Contents x
      Classification
      • 19 Sep 2024
      • 1 Minute to read
      • Print
      • Dark
        Light
      Contents

      Classification

      • Updated on 19 Sep 2024
      • 1 Minute to read
      • Print
      • Dark
        Light
      Article summary

      How you can use a Classification Model in your control system.

      Classification models are ideal for analyzing data and determining which discrete state or condition a process is in. A common application of this is anomaly detection. In the chart below, each detected anomaly is highlighted by an orange dot. This warning can be used by the control system or operations as a form of “active preventative maintenance” to steer the process away from anomalous conditions.

      What is an anomaly detection model?

      An anomaly detection model identifies unusual or abnormal patterns in data that deviate from the norm. It is commonly used to detect outliers or unexpected behaviors in systems, such as equipment malfunctions or unusual process values. These models learn what is considered “normal” and flag data points that fall outside this range, helping to identify potential issues early. Anomaly detection is widely used in applications like fraud detection, fault detection in machinery, and monitoring of industrial processes.

      Typical Use Cases

      When implementing a anomaly style classification model into your control system, you must first determine what actions you would like to happen once an anomaly is detected. Here we will go over typical PLC or DCS use cases:

      Alerting the operator

      The most common use case for an anomaly detection model will start with an alert to the operator. The implementation of this will vary based on the control system and the organization's standards, but it is typically best practice to have a visual indication of the alarm on the HMI near the piece of equipment that is experiencing the anomaly.

      Taking Preventative Action

      It may also be advisable to take a more aggressive approach to mitigating the anomaly by programming in a safety override that can steer the process out of harms way. This may involve shutting down a piece of equipment that is causing the upset or turning on another piece of equipment that can help resolve the issue.

      Assisting a DRL Model

      One of the most powerful techniques for using a classification model is to utilize its state prediction when training a Deep Reinforcement Learning (DRL) model. This can allow the DRL model to "learn" the different nuances in how to control the process during anomalous conditions.

      Example Logic Diagram

      The following diagram is an extremely simplified use case that:

      • Alerts the operator using an HMI alert.

      • Works with the existing interlocks to shut down the equipment during an anomalous condition.


      What's Next