A pump stops, pressure falls, flow is lost and within three seconds the alarm list holds twenty rows. The operator acknowledges and restarts. The next morning someone asks why the line stopped. The answer is in the alarm history, provided the history has been kept and its order can be trusted.
What is root cause analysis in a process?
Root cause analysis is a method for finding the factor whose removal stops a disturbance from happening again. In quality management it uses tools such as the five whys and the fishbone diagram. In a process disturbance the first and most important question is more concrete: what happened first? Only once the first event is found is it worth asking why it happened.
Why alarm history?
An alarm is an event with a timestamp given by the automation system. When alarms are stored with that original timestamp, their order tells the story of what happened. In an alarm flood the consequences often arrive in the same second as the cause, so the order shows only in the milliseconds.
If the timestamp is rounded to seconds or replaced with the moment of storage, no tool can bring the order back. The most important requirement for storage is therefore that the source system's timestamp is kept intact.
The method step by step
- Set the time window. Choose a period that starts a little before the disturbance and ends when the process is stable again.
- Sort the alarms by their original timestamp. Show milliseconds if the source system provides them.
- Find the first one. The first alarm is not always the cause, but it is the best candidate. Look also at what happened just before it.
- Separate the consequences. Group alarms that belong to the same process area or equipment. Consequences typically follow the cause in a recognisable order.
- Check the measurements. An alarm trips only when a limit is crossed. The measurement may have started to move minutes earlier, and the trend shows which measurement moved first.
- See whether it recurs. A Pareto analysis shows which alarms become active most often, and a duration analysis which take the most time. A one-off and a weekly repeat call for different actions.
What the method does not do
- Precision cannot be improved afterwards. If the source system stamps to the second, there is no millisecond order to recover.
- The clocks of several systems must be synchronised. Otherwise the order of alarms between two systems cannot be trusted.
- The daily rhythm confuses. If a pump starts at six every morning, its movement is not the cause of a disturbance even if it is the first thing in the window.
- A person draws the conclusion. The analysis gives the order, the counts and the durations. What they mean in the process is known by someone who knows the plant.
How does DataPortia™ support root cause analysis?
DataPortia reads alarms and events from the automation system's OPC UA server through an Alarms & Conditions subscription and stores them with the timestamp the server gave. The alarm history offers:
- filtering by time range, severity and area, and a setting to show milliseconds (seconds by default)
- a Pareto analysis by activations, a duration analysis, a severity distribution and a timeline
- export to CSV, Excel and PDF, and scheduled alarm reports.
AI root cause analysis
In the root cause analysis of the AI analysis add-on you choose a station and a time window, not individual measurement points. The server goes through every logged measurement point of the station, works out which one moved first and when, fetches the alarms of the same period and hands the local language model a ready-made sequence of events to interpret. The findings cards at the top of the result are computed directly from the measured data; the language model writes nothing in them.
The daily rhythm is taken into account. Each measurement point is compared with how it moved at the same time of day on up to seven earlier days:
- A change that happened at about the same time on at least two of those days counts as usual and is not reported as an event. A morning start-up, for example, is not shown as the first thing that moved in a disturbance.
- A usual change that did not happen, such as a morning start-up that never came, is listed separately.
- A measurement point with less than four days of history is compared only with the hours before the period, and the analysis says so.
The root cause analysis uses measurement data at one-minute resolution. If several measurement points moved within the same minute, the analysis says they moved simultaneously instead of guessing an order. Order finer than that comes from the alarm timestamps. The language model runs on the plant's own server, so the data does not leave the network.
Reading alarms is covered in more detail in OPC UA Alarms & Conditions, and the AI analysis on the local AI analysis page.
Try DataPortia for 30 days: the trial includes every feature except the HA add-on and commits you to nothing. Collect a week of alarm history and see whether the first event of the next disturbance stands out from the flood.