Alarm analysisDevelopment of alarm analysis algorithms
Since the introduction of distributed control systems (DCS), a problem that has arisen in the process industry is that the high frequency of alarms has a negative impact on the work efficiency of on-site staff. Because of the redundancy built into these alarm systems, there are many unnecessary alarms and also alarms that merely result from having improper settings. What this means is that a disturbance in the process can lead to a barrage of alarms, making it difficult to identify the real cause of the fault and leading to delays in implementing corrective countermeasures.Until now the only way to deal with these problems was to collect the alarm log files and, starting with those points with the highest alarm frequency, investigate each possible cause. Naturally this was a very time-consuming procedure.
What we have done at Yamatake is to develop a type of event analysis that will facilitate alarm reduction. This event analysis involves the application of data mining techniques to a body of event data comprising log data and the time of occurrence of discrete events, such as equipment malfunctions and natural events like earthquakes and floods. This technique employs a point process model, a type of probability model, to calculate cross-correlation function values between discrete events. In this way, we are able to quantify the relevance and time lag observed between one event and another. Adopting this approach for situations in which multiple events arise from continuous processes, such as those used in the chemical industry, means that we can group chain-reaction events and analyze the order in which they occur. Two patent applications have already been made for this technology.
As a result of applying this to an actual process, in just a few hours it was possible to develop a proposal that would reduce by over half the number of alarms at a plant (these were occurring many thousand times a month). This included prioritization of those events to eliminate, based on grouping, and an analysis of the causes of these events, based on the order of their occurrence. This test case confirmed our ability to conduct efficient alarm analysis using this technique.
This application of event data is not of a specialized nature, and therefore it should be applicable to other fields, using environmental or medical data, for example. We are currently exploring these possibilities.
 Event analysis technology |
<<Measurement data processing
