From Dynamics to Relationships in Data.
Most monitoring solutions focus on high-value equipment. By the time an alarm is triggered, the fault has already developed, and downtime is often unavoidable. Yet the majority of equipment failures are not caused by simple wear and tear. They arise from how assets are operated, controlled, and exposed to varying process conditions.
This implies that we should monitor the process, not just individual machines. Most machine learning-based process monitoring solutions model system dynamics captured by sensors. However, when a process occurs, it first disrupts the relationships between process variables. It often takes time until any process variable sigificantly deviate from its normal operating behavior. At that stage, multiple sensors may deviate simultaneously, making it difficult to identify root causes from secondary effects and determine appropriate mitigation actions.
By modeling and montiroing the relationships between process variables insread, we can detect faults at an ealrly stage, before abnormal process conditions impact equipment performance or product quality. At the same time, tracking shifts in these relationships give us insights into how the faults propogate through the process, helping to identify the root causes and guide effective migitation.
