Automated active fault detection in fouled dissolved oxygen sensors

Area of Science:

• Environmental Engineering
• Sensor Technology
• Wastewater Treatment

Background:

• Biofilm formation on dissolved oxygen (DO) sensors introduces bias, compromising automatic control and efficiency in wastewater treatment.
• Accurate DO monitoring is essential for balancing energy consumption and treatment effectiveness.

Purpose of the Study:

• To evaluate the automatic interpretation of a perturbed dataset for detecting biofilm-induced bias in DO sensors.
• To assess the accessibility of fault detection (FD) methods for operators through automated training and tuning.

Main Methods:

• Utilized a challenging setup with realistic conditions for full-scale application.
• Implemented automated training for adapting to changing normal conditions and automated tuning for alarm thresholds.
• Applied fault detection methods to a dataset generated with deliberate perturbations.

Main Results:

• Automatic usage of FD methods proved difficult, particularly automatic tuning of alarm thresholds with limited training data.
• Two FD methods successfully detected DO sensor bias caused by biofilm formation down to 0.5 mg DO/L.
• Simpler FD methods interpreted the data as effectively as advanced machine learning algorithms.

Conclusions:

• Actively generated data quality is more critical than the complexity of interpretation algorithms for detecting biofilm bias.
• Fault detection methods can be made accessible to operators, improving the reliability of DO sensor measurements.
• Effective detection of biofilm-induced bias enhances the efficiency and control of wastewater treatment processes.