Fault detection for IoT hydrogen refueling station system using a combined hidden Markov model mixed with Gaussian

To solve environmental problems, a major shift in energy sources for vehicles, from coal fuel to hydrogen, has been underway. Thus, the number of hydrogen refueling stations are increasing rapidly. However, the burn in failure caused by inherent quality problems, installation problems, or inexperienced operation at the initial stage of an asset's life results in huge maintenance costs and further impedes the development of the hydrogen car industry. This paper proposes an approach to detect abnormal states that can be implemented in a plant information system that collects and manages real-time data and events from SCADA. First, events, such as on or off states of valves and switches, are used to define the operating modes. From multiple real-time variables, the Gaussian mixture model (GMM) is used as the priori probability distribution for each mode. Finally, as a measure to evaluate abnormality, the likelihoods of situations are calculated from the posteriori probabilities that are estimated by the hidden Markov model (HMM). The performance of the proposed approach was tested with data collected from the currently operating Seosan Hydrogen Refueling Station of Korea Gas Technology Corporation. Additionally, the strength of the proposed GMM-HMM method is compared.