Short-term memory artificial neural network modelling to predict concrete corrosion in wastewater treatment plant inlet chambers using sulphide sensors

Sulphide accumulation in lengthy rising mains can lead to significant concrete corrosion and odour issues at manholes and wastewater treatment plants (WWTPs). Monitoring dissolved sulphide, typically relies on auto-sampling or grab samples followed by laboratory analysis, remains underdeveloped. This study aimed to identify sources of concrete corrosion sources at a WWTP inlet chamber and develop a sulphide prediction model using artificial intelligence (AI). A dissolved sulphide sensor was installed at three rising mains (RM1 to RM3) and the combined inlet at a full-scale WWTP, providing a 5-minute resolution data that revealed a daily hydrogen sulphide (H2S) pattern that inversely correlated with the flow rate. RM1 exhibited the highest sulphide load, peaking at 3.6 kg/d during cold months and 4.2 kg/d during warm months. RM3 and RM2 recorded loads of 2.96 kg/d and 0.98 kg/d, respectively, during cold months. A long short-term memory (LSTM) artificial neural network (ANN) model was developed to predict H₂S concentrations at RM1, using flow rate, temperature, and time of day as inputs. The model achieved a root mean square error (RMSE) of 0.34 and a Nash-Sutcliffe efficiency (NSE) of 0.57, accurately predicting the daily H2S pattern. This study's main contributions include insights into sulphide dynamics from high-resolution sensor data, which could support corrosion management as part of a septicity warning system or feedforward control for sulphide treatment. Additionally, the AI-based prediction model offers potential for sensor repurposing, saving both capital and operational costs.