Incipient fault detection approach based on piecewise linear shape-based global embedding for steam turbine plants

The advancement of modern thermal power plants has led to higher requirements for real-time monitoring of industrial process safety status. Therefore, the last two decades have also witnessed a significant growth in fault detection strategies for industrial processes. To the best of our knowledge, most existing fault detection techniques can effectively handle abrupt faults. Most faults in power plants, however, originate from the prolonged evolution of the initial abnormal event. This necessitates early detection by the operator to anticipate incipient faults and schedule maintenance promptly. To achieve this goal, this paper introduces a novel method for detecting incipient faults in steam turbine plants called Piecewise Linear Shape-Based Global Embedding (PLSGE). This method involves integrating the construction of piecewise linear shapes into a global structure-based data projection. The primary purpose of constructing piecewise linear shapes is to uncover the nonlinear underlying structure of the process data collected during the evolution of the turbine system from normal operating conditions to faulty operating conditions. The framework of the incipient fault detection scheme is then established based on the constructed shape. Two fault indicators are provided to characterize the degree of deviation between the current data and the shape model, as well as the degree of loss of efficacy of the established shape model, respectively. Finally, the case studies of simulated and real steam turbine plants validate that the proposed method offers superior early detection performance for slowly developing additive faults compared to existing methods.