Our algorithm is based on: artificial intelligence, machine learning and data mining.
We study the monitored system at its normal “healthy” operation. It could be any stationary process – machines or pipelines. We check and learn the correlation between the different parameters collected by the sensors installed on the system. Any novelty from normal behavior is considered at first as a new case study. If we detect the situation continues – meaning the correlation is broken and we will provide an alert. Since our system is adaptive and learning it will not alert in transient stages, in this case it will continue operating normally with no need for calibration.
On the contrary to other existing monitoring systems our technique is based on the systems normal behavior characterization. While other systems compare the current state of the system to a large DB of faulty data. As we know, normal data is always available and can be well characterized using statistical training techniques. While a DB of faulty data as big as it is cannot include all the malfunctions that can occur in complex systems.
Any deviation from the “normal behavior” is considered a “novelty” which mostly will be caused by an appearing defect that eventually becomes a failure.
Benefit: Our algorithm tracks your system on-line. Thanks to its ability to learn, it can alert an abnormal behavior which predicts an evolving malfunction.