NU?

Accepted Answer

In this paper, the authors propose a novel sensor validation architecture, which performs sensor fault detection, isolation and accommodation ( SFDIA ).. More specifically, a machinelearning based architecture is presented to detect faults in sensors measurements within the system, identify the faulty ones and replace them with estimated values.

Accepted Answer

Future directions will include the use of deep networks for the modules of the proposed SFDIA and type-of-fault classification.

Accepted Answer

Sensors measurements constitute the input of the proposed SFDIA system, where measurements are divided into twosets: NR reliable sensors (set SR), which represent supportive data, and NU unreliable sensors (set SU ), which are prone to failure.

Accepted Answer

Five MLP virtual sensors (estimators) with one single hidden layer (made of 10 neurons) are trained to provide estimation of the NU = 5 unreliable metal oxide chemical sensors.

Accepted Answer

MLP-based virtual sensors provided appropriate estimates of unreliable sensors to replace corresponding corupted measurements in presence of faults, while an MLP-based classifier was responsible for detection and isolation of faults.

Accepted Answer

The classifier softmax output gives a decision vector representing the probability distributions of the vector of potential outcomes.

Accepted Answer

The proposed architecture is validated by rea-world data from AQ monitoring sensors, and results illustrate the prompt detection, isolation and accommodation of sensors’ failures with less than 2.6% of faults on average remained undetected on the test set.

Accepted Answer

Each virtual sensor receives all sensors’ data except for the sensor under estimation from time instant n to n−m (i.e. using a sliding window of length m + 1) as input and estimates the measurement of the sensor under estimation at time n as output.

Accepted Answer

In detail, the classification stage accepts the error vectors inputs at time instants n to n − k, namely e(n), . . . , e(n − k) (i.e. a sliding window of length k + 1).

Accepted Answer

if an unreliable sensor is declared in failure, its measurements are replaced with the estimated values from the corresponding virtual sensor.

Accepted Answer

As can be seen in Fig. 3(b), after fault identification, system accommodates isolated faulty data with its estimation to ensure the fault-free performance of the system.

Accepted Answer

Hourly averaged measurements of the multi-sensor device consisting of carbon monoxide (CO), Non-Metanic Hydrocarbons (NMH), Nitrogen Oxides (NOx), Nitrogen Dioxide (NO2) and ozone (O3) gas concentrations are considered as unreliable set.

Accepted Answer

In the case im 6= 0, the specific values of im performs the isolation task and replacing faulty sensor measurements with corresponding virtual sensor estimates employs the accommodation task.

Accepted Answer

3. More specifically, Fig. 3(a) monitors the faults on different sensors where the proposed architecture successfully detects and identifies all faults without delay in the system (dashed line) with only two false declaration samples (false positive) in the first and fifth days.

Accepted Answer

All classes show high precision over 90%, with the lowest precision exhibited on O3 and NOx sensors with values 93.75% and 93.64%, respectively.

Accepted Answer

the size of the sliding window is assumed to span 10 samples for both the estimators and the classifier (i.e. m = 10 and k = 10).

A Data-Driven Architecture for Sensor Validation Based on Neural Networks

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What have the authors contributed in "A data-driven architecture for sensor validation based on neural networks" ?

2. What are the future works in "A data-driven architecture for sensor validation based on neural networks" ?

3. What is the purpose of the proposed SFDIA system?

4. How many MLP virtual sensors are used to estimate the NU?

5. What is the purpose of this paper?

6. What is the definition of the classifier?

7. how many faults did the proposed architecture detect?

8. What is the purpose of the proposed architecture?

9. What is the classification stage of the proposed SFDIA system?

10. What is the function of the classifier?

11. What is the purpose of the paper?

12. What is the definition of a reliable set?

13. What is the way to replace faulty sensor measurements with corresponding virtual ones?

14. How many false declaration samples are there in the first and fifth days?

15. What is the description of the SFDIA architecture?

16. How many layers are used to estimate the NU?

Figures

Citations

Sensor-Fault Detection, Isolation and Accommodation for Digital Twins via Modular Data-Driven Architecture

A Data-Driven Soft Sensor to Forecast Energy Consumption in Wastewater Treatment Plants: A Case Study

A Machine-Learning Architecture for Sensor Fault Detection, Isolation, and Accommodation in Digital Twins

Wasserstein GAN-Based Digital Twin-Inspired Model for Early Drift Fault Detection in Wireless Sensor Networks

Data‐driven sensor fault detection and isolation of nonlinear systems: Deep neural‐network Koopman operator

References

Sediment Transport and Movable Bed s

On field calibration of an electronic nose for benzene estimation in an urban pollution monitoring scenario

Cyber Security of Water SCADA Systems—Part I: Analysis and Experimentation of Stealthy Deception Attacks

Sensor Fault Detection, Isolation, and Identification Using Multiple-Model-Based Hybrid Kalman Filter for Gas Turbine Engines

Fault detection using multiscale PCA-based moving window GLRT

Related Papers (5)

Sensor-Fault Detection, Isolation and Accommodation for Digital Twins via Modular Data-Driven Architecture

A hybrid fault detection approach for context-aware wireless sensor networks

A Sensor Fault Detection Algorithm for Structural Health Monitoring Using Adaptive Differential Evolution

Fault Detection in Wireless Sensor Networks: A Machine Learning Approach

A Machine Learning Approach for Identifying and Classifying Faults in Wireless Sensor Network