There are floods of alarms—and then there are so many alarms that you can't even see the floods. This used to be the situation in the leaching process at AREVA Resource Canada Inc.'s uranium mill in McClean Lake, Saskatchewan, where operators had been dealing with 60 alarms every 10 minutes as recently as January 2016.
Starting operations in 1995, the mill and its process application, systems and equipment were built by several different contractors with different process control methods and technologies, which led to a multiplicity of alarm systems that grew increasingly unwieldy over the years. "There was an absence of alarm guidelines and specifications," said Adrien Rudulier, P. Eng., AREVA. "Alarms were designed and implemented differently and subjectively because the original build had many consultants, subsequent capital projects had numerous more consultants, and there were hundreds of site change controls.
The resulting one alarm every 10 seconds became white noise in the minds of AREVA's operators, and motivated Rudulier and his colleagues to undertake an alarm rationalization and management project in 2015. "We used DeltaV Analyze software from Emerson to obtain baseline statistics for project justification, and sort out alarms based on their DeltaV area assignments," said Rudulier. "We also sought assistance from safety and security consultant exida, which helped us develop a customized alarm philosophy, set up a SILAlarm master alarm database, and provide onsite training."
Rudulier and Darren Wells of Spartan Controls presented "Everything is alarming: Implementing alarm management at McClean Lake" at the 2017 Emerson Global Users Exchange in Minneapolis.
Process yellowcake safely
The McClean Lake mill is unusual in that it can handle low- to very high-grade ores. During 1999 to 2010, its average ore grade was about 1%, but from 2015 to the present, its average ore grade was 17%. "The mill is presently configured and licensed to produce 24 million pounds of U3O8 (yellowcake) per year," added Rudulier. "It can accept ore that can range from 0.5% U to 30% U, and produce 80-84% U yellowcake."
To achieve these results, Rudulier reported that the mill's operators interact with its processes in several ways. "They're responsible for the safe operation of circuits," he explained. "There's some manual control and monitoring (rounds, sampling, etc.), and some remote control and monitoring from a control room via DeltaV. Alarming is inherent to all systems, both local and remote, to inform operators of any problems. Each circuit has a dedicated control room and operator. All circuits are monitored by the central control room for additional oversight."
Rudulier added, the intent of AREVA's alarm management project was to give it a reliable tool to keep its operators and processes safe. "The key was to get buy-in from management and everyone else," said Rudulier. "We secured that when we issued our first alarm philosophy, and defined roles and responsibilities. Once we got an estimate of the rationalization project, it began to smooth out."
Appropriate alarm design
To rationalize its alarms and develop a new alarm management system, Rudulier reported that his team followed the ISA 18.2 standard and its design principles. "Each alarm should inform and guide," he said. "Every alarm presented to the operator must be unique, and every alarm must have a defined/required response. Alarms should also be explicitly designed to take into account human limitations. We also had to allow adequate time for operators to analyze the situation, and carry out a defined response."
Consisting of members from operations, training, instrumentation, metallurgy, health and safety, environment and engineering, AREVA's cross-functional alarm team sought a target alarm distribution of 60-80% low, 10-20% medium, 3-7% high, and less than 1% emergency. Its industrial target for alarms per 10 minutes was that zero to one alarm was acceptable, two to four was manageable, five to nine was demanding, and more than 10 was unacceptable and impossible.
First attempt lessons
Despite gaining crucial buy-in, Rudulier added that the team's first attempt at alarm rationalization experienced some unexpected challenges, even though it made good use of DeltaV Analyze and exida SILAlarm rationalization software, which were used to step though each of the mill's approximately 12,000 configured alarms.
"We tried to deal with the bad actors first, but this was a bad idea because we got bogged down trying to refine our consequence table and rationalize every part of an alarm as per our alarm philosophy, and our sessions got too long—about eight hours a day for five days," says Rudulier. "We opted for a first-pass strategy that limited alarms by simply asking ‘is it an alarm?’ Our alarm priorities were that every alarm must also have a cause and consequence. If there's no consequence, then it's not an alarm."
The team eventually got into a groove of monthly two-day sessions, and found they were able to rationalize about 300 alarms during each 12-hour day. Its rationalization process:
"Our first changes were implemented in November 2016, and so far, we've rationalized about 46% of our alarms, or about 5,400 of our total 12,000 alarms," added Rudulier. "Our first-pass rationalization will be done by December 2018, but we'll keep on doing a continuous monitoring and improvement process. The good news is that the bad actors are now easily identifiable.
"We've not only reduced alarms in our leaching process by 98.6% from 17,280 in October 2016 to 245 in November 2016, but our operators who used to be faulted for not reacting to alarms can now react to them, and they aren't in a constant flood state."