Joe Alford on Alarm Management

Alford, who is a retired engineering advisor (Process God) at Eli Lilly, started by talking about automobile alarms. In the industrial plant situation:
  • large #of configured alarms
  • many nuisance alarms
  • cost of configuring a new alarm is essentially free
  • situation getting worse with indreasing numbers of smart field devices
  • minimal creation of "intelligent" alarms
  • little data mining pursued to extract information and knowledge content of data
According to the PAS alarm management handbook, alarms have gone from about 50 to over 10,000 per operator since the 1960s. Batch Processes:
  • multiple process steps, phases
  • many transients
  • process loads, setpoints = f(time) little if any steady state operation
  • desire to sort and query alarm records by lot number, step, phase, etc.
  • desire to visualize data via "relative time" rather than calendar time
Alarms can occur at many levels. Safety critical elements can generate alarms in the field, alarms can be generated in the PLC, PC, DCS, SIS and even third party systems like data historians. Viision in creating batch alarms
  • adhere to definition of alarms: i.e., an alert to "abnormal situations requiring a response."
  • Tag alarm records with sufficient information needed for display, sorts, queries, and report generation...including lot number, categor, priorit, process step/phase
  • generate more intelligent alarms-- make use of all available relevant information (e.g. "if-then" rules incorporating redundant sensors, trend slopes, other correlated variables)
Vision for displaying batch alarms
  • eliminate nuisance alarms
  • do not include 'notifications' on alarm displays
  • use unique colors in displaying alarms (if red is used to display high priority alarms, avoid red for other info on console)
  • minimize multiple alarms per "abnormal" event
  • ensure that alarms ALERT, INFORM, and GUIDE
Alford presented a method to compute alarm bandpass from data mining historical alarm information. He presented an example of how an alarm can "guide" the operator. Vision for mining gold from batch alarm records:
  • records easy to access and understand-- non crypitc
  • sufficient alarm tag information to accommodate expected queries, sorts and batch reports: e.g., lot number, category, process step and phase
  • displayable in relative time
  • utilities to include Pareto chart (alarm frequency) and other alarm metrics: e.g., alarms per day
  • ability to combine discrete and continuous trend data to facilitate abnormal event analysis
  • abilityto use of PCA, PLS, neural nets., etc. in developing virtual sensors and or identifying a small number of key independent variables that are predictive of important process outputs.
Alford showed a combined continuous trend and discrete event data to show how much time you can save. Role of the batch standard
  • separates equipment from process
  • encourages sequential flow diagrams
  • identifies process states
Role of S95
  • connect alarms to the enterprise
Role of S-18 you'll see more as this standard is released for comment