Getting More Out of Your Time Series Data

Getting More Out of Your Time Series Data

Plant operating profits are a function of thermal performance and reliability. In many industries liked defense, power, and oil and gas, reliability of equipment is a very important factor and it is common to see machines operating for long durations without any outage. Forced outages of any critical equipment in such industries can result in huge financial losses and reduce operational preparedness.

Reliability of equipment can be maintained to a certain extent by following best practices in O&M and RCM techniques. However, with the limitations of existing monitoring systems, industries are losing significant amounts of money every year due to failure of equipment. Statistics indicate that such losses are in the millions of dollars per year in forced outages. It is interesting to note that the initiating factors in many equipment failures originate as minor anomalies.


An anomaly can be described as a discrepancy or deviation from an established rule or trend. Every machine/process has a unique operating profile or “signature”. This signature is the response of the machine or process to various internal and external factors under given conditions. Various instruments mounted on machines (like temperature, pressure, vibration, flow, current, and voltage sensors) indicate the response of the machine to such conditions.

Various types of anomalies occur in industries –

  • Equipment Anomalies
  • Process Anomalies
  • Instrument Anomalies
  • Data Anomalies


Anomaly Detection – Predictive Analytics

Advances in data analytics give us the capability to use a combination of modeling tools and predictive analytic techniques to build behavioral “signatures”. These signatures in turn are used to “predict” various process parameters. Such signatures are built using thermodynamics analysis, artificial intelligence and/or statistical methods depending on the parameter we are trying to predict. They are very sensitive and are extremely accurate at modeling even complex processes.

A significant deviation between “Actual (measured) value” and “Predicted value” indicates a deviation in its behavior. This deviation usually is the onset of an anomaly. Proprietary algorithms differentiate between sensor issues and real anomalies and calculate the “confidence level” of prediction to reduce false positives.

Advantages of Anomaly Detection Systems.

1. Identify problems that go undetected by normal procedures

  • Reduce forced outages
  • Improve safety


2. Aids in planning maintenance actions and improving operational preparedness

  • Man-power planning
  • Procurement of spares


Hybrid Modeling Techniques

We use a combination of neural networks, statistics and detailed component-level thermodynamic models to analyze data and identify anomalies. Such hybrid models reduce false positives and provide additional information on the current health and projected degradation of various major and minor components of any industrial equipment.