DESTRUCTIVE PROCESS EVENT DETECTION AND MITIGATION

Problem background

The client wished to use existing instrumentation data to detect, measure and record subtle but unfavourable operational conditions, which posed a cumulative risk of damage to process infrastructure, eventually necessitating significant and expensive down-time (in terms of parts, loss of production and staff time).

​

How the problem was approached

Various positions in the process chain were outfitted with appropriate sensing technologies; the data from which were then compiled, post-processed and correlated.  A unique signature was identified in the captured data frequency information which strongly correlated to the onset of dysfunction.  An artificial neural network was written and trained to identify the appropriate patterns of interest and to provide inference on the relative severity and approximate location of the primary sensing detector.

​

Results

The system was able to successfully discriminate temporal domain data and provide usable, practical information to the customer.

​

Key terms

• Augmentation of existing process

• Custom data processing engine

• Frequency analysis (DFT/FFT)

• 'Real world' useable output