Framework with Data-Analytic for Fault Detection and Performance Prediction of a Steam Boiler: A Case Study

This work reports data-driven fault diagnosis and performance prediction framework for a boiler. In-house experimental setup of a boiler and heat exchanger is used to generate the data for various parameters. Various combinations of two, three and four state variables are used to develop eight different regression models using PYTHON in-house code. Performance of all these models is analyzed by comparing different statistical parameters like sum squared residual, mean squared residual and standard error. It is noticed that models with two state variables namely, steam temperature and volume of water or steam temperature and volume of steam predict the steam pressure with optimum error. These models are further tested to diagnose the fault which is added numerically in terms of upward and downward temperature drifts. The models are statistically robust and can diagnose the fault and predict the pressure which is within the target as evidenced through EWMA charts.