A sub-principal component of fault detection (PCFD) modeling method and its application to online fault diagnosis

A sub-principal component of fault detection (PCFD) modeling method is proposed for online fault diagnosis for multiphase batch processes. Without the requirement of priori process knowledge, an automatic phase division algorithm is proposed to separate the abnormal batch process into multiple phases by capturing the changes of fault deviations throughout the batch. The similar fault characteristics are grouped into the same phase while different fault characteristics are classified into different phases. PCFD algorithm is then used to decompose the fault deviations relative to normal in different phases. Phase-representative fault diagnosis model is developed to capture the similar fault characteristics within the same phase and multiphase sub-phase models across different phases. Critical-to-diagnosis fault phases are defined and identified which have significant contributions to online fault diagnosis. Based on the identified phase nature and fault diagnosis relationships, an online fault diagnosis strategy is developed to isolate the possible abnormality cause realtime. The applications of the proposed scheme to a typical multiphase batch process, injection molding, show that the proposed analysis and fault diagnosis are not only effective but are also able to enhance fault process understanding and identify specific periods for fault diagnosis in time.