Data-driven qualitative and model-based quantitative approaches to fault diagnosis

Abnormal Situation Management (ASM) and Process Hazards Analysis (PHA) are two tasks aimed at improving the design and performance of a process, while ensuring safety of people and property involved and addressing environmental, occupational safety and health related concerns. ASM deals with timely detection and diagnosis of faults, situation assessment and countermeasure planning. PHA is concerned with safety issues during design/modifications stages. The problem of fault diagnosis is made considerably difficult by the scale and complexity of modern operations, and as such, there is considerable incentive to develop intelligent systems that assist the operator in making informed decisions and taking actions quickly. In this work, we look at two diagnostic methodologies: Qualitative Trend Analysis (QTA), a data-driven qualitative technique and Observers a model-based quantitative method. QTA aims to utilize the trend signatures in the measurements arising from malfunctions to reason about process behavior. The first step of feature-extraction deals with adaptive identification of important trends from real-time noisy data followed by feature-classification maps the features to the process states. We present a novel QTA technique based on interval-halving to extract trends followed by fuzzy-inferencing for trend-to-state mapping. The first stage effectively addresses the issue of time-scale trend identification and the second deals with the inherent uncertainty in these features. While QTA provides for qualitative trend-based temporal reasoning, it is by design data-driven and, thus, restricted in its capability. As such, the other area of focus is the model-based quantitative approach, in particular observer-based diagnosis which exploits the analytical redundancy inherent in a process model. We evaluate the efficacy of a linear, an extended linear and a nonlinear observer in diagnosing multiple faults. Finally, we note that the inherent objectives in ASM and PHA are very similar i.e., to identify hazards, to avoid/mitigate them and plan for emergencies. The PHA results contain valuable cause-consequence information, safeguards and other operability issues from which ASM can potentially benefit. To this end, an integrated framework combining both these tasks in a synergistic manner is proposed