Optimization of the jar testing process with the internetof things and machine learning

Although 71% of the Earth's surface is covered by water, only 2.5% of it is freshwater. According to the World Health Organization, approximately 844 million people do not have access to a safe drinking water source, while 159 million people use surface water. This situation necessitates continuous monitoring and improvement of drinking water quality. The water quality index is widely used in the assessment of drinking water quality and is based on the measurement of various physical and chemical parameters. In this study, taking the drinking water treatment plant in Ereğli district of Zonguldak province of Turkey as an example, it is aimed to digitize and optimize the classical jar testing process. The system is designed to be integrated with iot, fog computing and cloud-based machine learning infrastructures. With the sensors used in the iot layer, parameters such as turbidity, ph, tds, temperature, water level and rainfall status of the inlet water were measured at 5-minute intervals and transmitted to the raspberry pi device via nodemcu esp8266 via mqtt protocol. With the nodered application running on the raspberry pi, the data was saved to the influxdb cloud database and visualized with the grafana interface. Thanks to this structure, the remote traceability of the system has been increased, intervention times have been reduced and operational security has been ensured. The jar tests performed in the laboratory environment were digitized with a web-based interface and these data were analyzed with machine learning algorithms. In particular, the support vector regression algorithm showed the highest performance in tests on 502 samples (MAE: 0.29, MAPE: 0.11, MSE: 0.16, RMSE: 0.40, R²: 0.61). The fact that the model produces results in only 1.27 seconds shows that it is sufficient for real-time forecasting applications. The web-based user interface allows operators to enter input parameters and see the output turbidity instantly, and the "start training" module in the system allows the model to be updated automatically with new data.